blob: 671dc05c0b0fd6b732cf03d6efec6a5dd5a3557d [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001/*
Paul Menageddbcc7e2007-10-18 23:39:30 -07002 * 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. Shutemov0dea1162010-03-10 15:22:20 -08007 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
10 *
Paul Menageddbcc7e2007-10-18 23:39:30 -070011 * 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 Perchesed3d2612014-04-25 18:28:03 -040029#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
Paul Menageddbcc7e2007-10-18 23:39:30 -070031#include <linux/cgroup.h>
eparis@redhat2ce97382011-06-02 21:20:51 +100032#include <linux/cred.h>
Paul Menagec6d57f332009-09-23 15:56:19 -070033#include <linux/ctype.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070034#include <linux/errno.h>
eparis@redhat2ce97382011-06-02 21:20:51 +100035#include <linux/init_task.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070036#include <linux/kernel.h>
37#include <linux/list.h>
Jianyu Zhanc9482a52014-04-26 15:40:28 +080038#include <linux/magic.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070039#include <linux/mm.h>
40#include <linux/mutex.h>
41#include <linux/mount.h>
42#include <linux/pagemap.h>
Paul Menagea4243162007-10-18 23:39:35 -070043#include <linux/proc_fs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070044#include <linux/rcupdate.h>
45#include <linux/sched.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070046#include <linux/slab.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070047#include <linux/spinlock.h>
Tejun Heo1ed13282015-09-16 12:53:17 -040048#include <linux/percpu-rwsem.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070049#include <linux/string.h>
Paul Menagebbcb81d2007-10-18 23:39:32 -070050#include <linux/sort.h>
Paul Menage81a6a5c2007-10-18 23:39:38 -070051#include <linux/kmod.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070052#include <linux/delayacct.h>
53#include <linux/cgroupstats.h>
Li Zefan0ac801f2013-01-10 11:49:27 +080054#include <linux/hashtable.h>
Li Zefan096b7fe2009-07-29 15:04:04 -070055#include <linux/pid_namespace.h>
Paul Menage2c6ab6d2009-09-23 15:56:23 -070056#include <linux/idr.h>
Ben Blumd1d9fd32009-09-23 15:56:28 -070057#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
Mike Galbraithc4c27fb2012-04-21 09:13:46 +020058#include <linux/kthread.h>
Tejun Heo776f02f2014-02-12 09:29:50 -050059#include <linux/delay.h>
Arun Sharma600634972011-07-26 16:09:06 -070060#include <linux/atomic.h>
Tejun Heoe93ad192016-01-19 12:18:41 -050061#include <linux/cpuset.h>
Aditya Kalia79a9082016-01-29 02:54:06 -060062#include <linux/proc_ns.h>
63#include <linux/nsproxy.h>
64#include <linux/proc_ns.h>
Tejun Heobd1060a2015-12-07 17:38:53 -050065#include <net/sock.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070066
Tejun Heoe25e2cbb2011-12-12 18:12:21 -080067/*
Tejun Heob1a21362013-11-29 10:42:58 -050068 * 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 Heo8d7e6fb2014-02-11 11:52:48 -050075#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
76 MAX_CFTYPE_NAME + 2)
77
Tejun Heob1a21362013-11-29 10:42:58 -050078/*
Tejun Heoe25e2cbb2011-12-12 18:12:21 -080079 * cgroup_mutex is the master lock. Any modification to cgroup or its
80 * hierarchy must be performed while holding it.
81 *
Tejun Heof0d9a5f2015-10-15 16:41:53 -040082 * css_set_lock protects task->cgroups pointer, the list of css_set
Tejun Heo0e1d7682014-02-25 10:04:03 -050083 * objects, and the chain of tasks off each css_set.
Tejun Heoe25e2cbb2011-12-12 18:12:21 -080084 *
Tejun Heo0e1d7682014-02-25 10:04:03 -050085 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
86 * cgroup.h can use them for lockdep annotations.
Tejun Heoe25e2cbb2011-12-12 18:12:21 -080087 */
Tejun Heo22194492013-04-07 09:29:51 -070088#ifdef CONFIG_PROVE_RCU
89DEFINE_MUTEX(cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -040090DEFINE_SPINLOCK(css_set_lock);
Tejun Heo0e1d7682014-02-25 10:04:03 -050091EXPORT_SYMBOL_GPL(cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -040092EXPORT_SYMBOL_GPL(css_set_lock);
Tejun Heo22194492013-04-07 09:29:51 -070093#else
Paul Menage81a6a5c2007-10-18 23:39:38 -070094static DEFINE_MUTEX(cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -040095static DEFINE_SPINLOCK(css_set_lock);
Tejun Heo22194492013-04-07 09:29:51 -070096#endif
97
Tejun Heo69e943b2014-02-08 10:36:58 -050098/*
Tejun Heo15a4c832014-05-04 15:09:14 -040099 * Protects cgroup_idr and css_idr so that IDs can be released without
100 * grabbing cgroup_mutex.
Tejun Heo6fa49182014-05-04 15:09:13 -0400101 */
102static DEFINE_SPINLOCK(cgroup_idr_lock);
103
104/*
Tejun Heo34c06252015-11-05 00:12:24 -0500105 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
106 * against file removal/re-creation across css hiding.
107 */
108static DEFINE_SPINLOCK(cgroup_file_kn_lock);
109
110/*
Tejun Heo69e943b2014-02-08 10:36:58 -0500111 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
112 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
113 */
114static DEFINE_SPINLOCK(release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700115
Tejun Heo1ed13282015-09-16 12:53:17 -0400116struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
117
Tejun Heo8353da12014-05-13 12:19:23 -0400118#define cgroup_assert_mutex_or_rcu_locked() \
Paul E. McKenneyf78f5b92015-06-18 15:50:02 -0700119 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
120 !lockdep_is_held(&cgroup_mutex), \
Tejun Heo8353da12014-05-13 12:19:23 -0400121 "cgroup_mutex or RCU read lock required");
Tejun Heo780cd8b2013-12-06 15:11:56 -0500122
Ben Blumaae8aab2010-03-10 15:22:07 -0800123/*
Tejun Heoe5fca242013-11-22 17:14:39 -0500124 * 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 */
129static struct workqueue_struct *cgroup_destroy_wq;
130
131/*
Tejun Heob1a21362013-11-29 10:42:58 -0500132 * pidlist destructions need to be flushed on cgroup destruction. Use a
133 * separate workqueue as flush domain.
134 */
135static struct workqueue_struct *cgroup_pidlist_destroy_wq;
136
Tejun Heo3ed80a62014-02-08 10:36:58 -0500137/* generate an array of cgroup subsystem pointers */
Tejun Heo073219e2014-02-08 10:36:58 -0500138#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
Tejun Heo3ed80a62014-02-08 10:36:58 -0500139static struct cgroup_subsys *cgroup_subsys[] = {
Paul Menageddbcc7e2007-10-18 23:39:30 -0700140#include <linux/cgroup_subsys.h>
141};
Tejun Heo073219e2014-02-08 10:36:58 -0500142#undef SUBSYS
143
144/* array of cgroup subsystem names */
145#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
146static const char *cgroup_subsys_name[] = {
147#include <linux/cgroup_subsys.h>
148};
149#undef SUBSYS
Paul Menageddbcc7e2007-10-18 23:39:30 -0700150
Tejun Heo49d1dc42015-09-18 11:56:28 -0400151/* 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,
161static 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,
167static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
168#include <linux/cgroup_subsys.h>
169};
170#undef SUBSYS
171
Paul Menageddbcc7e2007-10-18 23:39:30 -0700172/*
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400173 * The default hierarchy, reserved for the subsystems that are otherwise
Tejun Heo9871bf92013-06-24 15:21:47 -0700174 * unattached - it never has more than a single cgroup, and all tasks are
175 * part of that cgroup.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700176 */
Tejun Heoa2dd4242014-03-19 10:23:55 -0400177struct cgroup_root cgrp_dfl_root;
Tejun Heod0ec4232015-08-05 16:03:19 -0400178EXPORT_SYMBOL_GPL(cgrp_dfl_root);
Tejun Heo9871bf92013-06-24 15:21:47 -0700179
Tejun Heoa2dd4242014-03-19 10:23:55 -0400180/*
181 * The default hierarchy always exists but is hidden until mounted for the
182 * first time. This is for backward compatibility.
183 */
Tejun Heoa7165262016-02-23 10:00:50 -0500184static bool cgrp_dfl_visible;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700185
Johannes Weiner223ffb22016-02-11 13:34:49 -0500186/* Controllers blocked by the commandline in v1 */
Tejun Heo6e5c8302016-02-22 22:25:47 -0500187static u16 cgroup_no_v1_mask;
Johannes Weiner223ffb22016-02-11 13:34:49 -0500188
Tejun Heo5533e012014-05-14 19:33:07 -0400189/* some controllers are not supported in the default hierarchy */
Tejun Heoa7165262016-02-23 10:00:50 -0500190static u16 cgrp_dfl_inhibit_ss_mask;
Tejun Heo5533e012014-05-14 19:33:07 -0400191
Tejun Heof6d635ad2016-03-08 11:51:26 -0500192/* some controllers are implicitly enabled on the default hierarchy */
193static unsigned long cgrp_dfl_implicit_ss_mask;
194
Paul Menageddbcc7e2007-10-18 23:39:30 -0700195/* The list of hierarchy roots */
196
Tejun Heo9871bf92013-06-24 15:21:47 -0700197static LIST_HEAD(cgroup_roots);
198static int cgroup_root_count;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700199
Tejun Heo3417ae12014-02-08 10:37:01 -0500200/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
Tejun Heo1a574232013-04-14 11:36:58 -0700201static DEFINE_IDR(cgroup_hierarchy_idr);
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700202
Li Zefan794611a2013-06-18 18:53:53 +0800203/*
Tejun Heo0cb51d72014-05-16 13:22:49 -0400204 * 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 Zefan794611a2013-06-18 18:53:53 +0800209 */
Tejun Heo0cb51d72014-05-16 13:22:49 -0400210static u64 css_serial_nr_next = 1;
Li Zefan794611a2013-06-18 18:53:53 +0800211
Aleksa Saraicb4a3162015-06-06 10:02:14 +1000212/*
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 Menageddbcc7e2007-10-18 23:39:30 -0700216 */
Tejun Heo6e5c8302016-02-22 22:25:47 -0500217static u16 have_fork_callback __read_mostly;
218static u16 have_exit_callback __read_mostly;
219static u16 have_free_callback __read_mostly;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700220
Aditya Kalia79a9082016-01-29 02:54:06 -0600221/* cgroup namespace for init task */
222struct 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 Sarai7e476822015-06-09 21:32:09 +1000230/* Ditto for the can_fork callback. */
Tejun Heo6e5c8302016-02-22 22:25:47 -0500231static u16 have_canfork_callback __read_mostly;
Aleksa Sarai7e476822015-06-09 21:32:09 +1000232
Tejun Heo67e9c742015-11-16 11:13:34 -0500233static struct file_system_type cgroup2_fs_type;
Tejun Heoa14c6872014-07-15 11:05:09 -0400234static struct cftype cgroup_dfl_base_files[];
235static struct cftype cgroup_legacy_base_files[];
Tejun Heo628f7cd2013-06-28 16:24:11 -0700236
Tejun Heo6e5c8302016-02-22 22:25:47 -0500237static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
Tejun Heo945ba192016-03-03 09:58:00 -0500238static void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
Tejun Heo334c3672016-03-03 09:58:01 -0500239static int cgroup_apply_control(struct cgroup *cgrp);
240static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
Tejun Heoed27b9f2015-10-15 16:41:52 -0400241static void css_task_iter_advance(struct css_task_iter *it);
Tejun Heo42809dd2012-11-19 08:13:37 -0800242static int cgroup_destroy_locked(struct cgroup *cgrp);
Tejun Heo6cd0f5b2016-03-03 09:57:58 -0500243static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
244 struct cgroup_subsys *ss);
Tejun Heo9d755d32014-05-14 09:15:02 -0400245static void css_release(struct percpu_ref *ref);
Tejun Heof8f22e52014-04-23 11:13:16 -0400246static void kill_css(struct cgroup_subsys_state *css);
Tejun Heo4df8dc92015-09-18 17:54:23 -0400247static int cgroup_addrm_files(struct cgroup_subsys_state *css,
248 struct cgroup *cgrp, struct cftype cfts[],
Tejun Heo2bb566c2013-08-08 20:11:23 -0400249 bool is_add);
Tejun Heo42809dd2012-11-19 08:13:37 -0800250
Tejun Heofc5ed1e2015-09-18 11:56:28 -0400251/**
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 */
259static bool cgroup_ssid_enabled(int ssid)
260{
Arnd Bergmanncfe02a82016-03-15 00:21:06 +0100261 if (CGROUP_SUBSYS_COUNT == 0)
262 return false;
263
Tejun Heofc5ed1e2015-09-18 11:56:28 -0400264 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
265}
266
Johannes Weiner223ffb22016-02-11 13:34:49 -0500267static bool cgroup_ssid_no_v1(int ssid)
268{
269 return cgroup_no_v1_mask & (1 << ssid);
270}
271
Tejun Heo9e10a132015-09-18 11:56:28 -0400272/**
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 */
325static bool cgroup_on_dfl(const struct cgroup *cgrp)
326{
327 return cgrp->root == &cgrp_dfl_root;
328}
329
Tejun Heo6fa49182014-05-04 15:09:13 -0400330/* IDR wrappers which synchronize using cgroup_idr_lock */
331static 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 Heo54504e92014-05-13 12:10:59 -0400337 spin_lock_bh(&cgroup_idr_lock);
Mel Gormand0164ad2015-11-06 16:28:21 -0800338 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
Tejun Heo54504e92014-05-13 12:10:59 -0400339 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400340 idr_preload_end();
341 return ret;
342}
343
344static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
345{
346 void *ret;
347
Tejun Heo54504e92014-05-13 12:10:59 -0400348 spin_lock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400349 ret = idr_replace(idr, ptr, id);
Tejun Heo54504e92014-05-13 12:10:59 -0400350 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400351 return ret;
352}
353
354static void cgroup_idr_remove(struct idr *idr, int id)
355{
Tejun Heo54504e92014-05-13 12:10:59 -0400356 spin_lock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400357 idr_remove(idr, id);
Tejun Heo54504e92014-05-13 12:10:59 -0400358 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400359}
360
Tejun Heod51f39b2014-05-16 13:22:48 -0400361static 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 Heo5531dc92016-03-03 09:57:58 -0500370/* subsystems visibly enabled on a cgroup */
371static 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 Heof6d635ad2016-03-08 11:51:26 -0500380 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
381 cgrp_dfl_implicit_ss_mask);
Tejun Heo5531dc92016-03-03 09:57:58 -0500382 return root_ss_mask;
383}
384
385/* subsystems enabled on a cgroup */
386static 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 Heo95109b62013-08-08 20:11:27 -0400396/**
397 * cgroup_css - obtain a cgroup's css for the specified subsystem
398 * @cgrp: the cgroup of interest
Tejun Heo9d800df2014-05-14 09:15:00 -0400399 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
Tejun Heo95109b62013-08-08 20:11:27 -0400400 *
Tejun Heoca8bdca2013-08-26 18:40:56 -0400401 * 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 Heo95109b62013-08-08 20:11:27 -0400406 */
407static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
Tejun Heoca8bdca2013-08-26 18:40:56 -0400408 struct cgroup_subsys *ss)
Tejun Heo95109b62013-08-08 20:11:27 -0400409{
Tejun Heoca8bdca2013-08-26 18:40:56 -0400410 if (ss)
Tejun Heoaec25022014-02-08 10:36:58 -0500411 return rcu_dereference_check(cgrp->subsys[ss->id],
Tejun Heoace2bee812014-02-11 11:52:47 -0500412 lockdep_is_held(&cgroup_mutex));
Tejun Heoca8bdca2013-08-26 18:40:56 -0400413 else
Tejun Heo9d800df2014-05-14 09:15:00 -0400414 return &cgrp->self;
Tejun Heo95109b62013-08-08 20:11:27 -0400415}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700416
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400417/**
418 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
419 * @cgrp: the cgroup of interest
Tejun Heo9d800df2014-05-14 09:15:00 -0400420 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400421 *
Chen Hanxiaod0f702e2015-04-23 07:57:33 -0400422 * Similar to cgroup_css() but returns the effective css, which is defined
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400423 * 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 */
427static 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 Heo9d800df2014-05-14 09:15:00 -0400433 return &cgrp->self;
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400434
Tejun Heoeeecbd12014-11-18 02:49:52 -0500435 /*
436 * This function is used while updating css associations and thus
Tejun Heo5531dc92016-03-03 09:57:58 -0500437 * can't test the csses directly. Test ss_mask.
Tejun Heoeeecbd12014-11-18 02:49:52 -0500438 */
Tejun Heo5531dc92016-03-03 09:57:58 -0500439 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
Tejun Heod51f39b2014-05-16 13:22:48 -0400440 cgrp = cgroup_parent(cgrp);
Tejun Heo5531dc92016-03-03 09:57:58 -0500441 if (!cgrp)
442 return NULL;
443 }
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400444
445 return cgroup_css(cgrp, ss);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700446}
447
Tejun Heoeeecbd12014-11-18 02:49:52 -0500448/**
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 */
459struct 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);
476out_unlock:
477 rcu_read_unlock();
478 return css;
479}
480
Paul Menageddbcc7e2007-10-18 23:39:30 -0700481/* convenient tests for these bits */
Tejun Heo54766d42013-06-12 21:04:53 -0700482static inline bool cgroup_is_dead(const struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700483{
Tejun Heo184faf32014-05-16 13:22:51 -0400484 return !(cgrp->self.flags & CSS_ONLINE);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700485}
486
Tejun Heo052c3f32015-10-15 16:41:50 -0400487static void cgroup_get(struct cgroup *cgrp)
488{
489 WARN_ON_ONCE(cgroup_is_dead(cgrp));
490 css_get(&cgrp->self);
491}
492
493static bool cgroup_tryget(struct cgroup *cgrp)
494{
495 return css_tryget(&cgrp->self);
496}
497
Tejun Heob4168642014-05-13 12:16:21 -0400498struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
Tejun Heo59f52962014-02-11 11:52:49 -0500499{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500500 struct cgroup *cgrp = of->kn->parent->priv;
Tejun Heob4168642014-05-13 12:16:21 -0400501 struct cftype *cft = of_cft(of);
Tejun Heo2bd59d42014-02-11 11:52:49 -0500502
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 Heo9d800df2014-05-14 09:15:00 -0400514 return &cgrp->self;
Tejun Heo59f52962014-02-11 11:52:49 -0500515}
Tejun Heob4168642014-05-13 12:16:21 -0400516EXPORT_SYMBOL_GPL(of_css);
Tejun Heo59f52962014-02-11 11:52:49 -0500517
Adrian Bunke9685a02008-02-07 00:13:46 -0800518static int notify_on_release(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700519{
Paul Menagebd89aab2007-10-18 23:40:44 -0700520 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700521}
522
Tejun Heo30159ec2013-06-25 11:53:37 -0700523/**
Tejun Heo1c6727a2013-12-06 15:11:56 -0500524 * 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 Heo30159ec2013-06-25 11:53:37 -0700528 *
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400529 * Should be called under cgroup_[tree_]mutex.
Tejun Heo30159ec2013-06-25 11:53:37 -0700530 */
Tejun Heo1c6727a2013-12-06 15:11:56 -0500531#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 Heoaec3dfc2014-04-23 11:13:14 -0400539 * 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 Heo3ed80a62014-02-08 10:36:58 -0500553 * for_each_subsys - iterate all enabled cgroup subsystems
Tejun Heo30159ec2013-06-25 11:53:37 -0700554 * @ss: the iteration cursor
Tejun Heo780cd8b2013-12-06 15:11:56 -0500555 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
Tejun Heo30159ec2013-06-25 11:53:37 -0700556 */
Tejun Heo780cd8b2013-12-06 15:11:56 -0500557#define for_each_subsys(ss, ssid) \
Tejun Heo3ed80a62014-02-08 10:36:58 -0500558 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
559 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
Tejun Heo30159ec2013-06-25 11:53:37 -0700560
Aleksa Saraicb4a3162015-06-06 10:02:14 +1000561/**
Tejun Heob4e0eea2016-02-22 22:25:46 -0500562 * do_each_subsys_mask - filter for_each_subsys with a bitmask
Aleksa Saraicb4a3162015-06-06 10:02:14 +1000563 * @ss: the iteration cursor
564 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
Tejun Heob4e0eea2016-02-22 22:25:46 -0500565 * @ss_mask: the bitmask
Aleksa Saraicb4a3162015-06-06 10:02:14 +1000566 *
567 * The block will only run for cases where the ssid-th bit (1 << ssid) of
Tejun Heob4e0eea2016-02-22 22:25:46 -0500568 * @ss_mask is set.
Aleksa Saraicb4a3162015-06-06 10:02:14 +1000569 */
Tejun Heob4e0eea2016-02-22 22:25:46 -0500570#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 Sarai4a705c52015-06-09 21:32:07 +1000573 (ssid) = 0; \
Tejun Heob4e0eea2016-02-22 22:25:46 -0500574 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 Saraicb4a3162015-06-06 10:02:14 +1000584
Tejun Heo985ed672014-03-19 10:23:53 -0400585/* iterate across the hierarchies */
586#define for_each_root(root) \
Tejun Heo5549c4972013-06-24 15:21:48 -0700587 list_for_each_entry((root), &cgroup_roots, root_list)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700588
Tejun Heof8f22e52014-04-23 11:13:16 -0400589/* iterate over child cgrps, lock should be held throughout iteration */
590#define cgroup_for_each_live_child(child, cgrp) \
Tejun Heod5c419b2014-05-16 13:22:48 -0400591 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
Tejun Heo8353da12014-05-13 12:19:23 -0400592 if (({ lockdep_assert_held(&cgroup_mutex); \
Tejun Heof8f22e52014-04-23 11:13:16 -0400593 cgroup_is_dead(child); })) \
594 ; \
595 else
Tejun Heo7ae1bad2013-04-07 09:29:51 -0700596
Tejun Heoce3f1d92016-03-03 09:57:59 -0500597/* 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 Menage81a6a5c2007-10-18 23:39:38 -0700615static void cgroup_release_agent(struct work_struct *work);
Paul Menagebd89aab2007-10-18 23:40:44 -0700616static void check_for_release(struct cgroup *cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700617
Tejun Heo69d02062013-06-12 21:04:50 -0700618/*
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 */
626struct 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 Menage817929e2007-10-18 23:39:36 -0700636};
637
Tejun Heo172a2c062014-03-19 10:23:53 -0400638/*
639 * The default css_set - used by init and its children prior to any
Paul Menage817929e2007-10-18 23:39:36 -0700640 * 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 Heo5024ae22014-05-07 21:31:17 -0400645struct css_set init_css_set = {
Tejun Heo172a2c062014-03-19 10:23:53 -0400646 .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 Heoed27b9f2015-10-15 16:41:52 -0400652 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
Tejun Heo172a2c062014-03-19 10:23:53 -0400653};
Paul Menage817929e2007-10-18 23:39:36 -0700654
Tejun Heo172a2c062014-03-19 10:23:53 -0400655static int css_set_count = 1; /* 1 for init_css_set */
Paul Menage817929e2007-10-18 23:39:36 -0700656
Tejun Heo842b5972014-04-25 18:28:02 -0400657/**
Tejun Heo0de09422015-10-15 16:41:49 -0400658 * css_set_populated - does a css_set contain any tasks?
659 * @cset: target css_set
660 */
661static bool css_set_populated(struct css_set *cset)
662{
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400663 lockdep_assert_held(&css_set_lock);
Tejun Heo0de09422015-10-15 16:41:49 -0400664
665 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
666}
667
668/**
Tejun Heo842b5972014-04-25 18:28:02 -0400669 * cgroup_update_populated - updated populated count of a cgroup
670 * @cgrp: the target cgroup
671 * @populated: inc or dec populated count
672 *
Tejun Heo0de09422015-10-15 16:41:49 -0400673 * 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 Heo842b5972014-04-25 18:28:02 -0400677 *
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 */
684static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
685{
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400686 lockdep_assert_held(&css_set_lock);
Tejun Heo842b5972014-04-25 18:28:02 -0400687
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 Heoad2ed2b2015-10-15 16:41:50 -0400699 check_for_release(cgrp);
Tejun Heo6f60ead2015-09-18 17:54:23 -0400700 cgroup_file_notify(&cgrp->events_file);
701
Tejun Heod51f39b2014-05-16 13:22:48 -0400702 cgrp = cgroup_parent(cgrp);
Tejun Heo842b5972014-04-25 18:28:02 -0400703 } while (cgrp);
704}
705
Tejun Heo0de09422015-10-15 16:41:49 -0400706/**
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 */
714static void css_set_update_populated(struct css_set *cset, bool populated)
715{
716 struct cgrp_cset_link *link;
717
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400718 lockdep_assert_held(&css_set_lock);
Tejun Heo0de09422015-10-15 16:41:49 -0400719
720 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
721 cgroup_update_populated(link->cgrp, populated);
722}
723
Tejun Heof6d7d042015-10-15 16:41:52 -0400724/**
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 Heoed27b9f2015-10-15 16:41:52 -0400735 * 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 Heof6d7d042015-10-15 16:41:52 -0400738 */
739static 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 Heof0d9a5f2015-10-15 16:41:53 -0400743 lockdep_assert_held(&css_set_lock);
Tejun Heof6d7d042015-10-15 16:41:52 -0400744
Tejun Heo20b454a2016-03-03 09:57:57 -0500745 if (to_cset && !css_set_populated(to_cset))
746 css_set_update_populated(to_cset, true);
747
Tejun Heof6d7d042015-10-15 16:41:52 -0400748 if (from_cset) {
Tejun Heoed27b9f2015-10-15 16:41:52 -0400749 struct css_task_iter *it, *pos;
750
Tejun Heof6d7d042015-10-15 16:41:52 -0400751 WARN_ON_ONCE(list_empty(&task->cg_list));
Tejun Heoed27b9f2015-10-15 16:41:52 -0400752
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 Heof6d7d042015-10-15 16:41:52 -0400765 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 Heof6d7d042015-10-15 16:41:52 -0400781 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 Menage7717f7b2009-09-23 15:56:22 -0700787/*
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 Zefan472b10532008-04-29 01:00:11 -0700792#define CSS_SET_HASH_BITS 7
Li Zefan0ac801f2013-01-10 11:49:27 +0800793static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
Li Zefan472b10532008-04-29 01:00:11 -0700794
Li Zefan0ac801f2013-01-10 11:49:27 +0800795static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
Li Zefan472b10532008-04-29 01:00:11 -0700796{
Li Zefan0ac801f2013-01-10 11:49:27 +0800797 unsigned long key = 0UL;
Tejun Heo30159ec2013-06-25 11:53:37 -0700798 struct cgroup_subsys *ss;
799 int i;
Li Zefan472b10532008-04-29 01:00:11 -0700800
Tejun Heo30159ec2013-06-25 11:53:37 -0700801 for_each_subsys(ss, i)
Li Zefan0ac801f2013-01-10 11:49:27 +0800802 key += (unsigned long)css[i];
803 key = (key >> 16) ^ key;
Li Zefan472b10532008-04-29 01:00:11 -0700804
Li Zefan0ac801f2013-01-10 11:49:27 +0800805 return key;
Li Zefan472b10532008-04-29 01:00:11 -0700806}
807
Zefan Lia25eb522014-09-19 16:51:00 +0800808static void put_css_set_locked(struct css_set *cset)
Paul Menageb4f48b62007-10-18 23:39:33 -0700809{
Tejun Heo69d02062013-06-12 21:04:50 -0700810 struct cgrp_cset_link *link, *tmp_link;
Tejun Heo2d8f2432014-04-23 11:13:15 -0400811 struct cgroup_subsys *ss;
812 int ssid;
Tejun Heo5abb8852013-06-12 21:04:49 -0700813
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400814 lockdep_assert_held(&css_set_lock);
Tejun Heo89c55092014-02-13 06:58:40 -0500815
816 if (!atomic_dec_and_test(&cset->refcount))
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700817 return;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700818
Tejun Heo53254f92015-11-23 14:55:41 -0500819 /* This css_set is dead. unlink it and release cgroup and css refs */
820 for_each_subsys(ss, ssid) {
Tejun Heo2d8f2432014-04-23 11:13:15 -0400821 list_del(&cset->e_cset_node[ssid]);
Tejun Heo53254f92015-11-23 14:55:41 -0500822 css_put(cset->subsys[ssid]);
823 }
Tejun Heo5abb8852013-06-12 21:04:49 -0700824 hash_del(&cset->hlist);
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700825 css_set_count--;
826
Tejun Heo69d02062013-06-12 21:04:50 -0700827 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
Tejun Heo69d02062013-06-12 21:04:50 -0700828 list_del(&link->cset_link);
829 list_del(&link->cgrp_link);
Tejun Heo2ceb2312015-10-15 16:41:51 -0400830 if (cgroup_parent(link->cgrp))
831 cgroup_put(link->cgrp);
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700832 kfree(link);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700833 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700834
Tejun Heo5abb8852013-06-12 21:04:49 -0700835 kfree_rcu(cset, rcu_head);
Paul Menage817929e2007-10-18 23:39:36 -0700836}
837
Zefan Lia25eb522014-09-19 16:51:00 +0800838static void put_css_set(struct css_set *cset)
Tejun Heo89c55092014-02-13 06:58:40 -0500839{
840 /*
841 * Ensure that the refcount doesn't hit zero while any readers
842 * can see it. Similar to atomic_dec_and_lock(), but for an
843 * rwlock
844 */
845 if (atomic_add_unless(&cset->refcount, -1, 1))
846 return;
847
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400848 spin_lock_bh(&css_set_lock);
Zefan Lia25eb522014-09-19 16:51:00 +0800849 put_css_set_locked(cset);
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400850 spin_unlock_bh(&css_set_lock);
Tejun Heo89c55092014-02-13 06:58:40 -0500851}
852
Paul Menage817929e2007-10-18 23:39:36 -0700853/*
854 * refcounted get/put for css_set objects
855 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700856static inline void get_css_set(struct css_set *cset)
Paul Menage817929e2007-10-18 23:39:36 -0700857{
Tejun Heo5abb8852013-06-12 21:04:49 -0700858 atomic_inc(&cset->refcount);
Paul Menage817929e2007-10-18 23:39:36 -0700859}
860
Tejun Heob326f9d2013-06-24 15:21:48 -0700861/**
Paul Menage7717f7b2009-09-23 15:56:22 -0700862 * compare_css_sets - helper function for find_existing_css_set().
Tejun Heo5abb8852013-06-12 21:04:49 -0700863 * @cset: candidate css_set being tested
864 * @old_cset: existing css_set for a task
Paul Menage7717f7b2009-09-23 15:56:22 -0700865 * @new_cgrp: cgroup that's being entered by the task
866 * @template: desired set of css pointers in css_set (pre-calculated)
867 *
Li Zefan6f4b7e62013-07-31 16:18:36 +0800868 * Returns true if "cset" matches "old_cset" except for the hierarchy
Paul Menage7717f7b2009-09-23 15:56:22 -0700869 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
870 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700871static bool compare_css_sets(struct css_set *cset,
872 struct css_set *old_cset,
Paul Menage7717f7b2009-09-23 15:56:22 -0700873 struct cgroup *new_cgrp,
874 struct cgroup_subsys_state *template[])
875{
876 struct list_head *l1, *l2;
877
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400878 /*
879 * On the default hierarchy, there can be csets which are
880 * associated with the same set of cgroups but different csses.
881 * Let's first ensure that csses match.
882 */
883 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
Paul Menage7717f7b2009-09-23 15:56:22 -0700884 return false;
Paul Menage7717f7b2009-09-23 15:56:22 -0700885
886 /*
887 * Compare cgroup pointers in order to distinguish between
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400888 * different cgroups in hierarchies. As different cgroups may
889 * share the same effective css, this comparison is always
890 * necessary.
Paul Menage7717f7b2009-09-23 15:56:22 -0700891 */
Tejun Heo69d02062013-06-12 21:04:50 -0700892 l1 = &cset->cgrp_links;
893 l2 = &old_cset->cgrp_links;
Paul Menage7717f7b2009-09-23 15:56:22 -0700894 while (1) {
Tejun Heo69d02062013-06-12 21:04:50 -0700895 struct cgrp_cset_link *link1, *link2;
Tejun Heo5abb8852013-06-12 21:04:49 -0700896 struct cgroup *cgrp1, *cgrp2;
Paul Menage7717f7b2009-09-23 15:56:22 -0700897
898 l1 = l1->next;
899 l2 = l2->next;
900 /* See if we reached the end - both lists are equal length. */
Tejun Heo69d02062013-06-12 21:04:50 -0700901 if (l1 == &cset->cgrp_links) {
902 BUG_ON(l2 != &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700903 break;
904 } else {
Tejun Heo69d02062013-06-12 21:04:50 -0700905 BUG_ON(l2 == &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700906 }
907 /* Locate the cgroups associated with these links. */
Tejun Heo69d02062013-06-12 21:04:50 -0700908 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
909 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
910 cgrp1 = link1->cgrp;
911 cgrp2 = link2->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -0700912 /* Hierarchies should be linked in the same order. */
Tejun Heo5abb8852013-06-12 21:04:49 -0700913 BUG_ON(cgrp1->root != cgrp2->root);
Paul Menage7717f7b2009-09-23 15:56:22 -0700914
915 /*
916 * If this hierarchy is the hierarchy of the cgroup
917 * that's changing, then we need to check that this
918 * css_set points to the new cgroup; if it's any other
919 * hierarchy, then this css_set should point to the
920 * same cgroup as the old css_set.
921 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700922 if (cgrp1->root == new_cgrp->root) {
923 if (cgrp1 != new_cgrp)
Paul Menage7717f7b2009-09-23 15:56:22 -0700924 return false;
925 } else {
Tejun Heo5abb8852013-06-12 21:04:49 -0700926 if (cgrp1 != cgrp2)
Paul Menage7717f7b2009-09-23 15:56:22 -0700927 return false;
928 }
929 }
930 return true;
931}
932
Tejun Heob326f9d2013-06-24 15:21:48 -0700933/**
934 * find_existing_css_set - init css array and find the matching css_set
935 * @old_cset: the css_set that we're using before the cgroup transition
936 * @cgrp: the cgroup that we're moving into
937 * @template: out param for the new set of csses, should be clear on entry
Paul Menage817929e2007-10-18 23:39:36 -0700938 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700939static struct css_set *find_existing_css_set(struct css_set *old_cset,
940 struct cgroup *cgrp,
941 struct cgroup_subsys_state *template[])
Paul Menage817929e2007-10-18 23:39:36 -0700942{
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400943 struct cgroup_root *root = cgrp->root;
Tejun Heo30159ec2013-06-25 11:53:37 -0700944 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -0700945 struct css_set *cset;
Li Zefan0ac801f2013-01-10 11:49:27 +0800946 unsigned long key;
Tejun Heob326f9d2013-06-24 15:21:48 -0700947 int i;
Paul Menage817929e2007-10-18 23:39:36 -0700948
Ben Blumaae8aab2010-03-10 15:22:07 -0800949 /*
950 * Build the set of subsystem state objects that we want to see in the
951 * new css_set. while subsystems can change globally, the entries here
952 * won't change, so no need for locking.
953 */
Tejun Heo30159ec2013-06-25 11:53:37 -0700954 for_each_subsys(ss, i) {
Tejun Heof392e512014-04-23 11:13:14 -0400955 if (root->subsys_mask & (1UL << i)) {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400956 /*
957 * @ss is in this hierarchy, so we want the
958 * effective css from @cgrp.
959 */
960 template[i] = cgroup_e_css(cgrp, ss);
Paul Menage817929e2007-10-18 23:39:36 -0700961 } else {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400962 /*
963 * @ss is not in this hierarchy, so we don't want
964 * to change the css.
965 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700966 template[i] = old_cset->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700967 }
968 }
969
Li Zefan0ac801f2013-01-10 11:49:27 +0800970 key = css_set_hash(template);
Tejun Heo5abb8852013-06-12 21:04:49 -0700971 hash_for_each_possible(css_set_table, cset, hlist, key) {
972 if (!compare_css_sets(cset, old_cset, cgrp, template))
Paul Menage7717f7b2009-09-23 15:56:22 -0700973 continue;
974
975 /* This css_set matches what we need */
Tejun Heo5abb8852013-06-12 21:04:49 -0700976 return cset;
Li Zefan472b10532008-04-29 01:00:11 -0700977 }
Paul Menage817929e2007-10-18 23:39:36 -0700978
979 /* No existing cgroup group matched */
980 return NULL;
981}
982
Tejun Heo69d02062013-06-12 21:04:50 -0700983static void free_cgrp_cset_links(struct list_head *links_to_free)
Paul Menage817929e2007-10-18 23:39:36 -0700984{
Tejun Heo69d02062013-06-12 21:04:50 -0700985 struct cgrp_cset_link *link, *tmp_link;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700986
Tejun Heo69d02062013-06-12 21:04:50 -0700987 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
988 list_del(&link->cset_link);
Paul Menage817929e2007-10-18 23:39:36 -0700989 kfree(link);
990 }
991}
992
Tejun Heo69d02062013-06-12 21:04:50 -0700993/**
994 * allocate_cgrp_cset_links - allocate cgrp_cset_links
995 * @count: the number of links to allocate
996 * @tmp_links: list_head the allocated links are put on
997 *
998 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
999 * through ->cset_link. Returns 0 on success or -errno.
Li Zefan36553432008-07-29 22:33:19 -07001000 */
Tejun Heo69d02062013-06-12 21:04:50 -07001001static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
Li Zefan36553432008-07-29 22:33:19 -07001002{
Tejun Heo69d02062013-06-12 21:04:50 -07001003 struct cgrp_cset_link *link;
Li Zefan36553432008-07-29 22:33:19 -07001004 int i;
Tejun Heo69d02062013-06-12 21:04:50 -07001005
1006 INIT_LIST_HEAD(tmp_links);
1007
Li Zefan36553432008-07-29 22:33:19 -07001008 for (i = 0; i < count; i++) {
Tejun Heof4f4be22013-06-12 21:04:51 -07001009 link = kzalloc(sizeof(*link), GFP_KERNEL);
Li Zefan36553432008-07-29 22:33:19 -07001010 if (!link) {
Tejun Heo69d02062013-06-12 21:04:50 -07001011 free_cgrp_cset_links(tmp_links);
Li Zefan36553432008-07-29 22:33:19 -07001012 return -ENOMEM;
1013 }
Tejun Heo69d02062013-06-12 21:04:50 -07001014 list_add(&link->cset_link, tmp_links);
Li Zefan36553432008-07-29 22:33:19 -07001015 }
1016 return 0;
1017}
1018
Li Zefanc12f65d2009-01-07 18:07:42 -08001019/**
1020 * link_css_set - a helper function to link a css_set to a cgroup
Tejun Heo69d02062013-06-12 21:04:50 -07001021 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
Tejun Heo5abb8852013-06-12 21:04:49 -07001022 * @cset: the css_set to be linked
Li Zefanc12f65d2009-01-07 18:07:42 -08001023 * @cgrp: the destination cgroup
1024 */
Tejun Heo69d02062013-06-12 21:04:50 -07001025static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1026 struct cgroup *cgrp)
Li Zefanc12f65d2009-01-07 18:07:42 -08001027{
Tejun Heo69d02062013-06-12 21:04:50 -07001028 struct cgrp_cset_link *link;
Li Zefanc12f65d2009-01-07 18:07:42 -08001029
Tejun Heo69d02062013-06-12 21:04:50 -07001030 BUG_ON(list_empty(tmp_links));
Tejun Heo6803c002014-04-23 11:13:16 -04001031
1032 if (cgroup_on_dfl(cgrp))
1033 cset->dfl_cgrp = cgrp;
1034
Tejun Heo69d02062013-06-12 21:04:50 -07001035 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1036 link->cset = cset;
Paul Menage7717f7b2009-09-23 15:56:22 -07001037 link->cgrp = cgrp;
Tejun Heo842b5972014-04-25 18:28:02 -04001038
Paul Menage7717f7b2009-09-23 15:56:22 -07001039 /*
Tejun Heo389b9c12015-10-15 16:41:51 -04001040 * Always add links to the tail of the lists so that the lists are
1041 * in choronological order.
