Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1 | /* |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2 | * Generic process-grouping system. |
| 3 | * |
| 4 | * Based originally on the cpuset system, extracted by Paul Menage |
| 5 | * Copyright (C) 2006 Google, Inc |
| 6 | * |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 7 | * Notifications support |
| 8 | * Copyright (C) 2009 Nokia Corporation |
| 9 | * Author: Kirill A. Shutemov |
| 10 | * |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 11 | * Copyright notices from the original cpuset code: |
| 12 | * -------------------------------------------------- |
| 13 | * Copyright (C) 2003 BULL SA. |
| 14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. |
| 15 | * |
| 16 | * Portions derived from Patrick Mochel's sysfs code. |
| 17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel |
| 18 | * |
| 19 | * 2003-10-10 Written by Simon Derr. |
| 20 | * 2003-10-22 Updates by Stephen Hemminger. |
| 21 | * 2004 May-July Rework by Paul Jackson. |
| 22 | * --------------------------------------------------- |
| 23 | * |
| 24 | * This file is subject to the terms and conditions of the GNU General Public |
| 25 | * License. See the file COPYING in the main directory of the Linux |
| 26 | * distribution for more details. |
| 27 | */ |
| 28 | |
Joe Perches | ed3d261 | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 30 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 31 | #include <linux/cgroup.h> |
eparis@redhat | 2ce9738 | 2011-06-02 21:20:51 +1000 | [diff] [blame] | 32 | #include <linux/cred.h> |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 33 | #include <linux/ctype.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 34 | #include <linux/errno.h> |
eparis@redhat | 2ce9738 | 2011-06-02 21:20:51 +1000 | [diff] [blame] | 35 | #include <linux/init_task.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 36 | #include <linux/kernel.h> |
| 37 | #include <linux/list.h> |
Jianyu Zhan | c9482a5 | 2014-04-26 15:40:28 +0800 | [diff] [blame] | 38 | #include <linux/magic.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 39 | #include <linux/mm.h> |
| 40 | #include <linux/mutex.h> |
| 41 | #include <linux/mount.h> |
| 42 | #include <linux/pagemap.h> |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 43 | #include <linux/proc_fs.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 44 | #include <linux/rcupdate.h> |
| 45 | #include <linux/sched.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 46 | #include <linux/slab.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 47 | #include <linux/spinlock.h> |
Tejun Heo | 1ed1328 | 2015-09-16 12:53:17 -0400 | [diff] [blame] | 48 | #include <linux/percpu-rwsem.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 49 | #include <linux/string.h> |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 50 | #include <linux/sort.h> |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 51 | #include <linux/kmod.h> |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 52 | #include <linux/delayacct.h> |
| 53 | #include <linux/cgroupstats.h> |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 54 | #include <linux/hashtable.h> |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 55 | #include <linux/pid_namespace.h> |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 56 | #include <linux/idr.h> |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
Mike Galbraith | c4c27fb | 2012-04-21 09:13:46 +0200 | [diff] [blame] | 58 | #include <linux/kthread.h> |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 59 | #include <linux/delay.h> |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 60 | #include <linux/atomic.h> |
Tejun Heo | e93ad19 | 2016-01-19 12:18:41 -0500 | [diff] [blame] | 61 | #include <linux/cpuset.h> |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 62 | #include <linux/proc_ns.h> |
| 63 | #include <linux/nsproxy.h> |
Martin KaFai Lau | 1f3fe7e | 2016-06-30 10:28:42 -0700 | [diff] [blame] | 64 | #include <linux/file.h> |
Tejun Heo | bd1060a | 2015-12-07 17:38:53 -0500 | [diff] [blame] | 65 | #include <net/sock.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 66 | |
Tejun Heo | e25e2cbb | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 67 | /* |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 68 | * pidlists linger the following amount before being destroyed. The goal |
| 69 | * is avoiding frequent destruction in the middle of consecutive read calls |
| 70 | * Expiring in the middle is a performance problem not a correctness one. |
| 71 | * 1 sec should be enough. |
| 72 | */ |
| 73 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ |
| 74 | |
Tejun Heo | 8d7e6fb | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 75 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
| 76 | MAX_CFTYPE_NAME + 2) |
| 77 | |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 78 | /* |
Tejun Heo | e25e2cbb | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 79 | * cgroup_mutex is the master lock. Any modification to cgroup or its |
| 80 | * hierarchy must be performed while holding it. |
| 81 | * |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 82 | * css_set_lock protects task->cgroups pointer, the list of css_set |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 83 | * objects, and the chain of tasks off each css_set. |
Tejun Heo | e25e2cbb | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 84 | * |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 85 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
| 86 | * cgroup.h can use them for lockdep annotations. |
Tejun Heo | e25e2cbb | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 87 | */ |
Tejun Heo | 2219449 | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 88 | #ifdef CONFIG_PROVE_RCU |
| 89 | DEFINE_MUTEX(cgroup_mutex); |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 90 | DEFINE_SPINLOCK(css_set_lock); |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 91 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 92 | EXPORT_SYMBOL_GPL(css_set_lock); |
Tejun Heo | 2219449 | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 93 | #else |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 94 | static DEFINE_MUTEX(cgroup_mutex); |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 95 | static DEFINE_SPINLOCK(css_set_lock); |
Tejun Heo | 2219449 | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 96 | #endif |
| 97 | |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 98 | /* |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 99 | * Protects cgroup_idr and css_idr so that IDs can be released without |
| 100 | * grabbing cgroup_mutex. |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 101 | */ |
| 102 | static DEFINE_SPINLOCK(cgroup_idr_lock); |
| 103 | |
| 104 | /* |
Tejun Heo | 34c0625 | 2015-11-05 00:12:24 -0500 | [diff] [blame] | 105 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications |
| 106 | * against file removal/re-creation across css hiding. |
| 107 | */ |
| 108 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); |
| 109 | |
| 110 | /* |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 111 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires |
| 112 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. |
| 113 | */ |
| 114 | static DEFINE_SPINLOCK(release_agent_path_lock); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 115 | |
Tejun Heo | 1ed1328 | 2015-09-16 12:53:17 -0400 | [diff] [blame] | 116 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
| 117 | |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 118 | #define cgroup_assert_mutex_or_rcu_locked() \ |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 119 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
| 120 | !lockdep_is_held(&cgroup_mutex), \ |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 121 | "cgroup_mutex or RCU read lock required"); |
Tejun Heo | 780cd8b | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 122 | |
Ben Blum | aae8aab | 2010-03-10 15:22:07 -0800 | [diff] [blame] | 123 | /* |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 124 | * cgroup destruction makes heavy use of work items and there can be a lot |
| 125 | * of concurrent destructions. Use a separate workqueue so that cgroup |
| 126 | * destruction work items don't end up filling up max_active of system_wq |
| 127 | * which may lead to deadlock. |
| 128 | */ |
| 129 | static struct workqueue_struct *cgroup_destroy_wq; |
| 130 | |
| 131 | /* |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 132 | * pidlist destructions need to be flushed on cgroup destruction. Use a |
| 133 | * separate workqueue as flush domain. |
| 134 | */ |
| 135 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; |
| 136 | |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 137 | /* generate an array of cgroup subsystem pointers */ |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 138 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 139 | static struct cgroup_subsys *cgroup_subsys[] = { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 140 | #include <linux/cgroup_subsys.h> |
| 141 | }; |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 142 | #undef SUBSYS |
| 143 | |
| 144 | /* array of cgroup subsystem names */ |
| 145 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, |
| 146 | static const char *cgroup_subsys_name[] = { |
| 147 | #include <linux/cgroup_subsys.h> |
| 148 | }; |
| 149 | #undef SUBSYS |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 150 | |
Tejun Heo | 49d1dc4 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 151 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
| 152 | #define SUBSYS(_x) \ |
| 153 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ |
| 154 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ |
| 155 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ |
| 156 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); |
| 157 | #include <linux/cgroup_subsys.h> |
| 158 | #undef SUBSYS |
| 159 | |
| 160 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, |
| 161 | static struct static_key_true *cgroup_subsys_enabled_key[] = { |
| 162 | #include <linux/cgroup_subsys.h> |
| 163 | }; |
| 164 | #undef SUBSYS |
| 165 | |
| 166 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, |
| 167 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { |
| 168 | #include <linux/cgroup_subsys.h> |
| 169 | }; |
| 170 | #undef SUBSYS |
| 171 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 172 | /* |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 173 | * The default hierarchy, reserved for the subsystems that are otherwise |
Tejun Heo | 9871bf9 | 2013-06-24 15:21:47 -0700 | [diff] [blame] | 174 | * unattached - it never has more than a single cgroup, and all tasks are |
| 175 | * part of that cgroup. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 176 | */ |
Tejun Heo | a2dd424 | 2014-03-19 10:23:55 -0400 | [diff] [blame] | 177 | struct cgroup_root cgrp_dfl_root; |
Tejun Heo | d0ec423 | 2015-08-05 16:03:19 -0400 | [diff] [blame] | 178 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
Tejun Heo | 9871bf9 | 2013-06-24 15:21:47 -0700 | [diff] [blame] | 179 | |
Tejun Heo | a2dd424 | 2014-03-19 10:23:55 -0400 | [diff] [blame] | 180 | /* |
| 181 | * The default hierarchy always exists but is hidden until mounted for the |
| 182 | * first time. This is for backward compatibility. |
| 183 | */ |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 184 | static bool cgrp_dfl_visible; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 185 | |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 186 | /* Controllers blocked by the commandline in v1 */ |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 187 | static u16 cgroup_no_v1_mask; |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 188 | |
Tejun Heo | 5533e01 | 2014-05-14 19:33:07 -0400 | [diff] [blame] | 189 | /* some controllers are not supported in the default hierarchy */ |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 190 | static u16 cgrp_dfl_inhibit_ss_mask; |
Tejun Heo | 5533e01 | 2014-05-14 19:33:07 -0400 | [diff] [blame] | 191 | |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 192 | /* some controllers are implicitly enabled on the default hierarchy */ |
| 193 | static unsigned long cgrp_dfl_implicit_ss_mask; |
| 194 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 195 | /* The list of hierarchy roots */ |
| 196 | |
Tejun Heo | 9871bf9 | 2013-06-24 15:21:47 -0700 | [diff] [blame] | 197 | static LIST_HEAD(cgroup_roots); |
| 198 | static int cgroup_root_count; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 199 | |
Tejun Heo | 3417ae1 | 2014-02-08 10:37:01 -0500 | [diff] [blame] | 200 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
Tejun Heo | 1a57423 | 2013-04-14 11:36:58 -0700 | [diff] [blame] | 201 | static DEFINE_IDR(cgroup_hierarchy_idr); |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 202 | |
Li Zefan | 794611a | 2013-06-18 18:53:53 +0800 | [diff] [blame] | 203 | /* |
Tejun Heo | 0cb51d7 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 204 | * Assign a monotonically increasing serial number to csses. It guarantees |
| 205 | * cgroups with bigger numbers are newer than those with smaller numbers. |
| 206 | * Also, as csses are always appended to the parent's ->children list, it |
| 207 | * guarantees that sibling csses are always sorted in the ascending serial |
| 208 | * number order on the list. Protected by cgroup_mutex. |
Li Zefan | 794611a | 2013-06-18 18:53:53 +0800 | [diff] [blame] | 209 | */ |
Tejun Heo | 0cb51d7 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 210 | static u64 css_serial_nr_next = 1; |
Li Zefan | 794611a | 2013-06-18 18:53:53 +0800 | [diff] [blame] | 211 | |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 212 | /* |
| 213 | * These bitmask flags indicate whether tasks in the fork and exit paths have |
| 214 | * fork/exit handlers to call. This avoids us having to do extra work in the |
| 215 | * fork/exit path to check which subsystems have fork/exit callbacks. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 216 | */ |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 217 | static u16 have_fork_callback __read_mostly; |
| 218 | static u16 have_exit_callback __read_mostly; |
| 219 | static u16 have_free_callback __read_mostly; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 220 | |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 221 | /* cgroup namespace for init task */ |
| 222 | struct cgroup_namespace init_cgroup_ns = { |
| 223 | .count = { .counter = 2, }, |
| 224 | .user_ns = &init_user_ns, |
| 225 | .ns.ops = &cgroupns_operations, |
| 226 | .ns.inum = PROC_CGROUP_INIT_INO, |
| 227 | .root_cset = &init_css_set, |
| 228 | }; |
| 229 | |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 230 | /* Ditto for the can_fork callback. */ |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 231 | static u16 have_canfork_callback __read_mostly; |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 232 | |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 233 | static struct file_system_type cgroup2_fs_type; |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 234 | static struct cftype cgroup_dfl_base_files[]; |
| 235 | static struct cftype cgroup_legacy_base_files[]; |
Tejun Heo | 628f7cd | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 236 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 237 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 238 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 239 | static int cgroup_apply_control(struct cgroup *cgrp); |
| 240 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 241 | static void css_task_iter_advance(struct css_task_iter *it); |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 242 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 243 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
| 244 | struct cgroup_subsys *ss); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 245 | static void css_release(struct percpu_ref *ref); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 246 | static void kill_css(struct cgroup_subsys_state *css); |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 247 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
| 248 | struct cgroup *cgrp, struct cftype cfts[], |
Tejun Heo | 2bb566c | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 249 | bool is_add); |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 250 | |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 251 | /** |
| 252 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID |
| 253 | * @ssid: subsys ID of interest |
| 254 | * |
| 255 | * cgroup_subsys_enabled() can only be used with literal subsys names which |
| 256 | * is fine for individual subsystems but unsuitable for cgroup core. This |
| 257 | * is slower static_key_enabled() based test indexed by @ssid. |
| 258 | */ |
| 259 | static bool cgroup_ssid_enabled(int ssid) |
| 260 | { |
Arnd Bergmann | cfe02a8 | 2016-03-15 00:21:06 +0100 | [diff] [blame] | 261 | if (CGROUP_SUBSYS_COUNT == 0) |
| 262 | return false; |
| 263 | |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 264 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
| 265 | } |
| 266 | |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 267 | static bool cgroup_ssid_no_v1(int ssid) |
| 268 | { |
| 269 | return cgroup_no_v1_mask & (1 << ssid); |
| 270 | } |
| 271 | |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 272 | /** |
| 273 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy |
| 274 | * @cgrp: the cgroup of interest |
| 275 | * |
| 276 | * The default hierarchy is the v2 interface of cgroup and this function |
| 277 | * can be used to test whether a cgroup is on the default hierarchy for |
| 278 | * cases where a subsystem should behave differnetly depending on the |
| 279 | * interface version. |
| 280 | * |
| 281 | * The set of behaviors which change on the default hierarchy are still |
| 282 | * being determined and the mount option is prefixed with __DEVEL__. |
| 283 | * |
| 284 | * List of changed behaviors: |
| 285 | * |
| 286 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" |
| 287 | * and "name" are disallowed. |
| 288 | * |
| 289 | * - When mounting an existing superblock, mount options should match. |
| 290 | * |
| 291 | * - Remount is disallowed. |
| 292 | * |
| 293 | * - rename(2) is disallowed. |
| 294 | * |
| 295 | * - "tasks" is removed. Everything should be at process granularity. Use |
| 296 | * "cgroup.procs" instead. |
| 297 | * |
| 298 | * - "cgroup.procs" is not sorted. pids will be unique unless they got |
| 299 | * recycled inbetween reads. |
| 300 | * |
| 301 | * - "release_agent" and "notify_on_release" are removed. Replacement |
| 302 | * notification mechanism will be implemented. |
| 303 | * |
| 304 | * - "cgroup.clone_children" is removed. |
| 305 | * |
| 306 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup |
| 307 | * and its descendants contain no task; otherwise, 1. The file also |
| 308 | * generates kernfs notification which can be monitored through poll and |
| 309 | * [di]notify when the value of the file changes. |
| 310 | * |
| 311 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and |
| 312 | * take masks of ancestors with non-empty cpus/mems, instead of being |
| 313 | * moved to an ancestor. |
| 314 | * |
| 315 | * - cpuset: a task can be moved into an empty cpuset, and again it takes |
| 316 | * masks of ancestors. |
| 317 | * |
| 318 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag |
| 319 | * is not created. |
| 320 | * |
| 321 | * - blkcg: blk-throttle becomes properly hierarchical. |
| 322 | * |
| 323 | * - debug: disallowed on the default hierarchy. |
| 324 | */ |
| 325 | static bool cgroup_on_dfl(const struct cgroup *cgrp) |
| 326 | { |
| 327 | return cgrp->root == &cgrp_dfl_root; |
| 328 | } |
| 329 | |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 330 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
| 331 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, |
| 332 | gfp_t gfp_mask) |
| 333 | { |
| 334 | int ret; |
| 335 | |
| 336 | idr_preload(gfp_mask); |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 337 | spin_lock_bh(&cgroup_idr_lock); |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 338 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 339 | spin_unlock_bh(&cgroup_idr_lock); |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 340 | idr_preload_end(); |
| 341 | return ret; |
| 342 | } |
| 343 | |
| 344 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) |
| 345 | { |
| 346 | void *ret; |
| 347 | |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 348 | spin_lock_bh(&cgroup_idr_lock); |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 349 | ret = idr_replace(idr, ptr, id); |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 350 | spin_unlock_bh(&cgroup_idr_lock); |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 351 | return ret; |
| 352 | } |
| 353 | |
| 354 | static void cgroup_idr_remove(struct idr *idr, int id) |
| 355 | { |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 356 | spin_lock_bh(&cgroup_idr_lock); |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 357 | idr_remove(idr, id); |
Tejun Heo | 54504e9 | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 358 | spin_unlock_bh(&cgroup_idr_lock); |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 359 | } |
| 360 | |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 361 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
| 362 | { |
| 363 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; |
| 364 | |
| 365 | if (parent_css) |
| 366 | return container_of(parent_css, struct cgroup, self); |
| 367 | return NULL; |
| 368 | } |
| 369 | |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 370 | /* subsystems visibly enabled on a cgroup */ |
| 371 | static u16 cgroup_control(struct cgroup *cgrp) |
| 372 | { |
| 373 | struct cgroup *parent = cgroup_parent(cgrp); |
| 374 | u16 root_ss_mask = cgrp->root->subsys_mask; |
| 375 | |
| 376 | if (parent) |
| 377 | return parent->subtree_control; |
| 378 | |
| 379 | if (cgroup_on_dfl(cgrp)) |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 380 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
| 381 | cgrp_dfl_implicit_ss_mask); |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 382 | return root_ss_mask; |
| 383 | } |
| 384 | |
| 385 | /* subsystems enabled on a cgroup */ |
| 386 | static u16 cgroup_ss_mask(struct cgroup *cgrp) |
| 387 | { |
| 388 | struct cgroup *parent = cgroup_parent(cgrp); |
| 389 | |
| 390 | if (parent) |
| 391 | return parent->subtree_ss_mask; |
| 392 | |
| 393 | return cgrp->root->subsys_mask; |
| 394 | } |
| 395 | |
Tejun Heo | 95109b6 | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 396 | /** |
| 397 | * cgroup_css - obtain a cgroup's css for the specified subsystem |
| 398 | * @cgrp: the cgroup of interest |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 399 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
Tejun Heo | 95109b6 | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 400 | * |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 401 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
| 402 | * function must be called either under cgroup_mutex or rcu_read_lock() and |
| 403 | * the caller is responsible for pinning the returned css if it wants to |
| 404 | * keep accessing it outside the said locks. This function may return |
| 405 | * %NULL if @cgrp doesn't have @subsys_id enabled. |
Tejun Heo | 95109b6 | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 406 | */ |
| 407 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 408 | struct cgroup_subsys *ss) |
Tejun Heo | 95109b6 | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 409 | { |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 410 | if (ss) |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 411 | return rcu_dereference_check(cgrp->subsys[ss->id], |
Tejun Heo | ace2bee81 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 412 | lockdep_is_held(&cgroup_mutex)); |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 413 | else |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 414 | return &cgrp->self; |
Tejun Heo | 95109b6 | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 415 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 416 | |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 417 | /** |
| 418 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem |
| 419 | * @cgrp: the cgroup of interest |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 420 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 421 | * |
Chen Hanxiao | d0f702e | 2015-04-23 07:57:33 -0400 | [diff] [blame] | 422 | * Similar to cgroup_css() but returns the effective css, which is defined |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 423 | * as the matching css of the nearest ancestor including self which has @ss |
| 424 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this |
| 425 | * function is guaranteed to return non-NULL css. |
| 426 | */ |
| 427 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, |
| 428 | struct cgroup_subsys *ss) |
| 429 | { |
| 430 | lockdep_assert_held(&cgroup_mutex); |
| 431 | |
| 432 | if (!ss) |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 433 | return &cgrp->self; |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 434 | |
Tejun Heo | eeecbd1 | 2014-11-18 02:49:52 -0500 | [diff] [blame] | 435 | /* |
| 436 | * This function is used while updating css associations and thus |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 437 | * can't test the csses directly. Test ss_mask. |
Tejun Heo | eeecbd1 | 2014-11-18 02:49:52 -0500 | [diff] [blame] | 438 | */ |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 439 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 440 | cgrp = cgroup_parent(cgrp); |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 441 | if (!cgrp) |
| 442 | return NULL; |
| 443 | } |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 444 | |
| 445 | return cgroup_css(cgrp, ss); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 446 | } |
| 447 | |
Tejun Heo | eeecbd1 | 2014-11-18 02:49:52 -0500 | [diff] [blame] | 448 | /** |
| 449 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem |
| 450 | * @cgrp: the cgroup of interest |
| 451 | * @ss: the subsystem of interest |
| 452 | * |
| 453 | * Find and get the effective css of @cgrp for @ss. The effective css is |
| 454 | * defined as the matching css of the nearest ancestor including self which |
| 455 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, |
| 456 | * the root css is returned, so this function always returns a valid css. |
| 457 | * The returned css must be put using css_put(). |
| 458 | */ |
| 459 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, |
| 460 | struct cgroup_subsys *ss) |
| 461 | { |
| 462 | struct cgroup_subsys_state *css; |
| 463 | |
| 464 | rcu_read_lock(); |
| 465 | |
| 466 | do { |
| 467 | css = cgroup_css(cgrp, ss); |
| 468 | |
| 469 | if (css && css_tryget_online(css)) |
| 470 | goto out_unlock; |
| 471 | cgrp = cgroup_parent(cgrp); |
| 472 | } while (cgrp); |
| 473 | |
| 474 | css = init_css_set.subsys[ss->id]; |
| 475 | css_get(css); |
| 476 | out_unlock: |
| 477 | rcu_read_unlock(); |
| 478 | return css; |
| 479 | } |
| 480 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 481 | /* convenient tests for these bits */ |
Tejun Heo | 54766d4 | 2013-06-12 21:04:53 -0700 | [diff] [blame] | 482 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 483 | { |
Tejun Heo | 184faf3 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 484 | return !(cgrp->self.flags & CSS_ONLINE); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 485 | } |
| 486 | |
Tejun Heo | 052c3f3 | 2015-10-15 16:41:50 -0400 | [diff] [blame] | 487 | static void cgroup_get(struct cgroup *cgrp) |
| 488 | { |
| 489 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
| 490 | css_get(&cgrp->self); |
| 491 | } |
| 492 | |
| 493 | static bool cgroup_tryget(struct cgroup *cgrp) |
| 494 | { |
| 495 | return css_tryget(&cgrp->self); |
| 496 | } |
| 497 | |
Tejun Heo | b416864 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 498 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
Tejun Heo | 59f5296 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 499 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 500 | struct cgroup *cgrp = of->kn->parent->priv; |
Tejun Heo | b416864 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 501 | struct cftype *cft = of_cft(of); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 502 | |
| 503 | /* |
| 504 | * This is open and unprotected implementation of cgroup_css(). |
| 505 | * seq_css() is only called from a kernfs file operation which has |
| 506 | * an active reference on the file. Because all the subsystem |
| 507 | * files are drained before a css is disassociated with a cgroup, |
| 508 | * the matching css from the cgroup's subsys table is guaranteed to |
| 509 | * be and stay valid until the enclosing operation is complete. |
| 510 | */ |
| 511 | if (cft->ss) |
| 512 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); |
| 513 | else |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 514 | return &cgrp->self; |
Tejun Heo | 59f5296 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 515 | } |
Tejun Heo | b416864 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 516 | EXPORT_SYMBOL_GPL(of_css); |
Tejun Heo | 59f5296 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 517 | |
Adrian Bunk | e9685a0 | 2008-02-07 00:13:46 -0800 | [diff] [blame] | 518 | static int notify_on_release(const struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 519 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 520 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 521 | } |
| 522 | |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 523 | /** |
Tejun Heo | 1c6727a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 524 | * for_each_css - iterate all css's of a cgroup |
| 525 | * @css: the iteration cursor |
| 526 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end |
| 527 | * @cgrp: the target cgroup to iterate css's of |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 528 | * |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 529 | * Should be called under cgroup_[tree_]mutex. |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 530 | */ |
Tejun Heo | 1c6727a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 531 | #define for_each_css(css, ssid, cgrp) \ |
| 532 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ |
| 533 | if (!((css) = rcu_dereference_check( \ |
| 534 | (cgrp)->subsys[(ssid)], \ |
| 535 | lockdep_is_held(&cgroup_mutex)))) { } \ |
| 536 | else |
| 537 | |
| 538 | /** |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 539 | * for_each_e_css - iterate all effective css's of a cgroup |
| 540 | * @css: the iteration cursor |
| 541 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end |
| 542 | * @cgrp: the target cgroup to iterate css's of |
| 543 | * |
| 544 | * Should be called under cgroup_[tree_]mutex. |
| 545 | */ |
| 546 | #define for_each_e_css(css, ssid, cgrp) \ |
| 547 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ |
| 548 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ |
| 549 | ; \ |
| 550 | else |
| 551 | |
| 552 | /** |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 553 | * for_each_subsys - iterate all enabled cgroup subsystems |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 554 | * @ss: the iteration cursor |
Tejun Heo | 780cd8b | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 555 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 556 | */ |
Tejun Heo | 780cd8b | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 557 | #define for_each_subsys(ss, ssid) \ |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 558 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
| 559 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 560 | |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 561 | /** |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 562 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 563 | * @ss: the iteration cursor |
| 564 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 565 | * @ss_mask: the bitmask |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 566 | * |
| 567 | * The block will only run for cases where the ssid-th bit (1 << ssid) of |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 568 | * @ss_mask is set. |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 569 | */ |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 570 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
| 571 | unsigned long __ss_mask = (ss_mask); \ |
| 572 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ |
Aleksa Sarai | 4a705c5 | 2015-06-09 21:32:07 +1000 | [diff] [blame] | 573 | (ssid) = 0; \ |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 574 | break; \ |
| 575 | } \ |
| 576 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ |
| 577 | (ss) = cgroup_subsys[ssid]; \ |
| 578 | { |
| 579 | |
| 580 | #define while_each_subsys_mask() \ |
| 581 | } \ |
| 582 | } \ |
| 583 | } while (false) |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 584 | |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 585 | /* iterate across the hierarchies */ |
| 586 | #define for_each_root(root) \ |
Tejun Heo | 5549c49 | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 587 | list_for_each_entry((root), &cgroup_roots, root_list) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 588 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 589 | /* iterate over child cgrps, lock should be held throughout iteration */ |
| 590 | #define cgroup_for_each_live_child(child, cgrp) \ |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 591 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 592 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 593 | cgroup_is_dead(child); })) \ |
| 594 | ; \ |
| 595 | else |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 596 | |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 597 | /* walk live descendants in preorder */ |
| 598 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ |
| 599 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ |
| 600 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
| 601 | (dsct) = (d_css)->cgroup; \ |
| 602 | cgroup_is_dead(dsct); })) \ |
| 603 | ; \ |
| 604 | else |
| 605 | |
| 606 | /* walk live descendants in postorder */ |
| 607 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ |
| 608 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ |
| 609 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
| 610 | (dsct) = (d_css)->cgroup; \ |
| 611 | cgroup_is_dead(dsct); })) \ |
| 612 | ; \ |
| 613 | else |
| 614 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 615 | static void cgroup_release_agent(struct work_struct *work); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 616 | static void check_for_release(struct cgroup *cgrp); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 617 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 618 | /* |
| 619 | * A cgroup can be associated with multiple css_sets as different tasks may |
| 620 | * belong to different cgroups on different hierarchies. In the other |
| 621 | * direction, a css_set is naturally associated with multiple cgroups. |
| 622 | * This M:N relationship is represented by the following link structure |
| 623 | * which exists for each association and allows traversing the associations |
| 624 | * from both sides. |
| 625 | */ |
| 626 | struct cgrp_cset_link { |
| 627 | /* the cgroup and css_set this link associates */ |
| 628 | struct cgroup *cgrp; |
| 629 | struct css_set *cset; |
| 630 | |
| 631 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ |
| 632 | struct list_head cset_link; |
| 633 | |
| 634 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ |
| 635 | struct list_head cgrp_link; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 636 | }; |
| 637 | |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 638 | /* |
| 639 | * The default css_set - used by init and its children prior to any |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 640 | * hierarchies being mounted. It contains a pointer to the root state |
| 641 | * for each subsystem. Also used to anchor the list of css_sets. Not |
| 642 | * reference-counted, to improve performance when child cgroups |
| 643 | * haven't been created. |
| 644 | */ |
Tejun Heo | 5024ae2 | 2014-05-07 21:31:17 -0400 | [diff] [blame] | 645 | struct css_set init_css_set = { |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 646 | .refcount = ATOMIC_INIT(1), |
| 647 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), |
| 648 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
| 649 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), |
| 650 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), |
| 651 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 652 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 653 | }; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 654 | |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 655 | static int css_set_count = 1; /* 1 for init_css_set */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 656 | |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 657 | /** |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 658 | * css_set_populated - does a css_set contain any tasks? |
| 659 | * @cset: target css_set |
| 660 | */ |
| 661 | static bool css_set_populated(struct css_set *cset) |
| 662 | { |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 663 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 664 | |
| 665 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); |
| 666 | } |
| 667 | |
| 668 | /** |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 669 | * cgroup_update_populated - updated populated count of a cgroup |
| 670 | * @cgrp: the target cgroup |
| 671 | * @populated: inc or dec populated count |
| 672 | * |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 673 | * One of the css_sets associated with @cgrp is either getting its first |
| 674 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The |
| 675 | * count is propagated towards root so that a given cgroup's populated_cnt |
| 676 | * is zero iff the cgroup and all its descendants don't contain any tasks. |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 677 | * |
| 678 | * @cgrp's interface file "cgroup.populated" is zero if |
| 679 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt |
| 680 | * changes from or to zero, userland is notified that the content of the |
| 681 | * interface file has changed. This can be used to detect when @cgrp and |
| 682 | * its descendants become populated or empty. |
| 683 | */ |
| 684 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) |
| 685 | { |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 686 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 687 | |
| 688 | do { |
| 689 | bool trigger; |
| 690 | |
| 691 | if (populated) |
| 692 | trigger = !cgrp->populated_cnt++; |
| 693 | else |
| 694 | trigger = !--cgrp->populated_cnt; |
| 695 | |
| 696 | if (!trigger) |
| 697 | break; |
| 698 | |
Tejun Heo | ad2ed2b | 2015-10-15 16:41:50 -0400 | [diff] [blame] | 699 | check_for_release(cgrp); |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 700 | cgroup_file_notify(&cgrp->events_file); |
| 701 | |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 702 | cgrp = cgroup_parent(cgrp); |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 703 | } while (cgrp); |
| 704 | } |
| 705 | |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 706 | /** |
| 707 | * css_set_update_populated - update populated state of a css_set |
| 708 | * @cset: target css_set |
| 709 | * @populated: whether @cset is populated or depopulated |
| 710 | * |
| 711 | * @cset is either getting the first task or losing the last. Update the |
| 712 | * ->populated_cnt of all associated cgroups accordingly. |
| 713 | */ |
| 714 | static void css_set_update_populated(struct css_set *cset, bool populated) |
| 715 | { |
| 716 | struct cgrp_cset_link *link; |
| 717 | |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 718 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 719 | |
| 720 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) |
| 721 | cgroup_update_populated(link->cgrp, populated); |
| 722 | } |
| 723 | |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 724 | /** |
| 725 | * css_set_move_task - move a task from one css_set to another |
| 726 | * @task: task being moved |
| 727 | * @from_cset: css_set @task currently belongs to (may be NULL) |
| 728 | * @to_cset: new css_set @task is being moved to (may be NULL) |
| 729 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks |
| 730 | * |
| 731 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any |
| 732 | * css_set, @from_cset can be NULL. If @task is being disassociated |
| 733 | * instead of moved, @to_cset can be NULL. |
| 734 | * |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 735 | * This function automatically handles populated_cnt updates and |
| 736 | * css_task_iter adjustments but the caller is responsible for managing |
| 737 | * @from_cset and @to_cset's reference counts. |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 738 | */ |
| 739 | static void css_set_move_task(struct task_struct *task, |
| 740 | struct css_set *from_cset, struct css_set *to_cset, |
| 741 | bool use_mg_tasks) |
| 742 | { |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 743 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 744 | |
Tejun Heo | 20b454a | 2016-03-03 09:57:57 -0500 | [diff] [blame] | 745 | if (to_cset && !css_set_populated(to_cset)) |
| 746 | css_set_update_populated(to_cset, true); |
| 747 | |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 748 | if (from_cset) { |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 749 | struct css_task_iter *it, *pos; |
| 750 | |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 751 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 752 | |
| 753 | /* |
| 754 | * @task is leaving, advance task iterators which are |
| 755 | * pointing to it so that they can resume at the next |
| 756 | * position. Advancing an iterator might remove it from |
| 757 | * the list, use safe walk. See css_task_iter_advance*() |
| 758 | * for details. |
| 759 | */ |
| 760 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, |
| 761 | iters_node) |
| 762 | if (it->task_pos == &task->cg_list) |
| 763 | css_task_iter_advance(it); |
| 764 | |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 765 | list_del_init(&task->cg_list); |
| 766 | if (!css_set_populated(from_cset)) |
| 767 | css_set_update_populated(from_cset, false); |
| 768 | } else { |
| 769 | WARN_ON_ONCE(!list_empty(&task->cg_list)); |
| 770 | } |
| 771 | |
| 772 | if (to_cset) { |
| 773 | /* |
| 774 | * We are synchronized through cgroup_threadgroup_rwsem |
| 775 | * against PF_EXITING setting such that we can't race |
| 776 | * against cgroup_exit() changing the css_set to |
| 777 | * init_css_set and dropping the old one. |
| 778 | */ |
| 779 | WARN_ON_ONCE(task->flags & PF_EXITING); |
| 780 | |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 781 | rcu_assign_pointer(task->cgroups, to_cset); |
| 782 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : |
| 783 | &to_cset->tasks); |
| 784 | } |
| 785 | } |
| 786 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 787 | /* |
| 788 | * hash table for cgroup groups. This improves the performance to find |
| 789 | * an existing css_set. This hash doesn't (currently) take into |
| 790 | * account cgroups in empty hierarchies. |
| 791 | */ |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 792 | #define CSS_SET_HASH_BITS 7 |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 793 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 794 | |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 795 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 796 | { |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 797 | unsigned long key = 0UL; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 798 | struct cgroup_subsys *ss; |
| 799 | int i; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 800 | |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 801 | for_each_subsys(ss, i) |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 802 | key += (unsigned long)css[i]; |
| 803 | key = (key >> 16) ^ key; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 804 | |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 805 | return key; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 806 | } |
| 807 | |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 808 | static void put_css_set_locked(struct css_set *cset) |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 809 | { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 810 | struct cgrp_cset_link *link, *tmp_link; |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 811 | struct cgroup_subsys *ss; |
| 812 | int ssid; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 813 | |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 814 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 89c5509 | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 815 | |
| 816 | if (!atomic_dec_and_test(&cset->refcount)) |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 817 | return; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 818 | |
Tejun Heo | 53254f9 | 2015-11-23 14:55:41 -0500 | [diff] [blame] | 819 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
| 820 | for_each_subsys(ss, ssid) { |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 821 | list_del(&cset->e_cset_node[ssid]); |
Tejun Heo | 53254f9 | 2015-11-23 14:55:41 -0500 | [diff] [blame] | 822 | css_put(cset->subsys[ssid]); |
| 823 | } |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 824 | hash_del(&cset->hlist); |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 825 | css_set_count--; |
| 826 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 827 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 828 | list_del(&link->cset_link); |
| 829 | list_del(&link->cgrp_link); |
Tejun Heo | 2ceb231 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 830 | if (cgroup_parent(link->cgrp)) |
| 831 | cgroup_put(link->cgrp); |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 832 | kfree(link); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 833 | } |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 834 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 835 | kfree_rcu(cset, rcu_head); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 836 | } |
| 837 | |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 838 | static void put_css_set(struct css_set *cset) |
Tejun Heo | 89c5509 | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 839 | { |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 840 | unsigned long flags; |
| 841 | |
