blob: d1c51b7f5221fbfbb85ed3d12e4918a970713829 [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>
Martin KaFai Lau1f3fe7e2016-06-30 10:28:42 -070064#include <linux/file.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 Heo5549c492013-06-24 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{
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -0300840 unsigned long flags;
841
Tejun Heo89c55092014-02-13 06:58:40 -0500842 /*
843 * Ensure that the refcount doesn't hit zero while any readers
844 * can see it. Similar to atomic_dec_and_lock(), but for an
845 * rwlock
846 */
847 if (atomic_add_unless(&cset->refcount, -1, 1))
848 return;
849
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -0300850 spin_lock_irqsave(&css_set_lock, flags);
Zefan Lia25eb522014-09-19 16:51:00 +0800851 put_css_set_locked(cset);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -0300852 spin_unlock_irqrestore(&css_set_lock, flags);
Tejun Heo89c55092014-02-13 06:58:40 -0500853}
854
Paul Menage817929e2007-10-18 23:39:36 -0700855/*
856 * refcounted get/put for css_set objects
857 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700858static inline void get_css_set(struct css_set *cset)
Paul Menage817929e2007-10-18 23:39:36 -0700859{
Tejun Heo5abb8852013-06-12 21:04:49 -0700860 atomic_inc(&cset->refcount);
Paul Menage817929e2007-10-18 23:39:36 -0700861}
862
Tejun Heob326f9d2013-06-24 15:21:48 -0700863/**
Paul Menage7717f7b2009-09-23 15:56:22 -0700864 * compare_css_sets - helper function for find_existing_css_set().
Tejun Heo5abb8852013-06-12 21:04:49 -0700865 * @cset: candidate css_set being tested
866 * @old_cset: existing css_set for a task
Paul Menage7717f7b2009-09-23 15:56:22 -0700867 * @new_cgrp: cgroup that's being entered by the task
868 * @template: desired set of css pointers in css_set (pre-calculated)
869 *
Li Zefan6f4b7e62013-07-31 16:18:36 +0800870 * Returns true if "cset" matches "old_cset" except for the hierarchy
Paul Menage7717f7b2009-09-23 15:56:22 -0700871 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
872 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700873static bool compare_css_sets(struct css_set *cset,
874 struct css_set *old_cset,
Paul Menage7717f7b2009-09-23 15:56:22 -0700875 struct cgroup *new_cgrp,
876 struct cgroup_subsys_state *template[])
877{
878 struct list_head *l1, *l2;
879
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400880 /*
881 * On the default hierarchy, there can be csets which are
882 * associated with the same set of cgroups but different csses.
883 * Let's first ensure that csses match.
884 */
885 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
Paul Menage7717f7b2009-09-23 15:56:22 -0700886 return false;
Paul Menage7717f7b2009-09-23 15:56:22 -0700887
888 /*
889 * Compare cgroup pointers in order to distinguish between
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400890 * different cgroups in hierarchies. As different cgroups may
891 * share the same effective css, this comparison is always
892 * necessary.
Paul Menage7717f7b2009-09-23 15:56:22 -0700893 */
Tejun Heo69d02062013-06-12 21:04:50 -0700894 l1 = &cset->cgrp_links;
895 l2 = &old_cset->cgrp_links;
Paul Menage7717f7b2009-09-23 15:56:22 -0700896 while (1) {
Tejun Heo69d02062013-06-12 21:04:50 -0700897 struct cgrp_cset_link *link1, *link2;
Tejun Heo5abb8852013-06-12 21:04:49 -0700898 struct cgroup *cgrp1, *cgrp2;
Paul Menage7717f7b2009-09-23 15:56:22 -0700899
900 l1 = l1->next;
901 l2 = l2->next;
902 /* See if we reached the end - both lists are equal length. */
Tejun Heo69d02062013-06-12 21:04:50 -0700903 if (l1 == &cset->cgrp_links) {
904 BUG_ON(l2 != &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700905 break;
906 } else {
Tejun Heo69d02062013-06-12 21:04:50 -0700907 BUG_ON(l2 == &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700908 }
909 /* Locate the cgroups associated with these links. */
Tejun Heo69d02062013-06-12 21:04:50 -0700910 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
911 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
912 cgrp1 = link1->cgrp;
913 cgrp2 = link2->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -0700914 /* Hierarchies should be linked in the same order. */
Tejun Heo5abb8852013-06-12 21:04:49 -0700915 BUG_ON(cgrp1->root != cgrp2->root);
Paul Menage7717f7b2009-09-23 15:56:22 -0700916
917 /*
918 * If this hierarchy is the hierarchy of the cgroup
919 * that's changing, then we need to check that this
920 * css_set points to the new cgroup; if it's any other
921 * hierarchy, then this css_set should point to the
922 * same cgroup as the old css_set.
923 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700924 if (cgrp1->root == new_cgrp->root) {
925 if (cgrp1 != new_cgrp)
Paul Menage7717f7b2009-09-23 15:56:22 -0700926 return false;
927 } else {
Tejun Heo5abb8852013-06-12 21:04:49 -0700928 if (cgrp1 != cgrp2)
Paul Menage7717f7b2009-09-23 15:56:22 -0700929 return false;
930 }
931 }
932 return true;
933}
934
Tejun Heob326f9d2013-06-24 15:21:48 -0700935/**
936 * find_existing_css_set - init css array and find the matching css_set
937 * @old_cset: the css_set that we're using before the cgroup transition
938 * @cgrp: the cgroup that we're moving into
939 * @template: out param for the new set of csses, should be clear on entry
Paul Menage817929e2007-10-18 23:39:36 -0700940 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700941static struct css_set *find_existing_css_set(struct css_set *old_cset,
942 struct cgroup *cgrp,
943 struct cgroup_subsys_state *template[])
Paul Menage817929e2007-10-18 23:39:36 -0700944{
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400945 struct cgroup_root *root = cgrp->root;
Tejun Heo30159ec2013-06-25 11:53:37 -0700946 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -0700947 struct css_set *cset;
Li Zefan0ac801f2013-01-10 11:49:27 +0800948 unsigned long key;
Tejun Heob326f9d2013-06-24 15:21:48 -0700949 int i;
Paul Menage817929e2007-10-18 23:39:36 -0700950
Ben Blumaae8aab2010-03-10 15:22:07 -0800951 /*
952 * Build the set of subsystem state objects that we want to see in the
953 * new css_set. while subsystems can change globally, the entries here
954 * won't change, so no need for locking.
955 */
Tejun Heo30159ec2013-06-25 11:53:37 -0700956 for_each_subsys(ss, i) {
Tejun Heof392e512014-04-23 11:13:14 -0400957 if (root->subsys_mask & (1UL << i)) {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400958 /*
959 * @ss is in this hierarchy, so we want the
960 * effective css from @cgrp.
961 */
962 template[i] = cgroup_e_css(cgrp, ss);
Paul Menage817929e2007-10-18 23:39:36 -0700963 } else {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400964 /*
965 * @ss is not in this hierarchy, so we don't want
966 * to change the css.
967 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700968 template[i] = old_cset->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700969 }
970 }
971
Li Zefan0ac801f2013-01-10 11:49:27 +0800972 key = css_set_hash(template);
Tejun Heo5abb8852013-06-12 21:04:49 -0700973 hash_for_each_possible(css_set_table, cset, hlist, key) {
974 if (!compare_css_sets(cset, old_cset, cgrp, template))
Paul Menage7717f7b2009-09-23 15:56:22 -0700975 continue;
976
977 /* This css_set matches what we need */
Tejun Heo5abb8852013-06-12 21:04:49 -0700978 return cset;
Li Zefan472b10532008-04-29 01:00:11 -0700979 }
Paul Menage817929e2007-10-18 23:39:36 -0700980
981 /* No existing cgroup group matched */
982 return NULL;
983}
984
Tejun Heo69d02062013-06-12 21:04:50 -0700985static void free_cgrp_cset_links(struct list_head *links_to_free)
Paul Menage817929e2007-10-18 23:39:36 -0700986{
Tejun Heo69d02062013-06-12 21:04:50 -0700987 struct cgrp_cset_link *link, *tmp_link;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700988
Tejun Heo69d02062013-06-12 21:04:50 -0700989 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
990 list_del(&link->cset_link);
Paul Menage817929e2007-10-18 23:39:36 -0700991 kfree(link);
992 }
993}
994
Tejun Heo69d02062013-06-12 21:04:50 -0700995/**
996 * allocate_cgrp_cset_links - allocate cgrp_cset_links
997 * @count: the number of links to allocate
998 * @tmp_links: list_head the allocated links are put on
999 *
1000 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1001 * through ->cset_link. Returns 0 on success or -errno.