Paul Menage7717f7b2009-09-23 15:56:22 -07001042 */
Tejun Heo389b9c12015-10-15 16:41:51 -04001043 list_move_tail(&link->cset_link, &cgrp->cset_links);
Tejun Heo69d02062013-06-12 21:04:50 -07001044 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
Tejun Heo2ceb2312015-10-15 16:41:51 -04001045
1046 if (cgroup_parent(cgrp))
1047 cgroup_get(cgrp);
Li Zefanc12f65d2009-01-07 18:07:42 -08001048}
1049
Tejun Heob326f9d2013-06-24 15:21:48 -07001050/**
1051 * find_css_set - return a new css_set with one cgroup updated
1052 * @old_cset: the baseline css_set
1053 * @cgrp: the cgroup to be updated
1054 *
1055 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1056 * substituted into the appropriate hierarchy.
Paul Menage817929e2007-10-18 23:39:36 -07001057 */
Tejun Heo5abb8852013-06-12 21:04:49 -07001058static struct css_set *find_css_set(struct css_set *old_cset,
1059 struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -07001060{
Tejun Heob326f9d2013-06-24 15:21:48 -07001061 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
Tejun Heo5abb8852013-06-12 21:04:49 -07001062 struct css_set *cset;
Tejun Heo69d02062013-06-12 21:04:50 -07001063 struct list_head tmp_links;
1064 struct cgrp_cset_link *link;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001065 struct cgroup_subsys *ss;
Li Zefan0ac801f2013-01-10 11:49:27 +08001066 unsigned long key;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001067 int ssid;
Li Zefan472b10532008-04-29 01:00:11 -07001068
Tejun Heob326f9d2013-06-24 15:21:48 -07001069 lockdep_assert_held(&cgroup_mutex);
1070
Paul Menage817929e2007-10-18 23:39:36 -07001071 /* First see if we already have a cgroup group that matches
1072 * the desired set */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001073 spin_lock_bh(&css_set_lock);
Tejun Heo5abb8852013-06-12 21:04:49 -07001074 cset = find_existing_css_set(old_cset, cgrp, template);
1075 if (cset)
1076 get_css_set(cset);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001077 spin_unlock_bh(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001078
Tejun Heo5abb8852013-06-12 21:04:49 -07001079 if (cset)
1080 return cset;
Paul Menage817929e2007-10-18 23:39:36 -07001081
Tejun Heof4f4be22013-06-12 21:04:51 -07001082 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
Tejun Heo5abb8852013-06-12 21:04:49 -07001083 if (!cset)
Paul Menage817929e2007-10-18 23:39:36 -07001084 return NULL;
1085
Tejun Heo69d02062013-06-12 21:04:50 -07001086 /* Allocate all the cgrp_cset_link objects that we'll need */
Tejun Heo9871bf92013-06-24 15:21:47 -07001087 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
Tejun Heo5abb8852013-06-12 21:04:49 -07001088 kfree(cset);
Paul Menage817929e2007-10-18 23:39:36 -07001089 return NULL;
1090 }
1091
Tejun Heo5abb8852013-06-12 21:04:49 -07001092 atomic_set(&cset->refcount, 1);
Tejun Heo69d02062013-06-12 21:04:50 -07001093 INIT_LIST_HEAD(&cset->cgrp_links);
Tejun Heo5abb8852013-06-12 21:04:49 -07001094 INIT_LIST_HEAD(&cset->tasks);
Tejun Heoc7561122014-02-25 10:04:01 -05001095 INIT_LIST_HEAD(&cset->mg_tasks);
Tejun Heo1958d2d2014-02-25 10:04:03 -05001096 INIT_LIST_HEAD(&cset->mg_preload_node);
Tejun Heob3dc0942014-02-25 10:04:01 -05001097 INIT_LIST_HEAD(&cset->mg_node);
Tejun Heoed27b9f2015-10-15 16:41:52 -04001098 INIT_LIST_HEAD(&cset->task_iters);
Tejun Heo5abb8852013-06-12 21:04:49 -07001099 INIT_HLIST_NODE(&cset->hlist);
Paul Menage817929e2007-10-18 23:39:36 -07001100
1101 /* Copy the set of subsystem state objects generated in
1102 * find_existing_css_set() */
Tejun Heo5abb8852013-06-12 21:04:49 -07001103 memcpy(cset->subsys, template, sizeof(cset->subsys));
Paul Menage817929e2007-10-18 23:39:36 -07001104
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001105 spin_lock_bh(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001106 /* Add reference counts and links from the new css_set. */
Tejun Heo69d02062013-06-12 21:04:50 -07001107 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07001108 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001109
Paul Menage7717f7b2009-09-23 15:56:22 -07001110 if (c->root == cgrp->root)
1111 c = cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001112 link_css_set(&tmp_links, cset, c);
Paul Menage7717f7b2009-09-23 15:56:22 -07001113 }
Paul Menage817929e2007-10-18 23:39:36 -07001114
Tejun Heo69d02062013-06-12 21:04:50 -07001115 BUG_ON(!list_empty(&tmp_links));
Paul Menage817929e2007-10-18 23:39:36 -07001116
Paul Menage817929e2007-10-18 23:39:36 -07001117 css_set_count++;
Li Zefan472b10532008-04-29 01:00:11 -07001118
Tejun Heo2d8f2432014-04-23 11:13:15 -04001119 /* Add @cset to the hash table */
Tejun Heo5abb8852013-06-12 21:04:49 -07001120 key = css_set_hash(cset->subsys);
1121 hash_add(css_set_table, &cset->hlist, key);
Li Zefan472b10532008-04-29 01:00:11 -07001122
Tejun Heo53254f92015-11-23 14:55:41 -05001123 for_each_subsys(ss, ssid) {
1124 struct cgroup_subsys_state *css = cset->subsys[ssid];
1125
Tejun Heo2d8f2432014-04-23 11:13:15 -04001126 list_add_tail(&cset->e_cset_node[ssid],
Tejun Heo53254f92015-11-23 14:55:41 -05001127 &css->cgroup->e_csets[ssid]);
1128 css_get(css);
1129 }
Tejun Heo2d8f2432014-04-23 11:13:15 -04001130
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001131 spin_unlock_bh(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001132
Tejun Heo5abb8852013-06-12 21:04:49 -07001133 return cset;
Paul Menageb4f48b62007-10-18 23:39:33 -07001134}
1135
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001136static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
Paul Menage7717f7b2009-09-23 15:56:22 -07001137{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001138 struct cgroup *root_cgrp = kf_root->kn->priv;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001139
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001140 return root_cgrp->root;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001141}
1142
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001143static int cgroup_init_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001144{
1145 int id;
1146
1147 lockdep_assert_held(&cgroup_mutex);
1148
Tejun Heo985ed672014-03-19 10:23:53 -04001149 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
Tejun Heof2e85d52014-02-11 11:52:49 -05001150 if (id < 0)
1151 return id;
1152
1153 root->hierarchy_id = id;
1154 return 0;
1155}
1156
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001157static void cgroup_exit_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001158{
1159 lockdep_assert_held(&cgroup_mutex);
1160
1161 if (root->hierarchy_id) {
1162 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1163 root->hierarchy_id = 0;
1164 }
1165}
1166
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001167static void cgroup_free_root(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001168{
1169 if (root) {
Chen Hanxiaod0f702e2015-04-23 07:57:33 -04001170 /* hierarchy ID should already have been released */
Tejun Heof2e85d52014-02-11 11:52:49 -05001171 WARN_ON_ONCE(root->hierarchy_id);
1172
1173 idr_destroy(&root->cgroup_idr);
1174 kfree(root);
1175 }
1176}
1177
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001178static void cgroup_destroy_root(struct cgroup_root *root)
Tejun Heo59f52962014-02-11 11:52:49 -05001179{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001180 struct cgroup *cgrp = &root->cgrp;
Tejun Heof2e85d52014-02-11 11:52:49 -05001181 struct cgrp_cset_link *link, *tmp_link;
Tejun Heof2e85d52014-02-11 11:52:49 -05001182
Tejun Heo334c3672016-03-03 09:58:01 -05001183 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Tejun Heof2e85d52014-02-11 11:52:49 -05001184
Tejun Heo776f02f2014-02-12 09:29:50 -05001185 BUG_ON(atomic_read(&root->nr_cgrps));
Tejun Heod5c419b2014-05-16 13:22:48 -04001186 BUG_ON(!list_empty(&cgrp->self.children));
Tejun Heof2e85d52014-02-11 11:52:49 -05001187
Tejun Heof2e85d52014-02-11 11:52:49 -05001188 /* Rebind all subsystems back to the default hierarchy */
Tejun Heo334c3672016-03-03 09:58:01 -05001189 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
Tejun Heof2e85d52014-02-11 11:52:49 -05001190
1191 /*
1192 * Release all the links from cset_links to this hierarchy's
1193 * root cgroup
1194 */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001195 spin_lock_bh(&css_set_lock);
Tejun Heof2e85d52014-02-11 11:52:49 -05001196
1197 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1198 list_del(&link->cset_link);
1199 list_del(&link->cgrp_link);
1200 kfree(link);
1201 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001202
1203 spin_unlock_bh(&css_set_lock);
Tejun Heof2e85d52014-02-11 11:52:49 -05001204
1205 if (!list_empty(&root->root_list)) {
1206 list_del(&root->root_list);
1207 cgroup_root_count--;
1208 }
1209
1210 cgroup_exit_root_id(root);
1211
1212 mutex_unlock(&cgroup_mutex);
Tejun Heof2e85d52014-02-11 11:52:49 -05001213
Tejun Heo2bd59d42014-02-11 11:52:49 -05001214 kernfs_destroy_root(root->kf_root);
Tejun Heof2e85d52014-02-11 11:52:49 -05001215 cgroup_free_root(root);
1216}
1217
Tejun Heoceb6a082014-02-25 10:04:02 -05001218/* look up cgroup associated with given css_set on the specified hierarchy */
1219static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001220 struct cgroup_root *root)
Paul Menage7717f7b2009-09-23 15:56:22 -07001221{
Paul Menage7717f7b2009-09-23 15:56:22 -07001222 struct cgroup *res = NULL;
1223
Tejun Heo96d365e2014-02-13 06:58:40 -05001224 lockdep_assert_held(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001225 lockdep_assert_held(&css_set_lock);
Tejun Heo96d365e2014-02-13 06:58:40 -05001226
Tejun Heo5abb8852013-06-12 21:04:49 -07001227 if (cset == &init_css_set) {
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001228 res = &root->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -07001229 } else {
Tejun Heo69d02062013-06-12 21:04:50 -07001230 struct cgrp_cset_link *link;
1231
1232 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07001233 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001234
Paul Menage7717f7b2009-09-23 15:56:22 -07001235 if (c->root == root) {
1236 res = c;
1237 break;
1238 }
1239 }
1240 }
Tejun Heo96d365e2014-02-13 06:58:40 -05001241
Paul Menage7717f7b2009-09-23 15:56:22 -07001242 BUG_ON(!res);
1243 return res;
1244}
1245
1246/*
Tejun Heoceb6a082014-02-25 10:04:02 -05001247 * Return the cgroup for "task" from the given hierarchy. Must be
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001248 * called with cgroup_mutex and css_set_lock held.
Tejun Heoceb6a082014-02-25 10:04:02 -05001249 */
1250static struct cgroup *task_cgroup_from_root(struct task_struct *task,
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001251 struct cgroup_root *root)
Tejun Heoceb6a082014-02-25 10:04:02 -05001252{
1253 /*
1254 * No need to lock the task - since we hold cgroup_mutex the
1255 * task can't change groups, so the only thing that can happen
1256 * is that it exits and its css is set back to init_css_set.
1257 */
1258 return cset_cgroup_from_root(task_css_set(task), root);
1259}
1260
1261/*
Paul Menageddbcc7e2007-10-18 23:39:30 -07001262 * A task must hold cgroup_mutex to modify cgroups.
1263 *
1264 * Any task can increment and decrement the count field without lock.
1265 * So in general, code holding cgroup_mutex can't rely on the count
1266 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -08001267 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -07001268 * means that no tasks are currently attached, therefore there is no
1269 * way a task attached to that cgroup can fork (the other way to
1270 * increment the count). So code holding cgroup_mutex can safely
1271 * assume that if the count is zero, it will stay zero. Similarly, if
1272 * a task holds cgroup_mutex on a cgroup with zero count, it
1273 * knows that the cgroup won't be removed, as cgroup_rmdir()
1274 * needs that mutex.
1275 *
Paul Menageddbcc7e2007-10-18 23:39:30 -07001276 * A cgroup can only be deleted if both its 'count' of using tasks
1277 * is zero, and its list of 'children' cgroups is empty. Since all
1278 * tasks in the system use _some_ cgroup, and since there is always at
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001279 * least one task in the system (init, pid == 1), therefore, root cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -07001280 * always has either children cgroups and/or using tasks. So we don't
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001281 * need a special hack to ensure that root cgroup cannot be deleted.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001282 *
1283 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -08001284 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -07001285 */
1286
Tejun Heo2bd59d42014-02-11 11:52:49 -05001287static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001288static const struct file_operations proc_cgroupstats_operations;
Paul Menagea4243162007-10-18 23:39:35 -07001289
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001290static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1291 char *buf)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001292{
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001293 struct cgroup_subsys *ss = cft->ss;
1294
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001295 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1296 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1297 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001298 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1299 cft->name);
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001300 else
1301 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1302 return buf;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001303}
1304
Tejun Heof2e85d52014-02-11 11:52:49 -05001305/**
1306 * cgroup_file_mode - deduce file mode of a control file
1307 * @cft: the control file in question
1308 *
Tejun Heo7dbdb192015-09-18 17:54:23 -04001309 * S_IRUGO for read, S_IWUSR for write.
Tejun Heof2e85d52014-02-11 11:52:49 -05001310 */
1311static umode_t cgroup_file_mode(const struct cftype *cft)
Li Zefan65dff752013-03-01 15:01:56 +08001312{
Tejun Heof2e85d52014-02-11 11:52:49 -05001313 umode_t mode = 0;
Li Zefan65dff752013-03-01 15:01:56 +08001314
Tejun Heof2e85d52014-02-11 11:52:49 -05001315 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1316 mode |= S_IRUGO;
1317
Tejun Heo7dbdb192015-09-18 17:54:23 -04001318 if (cft->write_u64 || cft->write_s64 || cft->write) {
1319 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1320 mode |= S_IWUGO;
1321 else
1322 mode |= S_IWUSR;
1323 }
Tejun Heof2e85d52014-02-11 11:52:49 -05001324
1325 return mode;
Li Zefan65dff752013-03-01 15:01:56 +08001326}
1327
Tejun Heoa9746d82014-05-13 12:19:22 -04001328/**
Tejun Heo8699b772016-02-22 22:25:46 -05001329 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
Tejun Heo0f060de2014-11-18 02:49:50 -05001330 * @subtree_control: the new subtree_control mask to consider
Tejun Heo5ced2512016-03-03 09:58:01 -05001331 * @this_ss_mask: available subsystems
Tejun Heoaf0ba672014-07-08 18:02:57 -04001332 *
1333 * On the default hierarchy, a subsystem may request other subsystems to be
1334 * enabled together through its ->depends_on mask. In such cases, more
1335 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1336 *
Tejun Heo0f060de2014-11-18 02:49:50 -05001337 * This function calculates which subsystems need to be enabled if
Tejun Heo5ced2512016-03-03 09:58:01 -05001338 * @subtree_control is to be applied while restricted to @this_ss_mask.
Tejun Heoaf0ba672014-07-08 18:02:57 -04001339 */
Tejun Heo5ced2512016-03-03 09:58:01 -05001340static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
Tejun Heo667c2492014-07-08 18:02:56 -04001341{
Tejun Heo6e5c8302016-02-22 22:25:47 -05001342 u16 cur_ss_mask = subtree_control;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001343 struct cgroup_subsys *ss;
1344 int ssid;
1345
1346 lockdep_assert_held(&cgroup_mutex);
1347
Tejun Heof6d635ad2016-03-08 11:51:26 -05001348 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1349
Tejun Heoaf0ba672014-07-08 18:02:57 -04001350 while (true) {
Tejun Heo6e5c8302016-02-22 22:25:47 -05001351 u16 new_ss_mask = cur_ss_mask;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001352
Tejun Heob4e0eea2016-02-22 22:25:46 -05001353 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
Aleksa Saraia966a4e2015-06-06 10:02:15 +10001354 new_ss_mask |= ss->depends_on;
Tejun Heob4e0eea2016-02-22 22:25:46 -05001355 } while_each_subsys_mask();
Tejun Heoaf0ba672014-07-08 18:02:57 -04001356
1357 /*
1358 * Mask out subsystems which aren't available. This can
1359 * happen only if some depended-upon subsystems were bound
1360 * to non-default hierarchies.
1361 */
Tejun Heo5ced2512016-03-03 09:58:01 -05001362 new_ss_mask &= this_ss_mask;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001363
1364 if (new_ss_mask == cur_ss_mask)
1365 break;
1366 cur_ss_mask = new_ss_mask;
1367 }
1368
Tejun Heo0f060de2014-11-18 02:49:50 -05001369 return cur_ss_mask;
1370}
1371
1372/**
Tejun Heoa9746d82014-05-13 12:19:22 -04001373 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1374 * @kn: the kernfs_node being serviced
1375 *
1376 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1377 * the method finishes if locking succeeded. Note that once this function
1378 * returns the cgroup returned by cgroup_kn_lock_live() may become
1379 * inaccessible any time. If the caller intends to continue to access the
1380 * cgroup, it should pin it before invoking this function.
1381 */
1382static void cgroup_kn_unlock(struct kernfs_node *kn)
1383{
1384 struct cgroup *cgrp;
1385
1386 if (kernfs_type(kn) == KERNFS_DIR)
1387 cgrp = kn->priv;
1388 else
1389 cgrp = kn->parent->priv;
1390
1391 mutex_unlock(&cgroup_mutex);
Tejun Heoa9746d82014-05-13 12:19:22 -04001392
1393 kernfs_unbreak_active_protection(kn);
1394 cgroup_put(cgrp);
1395}
1396
1397/**
1398 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1399 * @kn: the kernfs_node being serviced
Tejun Heo945ba192016-03-03 09:58:00 -05001400 * @drain_offline: perform offline draining on the cgroup
Tejun Heoa9746d82014-05-13 12:19:22 -04001401 *
1402 * This helper is to be used by a cgroup kernfs method currently servicing
1403 * @kn. It breaks the active protection, performs cgroup locking and
1404 * verifies that the associated cgroup is alive. Returns the cgroup if
1405 * alive; otherwise, %NULL. A successful return should be undone by a
Tejun Heo945ba192016-03-03 09:58:00 -05001406 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1407 * cgroup is drained of offlining csses before return.
Tejun Heoa9746d82014-05-13 12:19:22 -04001408 *
1409 * Any cgroup kernfs method implementation which requires locking the
1410 * associated cgroup should use this helper. It avoids nesting cgroup
1411 * locking under kernfs active protection and allows all kernfs operations
1412 * including self-removal.
1413 */
Tejun Heo945ba192016-03-03 09:58:00 -05001414static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn,
1415 bool drain_offline)
Tejun Heoa9746d82014-05-13 12:19:22 -04001416{
1417 struct cgroup *cgrp;
1418
1419 if (kernfs_type(kn) == KERNFS_DIR)
1420 cgrp = kn->priv;
1421 else
1422 cgrp = kn->parent->priv;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001423
Tejun Heo2bd59d42014-02-11 11:52:49 -05001424 /*
Tejun Heo01f64742014-05-13 12:19:23 -04001425 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoa9746d82014-05-13 12:19:22 -04001426 * active_ref. cgroup liveliness check alone provides enough
1427 * protection against removal. Ensure @cgrp stays accessible and
1428 * break the active_ref protection.
Tejun Heo2bd59d42014-02-11 11:52:49 -05001429 */
Li Zefanaa323622014-09-04 14:43:38 +08001430 if (!cgroup_tryget(cgrp))
1431 return NULL;
Tejun Heoa9746d82014-05-13 12:19:22 -04001432 kernfs_break_active_protection(kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001433
Tejun Heo945ba192016-03-03 09:58:00 -05001434 if (drain_offline)
1435 cgroup_lock_and_drain_offline(cgrp);
1436 else
1437 mutex_lock(&cgroup_mutex);
Tejun Heoa9746d82014-05-13 12:19:22 -04001438
1439 if (!cgroup_is_dead(cgrp))
1440 return cgrp;
1441
1442 cgroup_kn_unlock(kn);
1443 return NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001444}
1445
Li Zefan2739d3c2013-01-21 18:18:33 +08001446static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001447{
Tejun Heo2bd59d42014-02-11 11:52:49 -05001448 char name[CGROUP_FILE_NAME_MAX];
Paul Menageddbcc7e2007-10-18 23:39:30 -07001449
Tejun Heo01f64742014-05-13 12:19:23 -04001450 lockdep_assert_held(&cgroup_mutex);
Tejun Heo34c06252015-11-05 00:12:24 -05001451
1452 if (cft->file_offset) {
1453 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1454 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1455
1456 spin_lock_irq(&cgroup_file_kn_lock);
1457 cfile->kn = NULL;
1458 spin_unlock_irq(&cgroup_file_kn_lock);
1459 }
1460
Tejun Heo2bd59d42014-02-11 11:52:49 -05001461 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
Tejun Heo05ef1d72012-04-01 12:09:56 -07001462}
1463
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001464/**
Tejun Heo4df8dc92015-09-18 17:54:23 -04001465 * css_clear_dir - remove subsys files in a cgroup directory
1466 * @css: taget css
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001467 */
Tejun Heo334c3672016-03-03 09:58:01 -05001468static void css_clear_dir(struct cgroup_subsys_state *css)
Tejun Heo05ef1d72012-04-01 12:09:56 -07001469{
Tejun Heo334c3672016-03-03 09:58:01 -05001470 struct cgroup *cgrp = css->cgroup;
Tejun Heo4df8dc92015-09-18 17:54:23 -04001471 struct cftype *cfts;
Tejun Heo05ef1d72012-04-01 12:09:56 -07001472
Tejun Heo88cb04b2016-03-03 09:57:58 -05001473 if (!(css->flags & CSS_VISIBLE))
1474 return;
1475
1476 css->flags &= ~CSS_VISIBLE;
1477
Tejun Heo4df8dc92015-09-18 17:54:23 -04001478 list_for_each_entry(cfts, &css->ss->cfts, node)
1479 cgroup_addrm_files(css, cgrp, cfts, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001480}
1481
Tejun Heoccdca212015-09-18 17:54:23 -04001482/**
Tejun Heo4df8dc92015-09-18 17:54:23 -04001483 * css_populate_dir - create subsys files in a cgroup directory
1484 * @css: target css
Tejun Heoccdca212015-09-18 17:54:23 -04001485 *
1486 * On failure, no file is added.
1487 */
Tejun Heo334c3672016-03-03 09:58:01 -05001488static int css_populate_dir(struct cgroup_subsys_state *css)
Tejun Heoccdca212015-09-18 17:54:23 -04001489{
Tejun Heo334c3672016-03-03 09:58:01 -05001490 struct cgroup *cgrp = css->cgroup;
Tejun Heo4df8dc92015-09-18 17:54:23 -04001491 struct cftype *cfts, *failed_cfts;
1492 int ret;
Tejun Heoccdca212015-09-18 17:54:23 -04001493
Tejun Heo03970d32016-03-03 09:58:00 -05001494 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
Tejun Heo88cb04b2016-03-03 09:57:58 -05001495 return 0;
1496
Tejun Heo4df8dc92015-09-18 17:54:23 -04001497 if (!css->ss) {
1498 if (cgroup_on_dfl(cgrp))
1499 cfts = cgroup_dfl_base_files;
1500 else
1501 cfts = cgroup_legacy_base_files;
Tejun Heoccdca212015-09-18 17:54:23 -04001502
Tejun Heo4df8dc92015-09-18 17:54:23 -04001503 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1504 }
Tejun Heoccdca212015-09-18 17:54:23 -04001505
Tejun Heo4df8dc92015-09-18 17:54:23 -04001506 list_for_each_entry(cfts, &css->ss->cfts, node) {
1507 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1508 if (ret < 0) {
1509 failed_cfts = cfts;
1510 goto err;
Tejun Heoccdca212015-09-18 17:54:23 -04001511 }
1512 }
Tejun Heo88cb04b2016-03-03 09:57:58 -05001513
1514 css->flags |= CSS_VISIBLE;
1515
Tejun Heoccdca212015-09-18 17:54:23 -04001516 return 0;
1517err:
Tejun Heo4df8dc92015-09-18 17:54:23 -04001518 list_for_each_entry(cfts, &css->ss->cfts, node) {
1519 if (cfts == failed_cfts)
1520 break;
1521 cgroup_addrm_files(css, cgrp, cfts, false);
1522 }
Tejun Heoccdca212015-09-18 17:54:23 -04001523 return ret;
1524}
1525
Tejun Heo6e5c8302016-02-22 22:25:47 -05001526static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001527{
Tejun Heo1ada4832015-09-18 17:54:23 -04001528 struct cgroup *dcgrp = &dst_root->cgrp;
Tejun Heo30159ec2013-06-25 11:53:37 -07001529 struct cgroup_subsys *ss;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001530 int ssid, i, ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001531
Tejun Heoace2bee812014-02-11 11:52:47 -05001532 lockdep_assert_held(&cgroup_mutex);
Ben Blumaae8aab2010-03-10 15:22:07 -08001533
Tejun Heob4e0eea2016-02-22 22:25:46 -05001534 do_each_subsys_mask(ss, ssid, ss_mask) {
Tejun Heof6d635ad2016-03-08 11:51:26 -05001535 /*
1536 * If @ss has non-root csses attached to it, can't move.
1537 * If @ss is an implicit controller, it is exempt from this
1538 * rule and can be stolen.
1539 */
1540 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1541 !ss->implicit_on_dfl)
Tejun Heo3ed80a62014-02-08 10:36:58 -05001542 return -EBUSY;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001543
Tejun Heo5df36032014-03-19 10:23:54 -04001544 /* can't move between two non-dummy roots either */
Tejun Heo7fd8c562014-04-23 11:13:16 -04001545 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
Tejun Heo5df36032014-03-19 10:23:54 -04001546 return -EBUSY;
Tejun Heob4e0eea2016-02-22 22:25:46 -05001547 } while_each_subsys_mask();
Paul Menageddbcc7e2007-10-18 23:39:30 -07001548
Tejun Heob4e0eea2016-02-22 22:25:46 -05001549 do_each_subsys_mask(ss, ssid, ss_mask) {
Tejun Heo1ada4832015-09-18 17:54:23 -04001550 struct cgroup_root *src_root = ss->root;
1551 struct cgroup *scgrp = &src_root->cgrp;
1552 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001553 struct css_set *cset;
Tejun Heo30159ec2013-06-25 11:53:37 -07001554
Tejun Heo1ada4832015-09-18 17:54:23 -04001555 WARN_ON(!css || cgroup_css(dcgrp, ss));
Tejun Heo73e80ed2013-08-13 11:01:55 -04001556
Tejun Heo334c3672016-03-03 09:58:01 -05001557 /* disable from the source */
1558 src_root->subsys_mask &= ~(1 << ssid);
1559 WARN_ON(cgroup_apply_control(scgrp));
1560 cgroup_finalize_control(scgrp, 0);
Tejun Heo4df8dc92015-09-18 17:54:23 -04001561
Tejun Heo334c3672016-03-03 09:58:01 -05001562 /* rebind */
Tejun Heo1ada4832015-09-18 17:54:23 -04001563 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1564 rcu_assign_pointer(dcgrp->subsys[ssid], css);
Tejun Heo5df36032014-03-19 10:23:54 -04001565 ss->root = dst_root;
Tejun Heo1ada4832015-09-18 17:54:23 -04001566 css->cgroup = dcgrp;
Tejun Heoa8a648c42013-06-24 15:21:47 -07001567
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001568 spin_lock_bh(&css_set_lock);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001569 hash_for_each(css_set_table, i, cset, hlist)
1570 list_move_tail(&cset->e_cset_node[ss->id],
Tejun Heo1ada4832015-09-18 17:54:23 -04001571 &dcgrp->e_csets[ss->id]);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001572 spin_unlock_bh(&css_set_lock);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001573
Tejun Heobd53d612014-04-23 11:13:16 -04001574 /* default hierarchy doesn't enable controllers by default */
Tejun Heof392e512014-04-23 11:13:14 -04001575 dst_root->subsys_mask |= 1 << ssid;
Tejun Heo49d1dc42015-09-18 11:56:28 -04001576 if (dst_root == &cgrp_dfl_root) {
1577 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1578 } else {
Tejun Heo1ada4832015-09-18 17:54:23 -04001579 dcgrp->subtree_control |= 1 << ssid;
Tejun Heo49d1dc42015-09-18 11:56:28 -04001580 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
Tejun Heo667c2492014-07-08 18:02:56 -04001581 }
Tejun Heo73e80ed2013-08-13 11:01:55 -04001582
Tejun Heo334c3672016-03-03 09:58:01 -05001583 ret = cgroup_apply_control(dcgrp);
1584 if (ret)
1585 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1586 ss->name, ret);
1587
Tejun Heo5df36032014-03-19 10:23:54 -04001588 if (ss->bind)
1589 ss->bind(css);
Tejun Heob4e0eea2016-02-22 22:25:46 -05001590 } while_each_subsys_mask();
Paul Menageddbcc7e2007-10-18 23:39:30 -07001591
Tejun Heo1ada4832015-09-18 17:54:23 -04001592 kernfs_activate(dcgrp->kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001593 return 0;
1594}
1595
Tejun Heo2bd59d42014-02-11 11:52:49 -05001596static int cgroup_show_options(struct seq_file *seq,
1597 struct kernfs_root *kf_root)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001598{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001599 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001600 struct cgroup_subsys *ss;
Tejun Heob85d2042013-12-06 15:11:57 -05001601 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001602
Tejun Heod98817d2015-08-18 13:58:16 -07001603 if (root != &cgrp_dfl_root)
1604 for_each_subsys(ss, ssid)
1605 if (root->subsys_mask & (1 << ssid))
Kees Cook61e57c02015-09-08 14:58:22 -07001606 seq_show_option(seq, ss->legacy_name, NULL);
Tejun Heo93438622013-04-14 20:15:25 -07001607 if (root->flags & CGRP_ROOT_NOPREFIX)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001608 seq_puts(seq, ",noprefix");
Tejun Heo93438622013-04-14 20:15:25 -07001609 if (root->flags & CGRP_ROOT_XATTR)
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001610 seq_puts(seq, ",xattr");
Tejun Heo69e943b2014-02-08 10:36:58 -05001611
1612 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001613 if (strlen(root->release_agent_path))
Kees Cooka068acf2015-09-04 15:44:57 -07001614 seq_show_option(seq, "release_agent",
1615 root->release_agent_path);
Tejun Heo69e943b2014-02-08 10:36:58 -05001616 spin_unlock(&release_agent_path_lock);
1617
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001618 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
Daniel Lezcano97978e62010-10-27 15:33:35 -07001619 seq_puts(seq, ",clone_children");
Paul Menagec6d57f332009-09-23 15:56:19 -07001620 if (strlen(root->name))
Kees Cooka068acf2015-09-04 15:44:57 -07001621 seq_show_option(seq, "name", root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001622 return 0;
1623}
1624
1625struct cgroup_sb_opts {
Tejun Heo6e5c8302016-02-22 22:25:47 -05001626 u16 subsys_mask;
Tejun Heo69dfa002014-05-04 15:09:13 -04001627 unsigned int flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001628 char *release_agent;
Tejun Heo2260e7f2012-11-19 08:13:38 -08001629 bool cpuset_clone_children;
Paul Menagec6d57f332009-09-23 15:56:19 -07001630 char *name;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001631 /* User explicitly requested empty subsystem */
1632 bool none;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001633};
1634
Ben Blumcf5d5942010-03-10 15:22:09 -08001635static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001636{
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001637 char *token, *o = data;
1638 bool all_ss = false, one_ss = false;
Tejun Heo6e5c8302016-02-22 22:25:47 -05001639 u16 mask = U16_MAX;
Tejun Heo30159ec2013-06-25 11:53:37 -07001640 struct cgroup_subsys *ss;
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001641 int nr_opts = 0;
Tejun Heo30159ec2013-06-25 11:53:37 -07001642 int i;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001643
1644#ifdef CONFIG_CPUSETS
Tejun Heo6e5c8302016-02-22 22:25:47 -05001645 mask = ~((u16)1 << cpuset_cgrp_id);
Li Zefanf9ab5b52009-06-17 16:26:33 -07001646#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001647
Paul Menagec6d57f332009-09-23 15:56:19 -07001648 memset(opts, 0, sizeof(*opts));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001649
1650 while ((token = strsep(&o, ",")) != NULL) {
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001651 nr_opts++;
1652
Paul Menageddbcc7e2007-10-18 23:39:30 -07001653 if (!*token)
1654 return -EINVAL;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001655 if (!strcmp(token, "none")) {
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001656 /* Explicitly have no subsystems */
1657 opts->none = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001658 continue;
1659 }
1660 if (!strcmp(token, "all")) {
1661 /* Mutually exclusive option 'all' + subsystem name */
1662 if (one_ss)
1663 return -EINVAL;
1664 all_ss = true;
1665 continue;
1666 }
1667 if (!strcmp(token, "noprefix")) {
Tejun Heo93438622013-04-14 20:15:25 -07001668 opts->flags |= CGRP_ROOT_NOPREFIX;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001669 continue;
1670 }
1671 if (!strcmp(token, "clone_children")) {
Tejun Heo2260e7f2012-11-19 08:13:38 -08001672 opts->cpuset_clone_children = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001673 continue;
1674 }
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001675 if (!strcmp(token, "xattr")) {
Tejun Heo93438622013-04-14 20:15:25 -07001676 opts->flags |= CGRP_ROOT_XATTR;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001677 continue;
1678 }
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001679 if (!strncmp(token, "release_agent=", 14)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07001680 /* Specifying two release agents is forbidden */
1681 if (opts->release_agent)
1682 return -EINVAL;
Paul Menagec6d57f332009-09-23 15:56:19 -07001683 opts->release_agent =
Dan Carpentere400c282010-08-10 18:02:54 -07001684 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001685 if (!opts->release_agent)
1686 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001687 continue;
1688 }
1689 if (!strncmp(token, "name=", 5)) {
Paul Menagec6d57f332009-09-23 15:56:19 -07001690 const char *name = token + 5;
1691 /* Can't specify an empty name */
1692 if (!strlen(name))
1693 return -EINVAL;
1694 /* Must match [\w.-]+ */
1695 for (i = 0; i < strlen(name); i++) {
1696 char c = name[i];
1697 if (isalnum(c))
1698 continue;
1699 if ((c == '.') || (c == '-') || (c == '_'))
1700 continue;
1701 return -EINVAL;
1702 }
1703 /* Specifying two names is forbidden */
1704 if (opts->name)
1705 return -EINVAL;
1706 opts->name = kstrndup(name,
Dan Carpentere400c282010-08-10 18:02:54 -07001707 MAX_CGROUP_ROOT_NAMELEN - 1,
Paul Menagec6d57f332009-09-23 15:56:19 -07001708 GFP_KERNEL);
1709 if (!opts->name)
1710 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001711
1712 continue;
1713 }
1714
Tejun Heo30159ec2013-06-25 11:53:37 -07001715 for_each_subsys(ss, i) {
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001716 if (strcmp(token, ss->legacy_name))
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001717 continue;
Tejun Heofc5ed1e2015-09-18 11:56:28 -04001718 if (!cgroup_ssid_enabled(i))
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001719 continue;
Johannes Weiner223ffb22016-02-11 13:34:49 -05001720 if (cgroup_ssid_no_v1(i))
1721 continue;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001722
1723 /* Mutually exclusive option 'all' + subsystem name */
1724 if (all_ss)
1725 return -EINVAL;
Tejun Heo69dfa002014-05-04 15:09:13 -04001726 opts->subsys_mask |= (1 << i);
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001727 one_ss = true;
1728
1729 break;
1730 }
1731 if (i == CGROUP_SUBSYS_COUNT)
1732 return -ENOENT;
1733 }
1734
Li Zefanf9ab5b52009-06-17 16:26:33 -07001735 /*
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001736 * If the 'all' option was specified select all the subsystems,
1737 * otherwise if 'none', 'name=' and a subsystem name options were
1738 * not specified, let's default to 'all'
1739 */
1740 if (all_ss || (!one_ss && !opts->none && !opts->name))
1741 for_each_subsys(ss, i)
Johannes Weiner223ffb22016-02-11 13:34:49 -05001742 if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i))
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001743 opts->subsys_mask |= (1 << i);
1744
1745 /*
1746 * We either have to specify by name or by subsystems. (So all
1747 * empty hierarchies must have a name).
1748 */
1749 if (!opts->subsys_mask && !opts->name)
1750 return -EINVAL;
1751
1752 /*
Li Zefanf9ab5b52009-06-17 16:26:33 -07001753 * Option noprefix was introduced just for backward compatibility
1754 * with the old cpuset, so we allow noprefix only if mounting just
1755 * the cpuset subsystem.