Tejun Heo | 89c5509 | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 842 | /* |
| 843 | * Ensure that the refcount doesn't hit zero while any readers |
| 844 | * can see it. Similar to atomic_dec_and_lock(), but for an |
| 845 | * rwlock |
| 846 | */ |
| 847 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
| 848 | return; |
| 849 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 850 | spin_lock_irqsave(&css_set_lock, flags); |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 851 | put_css_set_locked(cset); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 852 | spin_unlock_irqrestore(&css_set_lock, flags); |
Tejun Heo | 89c5509 | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 853 | } |
| 854 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 855 | /* |
| 856 | * refcounted get/put for css_set objects |
| 857 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 858 | static inline void get_css_set(struct css_set *cset) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 859 | { |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 860 | atomic_inc(&cset->refcount); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 861 | } |
| 862 | |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 863 | /** |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 864 | * compare_css_sets - helper function for find_existing_css_set(). |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 865 | * @cset: candidate css_set being tested |
| 866 | * @old_cset: existing css_set for a task |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 867 | * @new_cgrp: cgroup that's being entered by the task |
| 868 | * @template: desired set of css pointers in css_set (pre-calculated) |
| 869 | * |
Li Zefan | 6f4b7e6 | 2013-07-31 16:18:36 +0800 | [diff] [blame] | 870 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 871 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
| 872 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 873 | static bool compare_css_sets(struct css_set *cset, |
| 874 | struct css_set *old_cset, |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 875 | struct cgroup *new_cgrp, |
| 876 | struct cgroup_subsys_state *template[]) |
| 877 | { |
| 878 | struct list_head *l1, *l2; |
| 879 | |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 880 | /* |
| 881 | * On the default hierarchy, there can be csets which are |
| 882 | * associated with the same set of cgroups but different csses. |
| 883 | * Let's first ensure that csses match. |
| 884 | */ |
| 885 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 886 | return false; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 887 | |
| 888 | /* |
| 889 | * Compare cgroup pointers in order to distinguish between |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 890 | * different cgroups in hierarchies. As different cgroups may |
| 891 | * share the same effective css, this comparison is always |
| 892 | * necessary. |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 893 | */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 894 | l1 = &cset->cgrp_links; |
| 895 | l2 = &old_cset->cgrp_links; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 896 | while (1) { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 897 | struct cgrp_cset_link *link1, *link2; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 898 | struct cgroup *cgrp1, *cgrp2; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 899 | |
| 900 | l1 = l1->next; |
| 901 | l2 = l2->next; |
| 902 | /* See if we reached the end - both lists are equal length. */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 903 | if (l1 == &cset->cgrp_links) { |
| 904 | BUG_ON(l2 != &old_cset->cgrp_links); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 905 | break; |
| 906 | } else { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 907 | BUG_ON(l2 == &old_cset->cgrp_links); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 908 | } |
| 909 | /* Locate the cgroups associated with these links. */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 910 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
| 911 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); |
| 912 | cgrp1 = link1->cgrp; |
| 913 | cgrp2 = link2->cgrp; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 914 | /* Hierarchies should be linked in the same order. */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 915 | BUG_ON(cgrp1->root != cgrp2->root); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 916 | |
| 917 | /* |
| 918 | * If this hierarchy is the hierarchy of the cgroup |
| 919 | * that's changing, then we need to check that this |
| 920 | * css_set points to the new cgroup; if it's any other |
| 921 | * hierarchy, then this css_set should point to the |
| 922 | * same cgroup as the old css_set. |
| 923 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 924 | if (cgrp1->root == new_cgrp->root) { |
| 925 | if (cgrp1 != new_cgrp) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 926 | return false; |
| 927 | } else { |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 928 | if (cgrp1 != cgrp2) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 929 | return false; |
| 930 | } |
| 931 | } |
| 932 | return true; |
| 933 | } |
| 934 | |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 935 | /** |
| 936 | * find_existing_css_set - init css array and find the matching css_set |
| 937 | * @old_cset: the css_set that we're using before the cgroup transition |
| 938 | * @cgrp: the cgroup that we're moving into |
| 939 | * @template: out param for the new set of csses, should be clear on entry |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 940 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 941 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
| 942 | struct cgroup *cgrp, |
| 943 | struct cgroup_subsys_state *template[]) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 944 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 945 | struct cgroup_root *root = cgrp->root; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 946 | struct cgroup_subsys *ss; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 947 | struct css_set *cset; |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 948 | unsigned long key; |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 949 | int i; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 950 | |
Ben Blum | aae8aab | 2010-03-10 15:22:07 -0800 | [diff] [blame] | 951 | /* |
| 952 | * Build the set of subsystem state objects that we want to see in the |
| 953 | * new css_set. while subsystems can change globally, the entries here |
| 954 | * won't change, so no need for locking. |
| 955 | */ |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 956 | for_each_subsys(ss, i) { |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 957 | if (root->subsys_mask & (1UL << i)) { |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 958 | /* |
| 959 | * @ss is in this hierarchy, so we want the |
| 960 | * effective css from @cgrp. |
| 961 | */ |
| 962 | template[i] = cgroup_e_css(cgrp, ss); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 963 | } else { |
Tejun Heo | aec3dfc | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 964 | /* |
| 965 | * @ss is not in this hierarchy, so we don't want |
| 966 | * to change the css. |
| 967 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 968 | template[i] = old_cset->subsys[i]; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 969 | } |
| 970 | } |
| 971 | |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 972 | key = css_set_hash(template); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 973 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
| 974 | if (!compare_css_sets(cset, old_cset, cgrp, template)) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 975 | continue; |
| 976 | |
| 977 | /* This css_set matches what we need */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 978 | return cset; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 979 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 980 | |
| 981 | /* No existing cgroup group matched */ |
| 982 | return NULL; |
| 983 | } |
| 984 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 985 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 986 | { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 987 | struct cgrp_cset_link *link, *tmp_link; |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 988 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 989 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
| 990 | list_del(&link->cset_link); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 991 | kfree(link); |
| 992 | } |
| 993 | } |
| 994 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 995 | /** |
| 996 | * allocate_cgrp_cset_links - allocate cgrp_cset_links |
| 997 | * @count: the number of links to allocate |
| 998 | * @tmp_links: list_head the allocated links are put on |
| 999 | * |
| 1000 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links |
| 1001 | * through ->cset_link. Returns 0 on success or -errno. |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1002 | */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1003 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1004 | { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1005 | struct cgrp_cset_link *link; |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1006 | int i; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1007 | |
| 1008 | INIT_LIST_HEAD(tmp_links); |
| 1009 | |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1010 | for (i = 0; i < count; i++) { |
Tejun Heo | f4f4be2 | 2013-06-12 21:04:51 -0700 | [diff] [blame] | 1011 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1012 | if (!link) { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1013 | free_cgrp_cset_links(tmp_links); |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1014 | return -ENOMEM; |
| 1015 | } |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1016 | list_add(&link->cset_link, tmp_links); |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 1017 | } |
| 1018 | return 0; |
| 1019 | } |
| 1020 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1021 | /** |
| 1022 | * link_css_set - a helper function to link a css_set to a cgroup |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1023 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1024 | * @cset: the css_set to be linked |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1025 | * @cgrp: the destination cgroup |
| 1026 | */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1027 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
| 1028 | struct cgroup *cgrp) |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1029 | { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1030 | struct cgrp_cset_link *link; |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1031 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1032 | BUG_ON(list_empty(tmp_links)); |
Tejun Heo | 6803c00 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 1033 | |
| 1034 | if (cgroup_on_dfl(cgrp)) |
| 1035 | cset->dfl_cgrp = cgrp; |
| 1036 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1037 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
| 1038 | link->cset = cset; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1039 | link->cgrp = cgrp; |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 1040 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1041 | /* |
Tejun Heo | 389b9c1 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 1042 | * Always add links to the tail of the lists so that the lists are |
| 1043 | * in choronological order. |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1044 | */ |
Tejun Heo | 389b9c1 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 1045 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1046 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
Tejun Heo | 2ceb231 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 1047 | |
| 1048 | if (cgroup_parent(cgrp)) |
| 1049 | cgroup_get(cgrp); |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1050 | } |
| 1051 | |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 1052 | /** |
| 1053 | * find_css_set - return a new css_set with one cgroup updated |
| 1054 | * @old_cset: the baseline css_set |
| 1055 | * @cgrp: the cgroup to be updated |
| 1056 | * |
| 1057 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp |
| 1058 | * substituted into the appropriate hierarchy. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1059 | */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1060 | static struct css_set *find_css_set(struct css_set *old_cset, |
| 1061 | struct cgroup *cgrp) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1062 | { |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 1063 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1064 | struct css_set *cset; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1065 | struct list_head tmp_links; |
| 1066 | struct cgrp_cset_link *link; |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1067 | struct cgroup_subsys *ss; |
Li Zefan | 0ac801f | 2013-01-10 11:49:27 +0800 | [diff] [blame] | 1068 | unsigned long key; |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1069 | int ssid; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 1070 | |
Tejun Heo | b326f9d | 2013-06-24 15:21:48 -0700 | [diff] [blame] | 1071 | lockdep_assert_held(&cgroup_mutex); |
| 1072 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1073 | /* First see if we already have a cgroup group that matches |
| 1074 | * the desired set */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1075 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1076 | cset = find_existing_css_set(old_cset, cgrp, template); |
| 1077 | if (cset) |
| 1078 | get_css_set(cset); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1079 | spin_unlock_irq(&css_set_lock); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1080 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1081 | if (cset) |
| 1082 | return cset; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1083 | |
Tejun Heo | f4f4be2 | 2013-06-12 21:04:51 -0700 | [diff] [blame] | 1084 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1085 | if (!cset) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1086 | return NULL; |
| 1087 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1088 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
Tejun Heo | 9871bf9 | 2013-06-24 15:21:47 -0700 | [diff] [blame] | 1089 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1090 | kfree(cset); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1091 | return NULL; |
| 1092 | } |
| 1093 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1094 | atomic_set(&cset->refcount, 1); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1095 | INIT_LIST_HEAD(&cset->cgrp_links); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1096 | INIT_LIST_HEAD(&cset->tasks); |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 1097 | INIT_LIST_HEAD(&cset->mg_tasks); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 1098 | INIT_LIST_HEAD(&cset->mg_preload_node); |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 1099 | INIT_LIST_HEAD(&cset->mg_node); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 1100 | INIT_LIST_HEAD(&cset->task_iters); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1101 | INIT_HLIST_NODE(&cset->hlist); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1102 | |
| 1103 | /* Copy the set of subsystem state objects generated in |
| 1104 | * find_existing_css_set() */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1105 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1106 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1107 | spin_lock_irq(&css_set_lock); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1108 | /* Add reference counts and links from the new css_set. */ |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1109 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1110 | struct cgroup *c = link->cgrp; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1111 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1112 | if (c->root == cgrp->root) |
| 1113 | c = cgrp; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1114 | link_css_set(&tmp_links, cset, c); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1115 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1116 | |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1117 | BUG_ON(!list_empty(&tmp_links)); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1118 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1119 | css_set_count++; |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 1120 | |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1121 | /* Add @cset to the hash table */ |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1122 | key = css_set_hash(cset->subsys); |
| 1123 | hash_add(css_set_table, &cset->hlist, key); |
Li Zefan | 472b1053 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 1124 | |
Tejun Heo | 53254f9 | 2015-11-23 14:55:41 -0500 | [diff] [blame] | 1125 | for_each_subsys(ss, ssid) { |
| 1126 | struct cgroup_subsys_state *css = cset->subsys[ssid]; |
| 1127 | |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1128 | list_add_tail(&cset->e_cset_node[ssid], |
Tejun Heo | 53254f9 | 2015-11-23 14:55:41 -0500 | [diff] [blame] | 1129 | &css->cgroup->e_csets[ssid]); |
| 1130 | css_get(css); |
| 1131 | } |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1132 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1133 | spin_unlock_irq(&css_set_lock); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1134 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1135 | return cset; |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1136 | } |
| 1137 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1138 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1139 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1140 | struct cgroup *root_cgrp = kf_root->kn->priv; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1141 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1142 | return root_cgrp->root; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1143 | } |
| 1144 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1145 | static int cgroup_init_root_id(struct cgroup_root *root) |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1146 | { |
| 1147 | int id; |
| 1148 | |
| 1149 | lockdep_assert_held(&cgroup_mutex); |
| 1150 | |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 1151 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1152 | if (id < 0) |
| 1153 | return id; |
| 1154 | |
| 1155 | root->hierarchy_id = id; |
| 1156 | return 0; |
| 1157 | } |
| 1158 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1159 | static void cgroup_exit_root_id(struct cgroup_root *root) |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1160 | { |
| 1161 | lockdep_assert_held(&cgroup_mutex); |
| 1162 | |
Johannes Weiner | 8c8a550 | 2016-06-17 12:23:59 -0400 | [diff] [blame] | 1163 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1164 | } |
| 1165 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1166 | static void cgroup_free_root(struct cgroup_root *root) |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1167 | { |
| 1168 | if (root) { |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1169 | idr_destroy(&root->cgroup_idr); |
| 1170 | kfree(root); |
| 1171 | } |
| 1172 | } |
| 1173 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1174 | static void cgroup_destroy_root(struct cgroup_root *root) |
Tejun Heo | 59f5296 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1175 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1176 | struct cgroup *cgrp = &root->cgrp; |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1177 | struct cgrp_cset_link *link, *tmp_link; |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1178 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1179 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1180 | |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1181 | BUG_ON(atomic_read(&root->nr_cgrps)); |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 1182 | BUG_ON(!list_empty(&cgrp->self.children)); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1183 | |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1184 | /* Rebind all subsystems back to the default hierarchy */ |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1185 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1186 | |
| 1187 | /* |
| 1188 | * Release all the links from cset_links to this hierarchy's |
| 1189 | * root cgroup |
| 1190 | */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1191 | spin_lock_irq(&css_set_lock); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1192 | |
| 1193 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { |
| 1194 | list_del(&link->cset_link); |
| 1195 | list_del(&link->cgrp_link); |
| 1196 | kfree(link); |
| 1197 | } |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 1198 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1199 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1200 | |
| 1201 | if (!list_empty(&root->root_list)) { |
| 1202 | list_del(&root->root_list); |
| 1203 | cgroup_root_count--; |
| 1204 | } |
| 1205 | |
| 1206 | cgroup_exit_root_id(root); |
| 1207 | |
| 1208 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1209 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1210 | kernfs_destroy_root(root->kf_root); |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1211 | cgroup_free_root(root); |
| 1212 | } |
| 1213 | |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 1214 | /* |
| 1215 | * look up cgroup associated with current task's cgroup namespace on the |
| 1216 | * specified hierarchy |
| 1217 | */ |
| 1218 | static struct cgroup * |
| 1219 | current_cgns_cgroup_from_root(struct cgroup_root *root) |
| 1220 | { |
| 1221 | struct cgroup *res = NULL; |
| 1222 | struct css_set *cset; |
| 1223 | |
| 1224 | lockdep_assert_held(&css_set_lock); |
| 1225 | |
| 1226 | rcu_read_lock(); |
| 1227 | |
| 1228 | cset = current->nsproxy->cgroup_ns->root_cset; |
| 1229 | if (cset == &init_css_set) { |
| 1230 | res = &root->cgrp; |
| 1231 | } else { |
| 1232 | struct cgrp_cset_link *link; |
| 1233 | |
| 1234 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
| 1235 | struct cgroup *c = link->cgrp; |
| 1236 | |
| 1237 | if (c->root == root) { |
| 1238 | res = c; |
| 1239 | break; |
| 1240 | } |
| 1241 | } |
| 1242 | } |
| 1243 | rcu_read_unlock(); |
| 1244 | |
| 1245 | BUG_ON(!res); |
| 1246 | return res; |
| 1247 | } |
| 1248 | |
Tejun Heo | ceb6a08 | 2014-02-25 10:04:02 -0500 | [diff] [blame] | 1249 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
| 1250 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1251 | struct cgroup_root *root) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1252 | { |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1253 | struct cgroup *res = NULL; |
| 1254 | |
Tejun Heo | 96d365e | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 1255 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 1256 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 96d365e | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 1257 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 1258 | if (cset == &init_css_set) { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1259 | res = &root->cgrp; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1260 | } else { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1261 | struct cgrp_cset_link *link; |
| 1262 | |
| 1263 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1264 | struct cgroup *c = link->cgrp; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1265 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1266 | if (c->root == root) { |
| 1267 | res = c; |
| 1268 | break; |
| 1269 | } |
| 1270 | } |
| 1271 | } |
Tejun Heo | 96d365e | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 1272 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 1273 | BUG_ON(!res); |
| 1274 | return res; |
| 1275 | } |
| 1276 | |
| 1277 | /* |
Tejun Heo | ceb6a08 | 2014-02-25 10:04:02 -0500 | [diff] [blame] | 1278 | * Return the cgroup for "task" from the given hierarchy. Must be |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 1279 | * called with cgroup_mutex and css_set_lock held. |
Tejun Heo | ceb6a08 | 2014-02-25 10:04:02 -0500 | [diff] [blame] | 1280 | */ |
| 1281 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1282 | struct cgroup_root *root) |
Tejun Heo | ceb6a08 | 2014-02-25 10:04:02 -0500 | [diff] [blame] | 1283 | { |
| 1284 | /* |
| 1285 | * No need to lock the task - since we hold cgroup_mutex the |
| 1286 | * task can't change groups, so the only thing that can happen |
| 1287 | * is that it exits and its css is set back to init_css_set. |
| 1288 | */ |
| 1289 | return cset_cgroup_from_root(task_css_set(task), root); |
| 1290 | } |
| 1291 | |
| 1292 | /* |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1293 | * A task must hold cgroup_mutex to modify cgroups. |
| 1294 | * |
| 1295 | * Any task can increment and decrement the count field without lock. |
| 1296 | * So in general, code holding cgroup_mutex can't rely on the count |
| 1297 | * field not changing. However, if the count goes to zero, then only |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1298 | * cgroup_attach_task() can increment it again. Because a count of zero |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1299 | * means that no tasks are currently attached, therefore there is no |
| 1300 | * way a task attached to that cgroup can fork (the other way to |
| 1301 | * increment the count). So code holding cgroup_mutex can safely |
| 1302 | * assume that if the count is zero, it will stay zero. Similarly, if |
| 1303 | * a task holds cgroup_mutex on a cgroup with zero count, it |
| 1304 | * knows that the cgroup won't be removed, as cgroup_rmdir() |
| 1305 | * needs that mutex. |
| 1306 | * |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1307 | * A cgroup can only be deleted if both its 'count' of using tasks |
| 1308 | * is zero, and its list of 'children' cgroups is empty. Since all |
| 1309 | * tasks in the system use _some_ cgroup, and since there is always at |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1310 | * least one task in the system (init, pid == 1), therefore, root cgroup |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1311 | * always has either children cgroups and/or using tasks. So we don't |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1312 | * need a special hack to ensure that root cgroup cannot be deleted. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1313 | * |
| 1314 | * P.S. One more locking exception. RCU is used to guard the |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1315 | * update of a tasks cgroup pointer by cgroup_attach_task() |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1316 | */ |
| 1317 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1318 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 1319 | static const struct file_operations proc_cgroupstats_operations; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 1320 | |
Tejun Heo | 8d7e6fb | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1321 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
| 1322 | char *buf) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1323 | { |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 1324 | struct cgroup_subsys *ss = cft->ss; |
| 1325 | |
Tejun Heo | 8d7e6fb | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1326 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
| 1327 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) |
| 1328 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 1329 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
| 1330 | cft->name); |
Tejun Heo | 8d7e6fb | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1331 | else |
| 1332 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); |
| 1333 | return buf; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1334 | } |
| 1335 | |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1336 | /** |
| 1337 | * cgroup_file_mode - deduce file mode of a control file |
| 1338 | * @cft: the control file in question |
| 1339 | * |
Tejun Heo | 7dbdb19 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1340 | * S_IRUGO for read, S_IWUSR for write. |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1341 | */ |
| 1342 | static umode_t cgroup_file_mode(const struct cftype *cft) |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 1343 | { |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1344 | umode_t mode = 0; |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 1345 | |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1346 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
| 1347 | mode |= S_IRUGO; |
| 1348 | |
Tejun Heo | 7dbdb19 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1349 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
| 1350 | if (cft->flags & CFTYPE_WORLD_WRITABLE) |
| 1351 | mode |= S_IWUGO; |
| 1352 | else |
| 1353 | mode |= S_IWUSR; |
| 1354 | } |
Tejun Heo | f2e85d5 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1355 | |
| 1356 | return mode; |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 1357 | } |
| 1358 | |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1359 | /** |
Tejun Heo | 8699b77 | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1360 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
Tejun Heo | 0f060de | 2014-11-18 02:49:50 -0500 | [diff] [blame] | 1361 | * @subtree_control: the new subtree_control mask to consider |
Tejun Heo | 5ced251 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1362 | * @this_ss_mask: available subsystems |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1363 | * |
| 1364 | * On the default hierarchy, a subsystem may request other subsystems to be |
| 1365 | * enabled together through its ->depends_on mask. In such cases, more |
| 1366 | * subsystems than specified in "cgroup.subtree_control" may be enabled. |
| 1367 | * |
Tejun Heo | 0f060de | 2014-11-18 02:49:50 -0500 | [diff] [blame] | 1368 | * This function calculates which subsystems need to be enabled if |
Tejun Heo | 5ced251 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1369 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1370 | */ |
Tejun Heo | 5ced251 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1371 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 1372 | { |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1373 | u16 cur_ss_mask = subtree_control; |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1374 | struct cgroup_subsys *ss; |
| 1375 | int ssid; |
| 1376 | |
| 1377 | lockdep_assert_held(&cgroup_mutex); |
| 1378 | |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 1379 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
| 1380 | |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1381 | while (true) { |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1382 | u16 new_ss_mask = cur_ss_mask; |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1383 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1384 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
Aleksa Sarai | a966a4e | 2015-06-06 10:02:15 +1000 | [diff] [blame] | 1385 | new_ss_mask |= ss->depends_on; |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1386 | } while_each_subsys_mask(); |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1387 | |
| 1388 | /* |
| 1389 | * Mask out subsystems which aren't available. This can |
| 1390 | * happen only if some depended-upon subsystems were bound |
| 1391 | * to non-default hierarchies. |
| 1392 | */ |
Tejun Heo | 5ced251 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1393 | new_ss_mask &= this_ss_mask; |
Tejun Heo | af0ba67 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1394 | |
| 1395 | if (new_ss_mask == cur_ss_mask) |
| 1396 | break; |
| 1397 | cur_ss_mask = new_ss_mask; |
| 1398 | } |
| 1399 | |
Tejun Heo | 0f060de | 2014-11-18 02:49:50 -0500 | [diff] [blame] | 1400 | return cur_ss_mask; |
| 1401 | } |
| 1402 | |
| 1403 | /** |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1404 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods |
| 1405 | * @kn: the kernfs_node being serviced |
| 1406 | * |
| 1407 | * This helper undoes cgroup_kn_lock_live() and should be invoked before |
| 1408 | * the method finishes if locking succeeded. Note that once this function |
| 1409 | * returns the cgroup returned by cgroup_kn_lock_live() may become |
| 1410 | * inaccessible any time. If the caller intends to continue to access the |
| 1411 | * cgroup, it should pin it before invoking this function. |
| 1412 | */ |
| 1413 | static void cgroup_kn_unlock(struct kernfs_node *kn) |
| 1414 | { |
| 1415 | struct cgroup *cgrp; |
| 1416 | |
| 1417 | if (kernfs_type(kn) == KERNFS_DIR) |
| 1418 | cgrp = kn->priv; |
| 1419 | else |
| 1420 | cgrp = kn->parent->priv; |
| 1421 | |
| 1422 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1423 | |
| 1424 | kernfs_unbreak_active_protection(kn); |
| 1425 | cgroup_put(cgrp); |
| 1426 | } |
| 1427 | |
| 1428 | /** |
| 1429 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods |
| 1430 | * @kn: the kernfs_node being serviced |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 1431 | * @drain_offline: perform offline draining on the cgroup |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1432 | * |
| 1433 | * This helper is to be used by a cgroup kernfs method currently servicing |
| 1434 | * @kn. It breaks the active protection, performs cgroup locking and |
| 1435 | * verifies that the associated cgroup is alive. Returns the cgroup if |
| 1436 | * alive; otherwise, %NULL. A successful return should be undone by a |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 1437 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
| 1438 | * cgroup is drained of offlining csses before return. |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1439 | * |
| 1440 | * Any cgroup kernfs method implementation which requires locking the |
| 1441 | * associated cgroup should use this helper. It avoids nesting cgroup |
| 1442 | * locking under kernfs active protection and allows all kernfs operations |
| 1443 | * including self-removal. |
| 1444 | */ |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 1445 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
| 1446 | bool drain_offline) |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1447 | { |
| 1448 | struct cgroup *cgrp; |
| 1449 | |
| 1450 | if (kernfs_type(kn) == KERNFS_DIR) |
| 1451 | cgrp = kn->priv; |
| 1452 | else |
| 1453 | cgrp = kn->parent->priv; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1454 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1455 | /* |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 1456 | * We're gonna grab cgroup_mutex which nests outside kernfs |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1457 | * active_ref. cgroup liveliness check alone provides enough |
| 1458 | * protection against removal. Ensure @cgrp stays accessible and |
| 1459 | * break the active_ref protection. |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1460 | */ |
Li Zefan | aa32362 | 2014-09-04 14:43:38 +0800 | [diff] [blame] | 1461 | if (!cgroup_tryget(cgrp)) |
| 1462 | return NULL; |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1463 | kernfs_break_active_protection(kn); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1464 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 1465 | if (drain_offline) |
| 1466 | cgroup_lock_and_drain_offline(cgrp); |
| 1467 | else |
| 1468 | mutex_lock(&cgroup_mutex); |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 1469 | |
| 1470 | if (!cgroup_is_dead(cgrp)) |
| 1471 | return cgrp; |
| 1472 | |
| 1473 | cgroup_kn_unlock(kn); |
| 1474 | return NULL; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1475 | } |
| 1476 | |
Li Zefan | 2739d3c | 2013-01-21 18:18:33 +0800 | [diff] [blame] | 1477 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1478 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1479 | char name[CGROUP_FILE_NAME_MAX]; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1480 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 1481 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | 34c0625 | 2015-11-05 00:12:24 -0500 | [diff] [blame] | 1482 | |
| 1483 | if (cft->file_offset) { |
| 1484 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); |
| 1485 | struct cgroup_file *cfile = (void *)css + cft->file_offset; |
| 1486 | |
| 1487 | spin_lock_irq(&cgroup_file_kn_lock); |
| 1488 | cfile->kn = NULL; |
| 1489 | spin_unlock_irq(&cgroup_file_kn_lock); |
| 1490 | } |
| 1491 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1492 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
Tejun Heo | 05ef1d7 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 1493 | } |
| 1494 | |
Aristeu Rozanski | 13af07d | 2012-08-23 16:53:29 -0400 | [diff] [blame] | 1495 | /** |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1496 | * css_clear_dir - remove subsys files in a cgroup directory |
| 1497 | * @css: taget css |
Aristeu Rozanski | 13af07d | 2012-08-23 16:53:29 -0400 | [diff] [blame] | 1498 | */ |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1499 | static void css_clear_dir(struct cgroup_subsys_state *css) |
Tejun Heo | 05ef1d7 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 1500 | { |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1501 | struct cgroup *cgrp = css->cgroup; |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1502 | struct cftype *cfts; |
Tejun Heo | 05ef1d7 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 1503 | |
Tejun Heo | 88cb04b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 1504 | if (!(css->flags & CSS_VISIBLE)) |
| 1505 | return; |
| 1506 | |
| 1507 | css->flags &= ~CSS_VISIBLE; |
| 1508 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1509 | list_for_each_entry(cfts, &css->ss->cfts, node) |
| 1510 | cgroup_addrm_files(css, cgrp, cfts, false); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1511 | } |
| 1512 | |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1513 | /** |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1514 | * css_populate_dir - create subsys files in a cgroup directory |
| 1515 | * @css: target css |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1516 | * |
| 1517 | * On failure, no file is added. |
| 1518 | */ |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1519 | static int css_populate_dir(struct cgroup_subsys_state *css) |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1520 | { |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1521 | struct cgroup *cgrp = css->cgroup; |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1522 | struct cftype *cfts, *failed_cfts; |
| 1523 | int ret; |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1524 | |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 1525 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
Tejun Heo | 88cb04b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 1526 | return 0; |
| 1527 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1528 | if (!css->ss) { |
| 1529 | if (cgroup_on_dfl(cgrp)) |
| 1530 | cfts = cgroup_dfl_base_files; |
| 1531 | else |
| 1532 | cfts = cgroup_legacy_base_files; |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1533 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1534 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
| 1535 | } |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1536 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1537 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
| 1538 | ret = cgroup_addrm_files(css, cgrp, cfts, true); |
| 1539 | if (ret < 0) { |
| 1540 | failed_cfts = cfts; |
| 1541 | goto err; |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1542 | } |
| 1543 | } |
Tejun Heo | 88cb04b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 1544 | |
| 1545 | css->flags |= CSS_VISIBLE; |
| 1546 | |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1547 | return 0; |
| 1548 | err: |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1549 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
| 1550 | if (cfts == failed_cfts) |
| 1551 | break; |
| 1552 | cgroup_addrm_files(css, cgrp, cfts, false); |
| 1553 | } |
Tejun Heo | ccdca21 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1554 | return ret; |
| 1555 | } |
| 1556 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1557 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1558 | { |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1559 | struct cgroup *dcgrp = &dst_root->cgrp; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 1560 | struct cgroup_subsys *ss; |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1561 | int ssid, i, ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1562 | |
Tejun Heo | ace2bee81 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 1563 | lockdep_assert_held(&cgroup_mutex); |
Ben Blum | aae8aab | 2010-03-10 15:22:07 -0800 | [diff] [blame] | 1564 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1565 | do_each_subsys_mask(ss, ssid, ss_mask) { |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 1566 | /* |
| 1567 | * If @ss has non-root csses attached to it, can't move. |
| 1568 | * If @ss is an implicit controller, it is exempt from this |
| 1569 | * rule and can be stolen. |
| 1570 | */ |
| 1571 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && |
| 1572 | !ss->implicit_on_dfl) |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1573 | return -EBUSY; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1574 | |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1575 | /* can't move between two non-dummy roots either */ |
Tejun Heo | 7fd8c56 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 1576 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1577 | return -EBUSY; |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1578 | } while_each_subsys_mask(); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1579 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1580 | do_each_subsys_mask(ss, ssid, ss_mask) { |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1581 | struct cgroup_root *src_root = ss->root; |
| 1582 | struct cgroup *scgrp = &src_root->cgrp; |
| 1583 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1584 | struct css_set *cset; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 1585 | |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1586 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
Tejun Heo | 73e80ed | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 1587 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1588 | /* disable from the source */ |
| 1589 | src_root->subsys_mask &= ~(1 << ssid); |
| 1590 | WARN_ON(cgroup_apply_control(scgrp)); |
| 1591 | cgroup_finalize_control(scgrp, 0); |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1592 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1593 | /* rebind */ |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1594 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
| 1595 | rcu_assign_pointer(dcgrp->subsys[ssid], css); |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1596 | ss->root = dst_root; |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1597 | css->cgroup = dcgrp; |
Tejun Heo | a8a648c4 | 2013-06-24 15:21:47 -0700 | [diff] [blame] | 1598 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1599 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1600 | hash_for_each(css_set_table, i, cset, hlist) |
| 1601 | list_move_tail(&cset->e_cset_node[ss->id], |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1602 | &dcgrp->e_csets[ss->id]); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1603 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1604 | |
Tejun Heo | bd53d61 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 1605 | /* default hierarchy doesn't enable controllers by default */ |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 1606 | dst_root->subsys_mask |= 1 << ssid; |
Tejun Heo | 49d1dc4 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 1607 | if (dst_root == &cgrp_dfl_root) { |
| 1608 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); |
| 1609 | } else { |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1610 | dcgrp->subtree_control |= 1 << ssid; |
Tejun Heo | 49d1dc4 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 1611 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 1612 | } |
Tejun Heo | 73e80ed | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 1613 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1614 | ret = cgroup_apply_control(dcgrp); |
| 1615 | if (ret) |
| 1616 | pr_warn("partial failure to rebind %s controller (err=%d)\n", |
| 1617 | ss->name, ret); |
| 1618 | |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1619 | if (ss->bind) |
| 1620 | ss->bind(css); |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 1621 | } while_each_subsys_mask(); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1622 | |
Tejun Heo | 1ada483 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 1623 | kernfs_activate(dcgrp->kn); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1624 | return 0; |
| 1625 | } |
| 1626 | |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 1627 | static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
| 1628 | struct kernfs_root *kf_root) |
| 1629 | { |
Felipe Balbi | 09be4c8 | 2016-05-12 12:34:38 +0300 | [diff] [blame] | 1630 | int len = 0; |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 1631 | char *buf = NULL; |
| 1632 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); |
| 1633 | struct cgroup *ns_cgroup; |
| 1634 | |
| 1635 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
| 1636 | if (!buf) |
| 1637 | return -ENOMEM; |
| 1638 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1639 | spin_lock_irq(&css_set_lock); |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 1640 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
| 1641 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1642 | spin_unlock_irq(&css_set_lock); |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 1643 | |
| 1644 | if (len >= PATH_MAX) |
| 1645 | len = -ERANGE; |
| 1646 | else if (len > 0) { |
| 1647 | seq_escape(sf, buf, " \t\n\\"); |
| 1648 | len = 0; |
| 1649 | } |
| 1650 | kfree(buf); |
| 1651 | return len; |
| 1652 | } |
| 1653 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1654 | static int cgroup_show_options(struct seq_file *seq, |
| 1655 | struct kernfs_root *kf_root) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1656 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1657 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1658 | struct cgroup_subsys *ss; |
Tejun Heo | b85d204 | 2013-12-06 15:11:57 -0500 | [diff] [blame] | 1659 | int ssid; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1660 | |