Li Zefan36553432008-07-29 22:33:19 -07001002 */
Tejun Heo69d02062013-06-12 21:04:50 -07001003static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
Li Zefan36553432008-07-29 22:33:19 -07001004{
Tejun Heo69d02062013-06-12 21:04:50 -07001005 struct cgrp_cset_link *link;
Li Zefan36553432008-07-29 22:33:19 -07001006 int i;
Tejun Heo69d02062013-06-12 21:04:50 -07001007
1008 INIT_LIST_HEAD(tmp_links);
1009
Li Zefan36553432008-07-29 22:33:19 -07001010 for (i = 0; i < count; i++) {
Tejun Heof4f4be22013-06-12 21:04:51 -07001011 link = kzalloc(sizeof(*link), GFP_KERNEL);
Li Zefan36553432008-07-29 22:33:19 -07001012 if (!link) {
Tejun Heo69d02062013-06-12 21:04:50 -07001013 free_cgrp_cset_links(tmp_links);
Li Zefan36553432008-07-29 22:33:19 -07001014 return -ENOMEM;
1015 }
Tejun Heo69d02062013-06-12 21:04:50 -07001016 list_add(&link->cset_link, tmp_links);
Li Zefan36553432008-07-29 22:33:19 -07001017 }
1018 return 0;
1019}
1020
Li Zefanc12f65d2009-01-07 18:07:42 -08001021/**
1022 * link_css_set - a helper function to link a css_set to a cgroup
Tejun Heo69d02062013-06-12 21:04:50 -07001023 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
Tejun Heo5abb8852013-06-12 21:04:49 -07001024 * @cset: the css_set to be linked
Li Zefanc12f65d2009-01-07 18:07:42 -08001025 * @cgrp: the destination cgroup
1026 */
Tejun Heo69d02062013-06-12 21:04:50 -07001027static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1028 struct cgroup *cgrp)
Li Zefanc12f65d2009-01-07 18:07:42 -08001029{
Tejun Heo69d02062013-06-12 21:04:50 -07001030 struct cgrp_cset_link *link;
Li Zefanc12f65d2009-01-07 18:07:42 -08001031
Tejun Heo69d02062013-06-12 21:04:50 -07001032 BUG_ON(list_empty(tmp_links));
Tejun Heo6803c002014-04-23 11:13:16 -04001033
1034 if (cgroup_on_dfl(cgrp))
1035 cset->dfl_cgrp = cgrp;
1036
Tejun Heo69d02062013-06-12 21:04:50 -07001037 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1038 link->cset = cset;
Paul Menage7717f7b2009-09-23 15:56:22 -07001039 link->cgrp = cgrp;
Tejun Heo842b5972014-04-25 18:28:02 -04001040
Paul Menage7717f7b2009-09-23 15:56:22 -07001041 /*
Tejun Heo389b9c12015-10-15 16:41:51 -04001042 * Always add links to the tail of the lists so that the lists are
1043 * in choronological order.
Paul Menage7717f7b2009-09-23 15:56:22 -07001044 */
Tejun Heo389b9c12015-10-15 16:41:51 -04001045 list_move_tail(&link->cset_link, &cgrp->cset_links);
Tejun Heo69d02062013-06-12 21:04:50 -07001046 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
Tejun Heo2ceb2312015-10-15 16:41:51 -04001047
1048 if (cgroup_parent(cgrp))
1049 cgroup_get(cgrp);
Li Zefanc12f65d2009-01-07 18:07:42 -08001050}
1051
Tejun Heob326f9d2013-06-24 15:21:48 -07001052/**
1053 * find_css_set - return a new css_set with one cgroup updated
1054 * @old_cset: the baseline css_set
1055 * @cgrp: the cgroup to be updated
1056 *
1057 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1058 * substituted into the appropriate hierarchy.
Paul Menage817929e2007-10-18 23:39:36 -07001059 */
Tejun Heo5abb8852013-06-12 21:04:49 -07001060static struct css_set *find_css_set(struct css_set *old_cset,
1061 struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -07001062{
Tejun Heob326f9d2013-06-24 15:21:48 -07001063 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
Tejun Heo5abb8852013-06-12 21:04:49 -07001064 struct css_set *cset;
Tejun Heo69d02062013-06-12 21:04:50 -07001065 struct list_head tmp_links;
1066 struct cgrp_cset_link *link;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001067 struct cgroup_subsys *ss;
Li Zefan0ac801f2013-01-10 11:49:27 +08001068 unsigned long key;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001069 int ssid;
Li Zefan472b10532008-04-29 01:00:11 -07001070
Tejun Heob326f9d2013-06-24 15:21:48 -07001071 lockdep_assert_held(&cgroup_mutex);
1072
Paul Menage817929e2007-10-18 23:39:36 -07001073 /* First see if we already have a cgroup group that matches
1074 * the desired set */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001075 spin_lock_irq(&css_set_lock);
Tejun Heo5abb8852013-06-12 21:04:49 -07001076 cset = find_existing_css_set(old_cset, cgrp, template);
1077 if (cset)
1078 get_css_set(cset);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001079 spin_unlock_irq(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001080
Tejun Heo5abb8852013-06-12 21:04:49 -07001081 if (cset)
1082 return cset;
Paul Menage817929e2007-10-18 23:39:36 -07001083
Tejun Heof4f4be22013-06-12 21:04:51 -07001084 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
Tejun Heo5abb8852013-06-12 21:04:49 -07001085 if (!cset)
Paul Menage817929e2007-10-18 23:39:36 -07001086 return NULL;
1087
Tejun Heo69d02062013-06-12 21:04:50 -07001088 /* Allocate all the cgrp_cset_link objects that we'll need */
Tejun Heo9871bf92013-06-24 15:21:47 -07001089 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
Tejun Heo5abb8852013-06-12 21:04:49 -07001090 kfree(cset);
Paul Menage817929e2007-10-18 23:39:36 -07001091 return NULL;
1092 }
1093
Tejun Heo5abb8852013-06-12 21:04:49 -07001094 atomic_set(&cset->refcount, 1);
Tejun Heo69d02062013-06-12 21:04:50 -07001095 INIT_LIST_HEAD(&cset->cgrp_links);
Tejun Heo5abb8852013-06-12 21:04:49 -07001096 INIT_LIST_HEAD(&cset->tasks);
Tejun Heoc7561122014-02-25 10:04:01 -05001097 INIT_LIST_HEAD(&cset->mg_tasks);
Tejun Heo1958d2d2014-02-25 10:04:03 -05001098 INIT_LIST_HEAD(&cset->mg_preload_node);
Tejun Heob3dc0942014-02-25 10:04:01 -05001099 INIT_LIST_HEAD(&cset->mg_node);
Tejun Heoed27b9f2015-10-15 16:41:52 -04001100 INIT_LIST_HEAD(&cset->task_iters);
Tejun Heo5abb8852013-06-12 21:04:49 -07001101 INIT_HLIST_NODE(&cset->hlist);
Paul Menage817929e2007-10-18 23:39:36 -07001102
1103 /* Copy the set of subsystem state objects generated in
1104 * find_existing_css_set() */
Tejun Heo5abb8852013-06-12 21:04:49 -07001105 memcpy(cset->subsys, template, sizeof(cset->subsys));
Paul Menage817929e2007-10-18 23:39:36 -07001106
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001107 spin_lock_irq(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001108 /* Add reference counts and links from the new css_set. */
Tejun Heo69d02062013-06-12 21:04:50 -07001109 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07001110 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001111
Paul Menage7717f7b2009-09-23 15:56:22 -07001112 if (c->root == cgrp->root)
1113 c = cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001114 link_css_set(&tmp_links, cset, c);
Paul Menage7717f7b2009-09-23 15:56:22 -07001115 }
Paul Menage817929e2007-10-18 23:39:36 -07001116
Tejun Heo69d02062013-06-12 21:04:50 -07001117 BUG_ON(!list_empty(&tmp_links));
Paul Menage817929e2007-10-18 23:39:36 -07001118
Paul Menage817929e2007-10-18 23:39:36 -07001119 css_set_count++;
Li Zefan472b10532008-04-29 01:00:11 -07001120
Tejun Heo2d8f2432014-04-23 11:13:15 -04001121 /* Add @cset to the hash table */
Tejun Heo5abb8852013-06-12 21:04:49 -07001122 key = css_set_hash(cset->subsys);
1123 hash_add(css_set_table, &cset->hlist, key);
Li Zefan472b10532008-04-29 01:00:11 -07001124
Tejun Heo53254f92015-11-23 14:55:41 -05001125 for_each_subsys(ss, ssid) {
1126 struct cgroup_subsys_state *css = cset->subsys[ssid];
1127
Tejun Heo2d8f2432014-04-23 11:13:15 -04001128 list_add_tail(&cset->e_cset_node[ssid],
Tejun Heo53254f92015-11-23 14:55:41 -05001129 &css->cgroup->e_csets[ssid]);
1130 css_get(css);
1131 }
Tejun Heo2d8f2432014-04-23 11:13:15 -04001132
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001133 spin_unlock_irq(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07001134
Tejun Heo5abb8852013-06-12 21:04:49 -07001135 return cset;
Paul Menageb4f48b62007-10-18 23:39:33 -07001136}
1137
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001138static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
Paul Menage7717f7b2009-09-23 15:56:22 -07001139{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001140 struct cgroup *root_cgrp = kf_root->kn->priv;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001141
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001142 return root_cgrp->root;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001143}
1144
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001145static int cgroup_init_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001146{
1147 int id;
1148
1149 lockdep_assert_held(&cgroup_mutex);
1150
Tejun Heo985ed672014-03-19 10:23:53 -04001151 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
Tejun Heof2e85d52014-02-11 11:52:49 -05001152 if (id < 0)
1153 return id;
1154
1155 root->hierarchy_id = id;
1156 return 0;
1157}
1158
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001159static void cgroup_exit_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001160{
1161 lockdep_assert_held(&cgroup_mutex);
1162
Johannes Weiner8c8a5502016-06-17 12:23:59 -04001163 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
Tejun Heof2e85d52014-02-11 11:52:49 -05001164}
1165
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001166static void cgroup_free_root(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -05001167{
1168 if (root) {
Tejun Heof2e85d52014-02-11 11:52:49 -05001169 idr_destroy(&root->cgroup_idr);
1170 kfree(root);
1171 }
1172}
1173