1756 */
Tejun Heo93438622013-04-14 20:15:25 -07001757 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
Li Zefanf9ab5b52009-06-17 16:26:33 -07001758 return -EINVAL;
1759
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001760 /* Can't specify "none" and some subsystems */
Aristeu Rozanskia1a71b42012-08-23 16:53:31 -04001761 if (opts->subsys_mask && opts->none)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001762 return -EINVAL;
1763
Paul Menageddbcc7e2007-10-18 23:39:30 -07001764 return 0;
1765}
1766
Tejun Heo2bd59d42014-02-11 11:52:49 -05001767static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001768{
1769 int ret = 0;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001770 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001771 struct cgroup_sb_opts opts;
Tejun Heo6e5c8302016-02-22 22:25:47 -05001772 u16 added_mask, removed_mask;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001773
Tejun Heoaa6ec292014-07-09 10:08:08 -04001774 if (root == &cgrp_dfl_root) {
1775 pr_err("remount is not allowed\n");
Tejun Heo873fe092013-04-14 20:15:26 -07001776 return -EINVAL;
1777 }
1778
Tejun Heo334c3672016-03-03 09:58:01 -05001779 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001780
1781 /* See what subsystems are wanted */
1782 ret = parse_cgroupfs_options(data, &opts);
1783 if (ret)
1784 goto out_unlock;
1785
Tejun Heof392e512014-04-23 11:13:14 -04001786 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
Joe Perchesed3d2612014-04-25 18:28:03 -04001787 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
Jianyu Zhana2a1f9e2014-04-25 18:28:03 -04001788 task_tgid_nr(current), current->comm);
Tejun Heo8b5a5a92012-04-01 12:09:54 -07001789
Tejun Heof392e512014-04-23 11:13:14 -04001790 added_mask = opts.subsys_mask & ~root->subsys_mask;
1791 removed_mask = root->subsys_mask & ~opts.subsys_mask;
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001792
Ben Blumcf5d5942010-03-10 15:22:09 -08001793 /* Don't allow flags or name to change at remount */
Tejun Heo7450e902014-07-09 10:08:07 -04001794 if ((opts.flags ^ root->flags) ||
Ben Blumcf5d5942010-03-10 15:22:09 -08001795 (opts.name && strcmp(opts.name, root->name))) {
Tejun Heo69dfa002014-05-04 15:09:13 -04001796 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
Tejun Heo7450e902014-07-09 10:08:07 -04001797 opts.flags, opts.name ?: "", root->flags, root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001798 ret = -EINVAL;
Paul Menagec6d57f332009-09-23 15:56:19 -07001799 goto out_unlock;
1800 }
1801
Tejun Heof172e672013-06-28 17:07:30 -07001802 /* remounting is not allowed for populated hierarchies */
Tejun Heod5c419b2014-05-16 13:22:48 -04001803 if (!list_empty(&root->cgrp.self.children)) {
Tejun Heof172e672013-06-28 17:07:30 -07001804 ret = -EBUSY;
Li Zefan0670e082009-04-02 16:57:30 -07001805 goto out_unlock;
Ben Blumcf5d5942010-03-10 15:22:09 -08001806 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001807
Tejun Heo5df36032014-03-19 10:23:54 -04001808 ret = rebind_subsystems(root, added_mask);
Tejun Heo31261212013-06-28 17:07:30 -07001809 if (ret)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001810 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001811
Tejun Heo334c3672016-03-03 09:58:01 -05001812 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
Tejun Heo5df36032014-03-19 10:23:54 -04001813
Tejun Heo69e943b2014-02-08 10:36:58 -05001814 if (opts.release_agent) {
1815 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001816 strcpy(root->release_agent_path, opts.release_agent);
Tejun Heo69e943b2014-02-08 10:36:58 -05001817 spin_unlock(&release_agent_path_lock);
1818 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001819 out_unlock:
Jesper Juhl66bdc9c2009-04-02 16:57:27 -07001820 kfree(opts.release_agent);
Paul Menagec6d57f332009-09-23 15:56:19 -07001821 kfree(opts.name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001822 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001823 return ret;
1824}
1825
Tejun Heoafeb0f92014-02-13 06:58:39 -05001826/*
1827 * To reduce the fork() overhead for systems that are not actually using
1828 * their cgroups capability, we don't maintain the lists running through
1829 * each css_set to its tasks until we see the list actually used - in other
1830 * words after the first mount.
1831 */
1832static bool use_task_css_set_links __read_mostly;
1833
1834static void cgroup_enable_task_cg_lists(void)
1835{
1836 struct task_struct *p, *g;
1837
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001838 spin_lock_bh(&css_set_lock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001839
1840 if (use_task_css_set_links)
1841 goto out_unlock;
1842
1843 use_task_css_set_links = true;
1844
1845 /*
1846 * We need tasklist_lock because RCU is not safe against
1847 * while_each_thread(). Besides, a forking task that has passed
1848 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1849 * is not guaranteed to have its child immediately visible in the
1850 * tasklist if we walk through it with RCU.
1851 */
1852 read_lock(&tasklist_lock);
1853 do_each_thread(g, p) {
Tejun Heoafeb0f92014-02-13 06:58:39 -05001854 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1855 task_css_set(p) != &init_css_set);
1856
1857 /*
1858 * We should check if the process is exiting, otherwise
1859 * it will race with cgroup_exit() in that the list
1860 * entry won't be deleted though the process has exited.
Tejun Heof153ad12014-02-25 09:56:49 -05001861 * Do it while holding siglock so that we don't end up
1862 * racing against cgroup_exit().
Tejun Heoafeb0f92014-02-13 06:58:39 -05001863 */
Tejun Heof153ad12014-02-25 09:56:49 -05001864 spin_lock_irq(&p->sighand->siglock);
Tejun Heoeaf797a2014-02-25 10:04:03 -05001865 if (!(p->flags & PF_EXITING)) {
1866 struct css_set *cset = task_css_set(p);
1867
Tejun Heo0de09422015-10-15 16:41:49 -04001868 if (!css_set_populated(cset))
1869 css_set_update_populated(cset, true);
Tejun Heo389b9c12015-10-15 16:41:51 -04001870 list_add_tail(&p->cg_list, &cset->tasks);
Tejun Heoeaf797a2014-02-25 10:04:03 -05001871 get_css_set(cset);
1872 }
Tejun Heof153ad12014-02-25 09:56:49 -05001873 spin_unlock_irq(&p->sighand->siglock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001874 } while_each_thread(g, p);
1875 read_unlock(&tasklist_lock);
1876out_unlock:
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001877 spin_unlock_bh(&css_set_lock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001878}
Paul Menageddbcc7e2007-10-18 23:39:30 -07001879
Paul Menagecc31edc2008-10-18 20:28:04 -07001880static void init_cgroup_housekeeping(struct cgroup *cgrp)
1881{
Tejun Heo2d8f2432014-04-23 11:13:15 -04001882 struct cgroup_subsys *ss;
1883 int ssid;
1884
Tejun Heod5c419b2014-05-16 13:22:48 -04001885 INIT_LIST_HEAD(&cgrp->self.sibling);
1886 INIT_LIST_HEAD(&cgrp->self.children);
Tejun Heo69d02062013-06-12 21:04:50 -07001887 INIT_LIST_HEAD(&cgrp->cset_links);
Ben Blum72a8cb32009-09-23 15:56:27 -07001888 INIT_LIST_HEAD(&cgrp->pidlists);
1889 mutex_init(&cgrp->pidlist_mutex);
Tejun Heo9d800df2014-05-14 09:15:00 -04001890 cgrp->self.cgroup = cgrp;
Tejun Heo184faf32014-05-16 13:22:51 -04001891 cgrp->self.flags |= CSS_ONLINE;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001892
1893 for_each_subsys(ss, ssid)
1894 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
Tejun Heof8f22e52014-04-23 11:13:16 -04001895
1896 init_waitqueue_head(&cgrp->offline_waitq);
Zefan Li971ff492014-09-18 16:06:19 +08001897 INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
Paul Menagecc31edc2008-10-18 20:28:04 -07001898}
Paul Menagec6d57f332009-09-23 15:56:19 -07001899
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001900static void init_cgroup_root(struct cgroup_root *root,
Tejun Heo172a2c062014-03-19 10:23:53 -04001901 struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001902{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001903 struct cgroup *cgrp = &root->cgrp;
Tejun Heob0ca5a82012-04-01 12:09:54 -07001904
Paul Menageddbcc7e2007-10-18 23:39:30 -07001905 INIT_LIST_HEAD(&root->root_list);
Tejun Heo3c9c8252014-02-12 09:29:50 -05001906 atomic_set(&root->nr_cgrps, 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001907 cgrp->root = root;
Paul Menagecc31edc2008-10-18 20:28:04 -07001908 init_cgroup_housekeeping(cgrp);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08001909 idr_init(&root->cgroup_idr);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001910
Paul Menagec6d57f332009-09-23 15:56:19 -07001911 root->flags = opts->flags;
1912 if (opts->release_agent)
1913 strcpy(root->release_agent_path, opts->release_agent);
1914 if (opts->name)
1915 strcpy(root->name, opts->name);
Tejun Heo2260e7f2012-11-19 08:13:38 -08001916 if (opts->cpuset_clone_children)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001917 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
Paul Menagec6d57f332009-09-23 15:56:19 -07001918}
1919
Tejun Heo6e5c8302016-02-22 22:25:47 -05001920static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001921{
Tejun Heod427dfe2014-02-11 11:52:48 -05001922 LIST_HEAD(tmp_links);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001923 struct cgroup *root_cgrp = &root->cgrp;
Tejun Heod427dfe2014-02-11 11:52:48 -05001924 struct css_set *cset;
Tejun Heod427dfe2014-02-11 11:52:48 -05001925 int i, ret;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001926
Tejun Heod427dfe2014-02-11 11:52:48 -05001927 lockdep_assert_held(&cgroup_mutex);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001928
Vladimir Davydovcf780b72015-08-03 15:32:26 +03001929 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
Tejun Heod427dfe2014-02-11 11:52:48 -05001930 if (ret < 0)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001931 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05001932 root_cgrp->id = ret;
Tejun Heob11cfb52015-11-20 15:55:52 -05001933 root_cgrp->ancestor_ids[0] = ret;
Paul Menagec6d57f332009-09-23 15:56:19 -07001934
Tejun Heo2aad2a82014-09-24 13:31:50 -04001935 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1936 GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04001937 if (ret)
1938 goto out;
1939
Tejun Heod427dfe2014-02-11 11:52:48 -05001940 /*
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001941 * We're accessing css_set_count without locking css_set_lock here,
Tejun Heod427dfe2014-02-11 11:52:48 -05001942 * but that's OK - it can only be increased by someone holding
Tejun Heo04313592016-03-03 09:58:01 -05001943 * cgroup_lock, and that's us. Later rebinding may disable
1944 * controllers on the default hierarchy and thus create new csets,
1945 * which can't be more than the existing ones. Allocate 2x.
Tejun Heod427dfe2014-02-11 11:52:48 -05001946 */
Tejun Heo04313592016-03-03 09:58:01 -05001947 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001948 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04001949 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001950
Tejun Heo985ed672014-03-19 10:23:53 -04001951 ret = cgroup_init_root_id(root);
Tejun Heod427dfe2014-02-11 11:52:48 -05001952 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04001953 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001954
Tejun Heo2bd59d42014-02-11 11:52:49 -05001955 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
1956 KERNFS_ROOT_CREATE_DEACTIVATED,
1957 root_cgrp);
1958 if (IS_ERR(root->kf_root)) {
1959 ret = PTR_ERR(root->kf_root);
1960 goto exit_root_id;
1961 }
1962 root_cgrp->kn = root->kf_root->kn;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001963
Tejun Heo334c3672016-03-03 09:58:01 -05001964 ret = css_populate_dir(&root_cgrp->self);
Tejun Heod427dfe2014-02-11 11:52:48 -05001965 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001966 goto destroy_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001967
Tejun Heo5df36032014-03-19 10:23:54 -04001968 ret = rebind_subsystems(root, ss_mask);
Tejun Heod427dfe2014-02-11 11:52:48 -05001969 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001970 goto destroy_root;
Al Viro0df6a632010-12-21 13:29:29 -05001971
Tejun Heod427dfe2014-02-11 11:52:48 -05001972 /*
1973 * There must be no failure case after here, since rebinding takes
1974 * care of subsystems' refcounts, which are explicitly dropped in
1975 * the failure exit path.
1976 */
1977 list_add(&root->root_list, &cgroup_roots);
1978 cgroup_root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001979
Tejun Heod427dfe2014-02-11 11:52:48 -05001980 /*
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001981 * Link the root cgroup in this hierarchy into all the css_set
Tejun Heod427dfe2014-02-11 11:52:48 -05001982 * objects.
1983 */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001984 spin_lock_bh(&css_set_lock);
Tejun Heo0de09422015-10-15 16:41:49 -04001985 hash_for_each(css_set_table, i, cset, hlist) {
Tejun Heod427dfe2014-02-11 11:52:48 -05001986 link_css_set(&tmp_links, cset, root_cgrp);
Tejun Heo0de09422015-10-15 16:41:49 -04001987 if (css_set_populated(cset))
1988 cgroup_update_populated(root_cgrp, true);
1989 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001990 spin_unlock_bh(&css_set_lock);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001991
Tejun Heod5c419b2014-05-16 13:22:48 -04001992 BUG_ON(!list_empty(&root_cgrp->self.children));
Tejun Heo3c9c8252014-02-12 09:29:50 -05001993 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
Tejun Heod427dfe2014-02-11 11:52:48 -05001994
Tejun Heo2bd59d42014-02-11 11:52:49 -05001995 kernfs_activate(root_cgrp->kn);
Tejun Heod427dfe2014-02-11 11:52:48 -05001996 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001997 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05001998
Tejun Heo2bd59d42014-02-11 11:52:49 -05001999destroy_root:
2000 kernfs_destroy_root(root->kf_root);
2001 root->kf_root = NULL;
2002exit_root_id:
Tejun Heod427dfe2014-02-11 11:52:48 -05002003 cgroup_exit_root_id(root);
Tejun Heo9d755d32014-05-14 09:15:02 -04002004cancel_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04002005 percpu_ref_exit(&root_cgrp->self.refcnt);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002006out:
Tejun Heod427dfe2014-02-11 11:52:48 -05002007 free_cgrp_cset_links(&tmp_links);
2008 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002009}
2010
Al Virof7e83572010-07-26 13:23:11 +04002011static struct dentry *cgroup_mount(struct file_system_type *fs_type,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002012 int flags, const char *unused_dev_name,
Al Virof7e83572010-07-26 13:23:11 +04002013 void *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002014{
Tejun Heo67e9c742015-11-16 11:13:34 -05002015 bool is_v2 = fs_type == &cgroup2_fs_type;
Li Zefan3a32bd72014-06-30 11:50:59 +08002016 struct super_block *pinned_sb = NULL;
Serge Hallyned825712016-01-29 02:54:09 -06002017 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
Li Zefan970317a2014-06-30 11:49:58 +08002018 struct cgroup_subsys *ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002019 struct cgroup_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002020 struct cgroup_sb_opts opts;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002021 struct dentry *dentry;
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002022 int ret;
Li Zefan970317a2014-06-30 11:49:58 +08002023 int i;
Li Zefanc6b3d5b2014-04-04 17:14:41 +08002024 bool new_sb;
Paul Menagec6d57f332009-09-23 15:56:19 -07002025
Serge Hallyned825712016-01-29 02:54:09 -06002026 get_cgroup_ns(ns);
2027
2028 /* Check if the caller has permission to mount. */
2029 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2030 put_cgroup_ns(ns);
2031 return ERR_PTR(-EPERM);
2032 }
2033
Paul Menagec6d57f332009-09-23 15:56:19 -07002034 /*
Tejun Heo56fde9e2014-02-13 06:58:38 -05002035 * The first time anyone tries to mount a cgroup, enable the list
2036 * linking each css_set to its tasks and fix up all existing tasks.
Paul Menagec6d57f332009-09-23 15:56:19 -07002037 */
Tejun Heo56fde9e2014-02-13 06:58:38 -05002038 if (!use_task_css_set_links)
2039 cgroup_enable_task_cg_lists();
Li Zefane37a06f2014-04-17 13:53:08 +08002040
Tejun Heo67e9c742015-11-16 11:13:34 -05002041 if (is_v2) {
2042 if (data) {
2043 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
Serge Hallyned825712016-01-29 02:54:09 -06002044 put_cgroup_ns(ns);
Tejun Heo67e9c742015-11-16 11:13:34 -05002045 return ERR_PTR(-EINVAL);
2046 }
Tejun Heoa7165262016-02-23 10:00:50 -05002047 cgrp_dfl_visible = true;
Tejun Heo67e9c742015-11-16 11:13:34 -05002048 root = &cgrp_dfl_root;
2049 cgroup_get(&root->cgrp);
2050 goto out_mount;
2051 }
2052
Tejun Heo334c3672016-03-03 09:58:01 -05002053 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002054
Paul Menageddbcc7e2007-10-18 23:39:30 -07002055 /* First find the desired set of subsystems */
Paul Menageddbcc7e2007-10-18 23:39:30 -07002056 ret = parse_cgroupfs_options(data, &opts);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002057 if (ret)
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002058 goto out_unlock;
Tejun Heoa015edd2014-05-14 09:15:00 -04002059
Li Zefan970317a2014-06-30 11:49:58 +08002060 /*
2061 * Destruction of cgroup root is asynchronous, so subsystems may
2062 * still be dying after the previous unmount. Let's drain the
2063 * dying subsystems. We just need to ensure that the ones
2064 * unmounted previously finish dying and don't care about new ones
2065 * starting. Testing ref liveliness is good enough.
2066 */
2067 for_each_subsys(ss, i) {
2068 if (!(opts.subsys_mask & (1 << i)) ||
2069 ss->root == &cgrp_dfl_root)
2070 continue;
2071
2072 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
2073 mutex_unlock(&cgroup_mutex);
2074 msleep(10);
2075 ret = restart_syscall();
2076 goto out_free;
2077 }
2078 cgroup_put(&ss->root->cgrp);
2079 }
2080
Tejun Heo985ed672014-03-19 10:23:53 -04002081 for_each_root(root) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05002082 bool name_match = false;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002083
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002084 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04002085 continue;
Paul Menagec6d57f332009-09-23 15:56:19 -07002086
Paul Menage817929e2007-10-18 23:39:36 -07002087 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05002088 * If we asked for a name then it must match. Also, if
2089 * name matches but sybsys_mask doesn't, we should fail.
2090 * Remember whether name matched.
Paul Menage817929e2007-10-18 23:39:36 -07002091 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05002092 if (opts.name) {
2093 if (strcmp(opts.name, root->name))
2094 continue;
2095 name_match = true;
2096 }
Tejun Heo31261212013-06-28 17:07:30 -07002097
2098 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05002099 * If we asked for subsystems (or explicitly for no
2100 * subsystems) then they must match.
Tejun Heo31261212013-06-28 17:07:30 -07002101 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05002102 if ((opts.subsys_mask || opts.none) &&
Tejun Heof392e512014-04-23 11:13:14 -04002103 (opts.subsys_mask != root->subsys_mask)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05002104 if (!name_match)
2105 continue;
2106 ret = -EBUSY;
2107 goto out_unlock;
2108 }
Tejun Heo873fe092013-04-14 20:15:26 -07002109
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04002110 if (root->flags ^ opts.flags)
2111 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
Tejun Heo2bd59d42014-02-11 11:52:49 -05002112
Tejun Heo776f02f2014-02-12 09:29:50 -05002113 /*
Li Zefan3a32bd72014-06-30 11:50:59 +08002114 * We want to reuse @root whose lifetime is governed by its
2115 * ->cgrp. Let's check whether @root is alive and keep it
2116 * that way. As cgroup_kill_sb() can happen anytime, we
2117 * want to block it by pinning the sb so that @root doesn't
2118 * get killed before mount is complete.
2119 *
2120 * With the sb pinned, tryget_live can reliably indicate
2121 * whether @root can be reused. If it's being killed,
2122 * drain it. We can use wait_queue for the wait but this
2123 * path is super cold. Let's just sleep a bit and retry.
Tejun Heo776f02f2014-02-12 09:29:50 -05002124 */
Li Zefan3a32bd72014-06-30 11:50:59 +08002125 pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
2126 if (IS_ERR(pinned_sb) ||
2127 !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
Tejun Heo776f02f2014-02-12 09:29:50 -05002128 mutex_unlock(&cgroup_mutex);
Li Zefan3a32bd72014-06-30 11:50:59 +08002129 if (!IS_ERR_OR_NULL(pinned_sb))
2130 deactivate_super(pinned_sb);
Tejun Heo776f02f2014-02-12 09:29:50 -05002131 msleep(10);
Tejun Heoa015edd2014-05-14 09:15:00 -04002132 ret = restart_syscall();
2133 goto out_free;
Tejun Heo776f02f2014-02-12 09:29:50 -05002134 }
2135
2136 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002137 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002138 }
2139
Tejun Heo172a2c062014-03-19 10:23:53 -04002140 /*
2141 * No such thing, create a new one. name= matching without subsys
2142 * specification is allowed for already existing hierarchies but we
2143 * can't create new one without subsys specification.
2144 */
2145 if (!opts.subsys_mask && !opts.none) {
2146 ret = -EINVAL;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002147 goto out_unlock;
2148 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07002149
Serge Hallyned825712016-01-29 02:54:09 -06002150 /*
2151 * We know this subsystem has not yet been bound. Users in a non-init
2152 * user namespace may only mount hierarchies with no bound subsystems,
2153 * i.e. 'none,name=user1'
2154 */
2155 if (!opts.none && !capable(CAP_SYS_ADMIN)) {
2156 ret = -EPERM;
2157 goto out_unlock;
2158 }
2159
Tejun Heo172a2c062014-03-19 10:23:53 -04002160 root = kzalloc(sizeof(*root), GFP_KERNEL);
2161 if (!root) {
2162 ret = -ENOMEM;
2163 goto out_unlock;
2164 }
2165
2166 init_cgroup_root(root, &opts);
2167
Tejun Heo35585572014-02-13 06:58:38 -05002168 ret = cgroup_setup_root(root, opts.subsys_mask);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002169 if (ret)
2170 cgroup_free_root(root);
2171
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002172out_unlock:
Tejun Heoe25e2cbb2011-12-12 18:12:21 -08002173 mutex_unlock(&cgroup_mutex);
Tejun Heoa015edd2014-05-14 09:15:00 -04002174out_free:
Paul Menagec6d57f332009-09-23 15:56:19 -07002175 kfree(opts.release_agent);
2176 kfree(opts.name);
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002177
Serge Hallyned825712016-01-29 02:54:09 -06002178 if (ret) {
2179 put_cgroup_ns(ns);
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002180 return ERR_PTR(ret);
Serge Hallyned825712016-01-29 02:54:09 -06002181 }
Tejun Heo67e9c742015-11-16 11:13:34 -05002182out_mount:
Jianyu Zhanc9482a52014-04-26 15:40:28 +08002183 dentry = kernfs_mount(fs_type, flags, root->kf_root,
Tejun Heo67e9c742015-11-16 11:13:34 -05002184 is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC,
2185 &new_sb);
Serge Hallyned825712016-01-29 02:54:09 -06002186
2187 /*
2188 * In non-init cgroup namespace, instead of root cgroup's
2189 * dentry, we return the dentry corresponding to the
2190 * cgroupns->root_cgrp.
2191 */
2192 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
2193 struct dentry *nsdentry;
2194 struct cgroup *cgrp;
2195
2196 mutex_lock(&cgroup_mutex);
2197 spin_lock_bh(&css_set_lock);
2198
2199 cgrp = cset_cgroup_from_root(ns->root_cset, root);
2200
2201 spin_unlock_bh(&css_set_lock);
2202 mutex_unlock(&cgroup_mutex);
2203
2204 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
2205 dput(dentry);
2206 dentry = nsdentry;
2207 }
2208
Li Zefanc6b3d5b2014-04-04 17:14:41 +08002209 if (IS_ERR(dentry) || !new_sb)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002210 cgroup_put(&root->cgrp);
Li Zefan3a32bd72014-06-30 11:50:59 +08002211
2212 /*
2213 * If @pinned_sb, we're reusing an existing root and holding an
2214 * extra ref on its sb. Mount is complete. Put the extra ref.
2215 */
2216 if (pinned_sb) {
2217 WARN_ON(new_sb);
2218 deactivate_super(pinned_sb);
2219 }
2220
Serge Hallyned825712016-01-29 02:54:09 -06002221 put_cgroup_ns(ns);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002222 return dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002223}
2224
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09002225static void cgroup_kill_sb(struct super_block *sb)
2226{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002227 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002228 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002229
Tejun Heo9d755d32014-05-14 09:15:02 -04002230 /*
2231 * If @root doesn't have any mounts or children, start killing it.
2232 * This prevents new mounts by disabling percpu_ref_tryget_live().
2233 * cgroup_mount() may wait for @root's release.
Li Zefan1f779fb2014-06-04 16:48:15 +08002234 *
2235 * And don't kill the default root.
Tejun Heo9d755d32014-05-14 09:15:02 -04002236 */
Johannes Weiner3c606d32015-01-22 10:19:43 -05002237 if (!list_empty(&root->cgrp.self.children) ||
Li Zefan1f779fb2014-06-04 16:48:15 +08002238 root == &cgrp_dfl_root)
Tejun Heo9d755d32014-05-14 09:15:02 -04002239 cgroup_put(&root->cgrp);
2240 else
2241 percpu_ref_kill(&root->cgrp.self.refcnt);
2242
Tejun Heo2bd59d42014-02-11 11:52:49 -05002243 kernfs_kill_sb(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002244}
2245
2246static struct file_system_type cgroup_fs_type = {
2247 .name = "cgroup",
Al Virof7e83572010-07-26 13:23:11 +04002248 .mount = cgroup_mount,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002249 .kill_sb = cgroup_kill_sb,
Serge Hallyn1c537532016-01-29 02:54:11 -06002250 .fs_flags = FS_USERNS_MOUNT,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002251};
2252
Tejun Heo67e9c742015-11-16 11:13:34 -05002253static struct file_system_type cgroup2_fs_type = {
2254 .name = "cgroup2",
2255 .mount = cgroup_mount,
2256 .kill_sb = cgroup_kill_sb,
Serge Hallyn1c537532016-01-29 02:54:11 -06002257 .fs_flags = FS_USERNS_MOUNT,
Tejun Heo67e9c742015-11-16 11:13:34 -05002258};
2259
Aditya Kalia79a9082016-01-29 02:54:06 -06002260static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2261 struct cgroup_namespace *ns)
2262{
2263 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2264 int ret;
2265
2266 ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2267 if (ret < 0 || ret >= buflen)
2268 return NULL;
2269 return buf;
2270}
2271
2272char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2273 struct cgroup_namespace *ns)
2274{
2275 char *ret;
2276
2277 mutex_lock(&cgroup_mutex);
2278 spin_lock_bh(&css_set_lock);
2279
2280 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2281
2282 spin_unlock_bh(&css_set_lock);
2283 mutex_unlock(&cgroup_mutex);
2284
2285 return ret;
2286}
2287EXPORT_SYMBOL_GPL(cgroup_path_ns);
2288
Li Zefana043e3b2008-02-23 15:24:09 -08002289/**
Tejun Heo913ffdb2013-07-11 16:34:48 -07002290 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
Tejun Heo857a2be2013-04-14 20:50:08 -07002291 * @task: target task
Tejun Heo857a2be2013-04-14 20:50:08 -07002292 * @buf: the buffer to write the path into
2293 * @buflen: the length of the buffer
2294 *
Tejun Heo913ffdb2013-07-11 16:34:48 -07002295 * Determine @task's cgroup on the first (the one with the lowest non-zero
2296 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2297 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2298 * cgroup controller callbacks.
2299 *
Tejun Heoe61734c2014-02-12 09:29:50 -05002300 * Return value is the same as kernfs_path().
Tejun Heo857a2be2013-04-14 20:50:08 -07002301 */
Tejun Heoe61734c2014-02-12 09:29:50 -05002302char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
Tejun Heo857a2be2013-04-14 20:50:08 -07002303{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002304 struct cgroup_root *root;
Tejun Heo913ffdb2013-07-11 16:34:48 -07002305 struct cgroup *cgrp;
Tejun Heoe61734c2014-02-12 09:29:50 -05002306 int hierarchy_id = 1;
2307 char *path = NULL;
Tejun Heo857a2be2013-04-14 20:50:08 -07002308
2309 mutex_lock(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002310 spin_lock_bh(&css_set_lock);
Tejun Heo857a2be2013-04-14 20:50:08 -07002311
Tejun Heo913ffdb2013-07-11 16:34:48 -07002312 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2313
Tejun Heo857a2be2013-04-14 20:50:08 -07002314 if (root) {
2315 cgrp = task_cgroup_from_root(task, root);
Aditya Kalia79a9082016-01-29 02:54:06 -06002316 path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
Tejun Heo913ffdb2013-07-11 16:34:48 -07002317 } else {
2318 /* if no hierarchy exists, everyone is in "/" */
Tejun Heoe61734c2014-02-12 09:29:50 -05002319 if (strlcpy(buf, "/", buflen) < buflen)
2320 path = buf;
Tejun Heo857a2be2013-04-14 20:50:08 -07002321 }
2322
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002323 spin_unlock_bh(&css_set_lock);
Tejun Heo857a2be2013-04-14 20:50:08 -07002324 mutex_unlock(&cgroup_mutex);
Tejun Heoe61734c2014-02-12 09:29:50 -05002325 return path;
Tejun Heo857a2be2013-04-14 20:50:08 -07002326}
Tejun Heo913ffdb2013-07-11 16:34:48 -07002327EXPORT_SYMBOL_GPL(task_cgroup_path);
Tejun Heo857a2be2013-04-14 20:50:08 -07002328
Tejun Heob3dc0942014-02-25 10:04:01 -05002329/* used to track tasks and other necessary states during migration */
Tejun Heo2f7ee562011-12-12 18:12:21 -08002330struct cgroup_taskset {
Tejun Heob3dc0942014-02-25 10:04:01 -05002331 /* the src and dst cset list running through cset->mg_node */
2332 struct list_head src_csets;
2333 struct list_head dst_csets;
2334
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002335 /* the subsys currently being processed */
2336 int ssid;
2337
Tejun Heob3dc0942014-02-25 10:04:01 -05002338 /*
2339 * Fields for cgroup_taskset_*() iteration.
2340 *
2341 * Before migration is committed, the target migration tasks are on
2342 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2343 * the csets on ->dst_csets. ->csets point to either ->src_csets
2344 * or ->dst_csets depending on whether migration is committed.
2345 *
2346 * ->cur_csets and ->cur_task point to the current task position
2347 * during iteration.
2348 */
2349 struct list_head *csets;
2350 struct css_set *cur_cset;
2351 struct task_struct *cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002352};
2353
Tejun Heoadaae5d2015-09-11 15:00:21 -04002354#define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2355 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2356 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2357 .csets = &tset.src_csets, \
2358}
2359
2360/**
2361 * cgroup_taskset_add - try to add a migration target task to a taskset
2362 * @task: target task
2363 * @tset: target taskset
2364 *
2365 * Add @task, which is a migration target, to @tset. This function becomes
2366 * noop if @task doesn't need to be migrated. @task's css_set should have
2367 * been added as a migration source and @task->cg_list will be moved from
2368 * the css_set's tasks list to mg_tasks one.
2369 */
2370static void cgroup_taskset_add(struct task_struct *task,
2371 struct cgroup_taskset *tset)
2372{
2373 struct css_set *cset;
2374
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002375 lockdep_assert_held(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002376
2377 /* @task either already exited or can't exit until the end */
2378 if (task->flags & PF_EXITING)
2379 return;
2380
2381 /* leave @task alone if post_fork() hasn't linked it yet */
2382 if (list_empty(&task->cg_list))
2383 return;
2384
2385 cset = task_css_set(task);
2386 if (!cset->mg_src_cgrp)
2387 return;
2388
2389 list_move_tail(&task->cg_list, &cset->mg_tasks);
2390 if (list_empty(&cset->mg_node))
2391 list_add_tail(&cset->mg_node, &tset->src_csets);
2392 if (list_empty(&cset->mg_dst_cset->mg_node))
2393 list_move_tail(&cset->mg_dst_cset->mg_node,
2394 &tset->dst_csets);
2395}
2396
Tejun Heo2f7ee562011-12-12 18:12:21 -08002397/**
2398 * cgroup_taskset_first - reset taskset and return the first task
2399 * @tset: taskset of interest
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002400 * @dst_cssp: output variable for the destination css
Tejun Heo2f7ee562011-12-12 18:12:21 -08002401 *
2402 * @tset iteration is initialized and the first task is returned.
2403 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002404struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2405 struct cgroup_subsys_state **dst_cssp)
Tejun Heo2f7ee562011-12-12 18:12:21 -08002406{
Tejun Heob3dc0942014-02-25 10:04:01 -05002407 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2408 tset->cur_task = NULL;
2409
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002410 return cgroup_taskset_next(tset, dst_cssp);
Tejun Heo2f7ee562011-12-12 18:12:21 -08002411}
Tejun Heo2f7ee562011-12-12 18:12:21 -08002412
2413/**
2414 * cgroup_taskset_next - iterate to the next task in taskset
2415 * @tset: taskset of interest
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002416 * @dst_cssp: output variable for the destination css
Tejun Heo2f7ee562011-12-12 18:12:21 -08002417 *
2418 * Return the next task in @tset. Iteration must have been initialized
2419 * with cgroup_taskset_first().
2420 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002421struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2422 struct cgroup_subsys_state **dst_cssp)
Tejun Heo2f7ee562011-12-12 18:12:21 -08002423{
Tejun Heob3dc0942014-02-25 10:04:01 -05002424 struct css_set *cset = tset->cur_cset;
2425 struct task_struct *task = tset->cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002426
Tejun Heob3dc0942014-02-25 10:04:01 -05002427 while (&cset->mg_node != tset->csets) {
2428 if (!task)
2429 task = list_first_entry(&cset->mg_tasks,
2430 struct task_struct, cg_list);
2431 else
2432 task = list_next_entry(task, cg_list);
Tejun Heo2f7ee562011-12-12 18:12:21 -08002433
Tejun Heob3dc0942014-02-25 10:04:01 -05002434 if (&task->cg_list != &cset->mg_tasks) {
2435 tset->cur_cset = cset;
2436 tset->cur_task = task;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002437
2438 /*
2439 * This function may be called both before and
2440 * after cgroup_taskset_migrate(). The two cases
2441 * can be distinguished by looking at whether @cset
2442 * has its ->mg_dst_cset set.
2443 */
2444 if (cset->mg_dst_cset)
2445 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2446 else
2447 *dst_cssp = cset->subsys[tset->ssid];
2448
Tejun Heob3dc0942014-02-25 10:04:01 -05002449 return task;
2450 }
2451
2452 cset = list_next_entry(cset, mg_node);
2453 task = NULL;
2454 }
2455
2456 return NULL;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002457}
Tejun Heo2f7ee562011-12-12 18:12:21 -08002458
2459/**
Tejun Heo37ff9f82016-03-08 11:51:26 -05002460 * cgroup_taskset_migrate - migrate a taskset
Tejun Heoadaae5d2015-09-11 15:00:21 -04002461 * @tset: taget taskset
Tejun Heo37ff9f82016-03-08 11:51:26 -05002462 * @root: cgroup root the migration is taking place on
Tejun Heoadaae5d2015-09-11 15:00:21 -04002463 *
Tejun Heo37ff9f82016-03-08 11:51:26 -05002464 * Migrate tasks in @tset as setup by migration preparation functions.
2465 * This function fails iff one of the ->can_attach callbacks fails and
2466 * guarantees that either all or none of the tasks in @tset are migrated.
2467 * @tset is consumed regardless of success.
Tejun Heoadaae5d2015-09-11 15:00:21 -04002468 */
2469static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
Tejun Heo37ff9f82016-03-08 11:51:26 -05002470 struct cgroup_root *root)
Tejun Heoadaae5d2015-09-11 15:00:21 -04002471{
Tejun Heo37ff9f82016-03-08 11:51:26 -05002472 struct cgroup_subsys *ss;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002473 struct task_struct *task, *tmp_task;
2474 struct css_set *cset, *tmp_cset;
Tejun Heo37ff9f82016-03-08 11:51:26 -05002475 int ssid, failed_ssid, ret;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002476
2477 /* methods shouldn't be called if no task is actually migrating */
2478 if (list_empty(&tset->src_csets))
2479 return 0;
2480
2481 /* check that we can legitimately attach to the cgroup */
Tejun Heo37ff9f82016-03-08 11:51:26 -05002482 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2483 if (ss->can_attach) {
2484 tset->ssid = ssid;
2485 ret = ss->can_attach(tset);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002486 if (ret) {
Tejun Heo37ff9f82016-03-08 11:51:26 -05002487 failed_ssid = ssid;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002488 goto out_cancel_attach;
2489 }
2490 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002491 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002492
2493 /*
2494 * Now that we're guaranteed success, proceed to move all tasks to
2495 * the new cgroup. There are no failure cases after here, so this
2496 * is the commit point.
2497 */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002498 spin_lock_bh(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002499 list_for_each_entry(cset, &tset->src_csets, mg_node) {
Tejun Heof6d7d042015-10-15 16:41:52 -04002500 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2501 struct css_set *from_cset = task_css_set(task);
2502 struct css_set *to_cset = cset->mg_dst_cset;
2503
2504 get_css_set(to_cset);
2505 css_set_move_task(task, from_cset, to_cset, true);
2506 put_css_set_locked(from_cset);
2507 }
Tejun Heoadaae5d2015-09-11 15:00:21 -04002508 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002509 spin_unlock_bh(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002510
2511 /*
2512 * Migration is committed, all target tasks are now on dst_csets.
2513 * Nothing is sensitive to fork() after this point. Notify
2514 * controllers that migration is complete.
2515 */
2516 tset->csets = &tset->dst_csets;
2517
Tejun Heo37ff9f82016-03-08 11:51:26 -05002518 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2519 if (ss->attach) {
2520 tset->ssid = ssid;
2521 ss->attach(tset);
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002522 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002523 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002524
2525 ret = 0;
2526 goto out_release_tset;
2527
2528out_cancel_attach:
Tejun Heo37ff9f82016-03-08 11:51:26 -05002529 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2530 if (ssid == failed_ssid)
Tejun Heoadaae5d2015-09-11 15:00:21 -04002531 break;
Tejun Heo37ff9f82016-03-08 11:51:26 -05002532 if (ss->cancel_attach) {
2533 tset->ssid = ssid;
2534 ss->cancel_attach(tset);
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002535 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002536 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002537out_release_tset:
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002538 spin_lock_bh(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002539 list_splice_init(&tset->dst_csets, &tset->src_csets);
2540 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2541 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2542 list_del_init(&cset->mg_node);
2543 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002544 spin_unlock_bh(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002545 return ret;
2546}
2547
2548/**
Tejun Heo6c694c82016-03-08 11:51:25 -05002549 * cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2550 * @dst_cgrp: destination cgroup to test
2551 *
2552 * On the default hierarchy, except for the root, subtree_control must be
2553 * zero for migration destination cgroups with tasks so that child cgroups
2554 * don't compete against tasks.