Tejun Heo | d98817d | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 1661 | if (root != &cgrp_dfl_root) |
| 1662 | for_each_subsys(ss, ssid) |
| 1663 | if (root->subsys_mask & (1 << ssid)) |
Kees Cook | 61e57c0 | 2015-09-08 14:58:22 -0700 | [diff] [blame] | 1664 | seq_show_option(seq, ss->legacy_name, NULL); |
Tejun Heo | 9343862 | 2013-04-14 20:15:25 -0700 | [diff] [blame] | 1665 | if (root->flags & CGRP_ROOT_NOPREFIX) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1666 | seq_puts(seq, ",noprefix"); |
Tejun Heo | 9343862 | 2013-04-14 20:15:25 -0700 | [diff] [blame] | 1667 | if (root->flags & CGRP_ROOT_XATTR) |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 1668 | seq_puts(seq, ",xattr"); |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1669 | |
| 1670 | spin_lock(&release_agent_path_lock); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1671 | if (strlen(root->release_agent_path)) |
Kees Cook | a068acf | 2015-09-04 15:44:57 -0700 | [diff] [blame] | 1672 | seq_show_option(seq, "release_agent", |
| 1673 | root->release_agent_path); |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1674 | spin_unlock(&release_agent_path_lock); |
| 1675 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1676 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 1677 | seq_puts(seq, ",clone_children"); |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1678 | if (strlen(root->name)) |
Kees Cook | a068acf | 2015-09-04 15:44:57 -0700 | [diff] [blame] | 1679 | seq_show_option(seq, "name", root->name); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1680 | return 0; |
| 1681 | } |
| 1682 | |
| 1683 | struct cgroup_sb_opts { |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1684 | u16 subsys_mask; |
Tejun Heo | 69dfa00 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 1685 | unsigned int flags; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1686 | char *release_agent; |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 1687 | bool cpuset_clone_children; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1688 | char *name; |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1689 | /* User explicitly requested empty subsystem */ |
| 1690 | bool none; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1691 | }; |
| 1692 | |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 1693 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1694 | { |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1695 | char *token, *o = data; |
| 1696 | bool all_ss = false, one_ss = false; |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1697 | u16 mask = U16_MAX; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 1698 | struct cgroup_subsys *ss; |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 1699 | int nr_opts = 0; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 1700 | int i; |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 1701 | |
| 1702 | #ifdef CONFIG_CPUSETS |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1703 | mask = ~((u16)1 << cpuset_cgrp_id); |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 1704 | #endif |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1705 | |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1706 | memset(opts, 0, sizeof(*opts)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1707 | |
| 1708 | while ((token = strsep(&o, ",")) != NULL) { |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 1709 | nr_opts++; |
| 1710 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1711 | if (!*token) |
| 1712 | return -EINVAL; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1713 | if (!strcmp(token, "none")) { |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1714 | /* Explicitly have no subsystems */ |
| 1715 | opts->none = true; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1716 | continue; |
| 1717 | } |
| 1718 | if (!strcmp(token, "all")) { |
| 1719 | /* Mutually exclusive option 'all' + subsystem name */ |
| 1720 | if (one_ss) |
| 1721 | return -EINVAL; |
| 1722 | all_ss = true; |
| 1723 | continue; |
| 1724 | } |
| 1725 | if (!strcmp(token, "noprefix")) { |
Tejun Heo | 9343862 | 2013-04-14 20:15:25 -0700 | [diff] [blame] | 1726 | opts->flags |= CGRP_ROOT_NOPREFIX; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1727 | continue; |
| 1728 | } |
| 1729 | if (!strcmp(token, "clone_children")) { |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 1730 | opts->cpuset_clone_children = true; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1731 | continue; |
| 1732 | } |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 1733 | if (!strcmp(token, "xattr")) { |
Tejun Heo | 9343862 | 2013-04-14 20:15:25 -0700 | [diff] [blame] | 1734 | opts->flags |= CGRP_ROOT_XATTR; |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 1735 | continue; |
| 1736 | } |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1737 | if (!strncmp(token, "release_agent=", 14)) { |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1738 | /* Specifying two release agents is forbidden */ |
| 1739 | if (opts->release_agent) |
| 1740 | return -EINVAL; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1741 | opts->release_agent = |
Dan Carpenter | e400c28 | 2010-08-10 18:02:54 -0700 | [diff] [blame] | 1742 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1743 | if (!opts->release_agent) |
| 1744 | return -ENOMEM; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1745 | continue; |
| 1746 | } |
| 1747 | if (!strncmp(token, "name=", 5)) { |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1748 | const char *name = token + 5; |
| 1749 | /* Can't specify an empty name */ |
| 1750 | if (!strlen(name)) |
| 1751 | return -EINVAL; |
| 1752 | /* Must match [\w.-]+ */ |
| 1753 | for (i = 0; i < strlen(name); i++) { |
| 1754 | char c = name[i]; |
| 1755 | if (isalnum(c)) |
| 1756 | continue; |
| 1757 | if ((c == '.') || (c == '-') || (c == '_')) |
| 1758 | continue; |
| 1759 | return -EINVAL; |
| 1760 | } |
| 1761 | /* Specifying two names is forbidden */ |
| 1762 | if (opts->name) |
| 1763 | return -EINVAL; |
| 1764 | opts->name = kstrndup(name, |
Dan Carpenter | e400c28 | 2010-08-10 18:02:54 -0700 | [diff] [blame] | 1765 | MAX_CGROUP_ROOT_NAMELEN - 1, |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1766 | GFP_KERNEL); |
| 1767 | if (!opts->name) |
| 1768 | return -ENOMEM; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1769 | |
| 1770 | continue; |
| 1771 | } |
| 1772 | |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 1773 | for_each_subsys(ss, i) { |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 1774 | if (strcmp(token, ss->legacy_name)) |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1775 | continue; |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 1776 | if (!cgroup_ssid_enabled(i)) |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1777 | continue; |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 1778 | if (cgroup_ssid_no_v1(i)) |
| 1779 | continue; |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1780 | |
| 1781 | /* Mutually exclusive option 'all' + subsystem name */ |
| 1782 | if (all_ss) |
| 1783 | return -EINVAL; |
Tejun Heo | 69dfa00 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 1784 | opts->subsys_mask |= (1 << i); |
Daniel Lezcano | 32a8cf23 | 2010-10-27 15:33:37 -0700 | [diff] [blame] | 1785 | one_ss = true; |
| 1786 | |
| 1787 | break; |
| 1788 | } |
| 1789 | if (i == CGROUP_SUBSYS_COUNT) |
| 1790 | return -ENOENT; |
| 1791 | } |
| 1792 | |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 1793 | /* |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 1794 | * If the 'all' option was specified select all the subsystems, |
| 1795 | * otherwise if 'none', 'name=' and a subsystem name options were |
| 1796 | * not specified, let's default to 'all' |
| 1797 | */ |
| 1798 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
| 1799 | for_each_subsys(ss, i) |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 1800 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 1801 | opts->subsys_mask |= (1 << i); |
| 1802 | |
| 1803 | /* |
| 1804 | * We either have to specify by name or by subsystems. (So all |
| 1805 | * empty hierarchies must have a name). |
| 1806 | */ |
| 1807 | if (!opts->subsys_mask && !opts->name) |
| 1808 | return -EINVAL; |
| 1809 | |
| 1810 | /* |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 1811 | * Option noprefix was introduced just for backward compatibility |
| 1812 | * with the old cpuset, so we allow noprefix only if mounting just |
| 1813 | * the cpuset subsystem. |
| 1814 | */ |
Tejun Heo | 9343862 | 2013-04-14 20:15:25 -0700 | [diff] [blame] | 1815 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 1816 | return -EINVAL; |
| 1817 | |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1818 | /* Can't specify "none" and some subsystems */ |
Aristeu Rozanski | a1a71b4 | 2012-08-23 16:53:31 -0400 | [diff] [blame] | 1819 | if (opts->subsys_mask && opts->none) |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1820 | return -EINVAL; |
| 1821 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1822 | return 0; |
| 1823 | } |
| 1824 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1825 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1826 | { |
| 1827 | int ret = 0; |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1828 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1829 | struct cgroup_sb_opts opts; |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1830 | u16 added_mask, removed_mask; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1831 | |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 1832 | if (root == &cgrp_dfl_root) { |
| 1833 | pr_err("remount is not allowed\n"); |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 1834 | return -EINVAL; |
| 1835 | } |
| 1836 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1837 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1838 | |
| 1839 | /* See what subsystems are wanted */ |
| 1840 | ret = parse_cgroupfs_options(data, &opts); |
| 1841 | if (ret) |
| 1842 | goto out_unlock; |
| 1843 | |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 1844 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
Joe Perches | ed3d261 | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 1845 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
Jianyu Zhan | a2a1f9e | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 1846 | task_tgid_nr(current), current->comm); |
Tejun Heo | 8b5a5a9 | 2012-04-01 12:09:54 -0700 | [diff] [blame] | 1847 | |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 1848 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
| 1849 | removed_mask = root->subsys_mask & ~opts.subsys_mask; |
Aristeu Rozanski | 13af07d | 2012-08-23 16:53:29 -0400 | [diff] [blame] | 1850 | |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 1851 | /* Don't allow flags or name to change at remount */ |
Tejun Heo | 7450e90 | 2014-07-09 10:08:07 -0400 | [diff] [blame] | 1852 | if ((opts.flags ^ root->flags) || |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 1853 | (opts.name && strcmp(opts.name, root->name))) { |
Tejun Heo | 69dfa00 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 1854 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
Tejun Heo | 7450e90 | 2014-07-09 10:08:07 -0400 | [diff] [blame] | 1855 | opts.flags, opts.name ?: "", root->flags, root->name); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1856 | ret = -EINVAL; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1857 | goto out_unlock; |
| 1858 | } |
| 1859 | |
Tejun Heo | f172e67 | 2013-06-28 17:07:30 -0700 | [diff] [blame] | 1860 | /* remounting is not allowed for populated hierarchies */ |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 1861 | if (!list_empty(&root->cgrp.self.children)) { |
Tejun Heo | f172e67 | 2013-06-28 17:07:30 -0700 | [diff] [blame] | 1862 | ret = -EBUSY; |
Li Zefan | 0670e08 | 2009-04-02 16:57:30 -0700 | [diff] [blame] | 1863 | goto out_unlock; |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 1864 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1865 | |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1866 | ret = rebind_subsystems(root, added_mask); |
Tejun Heo | 3126121 | 2013-06-28 17:07:30 -0700 | [diff] [blame] | 1867 | if (ret) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1868 | goto out_unlock; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1869 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 1870 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1871 | |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1872 | if (opts.release_agent) { |
| 1873 | spin_lock(&release_agent_path_lock); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1874 | strcpy(root->release_agent_path, opts.release_agent); |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1875 | spin_unlock(&release_agent_path_lock); |
| 1876 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1877 | out_unlock: |
Jesper Juhl | 66bdc9c | 2009-04-02 16:57:27 -0700 | [diff] [blame] | 1878 | kfree(opts.release_agent); |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1879 | kfree(opts.name); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1880 | mutex_unlock(&cgroup_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1881 | return ret; |
| 1882 | } |
| 1883 | |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1884 | /* |
| 1885 | * To reduce the fork() overhead for systems that are not actually using |
| 1886 | * their cgroups capability, we don't maintain the lists running through |
| 1887 | * each css_set to its tasks until we see the list actually used - in other |
| 1888 | * words after the first mount. |
| 1889 | */ |
| 1890 | static bool use_task_css_set_links __read_mostly; |
| 1891 | |
| 1892 | static void cgroup_enable_task_cg_lists(void) |
| 1893 | { |
| 1894 | struct task_struct *p, *g; |
| 1895 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1896 | spin_lock_irq(&css_set_lock); |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1897 | |
| 1898 | if (use_task_css_set_links) |
| 1899 | goto out_unlock; |
| 1900 | |
| 1901 | use_task_css_set_links = true; |
| 1902 | |
| 1903 | /* |
| 1904 | * We need tasklist_lock because RCU is not safe against |
| 1905 | * while_each_thread(). Besides, a forking task that has passed |
| 1906 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 |
| 1907 | * is not guaranteed to have its child immediately visible in the |
| 1908 | * tasklist if we walk through it with RCU. |
| 1909 | */ |
| 1910 | read_lock(&tasklist_lock); |
| 1911 | do_each_thread(g, p) { |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1912 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
| 1913 | task_css_set(p) != &init_css_set); |
| 1914 | |
| 1915 | /* |
| 1916 | * We should check if the process is exiting, otherwise |
| 1917 | * it will race with cgroup_exit() in that the list |
| 1918 | * entry won't be deleted though the process has exited. |
Tejun Heo | f153ad1 | 2014-02-25 09:56:49 -0500 | [diff] [blame] | 1919 | * Do it while holding siglock so that we don't end up |
| 1920 | * racing against cgroup_exit(). |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1921 | * |
| 1922 | * Interrupts were already disabled while acquiring |
| 1923 | * the css_set_lock, so we do not need to disable it |
| 1924 | * again when acquiring the sighand->siglock here. |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1925 | */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1926 | spin_lock(&p->sighand->siglock); |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 1927 | if (!(p->flags & PF_EXITING)) { |
| 1928 | struct css_set *cset = task_css_set(p); |
| 1929 | |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 1930 | if (!css_set_populated(cset)) |
| 1931 | css_set_update_populated(cset, true); |
Tejun Heo | 389b9c1 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 1932 | list_add_tail(&p->cg_list, &cset->tasks); |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 1933 | get_css_set(cset); |
| 1934 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1935 | spin_unlock(&p->sighand->siglock); |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1936 | } while_each_thread(g, p); |
| 1937 | read_unlock(&tasklist_lock); |
| 1938 | out_unlock: |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 1939 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | afeb0f9 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 1940 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1941 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 1942 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
| 1943 | { |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1944 | struct cgroup_subsys *ss; |
| 1945 | int ssid; |
| 1946 | |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 1947 | INIT_LIST_HEAD(&cgrp->self.sibling); |
| 1948 | INIT_LIST_HEAD(&cgrp->self.children); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 1949 | INIT_LIST_HEAD(&cgrp->cset_links); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 1950 | INIT_LIST_HEAD(&cgrp->pidlists); |
| 1951 | mutex_init(&cgrp->pidlist_mutex); |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 1952 | cgrp->self.cgroup = cgrp; |
Tejun Heo | 184faf3 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 1953 | cgrp->self.flags |= CSS_ONLINE; |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 1954 | |
| 1955 | for_each_subsys(ss, ssid) |
| 1956 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 1957 | |
| 1958 | init_waitqueue_head(&cgrp->offline_waitq); |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 1959 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 1960 | } |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1961 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1962 | static void init_cgroup_root(struct cgroup_root *root, |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 1963 | struct cgroup_sb_opts *opts) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1964 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1965 | struct cgroup *cgrp = &root->cgrp; |
Tejun Heo | b0ca5a8 | 2012-04-01 12:09:54 -0700 | [diff] [blame] | 1966 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1967 | INIT_LIST_HEAD(&root->root_list); |
Tejun Heo | 3c9c825 | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1968 | atomic_set(&root->nr_cgrps, 1); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1969 | cgrp->root = root; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 1970 | init_cgroup_housekeeping(cgrp); |
Li Zefan | 4e96ee8e | 2013-07-31 09:50:50 +0800 | [diff] [blame] | 1971 | idr_init(&root->cgroup_idr); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1972 | |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1973 | root->flags = opts->flags; |
| 1974 | if (opts->release_agent) |
| 1975 | strcpy(root->release_agent_path, opts->release_agent); |
| 1976 | if (opts->name) |
| 1977 | strcpy(root->name, opts->name); |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 1978 | if (opts->cpuset_clone_children) |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1979 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1980 | } |
| 1981 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 1982 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1983 | { |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1984 | LIST_HEAD(tmp_links); |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 1985 | struct cgroup *root_cgrp = &root->cgrp; |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1986 | struct css_set *cset; |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1987 | int i, ret; |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1988 | |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1989 | lockdep_assert_held(&cgroup_mutex); |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 1990 | |
Vladimir Davydov | cf780b7 | 2015-08-03 15:32:26 +0300 | [diff] [blame] | 1991 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1992 | if (ret < 0) |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 1993 | goto out; |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 1994 | root_cgrp->id = ret; |
Tejun Heo | b11cfb5 | 2015-11-20 15:55:52 -0500 | [diff] [blame] | 1995 | root_cgrp->ancestor_ids[0] = ret; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 1996 | |
Tejun Heo | 2aad2a8 | 2014-09-24 13:31:50 -0400 | [diff] [blame] | 1997 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
| 1998 | GFP_KERNEL); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 1999 | if (ret) |
| 2000 | goto out; |
| 2001 | |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2002 | /* |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 2003 | * We're accessing css_set_count without locking css_set_lock here, |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2004 | * but that's OK - it can only be increased by someone holding |
Tejun Heo | 0431359 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 2005 | * cgroup_lock, and that's us. Later rebinding may disable |
| 2006 | * controllers on the default hierarchy and thus create new csets, |
| 2007 | * which can't be more than the existing ones. Allocate 2x. |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2008 | */ |
Tejun Heo | 0431359 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 2009 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2010 | if (ret) |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2011 | goto cancel_ref; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2012 | |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2013 | ret = cgroup_init_root_id(root); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2014 | if (ret) |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2015 | goto cancel_ref; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2016 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2017 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
| 2018 | KERNFS_ROOT_CREATE_DEACTIVATED, |
| 2019 | root_cgrp); |
| 2020 | if (IS_ERR(root->kf_root)) { |
| 2021 | ret = PTR_ERR(root->kf_root); |
| 2022 | goto exit_root_id; |
| 2023 | } |
| 2024 | root_cgrp->kn = root->kf_root->kn; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2025 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 2026 | ret = css_populate_dir(&root_cgrp->self); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2027 | if (ret) |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2028 | goto destroy_root; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2029 | |
Tejun Heo | 5df3603 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2030 | ret = rebind_subsystems(root, ss_mask); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2031 | if (ret) |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2032 | goto destroy_root; |
Al Viro | 0df6a63 | 2010-12-21 13:29:29 -0500 | [diff] [blame] | 2033 | |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2034 | /* |
| 2035 | * There must be no failure case after here, since rebinding takes |
| 2036 | * care of subsystems' refcounts, which are explicitly dropped in |
| 2037 | * the failure exit path. |
| 2038 | */ |
| 2039 | list_add(&root->root_list, &cgroup_roots); |
| 2040 | cgroup_root_count++; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2041 | |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2042 | /* |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2043 | * Link the root cgroup in this hierarchy into all the css_set |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2044 | * objects. |
| 2045 | */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2046 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 2047 | hash_for_each(css_set_table, i, cset, hlist) { |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2048 | link_css_set(&tmp_links, cset, root_cgrp); |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 2049 | if (css_set_populated(cset)) |
| 2050 | cgroup_update_populated(root_cgrp, true); |
| 2051 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2052 | spin_unlock_irq(&css_set_lock); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2053 | |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2054 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
Tejun Heo | 3c9c825 | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2055 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2056 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2057 | kernfs_activate(root_cgrp->kn); |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2058 | ret = 0; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2059 | goto out; |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2060 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2061 | destroy_root: |
| 2062 | kernfs_destroy_root(root->kf_root); |
| 2063 | root->kf_root = NULL; |
| 2064 | exit_root_id: |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2065 | cgroup_exit_root_id(root); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2066 | cancel_ref: |
Tejun Heo | 9a1049d | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 2067 | percpu_ref_exit(&root_cgrp->self.refcnt); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2068 | out: |
Tejun Heo | d427dfe | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2069 | free_cgrp_cset_links(&tmp_links); |
| 2070 | return ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2071 | } |
| 2072 | |
Al Viro | f7e8357 | 2010-07-26 13:23:11 +0400 | [diff] [blame] | 2073 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2074 | int flags, const char *unused_dev_name, |
Al Viro | f7e8357 | 2010-07-26 13:23:11 +0400 | [diff] [blame] | 2075 | void *data) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2076 | { |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2077 | bool is_v2 = fs_type == &cgroup2_fs_type; |
Li Zefan | 3a32bd7 | 2014-06-30 11:50:59 +0800 | [diff] [blame] | 2078 | struct super_block *pinned_sb = NULL; |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2079 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
Li Zefan | 970317a | 2014-06-30 11:49:58 +0800 | [diff] [blame] | 2080 | struct cgroup_subsys *ss; |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2081 | struct cgroup_root *root; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2082 | struct cgroup_sb_opts opts; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2083 | struct dentry *dentry; |
Tejun Heo | 8e30e2b8 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2084 | int ret; |
Li Zefan | 970317a | 2014-06-30 11:49:58 +0800 | [diff] [blame] | 2085 | int i; |
Li Zefan | c6b3d5b | 2014-04-04 17:14:41 +0800 | [diff] [blame] | 2086 | bool new_sb; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 2087 | |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2088 | get_cgroup_ns(ns); |
| 2089 | |
| 2090 | /* Check if the caller has permission to mount. */ |
| 2091 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { |
| 2092 | put_cgroup_ns(ns); |
| 2093 | return ERR_PTR(-EPERM); |
| 2094 | } |
| 2095 | |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 2096 | /* |
Tejun Heo | 56fde9e | 2014-02-13 06:58:38 -0500 | [diff] [blame] | 2097 | * The first time anyone tries to mount a cgroup, enable the list |
| 2098 | * linking each css_set to its tasks and fix up all existing tasks. |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 2099 | */ |
Tejun Heo | 56fde9e | 2014-02-13 06:58:38 -0500 | [diff] [blame] | 2100 | if (!use_task_css_set_links) |
| 2101 | cgroup_enable_task_cg_lists(); |
Li Zefan | e37a06f | 2014-04-17 13:53:08 +0800 | [diff] [blame] | 2102 | |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2103 | if (is_v2) { |
| 2104 | if (data) { |
| 2105 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2106 | put_cgroup_ns(ns); |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2107 | return ERR_PTR(-EINVAL); |
| 2108 | } |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 2109 | cgrp_dfl_visible = true; |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2110 | root = &cgrp_dfl_root; |
| 2111 | cgroup_get(&root->cgrp); |
| 2112 | goto out_mount; |
| 2113 | } |
| 2114 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 2115 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2116 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2117 | /* First find the desired set of subsystems */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2118 | ret = parse_cgroupfs_options(data, &opts); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2119 | if (ret) |
Tejun Heo | 8e30e2b8 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2120 | goto out_unlock; |
Tejun Heo | a015edd | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 2121 | |
Li Zefan | 970317a | 2014-06-30 11:49:58 +0800 | [diff] [blame] | 2122 | /* |
| 2123 | * Destruction of cgroup root is asynchronous, so subsystems may |
| 2124 | * still be dying after the previous unmount. Let's drain the |
| 2125 | * dying subsystems. We just need to ensure that the ones |
| 2126 | * unmounted previously finish dying and don't care about new ones |
| 2127 | * starting. Testing ref liveliness is good enough. |
| 2128 | */ |
| 2129 | for_each_subsys(ss, i) { |
| 2130 | if (!(opts.subsys_mask & (1 << i)) || |
| 2131 | ss->root == &cgrp_dfl_root) |
| 2132 | continue; |
| 2133 | |
| 2134 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { |
| 2135 | mutex_unlock(&cgroup_mutex); |
| 2136 | msleep(10); |
| 2137 | ret = restart_syscall(); |
| 2138 | goto out_free; |
| 2139 | } |
| 2140 | cgroup_put(&ss->root->cgrp); |
| 2141 | } |
| 2142 | |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2143 | for_each_root(root) { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2144 | bool name_match = false; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2145 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2146 | if (root == &cgrp_dfl_root) |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2147 | continue; |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 2148 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2149 | /* |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2150 | * If we asked for a name then it must match. Also, if |
| 2151 | * name matches but sybsys_mask doesn't, we should fail. |
| 2152 | * Remember whether name matched. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2153 | */ |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2154 | if (opts.name) { |
| 2155 | if (strcmp(opts.name, root->name)) |
| 2156 | continue; |
| 2157 | name_match = true; |
| 2158 | } |
Tejun Heo | 3126121 | 2013-06-28 17:07:30 -0700 | [diff] [blame] | 2159 | |
| 2160 | /* |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2161 | * If we asked for subsystems (or explicitly for no |
| 2162 | * subsystems) then they must match. |
Tejun Heo | 3126121 | 2013-06-28 17:07:30 -0700 | [diff] [blame] | 2163 | */ |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2164 | if ((opts.subsys_mask || opts.none) && |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 2165 | (opts.subsys_mask != root->subsys_mask)) { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2166 | if (!name_match) |
| 2167 | continue; |
| 2168 | ret = -EBUSY; |
| 2169 | goto out_unlock; |
| 2170 | } |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 2171 | |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 2172 | if (root->flags ^ opts.flags) |
| 2173 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2174 | |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2175 | /* |
Li Zefan | 3a32bd7 | 2014-06-30 11:50:59 +0800 | [diff] [blame] | 2176 | * We want to reuse @root whose lifetime is governed by its |
| 2177 | * ->cgrp. Let's check whether @root is alive and keep it |
| 2178 | * that way. As cgroup_kill_sb() can happen anytime, we |
| 2179 | * want to block it by pinning the sb so that @root doesn't |
| 2180 | * get killed before mount is complete. |
| 2181 | * |
| 2182 | * With the sb pinned, tryget_live can reliably indicate |
| 2183 | * whether @root can be reused. If it's being killed, |
| 2184 | * drain it. We can use wait_queue for the wait but this |
| 2185 | * path is super cold. Let's just sleep a bit and retry. |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2186 | */ |
Li Zefan | 3a32bd7 | 2014-06-30 11:50:59 +0800 | [diff] [blame] | 2187 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
| 2188 | if (IS_ERR(pinned_sb) || |
| 2189 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2190 | mutex_unlock(&cgroup_mutex); |
Li Zefan | 3a32bd7 | 2014-06-30 11:50:59 +0800 | [diff] [blame] | 2191 | if (!IS_ERR_OR_NULL(pinned_sb)) |
| 2192 | deactivate_super(pinned_sb); |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2193 | msleep(10); |
Tejun Heo | a015edd | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 2194 | ret = restart_syscall(); |
| 2195 | goto out_free; |
Tejun Heo | 776f02f | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2196 | } |
| 2197 | |
| 2198 | ret = 0; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2199 | goto out_unlock; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2200 | } |
| 2201 | |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2202 | /* |
| 2203 | * No such thing, create a new one. name= matching without subsys |
| 2204 | * specification is allowed for already existing hierarchies but we |
| 2205 | * can't create new one without subsys specification. |
| 2206 | */ |
| 2207 | if (!opts.subsys_mask && !opts.none) { |
| 2208 | ret = -EINVAL; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2209 | goto out_unlock; |
| 2210 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2211 | |
Eric W. Biederman | 726a499 | 2016-07-15 06:36:44 -0500 | [diff] [blame] | 2212 | /* Hierarchies may only be created in the initial cgroup namespace. */ |
| 2213 | if (ns != &init_cgroup_ns) { |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2214 | ret = -EPERM; |
| 2215 | goto out_unlock; |
| 2216 | } |
| 2217 | |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2218 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
| 2219 | if (!root) { |
| 2220 | ret = -ENOMEM; |
| 2221 | goto out_unlock; |
| 2222 | } |
| 2223 | |
| 2224 | init_cgroup_root(root, &opts); |
| 2225 | |
Tejun Heo | 3558557 | 2014-02-13 06:58:38 -0500 | [diff] [blame] | 2226 | ret = cgroup_setup_root(root, opts.subsys_mask); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2227 | if (ret) |
| 2228 | cgroup_free_root(root); |
| 2229 | |
Tejun Heo | 8e30e2b8 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2230 | out_unlock: |
Tejun Heo | e25e2cbb | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2231 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | a015edd | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 2232 | out_free: |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 2233 | kfree(opts.release_agent); |
| 2234 | kfree(opts.name); |
Tejun Heo | 8e30e2b8 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2235 | |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2236 | if (ret) { |
| 2237 | put_cgroup_ns(ns); |
Tejun Heo | 8e30e2b8 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2238 | return ERR_PTR(ret); |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2239 | } |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2240 | out_mount: |
Jianyu Zhan | c9482a5 | 2014-04-26 15:40:28 +0800 | [diff] [blame] | 2241 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2242 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
| 2243 | &new_sb); |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2244 | |
| 2245 | /* |
| 2246 | * In non-init cgroup namespace, instead of root cgroup's |
| 2247 | * dentry, we return the dentry corresponding to the |
| 2248 | * cgroupns->root_cgrp. |
| 2249 | */ |
| 2250 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { |
| 2251 | struct dentry *nsdentry; |
| 2252 | struct cgroup *cgrp; |
| 2253 | |
| 2254 | mutex_lock(&cgroup_mutex); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2255 | spin_lock_irq(&css_set_lock); |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2256 | |
| 2257 | cgrp = cset_cgroup_from_root(ns->root_cset, root); |
| 2258 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2259 | spin_unlock_irq(&css_set_lock); |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2260 | mutex_unlock(&cgroup_mutex); |
| 2261 | |
| 2262 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); |
| 2263 | dput(dentry); |
| 2264 | dentry = nsdentry; |
| 2265 | } |
| 2266 | |
Li Zefan | c6b3d5b | 2014-04-04 17:14:41 +0800 | [diff] [blame] | 2267 | if (IS_ERR(dentry) || !new_sb) |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2268 | cgroup_put(&root->cgrp); |
Li Zefan | 3a32bd7 | 2014-06-30 11:50:59 +0800 | [diff] [blame] | 2269 | |
| 2270 | /* |
| 2271 | * If @pinned_sb, we're reusing an existing root and holding an |
| 2272 | * extra ref on its sb. Mount is complete. Put the extra ref. |
| 2273 | */ |
| 2274 | if (pinned_sb) { |
| 2275 | WARN_ON(new_sb); |
| 2276 | deactivate_super(pinned_sb); |
| 2277 | } |
| 2278 | |
Serge Hallyn | ed82571 | 2016-01-29 02:54:09 -0600 | [diff] [blame] | 2279 | put_cgroup_ns(ns); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2280 | return dentry; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2281 | } |
| 2282 | |
SeongJae Park | dd4b0a4 | 2014-01-18 16:56:47 +0900 | [diff] [blame] | 2283 | static void cgroup_kill_sb(struct super_block *sb) |
| 2284 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2285 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2286 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2287 | |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2288 | /* |
| 2289 | * If @root doesn't have any mounts or children, start killing it. |
| 2290 | * This prevents new mounts by disabling percpu_ref_tryget_live(). |
| 2291 | * cgroup_mount() may wait for @root's release. |
Li Zefan | 1f779fb | 2014-06-04 16:48:15 +0800 | [diff] [blame] | 2292 | * |
| 2293 | * And don't kill the default root. |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2294 | */ |
Johannes Weiner | 3c606d3 | 2015-01-22 10:19:43 -0500 | [diff] [blame] | 2295 | if (!list_empty(&root->cgrp.self.children) || |
Li Zefan | 1f779fb | 2014-06-04 16:48:15 +0800 | [diff] [blame] | 2296 | root == &cgrp_dfl_root) |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 2297 | cgroup_put(&root->cgrp); |
| 2298 | else |
| 2299 | percpu_ref_kill(&root->cgrp.self.refcnt); |
| 2300 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 2301 | kernfs_kill_sb(sb); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2302 | } |
| 2303 | |
| 2304 | static struct file_system_type cgroup_fs_type = { |
| 2305 | .name = "cgroup", |
Al Viro | f7e8357 | 2010-07-26 13:23:11 +0400 | [diff] [blame] | 2306 | .mount = cgroup_mount, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2307 | .kill_sb = cgroup_kill_sb, |
Serge Hallyn | 1c53753 | 2016-01-29 02:54:11 -0600 | [diff] [blame] | 2308 | .fs_flags = FS_USERNS_MOUNT, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2309 | }; |
| 2310 | |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2311 | static struct file_system_type cgroup2_fs_type = { |
| 2312 | .name = "cgroup2", |
| 2313 | .mount = cgroup_mount, |
| 2314 | .kill_sb = cgroup_kill_sb, |
Serge Hallyn | 1c53753 | 2016-01-29 02:54:11 -0600 | [diff] [blame] | 2315 | .fs_flags = FS_USERNS_MOUNT, |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 2316 | }; |
| 2317 | |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 2318 | static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
| 2319 | struct cgroup_namespace *ns) |
| 2320 | { |
| 2321 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); |
| 2322 | int ret; |
| 2323 | |
| 2324 | ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
| 2325 | if (ret < 0 || ret >= buflen) |
| 2326 | return NULL; |
| 2327 | return buf; |
| 2328 | } |
| 2329 | |
| 2330 | char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
| 2331 | struct cgroup_namespace *ns) |
| 2332 | { |
| 2333 | char *ret; |
| 2334 | |
| 2335 | mutex_lock(&cgroup_mutex); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2336 | spin_lock_irq(&css_set_lock); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 2337 | |
| 2338 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); |
| 2339 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2340 | spin_unlock_irq(&css_set_lock); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 2341 | mutex_unlock(&cgroup_mutex); |
| 2342 | |
| 2343 | return ret; |
| 2344 | } |
| 2345 | EXPORT_SYMBOL_GPL(cgroup_path_ns); |
| 2346 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2347 | /** |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2348 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2349 | * @task: target task |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2350 | * @buf: the buffer to write the path into |
| 2351 | * @buflen: the length of the buffer |
| 2352 | * |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2353 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
| 2354 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This |
| 2355 | * function grabs cgroup_mutex and shouldn't be used inside locks used by |
| 2356 | * cgroup controller callbacks. |
| 2357 | * |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2358 | * Return value is the same as kernfs_path(). |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2359 | */ |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2360 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2361 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2362 | struct cgroup_root *root; |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2363 | struct cgroup *cgrp; |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2364 | int hierarchy_id = 1; |
| 2365 | char *path = NULL; |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2366 | |
| 2367 | mutex_lock(&cgroup_mutex); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2368 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2369 | |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2370 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
| 2371 | |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2372 | if (root) { |
| 2373 | cgrp = task_cgroup_from_root(task, root); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 2374 | path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2375 | } else { |
| 2376 | /* if no hierarchy exists, everyone is in "/" */ |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2377 | if (strlcpy(buf, "/", buflen) < buflen) |
| 2378 | path = buf; |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2379 | } |
| 2380 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2381 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2382 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 2383 | return path; |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2384 | } |
Tejun Heo | 913ffdb | 2013-07-11 16:34:48 -0700 | [diff] [blame] | 2385 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
Tejun Heo | 857a2be | 2013-04-14 20:50:08 -0700 | [diff] [blame] | 2386 | |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2387 | /* used to track tasks and other necessary states during migration */ |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2388 | struct cgroup_taskset { |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2389 | /* the src and dst cset list running through cset->mg_node */ |
| 2390 | struct list_head src_csets; |
| 2391 | struct list_head dst_csets; |
| 2392 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2393 | /* the subsys currently being processed */ |
| 2394 | int ssid; |
| 2395 | |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2396 | /* |
| 2397 | * Fields for cgroup_taskset_*() iteration. |
| 2398 | * |
| 2399 | * Before migration is committed, the target migration tasks are on |
| 2400 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of |
| 2401 | * the csets on ->dst_csets. ->csets point to either ->src_csets |
| 2402 | * or ->dst_csets depending on whether migration is committed. |
| 2403 | * |
| 2404 | * ->cur_csets and ->cur_task point to the current task position |
| 2405 | * during iteration. |
| 2406 | */ |
| 2407 | struct list_head *csets; |
| 2408 | struct css_set *cur_cset; |
| 2409 | struct task_struct *cur_task; |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2410 | }; |
| 2411 | |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2412 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
| 2413 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ |
| 2414 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ |
| 2415 | .csets = &tset.src_csets, \ |
| 2416 | } |
| 2417 | |
| 2418 | /** |
| 2419 | * cgroup_taskset_add - try to add a migration target task to a taskset |
| 2420 | * @task: target task |
| 2421 | * @tset: target taskset |
| 2422 | * |
| 2423 | * Add @task, which is a migration target, to @tset. This function becomes |
| 2424 | * noop if @task doesn't need to be migrated. @task's css_set should have |
| 2425 | * been added as a migration source and @task->cg_list will be moved from |
| 2426 | * the css_set's tasks list to mg_tasks one. |
| 2427 | */ |
| 2428 | static void cgroup_taskset_add(struct task_struct *task, |
| 2429 | struct cgroup_taskset *tset) |
| 2430 | { |
| 2431 | struct css_set *cset; |
| 2432 | |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 2433 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2434 | |
| 2435 | /* @task either already exited or can't exit until the end */ |
| 2436 | if (task->flags & PF_EXITING) |
| 2437 | return; |
| 2438 | |
| 2439 | /* leave @task alone if post_fork() hasn't linked it yet */ |
| 2440 | if (list_empty(&task->cg_list)) |
| 2441 | return; |
| 2442 | |
| 2443 | cset = task_css_set(task); |
| 2444 | if (!cset->mg_src_cgrp) |
| 2445 | return; |
| 2446 | |
| 2447 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
| 2448 | if (list_empty(&cset->mg_node)) |
| 2449 | list_add_tail(&cset->mg_node, &tset->src_csets); |
| 2450 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
| 2451 | list_move_tail(&cset->mg_dst_cset->mg_node, |
| 2452 | &tset->dst_csets); |
| 2453 | } |
| 2454 | |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2455 | /** |
| 2456 | * cgroup_taskset_first - reset taskset and return the first task |
| 2457 | * @tset: taskset of interest |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2458 | * @dst_cssp: output variable for the destination css |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2459 | * |
| 2460 | * @tset iteration is initialized and the first task is returned. |
| 2461 | */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2462 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
| 2463 | struct cgroup_subsys_state **dst_cssp) |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2464 | { |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2465 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
| 2466 | tset->cur_task = NULL; |
| 2467 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2468 | return cgroup_taskset_next(tset, dst_cssp); |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2469 | } |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2470 | |