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001174static void cgroup_destroy_root(struct cgroup_root *root)
Tejun Heo59f52962014-02-11 11:52:49 -05001175{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001176 struct cgroup *cgrp = &root->cgrp;
Tejun Heof2e85d52014-02-11 11:52:49 -05001177 struct cgrp_cset_link *link, *tmp_link;
Tejun Heof2e85d52014-02-11 11:52:49 -05001178
Tejun Heo334c3672016-03-03 09:58:01 -05001179 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Tejun Heof2e85d52014-02-11 11:52:49 -05001180
Tejun Heo776f02f2014-02-12 09:29:50 -05001181 BUG_ON(atomic_read(&root->nr_cgrps));
Tejun Heod5c419b2014-05-16 13:22:48 -04001182 BUG_ON(!list_empty(&cgrp->self.children));
Tejun Heof2e85d52014-02-11 11:52:49 -05001183
Tejun Heof2e85d52014-02-11 11:52:49 -05001184 /* Rebind all subsystems back to the default hierarchy */
Tejun Heo334c3672016-03-03 09:58:01 -05001185 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
Tejun Heof2e85d52014-02-11 11:52:49 -05001186
1187 /*
1188 * Release all the links from cset_links to this hierarchy's
1189 * root cgroup
1190 */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001191 spin_lock_irq(&css_set_lock);
Tejun Heof2e85d52014-02-11 11:52:49 -05001192
1193 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1194 list_del(&link->cset_link);
1195 list_del(&link->cgrp_link);
1196 kfree(link);
1197 }
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001198
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001199 spin_unlock_irq(&css_set_lock);
Tejun Heof2e85d52014-02-11 11:52:49 -05001200
1201 if (!list_empty(&root->root_list)) {
1202 list_del(&root->root_list);
1203 cgroup_root_count--;
1204 }
1205
1206 cgroup_exit_root_id(root);
1207
1208 mutex_unlock(&cgroup_mutex);
Tejun Heof2e85d52014-02-11 11:52:49 -05001209
Tejun Heo2bd59d42014-02-11 11:52:49 -05001210 kernfs_destroy_root(root->kf_root);
Tejun Heof2e85d52014-02-11 11:52:49 -05001211 cgroup_free_root(root);
1212}
1213
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05001214/*
1215 * look up cgroup associated with current task's cgroup namespace on the
1216 * specified hierarchy
1217 */
1218static struct cgroup *
1219current_cgns_cgroup_from_root(struct cgroup_root *root)
1220{
1221 struct cgroup *res = NULL;
1222 struct css_set *cset;
1223
1224 lockdep_assert_held(&css_set_lock);
1225
1226 rcu_read_lock();
1227
1228 cset = current->nsproxy->cgroup_ns->root_cset;
1229 if (cset == &init_css_set) {
1230 res = &root->cgrp;
1231 } else {
1232 struct cgrp_cset_link *link;
1233
1234 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1235 struct cgroup *c = link->cgrp;
1236
1237 if (c->root == root) {
1238 res = c;
1239 break;
1240 }
1241 }
1242 }
1243 rcu_read_unlock();
1244
1245 BUG_ON(!res);
1246 return res;
1247}
1248
Tejun Heoceb6a082014-02-25 10:04:02 -05001249/* look up cgroup associated with given css_set on the specified hierarchy */
1250static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001251 struct cgroup_root *root)
Paul Menage7717f7b2009-09-23 15:56:22 -07001252{
Paul Menage7717f7b2009-09-23 15:56:22 -07001253 struct cgroup *res = NULL;
1254
Tejun Heo96d365e2014-02-13 06:58:40 -05001255 lockdep_assert_held(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001256 lockdep_assert_held(&css_set_lock);
Tejun Heo96d365e2014-02-13 06:58:40 -05001257
Tejun Heo5abb8852013-06-12 21:04:49 -07001258 if (cset == &init_css_set) {
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001259 res = &root->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -07001260 } else {
Tejun Heo69d02062013-06-12 21:04:50 -07001261 struct cgrp_cset_link *link;
1262
1263 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07001264 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -07001265
Paul Menage7717f7b2009-09-23 15:56:22 -07001266 if (c->root == root) {
1267 res = c;
1268 break;
1269 }
1270 }
1271 }
Tejun Heo96d365e2014-02-13 06:58:40 -05001272
Paul Menage7717f7b2009-09-23 15:56:22 -07001273 BUG_ON(!res);
1274 return res;
1275}
1276
1277/*
Tejun Heoceb6a082014-02-25 10:04:02 -05001278 * Return the cgroup for "task" from the given hierarchy. Must be
Tejun Heof0d9a5f2015-10-15 16:41:53 -04001279 * called with cgroup_mutex and css_set_lock held.
Tejun Heoceb6a082014-02-25 10:04:02 -05001280 */
1281static struct cgroup *task_cgroup_from_root(struct task_struct *task,
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001282 struct cgroup_root *root)
Tejun Heoceb6a082014-02-25 10:04:02 -05001283{
1284 /*
1285 * No need to lock the task - since we hold cgroup_mutex the
1286 * task can't change groups, so the only thing that can happen
1287 * is that it exits and its css is set back to init_css_set.
1288 */
1289 return cset_cgroup_from_root(task_css_set(task), root);
1290}
1291
1292/*
Paul Menageddbcc7e2007-10-18 23:39:30 -07001293 * A task must hold cgroup_mutex to modify cgroups.
1294 *
1295 * Any task can increment and decrement the count field without lock.
1296 * So in general, code holding cgroup_mutex can't rely on the count
1297 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -08001298 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -07001299 * means that no tasks are currently attached, therefore there is no
1300 * way a task attached to that cgroup can fork (the other way to
1301 * increment the count). So code holding cgroup_mutex can safely
1302 * assume that if the count is zero, it will stay zero. Similarly, if
1303 * a task holds cgroup_mutex on a cgroup with zero count, it
1304 * knows that the cgroup won't be removed, as cgroup_rmdir()
1305 * needs that mutex.
1306 *
Paul Menageddbcc7e2007-10-18 23:39:30 -07001307 * A cgroup can only be deleted if both its 'count' of using tasks
1308 * is zero, and its list of 'children' cgroups is empty. Since all
1309 * tasks in the system use _some_ cgroup, and since there is always at
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001310 * least one task in the system (init, pid == 1), therefore, root cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -07001311 * always has either children cgroups and/or using tasks. So we don't
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001312 * need a special hack to ensure that root cgroup cannot be deleted.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001313 *
1314 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -08001315 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -07001316 */
1317
Tejun Heo2bd59d42014-02-11 11:52:49 -05001318static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001319static const struct file_operations proc_cgroupstats_operations;
Paul Menagea4243162007-10-18 23:39:35 -07001320
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001321static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1322 char *buf)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001323{
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001324 struct cgroup_subsys *ss = cft->ss;
1325
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001326 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1327 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1328 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001329 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1330 cft->name);
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05001331 else
1332 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1333 return buf;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001334}
1335
Tejun Heof2e85d52014-02-11 11:52:49 -05001336/**
1337 * cgroup_file_mode - deduce file mode of a control file
1338 * @cft: the control file in question
1339 *
Tejun Heo7dbdb192015-09-18 17:54:23 -04001340 * S_IRUGO for read, S_IWUSR for write.
Tejun Heof2e85d52014-02-11 11:52:49 -05001341 */
1342static umode_t cgroup_file_mode(const struct cftype *cft)
Li Zefan65dff752013-03-01 15:01:56 +08001343{
Tejun Heof2e85d52014-02-11 11:52:49 -05001344 umode_t mode = 0;
Li Zefan65dff752013-03-01 15:01:56 +08001345
Tejun Heof2e85d52014-02-11 11:52:49 -05001346 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1347 mode |= S_IRUGO;
1348
Tejun Heo7dbdb192015-09-18 17:54:23 -04001349 if (cft->write_u64 || cft->write_s64 || cft->write) {
1350 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1351 mode |= S_IWUGO;
1352 else
1353 mode |= S_IWUSR;
1354 }
Tejun Heof2e85d52014-02-11 11:52:49 -05001355
1356 return mode;
Li Zefan65dff752013-03-01 15:01:56 +08001357}
1358
Tejun Heoa9746d82014-05-13 12:19:22 -04001359/**
Tejun Heo8699b772016-02-22 22:25:46 -05001360 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
Tejun Heo0f060de2014-11-18 02:49:50 -05001361 * @subtree_control: the new subtree_control mask to consider
Tejun Heo5ced2512016-03-03 09:58:01 -05001362 * @this_ss_mask: available subsystems
Tejun Heoaf0ba672014-07-08 18:02:57 -04001363 *
1364 * On the default hierarchy, a subsystem may request other subsystems to be
1365 * enabled together through its ->depends_on mask. In such cases, more
1366 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1367 *
Tejun Heo0f060de2014-11-18 02:49:50 -05001368 * This function calculates which subsystems need to be enabled if
Tejun Heo5ced2512016-03-03 09:58:01 -05001369 * @subtree_control is to be applied while restricted to @this_ss_mask.