2555 */
2556static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2557{
2558 return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) ||
2559 !dst_cgrp->subtree_control;
2560}
2561
2562/**
Tejun Heo1958d2d2014-02-25 10:04:03 -05002563 * cgroup_migrate_finish - cleanup after attach
2564 * @preloaded_csets: list of preloaded css_sets
Ben Blum74a11662011-05-26 16:25:20 -07002565 *
Tejun Heo1958d2d2014-02-25 10:04:03 -05002566 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2567 * those functions for details.
Ben Blum74a11662011-05-26 16:25:20 -07002568 */
Tejun Heo1958d2d2014-02-25 10:04:03 -05002569static void cgroup_migrate_finish(struct list_head *preloaded_csets)
Ben Blum74a11662011-05-26 16:25:20 -07002570{
Tejun Heo1958d2d2014-02-25 10:04:03 -05002571 struct css_set *cset, *tmp_cset;
2572
2573 lockdep_assert_held(&cgroup_mutex);
2574
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002575 spin_lock_bh(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002576 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2577 cset->mg_src_cgrp = NULL;
Tejun Heoe4857982016-03-08 11:51:26 -05002578 cset->mg_dst_cgrp = NULL;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002579 cset->mg_dst_cset = NULL;
2580 list_del_init(&cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002581 put_css_set_locked(cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002582 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002583 spin_unlock_bh(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002584}
2585
2586/**
2587 * cgroup_migrate_add_src - add a migration source css_set
2588 * @src_cset: the source css_set to add
2589 * @dst_cgrp: the destination cgroup
2590 * @preloaded_csets: list of preloaded css_sets
2591 *
2592 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2593 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2594 * up by cgroup_migrate_finish().
2595 *
Tejun Heo1ed13282015-09-16 12:53:17 -04002596 * This function may be called without holding cgroup_threadgroup_rwsem
2597 * even if the target is a process. Threads may be created and destroyed
2598 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2599 * into play and the preloaded css_sets are guaranteed to cover all
2600 * migrations.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002601 */
2602static void cgroup_migrate_add_src(struct css_set *src_cset,
2603 struct cgroup *dst_cgrp,
2604 struct list_head *preloaded_csets)
2605{
2606 struct cgroup *src_cgrp;
2607
2608 lockdep_assert_held(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002609 lockdep_assert_held(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002610
Tejun Heo2b021cb2016-03-15 20:43:04 -04002611 /*
2612 * If ->dead, @src_set is associated with one or more dead cgroups
2613 * and doesn't contain any migratable tasks. Ignore it early so
2614 * that the rest of migration path doesn't get confused by it.
2615 */
2616 if (src_cset->dead)
2617 return;
2618
Tejun Heo1958d2d2014-02-25 10:04:03 -05002619 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2620
Tejun Heo1958d2d2014-02-25 10:04:03 -05002621 if (!list_empty(&src_cset->mg_preload_node))
2622 return;
2623
2624 WARN_ON(src_cset->mg_src_cgrp);
Tejun Heoe4857982016-03-08 11:51:26 -05002625 WARN_ON(src_cset->mg_dst_cgrp);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002626 WARN_ON(!list_empty(&src_cset->mg_tasks));
2627 WARN_ON(!list_empty(&src_cset->mg_node));
2628
2629 src_cset->mg_src_cgrp = src_cgrp;
Tejun Heoe4857982016-03-08 11:51:26 -05002630 src_cset->mg_dst_cgrp = dst_cgrp;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002631 get_css_set(src_cset);
2632 list_add(&src_cset->mg_preload_node, preloaded_csets);
2633}
2634
2635/**
2636 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
Tejun Heo1958d2d2014-02-25 10:04:03 -05002637 * @preloaded_csets: list of preloaded source css_sets
2638 *
Tejun Heoe4857982016-03-08 11:51:26 -05002639 * Tasks are about to be moved and all the source css_sets have been
2640 * preloaded to @preloaded_csets. This function looks up and pins all
2641 * destination css_sets, links each to its source, and append them to
2642 * @preloaded_csets.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002643 *
2644 * This function must be called after cgroup_migrate_add_src() has been
2645 * called on each migration source css_set. After migration is performed
2646 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2647 * @preloaded_csets.
2648 */
Tejun Heoe4857982016-03-08 11:51:26 -05002649static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets)
Tejun Heo1958d2d2014-02-25 10:04:03 -05002650{
2651 LIST_HEAD(csets);
Tejun Heof817de92014-04-23 11:13:16 -04002652 struct css_set *src_cset, *tmp_cset;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002653
2654 lockdep_assert_held(&cgroup_mutex);
2655
2656 /* look up the dst cset for each src cset and link it to src */
Tejun Heof817de92014-04-23 11:13:16 -04002657 list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
Tejun Heo1958d2d2014-02-25 10:04:03 -05002658 struct css_set *dst_cset;
2659
Tejun Heoe4857982016-03-08 11:51:26 -05002660 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002661 if (!dst_cset)
2662 goto err;
2663
2664 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002665
2666 /*
2667 * If src cset equals dst, it's noop. Drop the src.
2668 * cgroup_migrate() will skip the cset too. Note that we
2669 * can't handle src == dst as some nodes are used by both.
2670 */
2671 if (src_cset == dst_cset) {
2672 src_cset->mg_src_cgrp = NULL;
Tejun Heoe4857982016-03-08 11:51:26 -05002673 src_cset->mg_dst_cgrp = NULL;
Tejun Heof817de92014-04-23 11:13:16 -04002674 list_del_init(&src_cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002675 put_css_set(src_cset);
2676 put_css_set(dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002677 continue;
2678 }
2679
Tejun Heo1958d2d2014-02-25 10:04:03 -05002680 src_cset->mg_dst_cset = dst_cset;
2681
2682 if (list_empty(&dst_cset->mg_preload_node))
2683 list_add(&dst_cset->mg_preload_node, &csets);
2684 else
Zefan Lia25eb522014-09-19 16:51:00 +08002685 put_css_set(dst_cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002686 }
2687
Tejun Heof817de92014-04-23 11:13:16 -04002688 list_splice_tail(&csets, preloaded_csets);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002689 return 0;
2690err:
2691 cgroup_migrate_finish(&csets);
2692 return -ENOMEM;
2693}
2694
2695/**
2696 * cgroup_migrate - migrate a process or task to a cgroup
Tejun Heo1958d2d2014-02-25 10:04:03 -05002697 * @leader: the leader of the process or the task to migrate
2698 * @threadgroup: whether @leader points to the whole process or a single task
Tejun Heo37ff9f82016-03-08 11:51:26 -05002699 * @root: cgroup root migration is taking place on
Tejun Heo1958d2d2014-02-25 10:04:03 -05002700 *
Tejun Heo37ff9f82016-03-08 11:51:26 -05002701 * Migrate a process or task denoted by @leader. If migrating a process,
2702 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2703 * responsible for invoking cgroup_migrate_add_src() and
Tejun Heo1958d2d2014-02-25 10:04:03 -05002704 * cgroup_migrate_prepare_dst() on the targets before invoking this
2705 * function and following up with cgroup_migrate_finish().
2706 *
2707 * As long as a controller's ->can_attach() doesn't fail, this function is
2708 * guaranteed to succeed. This means that, excluding ->can_attach()
2709 * failure, when migrating multiple targets, the success or failure can be
2710 * decided for all targets by invoking group_migrate_prepare_dst() before
2711 * actually starting migrating.
2712 */
Tejun Heo9af2ec42015-09-11 15:00:20 -04002713static int cgroup_migrate(struct task_struct *leader, bool threadgroup,
Tejun Heo37ff9f82016-03-08 11:51:26 -05002714 struct cgroup_root *root)
Ben Blum74a11662011-05-26 16:25:20 -07002715{
Tejun Heoadaae5d2015-09-11 15:00:21 -04002716 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2717 struct task_struct *task;
Ben Blum74a11662011-05-26 16:25:20 -07002718
2719 /*
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002720 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2721 * already PF_EXITING could be freed from underneath us unless we
2722 * take an rcu_read_lock.
2723 */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002724 spin_lock_bh(&css_set_lock);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002725 rcu_read_lock();
Tejun Heo9db8de32014-02-13 06:58:43 -05002726 task = leader;
Ben Blum74a11662011-05-26 16:25:20 -07002727 do {
Tejun Heoadaae5d2015-09-11 15:00:21 -04002728 cgroup_taskset_add(task, &tset);
Li Zefan081aa452013-03-13 09:17:09 +08002729 if (!threadgroup)
2730 break;
Tejun Heo9db8de32014-02-13 06:58:43 -05002731 } while_each_thread(leader, task);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002732 rcu_read_unlock();
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002733 spin_unlock_bh(&css_set_lock);
Ben Blum74a11662011-05-26 16:25:20 -07002734
Tejun Heo37ff9f82016-03-08 11:51:26 -05002735 return cgroup_taskset_migrate(&tset, root);
Ben Blum74a11662011-05-26 16:25:20 -07002736}
2737
Tejun Heo1958d2d2014-02-25 10:04:03 -05002738/**
2739 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2740 * @dst_cgrp: the cgroup to attach to
2741 * @leader: the task or the leader of the threadgroup to be attached
2742 * @threadgroup: attach the whole threadgroup?
2743 *
Tejun Heo1ed13282015-09-16 12:53:17 -04002744 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002745 */
2746static int cgroup_attach_task(struct cgroup *dst_cgrp,
2747 struct task_struct *leader, bool threadgroup)
2748{
2749 LIST_HEAD(preloaded_csets);
2750 struct task_struct *task;
2751 int ret;
2752
Tejun Heo6c694c82016-03-08 11:51:25 -05002753 if (!cgroup_may_migrate_to(dst_cgrp))
2754 return -EBUSY;
2755
Tejun Heo1958d2d2014-02-25 10:04:03 -05002756 /* look up all src csets */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002757 spin_lock_bh(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002758 rcu_read_lock();
2759 task = leader;
2760 do {
2761 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2762 &preloaded_csets);
2763 if (!threadgroup)
2764 break;
2765 } while_each_thread(leader, task);
2766 rcu_read_unlock();
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002767 spin_unlock_bh(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002768
2769 /* prepare dst csets and commit */
Tejun Heoe4857982016-03-08 11:51:26 -05002770 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002771 if (!ret)
Tejun Heo37ff9f82016-03-08 11:51:26 -05002772 ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002773
2774 cgroup_migrate_finish(&preloaded_csets);
2775 return ret;
Ben Blum74a11662011-05-26 16:25:20 -07002776}
2777
Tejun Heo187fe842015-06-18 16:54:28 -04002778static int cgroup_procs_write_permission(struct task_struct *task,
2779 struct cgroup *dst_cgrp,
2780 struct kernfs_open_file *of)
Tejun Heodedf22e2015-06-18 16:54:28 -04002781{
2782 const struct cred *cred = current_cred();
2783 const struct cred *tcred = get_task_cred(task);
2784 int ret = 0;
2785
2786 /*
2787 * even if we're attaching all tasks in the thread group, we only
2788 * need to check permissions on one of them.
2789 */
2790 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2791 !uid_eq(cred->euid, tcred->uid) &&
2792 !uid_eq(cred->euid, tcred->suid))
2793 ret = -EACCES;
2794
Tejun Heo187fe842015-06-18 16:54:28 -04002795 if (!ret && cgroup_on_dfl(dst_cgrp)) {
2796 struct super_block *sb = of->file->f_path.dentry->d_sb;
2797 struct cgroup *cgrp;
2798 struct inode *inode;
2799
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002800 spin_lock_bh(&css_set_lock);
Tejun Heo187fe842015-06-18 16:54:28 -04002801 cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002802 spin_unlock_bh(&css_set_lock);
Tejun Heo187fe842015-06-18 16:54:28 -04002803
2804 while (!cgroup_is_descendant(dst_cgrp, cgrp))
2805 cgrp = cgroup_parent(cgrp);
2806
2807 ret = -ENOMEM;
Tejun Heo6f60ead2015-09-18 17:54:23 -04002808 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
Tejun Heo187fe842015-06-18 16:54:28 -04002809 if (inode) {
2810 ret = inode_permission(inode, MAY_WRITE);
2811 iput(inode);
2812 }
2813 }
2814
Tejun Heodedf22e2015-06-18 16:54:28 -04002815 put_cred(tcred);
2816 return ret;
2817}
2818
Ben Blum74a11662011-05-26 16:25:20 -07002819/*
2820 * Find the task_struct of the task to attach by vpid and pass it along to the
Tejun Heocd3d0952011-12-12 18:12:21 -08002821 * function to attach either it or all tasks in its threadgroup. Will lock
Tejun Heo0e1d7682014-02-25 10:04:03 -05002822 * cgroup_mutex and threadgroup.
Ben Blum74a11662011-05-26 16:25:20 -07002823 */
Tejun Heoacbef752014-05-13 12:16:22 -04002824static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2825 size_t nbytes, loff_t off, bool threadgroup)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002826{
Paul Menagebbcb81d2007-10-18 23:39:32 -07002827 struct task_struct *tsk;
Tejun Heoe76ecae2014-05-13 12:19:23 -04002828 struct cgroup *cgrp;
Tejun Heoacbef752014-05-13 12:16:22 -04002829 pid_t pid;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002830 int ret;
2831
Tejun Heoacbef752014-05-13 12:16:22 -04002832 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2833 return -EINVAL;
2834
Tejun Heo945ba192016-03-03 09:58:00 -05002835 cgrp = cgroup_kn_lock_live(of->kn, false);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002836 if (!cgrp)
Ben Blum74a11662011-05-26 16:25:20 -07002837 return -ENODEV;
2838
Tejun Heo3014dde2015-09-16 13:03:02 -04002839 percpu_down_write(&cgroup_threadgroup_rwsem);
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002840 rcu_read_lock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07002841 if (pid) {
Pavel Emelyanov73507f32008-02-07 00:14:47 -08002842 tsk = find_task_by_vpid(pid);
Ben Blum74a11662011-05-26 16:25:20 -07002843 if (!tsk) {
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09002844 ret = -ESRCH;
Tejun Heo3014dde2015-09-16 13:03:02 -04002845 goto out_unlock_rcu;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002846 }
Tejun Heodedf22e2015-06-18 16:54:28 -04002847 } else {
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002848 tsk = current;
Tejun Heodedf22e2015-06-18 16:54:28 -04002849 }
Tejun Heocd3d0952011-12-12 18:12:21 -08002850
2851 if (threadgroup)
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002852 tsk = tsk->group_leader;
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002853
2854 /*
Tejun Heo14a40ff2013-03-19 13:45:20 -07002855 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002856 * trapped in a cpuset, or RT worker may be born in a cgroup
2857 * with no rt_runtime allocated. Just say no.
2858 */
Tejun Heo14a40ff2013-03-19 13:45:20 -07002859 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002860 ret = -EINVAL;
Tejun Heo3014dde2015-09-16 13:03:02 -04002861 goto out_unlock_rcu;
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002862 }
2863
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002864 get_task_struct(tsk);
2865 rcu_read_unlock();
Tejun Heocd3d0952011-12-12 18:12:21 -08002866
Tejun Heo187fe842015-06-18 16:54:28 -04002867 ret = cgroup_procs_write_permission(tsk, cgrp, of);
Tejun Heodedf22e2015-06-18 16:54:28 -04002868 if (!ret)
2869 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
Li Zefan081aa452013-03-13 09:17:09 +08002870
Tejun Heof9f9e7b2015-09-16 11:51:12 -04002871 put_task_struct(tsk);
Tejun Heo3014dde2015-09-16 13:03:02 -04002872 goto out_unlock_threadgroup;
2873
2874out_unlock_rcu:
2875 rcu_read_unlock();
2876out_unlock_threadgroup:
2877 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002878 cgroup_kn_unlock(of->kn);
Tejun Heoe93ad192016-01-19 12:18:41 -05002879 cpuset_post_attach_flush();
Tejun Heoacbef752014-05-13 12:16:22 -04002880 return ret ?: nbytes;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002881}
2882
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002883/**
2884 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2885 * @from: attach to all cgroups of a given task
2886 * @tsk: the task to be attached
2887 */
2888int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2889{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002890 struct cgroup_root *root;
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002891 int retval = 0;
2892
Tejun Heo47cfcd02013-04-07 09:29:51 -07002893 mutex_lock(&cgroup_mutex);
Tejun Heo985ed672014-03-19 10:23:53 -04002894 for_each_root(root) {
Tejun Heo96d365e2014-02-13 06:58:40 -05002895 struct cgroup *from_cgrp;
2896
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002897 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04002898 continue;
2899
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002900 spin_lock_bh(&css_set_lock);
Tejun Heo96d365e2014-02-13 06:58:40 -05002901 from_cgrp = task_cgroup_from_root(from, root);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002902 spin_unlock_bh(&css_set_lock);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002903
Li Zefan6f4b7e62013-07-31 16:18:36 +08002904 retval = cgroup_attach_task(from_cgrp, tsk, false);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002905 if (retval)
2906 break;
2907 }
Tejun Heo47cfcd02013-04-07 09:29:51 -07002908 mutex_unlock(&cgroup_mutex);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002909
2910 return retval;
2911}
2912EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2913
Tejun Heoacbef752014-05-13 12:16:22 -04002914static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2915 char *buf, size_t nbytes, loff_t off)
Paul Menageaf351022008-07-25 01:47:01 -07002916{
Tejun Heoacbef752014-05-13 12:16:22 -04002917 return __cgroup_procs_write(of, buf, nbytes, off, false);
Ben Blum74a11662011-05-26 16:25:20 -07002918}
2919
Tejun Heoacbef752014-05-13 12:16:22 -04002920static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2921 char *buf, size_t nbytes, loff_t off)
Ben Blum74a11662011-05-26 16:25:20 -07002922{
Tejun Heoacbef752014-05-13 12:16:22 -04002923 return __cgroup_procs_write(of, buf, nbytes, off, true);
Paul Menageaf351022008-07-25 01:47:01 -07002924}
2925
Tejun Heo451af5042014-05-13 12:16:21 -04002926static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2927 char *buf, size_t nbytes, loff_t off)
Paul Menagee788e0662008-07-25 01:46:59 -07002928{
Tejun Heoe76ecae2014-05-13 12:19:23 -04002929 struct cgroup *cgrp;
Tejun Heo5f469902014-02-11 11:52:48 -05002930
Tejun Heoe76ecae2014-05-13 12:19:23 -04002931 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2932
Tejun Heo945ba192016-03-03 09:58:00 -05002933 cgrp = cgroup_kn_lock_live(of->kn, false);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002934 if (!cgrp)
Paul Menagee788e0662008-07-25 01:46:59 -07002935 return -ENODEV;
Tejun Heo69e943b2014-02-08 10:36:58 -05002936 spin_lock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002937 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
2938 sizeof(cgrp->root->release_agent_path));
Tejun Heo69e943b2014-02-08 10:36:58 -05002939 spin_unlock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002940 cgroup_kn_unlock(of->kn);
Tejun Heo451af5042014-05-13 12:16:21 -04002941 return nbytes;
Paul Menagee788e0662008-07-25 01:46:59 -07002942}
2943
Tejun Heo2da8ca82013-12-05 12:28:04 -05002944static int cgroup_release_agent_show(struct seq_file *seq, void *v)
Paul Menagee788e0662008-07-25 01:46:59 -07002945{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002946 struct cgroup *cgrp = seq_css(seq)->cgroup;
Tejun Heo182446d2013-08-08 20:11:24 -04002947
Tejun Heo46cfeb02014-05-13 12:11:00 -04002948 spin_lock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07002949 seq_puts(seq, cgrp->root->release_agent_path);
Tejun Heo46cfeb02014-05-13 12:11:00 -04002950 spin_unlock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07002951 seq_putc(seq, '\n');
Paul Menagee788e0662008-07-25 01:46:59 -07002952 return 0;
2953}
2954
Tejun Heo2da8ca82013-12-05 12:28:04 -05002955static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
Tejun Heo873fe092013-04-14 20:15:26 -07002956{
Tejun Heoc1d5d422014-07-09 10:08:08 -04002957 seq_puts(seq, "0\n");
Paul Menage81a6a5c2007-10-18 23:39:38 -07002958 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002959}
2960
Tejun Heo6e5c8302016-02-22 22:25:47 -05002961static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
Tejun Heof8f22e52014-04-23 11:13:16 -04002962{
2963 struct cgroup_subsys *ss;
2964 bool printed = false;
2965 int ssid;
2966
Tejun Heob4e0eea2016-02-22 22:25:46 -05002967 do_each_subsys_mask(ss, ssid, ss_mask) {
Aleksa Saraia966a4e2015-06-06 10:02:15 +10002968 if (printed)
2969 seq_putc(seq, ' ');
2970 seq_printf(seq, "%s", ss->name);
2971 printed = true;
Tejun Heob4e0eea2016-02-22 22:25:46 -05002972 } while_each_subsys_mask();
Tejun Heof8f22e52014-04-23 11:13:16 -04002973 if (printed)
2974 seq_putc(seq, '\n');
2975}
2976
Tejun Heof8f22e52014-04-23 11:13:16 -04002977/* show controllers which are enabled from the parent */
2978static int cgroup_controllers_show(struct seq_file *seq, void *v)
2979{
2980 struct cgroup *cgrp = seq_css(seq)->cgroup;
2981
Tejun Heo5531dc92016-03-03 09:57:58 -05002982 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
Tejun Heof8f22e52014-04-23 11:13:16 -04002983 return 0;
2984}
2985
2986/* show controllers which are enabled for a given cgroup's children */
2987static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2988{
2989 struct cgroup *cgrp = seq_css(seq)->cgroup;
2990
Tejun Heo667c2492014-07-08 18:02:56 -04002991 cgroup_print_ss_mask(seq, cgrp->subtree_control);
Tejun Heof8f22e52014-04-23 11:13:16 -04002992 return 0;
2993}
2994
2995/**
2996 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2997 * @cgrp: root of the subtree to update csses for
2998 *
Tejun Heo54962602016-03-03 09:58:01 -05002999 * @cgrp's control masks have changed and its subtree's css associations
3000 * need to be updated accordingly. This function looks up all css_sets
3001 * which are attached to the subtree, creates the matching updated css_sets
3002 * and migrates the tasks to the new ones.
Tejun Heof8f22e52014-04-23 11:13:16 -04003003 */
3004static int cgroup_update_dfl_csses(struct cgroup *cgrp)
3005{
3006 LIST_HEAD(preloaded_csets);
Tejun Heo10265072015-09-11 15:00:22 -04003007 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
Tejun Heo54962602016-03-03 09:58:01 -05003008 struct cgroup_subsys_state *d_css;
3009 struct cgroup *dsct;
Tejun Heof8f22e52014-04-23 11:13:16 -04003010 struct css_set *src_cset;
3011 int ret;
3012
Tejun Heof8f22e52014-04-23 11:13:16 -04003013 lockdep_assert_held(&cgroup_mutex);
3014
Tejun Heo3014dde2015-09-16 13:03:02 -04003015 percpu_down_write(&cgroup_threadgroup_rwsem);
3016
Tejun Heof8f22e52014-04-23 11:13:16 -04003017 /* look up all csses currently attached to @cgrp's subtree */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003018 spin_lock_bh(&css_set_lock);
Tejun Heo54962602016-03-03 09:58:01 -05003019 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003020 struct cgrp_cset_link *link;
3021
Tejun Heo54962602016-03-03 09:58:01 -05003022 list_for_each_entry(link, &dsct->cset_links, cset_link)
Tejun Heo58cdb1c2016-03-08 11:51:25 -05003023 cgroup_migrate_add_src(link->cset, dsct,
Tejun Heof8f22e52014-04-23 11:13:16 -04003024 &preloaded_csets);
3025 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003026 spin_unlock_bh(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003027
3028 /* NULL dst indicates self on default hierarchy */
Tejun Heoe4857982016-03-08 11:51:26 -05003029 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heof8f22e52014-04-23 11:13:16 -04003030 if (ret)
3031 goto out_finish;
3032
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003033 spin_lock_bh(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003034 list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
Tejun Heo10265072015-09-11 15:00:22 -04003035 struct task_struct *task, *ntask;
Tejun Heof8f22e52014-04-23 11:13:16 -04003036
3037 /* src_csets precede dst_csets, break on the first dst_cset */
3038 if (!src_cset->mg_src_cgrp)
3039 break;
3040
Tejun Heo10265072015-09-11 15:00:22 -04003041 /* all tasks in src_csets need to be migrated */
3042 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3043 cgroup_taskset_add(task, &tset);
Tejun Heof8f22e52014-04-23 11:13:16 -04003044 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003045 spin_unlock_bh(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003046
Tejun Heo37ff9f82016-03-08 11:51:26 -05003047 ret = cgroup_taskset_migrate(&tset, cgrp->root);
Tejun Heof8f22e52014-04-23 11:13:16 -04003048out_finish:
3049 cgroup_migrate_finish(&preloaded_csets);
Tejun Heo3014dde2015-09-16 13:03:02 -04003050 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heof8f22e52014-04-23 11:13:16 -04003051 return ret;
3052}
3053
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003054/**
Tejun Heo945ba192016-03-03 09:58:00 -05003055 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
Tejun Heoce3f1d92016-03-03 09:57:59 -05003056 * @cgrp: root of the target subtree
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003057 *
3058 * Because css offlining is asynchronous, userland may try to re-enable a
Tejun Heo945ba192016-03-03 09:58:00 -05003059 * controller while the previous css is still around. This function grabs
3060 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003061 */
Tejun Heo945ba192016-03-03 09:58:00 -05003062static void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3063 __acquires(&cgroup_mutex)
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003064{
3065 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003066 struct cgroup_subsys_state *d_css;
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003067 struct cgroup_subsys *ss;
3068 int ssid;
3069
Tejun Heo945ba192016-03-03 09:58:00 -05003070restart:
3071 mutex_lock(&cgroup_mutex);
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003072
Tejun Heoce3f1d92016-03-03 09:57:59 -05003073 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003074 for_each_subsys(ss, ssid) {
3075 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3076 DEFINE_WAIT(wait);
3077
Tejun Heoce3f1d92016-03-03 09:57:59 -05003078 if (!css || !percpu_ref_is_dying(&css->refcnt))
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003079 continue;
3080
3081 cgroup_get(dsct);
3082 prepare_to_wait(&dsct->offline_waitq, &wait,
3083 TASK_UNINTERRUPTIBLE);
3084
3085 mutex_unlock(&cgroup_mutex);
3086 schedule();
3087 finish_wait(&dsct->offline_waitq, &wait);
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003088
3089 cgroup_put(dsct);
Tejun Heo945ba192016-03-03 09:58:00 -05003090 goto restart;
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003091 }
3092 }
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003093}
3094
Tejun Heo12b3bb62016-03-03 09:57:59 -05003095/**
Tejun Heo15a27c32016-03-03 09:57:59 -05003096 * cgroup_save_control - save control masks of a subtree
3097 * @cgrp: root of the target subtree
3098 *
3099 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
3100 * prefixed fields for @cgrp's subtree including @cgrp itself.
3101 */
3102static void cgroup_save_control(struct cgroup *cgrp)
3103{
3104 struct cgroup *dsct;
3105 struct cgroup_subsys_state *d_css;
3106
3107 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3108 dsct->old_subtree_control = dsct->subtree_control;
3109 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3110 }
3111}
3112
3113/**
3114 * cgroup_propagate_control - refresh control masks of a subtree
3115 * @cgrp: root of the target subtree
3116 *
3117 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3118 * ->subtree_control and propagate controller availability through the
3119 * subtree so that descendants don't have unavailable controllers enabled.
3120 */
3121static void cgroup_propagate_control(struct cgroup *cgrp)
3122{
3123 struct cgroup *dsct;
3124 struct cgroup_subsys_state *d_css;
3125
3126 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3127 dsct->subtree_control &= cgroup_control(dsct);
Tejun Heo5ced2512016-03-03 09:58:01 -05003128 dsct->subtree_ss_mask =
3129 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3130 cgroup_ss_mask(dsct));
Tejun Heo15a27c32016-03-03 09:57:59 -05003131 }
3132}
3133
3134/**
3135 * cgroup_restore_control - restore control masks of a subtree
3136 * @cgrp: root of the target subtree
3137 *
3138 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
3139 * prefixed fields for @cgrp's subtree including @cgrp itself.
3140 */
3141static void cgroup_restore_control(struct cgroup *cgrp)
3142{
3143 struct cgroup *dsct;
3144 struct cgroup_subsys_state *d_css;
3145
3146 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3147 dsct->subtree_control = dsct->old_subtree_control;
3148 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3149 }
3150}
3151
Tejun Heof6d635ad2016-03-08 11:51:26 -05003152static bool css_visible(struct cgroup_subsys_state *css)
3153{
3154 struct cgroup_subsys *ss = css->ss;
3155 struct cgroup *cgrp = css->cgroup;
3156
3157 if (cgroup_control(cgrp) & (1 << ss->id))
3158 return true;
3159 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3160 return false;
3161 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3162}
3163
Tejun Heo15a27c32016-03-03 09:57:59 -05003164/**
Tejun Heobdb53bd2016-03-03 09:57:59 -05003165 * cgroup_apply_control_enable - enable or show csses according to control
Tejun Heoce3f1d92016-03-03 09:57:59 -05003166 * @cgrp: root of the target subtree
Tejun Heobdb53bd2016-03-03 09:57:59 -05003167 *
Tejun Heoce3f1d92016-03-03 09:57:59 -05003168 * Walk @cgrp's subtree and create new csses or make the existing ones
Tejun Heobdb53bd2016-03-03 09:57:59 -05003169 * visible. A css is created invisible if it's being implicitly enabled
3170 * through dependency. An invisible css is made visible when the userland
3171 * explicitly enables it.
3172 *
3173 * Returns 0 on success, -errno on failure. On failure, csses which have
3174 * been processed already aren't cleaned up. The caller is responsible for
3175 * cleaning up with cgroup_apply_control_disble().
3176 */
3177static int cgroup_apply_control_enable(struct cgroup *cgrp)
3178{
3179 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003180 struct cgroup_subsys_state *d_css;
Tejun Heobdb53bd2016-03-03 09:57:59 -05003181 struct cgroup_subsys *ss;
3182 int ssid, ret;
3183
Tejun Heoce3f1d92016-03-03 09:57:59 -05003184 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
Tejun Heobdb53bd2016-03-03 09:57:59 -05003185 for_each_subsys(ss, ssid) {
3186 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3187
Tejun Heo945ba192016-03-03 09:58:00 -05003188 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3189
Tejun Heobdb53bd2016-03-03 09:57:59 -05003190 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3191 continue;
3192
3193 if (!css) {
3194 css = css_create(dsct, ss);
3195 if (IS_ERR(css))
3196 return PTR_ERR(css);
3197 }
3198
Tejun Heof6d635ad2016-03-08 11:51:26 -05003199 if (css_visible(css)) {
Tejun Heo334c3672016-03-03 09:58:01 -05003200 ret = css_populate_dir(css);
Tejun Heobdb53bd2016-03-03 09:57:59 -05003201 if (ret)
3202 return ret;
3203 }
3204 }
3205 }
3206
3207 return 0;
3208}
3209
3210/**
Tejun Heo12b3bb62016-03-03 09:57:59 -05003211 * cgroup_apply_control_disable - kill or hide csses according to control
Tejun Heoce3f1d92016-03-03 09:57:59 -05003212 * @cgrp: root of the target subtree
Tejun Heo12b3bb62016-03-03 09:57:59 -05003213 *
Tejun Heoce3f1d92016-03-03 09:57:59 -05003214 * Walk @cgrp's subtree and kill and hide csses so that they match
Tejun Heo12b3bb62016-03-03 09:57:59 -05003215 * cgroup_ss_mask() and cgroup_visible_mask().
3216 *
3217 * A css is hidden when the userland requests it to be disabled while other
3218 * subsystems are still depending on it. The css must not actively control
3219 * resources and be in the vanilla state if it's made visible again later.
3220 * Controllers which may be depended upon should provide ->css_reset() for
3221 * this purpose.
3222 */
3223static void cgroup_apply_control_disable(struct cgroup *cgrp)
3224{
3225 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003226 struct cgroup_subsys_state *d_css;
Tejun Heo12b3bb62016-03-03 09:57:59 -05003227 struct cgroup_subsys *ss;
3228 int ssid;
3229
Tejun Heoce3f1d92016-03-03 09:57:59 -05003230 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
Tejun Heo12b3bb62016-03-03 09:57:59 -05003231 for_each_subsys(ss, ssid) {
3232 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3233
Tejun Heo945ba192016-03-03 09:58:00 -05003234 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3235
Tejun Heo12b3bb62016-03-03 09:57:59 -05003236 if (!css)
3237 continue;
3238
Tejun Heo334c3672016-03-03 09:58:01 -05003239 if (css->parent &&
3240 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
Tejun Heo12b3bb62016-03-03 09:57:59 -05003241 kill_css(css);
Tejun Heof6d635ad2016-03-08 11:51:26 -05003242 } else if (!css_visible(css)) {
Tejun Heo334c3672016-03-03 09:58:01 -05003243 css_clear_dir(css);
Tejun Heo12b3bb62016-03-03 09:57:59 -05003244 if (ss->css_reset)
3245 ss->css_reset(css);
3246 }
3247 }
3248 }
3249}
3250
Tejun Heof7b28142016-03-03 09:58:00 -05003251/**
3252 * cgroup_apply_control - apply control mask updates to the subtree
3253 * @cgrp: root of the target subtree
3254 *
3255 * subsystems can be enabled and disabled in a subtree using the following
3256 * steps.
3257 *
3258 * 1. Call cgroup_save_control() to stash the current state.
3259 * 2. Update ->subtree_control masks in the subtree as desired.
3260 * 3. Call cgroup_apply_control() to apply the changes.
3261 * 4. Optionally perform other related operations.
3262 * 5. Call cgroup_finalize_control() to finish up.
3263 *
3264 * This function implements step 3 and propagates the mask changes
3265 * throughout @cgrp's subtree, updates csses accordingly and perform
3266 * process migrations.
3267 */
3268static int cgroup_apply_control(struct cgroup *cgrp)
3269{
3270 int ret;
3271
3272 cgroup_propagate_control(cgrp);
3273
3274 ret = cgroup_apply_control_enable(cgrp);
3275 if (ret)
3276 return ret;
3277
3278 /*
3279 * At this point, cgroup_e_css() results reflect the new csses
3280 * making the following cgroup_update_dfl_csses() properly update
3281 * css associations of all tasks in the subtree.
3282 */
3283 ret = cgroup_update_dfl_csses(cgrp);
3284 if (ret)
3285 return ret;
3286
3287 return 0;
3288}
3289
3290/**
3291 * cgroup_finalize_control - finalize control mask update
3292 * @cgrp: root of the target subtree
3293 * @ret: the result of the update
3294 *
3295 * Finalize control mask update. See cgroup_apply_control() for more info.
3296 */
3297static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3298{
3299 if (ret) {
3300 cgroup_restore_control(cgrp);
3301 cgroup_propagate_control(cgrp);
3302 }
3303
3304 cgroup_apply_control_disable(cgrp);
3305}
3306
Tejun Heof8f22e52014-04-23 11:13:16 -04003307/* change the enabled child controllers for a cgroup in the default hierarchy */
Tejun Heo451af5042014-05-13 12:16:21 -04003308static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3309 char *buf, size_t nbytes,
3310 loff_t off)
Tejun Heof8f22e52014-04-23 11:13:16 -04003311{
Tejun Heo6e5c8302016-02-22 22:25:47 -05003312 u16 enable = 0, disable = 0;
Tejun Heoa9746d82014-05-13 12:19:22 -04003313 struct cgroup *cgrp, *child;
Tejun Heof8f22e52014-04-23 11:13:16 -04003314 struct cgroup_subsys *ss;
Tejun Heo451af5042014-05-13 12:16:21 -04003315 char *tok;
Tejun Heof8f22e52014-04-23 11:13:16 -04003316 int ssid, ret;
3317
3318 /*
Tejun Heod37167a2014-05-13 12:10:59 -04003319 * Parse input - space separated list of subsystem names prefixed
3320 * with either + or -.
Tejun Heof8f22e52014-04-23 11:13:16 -04003321 */
Tejun Heo451af5042014-05-13 12:16:21 -04003322 buf = strstrip(buf);
3323 while ((tok = strsep(&buf, " "))) {
Tejun Heod37167a2014-05-13 12:10:59 -04003324 if (tok[0] == '\0')
3325 continue;
Tejun Heoa7165262016-02-23 10:00:50 -05003326 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
Tejun Heofc5ed1e2015-09-18 11:56:28 -04003327 if (!cgroup_ssid_enabled(ssid) ||
3328 strcmp(tok + 1, ss->name))
Tejun Heof8f22e52014-04-23 11:13:16 -04003329 continue;
3330
3331 if (*tok == '+') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04003332 enable |= 1 << ssid;
3333 disable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04003334 } else if (*tok == '-') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04003335 disable |= 1 << ssid;
3336 enable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04003337 } else {
3338 return -EINVAL;
3339 }
3340 break;
Tejun Heob4e0eea2016-02-22 22:25:46 -05003341 } while_each_subsys_mask();
Tejun Heof8f22e52014-04-23 11:13:16 -04003342 if (ssid == CGROUP_SUBSYS_COUNT)
3343 return -EINVAL;
3344 }
3345
Tejun Heo945ba192016-03-03 09:58:00 -05003346 cgrp = cgroup_kn_lock_live(of->kn, true);
Tejun Heoa9746d82014-05-13 12:19:22 -04003347 if (!cgrp)
3348 return -ENODEV;
Tejun Heof8f22e52014-04-23 11:13:16 -04003349
3350 for_each_subsys(ss, ssid) {
3351 if (enable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04003352 if (cgrp->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003353 enable &= ~(1 << ssid);
3354 continue;
3355 }
3356
Tejun Heo5531dc92016-03-03 09:57:58 -05003357 if (!(cgroup_control(cgrp) & (1 << ssid))) {
Tejun Heoc29adf22014-07-08 18:02:56 -04003358 ret = -ENOENT;
3359 goto out_unlock;
3360 }
Tejun Heof8f22e52014-04-23 11:13:16 -04003361 } else if (disable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04003362 if (!(cgrp->subtree_control & (1 << ssid))) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003363 disable &= ~(1 << ssid);
3364 continue;
3365 }
3366
3367 /* a child has it enabled? */
3368 cgroup_for_each_live_child(child, cgrp) {
Tejun Heo667c2492014-07-08 18:02:56 -04003369 if (child->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003370 ret = -EBUSY;
Tejun Heoddab2b62014-05-13 12:19:22 -04003371 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04003372 }
3373 }
3374 }
3375 }
3376
3377 if (!enable && !disable) {
3378 ret = 0;
Tejun Heoddab2b62014-05-13 12:19:22 -04003379 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04003380 }
3381
3382 /*
Tejun Heo667c2492014-07-08 18:02:56 -04003383 * Except for the root, subtree_control must be zero for a cgroup
Tejun Heof8f22e52014-04-23 11:13:16 -04003384 * with tasks so that child cgroups don't compete against tasks.