| 2471 | /** |
| 2472 | * cgroup_taskset_next - iterate to the next task in taskset |
| 2473 | * @tset: taskset of interest |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2474 | * @dst_cssp: output variable for the destination css |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2475 | * |
| 2476 | * Return the next task in @tset. Iteration must have been initialized |
| 2477 | * with cgroup_taskset_first(). |
| 2478 | */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2479 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
| 2480 | struct cgroup_subsys_state **dst_cssp) |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2481 | { |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2482 | struct css_set *cset = tset->cur_cset; |
| 2483 | struct task_struct *task = tset->cur_task; |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2484 | |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2485 | while (&cset->mg_node != tset->csets) { |
| 2486 | if (!task) |
| 2487 | task = list_first_entry(&cset->mg_tasks, |
| 2488 | struct task_struct, cg_list); |
| 2489 | else |
| 2490 | task = list_next_entry(task, cg_list); |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2491 | |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2492 | if (&task->cg_list != &cset->mg_tasks) { |
| 2493 | tset->cur_cset = cset; |
| 2494 | tset->cur_task = task; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2495 | |
| 2496 | /* |
| 2497 | * This function may be called both before and |
| 2498 | * after cgroup_taskset_migrate(). The two cases |
| 2499 | * can be distinguished by looking at whether @cset |
| 2500 | * has its ->mg_dst_cset set. |
| 2501 | */ |
| 2502 | if (cset->mg_dst_cset) |
| 2503 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; |
| 2504 | else |
| 2505 | *dst_cssp = cset->subsys[tset->ssid]; |
| 2506 | |
Tejun Heo | b3dc094 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 2507 | return task; |
| 2508 | } |
| 2509 | |
| 2510 | cset = list_next_entry(cset, mg_node); |
| 2511 | task = NULL; |
| 2512 | } |
| 2513 | |
| 2514 | return NULL; |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2515 | } |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2516 | |
| 2517 | /** |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2518 | * cgroup_taskset_migrate - migrate a taskset |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2519 | * @tset: taget taskset |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2520 | * @root: cgroup root the migration is taking place on |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2521 | * |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2522 | * Migrate tasks in @tset as setup by migration preparation functions. |
| 2523 | * This function fails iff one of the ->can_attach callbacks fails and |
| 2524 | * guarantees that either all or none of the tasks in @tset are migrated. |
| 2525 | * @tset is consumed regardless of success. |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2526 | */ |
| 2527 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2528 | struct cgroup_root *root) |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2529 | { |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2530 | struct cgroup_subsys *ss; |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2531 | struct task_struct *task, *tmp_task; |
| 2532 | struct css_set *cset, *tmp_cset; |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2533 | int ssid, failed_ssid, ret; |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2534 | |
| 2535 | /* methods shouldn't be called if no task is actually migrating */ |
| 2536 | if (list_empty(&tset->src_csets)) |
| 2537 | return 0; |
| 2538 | |
| 2539 | /* check that we can legitimately attach to the cgroup */ |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2540 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
| 2541 | if (ss->can_attach) { |
| 2542 | tset->ssid = ssid; |
| 2543 | ret = ss->can_attach(tset); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2544 | if (ret) { |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2545 | failed_ssid = ssid; |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2546 | goto out_cancel_attach; |
| 2547 | } |
| 2548 | } |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2549 | } while_each_subsys_mask(); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2550 | |
| 2551 | /* |
| 2552 | * Now that we're guaranteed success, proceed to move all tasks to |
| 2553 | * the new cgroup. There are no failure cases after here, so this |
| 2554 | * is the commit point. |
| 2555 | */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2556 | spin_lock_irq(&css_set_lock); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2557 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 2558 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
| 2559 | struct css_set *from_cset = task_css_set(task); |
| 2560 | struct css_set *to_cset = cset->mg_dst_cset; |
| 2561 | |
| 2562 | get_css_set(to_cset); |
| 2563 | css_set_move_task(task, from_cset, to_cset, true); |
| 2564 | put_css_set_locked(from_cset); |
| 2565 | } |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2566 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2567 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2568 | |
| 2569 | /* |
| 2570 | * Migration is committed, all target tasks are now on dst_csets. |
| 2571 | * Nothing is sensitive to fork() after this point. Notify |
| 2572 | * controllers that migration is complete. |
| 2573 | */ |
| 2574 | tset->csets = &tset->dst_csets; |
| 2575 | |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2576 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
| 2577 | if (ss->attach) { |
| 2578 | tset->ssid = ssid; |
| 2579 | ss->attach(tset); |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2580 | } |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2581 | } while_each_subsys_mask(); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2582 | |
| 2583 | ret = 0; |
| 2584 | goto out_release_tset; |
| 2585 | |
| 2586 | out_cancel_attach: |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2587 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
| 2588 | if (ssid == failed_ssid) |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2589 | break; |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2590 | if (ss->cancel_attach) { |
| 2591 | tset->ssid = ssid; |
| 2592 | ss->cancel_attach(tset); |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 2593 | } |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2594 | } while_each_subsys_mask(); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2595 | out_release_tset: |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2596 | spin_lock_irq(&css_set_lock); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2597 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
| 2598 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { |
| 2599 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
| 2600 | list_del_init(&cset->mg_node); |
| 2601 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2602 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2603 | return ret; |
| 2604 | } |
| 2605 | |
| 2606 | /** |
Tejun Heo | 6c694c8 | 2016-03-08 11:51:25 -0500 | [diff] [blame] | 2607 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination |
| 2608 | * @dst_cgrp: destination cgroup to test |
| 2609 | * |
| 2610 | * On the default hierarchy, except for the root, subtree_control must be |
| 2611 | * zero for migration destination cgroups with tasks so that child cgroups |
| 2612 | * don't compete against tasks. |
| 2613 | */ |
| 2614 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) |
| 2615 | { |
| 2616 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || |
| 2617 | !dst_cgrp->subtree_control; |
| 2618 | } |
| 2619 | |
| 2620 | /** |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2621 | * cgroup_migrate_finish - cleanup after attach |
| 2622 | * @preloaded_csets: list of preloaded css_sets |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2623 | * |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2624 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
| 2625 | * those functions for details. |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2626 | */ |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2627 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2628 | { |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2629 | struct css_set *cset, *tmp_cset; |
| 2630 | |
| 2631 | lockdep_assert_held(&cgroup_mutex); |
| 2632 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2633 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2634 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
| 2635 | cset->mg_src_cgrp = NULL; |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2636 | cset->mg_dst_cgrp = NULL; |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2637 | cset->mg_dst_cset = NULL; |
| 2638 | list_del_init(&cset->mg_preload_node); |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 2639 | put_css_set_locked(cset); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2640 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2641 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2642 | } |
| 2643 | |
| 2644 | /** |
| 2645 | * cgroup_migrate_add_src - add a migration source css_set |
| 2646 | * @src_cset: the source css_set to add |
| 2647 | * @dst_cgrp: the destination cgroup |
| 2648 | * @preloaded_csets: list of preloaded css_sets |
| 2649 | * |
| 2650 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin |
| 2651 | * @src_cset and add it to @preloaded_csets, which should later be cleaned |
| 2652 | * up by cgroup_migrate_finish(). |
| 2653 | * |
Tejun Heo | 1ed1328 | 2015-09-16 12:53:17 -0400 | [diff] [blame] | 2654 | * This function may be called without holding cgroup_threadgroup_rwsem |
| 2655 | * even if the target is a process. Threads may be created and destroyed |
| 2656 | * but as long as cgroup_mutex is not dropped, no new css_set can be put |
| 2657 | * into play and the preloaded css_sets are guaranteed to cover all |
| 2658 | * migrations. |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2659 | */ |
| 2660 | static void cgroup_migrate_add_src(struct css_set *src_cset, |
| 2661 | struct cgroup *dst_cgrp, |
| 2662 | struct list_head *preloaded_csets) |
| 2663 | { |
| 2664 | struct cgroup *src_cgrp; |
| 2665 | |
| 2666 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 2667 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2668 | |
Tejun Heo | 2b021cb | 2016-03-15 20:43:04 -0400 | [diff] [blame] | 2669 | /* |
| 2670 | * If ->dead, @src_set is associated with one or more dead cgroups |
| 2671 | * and doesn't contain any migratable tasks. Ignore it early so |
| 2672 | * that the rest of migration path doesn't get confused by it. |
| 2673 | */ |
| 2674 | if (src_cset->dead) |
| 2675 | return; |
| 2676 | |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2677 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
| 2678 | |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2679 | if (!list_empty(&src_cset->mg_preload_node)) |
| 2680 | return; |
| 2681 | |
| 2682 | WARN_ON(src_cset->mg_src_cgrp); |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2683 | WARN_ON(src_cset->mg_dst_cgrp); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2684 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
| 2685 | WARN_ON(!list_empty(&src_cset->mg_node)); |
| 2686 | |
| 2687 | src_cset->mg_src_cgrp = src_cgrp; |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2688 | src_cset->mg_dst_cgrp = dst_cgrp; |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2689 | get_css_set(src_cset); |
| 2690 | list_add(&src_cset->mg_preload_node, preloaded_csets); |
| 2691 | } |
| 2692 | |
| 2693 | /** |
| 2694 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2695 | * @preloaded_csets: list of preloaded source css_sets |
| 2696 | * |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2697 | * Tasks are about to be moved and all the source css_sets have been |
| 2698 | * preloaded to @preloaded_csets. This function looks up and pins all |
| 2699 | * destination css_sets, links each to its source, and append them to |
| 2700 | * @preloaded_csets. |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2701 | * |
| 2702 | * This function must be called after cgroup_migrate_add_src() has been |
| 2703 | * called on each migration source css_set. After migration is performed |
| 2704 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on |
| 2705 | * @preloaded_csets. |
| 2706 | */ |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2707 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2708 | { |
| 2709 | LIST_HEAD(csets); |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2710 | struct css_set *src_cset, *tmp_cset; |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2711 | |
| 2712 | lockdep_assert_held(&cgroup_mutex); |
| 2713 | |
| 2714 | /* look up the dst cset for each src cset and link it to src */ |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2715 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2716 | struct css_set *dst_cset; |
| 2717 | |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2718 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2719 | if (!dst_cset) |
| 2720 | goto err; |
| 2721 | |
| 2722 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2723 | |
| 2724 | /* |
| 2725 | * If src cset equals dst, it's noop. Drop the src. |
| 2726 | * cgroup_migrate() will skip the cset too. Note that we |
| 2727 | * can't handle src == dst as some nodes are used by both. |
| 2728 | */ |
| 2729 | if (src_cset == dst_cset) { |
| 2730 | src_cset->mg_src_cgrp = NULL; |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2731 | src_cset->mg_dst_cgrp = NULL; |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2732 | list_del_init(&src_cset->mg_preload_node); |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 2733 | put_css_set(src_cset); |
| 2734 | put_css_set(dst_cset); |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2735 | continue; |
| 2736 | } |
| 2737 | |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2738 | src_cset->mg_dst_cset = dst_cset; |
| 2739 | |
| 2740 | if (list_empty(&dst_cset->mg_preload_node)) |
| 2741 | list_add(&dst_cset->mg_preload_node, &csets); |
| 2742 | else |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 2743 | put_css_set(dst_cset); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2744 | } |
| 2745 | |
Tejun Heo | f817de9 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 2746 | list_splice_tail(&csets, preloaded_csets); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2747 | return 0; |
| 2748 | err: |
| 2749 | cgroup_migrate_finish(&csets); |
| 2750 | return -ENOMEM; |
| 2751 | } |
| 2752 | |
| 2753 | /** |
| 2754 | * cgroup_migrate - migrate a process or task to a cgroup |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2755 | * @leader: the leader of the process or the task to migrate |
| 2756 | * @threadgroup: whether @leader points to the whole process or a single task |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2757 | * @root: cgroup root migration is taking place on |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2758 | * |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2759 | * Migrate a process or task denoted by @leader. If migrating a process, |
| 2760 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also |
| 2761 | * responsible for invoking cgroup_migrate_add_src() and |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2762 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
| 2763 | * function and following up with cgroup_migrate_finish(). |
| 2764 | * |
| 2765 | * As long as a controller's ->can_attach() doesn't fail, this function is |
| 2766 | * guaranteed to succeed. This means that, excluding ->can_attach() |
| 2767 | * failure, when migrating multiple targets, the success or failure can be |
| 2768 | * decided for all targets by invoking group_migrate_prepare_dst() before |
| 2769 | * actually starting migrating. |
| 2770 | */ |
Tejun Heo | 9af2ec4 | 2015-09-11 15:00:20 -0400 | [diff] [blame] | 2771 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2772 | struct cgroup_root *root) |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2773 | { |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2774 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
| 2775 | struct task_struct *task; |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2776 | |
| 2777 | /* |
Mandeep Singh Baines | fb5d2b4 | 2012-01-03 21:18:31 -0800 | [diff] [blame] | 2778 | * Prevent freeing of tasks while we take a snapshot. Tasks that are |
| 2779 | * already PF_EXITING could be freed from underneath us unless we |
| 2780 | * take an rcu_read_lock. |
| 2781 | */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2782 | spin_lock_irq(&css_set_lock); |
Mandeep Singh Baines | fb5d2b4 | 2012-01-03 21:18:31 -0800 | [diff] [blame] | 2783 | rcu_read_lock(); |
Tejun Heo | 9db8de3 | 2014-02-13 06:58:43 -0500 | [diff] [blame] | 2784 | task = leader; |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2785 | do { |
Tejun Heo | adaae5d | 2015-09-11 15:00:21 -0400 | [diff] [blame] | 2786 | cgroup_taskset_add(task, &tset); |
Li Zefan | 081aa45 | 2013-03-13 09:17:09 +0800 | [diff] [blame] | 2787 | if (!threadgroup) |
| 2788 | break; |
Tejun Heo | 9db8de3 | 2014-02-13 06:58:43 -0500 | [diff] [blame] | 2789 | } while_each_thread(leader, task); |
Mandeep Singh Baines | fb5d2b4 | 2012-01-03 21:18:31 -0800 | [diff] [blame] | 2790 | rcu_read_unlock(); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2791 | spin_unlock_irq(&css_set_lock); |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2792 | |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2793 | return cgroup_taskset_migrate(&tset, root); |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2794 | } |
| 2795 | |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2796 | /** |
| 2797 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
| 2798 | * @dst_cgrp: the cgroup to attach to |
| 2799 | * @leader: the task or the leader of the threadgroup to be attached |
| 2800 | * @threadgroup: attach the whole threadgroup? |
| 2801 | * |
Tejun Heo | 1ed1328 | 2015-09-16 12:53:17 -0400 | [diff] [blame] | 2802 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2803 | */ |
| 2804 | static int cgroup_attach_task(struct cgroup *dst_cgrp, |
| 2805 | struct task_struct *leader, bool threadgroup) |
| 2806 | { |
| 2807 | LIST_HEAD(preloaded_csets); |
| 2808 | struct task_struct *task; |
| 2809 | int ret; |
| 2810 | |
Tejun Heo | 6c694c8 | 2016-03-08 11:51:25 -0500 | [diff] [blame] | 2811 | if (!cgroup_may_migrate_to(dst_cgrp)) |
| 2812 | return -EBUSY; |
| 2813 | |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2814 | /* look up all src csets */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2815 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2816 | rcu_read_lock(); |
| 2817 | task = leader; |
| 2818 | do { |
| 2819 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, |
| 2820 | &preloaded_csets); |
| 2821 | if (!threadgroup) |
| 2822 | break; |
| 2823 | } while_each_thread(leader, task); |
| 2824 | rcu_read_unlock(); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2825 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2826 | |
| 2827 | /* prepare dst csets and commit */ |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2828 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2829 | if (!ret) |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 2830 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
Tejun Heo | 1958d2d | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2831 | |
| 2832 | cgroup_migrate_finish(&preloaded_csets); |
| 2833 | return ret; |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2834 | } |
| 2835 | |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2836 | static int cgroup_procs_write_permission(struct task_struct *task, |
| 2837 | struct cgroup *dst_cgrp, |
| 2838 | struct kernfs_open_file *of) |
Tejun Heo | dedf22e | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2839 | { |
| 2840 | const struct cred *cred = current_cred(); |
| 2841 | const struct cred *tcred = get_task_cred(task); |
| 2842 | int ret = 0; |
| 2843 | |
| 2844 | /* |
| 2845 | * even if we're attaching all tasks in the thread group, we only |
| 2846 | * need to check permissions on one of them. |
| 2847 | */ |
| 2848 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
| 2849 | !uid_eq(cred->euid, tcred->uid) && |
| 2850 | !uid_eq(cred->euid, tcred->suid)) |
| 2851 | ret = -EACCES; |
| 2852 | |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2853 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
| 2854 | struct super_block *sb = of->file->f_path.dentry->d_sb; |
| 2855 | struct cgroup *cgrp; |
| 2856 | struct inode *inode; |
| 2857 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2858 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2859 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2860 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2861 | |
| 2862 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) |
| 2863 | cgrp = cgroup_parent(cgrp); |
| 2864 | |
| 2865 | ret = -ENOMEM; |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 2866 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2867 | if (inode) { |
| 2868 | ret = inode_permission(inode, MAY_WRITE); |
| 2869 | iput(inode); |
| 2870 | } |
| 2871 | } |
| 2872 | |
Tejun Heo | dedf22e | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2873 | put_cred(tcred); |
| 2874 | return ret; |
| 2875 | } |
| 2876 | |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2877 | /* |
| 2878 | * Find the task_struct of the task to attach by vpid and pass it along to the |
Tejun Heo | cd3d095 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2879 | * function to attach either it or all tasks in its threadgroup. Will lock |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 2880 | * cgroup_mutex and threadgroup. |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2881 | */ |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2882 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
| 2883 | size_t nbytes, loff_t off, bool threadgroup) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2884 | { |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2885 | struct task_struct *tsk; |
Tejun Heo | 5cf1cac | 2016-04-21 19:06:48 -0400 | [diff] [blame] | 2886 | struct cgroup_subsys *ss; |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2887 | struct cgroup *cgrp; |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2888 | pid_t pid; |
Tejun Heo | 5cf1cac | 2016-04-21 19:06:48 -0400 | [diff] [blame] | 2889 | int ssid, ret; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2890 | |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2891 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
| 2892 | return -EINVAL; |
| 2893 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 2894 | cgrp = cgroup_kn_lock_live(of->kn, false); |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2895 | if (!cgrp) |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2896 | return -ENODEV; |
| 2897 | |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 2898 | percpu_down_write(&cgroup_threadgroup_rwsem); |
Mandeep Singh Baines | b78949e | 2012-01-03 21:18:30 -0800 | [diff] [blame] | 2899 | rcu_read_lock(); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2900 | if (pid) { |
Pavel Emelyanov | 73507f3 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 2901 | tsk = find_task_by_vpid(pid); |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2902 | if (!tsk) { |
SeongJae Park | dd4b0a4 | 2014-01-18 16:56:47 +0900 | [diff] [blame] | 2903 | ret = -ESRCH; |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 2904 | goto out_unlock_rcu; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2905 | } |
Tejun Heo | dedf22e | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2906 | } else { |
Mandeep Singh Baines | b78949e | 2012-01-03 21:18:30 -0800 | [diff] [blame] | 2907 | tsk = current; |
Tejun Heo | dedf22e | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2908 | } |
Tejun Heo | cd3d095 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2909 | |
| 2910 | if (threadgroup) |
Mandeep Singh Baines | b78949e | 2012-01-03 21:18:30 -0800 | [diff] [blame] | 2911 | tsk = tsk->group_leader; |
Mike Galbraith | c4c27fb | 2012-04-21 09:13:46 +0200 | [diff] [blame] | 2912 | |
| 2913 | /* |
Tejun Heo | 14a40ff | 2013-03-19 13:45:20 -0700 | [diff] [blame] | 2914 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
Mike Galbraith | c4c27fb | 2012-04-21 09:13:46 +0200 | [diff] [blame] | 2915 | * trapped in a cpuset, or RT worker may be born in a cgroup |
| 2916 | * with no rt_runtime allocated. Just say no. |
| 2917 | */ |
Tejun Heo | 14a40ff | 2013-03-19 13:45:20 -0700 | [diff] [blame] | 2918 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
Mike Galbraith | c4c27fb | 2012-04-21 09:13:46 +0200 | [diff] [blame] | 2919 | ret = -EINVAL; |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 2920 | goto out_unlock_rcu; |
Mike Galbraith | c4c27fb | 2012-04-21 09:13:46 +0200 | [diff] [blame] | 2921 | } |
| 2922 | |
Mandeep Singh Baines | b78949e | 2012-01-03 21:18:30 -0800 | [diff] [blame] | 2923 | get_task_struct(tsk); |
| 2924 | rcu_read_unlock(); |
Tejun Heo | cd3d095 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 2925 | |
Tejun Heo | 187fe84 | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2926 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
Tejun Heo | dedf22e | 2015-06-18 16:54:28 -0400 | [diff] [blame] | 2927 | if (!ret) |
| 2928 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); |
Li Zefan | 081aa45 | 2013-03-13 09:17:09 +0800 | [diff] [blame] | 2929 | |
Tejun Heo | f9f9e7b | 2015-09-16 11:51:12 -0400 | [diff] [blame] | 2930 | put_task_struct(tsk); |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 2931 | goto out_unlock_threadgroup; |
| 2932 | |
| 2933 | out_unlock_rcu: |
| 2934 | rcu_read_unlock(); |
| 2935 | out_unlock_threadgroup: |
| 2936 | percpu_up_write(&cgroup_threadgroup_rwsem); |
Tejun Heo | 5cf1cac | 2016-04-21 19:06:48 -0400 | [diff] [blame] | 2937 | for_each_subsys(ss, ssid) |
| 2938 | if (ss->post_attach) |
| 2939 | ss->post_attach(); |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2940 | cgroup_kn_unlock(of->kn); |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2941 | return ret ?: nbytes; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2942 | } |
| 2943 | |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2944 | /** |
| 2945 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' |
| 2946 | * @from: attach to all cgroups of a given task |
| 2947 | * @tsk: the task to be attached |
| 2948 | */ |
| 2949 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) |
| 2950 | { |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2951 | struct cgroup_root *root; |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2952 | int retval = 0; |
| 2953 | |
Tejun Heo | 47cfcd0 | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2954 | mutex_lock(&cgroup_mutex); |
Eric W. Biederman | eedd0f4 | 2016-07-15 06:35:51 -0500 | [diff] [blame] | 2955 | percpu_down_write(&cgroup_threadgroup_rwsem); |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2956 | for_each_root(root) { |
Tejun Heo | 96d365e | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 2957 | struct cgroup *from_cgrp; |
| 2958 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 2959 | if (root == &cgrp_dfl_root) |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 2960 | continue; |
| 2961 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2962 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 96d365e | 2014-02-13 06:58:40 -0500 | [diff] [blame] | 2963 | from_cgrp = task_cgroup_from_root(from, root); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 2964 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2965 | |
Li Zefan | 6f4b7e6 | 2013-07-31 16:18:36 +0800 | [diff] [blame] | 2966 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2967 | if (retval) |
| 2968 | break; |
| 2969 | } |
Eric W. Biederman | eedd0f4 | 2016-07-15 06:35:51 -0500 | [diff] [blame] | 2970 | percpu_up_write(&cgroup_threadgroup_rwsem); |
Tejun Heo | 47cfcd0 | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2971 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | 7ae1bad | 2013-04-07 09:29:51 -0700 | [diff] [blame] | 2972 | |
| 2973 | return retval; |
| 2974 | } |
| 2975 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); |
| 2976 | |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2977 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
| 2978 | char *buf, size_t nbytes, loff_t off) |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 2979 | { |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2980 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2981 | } |
| 2982 | |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2983 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
| 2984 | char *buf, size_t nbytes, loff_t off) |
Ben Blum | 74a1166 | 2011-05-26 16:25:20 -0700 | [diff] [blame] | 2985 | { |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 2986 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 2987 | } |
| 2988 | |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2989 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
| 2990 | char *buf, size_t nbytes, loff_t off) |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 2991 | { |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2992 | struct cgroup *cgrp; |
Tejun Heo | 5f46990 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 2993 | |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2994 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
| 2995 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 2996 | cgrp = cgroup_kn_lock_live(of->kn, false); |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 2997 | if (!cgrp) |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 2998 | return -ENODEV; |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 2999 | spin_lock(&release_agent_path_lock); |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3000 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
| 3001 | sizeof(cgrp->root->release_agent_path)); |
Tejun Heo | 69e943b | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 3002 | spin_unlock(&release_agent_path_lock); |
Tejun Heo | e76ecae | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3003 | cgroup_kn_unlock(of->kn); |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3004 | return nbytes; |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3005 | } |
| 3006 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3007 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3008 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3009 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3010 | |
Tejun Heo | 46cfeb0 | 2014-05-13 12:11:00 -0400 | [diff] [blame] | 3011 | spin_lock(&release_agent_path_lock); |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3012 | seq_puts(seq, cgrp->root->release_agent_path); |
Tejun Heo | 46cfeb0 | 2014-05-13 12:11:00 -0400 | [diff] [blame] | 3013 | spin_unlock(&release_agent_path_lock); |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3014 | seq_putc(seq, '\n'); |
Paul Menage | e788e066 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3015 | return 0; |
| 3016 | } |
| 3017 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3018 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 3019 | { |
Tejun Heo | c1d5d42 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 3020 | seq_puts(seq, "0\n"); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3021 | return 0; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3022 | } |
| 3023 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 3024 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3025 | { |
| 3026 | struct cgroup_subsys *ss; |
| 3027 | bool printed = false; |
| 3028 | int ssid; |
| 3029 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 3030 | do_each_subsys_mask(ss, ssid, ss_mask) { |
Aleksa Sarai | a966a4e | 2015-06-06 10:02:15 +1000 | [diff] [blame] | 3031 | if (printed) |
| 3032 | seq_putc(seq, ' '); |
| 3033 | seq_printf(seq, "%s", ss->name); |
| 3034 | printed = true; |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 3035 | } while_each_subsys_mask(); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3036 | if (printed) |
| 3037 | seq_putc(seq, '\n'); |
| 3038 | } |
| 3039 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3040 | /* show controllers which are enabled from the parent */ |
| 3041 | static int cgroup_controllers_show(struct seq_file *seq, void *v) |
| 3042 | { |
| 3043 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
| 3044 | |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 3045 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3046 | return 0; |
| 3047 | } |
| 3048 | |
| 3049 | /* show controllers which are enabled for a given cgroup's children */ |
| 3050 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) |
| 3051 | { |
| 3052 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
| 3053 | |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3054 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3055 | return 0; |
| 3056 | } |
| 3057 | |
| 3058 | /** |
| 3059 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy |
| 3060 | * @cgrp: root of the subtree to update csses for |
| 3061 | * |
Tejun Heo | 5496260 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3062 | * @cgrp's control masks have changed and its subtree's css associations |
| 3063 | * need to be updated accordingly. This function looks up all css_sets |
| 3064 | * which are attached to the subtree, creates the matching updated css_sets |
| 3065 | * and migrates the tasks to the new ones. |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3066 | */ |
| 3067 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) |
| 3068 | { |
| 3069 | LIST_HEAD(preloaded_csets); |
Tejun Heo | 1026507 | 2015-09-11 15:00:22 -0400 | [diff] [blame] | 3070 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
Tejun Heo | 5496260 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3071 | struct cgroup_subsys_state *d_css; |
| 3072 | struct cgroup *dsct; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3073 | struct css_set *src_cset; |
| 3074 | int ret; |
| 3075 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3076 | lockdep_assert_held(&cgroup_mutex); |
| 3077 | |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 3078 | percpu_down_write(&cgroup_threadgroup_rwsem); |
| 3079 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3080 | /* look up all csses currently attached to @cgrp's subtree */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3081 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 5496260 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3082 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3083 | struct cgrp_cset_link *link; |
| 3084 | |
Tejun Heo | 5496260 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3085 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
Tejun Heo | 58cdb1c | 2016-03-08 11:51:25 -0500 | [diff] [blame] | 3086 | cgroup_migrate_add_src(link->cset, dsct, |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3087 | &preloaded_csets); |
| 3088 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3089 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3090 | |
| 3091 | /* NULL dst indicates self on default hierarchy */ |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 3092 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3093 | if (ret) |
| 3094 | goto out_finish; |
| 3095 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3096 | spin_lock_irq(&css_set_lock); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3097 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
Tejun Heo | 1026507 | 2015-09-11 15:00:22 -0400 | [diff] [blame] | 3098 | struct task_struct *task, *ntask; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3099 | |
| 3100 | /* src_csets precede dst_csets, break on the first dst_cset */ |
| 3101 | if (!src_cset->mg_src_cgrp) |
| 3102 | break; |
| 3103 | |
Tejun Heo | 1026507 | 2015-09-11 15:00:22 -0400 | [diff] [blame] | 3104 | /* all tasks in src_csets need to be migrated */ |
| 3105 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) |
| 3106 | cgroup_taskset_add(task, &tset); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3107 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3108 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3109 | |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 3110 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3111 | out_finish: |
| 3112 | cgroup_migrate_finish(&preloaded_csets); |
Tejun Heo | 3014dde | 2015-09-16 13:03:02 -0400 | [diff] [blame] | 3113 | percpu_up_write(&cgroup_threadgroup_rwsem); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3114 | return ret; |
| 3115 | } |
| 3116 | |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3117 | /** |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3118 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3119 | * @cgrp: root of the target subtree |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3120 | * |
| 3121 | * Because css offlining is asynchronous, userland may try to re-enable a |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3122 | * controller while the previous css is still around. This function grabs |
| 3123 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3124 | */ |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3125 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
| 3126 | __acquires(&cgroup_mutex) |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3127 | { |
| 3128 | struct cgroup *dsct; |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3129 | struct cgroup_subsys_state *d_css; |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3130 | struct cgroup_subsys *ss; |
| 3131 | int ssid; |
| 3132 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3133 | restart: |
| 3134 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3135 | |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3136 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3137 | for_each_subsys(ss, ssid) { |
| 3138 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); |
| 3139 | DEFINE_WAIT(wait); |
| 3140 | |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3141 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3142 | continue; |
| 3143 | |
| 3144 | cgroup_get(dsct); |
| 3145 | prepare_to_wait(&dsct->offline_waitq, &wait, |
| 3146 | TASK_UNINTERRUPTIBLE); |
| 3147 | |
| 3148 | mutex_unlock(&cgroup_mutex); |
| 3149 | schedule(); |
| 3150 | finish_wait(&dsct->offline_waitq, &wait); |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3151 | |
| 3152 | cgroup_put(dsct); |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3153 | goto restart; |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3154 | } |
| 3155 | } |
Tejun Heo | 1b9b96a | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3156 | } |
| 3157 | |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3158 | /** |
Tejun Heo | 15a27c3 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3159 | * cgroup_save_control - save control masks of a subtree |
| 3160 | * @cgrp: root of the target subtree |
| 3161 | * |
| 3162 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ |
| 3163 | * prefixed fields for @cgrp's subtree including @cgrp itself. |
| 3164 | */ |
| 3165 | static void cgroup_save_control(struct cgroup *cgrp) |
| 3166 | { |
| 3167 | struct cgroup *dsct; |
| 3168 | struct cgroup_subsys_state *d_css; |
| 3169 | |
| 3170 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
| 3171 | dsct->old_subtree_control = dsct->subtree_control; |
| 3172 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; |
| 3173 | } |
| 3174 | } |
| 3175 | |
| 3176 | /** |
| 3177 | * cgroup_propagate_control - refresh control masks of a subtree |
| 3178 | * @cgrp: root of the target subtree |
| 3179 | * |
| 3180 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches |
| 3181 | * ->subtree_control and propagate controller availability through the |
| 3182 | * subtree so that descendants don't have unavailable controllers enabled. |
| 3183 | */ |
| 3184 | static void cgroup_propagate_control(struct cgroup *cgrp) |
| 3185 | { |
| 3186 | struct cgroup *dsct; |
| 3187 | struct cgroup_subsys_state *d_css; |
| 3188 | |
| 3189 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
| 3190 | dsct->subtree_control &= cgroup_control(dsct); |
Tejun Heo | 5ced251 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3191 | dsct->subtree_ss_mask = |
| 3192 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, |
| 3193 | cgroup_ss_mask(dsct)); |
Tejun Heo | 15a27c3 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3194 | } |
| 3195 | } |
| 3196 | |
| 3197 | /** |
| 3198 | * cgroup_restore_control - restore control masks of a subtree |
| 3199 | * @cgrp: root of the target subtree |
| 3200 | * |
| 3201 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ |
| 3202 | * prefixed fields for @cgrp's subtree including @cgrp itself. |
| 3203 | */ |
| 3204 | static void cgroup_restore_control(struct cgroup *cgrp) |
| 3205 | { |
| 3206 | struct cgroup *dsct; |
| 3207 | struct cgroup_subsys_state *d_css; |
| 3208 | |
| 3209 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
| 3210 | dsct->subtree_control = dsct->old_subtree_control; |
| 3211 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; |
| 3212 | } |
| 3213 | } |
| 3214 | |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 3215 | static bool css_visible(struct cgroup_subsys_state *css) |
| 3216 | { |
| 3217 | struct cgroup_subsys *ss = css->ss; |
| 3218 | struct cgroup *cgrp = css->cgroup; |
| 3219 | |
| 3220 | if (cgroup_control(cgrp) & (1 << ss->id)) |
| 3221 | return true; |
| 3222 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) |
| 3223 | return false; |
| 3224 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; |
| 3225 | } |
| 3226 | |
Tejun Heo | 15a27c3 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3227 | /** |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3228 | * cgroup_apply_control_enable - enable or show csses according to control |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3229 | * @cgrp: root of the target subtree |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3230 | * |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3231 | * Walk @cgrp's subtree and create new csses or make the existing ones |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3232 | * visible. A css is created invisible if it's being implicitly enabled |
| 3233 | * through dependency. An invisible css is made visible when the userland |
| 3234 | * explicitly enables it. |
| 3235 | * |
| 3236 | * Returns 0 on success, -errno on failure. On failure, csses which have |
| 3237 | * been processed already aren't cleaned up. The caller is responsible for |
| 3238 | * cleaning up with cgroup_apply_control_disble(). |
| 3239 | */ |
| 3240 | static int cgroup_apply_control_enable(struct cgroup *cgrp) |
| 3241 | { |
| 3242 | struct cgroup *dsct; |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3243 | struct cgroup_subsys_state *d_css; |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3244 | struct cgroup_subsys *ss; |
| 3245 | int ssid, ret; |
| 3246 | |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3247 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3248 | for_each_subsys(ss, ssid) { |
| 3249 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); |
| 3250 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3251 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
| 3252 | |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3253 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
| 3254 | continue; |
| 3255 | |
| 3256 | if (!css) { |
| 3257 | css = css_create(dsct, ss); |
| 3258 | if (IS_ERR(css)) |
| 3259 | return PTR_ERR(css); |
| 3260 | } |
| 3261 | |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 3262 | if (css_visible(css)) { |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3263 | ret = css_populate_dir(css); |
Tejun Heo | bdb53bd | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3264 | if (ret) |
| 3265 | return ret; |
| 3266 | } |
| 3267 | } |
| 3268 | } |
| 3269 | |
| 3270 | return 0; |
| 3271 | } |
| 3272 | |
| 3273 | /** |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3274 | * cgroup_apply_control_disable - kill or hide csses according to control |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3275 | * @cgrp: root of the target subtree |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3276 | * |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3277 | * Walk @cgrp's subtree and kill and hide csses so that they match |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3278 | * cgroup_ss_mask() and cgroup_visible_mask(). |
| 3279 | * |
| 3280 | * A css is hidden when the userland requests it to be disabled while other |
| 3281 | * subsystems are still depending on it. The css must not actively control |
| 3282 | * resources and be in the vanilla state if it's made visible again later. |
| 3283 | * Controllers which may be depended upon should provide ->css_reset() for |
| 3284 | * this purpose. |
| 3285 | */ |
| 3286 | static void cgroup_apply_control_disable(struct cgroup *cgrp) |
| 3287 | { |
| 3288 | struct cgroup *dsct; |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3289 | struct cgroup_subsys_state *d_css; |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3290 | struct cgroup_subsys *ss; |
| 3291 | int ssid; |
| 3292 | |
Tejun Heo | ce3f1d9 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3293 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3294 | for_each_subsys(ss, ssid) { |
| 3295 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); |
| 3296 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3297 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
| 3298 | |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3299 | if (!css) |
| 3300 | continue; |
| 3301 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3302 | if (css->parent && |
| 3303 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3304 | kill_css(css); |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 3305 | } else if (!css_visible(css)) { |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 3306 | css_clear_dir(css); |
Tejun Heo | 12b3bb6 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3307 | if (ss->css_reset) |
| 3308 | ss->css_reset(css); |
| 3309 | } |
| 3310 | } |
| 3311 | } |
| 3312 | } |
| 3313 | |
Tejun Heo | f7b2814 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3314 | /** |
| 3315 | * cgroup_apply_control - apply control mask updates to the subtree |
| 3316 | * @cgrp: root of the target subtree |
| 3317 | * |
| 3318 | * subsystems can be enabled and disabled in a subtree using the following |