Tejun Heoaf0ba672014-07-08 18:02:57 -04001370 */
Tejun Heo5ced2512016-03-03 09:58:01 -05001371static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
Tejun Heo667c2492014-07-08 18:02:56 -04001372{
Tejun Heo6e5c8302016-02-22 22:25:47 -05001373 u16 cur_ss_mask = subtree_control;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001374 struct cgroup_subsys *ss;
1375 int ssid;
1376
1377 lockdep_assert_held(&cgroup_mutex);
1378
Tejun Heof6d635ad2016-03-08 11:51:26 -05001379 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1380
Tejun Heoaf0ba672014-07-08 18:02:57 -04001381 while (true) {
Tejun Heo6e5c8302016-02-22 22:25:47 -05001382 u16 new_ss_mask = cur_ss_mask;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001383
Tejun Heob4e0eea2016-02-22 22:25:46 -05001384 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
Aleksa Saraia966a4e2015-06-06 10:02:15 +10001385 new_ss_mask |= ss->depends_on;
Tejun Heob4e0eea2016-02-22 22:25:46 -05001386 } while_each_subsys_mask();
Tejun Heoaf0ba672014-07-08 18:02:57 -04001387
1388 /*
1389 * Mask out subsystems which aren't available. This can
1390 * happen only if some depended-upon subsystems were bound
1391 * to non-default hierarchies.
1392 */
Tejun Heo5ced2512016-03-03 09:58:01 -05001393 new_ss_mask &= this_ss_mask;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001394
1395 if (new_ss_mask == cur_ss_mask)
1396 break;
1397 cur_ss_mask = new_ss_mask;
1398 }
1399
Tejun Heo0f060de2014-11-18 02:49:50 -05001400 return cur_ss_mask;
1401}
1402
1403/**
Tejun Heoa9746d82014-05-13 12:19:22 -04001404 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1405 * @kn: the kernfs_node being serviced
1406 *
1407 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1408 * the method finishes if locking succeeded. Note that once this function
1409 * returns the cgroup returned by cgroup_kn_lock_live() may become
1410 * inaccessible any time. If the caller intends to continue to access the
1411 * cgroup, it should pin it before invoking this function.
1412 */
1413static void cgroup_kn_unlock(struct kernfs_node *kn)
1414{
1415 struct cgroup *cgrp;
1416
1417 if (kernfs_type(kn) == KERNFS_DIR)
1418 cgrp = kn->priv;
1419 else
1420 cgrp = kn->parent->priv;
1421
1422 mutex_unlock(&cgroup_mutex);
Tejun Heoa9746d82014-05-13 12:19:22 -04001423
1424 kernfs_unbreak_active_protection(kn);
1425 cgroup_put(cgrp);
1426}
1427
1428/**
1429 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1430 * @kn: the kernfs_node being serviced
Tejun Heo945ba192016-03-03 09:58:00 -05001431 * @drain_offline: perform offline draining on the cgroup
Tejun Heoa9746d82014-05-13 12:19:22 -04001432 *
1433 * This helper is to be used by a cgroup kernfs method currently servicing
1434 * @kn. It breaks the active protection, performs cgroup locking and
1435 * verifies that the associated cgroup is alive. Returns the cgroup if
1436 * alive; otherwise, %NULL. A successful return should be undone by a
Tejun Heo945ba192016-03-03 09:58:00 -05001437 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1438 * cgroup is drained of offlining csses before return.
Tejun Heoa9746d82014-05-13 12:19:22 -04001439 *
1440 * Any cgroup kernfs method implementation which requires locking the
1441 * associated cgroup should use this helper. It avoids nesting cgroup
1442 * locking under kernfs active protection and allows all kernfs operations
1443 * including self-removal.
1444 */
Tejun Heo945ba192016-03-03 09:58:00 -05001445static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn,
1446 bool drain_offline)
Tejun Heoa9746d82014-05-13 12:19:22 -04001447{
1448 struct cgroup *cgrp;
1449
1450 if (kernfs_type(kn) == KERNFS_DIR)
1451 cgrp = kn->priv;
1452 else
1453 cgrp = kn->parent->priv;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001454
Tejun Heo2bd59d42014-02-11 11:52:49 -05001455 /*
Tejun Heo01f64742014-05-13 12:19:23 -04001456 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoa9746d82014-05-13 12:19:22 -04001457 * active_ref. cgroup liveliness check alone provides enough
1458 * protection against removal. Ensure @cgrp stays accessible and
1459 * break the active_ref protection.
Tejun Heo2bd59d42014-02-11 11:52:49 -05001460 */
Li Zefanaa323622014-09-04 14:43:38 +08001461 if (!cgroup_tryget(cgrp))
1462 return NULL;
Tejun Heoa9746d82014-05-13 12:19:22 -04001463 kernfs_break_active_protection(kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001464
Tejun Heo945ba192016-03-03 09:58:00 -05001465 if (drain_offline)
1466 cgroup_lock_and_drain_offline(cgrp);
1467 else
1468 mutex_lock(&cgroup_mutex);
Tejun Heoa9746d82014-05-13 12:19:22 -04001469
1470 if (!cgroup_is_dead(cgrp))
1471 return cgrp;
1472
1473 cgroup_kn_unlock(kn);
1474 return NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001475}
1476
Li Zefan2739d3c2013-01-21 18:18:33 +08001477static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001478{
Tejun Heo2bd59d42014-02-11 11:52:49 -05001479 char name[CGROUP_FILE_NAME_MAX];
Paul Menageddbcc7e2007-10-18 23:39:30 -07001480
Tejun Heo01f64742014-05-13 12:19:23 -04001481 lockdep_assert_held(&cgroup_mutex);
Tejun Heo34c06252015-11-05 00:12:24 -05001482
1483 if (cft->file_offset) {
1484 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1485 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1486
1487 spin_lock_irq(&cgroup_file_kn_lock);
1488 cfile->kn = NULL;
1489 spin_unlock_irq(&cgroup_file_kn_lock);
1490 }
1491
Tejun Heo2bd59d42014-02-11 11:52:49 -05001492 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
Tejun Heo05ef1d72012-04-01 12:09:56 -07001493}
1494
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001495/**
Tejun Heo4df8dc92015-09-18 17:54:23 -04001496 * css_clear_dir - remove subsys files in a cgroup directory
1497 * @css: taget css
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001498 */
Tejun Heo334c3672016-03-03 09:58:01 -05001499static void css_clear_dir(struct cgroup_subsys_state *css)
Tejun Heo05ef1d72012-04-01 12:09:56 -07001500{
Tejun Heo334c3672016-03-03 09:58:01 -05001501 struct cgroup *cgrp = css->cgroup;
Tejun Heo4df8dc92015-09-18 17:54:23 -04001502 struct cftype *cfts;
Tejun Heo05ef1d72012-04-01 12:09:56 -07001503
Tejun Heo88cb04b2016-03-03 09:57:58 -05001504 if (!(css->flags & CSS_VISIBLE))
1505 return;
1506
1507 css->flags &= ~CSS_VISIBLE;
1508
Tejun Heo4df8dc92015-09-18 17:54:23 -04001509 list_for_each_entry(cfts, &css->ss->cfts, node)
1510 cgroup_addrm_files(css, cgrp, cfts, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001511}
1512
Tejun Heoccdca212015-09-18 17:54:23 -04001513/**
Tejun Heo4df8dc92015-09-18 17:54:23 -04001514 * css_populate_dir - create subsys files in a cgroup directory
1515 * @css: target css
Tejun Heoccdca212015-09-18 17:54:23 -04001516 *
1517 * On failure, no file is added.
1518 */
Tejun Heo334c3672016-03-03 09:58:01 -05001519static int css_populate_dir(struct cgroup_subsys_state *css)
Tejun Heoccdca212015-09-18 17:54:23 -04001520{
Tejun Heo334c3672016-03-03 09:58:01 -05001521 struct cgroup *cgrp = css->cgroup;
Tejun Heo4df8dc92015-09-18 17:54:23 -04001522 struct cftype *cfts, *failed_cfts;
1523 int ret;
Tejun Heoccdca212015-09-18 17:54:23 -04001524
Tejun Heo03970d32016-03-03 09:58:00 -05001525 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
Tejun Heo88cb04b2016-03-03 09:57:58 -05001526 return 0;
1527
Tejun Heo4df8dc92015-09-18 17:54:23 -04001528 if (!css->ss) {
1529 if (cgroup_on_dfl(cgrp))
1530 cfts = cgroup_dfl_base_files;
1531 else
1532 cfts = cgroup_legacy_base_files;
Tejun Heoccdca212015-09-18 17:54:23 -04001533
Tejun Heo4df8dc92015-09-18 17:54:23 -04001534 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1535 }
Tejun Heoccdca212015-09-18 17:54:23 -04001536
Tejun Heo4df8dc92015-09-18 17:54:23 -04001537 list_for_each_entry(cfts, &css->ss->cfts, node) {
1538 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1539 if (ret < 0) {
1540 failed_cfts = cfts;
1541 goto err;
Tejun Heoccdca212015-09-18 17:54:23 -04001542 }
1543 }
Tejun Heo88cb04b2016-03-03 09:57:58 -05001544
1545 css->flags |= CSS_VISIBLE;
1546
Tejun Heoccdca212015-09-18 17:54:23 -04001547 return 0;
1548err:
Tejun Heo4df8dc92015-09-18 17:54:23 -04001549 list_for_each_entry(cfts, &css->ss->cfts, node) {
1550 if (cfts == failed_cfts)
1551 break;
1552 cgroup_addrm_files(css, cgrp, cfts, false);
1553 }
Tejun Heoccdca212015-09-18 17:54:23 -04001554 return ret;
1555}
1556
Tejun Heo6e5c8302016-02-22 22:25:47 -05001557static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001558{
Tejun Heo1ada4832015-09-18 17:54:23 -04001559 struct cgroup *dcgrp = &dst_root->cgrp;
Tejun Heo30159ec2013-06-25 11:53:37 -07001560 struct cgroup_subsys *ss;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001561 int ssid, i, ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001562
Tejun Heoace2bee812014-02-11 11:52:47 -05001563 lockdep_assert_held(&cgroup_mutex);
Ben Blumaae8aab2010-03-10 15:22:07 -08001564
Tejun Heob4e0eea2016-02-22 22:25:46 -05001565 do_each_subsys_mask(ss, ssid, ss_mask) {
Tejun Heof6d635ad2016-03-08 11:51:26 -05001566 /*
1567 * If @ss has non-root csses attached to it, can't move.