3385 */
Tejun Heod51f39b2014-05-16 13:22:48 -04003386 if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003387 ret = -EBUSY;
3388 goto out_unlock;
3389 }
3390
Tejun Heo15a27c32016-03-03 09:57:59 -05003391 /* save and update control masks and prepare csses */
3392 cgroup_save_control(cgrp);
Tejun Heoc29adf22014-07-08 18:02:56 -04003393
Tejun Heo15a27c32016-03-03 09:57:59 -05003394 cgrp->subtree_control |= enable;
3395 cgrp->subtree_control &= ~disable;
Tejun Heof63070d2014-07-08 18:02:57 -04003396
Tejun Heof7b28142016-03-03 09:58:00 -05003397 ret = cgroup_apply_control(cgrp);
Tejun Heo755bf5e2014-11-18 02:49:50 -05003398
Tejun Heof7b28142016-03-03 09:58:00 -05003399 cgroup_finalize_control(cgrp, ret);
Tejun Heof8f22e52014-04-23 11:13:16 -04003400
3401 kernfs_activate(cgrp->kn);
3402 ret = 0;
3403out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04003404 cgroup_kn_unlock(of->kn);
Tejun Heo451af5042014-05-13 12:16:21 -04003405 return ret ?: nbytes;
Tejun Heof8f22e52014-04-23 11:13:16 -04003406}
3407
Tejun Heo4a07c222015-09-18 17:54:22 -04003408static int cgroup_events_show(struct seq_file *seq, void *v)
Tejun Heo842b5972014-04-25 18:28:02 -04003409{
Tejun Heo4a07c222015-09-18 17:54:22 -04003410 seq_printf(seq, "populated %d\n",
Tejun Heo27bd4db2015-10-15 16:41:50 -04003411 cgroup_is_populated(seq_css(seq)->cgroup));
Tejun Heo842b5972014-04-25 18:28:02 -04003412 return 0;
3413}
3414
Tejun Heo2bd59d42014-02-11 11:52:49 -05003415static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3416 size_t nbytes, loff_t off)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003417{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003418 struct cgroup *cgrp = of->kn->parent->priv;
3419 struct cftype *cft = of->kn->priv;
3420 struct cgroup_subsys_state *css;
Tejun Heoa742c592013-12-05 12:28:03 -05003421 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003422
Tejun Heob4168642014-05-13 12:16:21 -04003423 if (cft->write)
3424 return cft->write(of, buf, nbytes, off);
3425
Tejun Heo2bd59d42014-02-11 11:52:49 -05003426 /*
3427 * kernfs guarantees that a file isn't deleted with operations in
3428 * flight, which means that the matching css is and stays alive and
3429 * doesn't need to be pinned. The RCU locking is not necessary
3430 * either. It's just for the convenience of using cgroup_css().
3431 */
3432 rcu_read_lock();
3433 css = cgroup_css(cgrp, cft->ss);
3434 rcu_read_unlock();
Paul Menageddbcc7e2007-10-18 23:39:30 -07003435
Tejun Heo451af5042014-05-13 12:16:21 -04003436 if (cft->write_u64) {
Tejun Heoa742c592013-12-05 12:28:03 -05003437 unsigned long long v;
3438 ret = kstrtoull(buf, 0, &v);
3439 if (!ret)
3440 ret = cft->write_u64(css, cft, v);
3441 } else if (cft->write_s64) {
3442 long long v;
3443 ret = kstrtoll(buf, 0, &v);
3444 if (!ret)
3445 ret = cft->write_s64(css, cft, v);
Tejun Heoa742c592013-12-05 12:28:03 -05003446 } else {
3447 ret = -EINVAL;
3448 }
Tejun Heo2bd59d42014-02-11 11:52:49 -05003449
Tejun Heoa742c592013-12-05 12:28:03 -05003450 return ret ?: nbytes;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003451}
3452
Tejun Heo6612f052013-12-05 12:28:04 -05003453static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
Paul Menage91796562008-04-29 01:00:01 -07003454{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003455 return seq_cft(seq)->seq_start(seq, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05003456}
3457
3458static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3459{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003460 return seq_cft(seq)->seq_next(seq, v, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05003461}
3462
3463static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3464{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003465 seq_cft(seq)->seq_stop(seq, v);
Paul Menage91796562008-04-29 01:00:01 -07003466}
3467
3468static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3469{
Tejun Heo7da11272013-12-05 12:28:04 -05003470 struct cftype *cft = seq_cft(m);
3471 struct cgroup_subsys_state *css = seq_css(m);
Li Zefane0798ce2013-07-31 17:36:25 +08003472
Tejun Heo2da8ca82013-12-05 12:28:04 -05003473 if (cft->seq_show)
3474 return cft->seq_show(m, arg);
Paul Menage91796562008-04-29 01:00:01 -07003475
Tejun Heo896f5192013-12-05 12:28:04 -05003476 if (cft->read_u64)
3477 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3478 else if (cft->read_s64)
3479 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3480 else
3481 return -EINVAL;
3482 return 0;
Paul Menage91796562008-04-29 01:00:01 -07003483}
3484
Tejun Heo2bd59d42014-02-11 11:52:49 -05003485static struct kernfs_ops cgroup_kf_single_ops = {
3486 .atomic_write_len = PAGE_SIZE,
3487 .write = cgroup_file_write,
3488 .seq_show = cgroup_seqfile_show,
Paul Menage91796562008-04-29 01:00:01 -07003489};
3490
Tejun Heo2bd59d42014-02-11 11:52:49 -05003491static struct kernfs_ops cgroup_kf_ops = {
3492 .atomic_write_len = PAGE_SIZE,
3493 .write = cgroup_file_write,
3494 .seq_start = cgroup_seqfile_start,
3495 .seq_next = cgroup_seqfile_next,
3496 .seq_stop = cgroup_seqfile_stop,
3497 .seq_show = cgroup_seqfile_show,
3498};
Paul Menageddbcc7e2007-10-18 23:39:30 -07003499
3500/*
3501 * cgroup_rename - Only allow simple rename of directories in place.
3502 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05003503static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
3504 const char *new_name_str)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003505{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003506 struct cgroup *cgrp = kn->priv;
Li Zefan65dff752013-03-01 15:01:56 +08003507 int ret;
Li Zefan65dff752013-03-01 15:01:56 +08003508
Tejun Heo2bd59d42014-02-11 11:52:49 -05003509 if (kernfs_type(kn) != KERNFS_DIR)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003510 return -ENOTDIR;
Tejun Heo2bd59d42014-02-11 11:52:49 -05003511 if (kn->parent != new_parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003512 return -EIO;
Li Zefan65dff752013-03-01 15:01:56 +08003513
Tejun Heo6db8e852013-06-14 11:18:22 -07003514 /*
3515 * This isn't a proper migration and its usefulness is very
Tejun Heoaa6ec292014-07-09 10:08:08 -04003516 * limited. Disallow on the default hierarchy.
Tejun Heo6db8e852013-06-14 11:18:22 -07003517 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04003518 if (cgroup_on_dfl(cgrp))
Tejun Heo6db8e852013-06-14 11:18:22 -07003519 return -EPERM;
3520
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003521 /*
Tejun Heo8353da12014-05-13 12:19:23 -04003522 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003523 * active_ref. kernfs_rename() doesn't require active_ref
Tejun Heo8353da12014-05-13 12:19:23 -04003524 * protection. Break them before grabbing cgroup_mutex.
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003525 */
3526 kernfs_break_active_protection(new_parent);
3527 kernfs_break_active_protection(kn);
Li Zefan65dff752013-03-01 15:01:56 +08003528
Tejun Heo2bd59d42014-02-11 11:52:49 -05003529 mutex_lock(&cgroup_mutex);
Li Zefan65dff752013-03-01 15:01:56 +08003530
Tejun Heo2bd59d42014-02-11 11:52:49 -05003531 ret = kernfs_rename(kn, new_parent, new_name_str);
Li Zefan65dff752013-03-01 15:01:56 +08003532
Tejun Heo2bd59d42014-02-11 11:52:49 -05003533 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003534
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003535 kernfs_unbreak_active_protection(kn);
3536 kernfs_unbreak_active_protection(new_parent);
Tejun Heo2bd59d42014-02-11 11:52:49 -05003537 return ret;
Li Zefan099fca32009-04-02 16:57:29 -07003538}
3539
Tejun Heo49957f82014-04-07 16:44:47 -04003540/* set uid and gid of cgroup dirs and files to that of the creator */
3541static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3542{
3543 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3544 .ia_uid = current_fsuid(),
3545 .ia_gid = current_fsgid(), };
3546
3547 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3548 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3549 return 0;
3550
3551 return kernfs_setattr(kn, &iattr);
3552}
3553
Tejun Heo4df8dc92015-09-18 17:54:23 -04003554static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3555 struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003556{
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05003557 char name[CGROUP_FILE_NAME_MAX];
Tejun Heo2bd59d42014-02-11 11:52:49 -05003558 struct kernfs_node *kn;
3559 struct lock_class_key *key = NULL;
Tejun Heo49957f82014-04-07 16:44:47 -04003560 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003561
Tejun Heo2bd59d42014-02-11 11:52:49 -05003562#ifdef CONFIG_DEBUG_LOCK_ALLOC
3563 key = &cft->lockdep_key;
3564#endif
3565 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3566 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
Tejun Heodfeb07502015-02-13 14:36:31 -08003567 NULL, key);
Tejun Heo49957f82014-04-07 16:44:47 -04003568 if (IS_ERR(kn))
3569 return PTR_ERR(kn);
3570
3571 ret = cgroup_kn_set_ugid(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003572 if (ret) {
Tejun Heo49957f82014-04-07 16:44:47 -04003573 kernfs_remove(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003574 return ret;
3575 }
3576
Tejun Heo6f60ead2015-09-18 17:54:23 -04003577 if (cft->file_offset) {
3578 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3579
Tejun Heo34c06252015-11-05 00:12:24 -05003580 spin_lock_irq(&cgroup_file_kn_lock);
Tejun Heo6f60ead2015-09-18 17:54:23 -04003581 cfile->kn = kn;
Tejun Heo34c06252015-11-05 00:12:24 -05003582 spin_unlock_irq(&cgroup_file_kn_lock);
Tejun Heo6f60ead2015-09-18 17:54:23 -04003583 }
3584
Tejun Heof8f22e52014-04-23 11:13:16 -04003585 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003586}
3587
Tejun Heob1f28d32013-06-28 16:24:10 -07003588/**
3589 * cgroup_addrm_files - add or remove files to a cgroup directory
Tejun Heo4df8dc92015-09-18 17:54:23 -04003590 * @css: the target css
3591 * @cgrp: the target cgroup (usually css->cgroup)
Tejun Heob1f28d32013-06-28 16:24:10 -07003592 * @cfts: array of cftypes to be added
3593 * @is_add: whether to add or remove
3594 *
3595 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
Tejun Heo6732ed82015-09-18 17:54:23 -04003596 * For removals, this function never fails.
Tejun Heob1f28d32013-06-28 16:24:10 -07003597 */
Tejun Heo4df8dc92015-09-18 17:54:23 -04003598static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3599 struct cgroup *cgrp, struct cftype cfts[],
Tejun Heo2bb566c2013-08-08 20:11:23 -04003600 bool is_add)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003601{
Tejun Heo6732ed82015-09-18 17:54:23 -04003602 struct cftype *cft, *cft_end = NULL;
Tejun Heob598dde2016-02-22 22:25:45 -05003603 int ret = 0;
Tejun Heob1f28d32013-06-28 16:24:10 -07003604
Tejun Heo01f64742014-05-13 12:19:23 -04003605 lockdep_assert_held(&cgroup_mutex);
Tejun Heodb0416b2012-04-01 12:09:55 -07003606
Tejun Heo6732ed82015-09-18 17:54:23 -04003607restart:
3608 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
Gao fengf33fddc2012-12-06 14:38:57 +08003609 /* does cft->flags tell us to skip this file on @cgrp? */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003610 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
Tejun Heo8cbbf2c2014-03-19 10:23:55 -04003611 continue;
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003612 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
Tejun Heo873fe092013-04-14 20:15:26 -07003613 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003614 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003615 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003616 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003617 continue;
3618
Li Zefan2739d3c2013-01-21 18:18:33 +08003619 if (is_add) {
Tejun Heo4df8dc92015-09-18 17:54:23 -04003620 ret = cgroup_add_file(css, cgrp, cft);
Tejun Heob1f28d32013-06-28 16:24:10 -07003621 if (ret) {
Joe Perchesed3d2612014-04-25 18:28:03 -04003622 pr_warn("%s: failed to add %s, err=%d\n",
3623 __func__, cft->name, ret);
Tejun Heo6732ed82015-09-18 17:54:23 -04003624 cft_end = cft;
3625 is_add = false;
3626 goto restart;
Tejun Heob1f28d32013-06-28 16:24:10 -07003627 }
Li Zefan2739d3c2013-01-21 18:18:33 +08003628 } else {
3629 cgroup_rm_file(cgrp, cft);
Tejun Heodb0416b2012-04-01 12:09:55 -07003630 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003631 }
Tejun Heob598dde2016-02-22 22:25:45 -05003632 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003633}
3634
Tejun Heo21a2d342014-02-12 09:29:49 -05003635static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003636{
3637 LIST_HEAD(pending);
Tejun Heo2bb566c2013-08-08 20:11:23 -04003638 struct cgroup_subsys *ss = cfts[0].ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04003639 struct cgroup *root = &ss->root->cgrp;
Tejun Heo492eb212013-08-08 20:11:25 -04003640 struct cgroup_subsys_state *css;
Tejun Heo9ccece82013-06-28 16:24:11 -07003641 int ret = 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003642
Tejun Heo01f64742014-05-13 12:19:23 -04003643 lockdep_assert_held(&cgroup_mutex);
Li Zefane8c82d22013-06-18 18:48:37 +08003644
Li Zefane8c82d22013-06-18 18:48:37 +08003645 /* add/rm files for all cgroups created before */
Tejun Heoca8bdca2013-08-26 18:40:56 -04003646 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
Tejun Heo492eb212013-08-08 20:11:25 -04003647 struct cgroup *cgrp = css->cgroup;
3648
Tejun Heo88cb04b2016-03-03 09:57:58 -05003649 if (!(css->flags & CSS_VISIBLE))
Li Zefane8c82d22013-06-18 18:48:37 +08003650 continue;
3651
Tejun Heo4df8dc92015-09-18 17:54:23 -04003652 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
Tejun Heo9ccece82013-06-28 16:24:11 -07003653 if (ret)
3654 break;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003655 }
Tejun Heo21a2d342014-02-12 09:29:49 -05003656
3657 if (is_add && !ret)
3658 kernfs_activate(root->kn);
Tejun Heo9ccece82013-06-28 16:24:11 -07003659 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003660}
3661
Tejun Heo2da440a2014-02-11 11:52:48 -05003662static void cgroup_exit_cftypes(struct cftype *cfts)
3663{
3664 struct cftype *cft;
3665
Tejun Heo2bd59d42014-02-11 11:52:49 -05003666 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3667 /* free copy for custom atomic_write_len, see init_cftypes() */
3668 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3669 kfree(cft->kf_ops);
3670 cft->kf_ops = NULL;
Tejun Heo2da440a2014-02-11 11:52:48 -05003671 cft->ss = NULL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003672
3673 /* revert flags set by cgroup core while adding @cfts */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003674 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
Tejun Heo2bd59d42014-02-11 11:52:49 -05003675 }
Tejun Heo2da440a2014-02-11 11:52:48 -05003676}
3677
Tejun Heo2bd59d42014-02-11 11:52:49 -05003678static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo2da440a2014-02-11 11:52:48 -05003679{
3680 struct cftype *cft;
3681
Tejun Heo2bd59d42014-02-11 11:52:49 -05003682 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3683 struct kernfs_ops *kf_ops;
3684
Tejun Heo0adb0702014-02-12 09:29:48 -05003685 WARN_ON(cft->ss || cft->kf_ops);
3686
Tejun Heo2bd59d42014-02-11 11:52:49 -05003687 if (cft->seq_start)
3688 kf_ops = &cgroup_kf_ops;
3689 else
3690 kf_ops = &cgroup_kf_single_ops;
3691
3692 /*
3693 * Ugh... if @cft wants a custom max_write_len, we need to
3694 * make a copy of kf_ops to set its atomic_write_len.
3695 */
3696 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3697 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3698 if (!kf_ops) {
3699 cgroup_exit_cftypes(cfts);
3700 return -ENOMEM;
3701 }
3702 kf_ops->atomic_write_len = cft->max_write_len;
3703 }
3704
3705 cft->kf_ops = kf_ops;
Tejun Heo2da440a2014-02-11 11:52:48 -05003706 cft->ss = ss;
Tejun Heo2bd59d42014-02-11 11:52:49 -05003707 }
3708
3709 return 0;
Tejun Heo2da440a2014-02-11 11:52:48 -05003710}
3711
Tejun Heo21a2d342014-02-12 09:29:49 -05003712static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3713{
Tejun Heo01f64742014-05-13 12:19:23 -04003714 lockdep_assert_held(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003715
3716 if (!cfts || !cfts[0].ss)
3717 return -ENOENT;
3718
3719 list_del(&cfts->node);
3720 cgroup_apply_cftypes(cfts, false);
3721 cgroup_exit_cftypes(cfts);
3722 return 0;
3723}
3724
Tejun Heo8e3f6542012-04-01 12:09:55 -07003725/**
Tejun Heo80b13582014-02-12 09:29:48 -05003726 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3727 * @cfts: zero-length name terminated array of cftypes
3728 *
3729 * Unregister @cfts. Files described by @cfts are removed from all
3730 * existing cgroups and all future cgroups won't have them either. This
3731 * function can be called anytime whether @cfts' subsys is attached or not.
3732 *
3733 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3734 * registered.
3735 */
3736int cgroup_rm_cftypes(struct cftype *cfts)
3737{
Tejun Heo21a2d342014-02-12 09:29:49 -05003738 int ret;
Tejun Heo80b13582014-02-12 09:29:48 -05003739
Tejun Heo01f64742014-05-13 12:19:23 -04003740 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003741 ret = cgroup_rm_cftypes_locked(cfts);
Tejun Heo01f64742014-05-13 12:19:23 -04003742 mutex_unlock(&cgroup_mutex);
Tejun Heo8e3f6542012-04-01 12:09:55 -07003743 return ret;
3744}
3745
3746/**
3747 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3748 * @ss: target cgroup subsystem
3749 * @cfts: zero-length name terminated array of cftypes
3750 *
3751 * Register @cfts to @ss. Files described by @cfts are created for all
3752 * existing cgroups to which @ss is attached and all future cgroups will
3753 * have them too. This function can be called anytime whether @ss is
3754 * attached or not.
3755 *
3756 * Returns 0 on successful registration, -errno on failure. Note that this
3757 * function currently returns 0 as long as @cfts registration is successful
3758 * even if some file creation attempts on existing cgroups fail.
3759 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003760static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003761{
Tejun Heo9ccece82013-06-28 16:24:11 -07003762 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003763
Tejun Heofc5ed1e2015-09-18 11:56:28 -04003764 if (!cgroup_ssid_enabled(ss->id))
Li Zefanc731ae12014-06-05 17:16:30 +08003765 return 0;
3766
Li Zefandc5736e2014-02-17 10:41:50 +08003767 if (!cfts || cfts[0].name[0] == '\0')
3768 return 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003769
Tejun Heo2bd59d42014-02-11 11:52:49 -05003770 ret = cgroup_init_cftypes(ss, cfts);
Tejun Heo9ccece82013-06-28 16:24:11 -07003771 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05003772 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003773
Tejun Heo01f64742014-05-13 12:19:23 -04003774 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003775
Tejun Heo0adb0702014-02-12 09:29:48 -05003776 list_add_tail(&cfts->node, &ss->cfts);
Tejun Heo21a2d342014-02-12 09:29:49 -05003777 ret = cgroup_apply_cftypes(cfts, true);
Tejun Heo9ccece82013-06-28 16:24:11 -07003778 if (ret)
Tejun Heo21a2d342014-02-12 09:29:49 -05003779 cgroup_rm_cftypes_locked(cfts);
3780
Tejun Heo01f64742014-05-13 12:19:23 -04003781 mutex_unlock(&cgroup_mutex);
Tejun Heo9ccece82013-06-28 16:24:11 -07003782 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003783}
Tejun Heo79578622012-04-01 12:09:56 -07003784
3785/**
Tejun Heoa8ddc822014-07-15 11:05:10 -04003786 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3787 * @ss: target cgroup subsystem
3788 * @cfts: zero-length name terminated array of cftypes
3789 *
3790 * Similar to cgroup_add_cftypes() but the added files are only used for
3791 * the default hierarchy.
3792 */
3793int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3794{
3795 struct cftype *cft;
3796
3797 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003798 cft->flags |= __CFTYPE_ONLY_ON_DFL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003799 return cgroup_add_cftypes(ss, cfts);
3800}
3801
3802/**
3803 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3804 * @ss: target cgroup subsystem
3805 * @cfts: zero-length name terminated array of cftypes
3806 *
3807 * Similar to cgroup_add_cftypes() but the added files are only used for
3808 * the legacy hierarchies.
3809 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003810int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3811{
Tejun Heoa8ddc822014-07-15 11:05:10 -04003812 struct cftype *cft;
3813
Tejun Heoe4b70372015-10-15 17:00:43 -04003814 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3815 cft->flags |= __CFTYPE_NOT_ON_DFL;
Tejun Heo2cf669a2014-07-15 11:05:09 -04003816 return cgroup_add_cftypes(ss, cfts);
3817}
3818
Li Zefana043e3b2008-02-23 15:24:09 -08003819/**
Tejun Heo34c06252015-11-05 00:12:24 -05003820 * cgroup_file_notify - generate a file modified event for a cgroup_file
3821 * @cfile: target cgroup_file
3822 *
3823 * @cfile must have been obtained by setting cftype->file_offset.
3824 */
3825void cgroup_file_notify(struct cgroup_file *cfile)
3826{
3827 unsigned long flags;
3828
3829 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3830 if (cfile->kn)
3831 kernfs_notify(cfile->kn);
3832 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3833}
3834
3835/**
Li Zefana043e3b2008-02-23 15:24:09 -08003836 * cgroup_task_count - count the number of tasks in a cgroup.
3837 * @cgrp: the cgroup in question
3838 *
3839 * Return the number of tasks in the cgroup.
3840 */
Tejun Heo07bc3562014-02-13 06:58:39 -05003841static int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07003842{
3843 int count = 0;
Tejun Heo69d02062013-06-12 21:04:50 -07003844 struct cgrp_cset_link *link;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003845
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003846 spin_lock_bh(&css_set_lock);
Tejun Heo69d02062013-06-12 21:04:50 -07003847 list_for_each_entry(link, &cgrp->cset_links, cset_link)
3848 count += atomic_read(&link->cset->refcount);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04003849 spin_unlock_bh(&css_set_lock);
Paul Menagebbcb81d2007-10-18 23:39:32 -07003850 return count;
3851}
3852
Tejun Heo574bd9f2012-11-09 09:12:29 -08003853/**
Tejun Heo492eb212013-08-08 20:11:25 -04003854 * css_next_child - find the next child of a given css
Tejun Heoc2931b72014-05-16 13:22:51 -04003855 * @pos: the current position (%NULL to initiate traversal)
3856 * @parent: css whose children to walk
Tejun Heo53fa5262013-05-24 10:55:38 +09003857 *
Tejun Heoc2931b72014-05-16 13:22:51 -04003858 * This function returns the next child of @parent and should be called
Tejun Heo87fb54f2013-12-06 15:11:55 -05003859 * under either cgroup_mutex or RCU read lock. The only requirement is
Tejun Heoc2931b72014-05-16 13:22:51 -04003860 * that @parent and @pos are accessible. The next sibling is guaranteed to
3861 * be returned regardless of their states.
3862 *
3863 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3864 * css which finished ->css_online() is guaranteed to be visible in the
3865 * future iterations and will stay visible until the last reference is put.
3866 * A css which hasn't finished ->css_online() or already finished
3867 * ->css_offline() may show up during traversal. It's each subsystem's
3868 * responsibility to synchronize against on/offlining.
Tejun Heo53fa5262013-05-24 10:55:38 +09003869 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003870struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3871 struct cgroup_subsys_state *parent)
Tejun Heo53fa5262013-05-24 10:55:38 +09003872{
Tejun Heoc2931b72014-05-16 13:22:51 -04003873 struct cgroup_subsys_state *next;
Tejun Heo53fa5262013-05-24 10:55:38 +09003874
Tejun Heo8353da12014-05-13 12:19:23 -04003875 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo53fa5262013-05-24 10:55:38 +09003876
3877 /*
Tejun Heode3f0342014-05-16 13:22:49 -04003878 * @pos could already have been unlinked from the sibling list.
3879 * Once a cgroup is removed, its ->sibling.next is no longer
3880 * updated when its next sibling changes. CSS_RELEASED is set when
3881 * @pos is taken off list, at which time its next pointer is valid,
3882 * and, as releases are serialized, the one pointed to by the next
3883 * pointer is guaranteed to not have started release yet. This
3884 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3885 * critical section, the one pointed to by its next pointer is
3886 * guaranteed to not have finished its RCU grace period even if we
3887 * have dropped rcu_read_lock() inbetween iterations.
Tejun Heo3b287a52013-08-08 20:11:24 -04003888 *
Tejun Heode3f0342014-05-16 13:22:49 -04003889 * If @pos has CSS_RELEASED set, its next pointer can't be
3890 * dereferenced; however, as each css is given a monotonically
3891 * increasing unique serial number and always appended to the
3892 * sibling list, the next one can be found by walking the parent's
3893 * children until the first css with higher serial number than
3894 * @pos's. While this path can be slower, it happens iff iteration
3895 * races against release and the race window is very small.
Tejun Heo53fa5262013-05-24 10:55:38 +09003896 */
Tejun Heo3b287a52013-08-08 20:11:24 -04003897 if (!pos) {
Tejun Heoc2931b72014-05-16 13:22:51 -04003898 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3899 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3900 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
Tejun Heo3b287a52013-08-08 20:11:24 -04003901 } else {
Tejun Heoc2931b72014-05-16 13:22:51 -04003902 list_for_each_entry_rcu(next, &parent->children, sibling)
Tejun Heo3b287a52013-08-08 20:11:24 -04003903 if (next->serial_nr > pos->serial_nr)
3904 break;
Tejun Heo53fa5262013-05-24 10:55:38 +09003905 }
3906
Tejun Heo3b281af2014-04-23 11:13:15 -04003907 /*
3908 * @next, if not pointing to the head, can be dereferenced and is
Tejun Heoc2931b72014-05-16 13:22:51 -04003909 * the next sibling.
Tejun Heo3b281af2014-04-23 11:13:15 -04003910 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003911 if (&next->sibling != &parent->children)
3912 return next;
Tejun Heo3b281af2014-04-23 11:13:15 -04003913 return NULL;
Tejun Heo53fa5262013-05-24 10:55:38 +09003914}
Tejun Heo53fa5262013-05-24 10:55:38 +09003915
3916/**
Tejun Heo492eb212013-08-08 20:11:25 -04003917 * css_next_descendant_pre - find the next descendant for pre-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003918 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04003919 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003920 *
Tejun Heo492eb212013-08-08 20:11:25 -04003921 * To be used by css_for_each_descendant_pre(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04003922 * to visit for pre-order traversal of @root's descendants. @root is
3923 * included in the iteration and the first node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09003924 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003925 * While this function requires cgroup_mutex or RCU read locking, it
3926 * doesn't require the whole traversal to be contained in a single critical
3927 * section. This function will return the correct next descendant as long
3928 * as both @pos and @root are accessible and @pos is a descendant of @root.
Tejun Heoc2931b72014-05-16 13:22:51 -04003929 *
3930 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3931 * css which finished ->css_online() is guaranteed to be visible in the
3932 * future iterations and will stay visible until the last reference is put.
3933 * A css which hasn't finished ->css_online() or already finished
3934 * ->css_offline() may show up during traversal. It's each subsystem's
3935 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08003936 */
Tejun Heo492eb212013-08-08 20:11:25 -04003937struct cgroup_subsys_state *
3938css_next_descendant_pre(struct cgroup_subsys_state *pos,
3939 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003940{
Tejun Heo492eb212013-08-08 20:11:25 -04003941 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003942
Tejun Heo8353da12014-05-13 12:19:23 -04003943 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08003944
Tejun Heobd8815a2013-08-08 20:11:27 -04003945 /* if first iteration, visit @root */
Tejun Heo7805d002013-05-24 10:50:24 +09003946 if (!pos)
Tejun Heobd8815a2013-08-08 20:11:27 -04003947 return root;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003948
3949 /* visit the first child if exists */
Tejun Heo492eb212013-08-08 20:11:25 -04003950 next = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08003951 if (next)
3952 return next;
3953
3954 /* no child, visit my or the closest ancestor's next sibling */
Tejun Heo492eb212013-08-08 20:11:25 -04003955 while (pos != root) {
Tejun Heo5c9d5352014-05-16 13:22:48 -04003956 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09003957 if (next)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003958 return next;
Tejun Heo5c9d5352014-05-16 13:22:48 -04003959 pos = pos->parent;
Tejun Heo7805d002013-05-24 10:50:24 +09003960 }
Tejun Heo574bd9f2012-11-09 09:12:29 -08003961
3962 return NULL;
3963}
Tejun Heo574bd9f2012-11-09 09:12:29 -08003964
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003965/**
Tejun Heo492eb212013-08-08 20:11:25 -04003966 * css_rightmost_descendant - return the rightmost descendant of a css
3967 * @pos: css of interest
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003968 *
Tejun Heo492eb212013-08-08 20:11:25 -04003969 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3970 * is returned. This can be used during pre-order traversal to skip
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003971 * subtree of @pos.
Tejun Heo75501a62013-05-24 10:55:38 +09003972 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003973 * While this function requires cgroup_mutex or RCU read locking, it
3974 * doesn't require the whole traversal to be contained in a single critical
3975 * section. This function will return the correct rightmost descendant as
3976 * long as @pos is accessible.
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003977 */
Tejun Heo492eb212013-08-08 20:11:25 -04003978struct cgroup_subsys_state *
3979css_rightmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003980{
Tejun Heo492eb212013-08-08 20:11:25 -04003981 struct cgroup_subsys_state *last, *tmp;
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003982
Tejun Heo8353da12014-05-13 12:19:23 -04003983 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003984
3985 do {
3986 last = pos;
3987 /* ->prev isn't RCU safe, walk ->next till the end */
3988 pos = NULL;
Tejun Heo492eb212013-08-08 20:11:25 -04003989 css_for_each_child(tmp, last)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003990 pos = tmp;
3991 } while (pos);
3992
3993 return last;
3994}
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003995
Tejun Heo492eb212013-08-08 20:11:25 -04003996static struct cgroup_subsys_state *
3997css_leftmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003998{
Tejun Heo492eb212013-08-08 20:11:25 -04003999 struct cgroup_subsys_state *last;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004000
4001 do {
4002 last = pos;
Tejun Heo492eb212013-08-08 20:11:25 -04004003 pos = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004004 } while (pos);
4005
4006 return last;
4007}
4008
4009/**
Tejun Heo492eb212013-08-08 20:11:25 -04004010 * css_next_descendant_post - find the next descendant for post-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08004011 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04004012 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08004013 *
Tejun Heo492eb212013-08-08 20:11:25 -04004014 * To be used by css_for_each_descendant_post(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04004015 * to visit for post-order traversal of @root's descendants. @root is
4016 * included in the iteration and the last node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09004017 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05004018 * While this function requires cgroup_mutex or RCU read locking, it
4019 * doesn't require the whole traversal to be contained in a single critical
4020 * section. This function will return the correct next descendant as long
4021 * as both @pos and @cgroup are accessible and @pos is a descendant of
4022 * @cgroup.
Tejun Heoc2931b72014-05-16 13:22:51 -04004023 *
4024 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4025 * css which finished ->css_online() is guaranteed to be visible in the
4026 * future iterations and will stay visible until the last reference is put.
4027 * A css which hasn't finished ->css_online() or already finished
4028 * ->css_offline() may show up during traversal. It's each subsystem's
4029 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08004030 */
Tejun Heo492eb212013-08-08 20:11:25 -04004031struct cgroup_subsys_state *
4032css_next_descendant_post(struct cgroup_subsys_state *pos,
4033 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08004034{
Tejun Heo492eb212013-08-08 20:11:25 -04004035 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004036
Tejun Heo8353da12014-05-13 12:19:23 -04004037 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08004038
Tejun Heo58b79a92013-09-06 15:31:08 -04004039 /* if first iteration, visit leftmost descendant which may be @root */
4040 if (!pos)
4041 return css_leftmost_descendant(root);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004042
Tejun Heobd8815a2013-08-08 20:11:27 -04004043 /* if we visited @root, we're done */
4044 if (pos == root)
4045 return NULL;
4046
Tejun Heo574bd9f2012-11-09 09:12:29 -08004047 /* if there's an unvisited sibling, visit its leftmost descendant */
Tejun Heo5c9d5352014-05-16 13:22:48 -04004048 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09004049 if (next)
Tejun Heo492eb212013-08-08 20:11:25 -04004050 return css_leftmost_descendant(next);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004051
4052 /* no sibling left, visit parent */
Tejun Heo5c9d5352014-05-16 13:22:48 -04004053 return pos->parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004054}
Tejun Heo574bd9f2012-11-09 09:12:29 -08004055
Tejun Heof3d46502014-05-16 13:22:52 -04004056/**
4057 * css_has_online_children - does a css have online children
4058 * @css: the target css
4059 *
4060 * Returns %true if @css has any online children; otherwise, %false. This
4061 * function can be called from any context but the caller is responsible
4062 * for synchronizing against on/offlining as necessary.
4063 */
4064bool css_has_online_children(struct cgroup_subsys_state *css)
Tejun Heocbc125e2014-05-14 09:15:01 -04004065{
Tejun Heof3d46502014-05-16 13:22:52 -04004066 struct cgroup_subsys_state *child;
4067 bool ret = false;
Tejun Heocbc125e2014-05-14 09:15:01 -04004068
4069 rcu_read_lock();
Tejun Heof3d46502014-05-16 13:22:52 -04004070 css_for_each_child(child, css) {
Li Zefan99bae5f2014-06-12 14:31:31 +08004071 if (child->flags & CSS_ONLINE) {
Tejun Heof3d46502014-05-16 13:22:52 -04004072 ret = true;
4073 break;
Tejun Heocbc125e2014-05-14 09:15:01 -04004074 }
4075 }
4076 rcu_read_unlock();
Tejun Heof3d46502014-05-16 13:22:52 -04004077 return ret;
Cliff Wickman31a7df02008-02-07 00:14:42 -08004078}
4079
Tejun Heo0942eee2013-08-08 20:11:26 -04004080/**
Tejun Heoecb9d532015-10-15 16:41:52 -04004081 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
Tejun Heo0942eee2013-08-08 20:11:26 -04004082 * @it: the iterator to advance
4083 *
4084 * Advance @it to the next css_set to walk.
Tejun Heod5158762013-08-08 20:11:26 -04004085 */
Tejun Heoecb9d532015-10-15 16:41:52 -04004086static void css_task_iter_advance_css_set(struct css_task_iter *it)
Tejun Heod5158762013-08-08 20:11:26 -04004087{
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004088 struct list_head *l = it->cset_pos;
Tejun Heod5158762013-08-08 20:11:26 -04004089 struct cgrp_cset_link *link;
4090 struct css_set *cset;
4091
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004092 lockdep_assert_held(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004093
Tejun Heod5158762013-08-08 20:11:26 -04004094 /* Advance to the next non-empty css_set */
4095 do {
4096 l = l->next;
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004097 if (l == it->cset_head) {
4098 it->cset_pos = NULL;
Tejun Heoecb9d532015-10-15 16:41:52 -04004099 it->task_pos = NULL;
Tejun Heod5158762013-08-08 20:11:26 -04004100 return;
4101 }
Tejun Heo3ebb2b62014-04-23 11:13:15 -04004102
4103 if (it->ss) {
4104 cset = container_of(l, struct css_set,
4105 e_cset_node[it->ss->id]);
4106 } else {
4107 link = list_entry(l, struct cgrp_cset_link, cset_link);
4108 cset = link->cset;
4109 }
Tejun Heo0de09422015-10-15 16:41:49 -04004110 } while (!css_set_populated(cset));
Tejun Heoc7561122014-02-25 10:04:01 -05004111
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004112 it->cset_pos = l;
Tejun Heoc7561122014-02-25 10:04:01 -05004113
4114 if (!list_empty(&cset->tasks))
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004115 it->task_pos = cset->tasks.next;
Tejun Heoc7561122014-02-25 10:04:01 -05004116 else
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004117 it->task_pos = cset->mg_tasks.next;
4118
4119 it->tasks_head = &cset->tasks;
4120 it->mg_tasks_head = &cset->mg_tasks;
Tejun Heoed27b9f2015-10-15 16:41:52 -04004121
4122 /*
4123 * We don't keep css_sets locked across iteration steps and thus
4124 * need to take steps to ensure that iteration can be resumed after
4125 * the lock is re-acquired. Iteration is performed at two levels -
4126 * css_sets and tasks in them.
4127 *
4128 * Once created, a css_set never leaves its cgroup lists, so a
4129 * pinned css_set is guaranteed to stay put and we can resume
4130 * iteration afterwards.
4131 *
4132 * Tasks may leave @cset across iteration steps. This is resolved
4133 * by registering each iterator with the css_set currently being
4134 * walked and making css_set_move_task() advance iterators whose
4135 * next task is leaving.
4136 */
4137 if (it->cur_cset) {
4138 list_del(&it->iters_node);
4139 put_css_set_locked(it->cur_cset);
4140 }
4141 get_css_set(cset);
4142 it->cur_cset = cset;
4143 list_add(&it->iters_node, &cset->task_iters);
Tejun Heod5158762013-08-08 20:11:26 -04004144}
4145
Tejun Heoecb9d532015-10-15 16:41:52 -04004146static void css_task_iter_advance(struct css_task_iter *it)
4147{
4148 struct list_head *l = it->task_pos;
4149
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004150 lockdep_assert_held(&css_set_lock);
Tejun Heoecb9d532015-10-15 16:41:52 -04004151 WARN_ON_ONCE(!l);
4152
4153 /*
4154 * Advance iterator to find next entry. cset->tasks is consumed
4155 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4156 * next cset.
4157 */
4158 l = l->next;
4159
4160 if (l == it->tasks_head)
4161 l = it->mg_tasks_head->next;
4162
4163 if (l == it->mg_tasks_head)
4164 css_task_iter_advance_css_set(it);
4165 else
4166 it->task_pos = l;
4167}
4168
Tejun Heo0942eee2013-08-08 20:11:26 -04004169/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004170 * css_task_iter_start - initiate task iteration
4171 * @css: the css to walk tasks of
Tejun Heo0942eee2013-08-08 20:11:26 -04004172 * @it: the task iterator to use
4173 *
Tejun Heo72ec7022013-08-08 20:11:26 -04004174 * Initiate iteration through the tasks of @css. The caller can call
4175 * css_task_iter_next() to walk through the tasks until the function
4176 * returns NULL. On completion of iteration, css_task_iter_end() must be
4177 * called.