| 3319 | * steps. |
| 3320 | * |
| 3321 | * 1. Call cgroup_save_control() to stash the current state. |
| 3322 | * 2. Update ->subtree_control masks in the subtree as desired. |
| 3323 | * 3. Call cgroup_apply_control() to apply the changes. |
| 3324 | * 4. Optionally perform other related operations. |
| 3325 | * 5. Call cgroup_finalize_control() to finish up. |
| 3326 | * |
| 3327 | * This function implements step 3 and propagates the mask changes |
| 3328 | * throughout @cgrp's subtree, updates csses accordingly and perform |
| 3329 | * process migrations. |
| 3330 | */ |
| 3331 | static int cgroup_apply_control(struct cgroup *cgrp) |
| 3332 | { |
| 3333 | int ret; |
| 3334 | |
| 3335 | cgroup_propagate_control(cgrp); |
| 3336 | |
| 3337 | ret = cgroup_apply_control_enable(cgrp); |
| 3338 | if (ret) |
| 3339 | return ret; |
| 3340 | |
| 3341 | /* |
| 3342 | * At this point, cgroup_e_css() results reflect the new csses |
| 3343 | * making the following cgroup_update_dfl_csses() properly update |
| 3344 | * css associations of all tasks in the subtree. |
| 3345 | */ |
| 3346 | ret = cgroup_update_dfl_csses(cgrp); |
| 3347 | if (ret) |
| 3348 | return ret; |
| 3349 | |
| 3350 | return 0; |
| 3351 | } |
| 3352 | |
| 3353 | /** |
| 3354 | * cgroup_finalize_control - finalize control mask update |
| 3355 | * @cgrp: root of the target subtree |
| 3356 | * @ret: the result of the update |
| 3357 | * |
| 3358 | * Finalize control mask update. See cgroup_apply_control() for more info. |
| 3359 | */ |
| 3360 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) |
| 3361 | { |
| 3362 | if (ret) { |
| 3363 | cgroup_restore_control(cgrp); |
| 3364 | cgroup_propagate_control(cgrp); |
| 3365 | } |
| 3366 | |
| 3367 | cgroup_apply_control_disable(cgrp); |
| 3368 | } |
| 3369 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3370 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3371 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
| 3372 | char *buf, size_t nbytes, |
| 3373 | loff_t off) |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3374 | { |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 3375 | u16 enable = 0, disable = 0; |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 3376 | struct cgroup *cgrp, *child; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3377 | struct cgroup_subsys *ss; |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3378 | char *tok; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3379 | int ssid, ret; |
| 3380 | |
| 3381 | /* |
Tejun Heo | d37167a | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 3382 | * Parse input - space separated list of subsystem names prefixed |
| 3383 | * with either + or -. |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3384 | */ |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3385 | buf = strstrip(buf); |
| 3386 | while ((tok = strsep(&buf, " "))) { |
Tejun Heo | d37167a | 2014-05-13 12:10:59 -0400 | [diff] [blame] | 3387 | if (tok[0] == '\0') |
| 3388 | continue; |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 3389 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 3390 | if (!cgroup_ssid_enabled(ssid) || |
| 3391 | strcmp(tok + 1, ss->name)) |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3392 | continue; |
| 3393 | |
| 3394 | if (*tok == '+') { |
Tejun Heo | 7d331fa | 2014-05-13 12:11:00 -0400 | [diff] [blame] | 3395 | enable |= 1 << ssid; |
| 3396 | disable &= ~(1 << ssid); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3397 | } else if (*tok == '-') { |
Tejun Heo | 7d331fa | 2014-05-13 12:11:00 -0400 | [diff] [blame] | 3398 | disable |= 1 << ssid; |
| 3399 | enable &= ~(1 << ssid); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3400 | } else { |
| 3401 | return -EINVAL; |
| 3402 | } |
| 3403 | break; |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 3404 | } while_each_subsys_mask(); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3405 | if (ssid == CGROUP_SUBSYS_COUNT) |
| 3406 | return -EINVAL; |
| 3407 | } |
| 3408 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3409 | cgrp = cgroup_kn_lock_live(of->kn, true); |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 3410 | if (!cgrp) |
| 3411 | return -ENODEV; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3412 | |
| 3413 | for_each_subsys(ss, ssid) { |
| 3414 | if (enable & (1 << ssid)) { |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3415 | if (cgrp->subtree_control & (1 << ssid)) { |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3416 | enable &= ~(1 << ssid); |
| 3417 | continue; |
| 3418 | } |
| 3419 | |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 3420 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
Tejun Heo | c29adf2 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3421 | ret = -ENOENT; |
| 3422 | goto out_unlock; |
| 3423 | } |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3424 | } else if (disable & (1 << ssid)) { |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3425 | if (!(cgrp->subtree_control & (1 << ssid))) { |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3426 | disable &= ~(1 << ssid); |
| 3427 | continue; |
| 3428 | } |
| 3429 | |
| 3430 | /* a child has it enabled? */ |
| 3431 | cgroup_for_each_live_child(child, cgrp) { |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3432 | if (child->subtree_control & (1 << ssid)) { |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3433 | ret = -EBUSY; |
Tejun Heo | ddab2b6 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 3434 | goto out_unlock; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3435 | } |
| 3436 | } |
| 3437 | } |
| 3438 | } |
| 3439 | |
| 3440 | if (!enable && !disable) { |
| 3441 | ret = 0; |
Tejun Heo | ddab2b6 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 3442 | goto out_unlock; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3443 | } |
| 3444 | |
| 3445 | /* |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3446 | * Except for the root, subtree_control must be zero for a cgroup |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3447 | * with tasks so that child cgroups don't compete against tasks. |
| 3448 | */ |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3449 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3450 | ret = -EBUSY; |
| 3451 | goto out_unlock; |
| 3452 | } |
| 3453 | |
Tejun Heo | 15a27c3 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3454 | /* save and update control masks and prepare csses */ |
| 3455 | cgroup_save_control(cgrp); |
Tejun Heo | c29adf2 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 3456 | |
Tejun Heo | 15a27c3 | 2016-03-03 09:57:59 -0500 | [diff] [blame] | 3457 | cgrp->subtree_control |= enable; |
| 3458 | cgrp->subtree_control &= ~disable; |
Tejun Heo | f63070d | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 3459 | |
Tejun Heo | f7b2814 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3460 | ret = cgroup_apply_control(cgrp); |
Tejun Heo | 755bf5e | 2014-11-18 02:49:50 -0500 | [diff] [blame] | 3461 | |
Tejun Heo | f7b2814 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 3462 | cgroup_finalize_control(cgrp, ret); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3463 | |
| 3464 | kernfs_activate(cgrp->kn); |
| 3465 | ret = 0; |
| 3466 | out_unlock: |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 3467 | cgroup_kn_unlock(of->kn); |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3468 | return ret ?: nbytes; |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3469 | } |
| 3470 | |
Tejun Heo | 4a07c22 | 2015-09-18 17:54:22 -0400 | [diff] [blame] | 3471 | static int cgroup_events_show(struct seq_file *seq, void *v) |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 3472 | { |
Tejun Heo | 4a07c22 | 2015-09-18 17:54:22 -0400 | [diff] [blame] | 3473 | seq_printf(seq, "populated %d\n", |
Tejun Heo | 27bd4db | 2015-10-15 16:41:50 -0400 | [diff] [blame] | 3474 | cgroup_is_populated(seq_css(seq)->cgroup)); |
Tejun Heo | 842b597 | 2014-04-25 18:28:02 -0400 | [diff] [blame] | 3475 | return 0; |
| 3476 | } |
| 3477 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3478 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
| 3479 | size_t nbytes, loff_t off) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3480 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3481 | struct cgroup *cgrp = of->kn->parent->priv; |
| 3482 | struct cftype *cft = of->kn->priv; |
| 3483 | struct cgroup_subsys_state *css; |
Tejun Heo | a742c59 | 2013-12-05 12:28:03 -0500 | [diff] [blame] | 3484 | int ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3485 | |
Tejun Heo | b416864 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3486 | if (cft->write) |
| 3487 | return cft->write(of, buf, nbytes, off); |
| 3488 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3489 | /* |
| 3490 | * kernfs guarantees that a file isn't deleted with operations in |
| 3491 | * flight, which means that the matching css is and stays alive and |
| 3492 | * doesn't need to be pinned. The RCU locking is not necessary |
| 3493 | * either. It's just for the convenience of using cgroup_css(). |
| 3494 | */ |
| 3495 | rcu_read_lock(); |
| 3496 | css = cgroup_css(cgrp, cft->ss); |
| 3497 | rcu_read_unlock(); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3498 | |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3499 | if (cft->write_u64) { |
Tejun Heo | a742c59 | 2013-12-05 12:28:03 -0500 | [diff] [blame] | 3500 | unsigned long long v; |
| 3501 | ret = kstrtoull(buf, 0, &v); |
| 3502 | if (!ret) |
| 3503 | ret = cft->write_u64(css, cft, v); |
| 3504 | } else if (cft->write_s64) { |
| 3505 | long long v; |
| 3506 | ret = kstrtoll(buf, 0, &v); |
| 3507 | if (!ret) |
| 3508 | ret = cft->write_s64(css, cft, v); |
Tejun Heo | a742c59 | 2013-12-05 12:28:03 -0500 | [diff] [blame] | 3509 | } else { |
| 3510 | ret = -EINVAL; |
| 3511 | } |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3512 | |
Tejun Heo | a742c59 | 2013-12-05 12:28:03 -0500 | [diff] [blame] | 3513 | return ret ?: nbytes; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3514 | } |
| 3515 | |
Tejun Heo | 6612f05 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3516 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 3517 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3518 | return seq_cft(seq)->seq_start(seq, ppos); |
Tejun Heo | 6612f05 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3519 | } |
| 3520 | |
| 3521 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
| 3522 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3523 | return seq_cft(seq)->seq_next(seq, v, ppos); |
Tejun Heo | 6612f05 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3524 | } |
| 3525 | |
| 3526 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
| 3527 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3528 | seq_cft(seq)->seq_stop(seq, v); |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 3529 | } |
| 3530 | |
| 3531 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
| 3532 | { |
Tejun Heo | 7da1127 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3533 | struct cftype *cft = seq_cft(m); |
| 3534 | struct cgroup_subsys_state *css = seq_css(m); |
Li Zefan | e0798ce | 2013-07-31 17:36:25 +0800 | [diff] [blame] | 3535 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3536 | if (cft->seq_show) |
| 3537 | return cft->seq_show(m, arg); |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 3538 | |
Tejun Heo | 896f519 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3539 | if (cft->read_u64) |
| 3540 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
| 3541 | else if (cft->read_s64) |
| 3542 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); |
| 3543 | else |
| 3544 | return -EINVAL; |
| 3545 | return 0; |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 3546 | } |
| 3547 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3548 | static struct kernfs_ops cgroup_kf_single_ops = { |
| 3549 | .atomic_write_len = PAGE_SIZE, |
| 3550 | .write = cgroup_file_write, |
| 3551 | .seq_show = cgroup_seqfile_show, |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 3552 | }; |
| 3553 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3554 | static struct kernfs_ops cgroup_kf_ops = { |
| 3555 | .atomic_write_len = PAGE_SIZE, |
| 3556 | .write = cgroup_file_write, |
| 3557 | .seq_start = cgroup_seqfile_start, |
| 3558 | .seq_next = cgroup_seqfile_next, |
| 3559 | .seq_stop = cgroup_seqfile_stop, |
| 3560 | .seq_show = cgroup_seqfile_show, |
| 3561 | }; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3562 | |
| 3563 | /* |
| 3564 | * cgroup_rename - Only allow simple rename of directories in place. |
| 3565 | */ |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3566 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
| 3567 | const char *new_name_str) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3568 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3569 | struct cgroup *cgrp = kn->priv; |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3570 | int ret; |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3571 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3572 | if (kernfs_type(kn) != KERNFS_DIR) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3573 | return -ENOTDIR; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3574 | if (kn->parent != new_parent) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3575 | return -EIO; |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3576 | |
Tejun Heo | 6db8e85 | 2013-06-14 11:18:22 -0700 | [diff] [blame] | 3577 | /* |
| 3578 | * This isn't a proper migration and its usefulness is very |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 3579 | * limited. Disallow on the default hierarchy. |
Tejun Heo | 6db8e85 | 2013-06-14 11:18:22 -0700 | [diff] [blame] | 3580 | */ |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 3581 | if (cgroup_on_dfl(cgrp)) |
Tejun Heo | 6db8e85 | 2013-06-14 11:18:22 -0700 | [diff] [blame] | 3582 | return -EPERM; |
| 3583 | |
Tejun Heo | e1b2dc1 | 2014-03-20 11:10:15 -0400 | [diff] [blame] | 3584 | /* |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3585 | * We're gonna grab cgroup_mutex which nests outside kernfs |
Tejun Heo | e1b2dc1 | 2014-03-20 11:10:15 -0400 | [diff] [blame] | 3586 | * active_ref. kernfs_rename() doesn't require active_ref |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3587 | * protection. Break them before grabbing cgroup_mutex. |
Tejun Heo | e1b2dc1 | 2014-03-20 11:10:15 -0400 | [diff] [blame] | 3588 | */ |
| 3589 | kernfs_break_active_protection(new_parent); |
| 3590 | kernfs_break_active_protection(kn); |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3591 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3592 | mutex_lock(&cgroup_mutex); |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3593 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3594 | ret = kernfs_rename(kn, new_parent, new_name_str); |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 3595 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3596 | mutex_unlock(&cgroup_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3597 | |
Tejun Heo | e1b2dc1 | 2014-03-20 11:10:15 -0400 | [diff] [blame] | 3598 | kernfs_unbreak_active_protection(kn); |
| 3599 | kernfs_unbreak_active_protection(new_parent); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3600 | return ret; |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 3601 | } |
| 3602 | |
Tejun Heo | 49957f8 | 2014-04-07 16:44:47 -0400 | [diff] [blame] | 3603 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
| 3604 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) |
| 3605 | { |
| 3606 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, |
| 3607 | .ia_uid = current_fsuid(), |
| 3608 | .ia_gid = current_fsgid(), }; |
| 3609 | |
| 3610 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && |
| 3611 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) |
| 3612 | return 0; |
| 3613 | |
| 3614 | return kernfs_setattr(kn, &iattr); |
| 3615 | } |
| 3616 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3617 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
| 3618 | struct cftype *cft) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3619 | { |
Tejun Heo | 8d7e6fb | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3620 | char name[CGROUP_FILE_NAME_MAX]; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3621 | struct kernfs_node *kn; |
| 3622 | struct lock_class_key *key = NULL; |
Tejun Heo | 49957f8 | 2014-04-07 16:44:47 -0400 | [diff] [blame] | 3623 | int ret; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3624 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3625 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 3626 | key = &cft->lockdep_key; |
| 3627 | #endif |
| 3628 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), |
| 3629 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, |
Tejun Heo | dfeb0750 | 2015-02-13 14:36:31 -0800 | [diff] [blame] | 3630 | NULL, key); |
Tejun Heo | 49957f8 | 2014-04-07 16:44:47 -0400 | [diff] [blame] | 3631 | if (IS_ERR(kn)) |
| 3632 | return PTR_ERR(kn); |
| 3633 | |
| 3634 | ret = cgroup_kn_set_ugid(kn); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3635 | if (ret) { |
Tejun Heo | 49957f8 | 2014-04-07 16:44:47 -0400 | [diff] [blame] | 3636 | kernfs_remove(kn); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3637 | return ret; |
| 3638 | } |
| 3639 | |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3640 | if (cft->file_offset) { |
| 3641 | struct cgroup_file *cfile = (void *)css + cft->file_offset; |
| 3642 | |
Tejun Heo | 34c0625 | 2015-11-05 00:12:24 -0500 | [diff] [blame] | 3643 | spin_lock_irq(&cgroup_file_kn_lock); |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3644 | cfile->kn = kn; |
Tejun Heo | 34c0625 | 2015-11-05 00:12:24 -0500 | [diff] [blame] | 3645 | spin_unlock_irq(&cgroup_file_kn_lock); |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3646 | } |
| 3647 | |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 3648 | return 0; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3649 | } |
| 3650 | |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3651 | /** |
| 3652 | * cgroup_addrm_files - add or remove files to a cgroup directory |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3653 | * @css: the target css |
| 3654 | * @cgrp: the target cgroup (usually css->cgroup) |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3655 | * @cfts: array of cftypes to be added |
| 3656 | * @is_add: whether to add or remove |
| 3657 | * |
| 3658 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. |
Tejun Heo | 6732ed8 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3659 | * For removals, this function never fails. |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3660 | */ |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3661 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
| 3662 | struct cgroup *cgrp, struct cftype cfts[], |
Tejun Heo | 2bb566c | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 3663 | bool is_add) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3664 | { |
Tejun Heo | 6732ed8 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3665 | struct cftype *cft, *cft_end = NULL; |
Tejun Heo | b598dde | 2016-02-22 22:25:45 -0500 | [diff] [blame] | 3666 | int ret = 0; |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3667 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3668 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | db0416b | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3669 | |
Tejun Heo | 6732ed8 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3670 | restart: |
| 3671 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { |
Gao feng | f33fddc | 2012-12-06 14:38:57 +0800 | [diff] [blame] | 3672 | /* does cft->flags tell us to skip this file on @cgrp? */ |
Tejun Heo | 05ebb6e | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3673 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
Tejun Heo | 8cbbf2c | 2014-03-19 10:23:55 -0400 | [diff] [blame] | 3674 | continue; |
Tejun Heo | 05ebb6e | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3675 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 3676 | continue; |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3677 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
Gao feng | f33fddc | 2012-12-06 14:38:57 +0800 | [diff] [blame] | 3678 | continue; |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3679 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
Gao feng | f33fddc | 2012-12-06 14:38:57 +0800 | [diff] [blame] | 3680 | continue; |
| 3681 | |
Li Zefan | 2739d3c | 2013-01-21 18:18:33 +0800 | [diff] [blame] | 3682 | if (is_add) { |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3683 | ret = cgroup_add_file(css, cgrp, cft); |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3684 | if (ret) { |
Joe Perches | ed3d261 | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 3685 | pr_warn("%s: failed to add %s, err=%d\n", |
| 3686 | __func__, cft->name, ret); |
Tejun Heo | 6732ed8 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3687 | cft_end = cft; |
| 3688 | is_add = false; |
| 3689 | goto restart; |
Tejun Heo | b1f28d3 | 2013-06-28 16:24:10 -0700 | [diff] [blame] | 3690 | } |
Li Zefan | 2739d3c | 2013-01-21 18:18:33 +0800 | [diff] [blame] | 3691 | } else { |
| 3692 | cgroup_rm_file(cgrp, cft); |
Tejun Heo | db0416b | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3693 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3694 | } |
Tejun Heo | b598dde | 2016-02-22 22:25:45 -0500 | [diff] [blame] | 3695 | return ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 3696 | } |
| 3697 | |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3698 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3699 | { |
| 3700 | LIST_HEAD(pending); |
Tejun Heo | 2bb566c | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 3701 | struct cgroup_subsys *ss = cfts[0].ss; |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 3702 | struct cgroup *root = &ss->root->cgrp; |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3703 | struct cgroup_subsys_state *css; |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3704 | int ret = 0; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3705 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3706 | lockdep_assert_held(&cgroup_mutex); |
Li Zefan | e8c82d2 | 2013-06-18 18:48:37 +0800 | [diff] [blame] | 3707 | |
Li Zefan | e8c82d2 | 2013-06-18 18:48:37 +0800 | [diff] [blame] | 3708 | /* add/rm files for all cgroups created before */ |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 3709 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3710 | struct cgroup *cgrp = css->cgroup; |
| 3711 | |
Tejun Heo | 88cb04b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 3712 | if (!(css->flags & CSS_VISIBLE)) |
Li Zefan | e8c82d2 | 2013-06-18 18:48:37 +0800 | [diff] [blame] | 3713 | continue; |
| 3714 | |
Tejun Heo | 4df8dc9 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3715 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3716 | if (ret) |
| 3717 | break; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3718 | } |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3719 | |
| 3720 | if (is_add && !ret) |
| 3721 | kernfs_activate(root->kn); |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3722 | return ret; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3723 | } |
| 3724 | |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3725 | static void cgroup_exit_cftypes(struct cftype *cfts) |
| 3726 | { |
| 3727 | struct cftype *cft; |
| 3728 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3729 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
| 3730 | /* free copy for custom atomic_write_len, see init_cftypes() */ |
| 3731 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) |
| 3732 | kfree(cft->kf_ops); |
| 3733 | cft->kf_ops = NULL; |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3734 | cft->ss = NULL; |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3735 | |
| 3736 | /* revert flags set by cgroup core while adding @cfts */ |
Tejun Heo | 05ebb6e | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3737 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3738 | } |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3739 | } |
| 3740 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3741 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3742 | { |
| 3743 | struct cftype *cft; |
| 3744 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3745 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
| 3746 | struct kernfs_ops *kf_ops; |
| 3747 | |
Tejun Heo | 0adb070 | 2014-02-12 09:29:48 -0500 | [diff] [blame] | 3748 | WARN_ON(cft->ss || cft->kf_ops); |
| 3749 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3750 | if (cft->seq_start) |
| 3751 | kf_ops = &cgroup_kf_ops; |
| 3752 | else |
| 3753 | kf_ops = &cgroup_kf_single_ops; |
| 3754 | |
| 3755 | /* |
| 3756 | * Ugh... if @cft wants a custom max_write_len, we need to |
| 3757 | * make a copy of kf_ops to set its atomic_write_len. |
| 3758 | */ |
| 3759 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { |
| 3760 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); |
| 3761 | if (!kf_ops) { |
| 3762 | cgroup_exit_cftypes(cfts); |
| 3763 | return -ENOMEM; |
| 3764 | } |
| 3765 | kf_ops->atomic_write_len = cft->max_write_len; |
| 3766 | } |
| 3767 | |
| 3768 | cft->kf_ops = kf_ops; |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3769 | cft->ss = ss; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3770 | } |
| 3771 | |
| 3772 | return 0; |
Tejun Heo | 2da440a | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 3773 | } |
| 3774 | |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3775 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
| 3776 | { |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3777 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3778 | |
| 3779 | if (!cfts || !cfts[0].ss) |
| 3780 | return -ENOENT; |
| 3781 | |
| 3782 | list_del(&cfts->node); |
| 3783 | cgroup_apply_cftypes(cfts, false); |
| 3784 | cgroup_exit_cftypes(cfts); |
| 3785 | return 0; |
| 3786 | } |
| 3787 | |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3788 | /** |
Tejun Heo | 80b1358 | 2014-02-12 09:29:48 -0500 | [diff] [blame] | 3789 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem |
| 3790 | * @cfts: zero-length name terminated array of cftypes |
| 3791 | * |
| 3792 | * Unregister @cfts. Files described by @cfts are removed from all |
| 3793 | * existing cgroups and all future cgroups won't have them either. This |
| 3794 | * function can be called anytime whether @cfts' subsys is attached or not. |
| 3795 | * |
| 3796 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not |
| 3797 | * registered. |
| 3798 | */ |
| 3799 | int cgroup_rm_cftypes(struct cftype *cfts) |
| 3800 | { |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3801 | int ret; |
Tejun Heo | 80b1358 | 2014-02-12 09:29:48 -0500 | [diff] [blame] | 3802 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3803 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3804 | ret = cgroup_rm_cftypes_locked(cfts); |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3805 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3806 | return ret; |
| 3807 | } |
| 3808 | |
| 3809 | /** |
| 3810 | * cgroup_add_cftypes - add an array of cftypes to a subsystem |
| 3811 | * @ss: target cgroup subsystem |
| 3812 | * @cfts: zero-length name terminated array of cftypes |
| 3813 | * |
| 3814 | * Register @cfts to @ss. Files described by @cfts are created for all |
| 3815 | * existing cgroups to which @ss is attached and all future cgroups will |
| 3816 | * have them too. This function can be called anytime whether @ss is |
| 3817 | * attached or not. |
| 3818 | * |
| 3819 | * Returns 0 on successful registration, -errno on failure. Note that this |
| 3820 | * function currently returns 0 as long as @cfts registration is successful |
| 3821 | * even if some file creation attempts on existing cgroups fail. |
| 3822 | */ |
Tejun Heo | 2cf669a | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 3823 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3824 | { |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3825 | int ret; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3826 | |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 3827 | if (!cgroup_ssid_enabled(ss->id)) |
Li Zefan | c731ae1 | 2014-06-05 17:16:30 +0800 | [diff] [blame] | 3828 | return 0; |
| 3829 | |
Li Zefan | dc5736e | 2014-02-17 10:41:50 +0800 | [diff] [blame] | 3830 | if (!cfts || cfts[0].name[0] == '\0') |
| 3831 | return 0; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3832 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3833 | ret = cgroup_init_cftypes(ss, cfts); |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3834 | if (ret) |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 3835 | return ret; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3836 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3837 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3838 | |
Tejun Heo | 0adb070 | 2014-02-12 09:29:48 -0500 | [diff] [blame] | 3839 | list_add_tail(&cfts->node, &ss->cfts); |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3840 | ret = cgroup_apply_cftypes(cfts, true); |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3841 | if (ret) |
Tejun Heo | 21a2d34 | 2014-02-12 09:29:49 -0500 | [diff] [blame] | 3842 | cgroup_rm_cftypes_locked(cfts); |
| 3843 | |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3844 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | 9ccece8 | 2013-06-28 16:24:11 -0700 | [diff] [blame] | 3845 | return ret; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3846 | } |
Tejun Heo | 7957862 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 3847 | |
| 3848 | /** |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3849 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy |
| 3850 | * @ss: target cgroup subsystem |
| 3851 | * @cfts: zero-length name terminated array of cftypes |
| 3852 | * |
| 3853 | * Similar to cgroup_add_cftypes() but the added files are only used for |
| 3854 | * the default hierarchy. |
| 3855 | */ |
| 3856 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
| 3857 | { |
| 3858 | struct cftype *cft; |
| 3859 | |
| 3860 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
Tejun Heo | 05ebb6e | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3861 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3862 | return cgroup_add_cftypes(ss, cfts); |
| 3863 | } |
| 3864 | |
| 3865 | /** |
| 3866 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies |
| 3867 | * @ss: target cgroup subsystem |
| 3868 | * @cfts: zero-length name terminated array of cftypes |
| 3869 | * |
| 3870 | * Similar to cgroup_add_cftypes() but the added files are only used for |
| 3871 | * the legacy hierarchies. |
| 3872 | */ |
Tejun Heo | 2cf669a | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 3873 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
| 3874 | { |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 3875 | struct cftype *cft; |
| 3876 | |
Tejun Heo | e4b7037 | 2015-10-15 17:00:43 -0400 | [diff] [blame] | 3877 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
| 3878 | cft->flags |= __CFTYPE_NOT_ON_DFL; |
Tejun Heo | 2cf669a | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 3879 | return cgroup_add_cftypes(ss, cfts); |
| 3880 | } |
| 3881 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3882 | /** |
Tejun Heo | 34c0625 | 2015-11-05 00:12:24 -0500 | [diff] [blame] | 3883 | * cgroup_file_notify - generate a file modified event for a cgroup_file |
| 3884 | * @cfile: target cgroup_file |
| 3885 | * |
| 3886 | * @cfile must have been obtained by setting cftype->file_offset. |
| 3887 | */ |
| 3888 | void cgroup_file_notify(struct cgroup_file *cfile) |
| 3889 | { |
| 3890 | unsigned long flags; |
| 3891 | |
| 3892 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); |
| 3893 | if (cfile->kn) |
| 3894 | kernfs_notify(cfile->kn); |
| 3895 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); |
| 3896 | } |
| 3897 | |
| 3898 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3899 | * cgroup_task_count - count the number of tasks in a cgroup. |
| 3900 | * @cgrp: the cgroup in question |
| 3901 | * |
| 3902 | * Return the number of tasks in the cgroup. |
| 3903 | */ |
Tejun Heo | 07bc356 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 3904 | static int cgroup_task_count(const struct cgroup *cgrp) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 3905 | { |
| 3906 | int count = 0; |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 3907 | struct cgrp_cset_link *link; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 3908 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3909 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 3910 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
| 3911 | count += atomic_read(&link->cset->refcount); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 3912 | spin_unlock_irq(&css_set_lock); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 3913 | return count; |
| 3914 | } |
| 3915 | |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 3916 | /** |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3917 | * css_next_child - find the next child of a given css |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3918 | * @pos: the current position (%NULL to initiate traversal) |
| 3919 | * @parent: css whose children to walk |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3920 | * |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3921 | * This function returns the next child of @parent and should be called |
Tejun Heo | 87fb54f | 2013-12-06 15:11:55 -0500 | [diff] [blame] | 3922 | * under either cgroup_mutex or RCU read lock. The only requirement is |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3923 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
| 3924 | * be returned regardless of their states. |
| 3925 | * |
| 3926 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 3927 | * css which finished ->css_online() is guaranteed to be visible in the |
| 3928 | * future iterations and will stay visible until the last reference is put. |
| 3929 | * A css which hasn't finished ->css_online() or already finished |
| 3930 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 3931 | * responsibility to synchronize against on/offlining. |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3932 | */ |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3933 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
| 3934 | struct cgroup_subsys_state *parent) |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3935 | { |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3936 | struct cgroup_subsys_state *next; |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3937 | |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 3938 | cgroup_assert_mutex_or_rcu_locked(); |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3939 | |
| 3940 | /* |
Tejun Heo | de3f034 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 3941 | * @pos could already have been unlinked from the sibling list. |
| 3942 | * Once a cgroup is removed, its ->sibling.next is no longer |
| 3943 | * updated when its next sibling changes. CSS_RELEASED is set when |
| 3944 | * @pos is taken off list, at which time its next pointer is valid, |
| 3945 | * and, as releases are serialized, the one pointed to by the next |
| 3946 | * pointer is guaranteed to not have started release yet. This |
| 3947 | * implies that if we observe !CSS_RELEASED on @pos in this RCU |
| 3948 | * critical section, the one pointed to by its next pointer is |
| 3949 | * guaranteed to not have finished its RCU grace period even if we |
| 3950 | * have dropped rcu_read_lock() inbetween iterations. |
Tejun Heo | 3b287a5 | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3951 | * |
Tejun Heo | de3f034 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 3952 | * If @pos has CSS_RELEASED set, its next pointer can't be |
| 3953 | * dereferenced; however, as each css is given a monotonically |
| 3954 | * increasing unique serial number and always appended to the |
| 3955 | * sibling list, the next one can be found by walking the parent's |
| 3956 | * children until the first css with higher serial number than |
| 3957 | * @pos's. While this path can be slower, it happens iff iteration |
| 3958 | * races against release and the race window is very small. |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3959 | */ |
Tejun Heo | 3b287a5 | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3960 | if (!pos) { |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3961 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
| 3962 | } else if (likely(!(pos->flags & CSS_RELEASED))) { |
| 3963 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); |
Tejun Heo | 3b287a5 | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3964 | } else { |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3965 | list_for_each_entry_rcu(next, &parent->children, sibling) |
Tejun Heo | 3b287a5 | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3966 | if (next->serial_nr > pos->serial_nr) |
| 3967 | break; |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3968 | } |
| 3969 | |
Tejun Heo | 3b281af | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 3970 | /* |
| 3971 | * @next, if not pointing to the head, can be dereferenced and is |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3972 | * the next sibling. |
Tejun Heo | 3b281af | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 3973 | */ |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3974 | if (&next->sibling != &parent->children) |
| 3975 | return next; |
Tejun Heo | 3b281af | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 3976 | return NULL; |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3977 | } |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3978 | |
| 3979 | /** |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3980 | * css_next_descendant_pre - find the next descendant for pre-order walk |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 3981 | * @pos: the current position (%NULL to initiate traversal) |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3982 | * @root: css whose descendants to walk |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 3983 | * |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 3984 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 3985 | * to visit for pre-order traversal of @root's descendants. @root is |
| 3986 | * included in the iteration and the first node to be visited. |
Tejun Heo | 75501a6 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 3987 | * |
Tejun Heo | 87fb54f | 2013-12-06 15:11:55 -0500 | [diff] [blame] | 3988 | * While this function requires cgroup_mutex or RCU read locking, it |
| 3989 | * doesn't require the whole traversal to be contained in a single critical |
| 3990 | * section. This function will return the correct next descendant as long |
| 3991 | * as both @pos and @root are accessible and @pos is a descendant of @root. |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 3992 | * |
| 3993 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 3994 | * css which finished ->css_online() is guaranteed to be visible in the |
| 3995 | * future iterations and will stay visible until the last reference is put. |
| 3996 | * A css which hasn't finished ->css_online() or already finished |
| 3997 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 3998 | * responsibility to synchronize against on/offlining. |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 3999 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4000 | struct cgroup_subsys_state * |
| 4001 | css_next_descendant_pre(struct cgroup_subsys_state *pos, |
| 4002 | struct cgroup_subsys_state *root) |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4003 | { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4004 | struct cgroup_subsys_state *next; |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4005 | |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 4006 | cgroup_assert_mutex_or_rcu_locked(); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4007 | |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 4008 | /* if first iteration, visit @root */ |
Tejun Heo | 7805d00 | 2013-05-24 10:50:24 +0900 | [diff] [blame] | 4009 | if (!pos) |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 4010 | return root; |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4011 | |
| 4012 | /* visit the first child if exists */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4013 | next = css_next_child(NULL, pos); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4014 | if (next) |
| 4015 | return next; |
| 4016 | |
| 4017 | /* no child, visit my or the closest ancestor's next sibling */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4018 | while (pos != root) { |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4019 | next = css_next_child(pos, pos->parent); |
Tejun Heo | 75501a6 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 4020 | if (next) |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4021 | return next; |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4022 | pos = pos->parent; |
Tejun Heo | 7805d00 | 2013-05-24 10:50:24 +0900 | [diff] [blame] | 4023 | } |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4024 | |
| 4025 | return NULL; |
| 4026 | } |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4027 | |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4028 | /** |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4029 | * css_rightmost_descendant - return the rightmost descendant of a css |
| 4030 | * @pos: css of interest |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4031 | * |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4032 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
| 4033 | * is returned. This can be used during pre-order traversal to skip |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4034 | * subtree of @pos. |
Tejun Heo | 75501a6 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 4035 | * |
Tejun Heo | 87fb54f | 2013-12-06 15:11:55 -0500 | [diff] [blame] | 4036 | * While this function requires cgroup_mutex or RCU read locking, it |
| 4037 | * doesn't require the whole traversal to be contained in a single critical |
| 4038 | * section. This function will return the correct rightmost descendant as |
| 4039 | * long as @pos is accessible. |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4040 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4041 | struct cgroup_subsys_state * |
| 4042 | css_rightmost_descendant(struct cgroup_subsys_state *pos) |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4043 | { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4044 | struct cgroup_subsys_state *last, *tmp; |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4045 | |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 4046 | cgroup_assert_mutex_or_rcu_locked(); |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4047 | |
| 4048 | do { |
| 4049 | last = pos; |
| 4050 | /* ->prev isn't RCU safe, walk ->next till the end */ |
| 4051 | pos = NULL; |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4052 | css_for_each_child(tmp, last) |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4053 | pos = tmp; |
| 4054 | } while (pos); |
| 4055 | |
| 4056 | return last; |
| 4057 | } |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 4058 | |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4059 | static struct cgroup_subsys_state * |
| 4060 | css_leftmost_descendant(struct cgroup_subsys_state *pos) |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4061 | { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4062 | struct cgroup_subsys_state *last; |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4063 | |
| 4064 | do { |
| 4065 | last = pos; |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4066 | pos = css_next_child(NULL, pos); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4067 | } while (pos); |
| 4068 | |
| 4069 | return last; |
| 4070 | } |
| 4071 | |
| 4072 | /** |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4073 | * css_next_descendant_post - find the next descendant for post-order walk |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4074 | * @pos: the current position (%NULL to initiate traversal) |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4075 | * @root: css whose descendants to walk |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4076 | * |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4077 | * To be used by css_for_each_descendant_post(). Find the next descendant |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 4078 | * to visit for post-order traversal of @root's descendants. @root is |
| 4079 | * included in the iteration and the last node to be visited. |
Tejun Heo | 75501a6 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 4080 | * |
Tejun Heo | 87fb54f | 2013-12-06 15:11:55 -0500 | [diff] [blame] | 4081 | * While this function requires cgroup_mutex or RCU read locking, it |
| 4082 | * doesn't require the whole traversal to be contained in a single critical |
| 4083 | * section. This function will return the correct next descendant as long |
| 4084 | * as both @pos and @cgroup are accessible and @pos is a descendant of |
| 4085 | * @cgroup. |
Tejun Heo | c2931b7 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 4086 | * |
| 4087 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 4088 | * css which finished ->css_online() is guaranteed to be visible in the |
| 4089 | * future iterations and will stay visible until the last reference is put. |
| 4090 | * A css which hasn't finished ->css_online() or already finished |
| 4091 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 4092 | * responsibility to synchronize against on/offlining. |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4093 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4094 | struct cgroup_subsys_state * |
| 4095 | css_next_descendant_post(struct cgroup_subsys_state *pos, |
| 4096 | struct cgroup_subsys_state *root) |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4097 | { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4098 | struct cgroup_subsys_state *next; |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4099 | |
Tejun Heo | 8353da1 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 4100 | cgroup_assert_mutex_or_rcu_locked(); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4101 | |
Tejun Heo | 58b79a9 | 2013-09-06 15:31:08 -0400 | [diff] [blame] | 4102 | /* if first iteration, visit leftmost descendant which may be @root */ |
| 4103 | if (!pos) |
| 4104 | return css_leftmost_descendant(root); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4105 | |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 4106 | /* if we visited @root, we're done */ |
| 4107 | if (pos == root) |