1568 * If @ss is an implicit controller, it is exempt from this
1569 * rule and can be stolen.
1570 */
1571 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1572 !ss->implicit_on_dfl)
Tejun Heo3ed80a62014-02-08 10:36:58 -05001573 return -EBUSY;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001574
Tejun Heo5df36032014-03-19 10:23:54 -04001575 /* can't move between two non-dummy roots either */
Tejun Heo7fd8c562014-04-23 11:13:16 -04001576 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
Tejun Heo5df36032014-03-19 10:23:54 -04001577 return -EBUSY;
Tejun Heob4e0eea2016-02-22 22:25:46 -05001578 } while_each_subsys_mask();
Paul Menageddbcc7e2007-10-18 23:39:30 -07001579
Tejun Heob4e0eea2016-02-22 22:25:46 -05001580 do_each_subsys_mask(ss, ssid, ss_mask) {
Tejun Heo1ada4832015-09-18 17:54:23 -04001581 struct cgroup_root *src_root = ss->root;
1582 struct cgroup *scgrp = &src_root->cgrp;
1583 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001584 struct css_set *cset;
Tejun Heo30159ec2013-06-25 11:53:37 -07001585
Tejun Heo1ada4832015-09-18 17:54:23 -04001586 WARN_ON(!css || cgroup_css(dcgrp, ss));
Tejun Heo73e80ed2013-08-13 11:01:55 -04001587
Tejun Heo334c3672016-03-03 09:58:01 -05001588 /* disable from the source */
1589 src_root->subsys_mask &= ~(1 << ssid);
1590 WARN_ON(cgroup_apply_control(scgrp));
1591 cgroup_finalize_control(scgrp, 0);
Tejun Heo4df8dc92015-09-18 17:54:23 -04001592
Tejun Heo334c3672016-03-03 09:58:01 -05001593 /* rebind */
Tejun Heo1ada4832015-09-18 17:54:23 -04001594 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1595 rcu_assign_pointer(dcgrp->subsys[ssid], css);
Tejun Heo5df36032014-03-19 10:23:54 -04001596 ss->root = dst_root;
Tejun Heo1ada4832015-09-18 17:54:23 -04001597 css->cgroup = dcgrp;
Tejun Heoa8a648c42013-06-24 15:21:47 -07001598
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001599 spin_lock_irq(&css_set_lock);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001600 hash_for_each(css_set_table, i, cset, hlist)
1601 list_move_tail(&cset->e_cset_node[ss->id],
Tejun Heo1ada4832015-09-18 17:54:23 -04001602 &dcgrp->e_csets[ss->id]);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001603 spin_unlock_irq(&css_set_lock);
Tejun Heo2d8f2432014-04-23 11:13:15 -04001604
Tejun Heobd53d612014-04-23 11:13:16 -04001605 /* default hierarchy doesn't enable controllers by default */
Tejun Heof392e512014-04-23 11:13:14 -04001606 dst_root->subsys_mask |= 1 << ssid;
Tejun Heo49d1dc42015-09-18 11:56:28 -04001607 if (dst_root == &cgrp_dfl_root) {
1608 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1609 } else {
Tejun Heo1ada4832015-09-18 17:54:23 -04001610 dcgrp->subtree_control |= 1 << ssid;
Tejun Heo49d1dc42015-09-18 11:56:28 -04001611 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
Tejun Heo667c2492014-07-08 18:02:56 -04001612 }
Tejun Heo73e80ed2013-08-13 11:01:55 -04001613
Tejun Heo334c3672016-03-03 09:58:01 -05001614 ret = cgroup_apply_control(dcgrp);
1615 if (ret)
1616 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1617 ss->name, ret);
1618
Tejun Heo5df36032014-03-19 10:23:54 -04001619 if (ss->bind)
1620 ss->bind(css);
Tejun Heob4e0eea2016-02-22 22:25:46 -05001621 } while_each_subsys_mask();
Paul Menageddbcc7e2007-10-18 23:39:30 -07001622
Tejun Heo1ada4832015-09-18 17:54:23 -04001623 kernfs_activate(dcgrp->kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001624 return 0;
1625}
1626
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05001627static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1628 struct kernfs_root *kf_root)
1629{
Felipe Balbi09be4c82016-05-12 12:34:38 +03001630 int len = 0;
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05001631 char *buf = NULL;
1632 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1633 struct cgroup *ns_cgroup;
1634
1635 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1636 if (!buf)
1637 return -ENOMEM;
1638
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001639 spin_lock_irq(&css_set_lock);
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05001640 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1641 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001642 spin_unlock_irq(&css_set_lock);
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05001643
1644 if (len >= PATH_MAX)
1645 len = -ERANGE;
1646 else if (len > 0) {
1647 seq_escape(sf, buf, " \t\n\\");
1648 len = 0;
1649 }
1650 kfree(buf);
1651 return len;
1652}
1653
Tejun Heo2bd59d42014-02-11 11:52:49 -05001654static int cgroup_show_options(struct seq_file *seq,
1655 struct kernfs_root *kf_root)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001656{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001657 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001658 struct cgroup_subsys *ss;
Tejun Heob85d2042013-12-06 15:11:57 -05001659 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001660
Tejun Heod98817d2015-08-18 13:58:16 -07001661 if (root != &cgrp_dfl_root)
1662 for_each_subsys(ss, ssid)
1663 if (root->subsys_mask & (1 << ssid))
Kees Cook61e57c02015-09-08 14:58:22 -07001664 seq_show_option(seq, ss->legacy_name, NULL);
Tejun Heo93438622013-04-14 20:15:25 -07001665 if (root->flags & CGRP_ROOT_NOPREFIX)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001666 seq_puts(seq, ",noprefix");
Tejun Heo93438622013-04-14 20:15:25 -07001667 if (root->flags & CGRP_ROOT_XATTR)
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001668 seq_puts(seq, ",xattr");
Tejun Heo69e943b2014-02-08 10:36:58 -05001669
1670 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001671 if (strlen(root->release_agent_path))
Kees Cooka068acf2015-09-04 15:44:57 -07001672 seq_show_option(seq, "release_agent",
1673 root->release_agent_path);
Tejun Heo69e943b2014-02-08 10:36:58 -05001674 spin_unlock(&release_agent_path_lock);
1675
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001676 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
Daniel Lezcano97978e62010-10-27 15:33:35 -07001677 seq_puts(seq, ",clone_children");
Paul Menagec6d57f332009-09-23 15:56:19 -07001678 if (strlen(root->name))
Kees Cooka068acf2015-09-04 15:44:57 -07001679 seq_show_option(seq, "name", root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001680 return 0;
1681}
1682
1683struct cgroup_sb_opts {
Tejun Heo6e5c8302016-02-22 22:25:47 -05001684 u16 subsys_mask;
Tejun Heo69dfa002014-05-04 15:09:13 -04001685 unsigned int flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001686 char *release_agent;
Tejun Heo2260e7f2012-11-19 08:13:38 -08001687 bool cpuset_clone_children;
Paul Menagec6d57f332009-09-23 15:56:19 -07001688 char *name;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001689 /* User explicitly requested empty subsystem */
1690 bool none;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001691};
1692
Ben Blumcf5d5942010-03-10 15:22:09 -08001693static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001694{
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001695 char *token, *o = data;
1696 bool all_ss = false, one_ss = false;
Tejun Heo6e5c8302016-02-22 22:25:47 -05001697 u16 mask = U16_MAX;
Tejun Heo30159ec2013-06-25 11:53:37 -07001698 struct cgroup_subsys *ss;
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001699 int nr_opts = 0;
Tejun Heo30159ec2013-06-25 11:53:37 -07001700 int i;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001701
1702#ifdef CONFIG_CPUSETS
Tejun Heo6e5c8302016-02-22 22:25:47 -05001703 mask = ~((u16)1 << cpuset_cgrp_id);
Li Zefanf9ab5b52009-06-17 16:26:33 -07001704#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001705
Paul Menagec6d57f332009-09-23 15:56:19 -07001706 memset(opts, 0, sizeof(*opts));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001707
1708 while ((token = strsep(&o, ",")) != NULL) {
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001709 nr_opts++;
1710
Paul Menageddbcc7e2007-10-18 23:39:30 -07001711 if (!*token)
1712 return -EINVAL;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001713 if (!strcmp(token, "none")) {
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001714 /* Explicitly have no subsystems */
1715 opts->none = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001716 continue;
1717 }
1718 if (!strcmp(token, "all")) {
1719 /* Mutually exclusive option 'all' + subsystem name */
1720 if (one_ss)
1721 return -EINVAL;
1722 all_ss = true;
1723 continue;
1724 }
1725 if (!strcmp(token, "noprefix")) {
Tejun Heo93438622013-04-14 20:15:25 -07001726 opts->flags |= CGRP_ROOT_NOPREFIX;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001727 continue;
1728 }
1729 if (!strcmp(token, "clone_children")) {