Tejun Heo0942eee2013-08-08 20:11:26 -04004178 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004179void css_task_iter_start(struct cgroup_subsys_state *css,
4180 struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004181{
Tejun Heo56fde9e2014-02-13 06:58:38 -05004182 /* no one should try to iterate before mounting cgroups */
4183 WARN_ON_ONCE(!use_task_css_set_links);
Paul Menage817929e2007-10-18 23:39:36 -07004184
Tejun Heoed27b9f2015-10-15 16:41:52 -04004185 memset(it, 0, sizeof(*it));
4186
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004187 spin_lock_bh(&css_set_lock);
Tejun Heoc59cd3d2013-08-08 20:11:26 -04004188
Tejun Heo3ebb2b62014-04-23 11:13:15 -04004189 it->ss = css->ss;
4190
4191 if (it->ss)
4192 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4193 else
4194 it->cset_pos = &css->cgroup->cset_links;
4195
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004196 it->cset_head = it->cset_pos;
Tejun Heoc59cd3d2013-08-08 20:11:26 -04004197
Tejun Heoecb9d532015-10-15 16:41:52 -04004198 css_task_iter_advance_css_set(it);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004199
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004200 spin_unlock_bh(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07004201}
Paul Menage817929e2007-10-18 23:39:36 -07004202
Tejun Heo0942eee2013-08-08 20:11:26 -04004203/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004204 * css_task_iter_next - return the next task for the iterator
Tejun Heo0942eee2013-08-08 20:11:26 -04004205 * @it: the task iterator being iterated
4206 *
4207 * The "next" function for task iteration. @it should have been
Tejun Heo72ec7022013-08-08 20:11:26 -04004208 * initialized via css_task_iter_start(). Returns NULL when the iteration
4209 * reaches the end.
Tejun Heo0942eee2013-08-08 20:11:26 -04004210 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004211struct task_struct *css_task_iter_next(struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004212{
Tejun Heod5745672015-10-29 11:43:05 +09004213 if (it->cur_task) {
Tejun Heoed27b9f2015-10-15 16:41:52 -04004214 put_task_struct(it->cur_task);
Tejun Heod5745672015-10-29 11:43:05 +09004215 it->cur_task = NULL;
4216 }
Tejun Heoed27b9f2015-10-15 16:41:52 -04004217
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004218 spin_lock_bh(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004219
Tejun Heod5745672015-10-29 11:43:05 +09004220 if (it->task_pos) {
4221 it->cur_task = list_entry(it->task_pos, struct task_struct,
4222 cg_list);
4223 get_task_struct(it->cur_task);
4224 css_task_iter_advance(it);
4225 }
Tejun Heoed27b9f2015-10-15 16:41:52 -04004226
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004227 spin_unlock_bh(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004228
4229 return it->cur_task;
Paul Menage817929e2007-10-18 23:39:36 -07004230}
4231
Tejun Heo0942eee2013-08-08 20:11:26 -04004232/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004233 * css_task_iter_end - finish task iteration
Tejun Heo0942eee2013-08-08 20:11:26 -04004234 * @it: the task iterator to finish
4235 *
Tejun Heo72ec7022013-08-08 20:11:26 -04004236 * Finish task iteration started by css_task_iter_start().
Tejun Heo0942eee2013-08-08 20:11:26 -04004237 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004238void css_task_iter_end(struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004239{
Tejun Heoed27b9f2015-10-15 16:41:52 -04004240 if (it->cur_cset) {
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004241 spin_lock_bh(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004242 list_del(&it->iters_node);
4243 put_css_set_locked(it->cur_cset);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004244 spin_unlock_bh(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004245 }
4246
4247 if (it->cur_task)
4248 put_task_struct(it->cur_task);
Tejun Heo8cc99342013-04-07 09:29:50 -07004249}
4250
4251/**
4252 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4253 * @to: cgroup to which the tasks will be moved
4254 * @from: cgroup in which the tasks currently reside
Tejun Heoeaf797a2014-02-25 10:04:03 -05004255 *
4256 * Locking rules between cgroup_post_fork() and the migration path
4257 * guarantee that, if a task is forking while being migrated, the new child
4258 * is guaranteed to be either visible in the source cgroup after the
4259 * parent's migration is complete or put into the target cgroup. No task
4260 * can slip out of migration through forking.
Tejun Heo8cc99342013-04-07 09:29:50 -07004261 */
4262int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
4263{
Tejun Heo952aaa12014-02-25 10:04:03 -05004264 LIST_HEAD(preloaded_csets);
4265 struct cgrp_cset_link *link;
Tejun Heoe406d1c2014-02-13 06:58:39 -05004266 struct css_task_iter it;
4267 struct task_struct *task;
Tejun Heo952aaa12014-02-25 10:04:03 -05004268 int ret;
Tejun Heoe406d1c2014-02-13 06:58:39 -05004269
Tejun Heo6c694c82016-03-08 11:51:25 -05004270 if (!cgroup_may_migrate_to(to))
4271 return -EBUSY;
4272
Tejun Heo952aaa12014-02-25 10:04:03 -05004273 mutex_lock(&cgroup_mutex);
4274
4275 /* all tasks in @from are being moved, all csets are source */
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004276 spin_lock_bh(&css_set_lock);
Tejun Heo952aaa12014-02-25 10:04:03 -05004277 list_for_each_entry(link, &from->cset_links, cset_link)
4278 cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004279 spin_unlock_bh(&css_set_lock);
Tejun Heo952aaa12014-02-25 10:04:03 -05004280
Tejun Heoe4857982016-03-08 11:51:26 -05004281 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heo952aaa12014-02-25 10:04:03 -05004282 if (ret)
4283 goto out_err;
4284
4285 /*
Rami Rosen2cfa2b12016-01-09 23:33:06 +02004286 * Migrate tasks one-by-one until @from is empty. This fails iff
Tejun Heo952aaa12014-02-25 10:04:03 -05004287 * ->can_attach() fails.
4288 */
Tejun Heoe406d1c2014-02-13 06:58:39 -05004289 do {
Tejun Heo9d800df2014-05-14 09:15:00 -04004290 css_task_iter_start(&from->self, &it);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004291 task = css_task_iter_next(&it);
4292 if (task)
4293 get_task_struct(task);
4294 css_task_iter_end(&it);
4295
4296 if (task) {
Tejun Heo37ff9f82016-03-08 11:51:26 -05004297 ret = cgroup_migrate(task, false, to->root);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004298 put_task_struct(task);
4299 }
4300 } while (task && !ret);
Tejun Heo952aaa12014-02-25 10:04:03 -05004301out_err:
4302 cgroup_migrate_finish(&preloaded_csets);
4303 mutex_unlock(&cgroup_mutex);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004304 return ret;
Tejun Heo8cc99342013-04-07 09:29:50 -07004305}
4306
Paul Menage817929e2007-10-18 23:39:36 -07004307/*
Ben Blum102a7752009-09-23 15:56:26 -07004308 * Stuff for reading the 'tasks'/'procs' files.
Paul Menagebbcb81d2007-10-18 23:39:32 -07004309 *
4310 * Reading this file can return large amounts of data if a cgroup has
4311 * *lots* of attached tasks. So it may need several calls to read(),
4312 * but we cannot guarantee that the information we produce is correct
4313 * unless we produce it entirely atomically.
4314 *
Paul Menagebbcb81d2007-10-18 23:39:32 -07004315 */
Paul Menagebbcb81d2007-10-18 23:39:32 -07004316
Li Zefan24528252012-01-20 11:58:43 +08004317/* which pidlist file are we talking about? */
4318enum cgroup_filetype {
4319 CGROUP_FILE_PROCS,
4320 CGROUP_FILE_TASKS,
4321};
4322
4323/*
4324 * A pidlist is a list of pids that virtually represents the contents of one
4325 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4326 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4327 * to the cgroup.
4328 */
4329struct cgroup_pidlist {
4330 /*
4331 * used to find which pidlist is wanted. doesn't change as long as
4332 * this particular list stays in the list.
4333 */
4334 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
4335 /* array of xids */
4336 pid_t *list;
4337 /* how many elements the above list has */
4338 int length;
Li Zefan24528252012-01-20 11:58:43 +08004339 /* each of these stored in a list by its cgroup */
4340 struct list_head links;
4341 /* pointer to the cgroup we belong to, for list removal purposes */
4342 struct cgroup *owner;
Tejun Heob1a21362013-11-29 10:42:58 -05004343 /* for delayed destruction */
4344 struct delayed_work destroy_dwork;
Li Zefan24528252012-01-20 11:58:43 +08004345};
4346
Paul Menagebbcb81d2007-10-18 23:39:32 -07004347/*
Ben Blumd1d9fd32009-09-23 15:56:28 -07004348 * The following two functions "fix" the issue where there are more pids
4349 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4350 * TODO: replace with a kernel-wide solution to this problem
4351 */
4352#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4353static void *pidlist_allocate(int count)
4354{
4355 if (PIDLIST_TOO_LARGE(count))
4356 return vmalloc(count * sizeof(pid_t));
4357 else
4358 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
4359}
Tejun Heob1a21362013-11-29 10:42:58 -05004360
Ben Blumd1d9fd32009-09-23 15:56:28 -07004361static void pidlist_free(void *p)
4362{
Bandan Das58794512015-03-02 17:51:10 -05004363 kvfree(p);
Ben Blumd1d9fd32009-09-23 15:56:28 -07004364}
Ben Blumd1d9fd32009-09-23 15:56:28 -07004365
4366/*
Tejun Heob1a21362013-11-29 10:42:58 -05004367 * Used to destroy all pidlists lingering waiting for destroy timer. None
4368 * should be left afterwards.
4369 */
4370static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
4371{
4372 struct cgroup_pidlist *l, *tmp_l;
4373
4374 mutex_lock(&cgrp->pidlist_mutex);
4375 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
4376 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
4377 mutex_unlock(&cgrp->pidlist_mutex);
4378
4379 flush_workqueue(cgroup_pidlist_destroy_wq);
4380 BUG_ON(!list_empty(&cgrp->pidlists));
4381}
4382
4383static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
4384{
4385 struct delayed_work *dwork = to_delayed_work(work);
4386 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
4387 destroy_dwork);
4388 struct cgroup_pidlist *tofree = NULL;
4389
4390 mutex_lock(&l->owner->pidlist_mutex);
Tejun Heob1a21362013-11-29 10:42:58 -05004391
4392 /*
Tejun Heo04502362013-11-29 10:42:59 -05004393 * Destroy iff we didn't get queued again. The state won't change
4394 * as destroy_dwork can only be queued while locked.
Tejun Heob1a21362013-11-29 10:42:58 -05004395 */
Tejun Heo04502362013-11-29 10:42:59 -05004396 if (!delayed_work_pending(dwork)) {
Tejun Heob1a21362013-11-29 10:42:58 -05004397 list_del(&l->links);
4398 pidlist_free(l->list);
4399 put_pid_ns(l->key.ns);
4400 tofree = l;
4401 }
4402
Tejun Heob1a21362013-11-29 10:42:58 -05004403 mutex_unlock(&l->owner->pidlist_mutex);
4404 kfree(tofree);
4405}
4406
4407/*
Ben Blum102a7752009-09-23 15:56:26 -07004408 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
Li Zefan6ee211a2013-03-12 15:36:00 -07004409 * Returns the number of unique elements.
Paul Menagebbcb81d2007-10-18 23:39:32 -07004410 */
Li Zefan6ee211a2013-03-12 15:36:00 -07004411static int pidlist_uniq(pid_t *list, int length)
Paul Menagebbcb81d2007-10-18 23:39:32 -07004412{
Ben Blum102a7752009-09-23 15:56:26 -07004413 int src, dest = 1;
Ben Blum102a7752009-09-23 15:56:26 -07004414
4415 /*
4416 * we presume the 0th element is unique, so i starts at 1. trivial
4417 * edge cases first; no work needs to be done for either
4418 */
4419 if (length == 0 || length == 1)
4420 return length;
4421 /* src and dest walk down the list; dest counts unique elements */
4422 for (src = 1; src < length; src++) {
4423 /* find next unique element */
4424 while (list[src] == list[src-1]) {
4425 src++;
4426 if (src == length)
4427 goto after;
4428 }
4429 /* dest always points to where the next unique element goes */
4430 list[dest] = list[src];
4431 dest++;
4432 }
4433after:
Ben Blum102a7752009-09-23 15:56:26 -07004434 return dest;
4435}
4436
Tejun Heoafb2bc12013-11-29 10:42:59 -05004437/*
4438 * The two pid files - task and cgroup.procs - guaranteed that the result
4439 * is sorted, which forced this whole pidlist fiasco. As pid order is
4440 * different per namespace, each namespace needs differently sorted list,
4441 * making it impossible to use, for example, single rbtree of member tasks
4442 * sorted by task pointer. As pidlists can be fairly large, allocating one
4443 * per open file is dangerous, so cgroup had to implement shared pool of
4444 * pidlists keyed by cgroup and namespace.
4445 *
4446 * All this extra complexity was caused by the original implementation
4447 * committing to an entirely unnecessary property. In the long term, we
Tejun Heoaa6ec292014-07-09 10:08:08 -04004448 * want to do away with it. Explicitly scramble sort order if on the
4449 * default hierarchy so that no such expectation exists in the new
4450 * interface.
Tejun Heoafb2bc12013-11-29 10:42:59 -05004451 *
4452 * Scrambling is done by swapping every two consecutive bits, which is
4453 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4454 */
4455static pid_t pid_fry(pid_t pid)
4456{
4457 unsigned a = pid & 0x55555555;
4458 unsigned b = pid & 0xAAAAAAAA;
4459
4460 return (a << 1) | (b >> 1);
4461}
4462
4463static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
4464{
Tejun Heoaa6ec292014-07-09 10:08:08 -04004465 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05004466 return pid_fry(pid);
4467 else
4468 return pid;
4469}
4470
Ben Blum102a7752009-09-23 15:56:26 -07004471static int cmppid(const void *a, const void *b)
4472{
4473 return *(pid_t *)a - *(pid_t *)b;
4474}
4475
Tejun Heoafb2bc12013-11-29 10:42:59 -05004476static int fried_cmppid(const void *a, const void *b)
4477{
4478 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
4479}
4480
Ben Blum72a8cb32009-09-23 15:56:27 -07004481static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
4482 enum cgroup_filetype type)
4483{
4484 struct cgroup_pidlist *l;
4485 /* don't need task_nsproxy() if we're looking at ourself */
Eric W. Biederman17cf22c2010-03-02 14:51:53 -08004486 struct pid_namespace *ns = task_active_pid_ns(current);
Li Zefanb70cc5f2010-03-10 15:22:12 -08004487
Tejun Heoe6b81712013-11-29 10:42:59 -05004488 lockdep_assert_held(&cgrp->pidlist_mutex);
4489
4490 list_for_each_entry(l, &cgrp->pidlists, links)
4491 if (l->key.type == type && l->key.ns == ns)
Ben Blum72a8cb32009-09-23 15:56:27 -07004492 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05004493 return NULL;
4494}
4495
Ben Blum72a8cb32009-09-23 15:56:27 -07004496/*
4497 * find the appropriate pidlist for our purpose (given procs vs tasks)
4498 * returns with the lock on that pidlist already held, and takes care
4499 * of the use count, or returns NULL with no locks held if we're out of
4500 * memory.
4501 */
Tejun Heoe6b81712013-11-29 10:42:59 -05004502static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
4503 enum cgroup_filetype type)
Ben Blum72a8cb32009-09-23 15:56:27 -07004504{
4505 struct cgroup_pidlist *l;
Ben Blum72a8cb32009-09-23 15:56:27 -07004506
Tejun Heoe6b81712013-11-29 10:42:59 -05004507 lockdep_assert_held(&cgrp->pidlist_mutex);
4508
4509 l = cgroup_pidlist_find(cgrp, type);
4510 if (l)
4511 return l;
4512
Ben Blum72a8cb32009-09-23 15:56:27 -07004513 /* entry not found; create a new one */
Tejun Heof4f4be22013-06-12 21:04:51 -07004514 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
Tejun Heoe6b81712013-11-29 10:42:59 -05004515 if (!l)
Ben Blum72a8cb32009-09-23 15:56:27 -07004516 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05004517
Tejun Heob1a21362013-11-29 10:42:58 -05004518 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
Ben Blum72a8cb32009-09-23 15:56:27 -07004519 l->key.type = type;
Tejun Heoe6b81712013-11-29 10:42:59 -05004520 /* don't need task_nsproxy() if we're looking at ourself */
4521 l->key.ns = get_pid_ns(task_active_pid_ns(current));
Ben Blum72a8cb32009-09-23 15:56:27 -07004522 l->owner = cgrp;
4523 list_add(&l->links, &cgrp->pidlists);
Ben Blum72a8cb32009-09-23 15:56:27 -07004524 return l;
4525}
4526
4527/*
Ben Blum102a7752009-09-23 15:56:26 -07004528 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4529 */
Ben Blum72a8cb32009-09-23 15:56:27 -07004530static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
4531 struct cgroup_pidlist **lp)
Ben Blum102a7752009-09-23 15:56:26 -07004532{
4533 pid_t *array;
4534 int length;
4535 int pid, n = 0; /* used for populating the array */
Tejun Heo72ec7022013-08-08 20:11:26 -04004536 struct css_task_iter it;
Paul Menage817929e2007-10-18 23:39:36 -07004537 struct task_struct *tsk;
Ben Blum102a7752009-09-23 15:56:26 -07004538 struct cgroup_pidlist *l;
4539
Tejun Heo4bac00d2013-11-29 10:42:59 -05004540 lockdep_assert_held(&cgrp->pidlist_mutex);
4541
Ben Blum102a7752009-09-23 15:56:26 -07004542 /*
4543 * If cgroup gets more users after we read count, we won't have
4544 * enough space - tough. This race is indistinguishable to the
4545 * caller from the case that the additional cgroup users didn't
4546 * show up until sometime later on.
4547 */
4548 length = cgroup_task_count(cgrp);
Ben Blumd1d9fd32009-09-23 15:56:28 -07004549 array = pidlist_allocate(length);
Ben Blum102a7752009-09-23 15:56:26 -07004550 if (!array)
4551 return -ENOMEM;
4552 /* now, populate the array */
Tejun Heo9d800df2014-05-14 09:15:00 -04004553 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04004554 while ((tsk = css_task_iter_next(&it))) {
Ben Blum102a7752009-09-23 15:56:26 -07004555 if (unlikely(n == length))
Paul Menage817929e2007-10-18 23:39:36 -07004556 break;
Ben Blum102a7752009-09-23 15:56:26 -07004557 /* get tgid or pid for procs or tasks file respectively */
Ben Blum72a8cb32009-09-23 15:56:27 -07004558 if (type == CGROUP_FILE_PROCS)
4559 pid = task_tgid_vnr(tsk);
4560 else
4561 pid = task_pid_vnr(tsk);
Ben Blum102a7752009-09-23 15:56:26 -07004562 if (pid > 0) /* make sure to only use valid results */
4563 array[n++] = pid;
Paul Menage817929e2007-10-18 23:39:36 -07004564 }
Tejun Heo72ec7022013-08-08 20:11:26 -04004565 css_task_iter_end(&it);
Ben Blum102a7752009-09-23 15:56:26 -07004566 length = n;
4567 /* now sort & (if procs) strip out duplicates */
Tejun Heoaa6ec292014-07-09 10:08:08 -04004568 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05004569 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
4570 else
4571 sort(array, length, sizeof(pid_t), cmppid, NULL);
Ben Blum72a8cb32009-09-23 15:56:27 -07004572 if (type == CGROUP_FILE_PROCS)
Li Zefan6ee211a2013-03-12 15:36:00 -07004573 length = pidlist_uniq(array, length);
Tejun Heoe6b81712013-11-29 10:42:59 -05004574
Tejun Heoe6b81712013-11-29 10:42:59 -05004575 l = cgroup_pidlist_find_create(cgrp, type);
Ben Blum72a8cb32009-09-23 15:56:27 -07004576 if (!l) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07004577 pidlist_free(array);
Ben Blum72a8cb32009-09-23 15:56:27 -07004578 return -ENOMEM;
Ben Blum102a7752009-09-23 15:56:26 -07004579 }
Tejun Heoe6b81712013-11-29 10:42:59 -05004580
4581 /* store array, freeing old if necessary */
Ben Blumd1d9fd32009-09-23 15:56:28 -07004582 pidlist_free(l->list);
Ben Blum102a7752009-09-23 15:56:26 -07004583 l->list = array;
4584 l->length = length;
Ben Blum72a8cb32009-09-23 15:56:27 -07004585 *lp = l;
Ben Blum102a7752009-09-23 15:56:26 -07004586 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004587}
4588
Balbir Singh846c7bb2007-10-18 23:39:44 -07004589/**
Li Zefana043e3b2008-02-23 15:24:09 -08004590 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07004591 * @stats: cgroupstats to fill information into
4592 * @dentry: A dentry entry belonging to the cgroup for which stats have
4593 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08004594 *
4595 * Build and fill cgroupstats so that taskstats can export it to user
4596 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07004597 */
4598int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
4599{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004600 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07004601 struct cgroup *cgrp;
Tejun Heo72ec7022013-08-08 20:11:26 -04004602 struct css_task_iter it;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004603 struct task_struct *tsk;
Li Zefan33d283b2008-11-19 15:36:48 -08004604
Tejun Heo2bd59d42014-02-11 11:52:49 -05004605 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4606 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4607 kernfs_type(kn) != KERNFS_DIR)
4608 return -EINVAL;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004609
Li Zefanbad34662014-02-14 16:54:28 +08004610 mutex_lock(&cgroup_mutex);
4611
Tejun Heo2bd59d42014-02-11 11:52:49 -05004612 /*
4613 * We aren't being called from kernfs and there's no guarantee on
Tejun Heoec903c02014-05-13 12:11:01 -04004614 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
Tejun Heo2bd59d42014-02-11 11:52:49 -05004615 * @kn->priv is RCU safe. Let's do the RCU dancing.
4616 */
4617 rcu_read_lock();
4618 cgrp = rcu_dereference(kn->priv);
Li Zefanbad34662014-02-14 16:54:28 +08004619 if (!cgrp || cgroup_is_dead(cgrp)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05004620 rcu_read_unlock();
Li Zefanbad34662014-02-14 16:54:28 +08004621 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004622 return -ENOENT;
4623 }
Li Zefanbad34662014-02-14 16:54:28 +08004624 rcu_read_unlock();
Balbir Singh846c7bb2007-10-18 23:39:44 -07004625
Tejun Heo9d800df2014-05-14 09:15:00 -04004626 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04004627 while ((tsk = css_task_iter_next(&it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07004628 switch (tsk->state) {
4629 case TASK_RUNNING:
4630 stats->nr_running++;
4631 break;
4632 case TASK_INTERRUPTIBLE:
4633 stats->nr_sleeping++;
4634 break;
4635 case TASK_UNINTERRUPTIBLE:
4636 stats->nr_uninterruptible++;
4637 break;
4638 case TASK_STOPPED:
4639 stats->nr_stopped++;
4640 break;
4641 default:
4642 if (delayacct_is_task_waiting_on_io(tsk))
4643 stats->nr_io_wait++;
4644 break;
4645 }
4646 }
Tejun Heo72ec7022013-08-08 20:11:26 -04004647 css_task_iter_end(&it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07004648
Li Zefanbad34662014-02-14 16:54:28 +08004649 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004650 return 0;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004651}
4652
Paul Menage8f3ff202009-09-23 15:56:25 -07004653
Paul Menagecc31edc2008-10-18 20:28:04 -07004654/*
Ben Blum102a7752009-09-23 15:56:26 -07004655 * seq_file methods for the tasks/procs files. The seq_file position is the
Paul Menagecc31edc2008-10-18 20:28:04 -07004656 * next pid to display; the seq_file iterator is a pointer to the pid
Ben Blum102a7752009-09-23 15:56:26 -07004657 * in the cgroup->l->list array.
Paul Menagecc31edc2008-10-18 20:28:04 -07004658 */
4659
Ben Blum102a7752009-09-23 15:56:26 -07004660static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004661{
4662 /*
4663 * Initially we receive a position value that corresponds to
4664 * one more than the last pid shown (or 0 on the first call or
4665 * after a seek to the start). Use a binary-search to find the
4666 * next pid to display, if any
4667 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05004668 struct kernfs_open_file *of = s->private;
Tejun Heo7da11272013-12-05 12:28:04 -05004669 struct cgroup *cgrp = seq_css(s)->cgroup;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004670 struct cgroup_pidlist *l;
Tejun Heo7da11272013-12-05 12:28:04 -05004671 enum cgroup_filetype type = seq_cft(s)->private;
Paul Menagecc31edc2008-10-18 20:28:04 -07004672 int index = 0, pid = *pos;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004673 int *iter, ret;
Paul Menagecc31edc2008-10-18 20:28:04 -07004674
Tejun Heo4bac00d2013-11-29 10:42:59 -05004675 mutex_lock(&cgrp->pidlist_mutex);
4676
4677 /*
Tejun Heo5d224442013-12-05 12:28:04 -05004678 * !NULL @of->priv indicates that this isn't the first start()
Tejun Heo4bac00d2013-11-29 10:42:59 -05004679 * after open. If the matching pidlist is around, we can use that.
Tejun Heo5d224442013-12-05 12:28:04 -05004680 * Look for it. Note that @of->priv can't be used directly. It
Tejun Heo4bac00d2013-11-29 10:42:59 -05004681 * could already have been destroyed.
4682 */
Tejun Heo5d224442013-12-05 12:28:04 -05004683 if (of->priv)
4684 of->priv = cgroup_pidlist_find(cgrp, type);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004685
4686 /*
4687 * Either this is the first start() after open or the matching
4688 * pidlist has been destroyed inbetween. Create a new one.
4689 */
Tejun Heo5d224442013-12-05 12:28:04 -05004690 if (!of->priv) {
4691 ret = pidlist_array_load(cgrp, type,
4692 (struct cgroup_pidlist **)&of->priv);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004693 if (ret)
4694 return ERR_PTR(ret);
4695 }
Tejun Heo5d224442013-12-05 12:28:04 -05004696 l = of->priv;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004697
Paul Menagecc31edc2008-10-18 20:28:04 -07004698 if (pid) {
Ben Blum102a7752009-09-23 15:56:26 -07004699 int end = l->length;
Stephen Rothwell20777762008-10-21 16:11:20 +11004700
Paul Menagecc31edc2008-10-18 20:28:04 -07004701 while (index < end) {
4702 int mid = (index + end) / 2;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004703 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
Paul Menagecc31edc2008-10-18 20:28:04 -07004704 index = mid;
4705 break;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004706 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
Paul Menagecc31edc2008-10-18 20:28:04 -07004707 index = mid + 1;
4708 else
4709 end = mid;
4710 }
4711 }
4712 /* If we're off the end of the array, we're done */
Ben Blum102a7752009-09-23 15:56:26 -07004713 if (index >= l->length)
Paul Menagecc31edc2008-10-18 20:28:04 -07004714 return NULL;
4715 /* Update the abstract position to be the actual pid that we found */
Ben Blum102a7752009-09-23 15:56:26 -07004716 iter = l->list + index;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004717 *pos = cgroup_pid_fry(cgrp, *iter);
Paul Menagecc31edc2008-10-18 20:28:04 -07004718 return iter;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004719}
4720
Ben Blum102a7752009-09-23 15:56:26 -07004721static void cgroup_pidlist_stop(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004722{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004723 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004724 struct cgroup_pidlist *l = of->priv;
Tejun Heo62236852013-11-29 10:42:58 -05004725
Tejun Heo5d224442013-12-05 12:28:04 -05004726 if (l)
4727 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
Tejun Heo04502362013-11-29 10:42:59 -05004728 CGROUP_PIDLIST_DESTROY_DELAY);
Tejun Heo7da11272013-12-05 12:28:04 -05004729 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07004730}
4731
Ben Blum102a7752009-09-23 15:56:26 -07004732static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004733{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004734 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004735 struct cgroup_pidlist *l = of->priv;
Ben Blum102a7752009-09-23 15:56:26 -07004736 pid_t *p = v;
4737 pid_t *end = l->list + l->length;
Paul Menagecc31edc2008-10-18 20:28:04 -07004738 /*
4739 * Advance to the next pid in the array. If this goes off the
4740 * end, we're done
4741 */
4742 p++;
4743 if (p >= end) {
4744 return NULL;
4745 } else {
Tejun Heo7da11272013-12-05 12:28:04 -05004746 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
Paul Menagecc31edc2008-10-18 20:28:04 -07004747 return p;
4748 }
4749}
4750
Ben Blum102a7752009-09-23 15:56:26 -07004751static int cgroup_pidlist_show(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004752{
Joe Perches94ff2122015-04-15 16:18:20 -07004753 seq_printf(s, "%d\n", *(int *)v);
4754
4755 return 0;
Paul Menagecc31edc2008-10-18 20:28:04 -07004756}
4757
Tejun Heo182446d2013-08-08 20:11:24 -04004758static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4759 struct cftype *cft)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004760{
Tejun Heo182446d2013-08-08 20:11:24 -04004761 return notify_on_release(css->cgroup);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004762}
4763
Tejun Heo182446d2013-08-08 20:11:24 -04004764static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4765 struct cftype *cft, u64 val)
Paul Menage6379c102008-07-25 01:47:01 -07004766{
Paul Menage6379c102008-07-25 01:47:01 -07004767 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004768 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004769 else
Tejun Heo182446d2013-08-08 20:11:24 -04004770 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004771 return 0;
4772}
4773
Tejun Heo182446d2013-08-08 20:11:24 -04004774static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4775 struct cftype *cft)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004776{
Tejun Heo182446d2013-08-08 20:11:24 -04004777 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004778}
4779
Tejun Heo182446d2013-08-08 20:11:24 -04004780static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4781 struct cftype *cft, u64 val)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004782{
4783 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004784 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004785 else
Tejun Heo182446d2013-08-08 20:11:24 -04004786 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004787 return 0;
4788}
4789
Tejun Heoa14c6872014-07-15 11:05:09 -04004790/* cgroup core interface files for the default hierarchy */
4791static struct cftype cgroup_dfl_base_files[] = {
4792 {
4793 .name = "cgroup.procs",
Tejun Heo6f60ead2015-09-18 17:54:23 -04004794 .file_offset = offsetof(struct cgroup, procs_file),
Tejun Heoa14c6872014-07-15 11:05:09 -04004795 .seq_start = cgroup_pidlist_start,
4796 .seq_next = cgroup_pidlist_next,
4797 .seq_stop = cgroup_pidlist_stop,
4798 .seq_show = cgroup_pidlist_show,
4799 .private = CGROUP_FILE_PROCS,
4800 .write = cgroup_procs_write,
Tejun Heoa14c6872014-07-15 11:05:09 -04004801 },
4802 {
4803 .name = "cgroup.controllers",
Tejun Heoa14c6872014-07-15 11:05:09 -04004804 .seq_show = cgroup_controllers_show,
4805 },
4806 {
4807 .name = "cgroup.subtree_control",
4808 .seq_show = cgroup_subtree_control_show,
4809 .write = cgroup_subtree_control_write,
4810 },
4811 {
Tejun Heo4a07c222015-09-18 17:54:22 -04004812 .name = "cgroup.events",
Tejun Heoa14c6872014-07-15 11:05:09 -04004813 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo6f60ead2015-09-18 17:54:23 -04004814 .file_offset = offsetof(struct cgroup, events_file),
Tejun Heo4a07c222015-09-18 17:54:22 -04004815 .seq_show = cgroup_events_show,
Tejun Heoa14c6872014-07-15 11:05:09 -04004816 },
4817 { } /* terminate */
4818};
4819
4820/* cgroup core interface files for the legacy hierarchies */
4821static struct cftype cgroup_legacy_base_files[] = {
Paul Menage81a6a5c2007-10-18 23:39:38 -07004822 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004823 .name = "cgroup.procs",
Tejun Heo6612f052013-12-05 12:28:04 -05004824 .seq_start = cgroup_pidlist_start,
4825 .seq_next = cgroup_pidlist_next,
4826 .seq_stop = cgroup_pidlist_stop,
4827 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004828 .private = CGROUP_FILE_PROCS,
Tejun Heoacbef752014-05-13 12:16:22 -04004829 .write = cgroup_procs_write,
Ben Blum102a7752009-09-23 15:56:26 -07004830 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07004831 {
Daniel Lezcano97978e62010-10-27 15:33:35 -07004832 .name = "cgroup.clone_children",
4833 .read_u64 = cgroup_clone_children_read,
4834 .write_u64 = cgroup_clone_children_write,
4835 },
Tejun Heo6e6ff252012-04-01 12:09:55 -07004836 {
Tejun Heo873fe092013-04-14 20:15:26 -07004837 .name = "cgroup.sane_behavior",
4838 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004839 .seq_show = cgroup_sane_behavior_show,
Tejun Heo873fe092013-04-14 20:15:26 -07004840 },
Tejun Heof8f22e52014-04-23 11:13:16 -04004841 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004842 .name = "tasks",
Tejun Heo6612f052013-12-05 12:28:04 -05004843 .seq_start = cgroup_pidlist_start,
4844 .seq_next = cgroup_pidlist_next,
4845 .seq_stop = cgroup_pidlist_stop,
4846 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004847 .private = CGROUP_FILE_TASKS,
Tejun Heoacbef752014-05-13 12:16:22 -04004848 .write = cgroup_tasks_write,
Tejun Heod5c56ce2013-06-03 19:14:34 -07004849 },
4850 {
4851 .name = "notify_on_release",
Tejun Heod5c56ce2013-06-03 19:14:34 -07004852 .read_u64 = cgroup_read_notify_on_release,
4853 .write_u64 = cgroup_write_notify_on_release,
4854 },
Tejun Heo873fe092013-04-14 20:15:26 -07004855 {
Tejun Heo6e6ff252012-04-01 12:09:55 -07004856 .name = "release_agent",
Tejun Heoa14c6872014-07-15 11:05:09 -04004857 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004858 .seq_show = cgroup_release_agent_show,
Tejun Heo451af5042014-05-13 12:16:21 -04004859 .write = cgroup_release_agent_write,
Tejun Heo5f469902014-02-11 11:52:48 -05004860 .max_write_len = PATH_MAX - 1,
Tejun Heo6e6ff252012-04-01 12:09:55 -07004861 },
Tejun Heodb0416b2012-04-01 12:09:55 -07004862 { } /* terminate */
Paul Menagebbcb81d2007-10-18 23:39:32 -07004863};
4864
Tejun Heo0c21ead2013-08-13 20:22:51 -04004865/*
4866 * css destruction is four-stage process.
4867 *
4868 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4869 * Implemented in kill_css().
4870 *
4871 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
Tejun Heoec903c02014-05-13 12:11:01 -04004872 * and thus css_tryget_online() is guaranteed to fail, the css can be
4873 * offlined by invoking offline_css(). After offlining, the base ref is
4874 * put. Implemented in css_killed_work_fn().
Tejun Heo0c21ead2013-08-13 20:22:51 -04004875 *
4876 * 3. When the percpu_ref reaches zero, the only possible remaining
4877 * accessors are inside RCU read sections. css_release() schedules the
4878 * RCU callback.
4879 *
4880 * 4. After the grace period, the css can be freed. Implemented in
4881 * css_free_work_fn().
4882 *
4883 * It is actually hairier because both step 2 and 4 require process context
4884 * and thus involve punting to css->destroy_work adding two additional
4885 * steps to the already complex sequence.
4886 */
Tejun Heo35ef10d2013-08-13 11:01:54 -04004887static void css_free_work_fn(struct work_struct *work)
Tejun Heo48ddbe12012-04-01 12:09:56 -07004888{
4889 struct cgroup_subsys_state *css =
Tejun Heo35ef10d2013-08-13 11:01:54 -04004890 container_of(work, struct cgroup_subsys_state, destroy_work);
Vladimir Davydov01e58652015-02-12 14:59:26 -08004891 struct cgroup_subsys *ss = css->ss;
Tejun Heo0c21ead2013-08-13 20:22:51 -04004892 struct cgroup *cgrp = css->cgroup;
Tejun Heo48ddbe12012-04-01 12:09:56 -07004893
Tejun Heo9a1049d2014-06-28 08:10:14 -04004894 percpu_ref_exit(&css->refcnt);
4895
Vladimir Davydov01e58652015-02-12 14:59:26 -08004896 if (ss) {
Tejun Heo9d755d32014-05-14 09:15:02 -04004897 /* css free path */
Tejun Heo8bb5ef72016-01-21 15:32:15 -05004898 struct cgroup_subsys_state *parent = css->parent;
Vladimir Davydov01e58652015-02-12 14:59:26 -08004899 int id = css->id;
4900
Vladimir Davydov01e58652015-02-12 14:59:26 -08004901 ss->css_free(css);
4902 cgroup_idr_remove(&ss->css_idr, id);
Tejun Heo9d755d32014-05-14 09:15:02 -04004903 cgroup_put(cgrp);
Tejun Heo8bb5ef72016-01-21 15:32:15 -05004904
4905 if (parent)
4906 css_put(parent);
Tejun Heo9d755d32014-05-14 09:15:02 -04004907 } else {
4908 /* cgroup free path */
4909 atomic_dec(&cgrp->root->nr_cgrps);
4910 cgroup_pidlist_destroy_all(cgrp);
Zefan Li971ff492014-09-18 16:06:19 +08004911 cancel_work_sync(&cgrp->release_agent_work);
Tejun Heo9d755d32014-05-14 09:15:02 -04004912
Tejun Heod51f39b2014-05-16 13:22:48 -04004913 if (cgroup_parent(cgrp)) {
Tejun Heo9d755d32014-05-14 09:15:02 -04004914 /*
4915 * We get a ref to the parent, and put the ref when
4916 * this cgroup is being freed, so it's guaranteed
4917 * that the parent won't be destroyed before its
4918 * children.
4919 */
Tejun Heod51f39b2014-05-16 13:22:48 -04004920 cgroup_put(cgroup_parent(cgrp));
Tejun Heo9d755d32014-05-14 09:15:02 -04004921 kernfs_put(cgrp->kn);
4922 kfree(cgrp);
4923 } else {
4924 /*
4925 * This is root cgroup's refcnt reaching zero,
4926 * which indicates that the root should be
4927 * released.