| 4108 | return NULL; |
| 4109 | |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4110 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4111 | next = css_next_child(pos, pos->parent); |
Tejun Heo | 75501a6 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 4112 | if (next) |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 4113 | return css_leftmost_descendant(next); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4114 | |
| 4115 | /* no sibling left, visit parent */ |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4116 | return pos->parent; |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4117 | } |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 4118 | |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 4119 | /** |
| 4120 | * css_has_online_children - does a css have online children |
| 4121 | * @css: the target css |
| 4122 | * |
| 4123 | * Returns %true if @css has any online children; otherwise, %false. This |
| 4124 | * function can be called from any context but the caller is responsible |
| 4125 | * for synchronizing against on/offlining as necessary. |
| 4126 | */ |
| 4127 | bool css_has_online_children(struct cgroup_subsys_state *css) |
Tejun Heo | cbc125e | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 4128 | { |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 4129 | struct cgroup_subsys_state *child; |
| 4130 | bool ret = false; |
Tejun Heo | cbc125e | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 4131 | |
| 4132 | rcu_read_lock(); |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 4133 | css_for_each_child(child, css) { |
Li Zefan | 99bae5f | 2014-06-12 14:31:31 +0800 | [diff] [blame] | 4134 | if (child->flags & CSS_ONLINE) { |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 4135 | ret = true; |
| 4136 | break; |
Tejun Heo | cbc125e | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 4137 | } |
| 4138 | } |
| 4139 | rcu_read_unlock(); |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 4140 | return ret; |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 4141 | } |
| 4142 | |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4143 | /** |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4144 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4145 | * @it: the iterator to advance |
| 4146 | * |
| 4147 | * Advance @it to the next css_set to walk. |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4148 | */ |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4149 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4150 | { |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4151 | struct list_head *l = it->cset_pos; |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4152 | struct cgrp_cset_link *link; |
| 4153 | struct css_set *cset; |
| 4154 | |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 4155 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4156 | |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4157 | /* Advance to the next non-empty css_set */ |
| 4158 | do { |
| 4159 | l = l->next; |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4160 | if (l == it->cset_head) { |
| 4161 | it->cset_pos = NULL; |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4162 | it->task_pos = NULL; |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4163 | return; |
| 4164 | } |
Tejun Heo | 3ebb2b6 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4165 | |
| 4166 | if (it->ss) { |
| 4167 | cset = container_of(l, struct css_set, |
| 4168 | e_cset_node[it->ss->id]); |
| 4169 | } else { |
| 4170 | link = list_entry(l, struct cgrp_cset_link, cset_link); |
| 4171 | cset = link->cset; |
| 4172 | } |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 4173 | } while (!css_set_populated(cset)); |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 4174 | |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4175 | it->cset_pos = l; |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 4176 | |
| 4177 | if (!list_empty(&cset->tasks)) |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4178 | it->task_pos = cset->tasks.next; |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 4179 | else |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4180 | it->task_pos = cset->mg_tasks.next; |
| 4181 | |
| 4182 | it->tasks_head = &cset->tasks; |
| 4183 | it->mg_tasks_head = &cset->mg_tasks; |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4184 | |
| 4185 | /* |
| 4186 | * We don't keep css_sets locked across iteration steps and thus |
| 4187 | * need to take steps to ensure that iteration can be resumed after |
| 4188 | * the lock is re-acquired. Iteration is performed at two levels - |
| 4189 | * css_sets and tasks in them. |
| 4190 | * |
| 4191 | * Once created, a css_set never leaves its cgroup lists, so a |
| 4192 | * pinned css_set is guaranteed to stay put and we can resume |
| 4193 | * iteration afterwards. |
| 4194 | * |
| 4195 | * Tasks may leave @cset across iteration steps. This is resolved |
| 4196 | * by registering each iterator with the css_set currently being |
| 4197 | * walked and making css_set_move_task() advance iterators whose |
| 4198 | * next task is leaving. |
| 4199 | */ |
| 4200 | if (it->cur_cset) { |
| 4201 | list_del(&it->iters_node); |
| 4202 | put_css_set_locked(it->cur_cset); |
| 4203 | } |
| 4204 | get_css_set(cset); |
| 4205 | it->cur_cset = cset; |
| 4206 | list_add(&it->iters_node, &cset->task_iters); |
Tejun Heo | d515876 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4207 | } |
| 4208 | |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4209 | static void css_task_iter_advance(struct css_task_iter *it) |
| 4210 | { |
| 4211 | struct list_head *l = it->task_pos; |
| 4212 | |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 4213 | lockdep_assert_held(&css_set_lock); |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4214 | WARN_ON_ONCE(!l); |
| 4215 | |
| 4216 | /* |
| 4217 | * Advance iterator to find next entry. cset->tasks is consumed |
| 4218 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the |
| 4219 | * next cset. |
| 4220 | */ |
| 4221 | l = l->next; |
| 4222 | |
| 4223 | if (l == it->tasks_head) |
| 4224 | l = it->mg_tasks_head->next; |
| 4225 | |
| 4226 | if (l == it->mg_tasks_head) |
| 4227 | css_task_iter_advance_css_set(it); |
| 4228 | else |
| 4229 | it->task_pos = l; |
| 4230 | } |
| 4231 | |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4232 | /** |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4233 | * css_task_iter_start - initiate task iteration |
| 4234 | * @css: the css to walk tasks of |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4235 | * @it: the task iterator to use |
| 4236 | * |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4237 | * Initiate iteration through the tasks of @css. The caller can call |
| 4238 | * css_task_iter_next() to walk through the tasks until the function |
| 4239 | * returns NULL. On completion of iteration, css_task_iter_end() must be |
| 4240 | * called. |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4241 | */ |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4242 | void css_task_iter_start(struct cgroup_subsys_state *css, |
| 4243 | struct css_task_iter *it) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4244 | { |
Tejun Heo | 56fde9e | 2014-02-13 06:58:38 -0500 | [diff] [blame] | 4245 | /* no one should try to iterate before mounting cgroups */ |
| 4246 | WARN_ON_ONCE(!use_task_css_set_links); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4247 | |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4248 | memset(it, 0, sizeof(*it)); |
| 4249 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4250 | spin_lock_irq(&css_set_lock); |
Tejun Heo | c59cd3d | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4251 | |
Tejun Heo | 3ebb2b6 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4252 | it->ss = css->ss; |
| 4253 | |
| 4254 | if (it->ss) |
| 4255 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; |
| 4256 | else |
| 4257 | it->cset_pos = &css->cgroup->cset_links; |
| 4258 | |
Tejun Heo | 0f0a2b4 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 4259 | it->cset_head = it->cset_pos; |
Tejun Heo | c59cd3d | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4260 | |
Tejun Heo | ecb9d53 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4261 | css_task_iter_advance_css_set(it); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4262 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4263 | spin_unlock_irq(&css_set_lock); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 4264 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4265 | |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4266 | /** |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4267 | * css_task_iter_next - return the next task for the iterator |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4268 | * @it: the task iterator being iterated |
| 4269 | * |
| 4270 | * The "next" function for task iteration. @it should have been |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4271 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
| 4272 | * reaches the end. |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4273 | */ |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4274 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4275 | { |
Tejun Heo | d574567 | 2015-10-29 11:43:05 +0900 | [diff] [blame] | 4276 | if (it->cur_task) { |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4277 | put_task_struct(it->cur_task); |
Tejun Heo | d574567 | 2015-10-29 11:43:05 +0900 | [diff] [blame] | 4278 | it->cur_task = NULL; |
| 4279 | } |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4280 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4281 | spin_lock_irq(&css_set_lock); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4282 | |
Tejun Heo | d574567 | 2015-10-29 11:43:05 +0900 | [diff] [blame] | 4283 | if (it->task_pos) { |
| 4284 | it->cur_task = list_entry(it->task_pos, struct task_struct, |
| 4285 | cg_list); |
| 4286 | get_task_struct(it->cur_task); |
| 4287 | css_task_iter_advance(it); |
| 4288 | } |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4289 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4290 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4291 | |
| 4292 | return it->cur_task; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4293 | } |
| 4294 | |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4295 | /** |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4296 | * css_task_iter_end - finish task iteration |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4297 | * @it: the task iterator to finish |
| 4298 | * |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4299 | * Finish task iteration started by css_task_iter_start(). |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4300 | */ |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4301 | void css_task_iter_end(struct css_task_iter *it) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4302 | { |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4303 | if (it->cur_cset) { |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4304 | spin_lock_irq(&css_set_lock); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4305 | list_del(&it->iters_node); |
| 4306 | put_css_set_locked(it->cur_cset); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4307 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | ed27b9f | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 4308 | } |
| 4309 | |
| 4310 | if (it->cur_task) |
| 4311 | put_task_struct(it->cur_task); |
Tejun Heo | 8cc9934 | 2013-04-07 09:29:50 -0700 | [diff] [blame] | 4312 | } |
| 4313 | |
| 4314 | /** |
| 4315 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
| 4316 | * @to: cgroup to which the tasks will be moved |
| 4317 | * @from: cgroup in which the tasks currently reside |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4318 | * |
| 4319 | * Locking rules between cgroup_post_fork() and the migration path |
| 4320 | * guarantee that, if a task is forking while being migrated, the new child |
| 4321 | * is guaranteed to be either visible in the source cgroup after the |
| 4322 | * parent's migration is complete or put into the target cgroup. No task |
| 4323 | * can slip out of migration through forking. |
Tejun Heo | 8cc9934 | 2013-04-07 09:29:50 -0700 | [diff] [blame] | 4324 | */ |
| 4325 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
| 4326 | { |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4327 | LIST_HEAD(preloaded_csets); |
| 4328 | struct cgrp_cset_link *link; |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4329 | struct css_task_iter it; |
| 4330 | struct task_struct *task; |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4331 | int ret; |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4332 | |
Tejun Heo | 6c694c8 | 2016-03-08 11:51:25 -0500 | [diff] [blame] | 4333 | if (!cgroup_may_migrate_to(to)) |
| 4334 | return -EBUSY; |
| 4335 | |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4336 | mutex_lock(&cgroup_mutex); |
| 4337 | |
Eric W. Biederman | eedd0f4 | 2016-07-15 06:35:51 -0500 | [diff] [blame] | 4338 | percpu_down_write(&cgroup_threadgroup_rwsem); |
| 4339 | |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4340 | /* all tasks in @from are being moved, all csets are source */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4341 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4342 | list_for_each_entry(link, &from->cset_links, cset_link) |
| 4343 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 4344 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4345 | |
Tejun Heo | e485798 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 4346 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4347 | if (ret) |
| 4348 | goto out_err; |
| 4349 | |
| 4350 | /* |
Rami Rosen | 2cfa2b1 | 2016-01-09 23:33:06 +0200 | [diff] [blame] | 4351 | * Migrate tasks one-by-one until @from is empty. This fails iff |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4352 | * ->can_attach() fails. |
| 4353 | */ |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4354 | do { |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 4355 | css_task_iter_start(&from->self, &it); |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4356 | task = css_task_iter_next(&it); |
| 4357 | if (task) |
| 4358 | get_task_struct(task); |
| 4359 | css_task_iter_end(&it); |
| 4360 | |
| 4361 | if (task) { |
Tejun Heo | 37ff9f8 | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 4362 | ret = cgroup_migrate(task, false, to->root); |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4363 | put_task_struct(task); |
| 4364 | } |
| 4365 | } while (task && !ret); |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4366 | out_err: |
| 4367 | cgroup_migrate_finish(&preloaded_csets); |
Eric W. Biederman | eedd0f4 | 2016-07-15 06:35:51 -0500 | [diff] [blame] | 4368 | percpu_up_write(&cgroup_threadgroup_rwsem); |
Tejun Heo | 952aaa1 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 4369 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | e406d1c | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4370 | return ret; |
Tejun Heo | 8cc9934 | 2013-04-07 09:29:50 -0700 | [diff] [blame] | 4371 | } |
| 4372 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4373 | /* |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4374 | * Stuff for reading the 'tasks'/'procs' files. |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4375 | * |
| 4376 | * Reading this file can return large amounts of data if a cgroup has |
| 4377 | * *lots* of attached tasks. So it may need several calls to read(), |
| 4378 | * but we cannot guarantee that the information we produce is correct |
| 4379 | * unless we produce it entirely atomically. |
| 4380 | * |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4381 | */ |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4382 | |
Li Zefan | 2452825 | 2012-01-20 11:58:43 +0800 | [diff] [blame] | 4383 | /* which pidlist file are we talking about? */ |
| 4384 | enum cgroup_filetype { |
| 4385 | CGROUP_FILE_PROCS, |
| 4386 | CGROUP_FILE_TASKS, |
| 4387 | }; |
| 4388 | |
| 4389 | /* |
| 4390 | * A pidlist is a list of pids that virtually represents the contents of one |
| 4391 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, |
| 4392 | * a pair (one each for procs, tasks) for each pid namespace that's relevant |
| 4393 | * to the cgroup. |
| 4394 | */ |
| 4395 | struct cgroup_pidlist { |
| 4396 | /* |
| 4397 | * used to find which pidlist is wanted. doesn't change as long as |
| 4398 | * this particular list stays in the list. |
| 4399 | */ |
| 4400 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; |
| 4401 | /* array of xids */ |
| 4402 | pid_t *list; |
| 4403 | /* how many elements the above list has */ |
| 4404 | int length; |
Li Zefan | 2452825 | 2012-01-20 11:58:43 +0800 | [diff] [blame] | 4405 | /* each of these stored in a list by its cgroup */ |
| 4406 | struct list_head links; |
| 4407 | /* pointer to the cgroup we belong to, for list removal purposes */ |
| 4408 | struct cgroup *owner; |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4409 | /* for delayed destruction */ |
| 4410 | struct delayed_work destroy_dwork; |
Li Zefan | 2452825 | 2012-01-20 11:58:43 +0800 | [diff] [blame] | 4411 | }; |
| 4412 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4413 | /* |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4414 | * The following two functions "fix" the issue where there are more pids |
| 4415 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. |
| 4416 | * TODO: replace with a kernel-wide solution to this problem |
| 4417 | */ |
| 4418 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) |
| 4419 | static void *pidlist_allocate(int count) |
| 4420 | { |
| 4421 | if (PIDLIST_TOO_LARGE(count)) |
| 4422 | return vmalloc(count * sizeof(pid_t)); |
| 4423 | else |
| 4424 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); |
| 4425 | } |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4426 | |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4427 | static void pidlist_free(void *p) |
| 4428 | { |
Bandan Das | 5879451 | 2015-03-02 17:51:10 -0500 | [diff] [blame] | 4429 | kvfree(p); |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4430 | } |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4431 | |
| 4432 | /* |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4433 | * Used to destroy all pidlists lingering waiting for destroy timer. None |
| 4434 | * should be left afterwards. |
| 4435 | */ |
| 4436 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) |
| 4437 | { |
| 4438 | struct cgroup_pidlist *l, *tmp_l; |
| 4439 | |
| 4440 | mutex_lock(&cgrp->pidlist_mutex); |
| 4441 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) |
| 4442 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); |
| 4443 | mutex_unlock(&cgrp->pidlist_mutex); |
| 4444 | |
| 4445 | flush_workqueue(cgroup_pidlist_destroy_wq); |
| 4446 | BUG_ON(!list_empty(&cgrp->pidlists)); |
| 4447 | } |
| 4448 | |
| 4449 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) |
| 4450 | { |
| 4451 | struct delayed_work *dwork = to_delayed_work(work); |
| 4452 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, |
| 4453 | destroy_dwork); |
| 4454 | struct cgroup_pidlist *tofree = NULL; |
| 4455 | |
| 4456 | mutex_lock(&l->owner->pidlist_mutex); |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4457 | |
| 4458 | /* |
Tejun Heo | 0450236 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4459 | * Destroy iff we didn't get queued again. The state won't change |
| 4460 | * as destroy_dwork can only be queued while locked. |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4461 | */ |
Tejun Heo | 0450236 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4462 | if (!delayed_work_pending(dwork)) { |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4463 | list_del(&l->links); |
| 4464 | pidlist_free(l->list); |
| 4465 | put_pid_ns(l->key.ns); |
| 4466 | tofree = l; |
| 4467 | } |
| 4468 | |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4469 | mutex_unlock(&l->owner->pidlist_mutex); |
| 4470 | kfree(tofree); |
| 4471 | } |
| 4472 | |
| 4473 | /* |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4474 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
Li Zefan | 6ee211a | 2013-03-12 15:36:00 -0700 | [diff] [blame] | 4475 | * Returns the number of unique elements. |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4476 | */ |
Li Zefan | 6ee211a | 2013-03-12 15:36:00 -0700 | [diff] [blame] | 4477 | static int pidlist_uniq(pid_t *list, int length) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4478 | { |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4479 | int src, dest = 1; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4480 | |
| 4481 | /* |
| 4482 | * we presume the 0th element is unique, so i starts at 1. trivial |
| 4483 | * edge cases first; no work needs to be done for either |
| 4484 | */ |
| 4485 | if (length == 0 || length == 1) |
| 4486 | return length; |
| 4487 | /* src and dest walk down the list; dest counts unique elements */ |
| 4488 | for (src = 1; src < length; src++) { |
| 4489 | /* find next unique element */ |
| 4490 | while (list[src] == list[src-1]) { |
| 4491 | src++; |
| 4492 | if (src == length) |
| 4493 | goto after; |
| 4494 | } |
| 4495 | /* dest always points to where the next unique element goes */ |
| 4496 | list[dest] = list[src]; |
| 4497 | dest++; |
| 4498 | } |
| 4499 | after: |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4500 | return dest; |
| 4501 | } |
| 4502 | |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4503 | /* |
| 4504 | * The two pid files - task and cgroup.procs - guaranteed that the result |
| 4505 | * is sorted, which forced this whole pidlist fiasco. As pid order is |
| 4506 | * different per namespace, each namespace needs differently sorted list, |
| 4507 | * making it impossible to use, for example, single rbtree of member tasks |
| 4508 | * sorted by task pointer. As pidlists can be fairly large, allocating one |
| 4509 | * per open file is dangerous, so cgroup had to implement shared pool of |
| 4510 | * pidlists keyed by cgroup and namespace. |
| 4511 | * |
| 4512 | * All this extra complexity was caused by the original implementation |
| 4513 | * committing to an entirely unnecessary property. In the long term, we |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 4514 | * want to do away with it. Explicitly scramble sort order if on the |
| 4515 | * default hierarchy so that no such expectation exists in the new |
| 4516 | * interface. |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4517 | * |
| 4518 | * Scrambling is done by swapping every two consecutive bits, which is |
| 4519 | * non-identity one-to-one mapping which disturbs sort order sufficiently. |
| 4520 | */ |
| 4521 | static pid_t pid_fry(pid_t pid) |
| 4522 | { |
| 4523 | unsigned a = pid & 0x55555555; |
| 4524 | unsigned b = pid & 0xAAAAAAAA; |
| 4525 | |
| 4526 | return (a << 1) | (b >> 1); |
| 4527 | } |
| 4528 | |
| 4529 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) |
| 4530 | { |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 4531 | if (cgroup_on_dfl(cgrp)) |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4532 | return pid_fry(pid); |
| 4533 | else |
| 4534 | return pid; |
| 4535 | } |
| 4536 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4537 | static int cmppid(const void *a, const void *b) |
| 4538 | { |
| 4539 | return *(pid_t *)a - *(pid_t *)b; |
| 4540 | } |
| 4541 | |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4542 | static int fried_cmppid(const void *a, const void *b) |
| 4543 | { |
| 4544 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); |
| 4545 | } |
| 4546 | |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4547 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
| 4548 | enum cgroup_filetype type) |
| 4549 | { |
| 4550 | struct cgroup_pidlist *l; |
| 4551 | /* don't need task_nsproxy() if we're looking at ourself */ |
Eric W. Biederman | 17cf22c | 2010-03-02 14:51:53 -0800 | [diff] [blame] | 4552 | struct pid_namespace *ns = task_active_pid_ns(current); |
Li Zefan | b70cc5f | 2010-03-10 15:22:12 -0800 | [diff] [blame] | 4553 | |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4554 | lockdep_assert_held(&cgrp->pidlist_mutex); |
| 4555 | |
| 4556 | list_for_each_entry(l, &cgrp->pidlists, links) |
| 4557 | if (l->key.type == type && l->key.ns == ns) |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4558 | return l; |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4559 | return NULL; |
| 4560 | } |
| 4561 | |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4562 | /* |
| 4563 | * find the appropriate pidlist for our purpose (given procs vs tasks) |
| 4564 | * returns with the lock on that pidlist already held, and takes care |
| 4565 | * of the use count, or returns NULL with no locks held if we're out of |
| 4566 | * memory. |
| 4567 | */ |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4568 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
| 4569 | enum cgroup_filetype type) |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4570 | { |
| 4571 | struct cgroup_pidlist *l; |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4572 | |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4573 | lockdep_assert_held(&cgrp->pidlist_mutex); |
| 4574 | |
| 4575 | l = cgroup_pidlist_find(cgrp, type); |
| 4576 | if (l) |
| 4577 | return l; |
| 4578 | |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4579 | /* entry not found; create a new one */ |
Tejun Heo | f4f4be2 | 2013-06-12 21:04:51 -0700 | [diff] [blame] | 4580 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4581 | if (!l) |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4582 | return l; |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4583 | |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4584 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4585 | l->key.type = type; |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4586 | /* don't need task_nsproxy() if we're looking at ourself */ |
| 4587 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4588 | l->owner = cgrp; |
| 4589 | list_add(&l->links, &cgrp->pidlists); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4590 | return l; |
| 4591 | } |
| 4592 | |
| 4593 | /* |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4594 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids |
| 4595 | */ |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4596 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
| 4597 | struct cgroup_pidlist **lp) |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4598 | { |
| 4599 | pid_t *array; |
| 4600 | int length; |
| 4601 | int pid, n = 0; /* used for populating the array */ |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4602 | struct css_task_iter it; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4603 | struct task_struct *tsk; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4604 | struct cgroup_pidlist *l; |
| 4605 | |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4606 | lockdep_assert_held(&cgrp->pidlist_mutex); |
| 4607 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4608 | /* |
| 4609 | * If cgroup gets more users after we read count, we won't have |
| 4610 | * enough space - tough. This race is indistinguishable to the |
| 4611 | * caller from the case that the additional cgroup users didn't |
| 4612 | * show up until sometime later on. |
| 4613 | */ |
| 4614 | length = cgroup_task_count(cgrp); |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4615 | array = pidlist_allocate(length); |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4616 | if (!array) |
| 4617 | return -ENOMEM; |
| 4618 | /* now, populate the array */ |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 4619 | css_task_iter_start(&cgrp->self, &it); |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4620 | while ((tsk = css_task_iter_next(&it))) { |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4621 | if (unlikely(n == length)) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4622 | break; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4623 | /* get tgid or pid for procs or tasks file respectively */ |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4624 | if (type == CGROUP_FILE_PROCS) |
| 4625 | pid = task_tgid_vnr(tsk); |
| 4626 | else |
| 4627 | pid = task_pid_vnr(tsk); |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4628 | if (pid > 0) /* make sure to only use valid results */ |
| 4629 | array[n++] = pid; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 4630 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4631 | css_task_iter_end(&it); |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4632 | length = n; |
| 4633 | /* now sort & (if procs) strip out duplicates */ |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 4634 | if (cgroup_on_dfl(cgrp)) |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4635 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
| 4636 | else |
| 4637 | sort(array, length, sizeof(pid_t), cmppid, NULL); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4638 | if (type == CGROUP_FILE_PROCS) |
Li Zefan | 6ee211a | 2013-03-12 15:36:00 -0700 | [diff] [blame] | 4639 | length = pidlist_uniq(array, length); |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4640 | |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4641 | l = cgroup_pidlist_find_create(cgrp, type); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4642 | if (!l) { |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4643 | pidlist_free(array); |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4644 | return -ENOMEM; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4645 | } |
Tejun Heo | e6b8171 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4646 | |
| 4647 | /* store array, freeing old if necessary */ |
Ben Blum | d1d9fd3 | 2009-09-23 15:56:28 -0700 | [diff] [blame] | 4648 | pidlist_free(l->list); |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4649 | l->list = array; |
| 4650 | l->length = length; |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 4651 | *lp = l; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4652 | return 0; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4653 | } |
| 4654 | |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4655 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 4656 | * cgroupstats_build - build and fill cgroupstats |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4657 | * @stats: cgroupstats to fill information into |
| 4658 | * @dentry: A dentry entry belonging to the cgroup for which stats have |
| 4659 | * been requested. |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 4660 | * |
| 4661 | * Build and fill cgroupstats so that taskstats can export it to user |
| 4662 | * space. |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4663 | */ |
| 4664 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) |
| 4665 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4666 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 4667 | struct cgroup *cgrp; |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4668 | struct css_task_iter it; |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4669 | struct task_struct *tsk; |
Li Zefan | 33d283b | 2008-11-19 15:36:48 -0800 | [diff] [blame] | 4670 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4671 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
| 4672 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
| 4673 | kernfs_type(kn) != KERNFS_DIR) |
| 4674 | return -EINVAL; |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4675 | |
Li Zefan | bad3466 | 2014-02-14 16:54:28 +0800 | [diff] [blame] | 4676 | mutex_lock(&cgroup_mutex); |
| 4677 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4678 | /* |
| 4679 | * We aren't being called from kernfs and there's no guarantee on |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 4680 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4681 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
| 4682 | */ |
| 4683 | rcu_read_lock(); |
| 4684 | cgrp = rcu_dereference(kn->priv); |
Li Zefan | bad3466 | 2014-02-14 16:54:28 +0800 | [diff] [blame] | 4685 | if (!cgrp || cgroup_is_dead(cgrp)) { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4686 | rcu_read_unlock(); |
Li Zefan | bad3466 | 2014-02-14 16:54:28 +0800 | [diff] [blame] | 4687 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4688 | return -ENOENT; |
| 4689 | } |
Li Zefan | bad3466 | 2014-02-14 16:54:28 +0800 | [diff] [blame] | 4690 | rcu_read_unlock(); |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4691 | |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 4692 | css_task_iter_start(&cgrp->self, &it); |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4693 | while ((tsk = css_task_iter_next(&it))) { |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4694 | switch (tsk->state) { |
| 4695 | case TASK_RUNNING: |
| 4696 | stats->nr_running++; |
| 4697 | break; |
| 4698 | case TASK_INTERRUPTIBLE: |
| 4699 | stats->nr_sleeping++; |
| 4700 | break; |
| 4701 | case TASK_UNINTERRUPTIBLE: |
| 4702 | stats->nr_uninterruptible++; |
| 4703 | break; |
| 4704 | case TASK_STOPPED: |
| 4705 | stats->nr_stopped++; |
| 4706 | break; |
| 4707 | default: |
| 4708 | if (delayacct_is_task_waiting_on_io(tsk)) |
| 4709 | stats->nr_io_wait++; |
| 4710 | break; |
| 4711 | } |
| 4712 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 4713 | css_task_iter_end(&it); |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4714 | |
Li Zefan | bad3466 | 2014-02-14 16:54:28 +0800 | [diff] [blame] | 4715 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4716 | return 0; |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 4717 | } |
| 4718 | |
Paul Menage | 8f3ff20 | 2009-09-23 15:56:25 -0700 | [diff] [blame] | 4719 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4720 | /* |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4721 | * seq_file methods for the tasks/procs files. The seq_file position is the |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4722 | * next pid to display; the seq_file iterator is a pointer to the pid |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4723 | * in the cgroup->l->list array. |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4724 | */ |
| 4725 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4726 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4727 | { |
| 4728 | /* |
| 4729 | * Initially we receive a position value that corresponds to |
| 4730 | * one more than the last pid shown (or 0 on the first call or |
| 4731 | * after a seek to the start). Use a binary-search to find the |
| 4732 | * next pid to display, if any |
| 4733 | */ |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4734 | struct kernfs_open_file *of = s->private; |
Tejun Heo | 7da1127 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4735 | struct cgroup *cgrp = seq_css(s)->cgroup; |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4736 | struct cgroup_pidlist *l; |
Tejun Heo | 7da1127 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4737 | enum cgroup_filetype type = seq_cft(s)->private; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4738 | int index = 0, pid = *pos; |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4739 | int *iter, ret; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4740 | |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4741 | mutex_lock(&cgrp->pidlist_mutex); |
| 4742 | |
| 4743 | /* |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4744 | * !NULL @of->priv indicates that this isn't the first start() |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4745 | * after open. If the matching pidlist is around, we can use that. |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4746 | * Look for it. Note that @of->priv can't be used directly. It |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4747 | * could already have been destroyed. |
| 4748 | */ |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4749 | if (of->priv) |
| 4750 | of->priv = cgroup_pidlist_find(cgrp, type); |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4751 | |
| 4752 | /* |
| 4753 | * Either this is the first start() after open or the matching |
| 4754 | * pidlist has been destroyed inbetween. Create a new one. |
| 4755 | */ |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4756 | if (!of->priv) { |
| 4757 | ret = pidlist_array_load(cgrp, type, |
| 4758 | (struct cgroup_pidlist **)&of->priv); |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4759 | if (ret) |
| 4760 | return ERR_PTR(ret); |
| 4761 | } |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4762 | l = of->priv; |
Tejun Heo | 4bac00d | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4763 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4764 | if (pid) { |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4765 | int end = l->length; |
Stephen Rothwell | 2077776 | 2008-10-21 16:11:20 +1100 | [diff] [blame] | 4766 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4767 | while (index < end) { |
| 4768 | int mid = (index + end) / 2; |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4769 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4770 | index = mid; |
| 4771 | break; |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4772 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4773 | index = mid + 1; |
| 4774 | else |
| 4775 | end = mid; |
| 4776 | } |
| 4777 | } |
| 4778 | /* If we're off the end of the array, we're done */ |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4779 | if (index >= l->length) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4780 | return NULL; |
| 4781 | /* Update the abstract position to be the actual pid that we found */ |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4782 | iter = l->list + index; |
Tejun Heo | afb2bc1 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4783 | *pos = cgroup_pid_fry(cgrp, *iter); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4784 | return iter; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4785 | } |
| 4786 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4787 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4788 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4789 | struct kernfs_open_file *of = s->private; |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4790 | struct cgroup_pidlist *l = of->priv; |
Tejun Heo | 6223685 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 4791 | |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4792 | if (l) |
| 4793 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, |
Tejun Heo | 0450236 | 2013-11-29 10:42:59 -0500 | [diff] [blame] | 4794 | CGROUP_PIDLIST_DESTROY_DELAY); |
Tejun Heo | 7da1127 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4795 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4796 | } |
| 4797 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4798 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4799 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 4800 | struct kernfs_open_file *of = s->private; |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4801 | struct cgroup_pidlist *l = of->priv; |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4802 | pid_t *p = v; |
| 4803 | pid_t *end = l->list + l->length; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4804 | /* |
| 4805 | * Advance to the next pid in the array. If this goes off the |
| 4806 | * end, we're done |
| 4807 | */ |
| 4808 | p++; |
| 4809 | if (p >= end) { |
| 4810 | return NULL; |
| 4811 | } else { |
Tejun Heo | 7da1127 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4812 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4813 | return p; |
| 4814 | } |
| 4815 | } |
| 4816 | |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4817 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4818 | { |
Joe Perches | 94ff212 | 2015-04-15 16:18:20 -0700 | [diff] [blame] | 4819 | seq_printf(s, "%d\n", *(int *)v); |
| 4820 | |
| 4821 | return 0; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 4822 | } |
| 4823 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4824 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
| 4825 | struct cftype *cft) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 4826 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4827 | return notify_on_release(css->cgroup); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 4828 | } |
| 4829 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4830 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
| 4831 | struct cftype *cft, u64 val) |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 4832 | { |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 4833 | if (val) |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4834 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 4835 | else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4836 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 4837 | return 0; |
| 4838 | } |
| 4839 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4840 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
| 4841 | struct cftype *cft) |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4842 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4843 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4844 | } |
| 4845 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4846 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
| 4847 | struct cftype *cft, u64 val) |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4848 | { |
| 4849 | if (val) |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4850 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4851 | else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4852 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4853 | return 0; |
| 4854 | } |
| 4855 | |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4856 | /* cgroup core interface files for the default hierarchy */ |
| 4857 | static struct cftype cgroup_dfl_base_files[] = { |
| 4858 | { |
| 4859 | .name = "cgroup.procs", |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 4860 | .file_offset = offsetof(struct cgroup, procs_file), |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4861 | .seq_start = cgroup_pidlist_start, |
| 4862 | .seq_next = cgroup_pidlist_next, |
| 4863 | .seq_stop = cgroup_pidlist_stop, |
| 4864 | .seq_show = cgroup_pidlist_show, |
| 4865 | .private = CGROUP_FILE_PROCS, |
| 4866 | .write = cgroup_procs_write, |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4867 | }, |
| 4868 | { |
| 4869 | .name = "cgroup.controllers", |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4870 | .seq_show = cgroup_controllers_show, |
| 4871 | }, |
| 4872 | { |
| 4873 | .name = "cgroup.subtree_control", |
| 4874 | .seq_show = cgroup_subtree_control_show, |
| 4875 | .write = cgroup_subtree_control_write, |
| 4876 | }, |
| 4877 | { |
Tejun Heo | 4a07c22 | 2015-09-18 17:54:22 -0400 | [diff] [blame] | 4878 | .name = "cgroup.events", |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4879 | .flags = CFTYPE_NOT_ON_ROOT, |
Tejun Heo | 6f60ead | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 4880 | .file_offset = offsetof(struct cgroup, events_file), |
Tejun Heo | 4a07c22 | 2015-09-18 17:54:22 -0400 | [diff] [blame] | 4881 | .seq_show = cgroup_events_show, |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4882 | }, |
| 4883 | { } /* terminate */ |
| 4884 | }; |
| 4885 | |
| 4886 | /* cgroup core interface files for the legacy hierarchies */ |
| 4887 | static struct cftype cgroup_legacy_base_files[] = { |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 4888 | { |
Tejun Heo | d5c56ce | 2013-06-03 19:14:34 -0700 | [diff] [blame] | 4889 | .name = "cgroup.procs", |
Tejun Heo | 6612f05 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4890 | .seq_start = cgroup_pidlist_start, |
| 4891 | .seq_next = cgroup_pidlist_next, |
| 4892 | .seq_stop = cgroup_pidlist_stop, |
| 4893 | .seq_show = cgroup_pidlist_show, |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4894 | .private = CGROUP_FILE_PROCS, |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 4895 | .write = cgroup_procs_write, |
Ben Blum | 102a775 | 2009-09-23 15:56:26 -0700 | [diff] [blame] | 4896 | }, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 4897 | { |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 4898 | .name = "cgroup.clone_children", |
| 4899 | .read_u64 = cgroup_clone_children_read, |
| 4900 | .write_u64 = cgroup_clone_children_write, |
| 4901 | }, |
Tejun Heo | 6e6ff25 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4902 | { |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 4903 | .name = "cgroup.sane_behavior", |
| 4904 | .flags = CFTYPE_ONLY_ON_ROOT, |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4905 | .seq_show = cgroup_sane_behavior_show, |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 4906 | }, |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 4907 | { |
Tejun Heo | d5c56ce | 2013-06-03 19:14:34 -0700 | [diff] [blame] | 4908 | .name = "tasks", |
Tejun Heo | 6612f05 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4909 | .seq_start = cgroup_pidlist_start, |
| 4910 | .seq_next = cgroup_pidlist_next, |
| 4911 | .seq_stop = cgroup_pidlist_stop, |
| 4912 | .seq_show = cgroup_pidlist_show, |
Tejun Heo | 5d22444 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4913 | .private = CGROUP_FILE_TASKS, |
Tejun Heo | acbef75 | 2014-05-13 12:16:22 -0400 | [diff] [blame] | 4914 | .write = cgroup_tasks_write, |
Tejun Heo | d5c56ce | 2013-06-03 19:14:34 -0700 | [diff] [blame] | 4915 | }, |
| 4916 | { |
| 4917 | .name = "notify_on_release", |
Tejun Heo | d5c56ce | 2013-06-03 19:14:34 -0700 | [diff] [blame] | 4918 | .read_u64 = cgroup_read_notify_on_release, |
| 4919 | .write_u64 = cgroup_write_notify_on_release, |
| 4920 | }, |
Tejun Heo | 873fe09 | 2013-04-14 20:15:26 -0700 | [diff] [blame] | 4921 | { |
Tejun Heo | 6e6ff25 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4922 | .name = "release_agent", |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 4923 | .flags = CFTYPE_ONLY_ON_ROOT, |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4924 | .seq_show = cgroup_release_agent_show, |
Tejun Heo | 451af504 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4925 | .write = cgroup_release_agent_write, |
Tejun Heo | 5f46990 | 2014-02-11 11:52:48 -0500 | [diff] [blame] | 4926 | .max_write_len = PATH_MAX - 1, |
Tejun Heo | 6e6ff25 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4927 | }, |
Tejun Heo | db0416b | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4928 | { } /* terminate */ |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 4929 | }; |
| 4930 | |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 4931 | /* |
| 4932 | * css destruction is four-stage process. |
| 4933 | * |
| 4934 | * 1. Destruction starts. Killing of the percpu_ref is initiated. |
| 4935 | * Implemented in kill_css(). |
| 4936 | * |