Tejun Heo2260e7f2012-11-19 08:13:38 -08001730 opts->cpuset_clone_children = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001731 continue;
1732 }
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001733 if (!strcmp(token, "xattr")) {
Tejun Heo93438622013-04-14 20:15:25 -07001734 opts->flags |= CGRP_ROOT_XATTR;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001735 continue;
1736 }
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001737 if (!strncmp(token, "release_agent=", 14)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07001738 /* Specifying two release agents is forbidden */
1739 if (opts->release_agent)
1740 return -EINVAL;
Paul Menagec6d57f332009-09-23 15:56:19 -07001741 opts->release_agent =
Dan Carpentere400c282010-08-10 18:02:54 -07001742 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001743 if (!opts->release_agent)
1744 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001745 continue;
1746 }
1747 if (!strncmp(token, "name=", 5)) {
Paul Menagec6d57f332009-09-23 15:56:19 -07001748 const char *name = token + 5;
1749 /* Can't specify an empty name */
1750 if (!strlen(name))
1751 return -EINVAL;
1752 /* Must match [\w.-]+ */
1753 for (i = 0; i < strlen(name); i++) {
1754 char c = name[i];
1755 if (isalnum(c))
1756 continue;
1757 if ((c == '.') || (c == '-') || (c == '_'))
1758 continue;
1759 return -EINVAL;
1760 }
1761 /* Specifying two names is forbidden */
1762 if (opts->name)
1763 return -EINVAL;
1764 opts->name = kstrndup(name,
Dan Carpentere400c282010-08-10 18:02:54 -07001765 MAX_CGROUP_ROOT_NAMELEN - 1,
Paul Menagec6d57f332009-09-23 15:56:19 -07001766 GFP_KERNEL);
1767 if (!opts->name)
1768 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001769
1770 continue;
1771 }
1772
Tejun Heo30159ec2013-06-25 11:53:37 -07001773 for_each_subsys(ss, i) {
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07001774 if (strcmp(token, ss->legacy_name))
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001775 continue;
Tejun Heofc5ed1e2015-09-18 11:56:28 -04001776 if (!cgroup_ssid_enabled(i))
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001777 continue;
Johannes Weiner223ffb22016-02-11 13:34:49 -05001778 if (cgroup_ssid_no_v1(i))
1779 continue;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001780
1781 /* Mutually exclusive option 'all' + subsystem name */
1782 if (all_ss)
1783 return -EINVAL;
Tejun Heo69dfa002014-05-04 15:09:13 -04001784 opts->subsys_mask |= (1 << i);
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001785 one_ss = true;
1786
1787 break;
1788 }
1789 if (i == CGROUP_SUBSYS_COUNT)
1790 return -ENOENT;
1791 }
1792
Li Zefanf9ab5b52009-06-17 16:26:33 -07001793 /*
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001794 * If the 'all' option was specified select all the subsystems,
1795 * otherwise if 'none', 'name=' and a subsystem name options were
1796 * not specified, let's default to 'all'
1797 */
1798 if (all_ss || (!one_ss && !opts->none && !opts->name))
1799 for_each_subsys(ss, i)
Johannes Weiner223ffb22016-02-11 13:34:49 -05001800 if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i))
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001801 opts->subsys_mask |= (1 << i);
1802
1803 /*
1804 * We either have to specify by name or by subsystems. (So all
1805 * empty hierarchies must have a name).
1806 */
1807 if (!opts->subsys_mask && !opts->name)
1808 return -EINVAL;
1809
1810 /*
Li Zefanf9ab5b52009-06-17 16:26:33 -07001811 * Option noprefix was introduced just for backward compatibility
1812 * with the old cpuset, so we allow noprefix only if mounting just
1813 * the cpuset subsystem.
1814 */
Tejun Heo93438622013-04-14 20:15:25 -07001815 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
Li Zefanf9ab5b52009-06-17 16:26:33 -07001816 return -EINVAL;
1817
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001818 /* Can't specify "none" and some subsystems */
Aristeu Rozanskia1a71b42012-08-23 16:53:31 -04001819 if (opts->subsys_mask && opts->none)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001820 return -EINVAL;
1821
Paul Menageddbcc7e2007-10-18 23:39:30 -07001822 return 0;
1823}
1824
Tejun Heo2bd59d42014-02-11 11:52:49 -05001825static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001826{
1827 int ret = 0;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001828 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001829 struct cgroup_sb_opts opts;
Tejun Heo6e5c8302016-02-22 22:25:47 -05001830 u16 added_mask, removed_mask;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001831
Tejun Heoaa6ec292014-07-09 10:08:08 -04001832 if (root == &cgrp_dfl_root) {
1833 pr_err("remount is not allowed\n");
Tejun Heo873fe092013-04-14 20:15:26 -07001834 return -EINVAL;
1835 }
1836
Tejun Heo334c3672016-03-03 09:58:01 -05001837 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001838
1839 /* See what subsystems are wanted */
1840 ret = parse_cgroupfs_options(data, &opts);
1841 if (ret)
1842 goto out_unlock;
1843
Tejun Heof392e512014-04-23 11:13:14 -04001844 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
Joe Perchesed3d2612014-04-25 18:28:03 -04001845 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
Jianyu Zhana2a1f9e2014-04-25 18:28:03 -04001846 task_tgid_nr(current), current->comm);
Tejun Heo8b5a5a92012-04-01 12:09:54 -07001847
Tejun Heof392e512014-04-23 11:13:14 -04001848 added_mask = opts.subsys_mask & ~root->subsys_mask;
1849 removed_mask = root->subsys_mask & ~opts.subsys_mask;
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001850
Ben Blumcf5d5942010-03-10 15:22:09 -08001851 /* Don't allow flags or name to change at remount */
Tejun Heo7450e902014-07-09 10:08:07 -04001852 if ((opts.flags ^ root->flags) ||
Ben Blumcf5d5942010-03-10 15:22:09 -08001853 (opts.name && strcmp(opts.name, root->name))) {
Tejun Heo69dfa002014-05-04 15:09:13 -04001854 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
Tejun Heo7450e902014-07-09 10:08:07 -04001855 opts.flags, opts.name ?: "", root->flags, root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001856 ret = -EINVAL;
Paul Menagec6d57f332009-09-23 15:56:19 -07001857 goto out_unlock;
1858 }
1859
Tejun Heof172e672013-06-28 17:07:30 -07001860 /* remounting is not allowed for populated hierarchies */
Tejun Heod5c419b2014-05-16 13:22:48 -04001861 if (!list_empty(&root->cgrp.self.children)) {
Tejun Heof172e672013-06-28 17:07:30 -07001862 ret = -EBUSY;
Li Zefan0670e082009-04-02 16:57:30 -07001863 goto out_unlock;
Ben Blumcf5d5942010-03-10 15:22:09 -08001864 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001865
Tejun Heo5df36032014-03-19 10:23:54 -04001866 ret = rebind_subsystems(root, added_mask);
Tejun Heo31261212013-06-28 17:07:30 -07001867 if (ret)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001868 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001869
Tejun Heo334c3672016-03-03 09:58:01 -05001870 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
Tejun Heo5df36032014-03-19 10:23:54 -04001871
Tejun Heo69e943b2014-02-08 10:36:58 -05001872 if (opts.release_agent) {
1873 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001874 strcpy(root->release_agent_path, opts.release_agent);
Tejun Heo69e943b2014-02-08 10:36:58 -05001875 spin_unlock(&release_agent_path_lock);
1876 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001877 out_unlock:
Jesper Juhl66bdc9c2009-04-02 16:57:27 -07001878 kfree(opts.release_agent);
Paul Menagec6d57f332009-09-23 15:56:19 -07001879 kfree(opts.name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001880 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001881 return ret;
1882}
1883
Tejun Heoafeb0f92014-02-13 06:58:39 -05001884/*
1885 * To reduce the fork() overhead for systems that are not actually using
1886 * their cgroups capability, we don't maintain the lists running through
1887 * each css_set to its tasks until we see the list actually used - in other
1888 * words after the first mount.
1889 */
1890static bool use_task_css_set_links __read_mostly;
1891
1892static void cgroup_enable_task_cg_lists(void)
1893{
1894 struct task_struct *p, *g;
1895
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001896 spin_lock_irq(&css_set_lock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001897
1898 if (use_task_css_set_links)
1899 goto out_unlock;
1900
1901 use_task_css_set_links = true;
1902
1903 /*
1904 * We need tasklist_lock because RCU is not safe against
1905 * while_each_thread(). Besides, a forking task that has passed
1906 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1907 * is not guaranteed to have its child immediately visible in the
1908 * tasklist if we walk through it with RCU.