4928 */
4929 cgroup_destroy_root(cgrp->root);
4930 }
4931 }
Tejun Heo0c21ead2013-08-13 20:22:51 -04004932}
4933
4934static void css_free_rcu_fn(struct rcu_head *rcu_head)
4935{
4936 struct cgroup_subsys_state *css =
4937 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4938
Tejun Heo0c21ead2013-08-13 20:22:51 -04004939 INIT_WORK(&css->destroy_work, css_free_work_fn);
Tejun Heoe5fca242013-11-22 17:14:39 -05004940 queue_work(cgroup_destroy_wq, &css->destroy_work);
Tejun Heo48ddbe12012-04-01 12:09:56 -07004941}
4942
Tejun Heo25e15d82014-05-14 09:15:02 -04004943static void css_release_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07004944{
4945 struct cgroup_subsys_state *css =
Tejun Heo25e15d82014-05-14 09:15:02 -04004946 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heo15a4c832014-05-04 15:09:14 -04004947 struct cgroup_subsys *ss = css->ss;
Tejun Heo9d755d32014-05-14 09:15:02 -04004948 struct cgroup *cgrp = css->cgroup;
Tejun Heod3daf282013-06-13 19:39:16 -07004949
Tejun Heo1fed1b22014-05-16 13:22:49 -04004950 mutex_lock(&cgroup_mutex);
4951
Tejun Heode3f0342014-05-16 13:22:49 -04004952 css->flags |= CSS_RELEASED;
Tejun Heo1fed1b22014-05-16 13:22:49 -04004953 list_del_rcu(&css->sibling);
4954
Tejun Heo9d755d32014-05-14 09:15:02 -04004955 if (ss) {
4956 /* css release path */
Vladimir Davydov01e58652015-02-12 14:59:26 -08004957 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
Tejun Heo7d172cc2014-11-18 02:49:51 -05004958 if (ss->css_released)
4959 ss->css_released(css);
Tejun Heo9d755d32014-05-14 09:15:02 -04004960 } else {
4961 /* cgroup release path */
Tejun Heo9d755d32014-05-14 09:15:02 -04004962 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4963 cgrp->id = -1;
Li Zefana4189482014-09-04 14:43:07 +08004964
4965 /*
4966 * There are two control paths which try to determine
4967 * cgroup from dentry without going through kernfs -
4968 * cgroupstats_build() and css_tryget_online_from_dir().
4969 * Those are supported by RCU protecting clearing of
4970 * cgrp->kn->priv backpointer.
4971 */
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05004972 if (cgrp->kn)
4973 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
4974 NULL);
Tejun Heo9d755d32014-05-14 09:15:02 -04004975 }
Tejun Heo15a4c832014-05-04 15:09:14 -04004976
Tejun Heo1fed1b22014-05-16 13:22:49 -04004977 mutex_unlock(&cgroup_mutex);
4978
Tejun Heo0c21ead2013-08-13 20:22:51 -04004979 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07004980}
4981
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004982static void css_release(struct percpu_ref *ref)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004983{
4984 struct cgroup_subsys_state *css =
4985 container_of(ref, struct cgroup_subsys_state, refcnt);
4986
Tejun Heo25e15d82014-05-14 09:15:02 -04004987 INIT_WORK(&css->destroy_work, css_release_work_fn);
4988 queue_work(cgroup_destroy_wq, &css->destroy_work);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004989}
4990
Tejun Heoddfcada2014-05-04 15:09:14 -04004991static void init_and_link_css(struct cgroup_subsys_state *css,
4992 struct cgroup_subsys *ss, struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004993{
Tejun Heo0cb51d72014-05-16 13:22:49 -04004994 lockdep_assert_held(&cgroup_mutex);
4995
Tejun Heoddfcada2014-05-04 15:09:14 -04004996 cgroup_get(cgrp);
4997
Tejun Heod5c419b2014-05-16 13:22:48 -04004998 memset(css, 0, sizeof(*css));
Paul Menagebd89aab2007-10-18 23:40:44 -07004999 css->cgroup = cgrp;
Tejun Heo72c97e52013-08-08 20:11:22 -04005000 css->ss = ss;
Tejun Heod5c419b2014-05-16 13:22:48 -04005001 INIT_LIST_HEAD(&css->sibling);
5002 INIT_LIST_HEAD(&css->children);
Tejun Heo0cb51d72014-05-16 13:22:49 -04005003 css->serial_nr = css_serial_nr_next++;
Tejun Heoaa226ff2016-01-21 15:31:11 -05005004 atomic_set(&css->online_cnt, 0);
Tejun Heo48ddbe12012-04-01 12:09:56 -07005005
Tejun Heod51f39b2014-05-16 13:22:48 -04005006 if (cgroup_parent(cgrp)) {
5007 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
Tejun Heoddfcada2014-05-04 15:09:14 -04005008 css_get(css->parent);
Tejun Heoddfcada2014-05-04 15:09:14 -04005009 }
Tejun Heo0ae78e02013-08-13 11:01:54 -04005010
Tejun Heoca8bdca2013-08-26 18:40:56 -04005011 BUG_ON(cgroup_css(cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005012}
5013
Li Zefan2a4ac632013-07-31 16:16:40 +08005014/* invoke ->css_online() on a new CSS and mark it online if successful */
Tejun Heo623f9262013-08-13 11:01:55 -04005015static int online_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08005016{
Tejun Heo623f9262013-08-13 11:01:55 -04005017 struct cgroup_subsys *ss = css->ss;
Tejun Heob1929db2012-11-19 08:13:38 -08005018 int ret = 0;
5019
Tejun Heoa31f2d32012-11-19 08:13:37 -08005020 lockdep_assert_held(&cgroup_mutex);
5021
Tejun Heo92fb9742012-11-19 08:13:38 -08005022 if (ss->css_online)
Tejun Heoeb954192013-08-08 20:11:23 -04005023 ret = ss->css_online(css);
Tejun Heoae7f1642013-08-13 20:22:50 -04005024 if (!ret) {
Tejun Heoeb954192013-08-08 20:11:23 -04005025 css->flags |= CSS_ONLINE;
Tejun Heoaec25022014-02-08 10:36:58 -05005026 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
Tejun Heoaa226ff2016-01-21 15:31:11 -05005027
5028 atomic_inc(&css->online_cnt);
5029 if (css->parent)
5030 atomic_inc(&css->parent->online_cnt);
Tejun Heoae7f1642013-08-13 20:22:50 -04005031 }
Tejun Heob1929db2012-11-19 08:13:38 -08005032 return ret;
Tejun Heoa31f2d32012-11-19 08:13:37 -08005033}
5034
Li Zefan2a4ac632013-07-31 16:16:40 +08005035/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
Tejun Heo623f9262013-08-13 11:01:55 -04005036static void offline_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08005037{
Tejun Heo623f9262013-08-13 11:01:55 -04005038 struct cgroup_subsys *ss = css->ss;
Tejun Heoa31f2d32012-11-19 08:13:37 -08005039
5040 lockdep_assert_held(&cgroup_mutex);
5041
5042 if (!(css->flags & CSS_ONLINE))
5043 return;
5044
Vladimir Davydovfa062352016-03-01 19:56:30 +03005045 if (ss->css_reset)
5046 ss->css_reset(css);
5047
Li Zefand7eeac12013-03-12 15:35:59 -07005048 if (ss->css_offline)
Tejun Heoeb954192013-08-08 20:11:23 -04005049 ss->css_offline(css);
Tejun Heoa31f2d32012-11-19 08:13:37 -08005050
Tejun Heoeb954192013-08-08 20:11:23 -04005051 css->flags &= ~CSS_ONLINE;
Tejun Heoe3297802014-04-23 11:13:15 -04005052 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
Tejun Heof8f22e52014-04-23 11:13:16 -04005053
5054 wake_up_all(&css->cgroup->offline_waitq);
Tejun Heoa31f2d32012-11-19 08:13:37 -08005055}
5056
Tejun Heoc81c925a2013-12-06 15:11:56 -05005057/**
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005058 * css_create - create a cgroup_subsys_state
Tejun Heoc81c925a2013-12-06 15:11:56 -05005059 * @cgrp: the cgroup new css will be associated with
5060 * @ss: the subsys of new css
5061 *
5062 * Create a new css associated with @cgrp - @ss pair. On success, the new
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005063 * css is online and installed in @cgrp. This function doesn't create the
5064 * interface files. Returns 0 on success, -errno on failure.
Tejun Heoc81c925a2013-12-06 15:11:56 -05005065 */
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005066static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5067 struct cgroup_subsys *ss)
Tejun Heoc81c925a2013-12-06 15:11:56 -05005068{
Tejun Heod51f39b2014-05-16 13:22:48 -04005069 struct cgroup *parent = cgroup_parent(cgrp);
Tejun Heo1fed1b22014-05-16 13:22:49 -04005070 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005071 struct cgroup_subsys_state *css;
5072 int err;
5073
Tejun Heoc81c925a2013-12-06 15:11:56 -05005074 lockdep_assert_held(&cgroup_mutex);
5075
Tejun Heo1fed1b22014-05-16 13:22:49 -04005076 css = ss->css_alloc(parent_css);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005077 if (IS_ERR(css))
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005078 return css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005079
Tejun Heoddfcada2014-05-04 15:09:14 -04005080 init_and_link_css(css, ss, cgrp);
Tejun Heoa2bed822014-05-04 15:09:14 -04005081
Tejun Heo2aad2a82014-09-24 13:31:50 -04005082 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005083 if (err)
Li Zefan3eb59ec2014-03-18 17:02:36 +08005084 goto err_free_css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005085
Vladimir Davydovcf780b72015-08-03 15:32:26 +03005086 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
Tejun Heo15a4c832014-05-04 15:09:14 -04005087 if (err < 0)
5088 goto err_free_percpu_ref;
5089 css->id = err;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005090
Tejun Heo15a4c832014-05-04 15:09:14 -04005091 /* @css is ready to be brought online now, make it visible */
Tejun Heo1fed1b22014-05-16 13:22:49 -04005092 list_add_tail_rcu(&css->sibling, &parent_css->children);
Tejun Heo15a4c832014-05-04 15:09:14 -04005093 cgroup_idr_replace(&ss->css_idr, css, css->id);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005094
5095 err = online_css(css);
5096 if (err)
Tejun Heo1fed1b22014-05-16 13:22:49 -04005097 goto err_list_del;
Tejun Heo94419622014-03-19 10:23:54 -04005098
Tejun Heoc81c925a2013-12-06 15:11:56 -05005099 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
Tejun Heod51f39b2014-05-16 13:22:48 -04005100 cgroup_parent(parent)) {
Joe Perchesed3d2612014-04-25 18:28:03 -04005101 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 Zhana2a1f9e2014-04-25 18:28:03 -04005102 current->comm, current->pid, ss->name);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005103 if (!strcmp(ss->name, "memory"))
Joe Perchesed3d2612014-04-25 18:28:03 -04005104 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
Tejun Heoc81c925a2013-12-06 15:11:56 -05005105 ss->warned_broken_hierarchy = true;
5106 }
5107
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005108 return css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005109
Tejun Heo1fed1b22014-05-16 13:22:49 -04005110err_list_del:
5111 list_del_rcu(&css->sibling);
Tejun Heo15a4c832014-05-04 15:09:14 -04005112 cgroup_idr_remove(&ss->css_idr, css->id);
Li Zefan3eb59ec2014-03-18 17:02:36 +08005113err_free_percpu_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04005114 percpu_ref_exit(&css->refcnt);
Li Zefan3eb59ec2014-03-18 17:02:36 +08005115err_free_css:
Tejun Heoa2bed822014-05-04 15:09:14 -04005116 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005117 return ERR_PTR(err);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005118}
5119
Tejun Heoa5bca212016-03-03 09:57:58 -05005120static struct cgroup *cgroup_create(struct cgroup *parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005121{
Tejun Heoa5bca212016-03-03 09:57:58 -05005122 struct cgroup_root *root = parent->root;
Tejun Heoa5bca212016-03-03 09:57:58 -05005123 struct cgroup *cgrp, *tcgrp;
5124 int level = parent->level + 1;
Tejun Heo03970d32016-03-03 09:58:00 -05005125 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005126
Tejun Heo0a950f62012-11-19 09:02:12 -08005127 /* allocate the cgroup and its ID, 0 is reserved for the root */
Tejun Heob11cfb52015-11-20 15:55:52 -05005128 cgrp = kzalloc(sizeof(*cgrp) +
5129 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
Tejun Heoa5bca212016-03-03 09:57:58 -05005130 if (!cgrp)
5131 return ERR_PTR(-ENOMEM);
Li Zefan0ab02ca2014-02-11 16:05:46 +08005132
Tejun Heo2aad2a82014-09-24 13:31:50 -04005133 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04005134 if (ret)
5135 goto out_free_cgrp;
5136
Li Zefan0ab02ca2014-02-11 16:05:46 +08005137 /*
5138 * Temporarily set the pointer to NULL, so idr_find() won't return
5139 * a half-baked cgroup.
5140 */
Vladimir Davydovcf780b72015-08-03 15:32:26 +03005141 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
Li Zefan0ab02ca2014-02-11 16:05:46 +08005142 if (cgrp->id < 0) {
Tejun Heoba0f4d72014-05-13 12:19:22 -04005143 ret = -ENOMEM;
Tejun Heo9d755d32014-05-14 09:15:02 -04005144 goto out_cancel_ref;
Tejun Heo976c06b2012-11-05 09:16:59 -08005145 }
5146
Paul Menagecc31edc2008-10-18 20:28:04 -07005147 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005148
Tejun Heo9d800df2014-05-14 09:15:00 -04005149 cgrp->self.parent = &parent->self;
Tejun Heoba0f4d72014-05-13 12:19:22 -04005150 cgrp->root = root;
Tejun Heob11cfb52015-11-20 15:55:52 -05005151 cgrp->level = level;
5152
5153 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
5154 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005155
Li Zefanb6abdb02008-03-04 14:28:19 -08005156 if (notify_on_release(parent))
5157 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5158
Tejun Heo2260e7f2012-11-19 08:13:38 -08005159 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5160 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07005161
Tejun Heo0cb51d72014-05-16 13:22:49 -04005162 cgrp->self.serial_nr = css_serial_nr_next++;
Tejun Heo53fa5262013-05-24 10:55:38 +09005163
Tejun Heo4e139af2012-11-19 08:13:36 -08005164 /* allocation complete, commit to creation */
Tejun Heod5c419b2014-05-16 13:22:48 -04005165 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
Tejun Heo3c9c8252014-02-12 09:29:50 -05005166 atomic_inc(&root->nr_cgrps);
Tejun Heo59f52962014-02-11 11:52:49 -05005167 cgroup_get(parent);
Li Zefan415cf072013-04-08 14:35:02 +08005168
Tejun Heo0d802552013-12-06 15:11:56 -05005169 /*
5170 * @cgrp is now fully operational. If something fails after this
5171 * point, it'll be released via the normal destruction path.
5172 */
Tejun Heo6fa49182014-05-04 15:09:13 -04005173 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08005174
Tejun Heobd53d612014-04-23 11:13:16 -04005175 /*
5176 * On the default hierarchy, a child doesn't automatically inherit
Tejun Heo667c2492014-07-08 18:02:56 -04005177 * subtree_control from the parent. Each is configured manually.
Tejun Heobd53d612014-04-23 11:13:16 -04005178 */
Tejun Heo03970d32016-03-03 09:58:00 -05005179 if (!cgroup_on_dfl(cgrp))
Tejun Heo5531dc92016-03-03 09:57:58 -05005180 cgrp->subtree_control = cgroup_control(cgrp);
Tejun Heo03970d32016-03-03 09:58:00 -05005181
5182 cgroup_propagate_control(cgrp);
5183
5184 /* @cgrp doesn't have dir yet so the following will only create csses */
5185 ret = cgroup_apply_control_enable(cgrp);
5186 if (ret)
5187 goto out_destroy;
Tejun Heof392e512014-04-23 11:13:14 -04005188
Tejun Heoa5bca212016-03-03 09:57:58 -05005189 return cgrp;
5190
5191out_cancel_ref:
5192 percpu_ref_exit(&cgrp->self.refcnt);
5193out_free_cgrp:
5194 kfree(cgrp);
5195 return ERR_PTR(ret);
5196out_destroy:
5197 cgroup_destroy_locked(cgrp);
5198 return ERR_PTR(ret);
5199}
5200
5201static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
5202 umode_t mode)
5203{
5204 struct cgroup *parent, *cgrp;
Tejun Heoa5bca212016-03-03 09:57:58 -05005205 struct kernfs_node *kn;
Tejun Heo03970d32016-03-03 09:58:00 -05005206 int ret;
Tejun Heoa5bca212016-03-03 09:57:58 -05005207
5208 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5209 if (strchr(name, '\n'))
5210 return -EINVAL;
5211
Tejun Heo945ba192016-03-03 09:58:00 -05005212 parent = cgroup_kn_lock_live(parent_kn, false);
Tejun Heoa5bca212016-03-03 09:57:58 -05005213 if (!parent)
5214 return -ENODEV;
5215
5216 cgrp = cgroup_create(parent);
5217 if (IS_ERR(cgrp)) {
5218 ret = PTR_ERR(cgrp);
5219 goto out_unlock;
5220 }
5221
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005222 /* create the directory */
5223 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5224 if (IS_ERR(kn)) {
5225 ret = PTR_ERR(kn);
5226 goto out_destroy;
5227 }
5228 cgrp->kn = kn;
5229
5230 /*
5231 * This extra ref will be put in cgroup_free_fn() and guarantees
5232 * that @cgrp->kn is always accessible.
5233 */
5234 kernfs_get(kn);
5235
5236 ret = cgroup_kn_set_ugid(kn);
5237 if (ret)
5238 goto out_destroy;
5239
Tejun Heo334c3672016-03-03 09:58:01 -05005240 ret = css_populate_dir(&cgrp->self);
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005241 if (ret)
5242 goto out_destroy;
5243
Tejun Heo03970d32016-03-03 09:58:00 -05005244 ret = cgroup_apply_control_enable(cgrp);
5245 if (ret)
5246 goto out_destroy;
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005247
5248 /* let's create and online css's */
Tejun Heo2bd59d42014-02-11 11:52:49 -05005249 kernfs_activate(kn);
5250
Tejun Heoba0f4d72014-05-13 12:19:22 -04005251 ret = 0;
5252 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005253
Tejun Heoa5bca212016-03-03 09:57:58 -05005254out_destroy:
5255 cgroup_destroy_locked(cgrp);
Tejun Heoba0f4d72014-05-13 12:19:22 -04005256out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04005257 cgroup_kn_unlock(parent_kn);
Tejun Heoe1b2dc12014-03-20 11:10:15 -04005258 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005259}
5260
Tejun Heo223dbc32013-08-13 20:22:50 -04005261/*
5262 * This is called when the refcnt of a css is confirmed to be killed.
Tejun Heo249f3462014-05-14 09:15:01 -04005263 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5264 * initate destruction and put the css ref from kill_css().
Tejun Heo223dbc32013-08-13 20:22:50 -04005265 */
5266static void css_killed_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07005267{
Tejun Heo223dbc32013-08-13 20:22:50 -04005268 struct cgroup_subsys_state *css =
5269 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heod3daf282013-06-13 19:39:16 -07005270
Tejun Heof20104d2013-08-13 20:22:50 -04005271 mutex_lock(&cgroup_mutex);
Tejun Heo09a503ea2013-08-13 20:22:50 -04005272
Tejun Heoaa226ff2016-01-21 15:31:11 -05005273 do {
5274 offline_css(css);
5275 css_put(css);
5276 /* @css can't go away while we're holding cgroup_mutex */
5277 css = css->parent;
5278 } while (css && atomic_dec_and_test(&css->online_cnt));
5279
5280 mutex_unlock(&cgroup_mutex);
Tejun Heod3daf282013-06-13 19:39:16 -07005281}
5282
Tejun Heo223dbc32013-08-13 20:22:50 -04005283/* css kill confirmation processing requires process context, bounce */
5284static void css_killed_ref_fn(struct percpu_ref *ref)
Tejun Heod3daf282013-06-13 19:39:16 -07005285{
5286 struct cgroup_subsys_state *css =
5287 container_of(ref, struct cgroup_subsys_state, refcnt);
5288
Tejun Heoaa226ff2016-01-21 15:31:11 -05005289 if (atomic_dec_and_test(&css->online_cnt)) {
5290 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5291 queue_work(cgroup_destroy_wq, &css->destroy_work);
5292 }
Tejun Heod3daf282013-06-13 19:39:16 -07005293}
5294
Tejun Heof392e512014-04-23 11:13:14 -04005295/**
5296 * kill_css - destroy a css
5297 * @css: css to destroy
5298 *
5299 * This function initiates destruction of @css by removing cgroup interface
5300 * files and putting its base reference. ->css_offline() will be invoked
Tejun Heoec903c02014-05-13 12:11:01 -04005301 * asynchronously once css_tryget_online() is guaranteed to fail and when
5302 * the reference count reaches zero, @css will be released.
Tejun Heof392e512014-04-23 11:13:14 -04005303 */
5304static void kill_css(struct cgroup_subsys_state *css)
Tejun Heoedae0c32013-08-13 20:22:51 -04005305{
Tejun Heo01f64742014-05-13 12:19:23 -04005306 lockdep_assert_held(&cgroup_mutex);
Tejun Heo94419622014-03-19 10:23:54 -04005307
Tejun Heo2bd59d42014-02-11 11:52:49 -05005308 /*
5309 * This must happen before css is disassociated with its cgroup.
5310 * See seq_css() for details.
5311 */
Tejun Heo334c3672016-03-03 09:58:01 -05005312 css_clear_dir(css);
Tejun Heo3c14f8b2013-08-13 20:22:51 -04005313
Tejun Heoedae0c32013-08-13 20:22:51 -04005314 /*
5315 * Killing would put the base ref, but we need to keep it alive
5316 * until after ->css_offline().
5317 */
5318 css_get(css);
5319
5320 /*
5321 * cgroup core guarantees that, by the time ->css_offline() is
5322 * invoked, no new css reference will be given out via
Tejun Heoec903c02014-05-13 12:11:01 -04005323 * css_tryget_online(). We can't simply call percpu_ref_kill() and
Tejun Heoedae0c32013-08-13 20:22:51 -04005324 * proceed to offlining css's because percpu_ref_kill() doesn't
5325 * guarantee that the ref is seen as killed on all CPUs on return.
5326 *
5327 * Use percpu_ref_kill_and_confirm() to get notifications as each
5328 * css is confirmed to be seen as killed on all CPUs.
5329 */
5330 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07005331}
5332
5333/**
5334 * cgroup_destroy_locked - the first stage of cgroup destruction
5335 * @cgrp: cgroup to be destroyed
5336 *
5337 * css's make use of percpu refcnts whose killing latency shouldn't be
5338 * exposed to userland and are RCU protected. Also, cgroup core needs to
Tejun Heoec903c02014-05-13 12:11:01 -04005339 * guarantee that css_tryget_online() won't succeed by the time
5340 * ->css_offline() is invoked. To satisfy all the requirements,
5341 * destruction is implemented in the following two steps.
Tejun Heod3daf282013-06-13 19:39:16 -07005342 *
5343 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5344 * userland visible parts and start killing the percpu refcnts of
5345 * css's. Set up so that the next stage will be kicked off once all
5346 * the percpu refcnts are confirmed to be killed.
5347 *
5348 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5349 * rest of destruction. Once all cgroup references are gone, the
5350 * cgroup is RCU-freed.
5351 *
5352 * This function implements s1. After this step, @cgrp is gone as far as
5353 * the userland is concerned and a new cgroup with the same name may be
5354 * created. As cgroup doesn't care about the names internally, this
5355 * doesn't cause any problem.
5356 */
Tejun Heo42809dd2012-11-19 08:13:37 -08005357static int cgroup_destroy_locked(struct cgroup *cgrp)
5358 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005359{
Tejun Heo2bd59d42014-02-11 11:52:49 -05005360 struct cgroup_subsys_state *css;
Tejun Heo2b021cb2016-03-15 20:43:04 -04005361 struct cgrp_cset_link *link;
Tejun Heo1c6727a2013-12-06 15:11:56 -05005362 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005363
Tejun Heo42809dd2012-11-19 08:13:37 -08005364 lockdep_assert_held(&cgroup_mutex);
5365
Tejun Heo91486f62015-10-15 16:41:51 -04005366 /*
5367 * Only migration can raise populated from zero and we're already
5368 * holding cgroup_mutex.
5369 */
5370 if (cgroup_is_populated(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07005371 return -EBUSY;
Tejun Heoed9577932012-11-05 09:16:58 -08005372
Tejun Heo1a90dd52012-11-05 09:16:59 -08005373 /*
Tejun Heod5c419b2014-05-16 13:22:48 -04005374 * Make sure there's no live children. We can't test emptiness of
5375 * ->self.children as dead children linger on it while being
5376 * drained; otherwise, "rmdir parent/child parent" may fail.
Hugh Dickinsbb78a922013-08-28 16:31:23 -07005377 */
Tejun Heof3d46502014-05-16 13:22:52 -04005378 if (css_has_online_children(&cgrp->self))
Hugh Dickinsbb78a922013-08-28 16:31:23 -07005379 return -EBUSY;
5380
5381 /*
Tejun Heo2b021cb2016-03-15 20:43:04 -04005382 * Mark @cgrp and the associated csets dead. The former prevents
5383 * further task migration and child creation by disabling
5384 * cgroup_lock_live_group(). The latter makes the csets ignored by
5385 * the migration path.
Tejun Heo455050d2013-06-13 19:27:41 -07005386 */
Tejun Heo184faf32014-05-16 13:22:51 -04005387 cgrp->self.flags &= ~CSS_ONLINE;
Tejun Heo1a90dd52012-11-05 09:16:59 -08005388
Tejun Heo2b021cb2016-03-15 20:43:04 -04005389 spin_lock_bh(&css_set_lock);
5390 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5391 link->cset->dead = true;
5392 spin_unlock_bh(&css_set_lock);
5393
Tejun Heo249f3462014-05-14 09:15:01 -04005394 /* initiate massacre of all css's */
Tejun Heo1a90dd52012-11-05 09:16:59 -08005395 for_each_css(css, ssid, cgrp)
Tejun Heo455050d2013-06-13 19:27:41 -07005396 kill_css(css);
5397
Tejun Heo455050d2013-06-13 19:27:41 -07005398 /*
Tejun Heo01f64742014-05-13 12:19:23 -04005399 * Remove @cgrp directory along with the base files. @cgrp has an
5400 * extra ref on its kn.
Tejun Heo455050d2013-06-13 19:27:41 -07005401 */
Tejun Heo01f64742014-05-13 12:19:23 -04005402 kernfs_remove(cgrp->kn);
Tejun Heof20104d2013-08-13 20:22:50 -04005403
Tejun Heod51f39b2014-05-16 13:22:48 -04005404 check_for_release(cgroup_parent(cgrp));
Tejun Heo2bd59d42014-02-11 11:52:49 -05005405
Tejun Heo249f3462014-05-14 09:15:01 -04005406 /* put the base reference */
Tejun Heo9d755d32014-05-14 09:15:02 -04005407 percpu_ref_kill(&cgrp->self.refcnt);
Tejun Heo455050d2013-06-13 19:27:41 -07005408
Tejun Heoea15f8c2013-06-13 19:27:42 -07005409 return 0;
5410};
5411
Tejun Heo2bd59d42014-02-11 11:52:49 -05005412static int cgroup_rmdir(struct kernfs_node *kn)
Tejun Heo42809dd2012-11-19 08:13:37 -08005413{
Tejun Heoa9746d82014-05-13 12:19:22 -04005414 struct cgroup *cgrp;
Tejun Heo2bd59d42014-02-11 11:52:49 -05005415 int ret = 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08005416
Tejun Heo945ba192016-03-03 09:58:00 -05005417 cgrp = cgroup_kn_lock_live(kn, false);
Tejun Heoa9746d82014-05-13 12:19:22 -04005418 if (!cgrp)
5419 return 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08005420
Tejun Heoa9746d82014-05-13 12:19:22 -04005421 ret = cgroup_destroy_locked(cgrp);
Tejun Heo42809dd2012-11-19 08:13:37 -08005422
Tejun Heoa9746d82014-05-13 12:19:22 -04005423 cgroup_kn_unlock(kn);
Tejun Heo42809dd2012-11-19 08:13:37 -08005424 return ret;
5425}
5426
Tejun Heo2bd59d42014-02-11 11:52:49 -05005427static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5428 .remount_fs = cgroup_remount,
5429 .show_options = cgroup_show_options,
5430 .mkdir = cgroup_mkdir,
5431 .rmdir = cgroup_rmdir,
5432 .rename = cgroup_rename,
5433};
Tejun Heo8e3f6542012-04-01 12:09:55 -07005434
Tejun Heo15a4c832014-05-04 15:09:14 -04005435static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005436{
Paul Menageddbcc7e2007-10-18 23:39:30 -07005437 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08005438
Tejun Heoa5ae9892015-12-29 14:53:56 -05005439 pr_debug("Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005440
Tejun Heo648bb562012-11-19 08:13:36 -08005441 mutex_lock(&cgroup_mutex);
5442
Tejun Heo15a4c832014-05-04 15:09:14 -04005443 idr_init(&ss->css_idr);
Tejun Heo0adb0702014-02-12 09:29:48 -05005444 INIT_LIST_HEAD(&ss->cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -07005445
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005446 /* Create the root cgroup state for this subsystem */
5447 ss->root = &cgrp_dfl_root;
5448 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005449 /* We don't handle early failures gracefully */
5450 BUG_ON(IS_ERR(css));
Tejun Heoddfcada2014-05-04 15:09:14 -04005451 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
Tejun Heo3b514d22014-05-16 13:22:47 -04005452
5453 /*
5454 * Root csses are never destroyed and we can't initialize
5455 * percpu_ref during early init. Disable refcnting.
5456 */
5457 css->flags |= CSS_NO_REF;
5458
Tejun Heo15a4c832014-05-04 15:09:14 -04005459 if (early) {
Tejun Heo9395a452014-05-14 09:15:02 -04005460 /* allocation can't be done safely during early init */
Tejun Heo15a4c832014-05-04 15:09:14 -04005461 css->id = 1;
5462 } else {
5463 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5464 BUG_ON(css->id < 0);
5465 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07005466
Li Zefane8d55fd2008-04-29 01:00:13 -07005467 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07005468 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07005469 * newly registered, all tasks and hence the
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005470 * init_css_set is in the subsystem's root cgroup. */
Tejun Heoaec25022014-02-08 10:36:58 -05005471 init_css_set.subsys[ss->id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005472
Aleksa Saraicb4a3162015-06-06 10:02:14 +10005473 have_fork_callback |= (bool)ss->fork << ss->id;
5474 have_exit_callback |= (bool)ss->exit << ss->id;
Tejun Heoafcf6c82015-10-15 16:41:53 -04005475 have_free_callback |= (bool)ss->free << ss->id;
Aleksa Sarai7e476822015-06-09 21:32:09 +10005476 have_canfork_callback |= (bool)ss->can_fork << ss->id;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005477
Li Zefane8d55fd2008-04-29 01:00:13 -07005478 /* At system boot, before all subsystems have been
5479 * registered, no tasks have been forked, so we don't
5480 * need to invoke fork callbacks here. */
5481 BUG_ON(!list_empty(&init_task.tasks));
5482
Tejun Heoae7f1642013-08-13 20:22:50 -04005483 BUG_ON(online_css(css));
Tejun Heoa8638032012-11-09 09:12:29 -08005484
Tejun Heo648bb562012-11-19 08:13:36 -08005485 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005486}
5487
5488/**
Li Zefana043e3b2008-02-23 15:24:09 -08005489 * cgroup_init_early - cgroup initialization at system boot
5490 *
5491 * Initialize cgroups at system boot, and initialize any
5492 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07005493 */
5494int __init cgroup_init_early(void)
5495{
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04005496 static struct cgroup_sb_opts __initdata opts;
Tejun Heo30159ec2013-06-25 11:53:37 -07005497 struct cgroup_subsys *ss;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005498 int i;
Tejun Heo30159ec2013-06-25 11:53:37 -07005499
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005500 init_cgroup_root(&cgrp_dfl_root, &opts);
Tejun Heo3b514d22014-05-16 13:22:47 -04005501 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5502
Tejun Heoa4ea1cc2013-06-21 15:52:33 -07005503 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07005504
Tejun Heo3ed80a62014-02-08 10:36:58 -05005505 for_each_subsys(ss, i) {
Tejun Heoaec25022014-02-08 10:36:58 -05005506 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
Xiubo Li63253ad2016-02-26 13:07:38 +08005507 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
Tejun Heo073219e2014-02-08 10:36:58 -05005508 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
Tejun Heoaec25022014-02-08 10:36:58 -05005509 ss->id, ss->name);
Tejun Heo073219e2014-02-08 10:36:58 -05005510 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5511 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005512
Tejun Heoaec25022014-02-08 10:36:58 -05005513 ss->id = i;
Tejun Heo073219e2014-02-08 10:36:58 -05005514 ss->name = cgroup_subsys_name[i];
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005515 if (!ss->legacy_name)
5516 ss->legacy_name = cgroup_subsys_name[i];
Paul Menageddbcc7e2007-10-18 23:39:30 -07005517
5518 if (ss->early_init)
Tejun Heo15a4c832014-05-04 15:09:14 -04005519 cgroup_init_subsys(ss, true);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005520 }
5521 return 0;
5522}
5523
Tejun Heo6e5c8302016-02-22 22:25:47 -05005524static u16 cgroup_disable_mask __initdata;
Tejun Heoa3e72732015-09-25 16:24:27 -04005525
Paul Menageddbcc7e2007-10-18 23:39:30 -07005526/**
Li Zefana043e3b2008-02-23 15:24:09 -08005527 * cgroup_init - cgroup initialization
5528 *
5529 * Register cgroup filesystem and /proc file, and initialize
5530 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07005531 */
5532int __init cgroup_init(void)
5533{
Tejun Heo30159ec2013-06-25 11:53:37 -07005534 struct cgroup_subsys *ss;
Tejun Heo035f4f52015-10-15 17:00:43 -04005535 int ssid;
Paul Menagea4243162007-10-18 23:39:35 -07005536
Tejun Heo6e5c8302016-02-22 22:25:47 -05005537 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
Tejun Heo1ed13282015-09-16 12:53:17 -04005538 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
Tejun Heoa14c6872014-07-15 11:05:09 -04005539 BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
5540 BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005541
Aditya Kalia79a9082016-01-29 02:54:06 -06005542 get_user_ns(init_cgroup_ns.user_ns);
5543
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005544 mutex_lock(&cgroup_mutex);
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005545
Tejun Heo2378d8b2016-03-03 09:57:57 -05005546 /*
5547 * Add init_css_set to the hash table so that dfl_root can link to
5548 * it during init.
5549 */
5550 hash_add(css_set_table, &init_css_set.hlist,
5551 css_set_hash(init_css_set.subsys));
Tejun Heo82fe9b02013-06-25 11:53:37 -07005552
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005553 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
Greg KH676db4a2010-08-05 13:53:35 -07005554
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005555 mutex_unlock(&cgroup_mutex);
5556
Tejun Heo172a2c062014-03-19 10:23:53 -04005557 for_each_subsys(ss, ssid) {
Tejun Heo15a4c832014-05-04 15:09:14 -04005558 if (ss->early_init) {
5559 struct cgroup_subsys_state *css =
5560 init_css_set.subsys[ss->id];
5561
5562 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5563 GFP_KERNEL);
5564 BUG_ON(css->id < 0);
5565 } else {
5566 cgroup_init_subsys(ss, false);
5567 }
Tejun Heo172a2c062014-03-19 10:23:53 -04005568
Tejun Heo2d8f2432014-04-23 11:13:15 -04005569 list_add_tail(&init_css_set.e_cset_node[ssid],
5570 &cgrp_dfl_root.cgrp.e_csets[ssid]);
Tejun Heo172a2c062014-03-19 10:23:53 -04005571
5572 /*
Li Zefanc731ae12014-06-05 17:16:30 +08005573 * Setting dfl_root subsys_mask needs to consider the
5574 * disabled flag and cftype registration needs kmalloc,
5575 * both of which aren't available during early_init.
Tejun Heo172a2c062014-03-19 10:23:53 -04005576 */
Tejun Heoa3e72732015-09-25 16:24:27 -04005577 if (cgroup_disable_mask & (1 << ssid)) {
5578 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5579 printk(KERN_INFO "Disabling %s control group subsystem\n",
5580 ss->name);
Tejun Heoa8ddc822014-07-15 11:05:10 -04005581 continue;
Tejun Heoa3e72732015-09-25 16:24:27 -04005582 }
Tejun Heoa8ddc822014-07-15 11:05:10 -04005583
Johannes Weiner223ffb22016-02-11 13:34:49 -05005584 if (cgroup_ssid_no_v1(ssid))
5585 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5586 ss->name);
5587
Tejun Heoa8ddc822014-07-15 11:05:10 -04005588 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5589
Tejun Heof6d635ad2016-03-08 11:51:26 -05005590 if (ss->implicit_on_dfl)
5591 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5592 else if (!ss->dfl_cftypes)
Tejun Heoa7165262016-02-23 10:00:50 -05005593 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
Tejun Heo5de4fa12014-07-15 11:05:10 -04005594
Tejun Heoa8ddc822014-07-15 11:05:10 -04005595 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5596 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5597 } else {
5598 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5599 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
Li Zefanc731ae12014-06-05 17:16:30 +08005600 }
Vladimir Davydov295458e2015-02-19 17:34:46 +03005601
5602 if (ss->bind)
5603 ss->bind(init_css_set.subsys[ssid]);
Tejun Heo172a2c062014-03-19 10:23:53 -04005604 }
Greg KH676db4a2010-08-05 13:53:35 -07005605
Tejun Heo2378d8b2016-03-03 09:57:57 -05005606 /* init_css_set.subsys[] has been updated, re-hash */
5607 hash_del(&init_css_set.hlist);
5608 hash_add(css_set_table, &init_css_set.hlist,
5609 css_set_hash(init_css_set.subsys));
5610
Tejun Heo035f4f52015-10-15 17:00:43 -04005611 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5612 WARN_ON(register_filesystem(&cgroup_fs_type));
Tejun Heo67e9c742015-11-16 11:13:34 -05005613 WARN_ON(register_filesystem(&cgroup2_fs_type));
Tejun Heo035f4f52015-10-15 17:00:43 -04005614 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
Paul Menagea4243162007-10-18 23:39:35 -07005615
Tejun Heo2bd59d42014-02-11 11:52:49 -05005616 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005617}
Paul Menageb4f48b62007-10-18 23:39:33 -07005618
Tejun Heoe5fca242013-11-22 17:14:39 -05005619static int __init cgroup_wq_init(void)
5620{
5621 /*
5622 * There isn't much point in executing destruction path in
5623 * parallel. Good chunk is serialized with cgroup_mutex anyway.
Tejun Heo1a115332014-02-12 19:06:19 -05005624 * Use 1 for @max_active.
Tejun Heoe5fca242013-11-22 17:14:39 -05005625 *
5626 * We would prefer to do this in cgroup_init() above, but that
5627 * is called before init_workqueues(): so leave this until after.
5628 */
Tejun Heo1a115332014-02-12 19:06:19 -05005629 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
Tejun Heoe5fca242013-11-22 17:14:39 -05005630 BUG_ON(!cgroup_destroy_wq);
Tejun Heob1a21362013-11-29 10:42:58 -05005631
5632 /*
5633 * Used to destroy pidlists and separate to serve as flush domain.