| 4937 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 4938 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
| 4939 | * offlined by invoking offline_css(). After offlining, the base ref is |
| 4940 | * put. Implemented in css_killed_work_fn(). |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 4941 | * |
| 4942 | * 3. When the percpu_ref reaches zero, the only possible remaining |
| 4943 | * accessors are inside RCU read sections. css_release() schedules the |
| 4944 | * RCU callback. |
| 4945 | * |
| 4946 | * 4. After the grace period, the css can be freed. Implemented in |
| 4947 | * css_free_work_fn(). |
| 4948 | * |
| 4949 | * It is actually hairier because both step 2 and 4 require process context |
| 4950 | * and thus involve punting to css->destroy_work adding two additional |
| 4951 | * steps to the already complex sequence. |
| 4952 | */ |
Tejun Heo | 35ef10d | 2013-08-13 11:01:54 -0400 | [diff] [blame] | 4953 | static void css_free_work_fn(struct work_struct *work) |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 4954 | { |
| 4955 | struct cgroup_subsys_state *css = |
Tejun Heo | 35ef10d | 2013-08-13 11:01:54 -0400 | [diff] [blame] | 4956 | container_of(work, struct cgroup_subsys_state, destroy_work); |
Vladimir Davydov | 01e5865 | 2015-02-12 14:59:26 -0800 | [diff] [blame] | 4957 | struct cgroup_subsys *ss = css->ss; |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 4958 | struct cgroup *cgrp = css->cgroup; |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 4959 | |
Tejun Heo | 9a1049d | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 4960 | percpu_ref_exit(&css->refcnt); |
| 4961 | |
Vladimir Davydov | 01e5865 | 2015-02-12 14:59:26 -0800 | [diff] [blame] | 4962 | if (ss) { |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4963 | /* css free path */ |
Tejun Heo | 8bb5ef7 | 2016-01-21 15:32:15 -0500 | [diff] [blame] | 4964 | struct cgroup_subsys_state *parent = css->parent; |
Vladimir Davydov | 01e5865 | 2015-02-12 14:59:26 -0800 | [diff] [blame] | 4965 | int id = css->id; |
| 4966 | |
Vladimir Davydov | 01e5865 | 2015-02-12 14:59:26 -0800 | [diff] [blame] | 4967 | ss->css_free(css); |
| 4968 | cgroup_idr_remove(&ss->css_idr, id); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4969 | cgroup_put(cgrp); |
Tejun Heo | 8bb5ef7 | 2016-01-21 15:32:15 -0500 | [diff] [blame] | 4970 | |
| 4971 | if (parent) |
| 4972 | css_put(parent); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4973 | } else { |
| 4974 | /* cgroup free path */ |
| 4975 | atomic_dec(&cgrp->root->nr_cgrps); |
| 4976 | cgroup_pidlist_destroy_all(cgrp); |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 4977 | cancel_work_sync(&cgrp->release_agent_work); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4978 | |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4979 | if (cgroup_parent(cgrp)) { |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4980 | /* |
| 4981 | * We get a ref to the parent, and put the ref when |
| 4982 | * this cgroup is being freed, so it's guaranteed |
| 4983 | * that the parent won't be destroyed before its |
| 4984 | * children. |
| 4985 | */ |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4986 | cgroup_put(cgroup_parent(cgrp)); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 4987 | kernfs_put(cgrp->kn); |
| 4988 | kfree(cgrp); |
| 4989 | } else { |
| 4990 | /* |
| 4991 | * This is root cgroup's refcnt reaching zero, |
| 4992 | * which indicates that the root should be |
| 4993 | * released. |
| 4994 | */ |
| 4995 | cgroup_destroy_root(cgrp->root); |
| 4996 | } |
| 4997 | } |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 4998 | } |
| 4999 | |
| 5000 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
| 5001 | { |
| 5002 | struct cgroup_subsys_state *css = |
| 5003 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
| 5004 | |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5005 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 5006 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 5007 | } |
| 5008 | |
Tejun Heo | 25e15d8 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5009 | static void css_release_work_fn(struct work_struct *work) |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5010 | { |
| 5011 | struct cgroup_subsys_state *css = |
Tejun Heo | 25e15d8 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5012 | container_of(work, struct cgroup_subsys_state, destroy_work); |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5013 | struct cgroup_subsys *ss = css->ss; |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5014 | struct cgroup *cgrp = css->cgroup; |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5015 | |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5016 | mutex_lock(&cgroup_mutex); |
| 5017 | |
Tejun Heo | de3f034 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5018 | css->flags |= CSS_RELEASED; |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5019 | list_del_rcu(&css->sibling); |
| 5020 | |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5021 | if (ss) { |
| 5022 | /* css release path */ |
Vladimir Davydov | 01e5865 | 2015-02-12 14:59:26 -0800 | [diff] [blame] | 5023 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
Tejun Heo | 7d172cc | 2014-11-18 02:49:51 -0500 | [diff] [blame] | 5024 | if (ss->css_released) |
| 5025 | ss->css_released(css); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5026 | } else { |
| 5027 | /* cgroup release path */ |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5028 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
| 5029 | cgrp->id = -1; |
Li Zefan | a418948 | 2014-09-04 14:43:07 +0800 | [diff] [blame] | 5030 | |
| 5031 | /* |
| 5032 | * There are two control paths which try to determine |
| 5033 | * cgroup from dentry without going through kernfs - |
| 5034 | * cgroupstats_build() and css_tryget_online_from_dir(). |
| 5035 | * Those are supported by RCU protecting clearing of |
| 5036 | * cgrp->kn->priv backpointer. |
| 5037 | */ |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5038 | if (cgrp->kn) |
| 5039 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, |
| 5040 | NULL); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5041 | } |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5042 | |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5043 | mutex_unlock(&cgroup_mutex); |
| 5044 | |
Tejun Heo | 0c21ead | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5045 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5046 | } |
| 5047 | |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 5048 | static void css_release(struct percpu_ref *ref) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5049 | { |
| 5050 | struct cgroup_subsys_state *css = |
| 5051 | container_of(ref, struct cgroup_subsys_state, refcnt); |
| 5052 | |
Tejun Heo | 25e15d8 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5053 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
| 5054 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5055 | } |
| 5056 | |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5057 | static void init_and_link_css(struct cgroup_subsys_state *css, |
| 5058 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5059 | { |
Tejun Heo | 0cb51d7 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5060 | lockdep_assert_held(&cgroup_mutex); |
| 5061 | |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5062 | cgroup_get(cgrp); |
| 5063 | |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5064 | memset(css, 0, sizeof(*css)); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 5065 | css->cgroup = cgrp; |
Tejun Heo | 72c97e5 | 2013-08-08 20:11:22 -0400 | [diff] [blame] | 5066 | css->ss = ss; |
Tejun Heo | 8fa3b8d | 2016-05-26 15:42:13 -0400 | [diff] [blame] | 5067 | css->id = -1; |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5068 | INIT_LIST_HEAD(&css->sibling); |
| 5069 | INIT_LIST_HEAD(&css->children); |
Tejun Heo | 0cb51d7 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5070 | css->serial_nr = css_serial_nr_next++; |
Tejun Heo | aa226ff | 2016-01-21 15:31:11 -0500 | [diff] [blame] | 5071 | atomic_set(&css->online_cnt, 0); |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 5072 | |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5073 | if (cgroup_parent(cgrp)) { |
| 5074 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5075 | css_get(css->parent); |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5076 | } |
Tejun Heo | 0ae78e0 | 2013-08-13 11:01:54 -0400 | [diff] [blame] | 5077 | |
Tejun Heo | ca8bdca | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 5078 | BUG_ON(cgroup_css(cgrp, ss)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5079 | } |
| 5080 | |
Li Zefan | 2a4ac63 | 2013-07-31 16:16:40 +0800 | [diff] [blame] | 5081 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
Tejun Heo | 623f926 | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 5082 | static int online_css(struct cgroup_subsys_state *css) |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5083 | { |
Tejun Heo | 623f926 | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 5084 | struct cgroup_subsys *ss = css->ss; |
Tejun Heo | b1929db | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5085 | int ret = 0; |
| 5086 | |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5087 | lockdep_assert_held(&cgroup_mutex); |
| 5088 | |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5089 | if (ss->css_online) |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5090 | ret = ss->css_online(css); |
Tejun Heo | ae7f164 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5091 | if (!ret) { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5092 | css->flags |= CSS_ONLINE; |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5093 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
Tejun Heo | aa226ff | 2016-01-21 15:31:11 -0500 | [diff] [blame] | 5094 | |
| 5095 | atomic_inc(&css->online_cnt); |
| 5096 | if (css->parent) |
| 5097 | atomic_inc(&css->parent->online_cnt); |
Tejun Heo | ae7f164 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5098 | } |
Tejun Heo | b1929db | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5099 | return ret; |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5100 | } |
| 5101 | |
Li Zefan | 2a4ac63 | 2013-07-31 16:16:40 +0800 | [diff] [blame] | 5102 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
Tejun Heo | 623f926 | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 5103 | static void offline_css(struct cgroup_subsys_state *css) |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5104 | { |
Tejun Heo | 623f926 | 2013-08-13 11:01:55 -0400 | [diff] [blame] | 5105 | struct cgroup_subsys *ss = css->ss; |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5106 | |
| 5107 | lockdep_assert_held(&cgroup_mutex); |
| 5108 | |
| 5109 | if (!(css->flags & CSS_ONLINE)) |
| 5110 | return; |
| 5111 | |
Vladimir Davydov | fa06235 | 2016-03-01 19:56:30 +0300 | [diff] [blame] | 5112 | if (ss->css_reset) |
| 5113 | ss->css_reset(css); |
| 5114 | |
Li Zefan | d7eeac1 | 2013-03-12 15:35:59 -0700 | [diff] [blame] | 5115 | if (ss->css_offline) |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5116 | ss->css_offline(css); |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5117 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5118 | css->flags &= ~CSS_ONLINE; |
Tejun Heo | e329780 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 5119 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
Tejun Heo | f8f22e5 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 5120 | |
| 5121 | wake_up_all(&css->cgroup->offline_waitq); |
Tejun Heo | a31f2d3 | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5122 | } |
| 5123 | |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5124 | /** |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5125 | * css_create - create a cgroup_subsys_state |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5126 | * @cgrp: the cgroup new css will be associated with |
| 5127 | * @ss: the subsys of new css |
| 5128 | * |
| 5129 | * Create a new css associated with @cgrp - @ss pair. On success, the new |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5130 | * css is online and installed in @cgrp. This function doesn't create the |
| 5131 | * interface files. Returns 0 on success, -errno on failure. |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5132 | */ |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5133 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
| 5134 | struct cgroup_subsys *ss) |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5135 | { |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5136 | struct cgroup *parent = cgroup_parent(cgrp); |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5137 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5138 | struct cgroup_subsys_state *css; |
| 5139 | int err; |
| 5140 | |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5141 | lockdep_assert_held(&cgroup_mutex); |
| 5142 | |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5143 | css = ss->css_alloc(parent_css); |
Tejun Heo | e7e15b8 | 2016-06-21 13:06:24 -0400 | [diff] [blame] | 5144 | if (!css) |
| 5145 | css = ERR_PTR(-ENOMEM); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5146 | if (IS_ERR(css)) |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5147 | return css; |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5148 | |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5149 | init_and_link_css(css, ss, cgrp); |
Tejun Heo | a2bed82 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5150 | |
Tejun Heo | 2aad2a8 | 2014-09-24 13:31:50 -0400 | [diff] [blame] | 5151 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5152 | if (err) |
Li Zefan | 3eb59ec | 2014-03-18 17:02:36 +0800 | [diff] [blame] | 5153 | goto err_free_css; |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5154 | |
Vladimir Davydov | cf780b7 | 2015-08-03 15:32:26 +0300 | [diff] [blame] | 5155 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5156 | if (err < 0) |
Wenwei Tao | b00c52d | 2016-05-13 22:59:20 +0800 | [diff] [blame] | 5157 | goto err_free_css; |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5158 | css->id = err; |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5159 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5160 | /* @css is ready to be brought online now, make it visible */ |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5161 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5162 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5163 | |
| 5164 | err = online_css(css); |
| 5165 | if (err) |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5166 | goto err_list_del; |
Tejun Heo | 9441962 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5167 | |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5168 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5169 | cgroup_parent(parent)) { |
Joe Perches | ed3d261 | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 5170 | pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", |
Jianyu Zhan | a2a1f9e | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 5171 | current->comm, current->pid, ss->name); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5172 | if (!strcmp(ss->name, "memory")) |
Joe Perches | ed3d261 | 2014-04-25 18:28:03 -0400 | [diff] [blame] | 5173 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5174 | ss->warned_broken_hierarchy = true; |
| 5175 | } |
| 5176 | |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5177 | return css; |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5178 | |
Tejun Heo | 1fed1b2 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5179 | err_list_del: |
| 5180 | list_del_rcu(&css->sibling); |
Li Zefan | 3eb59ec | 2014-03-18 17:02:36 +0800 | [diff] [blame] | 5181 | err_free_css: |
Tejun Heo | a2bed82 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5182 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
Tejun Heo | 6cd0f5b | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5183 | return ERR_PTR(err); |
Tejun Heo | c81c925a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5184 | } |
| 5185 | |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5186 | static struct cgroup *cgroup_create(struct cgroup *parent) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5187 | { |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5188 | struct cgroup_root *root = parent->root; |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5189 | struct cgroup *cgrp, *tcgrp; |
| 5190 | int level = parent->level + 1; |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5191 | int ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5192 | |
Tejun Heo | 0a950f6 | 2012-11-19 09:02:12 -0800 | [diff] [blame] | 5193 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
Tejun Heo | b11cfb5 | 2015-11-20 15:55:52 -0500 | [diff] [blame] | 5194 | cgrp = kzalloc(sizeof(*cgrp) + |
| 5195 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5196 | if (!cgrp) |
| 5197 | return ERR_PTR(-ENOMEM); |
Li Zefan | 0ab02ca | 2014-02-11 16:05:46 +0800 | [diff] [blame] | 5198 | |
Tejun Heo | 2aad2a8 | 2014-09-24 13:31:50 -0400 | [diff] [blame] | 5199 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5200 | if (ret) |
| 5201 | goto out_free_cgrp; |
| 5202 | |
Li Zefan | 0ab02ca | 2014-02-11 16:05:46 +0800 | [diff] [blame] | 5203 | /* |
| 5204 | * Temporarily set the pointer to NULL, so idr_find() won't return |
| 5205 | * a half-baked cgroup. |
| 5206 | */ |
Vladimir Davydov | cf780b7 | 2015-08-03 15:32:26 +0300 | [diff] [blame] | 5207 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
Li Zefan | 0ab02ca | 2014-02-11 16:05:46 +0800 | [diff] [blame] | 5208 | if (cgrp->id < 0) { |
Tejun Heo | ba0f4d7 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5209 | ret = -ENOMEM; |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5210 | goto out_cancel_ref; |
Tejun Heo | 976c06b | 2012-11-05 09:16:59 -0800 | [diff] [blame] | 5211 | } |
| 5212 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 5213 | init_cgroup_housekeeping(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5214 | |
Tejun Heo | 9d800df | 2014-05-14 09:15:00 -0400 | [diff] [blame] | 5215 | cgrp->self.parent = &parent->self; |
Tejun Heo | ba0f4d7 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5216 | cgrp->root = root; |
Tejun Heo | b11cfb5 | 2015-11-20 15:55:52 -0500 | [diff] [blame] | 5217 | cgrp->level = level; |
| 5218 | |
| 5219 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) |
| 5220 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5221 | |
Li Zefan | b6abdb0 | 2008-03-04 14:28:19 -0800 | [diff] [blame] | 5222 | if (notify_on_release(parent)) |
| 5223 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
| 5224 | |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5225 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
| 5226 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 5227 | |
Tejun Heo | 0cb51d7 | 2014-05-16 13:22:49 -0400 | [diff] [blame] | 5228 | cgrp->self.serial_nr = css_serial_nr_next++; |
Tejun Heo | 53fa526 | 2013-05-24 10:55:38 +0900 | [diff] [blame] | 5229 | |
Tejun Heo | 4e139af | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 5230 | /* allocation complete, commit to creation */ |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5231 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
Tejun Heo | 3c9c825 | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 5232 | atomic_inc(&root->nr_cgrps); |
Tejun Heo | 59f5296 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5233 | cgroup_get(parent); |
Li Zefan | 415cf07 | 2013-04-08 14:35:02 +0800 | [diff] [blame] | 5234 | |
Tejun Heo | 0d80255 | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5235 | /* |
| 5236 | * @cgrp is now fully operational. If something fails after this |
| 5237 | * point, it'll be released via the normal destruction path. |
| 5238 | */ |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 5239 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
Li Zefan | 4e96ee8e | 2013-07-31 09:50:50 +0800 | [diff] [blame] | 5240 | |
Tejun Heo | bd53d61 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 5241 | /* |
| 5242 | * On the default hierarchy, a child doesn't automatically inherit |
Tejun Heo | 667c249 | 2014-07-08 18:02:56 -0400 | [diff] [blame] | 5243 | * subtree_control from the parent. Each is configured manually. |
Tejun Heo | bd53d61 | 2014-04-23 11:13:16 -0400 | [diff] [blame] | 5244 | */ |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5245 | if (!cgroup_on_dfl(cgrp)) |
Tejun Heo | 5531dc9 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5246 | cgrp->subtree_control = cgroup_control(cgrp); |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5247 | |
| 5248 | cgroup_propagate_control(cgrp); |
| 5249 | |
| 5250 | /* @cgrp doesn't have dir yet so the following will only create csses */ |
| 5251 | ret = cgroup_apply_control_enable(cgrp); |
| 5252 | if (ret) |
| 5253 | goto out_destroy; |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 5254 | |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5255 | return cgrp; |
| 5256 | |
| 5257 | out_cancel_ref: |
| 5258 | percpu_ref_exit(&cgrp->self.refcnt); |
| 5259 | out_free_cgrp: |
| 5260 | kfree(cgrp); |
| 5261 | return ERR_PTR(ret); |
| 5262 | out_destroy: |
| 5263 | cgroup_destroy_locked(cgrp); |
| 5264 | return ERR_PTR(ret); |
| 5265 | } |
| 5266 | |
| 5267 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
| 5268 | umode_t mode) |
| 5269 | { |
| 5270 | struct cgroup *parent, *cgrp; |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5271 | struct kernfs_node *kn; |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5272 | int ret; |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5273 | |
| 5274 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ |
| 5275 | if (strchr(name, '\n')) |
| 5276 | return -EINVAL; |
| 5277 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5278 | parent = cgroup_kn_lock_live(parent_kn, false); |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5279 | if (!parent) |
| 5280 | return -ENODEV; |
| 5281 | |
| 5282 | cgrp = cgroup_create(parent); |
| 5283 | if (IS_ERR(cgrp)) { |
| 5284 | ret = PTR_ERR(cgrp); |
| 5285 | goto out_unlock; |
| 5286 | } |
| 5287 | |
Tejun Heo | 195e9b6c | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5288 | /* create the directory */ |
| 5289 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
| 5290 | if (IS_ERR(kn)) { |
| 5291 | ret = PTR_ERR(kn); |
| 5292 | goto out_destroy; |
| 5293 | } |
| 5294 | cgrp->kn = kn; |
| 5295 | |
| 5296 | /* |
| 5297 | * This extra ref will be put in cgroup_free_fn() and guarantees |
| 5298 | * that @cgrp->kn is always accessible. |
| 5299 | */ |
| 5300 | kernfs_get(kn); |
| 5301 | |
| 5302 | ret = cgroup_kn_set_ugid(kn); |
| 5303 | if (ret) |
| 5304 | goto out_destroy; |
| 5305 | |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 5306 | ret = css_populate_dir(&cgrp->self); |
Tejun Heo | 195e9b6c | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5307 | if (ret) |
| 5308 | goto out_destroy; |
| 5309 | |
Tejun Heo | 03970d3 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5310 | ret = cgroup_apply_control_enable(cgrp); |
| 5311 | if (ret) |
| 5312 | goto out_destroy; |
Tejun Heo | 195e9b6c | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5313 | |
| 5314 | /* let's create and online css's */ |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5315 | kernfs_activate(kn); |
| 5316 | |
Tejun Heo | ba0f4d7 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5317 | ret = 0; |
| 5318 | goto out_unlock; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5319 | |
Tejun Heo | a5bca21 | 2016-03-03 09:57:58 -0500 | [diff] [blame] | 5320 | out_destroy: |
| 5321 | cgroup_destroy_locked(cgrp); |
Tejun Heo | ba0f4d7 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5322 | out_unlock: |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5323 | cgroup_kn_unlock(parent_kn); |
Tejun Heo | e1b2dc1 | 2014-03-20 11:10:15 -0400 | [diff] [blame] | 5324 | return ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5325 | } |
| 5326 | |
Tejun Heo | 223dbc3 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5327 | /* |
| 5328 | * This is called when the refcnt of a css is confirmed to be killed. |
Tejun Heo | 249f346 | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 5329 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
| 5330 | * initate destruction and put the css ref from kill_css(). |
Tejun Heo | 223dbc3 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5331 | */ |
| 5332 | static void css_killed_work_fn(struct work_struct *work) |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5333 | { |
Tejun Heo | 223dbc3 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5334 | struct cgroup_subsys_state *css = |
| 5335 | container_of(work, struct cgroup_subsys_state, destroy_work); |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5336 | |
Tejun Heo | f20104d | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5337 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 09a503ea | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5338 | |
Tejun Heo | aa226ff | 2016-01-21 15:31:11 -0500 | [diff] [blame] | 5339 | do { |
| 5340 | offline_css(css); |
| 5341 | css_put(css); |
| 5342 | /* @css can't go away while we're holding cgroup_mutex */ |
| 5343 | css = css->parent; |
| 5344 | } while (css && atomic_dec_and_test(&css->online_cnt)); |
| 5345 | |
| 5346 | mutex_unlock(&cgroup_mutex); |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5347 | } |
| 5348 | |
Tejun Heo | 223dbc3 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5349 | /* css kill confirmation processing requires process context, bounce */ |
| 5350 | static void css_killed_ref_fn(struct percpu_ref *ref) |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5351 | { |
| 5352 | struct cgroup_subsys_state *css = |
| 5353 | container_of(ref, struct cgroup_subsys_state, refcnt); |
| 5354 | |
Tejun Heo | aa226ff | 2016-01-21 15:31:11 -0500 | [diff] [blame] | 5355 | if (atomic_dec_and_test(&css->online_cnt)) { |
| 5356 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
| 5357 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
| 5358 | } |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5359 | } |
| 5360 | |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 5361 | /** |
| 5362 | * kill_css - destroy a css |
| 5363 | * @css: css to destroy |
| 5364 | * |
| 5365 | * This function initiates destruction of @css by removing cgroup interface |
| 5366 | * files and putting its base reference. ->css_offline() will be invoked |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 5367 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
| 5368 | * the reference count reaches zero, @css will be released. |
Tejun Heo | f392e51 | 2014-04-23 11:13:14 -0400 | [diff] [blame] | 5369 | */ |
| 5370 | static void kill_css(struct cgroup_subsys_state *css) |
Tejun Heo | edae0c3 | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5371 | { |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 5372 | lockdep_assert_held(&cgroup_mutex); |
Tejun Heo | 9441962 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5373 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5374 | /* |
| 5375 | * This must happen before css is disassociated with its cgroup. |
| 5376 | * See seq_css() for details. |
| 5377 | */ |
Tejun Heo | 334c367 | 2016-03-03 09:58:01 -0500 | [diff] [blame] | 5378 | css_clear_dir(css); |
Tejun Heo | 3c14f8b | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5379 | |
Tejun Heo | edae0c3 | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5380 | /* |
| 5381 | * Killing would put the base ref, but we need to keep it alive |
| 5382 | * until after ->css_offline(). |
| 5383 | */ |
| 5384 | css_get(css); |
| 5385 | |
| 5386 | /* |
| 5387 | * cgroup core guarantees that, by the time ->css_offline() is |
| 5388 | * invoked, no new css reference will be given out via |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 5389 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
Tejun Heo | edae0c3 | 2013-08-13 20:22:51 -0400 | [diff] [blame] | 5390 | * proceed to offlining css's because percpu_ref_kill() doesn't |
| 5391 | * guarantee that the ref is seen as killed on all CPUs on return. |
| 5392 | * |
| 5393 | * Use percpu_ref_kill_and_confirm() to get notifications as each |
| 5394 | * css is confirmed to be seen as killed on all CPUs. |
| 5395 | */ |
| 5396 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5397 | } |
| 5398 | |
| 5399 | /** |
| 5400 | * cgroup_destroy_locked - the first stage of cgroup destruction |
| 5401 | * @cgrp: cgroup to be destroyed |
| 5402 | * |
| 5403 | * css's make use of percpu refcnts whose killing latency shouldn't be |
| 5404 | * exposed to userland and are RCU protected. Also, cgroup core needs to |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 5405 | * guarantee that css_tryget_online() won't succeed by the time |
| 5406 | * ->css_offline() is invoked. To satisfy all the requirements, |
| 5407 | * destruction is implemented in the following two steps. |
Tejun Heo | d3daf28 | 2013-06-13 19:39:16 -0700 | [diff] [blame] | 5408 | * |
| 5409 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all |
| 5410 | * userland visible parts and start killing the percpu refcnts of |
| 5411 | * css's. Set up so that the next stage will be kicked off once all |
| 5412 | * the percpu refcnts are confirmed to be killed. |
| 5413 | * |
| 5414 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the |
| 5415 | * rest of destruction. Once all cgroup references are gone, the |
| 5416 | * cgroup is RCU-freed. |
| 5417 | * |
| 5418 | * This function implements s1. After this step, @cgrp is gone as far as |
| 5419 | * the userland is concerned and a new cgroup with the same name may be |
| 5420 | * created. As cgroup doesn't care about the names internally, this |
| 5421 | * doesn't cause any problem. |
| 5422 | */ |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5423 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
| 5424 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5425 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5426 | struct cgroup_subsys_state *css; |
Tejun Heo | 2b021cb | 2016-03-15 20:43:04 -0400 | [diff] [blame] | 5427 | struct cgrp_cset_link *link; |
Tejun Heo | 1c6727a | 2013-12-06 15:11:56 -0500 | [diff] [blame] | 5428 | int ssid; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5429 | |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5430 | lockdep_assert_held(&cgroup_mutex); |
| 5431 | |
Tejun Heo | 91486f6 | 2015-10-15 16:41:51 -0400 | [diff] [blame] | 5432 | /* |
| 5433 | * Only migration can raise populated from zero and we're already |
| 5434 | * holding cgroup_mutex. |
| 5435 | */ |
| 5436 | if (cgroup_is_populated(cgrp)) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5437 | return -EBUSY; |
Tejun Heo | ed957793 | 2012-11-05 09:16:58 -0800 | [diff] [blame] | 5438 | |
Tejun Heo | 1a90dd5 | 2012-11-05 09:16:59 -0800 | [diff] [blame] | 5439 | /* |
Tejun Heo | d5c419b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5440 | * Make sure there's no live children. We can't test emptiness of |
| 5441 | * ->self.children as dead children linger on it while being |
| 5442 | * drained; otherwise, "rmdir parent/child parent" may fail. |
Hugh Dickins | bb78a92 | 2013-08-28 16:31:23 -0700 | [diff] [blame] | 5443 | */ |
Tejun Heo | f3d4650 | 2014-05-16 13:22:52 -0400 | [diff] [blame] | 5444 | if (css_has_online_children(&cgrp->self)) |
Hugh Dickins | bb78a92 | 2013-08-28 16:31:23 -0700 | [diff] [blame] | 5445 | return -EBUSY; |
| 5446 | |
| 5447 | /* |
Tejun Heo | 2b021cb | 2016-03-15 20:43:04 -0400 | [diff] [blame] | 5448 | * Mark @cgrp and the associated csets dead. The former prevents |
| 5449 | * further task migration and child creation by disabling |
| 5450 | * cgroup_lock_live_group(). The latter makes the csets ignored by |
| 5451 | * the migration path. |
Tejun Heo | 455050d | 2013-06-13 19:27:41 -0700 | [diff] [blame] | 5452 | */ |
Tejun Heo | 184faf3 | 2014-05-16 13:22:51 -0400 | [diff] [blame] | 5453 | cgrp->self.flags &= ~CSS_ONLINE; |
Tejun Heo | 1a90dd5 | 2012-11-05 09:16:59 -0800 | [diff] [blame] | 5454 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5455 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 2b021cb | 2016-03-15 20:43:04 -0400 | [diff] [blame] | 5456 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
| 5457 | link->cset->dead = true; |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5458 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 2b021cb | 2016-03-15 20:43:04 -0400 | [diff] [blame] | 5459 | |
Tejun Heo | 249f346 | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 5460 | /* initiate massacre of all css's */ |
Tejun Heo | 1a90dd5 | 2012-11-05 09:16:59 -0800 | [diff] [blame] | 5461 | for_each_css(css, ssid, cgrp) |
Tejun Heo | 455050d | 2013-06-13 19:27:41 -0700 | [diff] [blame] | 5462 | kill_css(css); |
| 5463 | |
Tejun Heo | 455050d | 2013-06-13 19:27:41 -0700 | [diff] [blame] | 5464 | /* |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 5465 | * Remove @cgrp directory along with the base files. @cgrp has an |
| 5466 | * extra ref on its kn. |
Tejun Heo | 455050d | 2013-06-13 19:27:41 -0700 | [diff] [blame] | 5467 | */ |
Tejun Heo | 01f6474 | 2014-05-13 12:19:23 -0400 | [diff] [blame] | 5468 | kernfs_remove(cgrp->kn); |
Tejun Heo | f20104d | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5469 | |
Tejun Heo | d51f39b | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 5470 | check_for_release(cgroup_parent(cgrp)); |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5471 | |
Tejun Heo | 249f346 | 2014-05-14 09:15:01 -0400 | [diff] [blame] | 5472 | /* put the base reference */ |
Tejun Heo | 9d755d3 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5473 | percpu_ref_kill(&cgrp->self.refcnt); |
Tejun Heo | 455050d | 2013-06-13 19:27:41 -0700 | [diff] [blame] | 5474 | |
Tejun Heo | ea15f8c | 2013-06-13 19:27:42 -0700 | [diff] [blame] | 5475 | return 0; |
| 5476 | }; |
| 5477 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5478 | static int cgroup_rmdir(struct kernfs_node *kn) |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5479 | { |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5480 | struct cgroup *cgrp; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5481 | int ret = 0; |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5482 | |
Tejun Heo | 945ba19 | 2016-03-03 09:58:00 -0500 | [diff] [blame] | 5483 | cgrp = cgroup_kn_lock_live(kn, false); |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5484 | if (!cgrp) |
| 5485 | return 0; |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5486 | |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5487 | ret = cgroup_destroy_locked(cgrp); |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5488 | |
Tejun Heo | a9746d8 | 2014-05-13 12:19:22 -0400 | [diff] [blame] | 5489 | cgroup_kn_unlock(kn); |
Tejun Heo | 42809dd | 2012-11-19 08:13:37 -0800 | [diff] [blame] | 5490 | return ret; |
| 5491 | } |
| 5492 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5493 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
| 5494 | .remount_fs = cgroup_remount, |
| 5495 | .show_options = cgroup_show_options, |
| 5496 | .mkdir = cgroup_mkdir, |
| 5497 | .rmdir = cgroup_rmdir, |
| 5498 | .rename = cgroup_rename, |
Serge E. Hallyn | 4f41fc5 | 2016-05-09 09:59:55 -0500 | [diff] [blame] | 5499 | .show_path = cgroup_show_path, |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5500 | }; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 5501 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5502 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5503 | { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5504 | struct cgroup_subsys_state *css; |
Diego Calleja | cfe36bd | 2007-11-14 16:58:54 -0800 | [diff] [blame] | 5505 | |
Tejun Heo | a5ae989 | 2015-12-29 14:53:56 -0500 | [diff] [blame] | 5506 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5507 | |
Tejun Heo | 648bb56 | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 5508 | mutex_lock(&cgroup_mutex); |
| 5509 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5510 | idr_init(&ss->css_idr); |
Tejun Heo | 0adb070 | 2014-02-12 09:29:48 -0500 | [diff] [blame] | 5511 | INIT_LIST_HEAD(&ss->cfts); |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 5512 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5513 | /* Create the root cgroup state for this subsystem */ |
| 5514 | ss->root = &cgrp_dfl_root; |
| 5515 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5516 | /* We don't handle early failures gracefully */ |
| 5517 | BUG_ON(IS_ERR(css)); |
Tejun Heo | ddfcada | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5518 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
Tejun Heo | 3b514d2 | 2014-05-16 13:22:47 -0400 | [diff] [blame] | 5519 | |
| 5520 | /* |
| 5521 | * Root csses are never destroyed and we can't initialize |
| 5522 | * percpu_ref during early init. Disable refcnting. |
| 5523 | */ |
| 5524 | css->flags |= CSS_NO_REF; |
| 5525 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5526 | if (early) { |
Tejun Heo | 9395a45 | 2014-05-14 09:15:02 -0400 | [diff] [blame] | 5527 | /* allocation can't be done safely during early init */ |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5528 | css->id = 1; |
| 5529 | } else { |
| 5530 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); |
| 5531 | BUG_ON(css->id < 0); |
| 5532 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5533 | |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 5534 | /* Update the init_css_set to contain a subsys |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5535 | * pointer to this state - since the subsystem is |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 5536 | * newly registered, all tasks and hence the |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5537 | * init_css_set is in the subsystem's root cgroup. */ |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5538 | init_css_set.subsys[ss->id] = css; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5539 | |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 5540 | have_fork_callback |= (bool)ss->fork << ss->id; |
| 5541 | have_exit_callback |= (bool)ss->exit << ss->id; |
Tejun Heo | afcf6c8 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5542 | have_free_callback |= (bool)ss->free << ss->id; |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5543 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5544 | |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 5545 | /* At system boot, before all subsystems have been |
| 5546 | * registered, no tasks have been forked, so we don't |
| 5547 | * need to invoke fork callbacks here. */ |
| 5548 | BUG_ON(!list_empty(&init_task.tasks)); |
| 5549 | |
Tejun Heo | ae7f164 | 2013-08-13 20:22:50 -0400 | [diff] [blame] | 5550 | BUG_ON(online_css(css)); |
Tejun Heo | a863803 | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 5551 | |
Tejun Heo | 648bb56 | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 5552 | mutex_unlock(&cgroup_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5553 | } |
| 5554 | |
| 5555 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 5556 | * cgroup_init_early - cgroup initialization at system boot |
| 5557 | * |
| 5558 | * Initialize cgroups at system boot, and initialize any |
| 5559 | * subsystems that request early init. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5560 | */ |
| 5561 | int __init cgroup_init_early(void) |
| 5562 | { |
Tejun Heo | 7b9a6ba | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5563 | static struct cgroup_sb_opts __initdata opts; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5564 | struct cgroup_subsys *ss; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5565 | int i; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5566 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5567 | init_cgroup_root(&cgrp_dfl_root, &opts); |
Tejun Heo | 3b514d2 | 2014-05-16 13:22:47 -0400 | [diff] [blame] | 5568 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
| 5569 | |
Tejun Heo | a4ea1cc | 2013-06-21 15:52:33 -0700 | [diff] [blame] | 5570 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5571 | |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5572 | for_each_subsys(ss, i) { |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5573 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
Xiubo Li | 63253ad | 2016-02-26 13:07:38 +0800 | [diff] [blame] | 5574 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5575 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5576 | ss->id, ss->name); |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5577 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
| 5578 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5579 | |
Tejun Heo | aec2502 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5580 | ss->id = i; |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5581 | ss->name = cgroup_subsys_name[i]; |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 5582 | if (!ss->legacy_name) |
| 5583 | ss->legacy_name = cgroup_subsys_name[i]; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5584 | |
| 5585 | if (ss->early_init) |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5586 | cgroup_init_subsys(ss, true); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5587 | } |
| 5588 | return 0; |
| 5589 | } |
| 5590 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 5591 | static u16 cgroup_disable_mask __initdata; |
Tejun Heo | a3e7273 | 2015-09-25 16:24:27 -0400 | [diff] [blame] | 5592 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5593 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 5594 | * cgroup_init - cgroup initialization |
| 5595 | * |
| 5596 | * Register cgroup filesystem and /proc file, and initialize |
| 5597 | * any subsystems that didn't request early init. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5598 | */ |
| 5599 | int __init cgroup_init(void) |
| 5600 | { |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5601 | struct cgroup_subsys *ss; |
Tejun Heo | 035f4f5 | 2015-10-15 17:00:43 -0400 | [diff] [blame] | 5602 | int ssid; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5603 | |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 5604 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
Tejun Heo | 1ed1328 | 2015-09-16 12:53:17 -0400 | [diff] [blame] | 5605 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
Tejun Heo | a14c687 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 5606 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
| 5607 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5608 | |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 5609 | get_user_ns(init_cgroup_ns.user_ns); |
| 5610 | |
Tejun Heo | 54e7b4e | 2013-04-14 11:36:57 -0700 | [diff] [blame] | 5611 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 54e7b4e | 2013-04-14 11:36:57 -0700 | [diff] [blame] | 5612 | |
Tejun Heo | 2378d8b | 2016-03-03 09:57:57 -0500 | [diff] [blame] | 5613 | /* |
| 5614 | * Add init_css_set to the hash table so that dfl_root can link to |
| 5615 | * it during init. |
| 5616 | */ |
| 5617 | hash_add(css_set_table, &init_css_set.hlist, |
| 5618 | css_set_hash(init_css_set.subsys)); |
Tejun Heo | 82fe9b0 | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5619 | |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5620 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
Greg KH | 676db4a | 2010-08-05 13:53:35 -0700 | [diff] [blame] | 5621 | |
Tejun Heo | 54e7b4e | 2013-04-14 11:36:57 -0700 | [diff] [blame] | 5622 | mutex_unlock(&cgroup_mutex); |
| 5623 | |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5624 | for_each_subsys(ss, ssid) { |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 5625 | if (ss->early_init) { |
| 5626 | struct cgroup_subsys_state *css = |
| 5627 | init_css_set.subsys[ss->id]; |
| 5628 | |
| 5629 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, |
| 5630 | GFP_KERNEL); |
| 5631 | BUG_ON(css->id < 0); |
| 5632 | } else { |
| 5633 | cgroup_init_subsys(ss, false); |
| 5634 | } |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5635 | |
Tejun Heo | 2d8f243 | 2014-04-23 11:13:15 -0400 | [diff] [blame] | 5636 | list_add_tail(&init_css_set.e_cset_node[ssid], |
| 5637 | &cgrp_dfl_root.cgrp.e_csets[ssid]); |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5638 | |
| 5639 | /* |
Li Zefan | c731ae1 | 2014-06-05 17:16:30 +0800 | [diff] [blame] | 5640 | * Setting dfl_root subsys_mask needs to consider the |
| 5641 | * disabled flag and cftype registration needs kmalloc, |
| 5642 | * both of which aren't available during early_init. |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5643 | */ |
Tejun Heo | a3e7273 | 2015-09-25 16:24:27 -0400 | [diff] [blame] | 5644 | if (cgroup_disable_mask & (1 << ssid)) { |
| 5645 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); |
| 5646 | printk(KERN_INFO "Disabling %s control group subsystem\n", |
| 5647 | ss->name); |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 5648 | continue; |
Tejun Heo | a3e7273 | 2015-09-25 16:24:27 -0400 | [diff] [blame] | 5649 | } |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 5650 | |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 5651 | if (cgroup_ssid_no_v1(ssid)) |
| 5652 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
| 5653 | ss->name); |
| 5654 | |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 5655 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
| 5656 | |
Tejun Heo | f6d635ad | 2016-03-08 11:51:26 -0500 | [diff] [blame] | 5657 | if (ss->implicit_on_dfl) |
| 5658 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; |
| 5659 | else if (!ss->dfl_cftypes) |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 5660 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
Tejun Heo | 5de4fa1 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 5661 | |
Tejun Heo | a8ddc82 | 2014-07-15 11:05:10 -0400 | [diff] [blame] | 5662 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
| 5663 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); |
| 5664 | } else { |
| 5665 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); |
| 5666 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); |
Li Zefan | c731ae1 | 2014-06-05 17:16:30 +0800 | [diff] [blame] | 5667 | } |
Vladimir Davydov | 295458e | 2015-02-19 17:34:46 +0300 | [diff] [blame] | 5668 | |
| 5669 | if (ss->bind) |
| 5670 | ss->bind(init_css_set.subsys[ssid]); |
Tejun Heo | 172a2c06 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5671 | } |
Greg KH | 676db4a | 2010-08-05 13:53:35 -0700 | [diff] [blame] | 5672 | |
Tejun Heo | 2378d8b | 2016-03-03 09:57:57 -0500 | [diff] [blame] | 5673 | /* init_css_set.subsys[] has been updated, re-hash */ |
| 5674 | hash_del(&init_css_set.hlist); |
| 5675 | hash_add(css_set_table, &init_css_set.hlist, |
| 5676 | css_set_hash(init_css_set.subsys)); |
| 5677 | |
Tejun Heo | 035f4f5 | 2015-10-15 17:00:43 -0400 | [diff] [blame] | 5678 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
| 5679 | WARN_ON(register_filesystem(&cgroup_fs_type)); |