1909 */
1910 read_lock(&tasklist_lock);
1911 do_each_thread(g, p) {
Tejun Heoafeb0f92014-02-13 06:58:39 -05001912 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1913 task_css_set(p) != &init_css_set);
1914
1915 /*
1916 * We should check if the process is exiting, otherwise
1917 * it will race with cgroup_exit() in that the list
1918 * entry won't be deleted though the process has exited.
Tejun Heof153ad12014-02-25 09:56:49 -05001919 * Do it while holding siglock so that we don't end up
1920 * racing against cgroup_exit().
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001921 *
1922 * Interrupts were already disabled while acquiring
1923 * the css_set_lock, so we do not need to disable it
1924 * again when acquiring the sighand->siglock here.
Tejun Heoafeb0f92014-02-13 06:58:39 -05001925 */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001926 spin_lock(&p->sighand->siglock);
Tejun Heoeaf797a2014-02-25 10:04:03 -05001927 if (!(p->flags & PF_EXITING)) {
1928 struct css_set *cset = task_css_set(p);
1929
Tejun Heo0de09422015-10-15 16:41:49 -04001930 if (!css_set_populated(cset))
1931 css_set_update_populated(cset, true);
Tejun Heo389b9c12015-10-15 16:41:51 -04001932 list_add_tail(&p->cg_list, &cset->tasks);
Tejun Heoeaf797a2014-02-25 10:04:03 -05001933 get_css_set(cset);
1934 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001935 spin_unlock(&p->sighand->siglock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001936 } while_each_thread(g, p);
1937 read_unlock(&tasklist_lock);
1938out_unlock:
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03001939 spin_unlock_irq(&css_set_lock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001940}
Paul Menageddbcc7e2007-10-18 23:39:30 -07001941
Paul Menagecc31edc2008-10-18 20:28:04 -07001942static void init_cgroup_housekeeping(struct cgroup *cgrp)
1943{
Tejun Heo2d8f2432014-04-23 11:13:15 -04001944 struct cgroup_subsys *ss;
1945 int ssid;
1946
Tejun Heod5c419b2014-05-16 13:22:48 -04001947 INIT_LIST_HEAD(&cgrp->self.sibling);
1948 INIT_LIST_HEAD(&cgrp->self.children);
Tejun Heo69d02062013-06-12 21:04:50 -07001949 INIT_LIST_HEAD(&cgrp->cset_links);
Ben Blum72a8cb32009-09-23 15:56:27 -07001950 INIT_LIST_HEAD(&cgrp->pidlists);
1951 mutex_init(&cgrp->pidlist_mutex);
Tejun Heo9d800df2014-05-14 09:15:00 -04001952 cgrp->self.cgroup = cgrp;
Tejun Heo184faf32014-05-16 13:22:51 -04001953 cgrp->self.flags |= CSS_ONLINE;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001954
1955 for_each_subsys(ss, ssid)
1956 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
Tejun Heof8f22e52014-04-23 11:13:16 -04001957
1958 init_waitqueue_head(&cgrp->offline_waitq);
Zefan Li971ff492014-09-18 16:06:19 +08001959 INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
Paul Menagecc31edc2008-10-18 20:28:04 -07001960}
Paul Menagec6d57f332009-09-23 15:56:19 -07001961
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001962static void init_cgroup_root(struct cgroup_root *root,
Tejun Heo172a2c062014-03-19 10:23:53 -04001963 struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001964{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001965 struct cgroup *cgrp = &root->cgrp;
Tejun Heob0ca5a82012-04-01 12:09:54 -07001966
Paul Menageddbcc7e2007-10-18 23:39:30 -07001967 INIT_LIST_HEAD(&root->root_list);
Tejun Heo3c9c8252014-02-12 09:29:50 -05001968 atomic_set(&root->nr_cgrps, 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001969 cgrp->root = root;
Paul Menagecc31edc2008-10-18 20:28:04 -07001970 init_cgroup_housekeeping(cgrp);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08001971 idr_init(&root->cgroup_idr);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001972
Paul Menagec6d57f332009-09-23 15:56:19 -07001973 root->flags = opts->flags;
1974 if (opts->release_agent)
1975 strcpy(root->release_agent_path, opts->release_agent);
1976 if (opts->name)
1977 strcpy(root->name, opts->name);
Tejun Heo2260e7f2012-11-19 08:13:38 -08001978 if (opts->cpuset_clone_children)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001979 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
Paul Menagec6d57f332009-09-23 15:56:19 -07001980}
1981
Tejun Heo6e5c8302016-02-22 22:25:47 -05001982static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001983{
Tejun Heod427dfe2014-02-11 11:52:48 -05001984 LIST_HEAD(tmp_links);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001985 struct cgroup *root_cgrp = &root->cgrp;
Tejun Heod427dfe2014-02-11 11:52:48 -05001986 struct css_set *cset;
Tejun Heod427dfe2014-02-11 11:52:48 -05001987 int i, ret;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001988
Tejun Heod427dfe2014-02-11 11:52:48 -05001989 lockdep_assert_held(&cgroup_mutex);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001990
Vladimir Davydovcf780b72015-08-03 15:32:26 +03001991 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
Tejun Heod427dfe2014-02-11 11:52:48 -05001992 if (ret < 0)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001993 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05001994 root_cgrp->id = ret;
Tejun Heob11cfb52015-11-20 15:55:52 -05001995 root_cgrp->ancestor_ids[0] = ret;
Paul Menagec6d57f332009-09-23 15:56:19 -07001996
Tejun Heo2aad2a82014-09-24 13:31:50 -04001997 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1998 GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04001999 if (ret)
2000 goto out;
2001
Tejun Heod427dfe2014-02-11 11:52:48 -05002002 /*
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002003 * We're accessing css_set_count without locking css_set_lock here,
Tejun Heod427dfe2014-02-11 11:52:48 -05002004 * but that's OK - it can only be increased by someone holding
Tejun Heo04313592016-03-03 09:58:01 -05002005 * cgroup_lock, and that's us. Later rebinding may disable
2006 * controllers on the default hierarchy and thus create new csets,
2007 * which can't be more than the existing ones. Allocate 2x.
Tejun Heod427dfe2014-02-11 11:52:48 -05002008 */
Tejun Heo04313592016-03-03 09:58:01 -05002009 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002010 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04002011 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002012
Tejun Heo985ed672014-03-19 10:23:53 -04002013 ret = cgroup_init_root_id(root);
Tejun Heod427dfe2014-02-11 11:52:48 -05002014 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04002015 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002016
Tejun Heo2bd59d42014-02-11 11:52:49 -05002017 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
2018 KERNFS_ROOT_CREATE_DEACTIVATED,
2019 root_cgrp);
2020 if (IS_ERR(root->kf_root)) {
2021 ret = PTR_ERR(root->kf_root);
2022 goto exit_root_id;
2023 }
2024 root_cgrp->kn = root->kf_root->kn;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002025
Tejun Heo334c3672016-03-03 09:58:01 -05002026 ret = css_populate_dir(&root_cgrp->self);
Tejun Heod427dfe2014-02-11 11:52:48 -05002027 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05002028 goto destroy_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002029
Tejun Heo5df36032014-03-19 10:23:54 -04002030 ret = rebind_subsystems(root, ss_mask);
Tejun Heod427dfe2014-02-11 11:52:48 -05002031 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05002032 goto destroy_root;
Al Viro0df6a632010-12-21 13:29:29 -05002033
Tejun Heod427dfe2014-02-11 11:52:48 -05002034 /*
2035 * There must be no failure case after here, since rebinding takes
2036 * care of subsystems' refcounts, which are explicitly dropped in
2037 * the failure exit path.
2038 */
2039 list_add(&root->root_list, &cgroup_roots);
2040 cgroup_root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002041
Tejun Heod427dfe2014-02-11 11:52:48 -05002042 /*
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002043 * Link the root cgroup in this hierarchy into all the css_set
Tejun Heod427dfe2014-02-11 11:52:48 -05002044 * objects.