5634 * Cap @max_active to 1 too.
5635 */
5636 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5637 0, 1);
5638 BUG_ON(!cgroup_pidlist_destroy_wq);
5639
Tejun Heoe5fca242013-11-22 17:14:39 -05005640 return 0;
5641}
5642core_initcall(cgroup_wq_init);
5643
Paul Menagea4243162007-10-18 23:39:35 -07005644/*
5645 * proc_cgroup_show()
5646 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5647 * - Used for /proc/<pid>/cgroup.
Paul Menagea4243162007-10-18 23:39:35 -07005648 */
Zefan Li006f4ac2014-09-18 16:03:15 +08005649int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5650 struct pid *pid, struct task_struct *tsk)
Paul Menagea4243162007-10-18 23:39:35 -07005651{
Tejun Heoe61734c2014-02-12 09:29:50 -05005652 char *buf, *path;
Paul Menagea4243162007-10-18 23:39:35 -07005653 int retval;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005654 struct cgroup_root *root;
Paul Menagea4243162007-10-18 23:39:35 -07005655
5656 retval = -ENOMEM;
Tejun Heoe61734c2014-02-12 09:29:50 -05005657 buf = kmalloc(PATH_MAX, GFP_KERNEL);
Paul Menagea4243162007-10-18 23:39:35 -07005658 if (!buf)
5659 goto out;
5660
Paul Menagea4243162007-10-18 23:39:35 -07005661 mutex_lock(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005662 spin_lock_bh(&css_set_lock);
Paul Menagea4243162007-10-18 23:39:35 -07005663
Tejun Heo985ed672014-03-19 10:23:53 -04005664 for_each_root(root) {
Paul Menagea4243162007-10-18 23:39:35 -07005665 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07005666 struct cgroup *cgrp;
Tejun Heob85d2042013-12-06 15:11:57 -05005667 int ssid, count = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005668
Tejun Heoa7165262016-02-23 10:00:50 -05005669 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
Tejun Heo985ed672014-03-19 10:23:53 -04005670 continue;
5671
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005672 seq_printf(m, "%d:", root->hierarchy_id);
Tejun Heod98817d2015-08-18 13:58:16 -07005673 if (root != &cgrp_dfl_root)
5674 for_each_subsys(ss, ssid)
5675 if (root->subsys_mask & (1 << ssid))
5676 seq_printf(m, "%s%s", count++ ? "," : "",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005677 ss->legacy_name);
Paul Menagec6d57f332009-09-23 15:56:19 -07005678 if (strlen(root->name))
5679 seq_printf(m, "%sname=%s", count ? "," : "",
5680 root->name);
Paul Menagea4243162007-10-18 23:39:35 -07005681 seq_putc(m, ':');
Tejun Heo2e91fa72015-10-15 16:41:53 -04005682
Paul Menage7717f7b2009-09-23 15:56:22 -07005683 cgrp = task_cgroup_from_root(tsk, root);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005684
5685 /*
5686 * On traditional hierarchies, all zombie tasks show up as
5687 * belonging to the root cgroup. On the default hierarchy,
5688 * while a zombie doesn't show up in "cgroup.procs" and
5689 * thus can't be migrated, its /proc/PID/cgroup keeps
5690 * reporting the cgroup it belonged to before exiting. If
5691 * the cgroup is removed before the zombie is reaped,
5692 * " (deleted)" is appended to the cgroup path.
5693 */
5694 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
Aditya Kalia79a9082016-01-29 02:54:06 -06005695 path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5696 current->nsproxy->cgroup_ns);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005697 if (!path) {
5698 retval = -ENAMETOOLONG;
5699 goto out_unlock;
5700 }
5701 } else {
5702 path = "/";
Tejun Heoe61734c2014-02-12 09:29:50 -05005703 }
Tejun Heo2e91fa72015-10-15 16:41:53 -04005704
Tejun Heoe61734c2014-02-12 09:29:50 -05005705 seq_puts(m, path);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005706
5707 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5708 seq_puts(m, " (deleted)\n");
5709 else
5710 seq_putc(m, '\n');
Paul Menagea4243162007-10-18 23:39:35 -07005711 }
5712
Zefan Li006f4ac2014-09-18 16:03:15 +08005713 retval = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005714out_unlock:
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005715 spin_unlock_bh(&css_set_lock);
Paul Menagea4243162007-10-18 23:39:35 -07005716 mutex_unlock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07005717 kfree(buf);
5718out:
5719 return retval;
5720}
5721
Paul Menagea4243162007-10-18 23:39:35 -07005722/* Display information about each subsystem and each hierarchy */
5723static int proc_cgroupstats_show(struct seq_file *m, void *v)
5724{
Tejun Heo30159ec2013-06-25 11:53:37 -07005725 struct cgroup_subsys *ss;
Paul Menagea4243162007-10-18 23:39:35 -07005726 int i;
Paul Menagea4243162007-10-18 23:39:35 -07005727
Paul Menage8bab8dd2008-04-04 14:29:57 -07005728 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Ben Blumaae8aab2010-03-10 15:22:07 -08005729 /*
5730 * ideally we don't want subsystems moving around while we do this.
5731 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5732 * subsys/hierarchy state.
5733 */
Paul Menagea4243162007-10-18 23:39:35 -07005734 mutex_lock(&cgroup_mutex);
Tejun Heo30159ec2013-06-25 11:53:37 -07005735
5736 for_each_subsys(ss, i)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005737 seq_printf(m, "%s\t%d\t%d\t%d\n",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005738 ss->legacy_name, ss->root->hierarchy_id,
Tejun Heofc5ed1e2015-09-18 11:56:28 -04005739 atomic_read(&ss->root->nr_cgrps),
5740 cgroup_ssid_enabled(i));
Tejun Heo30159ec2013-06-25 11:53:37 -07005741
Paul Menagea4243162007-10-18 23:39:35 -07005742 mutex_unlock(&cgroup_mutex);
5743 return 0;
5744}
5745
5746static int cgroupstats_open(struct inode *inode, struct file *file)
5747{
Al Viro9dce07f2008-03-29 03:07:28 +00005748 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07005749}
5750
Alexey Dobriyan828c0952009-10-01 15:43:56 -07005751static const struct file_operations proc_cgroupstats_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07005752 .open = cgroupstats_open,
5753 .read = seq_read,
5754 .llseek = seq_lseek,
5755 .release = single_release,
5756};
5757
Paul Menageb4f48b62007-10-18 23:39:33 -07005758/**
Tejun Heoeaf797a2014-02-25 10:04:03 -05005759 * cgroup_fork - initialize cgroup related fields during copy_process()
Li Zefana043e3b2008-02-23 15:24:09 -08005760 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07005761 *
Tejun Heoeaf797a2014-02-25 10:04:03 -05005762 * A task is associated with the init_css_set until cgroup_post_fork()
5763 * attaches it to the parent's css_set. Empty cg_list indicates that
5764 * @child isn't holding reference to its css_set.
Paul Menageb4f48b62007-10-18 23:39:33 -07005765 */
5766void cgroup_fork(struct task_struct *child)
5767{
Tejun Heoeaf797a2014-02-25 10:04:03 -05005768 RCU_INIT_POINTER(child->cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07005769 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07005770}
5771
5772/**
Aleksa Sarai7e476822015-06-09 21:32:09 +10005773 * cgroup_can_fork - called on a new task before the process is exposed
5774 * @child: the task in question.
5775 *
5776 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5777 * returns an error, the fork aborts with that error code. This allows for
5778 * a cgroup subsystem to conditionally allow or deny new forks.
5779 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005780int cgroup_can_fork(struct task_struct *child)
Aleksa Sarai7e476822015-06-09 21:32:09 +10005781{
5782 struct cgroup_subsys *ss;
5783 int i, j, ret;
5784
Tejun Heob4e0eea2016-02-22 22:25:46 -05005785 do_each_subsys_mask(ss, i, have_canfork_callback) {
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005786 ret = ss->can_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005787 if (ret)
5788 goto out_revert;
Tejun Heob4e0eea2016-02-22 22:25:46 -05005789 } while_each_subsys_mask();
Aleksa Sarai7e476822015-06-09 21:32:09 +10005790
5791 return 0;
5792
5793out_revert:
5794 for_each_subsys(ss, j) {
5795 if (j >= i)
5796 break;
5797 if (ss->cancel_fork)
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005798 ss->cancel_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005799 }
5800
5801 return ret;
5802}
5803
5804/**
5805 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5806 * @child: the task in question
5807 *
5808 * This calls the cancel_fork() callbacks if a fork failed *after*
5809 * cgroup_can_fork() succeded.
5810 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005811void cgroup_cancel_fork(struct task_struct *child)
Aleksa Sarai7e476822015-06-09 21:32:09 +10005812{
5813 struct cgroup_subsys *ss;
5814 int i;
5815
5816 for_each_subsys(ss, i)
5817 if (ss->cancel_fork)
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005818 ss->cancel_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005819}
5820
5821/**
Li Zefana043e3b2008-02-23 15:24:09 -08005822 * cgroup_post_fork - called on a new task after adding it to the task list
5823 * @child: the task in question
5824 *
Tejun Heo5edee612012-10-16 15:03:14 -07005825 * Adds the task to the list running through its css_set if necessary and
5826 * call the subsystem fork() callbacks. Has to be after the task is
5827 * visible on the task list in case we race with the first call to
Tejun Heo0942eee2013-08-08 20:11:26 -04005828 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
Tejun Heo5edee612012-10-16 15:03:14 -07005829 * list.
Li Zefana043e3b2008-02-23 15:24:09 -08005830 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005831void cgroup_post_fork(struct task_struct *child)
Paul Menage817929e2007-10-18 23:39:36 -07005832{
Tejun Heo30159ec2013-06-25 11:53:37 -07005833 struct cgroup_subsys *ss;
Tejun Heo5edee612012-10-16 15:03:14 -07005834 int i;
5835
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005836 /*
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005837 * This may race against cgroup_enable_task_cg_lists(). As that
Tejun Heoeaf797a2014-02-25 10:04:03 -05005838 * function sets use_task_css_set_links before grabbing
5839 * tasklist_lock and we just went through tasklist_lock to add
5840 * @child, it's guaranteed that either we see the set
5841 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5842 * @child during its iteration.
5843 *
5844 * If we won the race, @child is associated with %current's
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005845 * css_set. Grabbing css_set_lock guarantees both that the
Tejun Heoeaf797a2014-02-25 10:04:03 -05005846 * association is stable, and, on completion of the parent's
5847 * migration, @child is visible in the source of migration or
5848 * already in the destination cgroup. This guarantee is necessary
5849 * when implementing operations which need to migrate all tasks of
5850 * a cgroup to another.
5851 *
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005852 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
Tejun Heoeaf797a2014-02-25 10:04:03 -05005853 * will remain in init_css_set. This is safe because all tasks are
5854 * in the init_css_set before cg_links is enabled and there's no
5855 * operation which transfers all tasks out of init_css_set.
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005856 */
Paul Menage817929e2007-10-18 23:39:36 -07005857 if (use_task_css_set_links) {
Tejun Heoeaf797a2014-02-25 10:04:03 -05005858 struct css_set *cset;
5859
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005860 spin_lock_bh(&css_set_lock);
Tejun Heo0e1d7682014-02-25 10:04:03 -05005861 cset = task_css_set(current);
Tejun Heoeaf797a2014-02-25 10:04:03 -05005862 if (list_empty(&child->cg_list)) {
Tejun Heoeaf797a2014-02-25 10:04:03 -05005863 get_css_set(cset);
Tejun Heof6d7d042015-10-15 16:41:52 -04005864 css_set_move_task(child, NULL, cset, false);
Tejun Heoeaf797a2014-02-25 10:04:03 -05005865 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005866 spin_unlock_bh(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07005867 }
Tejun Heo5edee612012-10-16 15:03:14 -07005868
5869 /*
5870 * Call ss->fork(). This must happen after @child is linked on
5871 * css_set; otherwise, @child might change state between ->fork()
5872 * and addition to css_set.
5873 */
Tejun Heob4e0eea2016-02-22 22:25:46 -05005874 do_each_subsys_mask(ss, i, have_fork_callback) {
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005875 ss->fork(child);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005876 } while_each_subsys_mask();
Paul Menage817929e2007-10-18 23:39:36 -07005877}
Tejun Heo5edee612012-10-16 15:03:14 -07005878
Paul Menage817929e2007-10-18 23:39:36 -07005879/**
Paul Menageb4f48b62007-10-18 23:39:33 -07005880 * cgroup_exit - detach cgroup from exiting task
5881 * @tsk: pointer to task_struct of exiting process
5882 *
5883 * Description: Detach cgroup from @tsk and release it.
5884 *
5885 * Note that cgroups marked notify_on_release force every task in
5886 * them to take the global cgroup_mutex mutex when exiting.
5887 * This could impact scaling on very large systems. Be reluctant to
5888 * use notify_on_release cgroups where very high task exit scaling
5889 * is required on large systems.
5890 *
Tejun Heo0e1d7682014-02-25 10:04:03 -05005891 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5892 * call cgroup_exit() while the task is still competent to handle
5893 * notify_on_release(), then leave the task attached to the root cgroup in
5894 * each hierarchy for the remainder of its exit. No need to bother with
5895 * init_css_set refcnting. init_css_set never goes away and we can't race
Li Zefane8604cb2014-03-28 15:18:27 +08005896 * with migration path - PF_EXITING is visible to migration path.
Paul Menageb4f48b62007-10-18 23:39:33 -07005897 */
Li Zefan1ec41832014-03-28 15:22:19 +08005898void cgroup_exit(struct task_struct *tsk)
Paul Menageb4f48b62007-10-18 23:39:33 -07005899{
Tejun Heo30159ec2013-06-25 11:53:37 -07005900 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -07005901 struct css_set *cset;
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005902 int i;
Paul Menage817929e2007-10-18 23:39:36 -07005903
5904 /*
Tejun Heo0e1d7682014-02-25 10:04:03 -05005905 * Unlink from @tsk from its css_set. As migration path can't race
Tejun Heo0de09422015-10-15 16:41:49 -04005906 * with us, we can check css_set and cg_list without synchronization.
Paul Menage817929e2007-10-18 23:39:36 -07005907 */
Tejun Heo0de09422015-10-15 16:41:49 -04005908 cset = task_css_set(tsk);
5909
Paul Menage817929e2007-10-18 23:39:36 -07005910 if (!list_empty(&tsk->cg_list)) {
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005911 spin_lock_bh(&css_set_lock);
Tejun Heof6d7d042015-10-15 16:41:52 -04005912 css_set_move_task(tsk, cset, NULL, false);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005913 spin_unlock_bh(&css_set_lock);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005914 } else {
5915 get_css_set(cset);
Paul Menage817929e2007-10-18 23:39:36 -07005916 }
5917
Aleksa Saraicb4a3162015-06-06 10:02:14 +10005918 /* see cgroup_post_fork() for details */
Tejun Heob4e0eea2016-02-22 22:25:46 -05005919 do_each_subsys_mask(ss, i, have_exit_callback) {
Tejun Heo2e91fa72015-10-15 16:41:53 -04005920 ss->exit(tsk);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005921 } while_each_subsys_mask();
Tejun Heo2e91fa72015-10-15 16:41:53 -04005922}
Tejun Heo30159ec2013-06-25 11:53:37 -07005923
Tejun Heo2e91fa72015-10-15 16:41:53 -04005924void cgroup_free(struct task_struct *task)
5925{
5926 struct css_set *cset = task_css_set(task);
Tejun Heoafcf6c82015-10-15 16:41:53 -04005927 struct cgroup_subsys *ss;
5928 int ssid;
5929
Tejun Heob4e0eea2016-02-22 22:25:46 -05005930 do_each_subsys_mask(ss, ssid, have_free_callback) {
Tejun Heoafcf6c82015-10-15 16:41:53 -04005931 ss->free(task);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005932 } while_each_subsys_mask();
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005933
Tejun Heo2e91fa72015-10-15 16:41:53 -04005934 put_css_set(cset);
Paul Menageb4f48b62007-10-18 23:39:33 -07005935}
Paul Menage697f4162007-10-18 23:39:34 -07005936
Paul Menagebd89aab2007-10-18 23:40:44 -07005937static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07005938{
Tejun Heo27bd4db2015-10-15 16:41:50 -04005939 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
Zefan Li971ff492014-09-18 16:06:19 +08005940 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
5941 schedule_work(&cgrp->release_agent_work);
Paul Menage81a6a5c2007-10-18 23:39:38 -07005942}
5943
Paul Menage81a6a5c2007-10-18 23:39:38 -07005944/*
5945 * Notify userspace when a cgroup is released, by running the
5946 * configured release agent with the name of the cgroup (path
5947 * relative to the root of cgroup file system) as the argument.
5948 *
5949 * Most likely, this user command will try to rmdir this cgroup.
5950 *
5951 * This races with the possibility that some other task will be
5952 * attached to this cgroup before it is removed, or that some other
5953 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5954 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5955 * unused, and this cgroup will be reprieved from its death sentence,
5956 * to continue to serve a useful existence. Next time it's released,
5957 * we will get notified again, if it still has 'notify_on_release' set.
5958 *
5959 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5960 * means only wait until the task is successfully execve()'d. The
5961 * separate release agent task is forked by call_usermodehelper(),
5962 * then control in this thread returns here, without waiting for the
5963 * release agent task. We don't bother to wait because the caller of
5964 * this routine has no use for the exit status of the release agent
5965 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07005966 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07005967static void cgroup_release_agent(struct work_struct *work)
5968{
Zefan Li971ff492014-09-18 16:06:19 +08005969 struct cgroup *cgrp =
5970 container_of(work, struct cgroup, release_agent_work);
5971 char *pathbuf = NULL, *agentbuf = NULL, *path;
5972 char *argv[3], *envp[3];
5973
Paul Menage81a6a5c2007-10-18 23:39:38 -07005974 mutex_lock(&cgroup_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07005975
Zefan Li971ff492014-09-18 16:06:19 +08005976 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
5977 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5978 if (!pathbuf || !agentbuf)
5979 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07005980
Aditya Kalia79a9082016-01-29 02:54:06 -06005981 spin_lock_bh(&css_set_lock);
5982 path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
5983 spin_unlock_bh(&css_set_lock);
Zefan Li971ff492014-09-18 16:06:19 +08005984 if (!path)
5985 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07005986
Zefan Li971ff492014-09-18 16:06:19 +08005987 argv[0] = agentbuf;
5988 argv[1] = path;
5989 argv[2] = NULL;
5990
5991 /* minimal command environment */
5992 envp[0] = "HOME=/";
5993 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5994 envp[2] = NULL;
5995
Paul Menage81a6a5c2007-10-18 23:39:38 -07005996 mutex_unlock(&cgroup_mutex);
Zefan Li971ff492014-09-18 16:06:19 +08005997 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
Zefan Li3e2cd912014-09-20 14:35:43 +08005998 goto out_free;
Zefan Li971ff492014-09-18 16:06:19 +08005999out:
Zefan Li3e2cd912014-09-20 14:35:43 +08006000 mutex_unlock(&cgroup_mutex);
6001out_free:
Zefan Li971ff492014-09-18 16:06:19 +08006002 kfree(agentbuf);
6003 kfree(pathbuf);
Paul Menage81a6a5c2007-10-18 23:39:38 -07006004}
Paul Menage8bab8dd2008-04-04 14:29:57 -07006005
6006static int __init cgroup_disable(char *str)
6007{
Tejun Heo30159ec2013-06-25 11:53:37 -07006008 struct cgroup_subsys *ss;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006009 char *token;
Tejun Heo30159ec2013-06-25 11:53:37 -07006010 int i;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006011
6012 while ((token = strsep(&str, ",")) != NULL) {
6013 if (!*token)
6014 continue;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006015
Tejun Heo3ed80a62014-02-08 10:36:58 -05006016 for_each_subsys(ss, i) {
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07006017 if (strcmp(token, ss->name) &&
6018 strcmp(token, ss->legacy_name))
6019 continue;
Tejun Heoa3e72732015-09-25 16:24:27 -04006020 cgroup_disable_mask |= 1 << i;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006021 }
6022 }
6023 return 1;
6024}
6025__setup("cgroup_disable=", cgroup_disable);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07006026
Johannes Weiner223ffb22016-02-11 13:34:49 -05006027static int __init cgroup_no_v1(char *str)
6028{
6029 struct cgroup_subsys *ss;
6030 char *token;
6031 int i;
6032
6033 while ((token = strsep(&str, ",")) != NULL) {
6034 if (!*token)
6035 continue;
6036
6037 if (!strcmp(token, "all")) {
Tejun Heo6e5c8302016-02-22 22:25:47 -05006038 cgroup_no_v1_mask = U16_MAX;
Johannes Weiner223ffb22016-02-11 13:34:49 -05006039 break;
6040 }
6041
6042 for_each_subsys(ss, i) {
6043 if (strcmp(token, ss->name) &&
6044 strcmp(token, ss->legacy_name))
6045 continue;
6046
6047 cgroup_no_v1_mask |= 1 << i;
6048 }
6049 }
6050 return 1;
6051}
6052__setup("cgroup_no_v1=", cgroup_no_v1);
6053
Tejun Heob77d7b62013-08-13 11:01:54 -04006054/**
Tejun Heoec903c02014-05-13 12:11:01 -04006055 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
Tejun Heo35cf0832013-08-26 18:40:56 -04006056 * @dentry: directory dentry of interest
6057 * @ss: subsystem of interest
Tejun Heob77d7b62013-08-13 11:01:54 -04006058 *
Tejun Heo5a17f542014-02-11 11:52:47 -05006059 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6060 * to get the corresponding css and return it. If such css doesn't exist
6061 * or can't be pinned, an ERR_PTR value is returned.
Stephane Eraniane5d13672011-02-14 11:20:01 +02006062 */
Tejun Heoec903c02014-05-13 12:11:01 -04006063struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6064 struct cgroup_subsys *ss)
Stephane Eraniane5d13672011-02-14 11:20:01 +02006065{
Tejun Heo2bd59d42014-02-11 11:52:49 -05006066 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
Tejun Heof17fc252016-02-23 10:00:51 -05006067 struct file_system_type *s_type = dentry->d_sb->s_type;
Tejun Heo2bd59d42014-02-11 11:52:49 -05006068 struct cgroup_subsys_state *css = NULL;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006069 struct cgroup *cgrp;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006070
Tejun Heo35cf0832013-08-26 18:40:56 -04006071 /* is @dentry a cgroup dir? */
Tejun Heof17fc252016-02-23 10:00:51 -05006072 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6073 !kn || kernfs_type(kn) != KERNFS_DIR)
Stephane Eraniane5d13672011-02-14 11:20:01 +02006074 return ERR_PTR(-EBADF);
6075
Tejun Heo5a17f542014-02-11 11:52:47 -05006076 rcu_read_lock();
6077
Tejun Heo2bd59d42014-02-11 11:52:49 -05006078 /*
6079 * This path doesn't originate from kernfs and @kn could already
6080 * have been or be removed at any point. @kn->priv is RCU
Li Zefana4189482014-09-04 14:43:07 +08006081 * protected for this access. See css_release_work_fn() for details.
Tejun Heo2bd59d42014-02-11 11:52:49 -05006082 */
6083 cgrp = rcu_dereference(kn->priv);
6084 if (cgrp)
6085 css = cgroup_css(cgrp, ss);
Tejun Heo5a17f542014-02-11 11:52:47 -05006086
Tejun Heoec903c02014-05-13 12:11:01 -04006087 if (!css || !css_tryget_online(css))
Tejun Heo5a17f542014-02-11 11:52:47 -05006088 css = ERR_PTR(-ENOENT);
6089
6090 rcu_read_unlock();
6091 return css;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006092}
Stephane Eraniane5d13672011-02-14 11:20:01 +02006093
Li Zefan1cb650b2013-08-19 10:05:24 +08006094/**
6095 * css_from_id - lookup css by id
6096 * @id: the cgroup id
6097 * @ss: cgroup subsys to be looked into
6098 *
6099 * Returns the css if there's valid one with @id, otherwise returns NULL.
6100 * Should be called under rcu_read_lock().
6101 */
6102struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6103{
Tejun Heo6fa49182014-05-04 15:09:13 -04006104 WARN_ON_ONCE(!rcu_read_lock_held());
Vladimir Davydovadbe4272015-04-15 16:13:00 -07006105 return id > 0 ? idr_find(&ss->css_idr, id) : NULL;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006106}
6107
Tejun Heo16af4392015-11-20 15:55:52 -05006108/**
6109 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6110 * @path: path on the default hierarchy
6111 *
6112 * Find the cgroup at @path on the default hierarchy, increment its
6113 * reference count and return it. Returns pointer to the found cgroup on
6114 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6115 * if @path points to a non-directory.
6116 */
6117struct cgroup *cgroup_get_from_path(const char *path)
6118{
6119 struct kernfs_node *kn;
6120 struct cgroup *cgrp;
6121
6122 mutex_lock(&cgroup_mutex);
6123
6124 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6125 if (kn) {
6126 if (kernfs_type(kn) == KERNFS_DIR) {
6127 cgrp = kn->priv;
6128 cgroup_get(cgrp);
6129 } else {
6130 cgrp = ERR_PTR(-ENOTDIR);
6131 }
6132 kernfs_put(kn);
6133 } else {
6134 cgrp = ERR_PTR(-ENOENT);
6135 }
6136
6137 mutex_unlock(&cgroup_mutex);
6138 return cgrp;
6139}
6140EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6141
Tejun Heobd1060a2015-12-07 17:38:53 -05006142/*
6143 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6144 * definition in cgroup-defs.h.
6145 */
6146#ifdef CONFIG_SOCK_CGROUP_DATA
6147
6148#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6149
Tejun Heo3fa4cc92015-12-14 11:24:06 -05006150DEFINE_SPINLOCK(cgroup_sk_update_lock);
Tejun Heobd1060a2015-12-07 17:38:53 -05006151static bool cgroup_sk_alloc_disabled __read_mostly;
6152
6153void cgroup_sk_alloc_disable(void)
6154{
6155 if (cgroup_sk_alloc_disabled)
6156 return;
6157 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6158 cgroup_sk_alloc_disabled = true;
6159}
6160
6161#else
6162
6163#define cgroup_sk_alloc_disabled false
6164
6165#endif
6166
6167void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6168{
6169 if (cgroup_sk_alloc_disabled)
6170 return;
6171
6172 rcu_read_lock();
6173
6174 while (true) {
6175 struct css_set *cset;
6176
6177 cset = task_css_set(current);
6178 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6179 skcd->val = (unsigned long)cset->dfl_cgrp;
6180 break;
6181 }
6182 cpu_relax();
6183 }
6184
6185 rcu_read_unlock();
6186}
6187
6188void cgroup_sk_free(struct sock_cgroup_data *skcd)
6189{
6190 cgroup_put(sock_cgroup_ptr(skcd));
6191}
6192
6193#endif /* CONFIG_SOCK_CGROUP_DATA */
6194
Aditya Kalia79a9082016-01-29 02:54:06 -06006195/* cgroup namespaces */
6196
6197static struct cgroup_namespace *alloc_cgroup_ns(void)
6198{
6199 struct cgroup_namespace *new_ns;
6200 int ret;
6201
6202 new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL);
6203 if (!new_ns)
6204 return ERR_PTR(-ENOMEM);
6205 ret = ns_alloc_inum(&new_ns->ns);
6206 if (ret) {
6207 kfree(new_ns);
6208 return ERR_PTR(ret);
6209 }
6210 atomic_set(&new_ns->count, 1);
6211 new_ns->ns.ops = &cgroupns_operations;
6212 return new_ns;
6213}
6214
6215void free_cgroup_ns(struct cgroup_namespace *ns)
6216{
6217 put_css_set(ns->root_cset);
6218 put_user_ns(ns->user_ns);
6219 ns_free_inum(&ns->ns);
6220 kfree(ns);
6221}
6222EXPORT_SYMBOL(free_cgroup_ns);
6223
6224struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
6225 struct user_namespace *user_ns,
6226 struct cgroup_namespace *old_ns)
6227{
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006228 struct cgroup_namespace *new_ns;
6229 struct css_set *cset;
Aditya Kalia79a9082016-01-29 02:54:06 -06006230
6231 BUG_ON(!old_ns);
6232
6233 if (!(flags & CLONE_NEWCGROUP)) {
6234 get_cgroup_ns(old_ns);
6235 return old_ns;
6236 }
6237
6238 /* Allow only sysadmin to create cgroup namespace. */
Aditya Kalia79a9082016-01-29 02:54:06 -06006239 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006240 return ERR_PTR(-EPERM);
Aditya Kalia79a9082016-01-29 02:54:06 -06006241
6242 mutex_lock(&cgroup_mutex);
6243 spin_lock_bh(&css_set_lock);
6244
6245 cset = task_css_set(current);
6246 get_css_set(cset);
6247
6248 spin_unlock_bh(&css_set_lock);
6249 mutex_unlock(&cgroup_mutex);
6250
Aditya Kalia79a9082016-01-29 02:54:06 -06006251 new_ns = alloc_cgroup_ns();
Tejun Heod2202552016-02-18 11:44:24 -05006252 if (IS_ERR(new_ns)) {
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006253 put_css_set(cset);
6254 return new_ns;
Tejun Heod2202552016-02-18 11:44:24 -05006255 }
Aditya Kalia79a9082016-01-29 02:54:06 -06006256
6257 new_ns->user_ns = get_user_ns(user_ns);
6258 new_ns->root_cset = cset;
6259
6260 return new_ns;
Aditya Kalia79a9082016-01-29 02:54:06 -06006261}
6262
6263static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns)
6264{
6265 return container_of(ns, struct cgroup_namespace, ns);
6266}
6267
Aditya Kalia0530e02016-01-29 02:54:07 -06006268static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Aditya Kalia79a9082016-01-29 02:54:06 -06006269{
Aditya Kalia0530e02016-01-29 02:54:07 -06006270 struct cgroup_namespace *cgroup_ns = to_cg_ns(ns);
6271
6272 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) ||
6273 !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN))
6274 return -EPERM;
6275
6276 /* Don't need to do anything if we are attaching to our own cgroupns. */
6277 if (cgroup_ns == nsproxy->cgroup_ns)
6278 return 0;
6279
6280 get_cgroup_ns(cgroup_ns);
6281 put_cgroup_ns(nsproxy->cgroup_ns);
6282 nsproxy->cgroup_ns = cgroup_ns;
6283
6284 return 0;
Aditya Kalia79a9082016-01-29 02:54:06 -06006285}
6286
6287static struct ns_common *cgroupns_get(struct task_struct *task)
6288{
6289 struct cgroup_namespace *ns = NULL;
6290 struct nsproxy *nsproxy;
6291
6292 task_lock(task);
6293 nsproxy = task->nsproxy;
6294 if (nsproxy) {
6295 ns = nsproxy->cgroup_ns;
6296 get_cgroup_ns(ns);
6297 }
6298 task_unlock(task);
6299
6300 return ns ? &ns->ns : NULL;
6301}
6302
6303static void cgroupns_put(struct ns_common *ns)
6304{
6305 put_cgroup_ns(to_cg_ns(ns));
6306}
6307
6308const struct proc_ns_operations cgroupns_operations = {
6309 .name = "cgroup",
6310 .type = CLONE_NEWCGROUP,
6311 .get = cgroupns_get,
6312 .put = cgroupns_put,
6313 .install = cgroupns_install,
6314};
6315
6316static __init int cgroup_namespaces_init(void)
6317{
6318 return 0;
6319}
6320subsys_initcall(cgroup_namespaces_init);
6321
Paul Menagefe693432009-09-23 15:56:20 -07006322#ifdef CONFIG_CGROUP_DEBUG
Tejun Heoeb954192013-08-08 20:11:23 -04006323static struct cgroup_subsys_state *
6324debug_css_alloc(struct cgroup_subsys_state *parent_css)
Paul Menagefe693432009-09-23 15:56:20 -07006325{
6326 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
6327
6328 if (!css)
6329 return ERR_PTR(-ENOMEM);
6330
6331 return css;
6332}
6333
Tejun Heoeb954192013-08-08 20:11:23 -04006334static void debug_css_free(struct cgroup_subsys_state *css)
Paul Menagefe693432009-09-23 15:56:20 -07006335{
Tejun Heoeb954192013-08-08 20:11:23 -04006336 kfree(css);
Paul Menagefe693432009-09-23 15:56:20 -07006337}
6338
Tejun Heo182446d2013-08-08 20:11:24 -04006339static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
6340 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006341{
Tejun Heo182446d2013-08-08 20:11:24 -04006342 return cgroup_task_count(css->cgroup);
Paul Menagefe693432009-09-23 15:56:20 -07006343}
6344
Tejun Heo182446d2013-08-08 20:11:24 -04006345static u64 current_css_set_read(struct cgroup_subsys_state *css,
6346 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006347{
6348 return (u64)(unsigned long)current->cgroups;
6349}
6350
Tejun Heo182446d2013-08-08 20:11:24 -04006351static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
Li Zefan03c78cb2013-06-14 11:17:19 +08006352 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006353{
6354 u64 count;
6355
6356 rcu_read_lock();
Tejun Heoa8ad8052013-06-21 15:52:04 -07006357 count = atomic_read(&task_css_set(current)->refcount);
Paul Menagefe693432009-09-23 15:56:20 -07006358 rcu_read_unlock();
6359 return count;
6360}
6361
Tejun Heo2da8ca82013-12-05 12:28:04 -05006362static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07006363{
Tejun Heo69d02062013-06-12 21:04:50 -07006364 struct cgrp_cset_link *link;
Tejun Heo5abb8852013-06-12 21:04:49 -07006365 struct css_set *cset;
Tejun Heoe61734c2014-02-12 09:29:50 -05006366 char *name_buf;
Paul Menage7717f7b2009-09-23 15:56:22 -07006367
Tejun Heoe61734c2014-02-12 09:29:50 -05006368 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
6369 if (!name_buf)
6370 return -ENOMEM;
Paul Menage7717f7b2009-09-23 15:56:22 -07006371
Tejun Heof0d9a5f2015-10-15 16:41:53 -04006372 spin_lock_bh(&css_set_lock);
Paul Menage7717f7b2009-09-23 15:56:22 -07006373 rcu_read_lock();
Tejun Heo5abb8852013-06-12 21:04:49 -07006374 cset = rcu_dereference(current->cgroups);
Tejun Heo69d02062013-06-12 21:04:50 -07006375 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07006376 struct cgroup *c = link->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -07006377
Tejun Heoa2dd4242014-03-19 10:23:55 -04006378 cgroup_name(c, name_buf, NAME_MAX + 1);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07006379 seq_printf(seq, "Root %d group %s\n",
Tejun Heoa2dd4242014-03-19 10:23:55 -04006380 c->root->hierarchy_id, name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07006381 }
6382 rcu_read_unlock();
Tejun Heof0d9a5f2015-10-15 16:41:53 -04006383 spin_unlock_bh(&css_set_lock);
Tejun Heoe61734c2014-02-12 09:29:50 -05006384 kfree(name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07006385 return 0;
6386}
6387
6388#define MAX_TASKS_SHOWN_PER_CSS 25
Tejun Heo2da8ca82013-12-05 12:28:04 -05006389static int cgroup_css_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07006390{
Tejun Heo2da8ca82013-12-05 12:28:04 -05006391 struct cgroup_subsys_state *css = seq_css(seq);
Tejun Heo69d02062013-06-12 21:04:50 -07006392 struct cgrp_cset_link *link;
Paul Menage7717f7b2009-09-23 15:56:22 -07006393
Tejun Heof0d9a5f2015-10-15 16:41:53 -04006394 spin_lock_bh(&css_set_lock);
Tejun Heo182446d2013-08-08 20:11:24 -04006395 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
Tejun Heo69d02062013-06-12 21:04:50 -07006396 struct css_set *cset = link->cset;
Paul Menage7717f7b2009-09-23 15:56:22 -07006397 struct task_struct *task;
6398 int count = 0;
Tejun Heoc7561122014-02-25 10:04:01 -05006399
Tejun Heo5abb8852013-06-12 21:04:49 -07006400 seq_printf(seq, "css_set %p\n", cset);
Tejun Heoc7561122014-02-25 10:04:01 -05006401
Tejun Heo5abb8852013-06-12 21:04:49 -07006402 list_for_each_entry(task, &cset->tasks, cg_list) {
Tejun Heoc7561122014-02-25 10:04:01 -05006403 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6404 goto overflow;
6405 seq_printf(seq, " task %d\n", task_pid_vnr(task));
Paul Menage7717f7b2009-09-23 15:56:22 -07006406 }
Tejun Heoc7561122014-02-25 10:04:01 -05006407
6408 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
6409 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6410 goto overflow;
6411 seq_printf(seq, " task %d\n", task_pid_vnr(task));
6412 }
6413 continue;
6414 overflow:
6415 seq_puts(seq, " ...\n");
Paul Menage7717f7b2009-09-23 15:56:22 -07006416 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04006417 spin_unlock_bh(&css_set_lock);
Paul Menage7717f7b2009-09-23 15:56:22 -07006418 return 0;
6419}
6420
Tejun Heo182446d2013-08-08 20:11:24 -04006421static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006422{
Tejun Heo27bd4db2015-10-15 16:41:50 -04006423 return (!cgroup_is_populated(css->cgroup) &&
Zefan Lia25eb522014-09-19 16:51:00 +08006424 !css_has_online_children(&css->cgroup->self));
Paul Menagefe693432009-09-23 15:56:20 -07006425}
6426
6427static struct cftype debug_files[] = {
6428 {
Paul Menagefe693432009-09-23 15:56:20 -07006429 .name = "taskcount",
6430 .read_u64 = debug_taskcount_read,
6431 },
6432
6433 {
6434 .name = "current_css_set",
6435 .read_u64 = current_css_set_read,
6436 },
6437
6438 {
6439 .name = "current_css_set_refcount",
6440 .read_u64 = current_css_set_refcount_read,
6441 },
6442
6443 {
Paul Menage7717f7b2009-09-23 15:56:22 -07006444 .name = "current_css_set_cg_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006445 .seq_show = current_css_set_cg_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07006446 },
6447
6448 {
6449 .name = "cgroup_css_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006450 .seq_show = cgroup_css_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07006451 },
6452
6453 {
Paul Menagefe693432009-09-23 15:56:20 -07006454 .name = "releasable",
6455 .read_u64 = releasable_read,
6456 },
Paul Menagefe693432009-09-23 15:56:20 -07006457
Tejun Heo4baf6e32012-04-01 12:09:55 -07006458 { } /* terminate */
6459};
Paul Menagefe693432009-09-23 15:56:20 -07006460
Tejun Heo073219e2014-02-08 10:36:58 -05006461struct cgroup_subsys debug_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006462 .css_alloc = debug_css_alloc,
6463 .css_free = debug_css_free,
Tejun Heo55779642014-07-15 11:05:09 -04006464 .legacy_cftypes = debug_files,
Paul Menagefe693432009-09-23 15:56:20 -07006465};
6466#endif /* CONFIG_CGROUP_DEBUG */