Tejun Heo | 67e9c74 | 2015-11-16 11:13:34 -0500 | [diff] [blame] | 5680 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
Tejun Heo | 035f4f5 | 2015-10-15 17:00:43 -0400 | [diff] [blame] | 5681 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5682 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 5683 | return 0; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 5684 | } |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5685 | |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 5686 | static int __init cgroup_wq_init(void) |
| 5687 | { |
| 5688 | /* |
| 5689 | * There isn't much point in executing destruction path in |
| 5690 | * parallel. Good chunk is serialized with cgroup_mutex anyway. |
Tejun Heo | 1a11533 | 2014-02-12 19:06:19 -0500 | [diff] [blame] | 5691 | * Use 1 for @max_active. |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 5692 | * |
| 5693 | * We would prefer to do this in cgroup_init() above, but that |
| 5694 | * is called before init_workqueues(): so leave this until after. |
| 5695 | */ |
Tejun Heo | 1a11533 | 2014-02-12 19:06:19 -0500 | [diff] [blame] | 5696 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 5697 | BUG_ON(!cgroup_destroy_wq); |
Tejun Heo | b1a2136 | 2013-11-29 10:42:58 -0500 | [diff] [blame] | 5698 | |
| 5699 | /* |
| 5700 | * Used to destroy pidlists and separate to serve as flush domain. |
| 5701 | * Cap @max_active to 1 too. |
| 5702 | */ |
| 5703 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", |
| 5704 | 0, 1); |
| 5705 | BUG_ON(!cgroup_pidlist_destroy_wq); |
| 5706 | |
Tejun Heo | e5fca24 | 2013-11-22 17:14:39 -0500 | [diff] [blame] | 5707 | return 0; |
| 5708 | } |
| 5709 | core_initcall(cgroup_wq_init); |
| 5710 | |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5711 | /* |
| 5712 | * proc_cgroup_show() |
| 5713 | * - Print task's cgroup paths into seq_file, one line for each hierarchy |
| 5714 | * - Used for /proc/<pid>/cgroup. |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5715 | */ |
Zefan Li | 006f4ac | 2014-09-18 16:03:15 +0800 | [diff] [blame] | 5716 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
| 5717 | struct pid *pid, struct task_struct *tsk) |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5718 | { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 5719 | char *buf, *path; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5720 | int retval; |
Tejun Heo | 3dd06ff | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5721 | struct cgroup_root *root; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5722 | |
| 5723 | retval = -ENOMEM; |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 5724 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5725 | if (!buf) |
| 5726 | goto out; |
| 5727 | |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5728 | mutex_lock(&cgroup_mutex); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5729 | spin_lock_irq(&css_set_lock); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5730 | |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5731 | for_each_root(root) { |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5732 | struct cgroup_subsys *ss; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 5733 | struct cgroup *cgrp; |
Tejun Heo | b85d204 | 2013-12-06 15:11:57 -0500 | [diff] [blame] | 5734 | int ssid, count = 0; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5735 | |
Tejun Heo | a716526 | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 5736 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
Tejun Heo | 985ed67 | 2014-03-19 10:23:53 -0400 | [diff] [blame] | 5737 | continue; |
| 5738 | |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 5739 | seq_printf(m, "%d:", root->hierarchy_id); |
Tejun Heo | d98817d | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 5740 | if (root != &cgrp_dfl_root) |
| 5741 | for_each_subsys(ss, ssid) |
| 5742 | if (root->subsys_mask & (1 << ssid)) |
| 5743 | seq_printf(m, "%s%s", count++ ? "," : "", |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 5744 | ss->legacy_name); |
Paul Menage | c6d57f33 | 2009-09-23 15:56:19 -0700 | [diff] [blame] | 5745 | if (strlen(root->name)) |
| 5746 | seq_printf(m, "%sname=%s", count ? "," : "", |
| 5747 | root->name); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5748 | seq_putc(m, ':'); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5749 | |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 5750 | cgrp = task_cgroup_from_root(tsk, root); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5751 | |
| 5752 | /* |
| 5753 | * On traditional hierarchies, all zombie tasks show up as |
| 5754 | * belonging to the root cgroup. On the default hierarchy, |
| 5755 | * while a zombie doesn't show up in "cgroup.procs" and |
| 5756 | * thus can't be migrated, its /proc/PID/cgroup keeps |
| 5757 | * reporting the cgroup it belonged to before exiting. If |
| 5758 | * the cgroup is removed before the zombie is reaped, |
| 5759 | * " (deleted)" is appended to the cgroup path. |
| 5760 | */ |
| 5761 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 5762 | path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
| 5763 | current->nsproxy->cgroup_ns); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5764 | if (!path) { |
| 5765 | retval = -ENAMETOOLONG; |
| 5766 | goto out_unlock; |
| 5767 | } |
| 5768 | } else { |
| 5769 | path = "/"; |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 5770 | } |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5771 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 5772 | seq_puts(m, path); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5773 | |
| 5774 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
| 5775 | seq_puts(m, " (deleted)\n"); |
| 5776 | else |
| 5777 | seq_putc(m, '\n'); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5778 | } |
| 5779 | |
Zefan Li | 006f4ac | 2014-09-18 16:03:15 +0800 | [diff] [blame] | 5780 | retval = 0; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5781 | out_unlock: |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5782 | spin_unlock_irq(&css_set_lock); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5783 | mutex_unlock(&cgroup_mutex); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5784 | kfree(buf); |
| 5785 | out: |
| 5786 | return retval; |
| 5787 | } |
| 5788 | |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5789 | /* Display information about each subsystem and each hierarchy */ |
| 5790 | static int proc_cgroupstats_show(struct seq_file *m, void *v) |
| 5791 | { |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5792 | struct cgroup_subsys *ss; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5793 | int i; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5794 | |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 5795 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
Ben Blum | aae8aab | 2010-03-10 15:22:07 -0800 | [diff] [blame] | 5796 | /* |
| 5797 | * ideally we don't want subsystems moving around while we do this. |
| 5798 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of |
| 5799 | * subsys/hierarchy state. |
| 5800 | */ |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5801 | mutex_lock(&cgroup_mutex); |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5802 | |
| 5803 | for_each_subsys(ss, i) |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 5804 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 5805 | ss->legacy_name, ss->root->hierarchy_id, |
Tejun Heo | fc5ed1e | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 5806 | atomic_read(&ss->root->nr_cgrps), |
| 5807 | cgroup_ssid_enabled(i)); |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5808 | |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5809 | mutex_unlock(&cgroup_mutex); |
| 5810 | return 0; |
| 5811 | } |
| 5812 | |
| 5813 | static int cgroupstats_open(struct inode *inode, struct file *file) |
| 5814 | { |
Al Viro | 9dce07f | 2008-03-29 03:07:28 +0000 | [diff] [blame] | 5815 | return single_open(file, proc_cgroupstats_show, NULL); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5816 | } |
| 5817 | |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 5818 | static const struct file_operations proc_cgroupstats_operations = { |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 5819 | .open = cgroupstats_open, |
| 5820 | .read = seq_read, |
| 5821 | .llseek = seq_lseek, |
| 5822 | .release = single_release, |
| 5823 | }; |
| 5824 | |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5825 | /** |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5826 | * cgroup_fork - initialize cgroup related fields during copy_process() |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 5827 | * @child: pointer to task_struct of forking parent process. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5828 | * |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5829 | * A task is associated with the init_css_set until cgroup_post_fork() |
| 5830 | * attaches it to the parent's css_set. Empty cg_list indicates that |
| 5831 | * @child isn't holding reference to its css_set. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5832 | */ |
| 5833 | void cgroup_fork(struct task_struct *child) |
| 5834 | { |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5835 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5836 | INIT_LIST_HEAD(&child->cg_list); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5837 | } |
| 5838 | |
| 5839 | /** |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5840 | * cgroup_can_fork - called on a new task before the process is exposed |
| 5841 | * @child: the task in question. |
| 5842 | * |
| 5843 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback |
| 5844 | * returns an error, the fork aborts with that error code. This allows for |
| 5845 | * a cgroup subsystem to conditionally allow or deny new forks. |
| 5846 | */ |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5847 | int cgroup_can_fork(struct task_struct *child) |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5848 | { |
| 5849 | struct cgroup_subsys *ss; |
| 5850 | int i, j, ret; |
| 5851 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5852 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5853 | ret = ss->can_fork(child); |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5854 | if (ret) |
| 5855 | goto out_revert; |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5856 | } while_each_subsys_mask(); |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5857 | |
| 5858 | return 0; |
| 5859 | |
| 5860 | out_revert: |
| 5861 | for_each_subsys(ss, j) { |
| 5862 | if (j >= i) |
| 5863 | break; |
| 5864 | if (ss->cancel_fork) |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5865 | ss->cancel_fork(child); |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5866 | } |
| 5867 | |
| 5868 | return ret; |
| 5869 | } |
| 5870 | |
| 5871 | /** |
| 5872 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() |
| 5873 | * @child: the task in question |
| 5874 | * |
| 5875 | * This calls the cancel_fork() callbacks if a fork failed *after* |
| 5876 | * cgroup_can_fork() succeded. |
| 5877 | */ |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5878 | void cgroup_cancel_fork(struct task_struct *child) |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5879 | { |
| 5880 | struct cgroup_subsys *ss; |
| 5881 | int i; |
| 5882 | |
| 5883 | for_each_subsys(ss, i) |
| 5884 | if (ss->cancel_fork) |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5885 | ss->cancel_fork(child); |
Aleksa Sarai | 7e47682 | 2015-06-09 21:32:09 +1000 | [diff] [blame] | 5886 | } |
| 5887 | |
| 5888 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 5889 | * cgroup_post_fork - called on a new task after adding it to the task list |
| 5890 | * @child: the task in question |
| 5891 | * |
Tejun Heo | 5edee61 | 2012-10-16 15:03:14 -0700 | [diff] [blame] | 5892 | * Adds the task to the list running through its css_set if necessary and |
| 5893 | * call the subsystem fork() callbacks. Has to be after the task is |
| 5894 | * visible on the task list in case we race with the first call to |
Tejun Heo | 0942eee | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 5895 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
Tejun Heo | 5edee61 | 2012-10-16 15:03:14 -0700 | [diff] [blame] | 5896 | * list. |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 5897 | */ |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5898 | void cgroup_post_fork(struct task_struct *child) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5899 | { |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5900 | struct cgroup_subsys *ss; |
Tejun Heo | 5edee61 | 2012-10-16 15:03:14 -0700 | [diff] [blame] | 5901 | int i; |
| 5902 | |
Frederic Weisbecker | 3ce3230 | 2012-02-08 03:37:27 +0100 | [diff] [blame] | 5903 | /* |
Dongsheng Yang | 251f8c0 | 2014-08-25 19:27:52 +0800 | [diff] [blame] | 5904 | * This may race against cgroup_enable_task_cg_lists(). As that |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5905 | * function sets use_task_css_set_links before grabbing |
| 5906 | * tasklist_lock and we just went through tasklist_lock to add |
| 5907 | * @child, it's guaranteed that either we see the set |
| 5908 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees |
| 5909 | * @child during its iteration. |
| 5910 | * |
| 5911 | * If we won the race, @child is associated with %current's |
Tejun Heo | f0d9a5f | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5912 | * css_set. Grabbing css_set_lock guarantees both that the |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5913 | * association is stable, and, on completion of the parent's |
| 5914 | * migration, @child is visible in the source of migration or |
| 5915 | * already in the destination cgroup. This guarantee is necessary |
| 5916 | * when implementing operations which need to migrate all tasks of |
| 5917 | * a cgroup to another. |
| 5918 | * |
Dongsheng Yang | 251f8c0 | 2014-08-25 19:27:52 +0800 | [diff] [blame] | 5919 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5920 | * will remain in init_css_set. This is safe because all tasks are |
| 5921 | * in the init_css_set before cg_links is enabled and there's no |
| 5922 | * operation which transfers all tasks out of init_css_set. |
Frederic Weisbecker | 3ce3230 | 2012-02-08 03:37:27 +0100 | [diff] [blame] | 5923 | */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5924 | if (use_task_css_set_links) { |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5925 | struct css_set *cset; |
| 5926 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5927 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5928 | cset = task_css_set(current); |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5929 | if (list_empty(&child->cg_list)) { |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5930 | get_css_set(cset); |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 5931 | css_set_move_task(child, NULL, cset, false); |
Tejun Heo | eaf797a | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5932 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5933 | spin_unlock_irq(&css_set_lock); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5934 | } |
Tejun Heo | 5edee61 | 2012-10-16 15:03:14 -0700 | [diff] [blame] | 5935 | |
| 5936 | /* |
| 5937 | * Call ss->fork(). This must happen after @child is linked on |
| 5938 | * css_set; otherwise, @child might change state between ->fork() |
| 5939 | * and addition to css_set. |
| 5940 | */ |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5941 | do_each_subsys_mask(ss, i, have_fork_callback) { |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 5942 | ss->fork(child); |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5943 | } while_each_subsys_mask(); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5944 | } |
Tejun Heo | 5edee61 | 2012-10-16 15:03:14 -0700 | [diff] [blame] | 5945 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5946 | /** |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5947 | * cgroup_exit - detach cgroup from exiting task |
| 5948 | * @tsk: pointer to task_struct of exiting process |
| 5949 | * |
| 5950 | * Description: Detach cgroup from @tsk and release it. |
| 5951 | * |
| 5952 | * Note that cgroups marked notify_on_release force every task in |
| 5953 | * them to take the global cgroup_mutex mutex when exiting. |
| 5954 | * This could impact scaling on very large systems. Be reluctant to |
| 5955 | * use notify_on_release cgroups where very high task exit scaling |
| 5956 | * is required on large systems. |
| 5957 | * |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5958 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
| 5959 | * call cgroup_exit() while the task is still competent to handle |
| 5960 | * notify_on_release(), then leave the task attached to the root cgroup in |
| 5961 | * each hierarchy for the remainder of its exit. No need to bother with |
| 5962 | * init_css_set refcnting. init_css_set never goes away and we can't race |
Li Zefan | e8604cb | 2014-03-28 15:18:27 +0800 | [diff] [blame] | 5963 | * with migration path - PF_EXITING is visible to migration path. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5964 | */ |
Li Zefan | 1ec4183 | 2014-03-28 15:22:19 +0800 | [diff] [blame] | 5965 | void cgroup_exit(struct task_struct *tsk) |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 5966 | { |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5967 | struct cgroup_subsys *ss; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 5968 | struct css_set *cset; |
Peter Zijlstra | d41d5a0 | 2011-02-07 17:02:20 +0100 | [diff] [blame] | 5969 | int i; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5970 | |
| 5971 | /* |
Tejun Heo | 0e1d768 | 2014-02-25 10:04:03 -0500 | [diff] [blame] | 5972 | * Unlink from @tsk from its css_set. As migration path can't race |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 5973 | * with us, we can check css_set and cg_list without synchronization. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5974 | */ |
Tejun Heo | 0de0942 | 2015-10-15 16:41:49 -0400 | [diff] [blame] | 5975 | cset = task_css_set(tsk); |
| 5976 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5977 | if (!list_empty(&tsk->cg_list)) { |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5978 | spin_lock_irq(&css_set_lock); |
Tejun Heo | f6d7d04 | 2015-10-15 16:41:52 -0400 | [diff] [blame] | 5979 | css_set_move_task(tsk, cset, NULL, false); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 5980 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5981 | } else { |
| 5982 | get_css_set(cset); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 5983 | } |
| 5984 | |
Aleksa Sarai | cb4a316 | 2015-06-06 10:02:14 +1000 | [diff] [blame] | 5985 | /* see cgroup_post_fork() for details */ |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5986 | do_each_subsys_mask(ss, i, have_exit_callback) { |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5987 | ss->exit(tsk); |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5988 | } while_each_subsys_mask(); |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5989 | } |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 5990 | |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5991 | void cgroup_free(struct task_struct *task) |
| 5992 | { |
| 5993 | struct css_set *cset = task_css_set(task); |
Tejun Heo | afcf6c8 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5994 | struct cgroup_subsys *ss; |
| 5995 | int ssid; |
| 5996 | |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5997 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
Tejun Heo | afcf6c8 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 5998 | ss->free(task); |
Tejun Heo | b4e0eea | 2016-02-22 22:25:46 -0500 | [diff] [blame] | 5999 | } while_each_subsys_mask(); |
Peter Zijlstra | d41d5a0 | 2011-02-07 17:02:20 +0100 | [diff] [blame] | 6000 | |
Tejun Heo | 2e91fa7 | 2015-10-15 16:41:53 -0400 | [diff] [blame] | 6001 | put_css_set(cset); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 6002 | } |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 6003 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 6004 | static void check_for_release(struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6005 | { |
Tejun Heo | 27bd4db | 2015-10-15 16:41:50 -0400 | [diff] [blame] | 6006 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6007 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
| 6008 | schedule_work(&cgrp->release_agent_work); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6009 | } |
| 6010 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6011 | /* |
| 6012 | * Notify userspace when a cgroup is released, by running the |
| 6013 | * configured release agent with the name of the cgroup (path |
| 6014 | * relative to the root of cgroup file system) as the argument. |
| 6015 | * |
| 6016 | * Most likely, this user command will try to rmdir this cgroup. |
| 6017 | * |
| 6018 | * This races with the possibility that some other task will be |
| 6019 | * attached to this cgroup before it is removed, or that some other |
| 6020 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. |
| 6021 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer |
| 6022 | * unused, and this cgroup will be reprieved from its death sentence, |
| 6023 | * to continue to serve a useful existence. Next time it's released, |
| 6024 | * we will get notified again, if it still has 'notify_on_release' set. |
| 6025 | * |
| 6026 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which |
| 6027 | * means only wait until the task is successfully execve()'d. The |
| 6028 | * separate release agent task is forked by call_usermodehelper(), |
| 6029 | * then control in this thread returns here, without waiting for the |
| 6030 | * release agent task. We don't bother to wait because the caller of |
| 6031 | * this routine has no use for the exit status of the release agent |
| 6032 | * task, so no sense holding our caller up for that. |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6033 | */ |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6034 | static void cgroup_release_agent(struct work_struct *work) |
| 6035 | { |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6036 | struct cgroup *cgrp = |
| 6037 | container_of(work, struct cgroup, release_agent_work); |
| 6038 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
| 6039 | char *argv[3], *envp[3]; |
| 6040 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6041 | mutex_lock(&cgroup_mutex); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6042 | |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6043 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
| 6044 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); |
| 6045 | if (!pathbuf || !agentbuf) |
| 6046 | goto out; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6047 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6048 | spin_lock_irq(&css_set_lock); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6049 | path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6050 | spin_unlock_irq(&css_set_lock); |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6051 | if (!path) |
| 6052 | goto out; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6053 | |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6054 | argv[0] = agentbuf; |
| 6055 | argv[1] = path; |
| 6056 | argv[2] = NULL; |
| 6057 | |
| 6058 | /* minimal command environment */ |
| 6059 | envp[0] = "HOME=/"; |
| 6060 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; |
| 6061 | envp[2] = NULL; |
| 6062 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6063 | mutex_unlock(&cgroup_mutex); |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6064 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
Zefan Li | 3e2cd91 | 2014-09-20 14:35:43 +0800 | [diff] [blame] | 6065 | goto out_free; |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6066 | out: |
Zefan Li | 3e2cd91 | 2014-09-20 14:35:43 +0800 | [diff] [blame] | 6067 | mutex_unlock(&cgroup_mutex); |
| 6068 | out_free: |
Zefan Li | 971ff49 | 2014-09-18 16:06:19 +0800 | [diff] [blame] | 6069 | kfree(agentbuf); |
| 6070 | kfree(pathbuf); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 6071 | } |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 6072 | |
| 6073 | static int __init cgroup_disable(char *str) |
| 6074 | { |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 6075 | struct cgroup_subsys *ss; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 6076 | char *token; |
Tejun Heo | 30159ec | 2013-06-25 11:53:37 -0700 | [diff] [blame] | 6077 | int i; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 6078 | |
| 6079 | while ((token = strsep(&str, ",")) != NULL) { |
| 6080 | if (!*token) |
| 6081 | continue; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 6082 | |
Tejun Heo | 3ed80a6 | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 6083 | for_each_subsys(ss, i) { |
Tejun Heo | 3e1d2ee | 2015-08-18 13:58:16 -0700 | [diff] [blame] | 6084 | if (strcmp(token, ss->name) && |
| 6085 | strcmp(token, ss->legacy_name)) |
| 6086 | continue; |
Tejun Heo | a3e7273 | 2015-09-25 16:24:27 -0400 | [diff] [blame] | 6087 | cgroup_disable_mask |= 1 << i; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 6088 | } |
| 6089 | } |
| 6090 | return 1; |
| 6091 | } |
| 6092 | __setup("cgroup_disable=", cgroup_disable); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 6093 | |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 6094 | static int __init cgroup_no_v1(char *str) |
| 6095 | { |
| 6096 | struct cgroup_subsys *ss; |
| 6097 | char *token; |
| 6098 | int i; |
| 6099 | |
| 6100 | while ((token = strsep(&str, ",")) != NULL) { |
| 6101 | if (!*token) |
| 6102 | continue; |
| 6103 | |
| 6104 | if (!strcmp(token, "all")) { |
Tejun Heo | 6e5c830 | 2016-02-22 22:25:47 -0500 | [diff] [blame] | 6105 | cgroup_no_v1_mask = U16_MAX; |
Johannes Weiner | 223ffb2 | 2016-02-11 13:34:49 -0500 | [diff] [blame] | 6106 | break; |
| 6107 | } |
| 6108 | |
| 6109 | for_each_subsys(ss, i) { |
| 6110 | if (strcmp(token, ss->name) && |
| 6111 | strcmp(token, ss->legacy_name)) |
| 6112 | continue; |
| 6113 | |
| 6114 | cgroup_no_v1_mask |= 1 << i; |
| 6115 | } |
| 6116 | } |
| 6117 | return 1; |
| 6118 | } |
| 6119 | __setup("cgroup_no_v1=", cgroup_no_v1); |
| 6120 | |
Tejun Heo | b77d7b6 | 2013-08-13 11:01:54 -0400 | [diff] [blame] | 6121 | /** |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 6122 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
Tejun Heo | 35cf083 | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 6123 | * @dentry: directory dentry of interest |
| 6124 | * @ss: subsystem of interest |
Tejun Heo | b77d7b6 | 2013-08-13 11:01:54 -0400 | [diff] [blame] | 6125 | * |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 6126 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
| 6127 | * to get the corresponding css and return it. If such css doesn't exist |
| 6128 | * or can't be pinned, an ERR_PTR value is returned. |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6129 | */ |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 6130 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
| 6131 | struct cgroup_subsys *ss) |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6132 | { |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 6133 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
Tejun Heo | f17fc25 | 2016-02-23 10:00:51 -0500 | [diff] [blame] | 6134 | struct file_system_type *s_type = dentry->d_sb->s_type; |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 6135 | struct cgroup_subsys_state *css = NULL; |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6136 | struct cgroup *cgrp; |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6137 | |
Tejun Heo | 35cf083 | 2013-08-26 18:40:56 -0400 | [diff] [blame] | 6138 | /* is @dentry a cgroup dir? */ |
Tejun Heo | f17fc25 | 2016-02-23 10:00:51 -0500 | [diff] [blame] | 6139 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
| 6140 | !kn || kernfs_type(kn) != KERNFS_DIR) |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6141 | return ERR_PTR(-EBADF); |
| 6142 | |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 6143 | rcu_read_lock(); |
| 6144 | |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 6145 | /* |
| 6146 | * This path doesn't originate from kernfs and @kn could already |
| 6147 | * have been or be removed at any point. @kn->priv is RCU |
Li Zefan | a418948 | 2014-09-04 14:43:07 +0800 | [diff] [blame] | 6148 | * protected for this access. See css_release_work_fn() for details. |
Tejun Heo | 2bd59d4 | 2014-02-11 11:52:49 -0500 | [diff] [blame] | 6149 | */ |
| 6150 | cgrp = rcu_dereference(kn->priv); |
| 6151 | if (cgrp) |
| 6152 | css = cgroup_css(cgrp, ss); |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 6153 | |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 6154 | if (!css || !css_tryget_online(css)) |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 6155 | css = ERR_PTR(-ENOENT); |
| 6156 | |
| 6157 | rcu_read_unlock(); |
| 6158 | return css; |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6159 | } |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6160 | |
Li Zefan | 1cb650b | 2013-08-19 10:05:24 +0800 | [diff] [blame] | 6161 | /** |
| 6162 | * css_from_id - lookup css by id |
| 6163 | * @id: the cgroup id |
| 6164 | * @ss: cgroup subsys to be looked into |
| 6165 | * |
| 6166 | * Returns the css if there's valid one with @id, otherwise returns NULL. |
| 6167 | * Should be called under rcu_read_lock(). |
| 6168 | */ |
| 6169 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) |
| 6170 | { |
Tejun Heo | 6fa4918 | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 6171 | WARN_ON_ONCE(!rcu_read_lock_held()); |
Johannes Weiner | d6ccc55 | 2016-06-17 12:24:27 -0400 | [diff] [blame] | 6172 | return idr_find(&ss->css_idr, id); |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 6173 | } |
| 6174 | |
Tejun Heo | 16af439 | 2015-11-20 15:55:52 -0500 | [diff] [blame] | 6175 | /** |
| 6176 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path |
| 6177 | * @path: path on the default hierarchy |
| 6178 | * |
| 6179 | * Find the cgroup at @path on the default hierarchy, increment its |
| 6180 | * reference count and return it. Returns pointer to the found cgroup on |
| 6181 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) |
| 6182 | * if @path points to a non-directory. |
| 6183 | */ |
| 6184 | struct cgroup *cgroup_get_from_path(const char *path) |
| 6185 | { |
| 6186 | struct kernfs_node *kn; |
| 6187 | struct cgroup *cgrp; |
| 6188 | |
| 6189 | mutex_lock(&cgroup_mutex); |
| 6190 | |
| 6191 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); |
| 6192 | if (kn) { |
| 6193 | if (kernfs_type(kn) == KERNFS_DIR) { |
| 6194 | cgrp = kn->priv; |
| 6195 | cgroup_get(cgrp); |
| 6196 | } else { |
| 6197 | cgrp = ERR_PTR(-ENOTDIR); |
| 6198 | } |
| 6199 | kernfs_put(kn); |
| 6200 | } else { |
| 6201 | cgrp = ERR_PTR(-ENOENT); |
| 6202 | } |
| 6203 | |
| 6204 | mutex_unlock(&cgroup_mutex); |
| 6205 | return cgrp; |
| 6206 | } |
| 6207 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); |
| 6208 | |
Martin KaFai Lau | 1f3fe7e | 2016-06-30 10:28:42 -0700 | [diff] [blame] | 6209 | /** |
| 6210 | * cgroup_get_from_fd - get a cgroup pointer from a fd |
| 6211 | * @fd: fd obtained by open(cgroup2_dir) |
| 6212 | * |
| 6213 | * Find the cgroup from a fd which should be obtained |
| 6214 | * by opening a cgroup directory. Returns a pointer to the |
| 6215 | * cgroup on success. ERR_PTR is returned if the cgroup |
| 6216 | * cannot be found. |
| 6217 | */ |
| 6218 | struct cgroup *cgroup_get_from_fd(int fd) |
| 6219 | { |
| 6220 | struct cgroup_subsys_state *css; |
| 6221 | struct cgroup *cgrp; |
| 6222 | struct file *f; |
| 6223 | |
| 6224 | f = fget_raw(fd); |
| 6225 | if (!f) |
| 6226 | return ERR_PTR(-EBADF); |
| 6227 | |
| 6228 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); |
| 6229 | fput(f); |
| 6230 | if (IS_ERR(css)) |
| 6231 | return ERR_CAST(css); |
| 6232 | |
| 6233 | cgrp = css->cgroup; |
| 6234 | if (!cgroup_on_dfl(cgrp)) { |
| 6235 | cgroup_put(cgrp); |
| 6236 | return ERR_PTR(-EBADF); |
| 6237 | } |
| 6238 | |
| 6239 | return cgrp; |
| 6240 | } |
| 6241 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); |
| 6242 | |
Tejun Heo | bd1060a | 2015-12-07 17:38:53 -0500 | [diff] [blame] | 6243 | /* |
| 6244 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data |
| 6245 | * definition in cgroup-defs.h. |
| 6246 | */ |
| 6247 | #ifdef CONFIG_SOCK_CGROUP_DATA |
| 6248 | |
| 6249 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) |
| 6250 | |
Tejun Heo | 3fa4cc9 | 2015-12-14 11:24:06 -0500 | [diff] [blame] | 6251 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
Tejun Heo | bd1060a | 2015-12-07 17:38:53 -0500 | [diff] [blame] | 6252 | static bool cgroup_sk_alloc_disabled __read_mostly; |
| 6253 | |
| 6254 | void cgroup_sk_alloc_disable(void) |
| 6255 | { |
| 6256 | if (cgroup_sk_alloc_disabled) |
| 6257 | return; |
| 6258 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); |
| 6259 | cgroup_sk_alloc_disabled = true; |
| 6260 | } |
| 6261 | |
| 6262 | #else |
| 6263 | |
| 6264 | #define cgroup_sk_alloc_disabled false |
| 6265 | |
| 6266 | #endif |
| 6267 | |
| 6268 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) |
| 6269 | { |
| 6270 | if (cgroup_sk_alloc_disabled) |
| 6271 | return; |
| 6272 | |
| 6273 | rcu_read_lock(); |
| 6274 | |
| 6275 | while (true) { |
| 6276 | struct css_set *cset; |
| 6277 | |
| 6278 | cset = task_css_set(current); |
| 6279 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { |
| 6280 | skcd->val = (unsigned long)cset->dfl_cgrp; |
| 6281 | break; |
| 6282 | } |
| 6283 | cpu_relax(); |
| 6284 | } |
| 6285 | |
| 6286 | rcu_read_unlock(); |
| 6287 | } |
| 6288 | |
| 6289 | void cgroup_sk_free(struct sock_cgroup_data *skcd) |
| 6290 | { |
| 6291 | cgroup_put(sock_cgroup_ptr(skcd)); |
| 6292 | } |
| 6293 | |
| 6294 | #endif /* CONFIG_SOCK_CGROUP_DATA */ |
| 6295 | |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6296 | /* cgroup namespaces */ |
| 6297 | |
| 6298 | static struct cgroup_namespace *alloc_cgroup_ns(void) |
| 6299 | { |
| 6300 | struct cgroup_namespace *new_ns; |
| 6301 | int ret; |
| 6302 | |
| 6303 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); |
| 6304 | if (!new_ns) |
| 6305 | return ERR_PTR(-ENOMEM); |
| 6306 | ret = ns_alloc_inum(&new_ns->ns); |
| 6307 | if (ret) { |
| 6308 | kfree(new_ns); |
| 6309 | return ERR_PTR(ret); |
| 6310 | } |
| 6311 | atomic_set(&new_ns->count, 1); |
| 6312 | new_ns->ns.ops = &cgroupns_operations; |
| 6313 | return new_ns; |
| 6314 | } |
| 6315 | |
| 6316 | void free_cgroup_ns(struct cgroup_namespace *ns) |
| 6317 | { |
| 6318 | put_css_set(ns->root_cset); |
| 6319 | put_user_ns(ns->user_ns); |
| 6320 | ns_free_inum(&ns->ns); |
| 6321 | kfree(ns); |
| 6322 | } |
| 6323 | EXPORT_SYMBOL(free_cgroup_ns); |
| 6324 | |
| 6325 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, |
| 6326 | struct user_namespace *user_ns, |
| 6327 | struct cgroup_namespace *old_ns) |
| 6328 | { |
Tejun Heo | fa5ff8a | 2016-02-28 08:59:33 -0500 | [diff] [blame] | 6329 | struct cgroup_namespace *new_ns; |
| 6330 | struct css_set *cset; |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6331 | |
| 6332 | BUG_ON(!old_ns); |
| 6333 | |
| 6334 | if (!(flags & CLONE_NEWCGROUP)) { |
| 6335 | get_cgroup_ns(old_ns); |
| 6336 | return old_ns; |
| 6337 | } |
| 6338 | |
| 6339 | /* Allow only sysadmin to create cgroup namespace. */ |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6340 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
Tejun Heo | fa5ff8a | 2016-02-28 08:59:33 -0500 | [diff] [blame] | 6341 | return ERR_PTR(-EPERM); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6342 | |
Eric W. Biederman | 7bd8830 | 2016-07-15 06:35:24 -0500 | [diff] [blame] | 6343 | /* It is not safe to take cgroup_mutex here */ |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6344 | spin_lock_irq(&css_set_lock); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6345 | cset = task_css_set(current); |
| 6346 | get_css_set(cset); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6347 | spin_unlock_irq(&css_set_lock); |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6348 | |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6349 | new_ns = alloc_cgroup_ns(); |
Tejun Heo | d220255 | 2016-02-18 11:44:24 -0500 | [diff] [blame] | 6350 | if (IS_ERR(new_ns)) { |
Tejun Heo | fa5ff8a | 2016-02-28 08:59:33 -0500 | [diff] [blame] | 6351 | put_css_set(cset); |
| 6352 | return new_ns; |
Tejun Heo | d220255 | 2016-02-18 11:44:24 -0500 | [diff] [blame] | 6353 | } |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6354 | |
| 6355 | new_ns->user_ns = get_user_ns(user_ns); |
| 6356 | new_ns->root_cset = cset; |
| 6357 | |
| 6358 | return new_ns; |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6359 | } |
| 6360 | |
| 6361 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) |
| 6362 | { |
| 6363 | return container_of(ns, struct cgroup_namespace, ns); |
| 6364 | } |
| 6365 | |
Aditya Kali | a0530e0 | 2016-01-29 02:54:07 -0600 | [diff] [blame] | 6366 | static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6367 | { |
Aditya Kali | a0530e0 | 2016-01-29 02:54:07 -0600 | [diff] [blame] | 6368 | struct cgroup_namespace *cgroup_ns = to_cg_ns(ns); |
| 6369 | |
| 6370 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) || |
| 6371 | !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN)) |
| 6372 | return -EPERM; |
| 6373 | |
| 6374 | /* Don't need to do anything if we are attaching to our own cgroupns. */ |
| 6375 | if (cgroup_ns == nsproxy->cgroup_ns) |
| 6376 | return 0; |
| 6377 | |
| 6378 | get_cgroup_ns(cgroup_ns); |
| 6379 | put_cgroup_ns(nsproxy->cgroup_ns); |
| 6380 | nsproxy->cgroup_ns = cgroup_ns; |
| 6381 | |
| 6382 | return 0; |
Aditya Kali | a79a908 | 2016-01-29 02:54:06 -0600 | [diff] [blame] | 6383 | } |
| 6384 | |
| 6385 | static struct ns_common *cgroupns_get(struct task_struct *task) |
| 6386 | { |
| 6387 | struct cgroup_namespace *ns = NULL; |
| 6388 | struct nsproxy *nsproxy; |
| 6389 | |
| 6390 | task_lock(task); |
| 6391 | nsproxy = task->nsproxy; |
| 6392 | if (nsproxy) { |
| 6393 | ns = nsproxy->cgroup_ns; |
| 6394 | get_cgroup_ns(ns); |
| 6395 | } |
| 6396 | task_unlock(task); |
| 6397 | |
| 6398 | return ns ? &ns->ns : NULL; |
| 6399 | } |
| 6400 | |
| 6401 | static void cgroupns_put(struct ns_common *ns) |
| 6402 | { |
| 6403 | put_cgroup_ns(to_cg_ns(ns)); |
| 6404 | } |
| 6405 | |
| 6406 | const struct proc_ns_operations cgroupns_operations = { |
| 6407 | .name = "cgroup", |
| 6408 | .type = CLONE_NEWCGROUP, |
| 6409 | .get = cgroupns_get, |
| 6410 | .put = cgroupns_put, |
| 6411 | .install = cgroupns_install, |
| 6412 | }; |
| 6413 | |
| 6414 | static __init int cgroup_namespaces_init(void) |
| 6415 | { |
| 6416 | return 0; |
| 6417 | } |
| 6418 | subsys_initcall(cgroup_namespaces_init); |
| 6419 | |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6420 | #ifdef CONFIG_CGROUP_DEBUG |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6421 | static struct cgroup_subsys_state * |
| 6422 | debug_css_alloc(struct cgroup_subsys_state *parent_css) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6423 | { |
| 6424 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); |
| 6425 | |
| 6426 | if (!css) |
| 6427 | return ERR_PTR(-ENOMEM); |
| 6428 | |
| 6429 | return css; |
| 6430 | } |
| 6431 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6432 | static void debug_css_free(struct cgroup_subsys_state *css) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6433 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6434 | kfree(css); |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6435 | } |
| 6436 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6437 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
| 6438 | struct cftype *cft) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6439 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6440 | return cgroup_task_count(css->cgroup); |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6441 | } |
| 6442 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6443 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
| 6444 | struct cftype *cft) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6445 | { |
| 6446 | return (u64)(unsigned long)current->cgroups; |
| 6447 | } |
| 6448 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6449 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
Li Zefan | 03c78cb | 2013-06-14 11:17:19 +0800 | [diff] [blame] | 6450 | struct cftype *cft) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6451 | { |
| 6452 | u64 count; |
| 6453 | |
| 6454 | rcu_read_lock(); |
Tejun Heo | a8ad805 | 2013-06-21 15:52:04 -0700 | [diff] [blame] | 6455 | count = atomic_read(&task_css_set(current)->refcount); |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6456 | rcu_read_unlock(); |
| 6457 | return count; |
| 6458 | } |
| 6459 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6460 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6461 | { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 6462 | struct cgrp_cset_link *link; |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 6463 | struct css_set *cset; |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 6464 | char *name_buf; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6465 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 6466 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); |
| 6467 | if (!name_buf) |
| 6468 | return -ENOMEM; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6469 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6470 | spin_lock_irq(&css_set_lock); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6471 | rcu_read_lock(); |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 6472 | cset = rcu_dereference(current->cgroups); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 6473 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6474 | struct cgroup *c = link->cgrp; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6475 | |
Tejun Heo | a2dd424 | 2014-03-19 10:23:55 -0400 | [diff] [blame] | 6476 | cgroup_name(c, name_buf, NAME_MAX + 1); |
Paul Menage | 2c6ab6d | 2009-09-23 15:56:23 -0700 | [diff] [blame] | 6477 | seq_printf(seq, "Root %d group %s\n", |
Tejun Heo | a2dd424 | 2014-03-19 10:23:55 -0400 | [diff] [blame] | 6478 | c->root->hierarchy_id, name_buf); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6479 | } |
| 6480 | rcu_read_unlock(); |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6481 | spin_unlock_irq(&css_set_lock); |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 6482 | kfree(name_buf); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6483 | return 0; |
| 6484 | } |
| 6485 | |
| 6486 | #define MAX_TASKS_SHOWN_PER_CSS 25 |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6487 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6488 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6489 | struct cgroup_subsys_state *css = seq_css(seq); |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 6490 | struct cgrp_cset_link *link; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6491 | |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6492 | spin_lock_irq(&css_set_lock); |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6493 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
Tejun Heo | 69d0206 | 2013-06-12 21:04:50 -0700 | [diff] [blame] | 6494 | struct css_set *cset = link->cset; |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6495 | struct task_struct *task; |
| 6496 | int count = 0; |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 6497 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 6498 | seq_printf(seq, "css_set %p\n", cset); |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 6499 | |
Tejun Heo | 5abb885 | 2013-06-12 21:04:49 -0700 | [diff] [blame] | 6500 | list_for_each_entry(task, &cset->tasks, cg_list) { |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 6501 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
| 6502 | goto overflow; |
| 6503 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6504 | } |
Tejun Heo | c756112 | 2014-02-25 10:04:01 -0500 | [diff] [blame] | 6505 | |
| 6506 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { |
| 6507 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
| 6508 | goto overflow; |
| 6509 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); |
| 6510 | } |
| 6511 | continue; |
| 6512 | overflow: |
| 6513 | seq_puts(seq, " ...\n"); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6514 | } |
Daniel Bristot de Oliveira | 82d6489 | 2016-06-22 17:28:41 -0300 | [diff] [blame] | 6515 | spin_unlock_irq(&css_set_lock); |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6516 | return 0; |
| 6517 | } |
| 6518 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 6519 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6520 | { |
Tejun Heo | 27bd4db | 2015-10-15 16:41:50 -0400 | [diff] [blame] | 6521 | return (!cgroup_is_populated(css->cgroup) && |
Zefan Li | a25eb52 | 2014-09-19 16:51:00 +0800 | [diff] [blame] | 6522 | !css_has_online_children(&css->cgroup->self)); |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6523 | } |
| 6524 | |
| 6525 | static struct cftype debug_files[] = { |
| 6526 | { |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6527 | .name = "taskcount", |
| 6528 | .read_u64 = debug_taskcount_read, |
| 6529 | }, |
| 6530 | |
| 6531 | { |
| 6532 | .name = "current_css_set", |
| 6533 | .read_u64 = current_css_set_read, |
| 6534 | }, |
| 6535 | |
| 6536 | { |
| 6537 | .name = "current_css_set_refcount", |
| 6538 | .read_u64 = current_css_set_refcount_read, |
| 6539 | }, |
| 6540 | |
| 6541 | { |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6542 | .name = "current_css_set_cg_links", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6543 | .seq_show = current_css_set_cg_links_read, |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6544 | }, |
| 6545 | |
| 6546 | { |
| 6547 | .name = "cgroup_css_links", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6548 | .seq_show = cgroup_css_links_read, |
Paul Menage | 7717f7b | 2009-09-23 15:56:22 -0700 | [diff] [blame] | 6549 | }, |
| 6550 | |
| 6551 | { |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6552 | .name = "releasable", |
| 6553 | .read_u64 = releasable_read, |
| 6554 | }, |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6555 | |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 6556 | { } /* terminate */ |
| 6557 | }; |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6558 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 6559 | struct cgroup_subsys debug_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 6560 | .css_alloc = debug_css_alloc, |
| 6561 | .css_free = debug_css_free, |
Tejun Heo | 5577964 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 6562 | .legacy_cftypes = debug_files, |
Paul Menage | fe69343 | 2009-09-23 15:56:20 -0700 | [diff] [blame] | 6563 | }; |
| 6564 | #endif /* CONFIG_CGROUP_DEBUG */ |