2045 */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002046 spin_lock_irq(&css_set_lock);
Tejun Heo0de09422015-10-15 16:41:49 -04002047 hash_for_each(css_set_table, i, cset, hlist) {
Tejun Heod427dfe2014-02-11 11:52:48 -05002048 link_css_set(&tmp_links, cset, root_cgrp);
Tejun Heo0de09422015-10-15 16:41:49 -04002049 if (css_set_populated(cset))
2050 cgroup_update_populated(root_cgrp, true);
2051 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002052 spin_unlock_irq(&css_set_lock);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002053
Tejun Heod5c419b2014-05-16 13:22:48 -04002054 BUG_ON(!list_empty(&root_cgrp->self.children));
Tejun Heo3c9c8252014-02-12 09:29:50 -05002055 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
Tejun Heod427dfe2014-02-11 11:52:48 -05002056
Tejun Heo2bd59d42014-02-11 11:52:49 -05002057 kernfs_activate(root_cgrp->kn);
Tejun Heod427dfe2014-02-11 11:52:48 -05002058 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002059 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05002060
Tejun Heo2bd59d42014-02-11 11:52:49 -05002061destroy_root:
2062 kernfs_destroy_root(root->kf_root);
2063 root->kf_root = NULL;
2064exit_root_id:
Tejun Heod427dfe2014-02-11 11:52:48 -05002065 cgroup_exit_root_id(root);
Tejun Heo9d755d32014-05-14 09:15:02 -04002066cancel_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04002067 percpu_ref_exit(&root_cgrp->self.refcnt);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002068out:
Tejun Heod427dfe2014-02-11 11:52:48 -05002069 free_cgrp_cset_links(&tmp_links);
2070 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002071}
2072
Al Virof7e83572010-07-26 13:23:11 +04002073static struct dentry *cgroup_mount(struct file_system_type *fs_type,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002074 int flags, const char *unused_dev_name,
Al Virof7e83572010-07-26 13:23:11 +04002075 void *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002076{
Tejun Heo67e9c742015-11-16 11:13:34 -05002077 bool is_v2 = fs_type == &cgroup2_fs_type;
Li Zefan3a32bd72014-06-30 11:50:59 +08002078 struct super_block *pinned_sb = NULL;
Serge Hallyned825712016-01-29 02:54:09 -06002079 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
Li Zefan970317a2014-06-30 11:49:58 +08002080 struct cgroup_subsys *ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002081 struct cgroup_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002082 struct cgroup_sb_opts opts;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002083 struct dentry *dentry;
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002084 int ret;
Li Zefan970317a2014-06-30 11:49:58 +08002085 int i;
Li Zefanc6b3d5b2014-04-04 17:14:41 +08002086 bool new_sb;
Paul Menagec6d57f332009-09-23 15:56:19 -07002087
Serge Hallyned825712016-01-29 02:54:09 -06002088 get_cgroup_ns(ns);
2089
2090 /* Check if the caller has permission to mount. */
2091 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2092 put_cgroup_ns(ns);
2093 return ERR_PTR(-EPERM);
2094 }
2095
Paul Menagec6d57f332009-09-23 15:56:19 -07002096 /*
Tejun Heo56fde9e2014-02-13 06:58:38 -05002097 * The first time anyone tries to mount a cgroup, enable the list
2098 * linking each css_set to its tasks and fix up all existing tasks.
Paul Menagec6d57f332009-09-23 15:56:19 -07002099 */
Tejun Heo56fde9e2014-02-13 06:58:38 -05002100 if (!use_task_css_set_links)
2101 cgroup_enable_task_cg_lists();
Li Zefane37a06f2014-04-17 13:53:08 +08002102
Tejun Heo67e9c742015-11-16 11:13:34 -05002103 if (is_v2) {
2104 if (data) {
2105 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
Serge Hallyned825712016-01-29 02:54:09 -06002106 put_cgroup_ns(ns);
Tejun Heo67e9c742015-11-16 11:13:34 -05002107 return ERR_PTR(-EINVAL);
2108 }
Tejun Heoa7165262016-02-23 10:00:50 -05002109 cgrp_dfl_visible = true;
Tejun Heo67e9c742015-11-16 11:13:34 -05002110 root = &cgrp_dfl_root;
2111 cgroup_get(&root->cgrp);
2112 goto out_mount;
2113 }
2114
Tejun Heo334c3672016-03-03 09:58:01 -05002115 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002116
Paul Menageddbcc7e2007-10-18 23:39:30 -07002117 /* First find the desired set of subsystems */
Paul Menageddbcc7e2007-10-18 23:39:30 -07002118 ret = parse_cgroupfs_options(data, &opts);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002119 if (ret)
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002120 goto out_unlock;
Tejun Heoa015edd2014-05-14 09:15:00 -04002121
Li Zefan970317a2014-06-30 11:49:58 +08002122 /*
2123 * Destruction of cgroup root is asynchronous, so subsystems may
2124 * still be dying after the previous unmount. Let's drain the
2125 * dying subsystems. We just need to ensure that the ones
2126 * unmounted previously finish dying and don't care about new ones
2127 * starting. Testing ref liveliness is good enough.
2128 */
2129 for_each_subsys(ss, i) {
2130 if (!(opts.subsys_mask & (1 << i)) ||
2131 ss->root == &cgrp_dfl_root)
2132 continue;
2133
2134 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
2135 mutex_unlock(&cgroup_mutex);
2136 msleep(10);
2137 ret = restart_syscall();
2138 goto out_free;
2139 }
2140 cgroup_put(&ss->root->cgrp);
2141 }
2142
Tejun Heo985ed672014-03-19 10:23:53 -04002143 for_each_root(root) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05002144 bool name_match = false;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002145
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002146 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04002147 continue;
Paul Menagec6d57f332009-09-23 15:56:19 -07002148
Paul Menage817929e2007-10-18 23:39:36 -07002149 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05002150 * If we asked for a name then it must match. Also, if
2151 * name matches but sybsys_mask doesn't, we should fail.
2152 * Remember whether name matched.
Paul Menage817929e2007-10-18 23:39:36 -07002153 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05002154 if (opts.name) {
2155 if (strcmp(opts.name, root->name))
2156 continue;
2157 name_match = true;
2158 }
Tejun Heo31261212013-06-28 17:07:30 -07002159
2160 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05002161 * If we asked for subsystems (or explicitly for no
2162 * subsystems) then they must match.
Tejun Heo31261212013-06-28 17:07:30 -07002163 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05002164 if ((opts.subsys_mask || opts.none) &&
Tejun Heof392e512014-04-23 11:13:14 -04002165 (opts.subsys_mask != root->subsys_mask)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05002166 if (!name_match)
2167 continue;
2168 ret = -EBUSY;
2169 goto out_unlock;
2170 }
Tejun Heo873fe092013-04-14 20:15:26 -07002171
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04002172 if (root->flags ^ opts.flags)
2173 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
Tejun Heo2bd59d42014-02-11 11:52:49 -05002174
Tejun Heo776f02f2014-02-12 09:29:50 -05002175 /*
Li Zefan3a32bd72014-06-30 11:50:59 +08002176 * We want to reuse @root whose lifetime is governed by its
2177 * ->cgrp. Let's check whether @root is alive and keep it
2178 * that way. As cgroup_kill_sb() can happen anytime, we
2179 * want to block it by pinning the sb so that @root doesn't
2180 * get killed before mount is complete.
2181 *
2182 * With the sb pinned, tryget_live can reliably indicate
2183 * whether @root can be reused. If it's being killed,
2184 * drain it. We can use wait_queue for the wait but this
2185 * path is super cold. Let's just sleep a bit and retry.
Tejun Heo776f02f2014-02-12 09:29:50 -05002186 */
Li Zefan3a32bd72014-06-30 11:50:59 +08002187 pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
2188 if (IS_ERR(pinned_sb) ||
2189 !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
Tejun Heo776f02f2014-02-12 09:29:50 -05002190 mutex_unlock(&cgroup_mutex);
Li Zefan3a32bd72014-06-30 11:50:59 +08002191 if (!IS_ERR_OR_NULL(pinned_sb))
2192 deactivate_super(pinned_sb);
Tejun Heo776f02f2014-02-12 09:29:50 -05002193 msleep(10);
Tejun Heoa015edd2014-05-14 09:15:00 -04002194 ret = restart_syscall();
2195 goto out_free;
Tejun Heo776f02f2014-02-12 09:29:50 -05002196 }
2197
2198 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002199 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002200 }
2201
Tejun Heo172a2c062014-03-19 10:23:53 -04002202 /*
2203 * No such thing, create a new one. name= matching without subsys
2204 * specification is allowed for already existing hierarchies but we
2205 * can't create new one without subsys specification.
2206 */
2207 if (!opts.subsys_mask && !opts.none) {
2208 ret = -EINVAL;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002209 goto out_unlock;
2210 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07002211
Eric W. Biederman726a4992016-07-15 06:36:44 -05002212 /* Hierarchies may only be created in the initial cgroup namespace. */
2213 if (ns != &init_cgroup_ns) {
Serge Hallyned825712016-01-29 02:54:09 -06002214 ret = -EPERM;
2215 goto out_unlock;
2216 }
2217
Tejun Heo172a2c062014-03-19 10:23:53 -04002218 root = kzalloc(sizeof(*root), GFP_KERNEL);
2219 if (!root) {
2220 ret = -ENOMEM;
2221 goto out_unlock;
2222 }
2223
2224 init_cgroup_root(root, &opts);
2225
Tejun Heo35585572014-02-13 06:58:38 -05002226 ret = cgroup_setup_root(root, opts.subsys_mask);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002227 if (ret)
2228 cgroup_free_root(root);
2229
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002230out_unlock:
Tejun Heoe25e2cbb2011-12-12 18:12:21 -08002231 mutex_unlock(&cgroup_mutex);
Tejun Heoa015edd2014-05-14 09:15:00 -04002232out_free:
Paul Menagec6d57f332009-09-23 15:56:19 -07002233 kfree(opts.release_agent);
2234 kfree(opts.name);
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002235
Serge Hallyned825712016-01-29 02:54:09 -06002236 if (ret) {
2237 put_cgroup_ns(ns);
Tejun Heo8e30e2b82014-02-11 11:52:48 -05002238 return ERR_PTR(ret);
Serge Hallyned825712016-01-29 02:54:09 -06002239 }
Tejun Heo67e9c742015-11-16 11:13:34 -05002240out_mount:
Jianyu Zhanc9482a52014-04-26 15:40:28 +08002241 dentry = kernfs_mount(fs_type, flags, root->kf_root,
Tejun Heo67e9c742015-11-16 11:13:34 -05002242 is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC,
2243 &new_sb);
Serge Hallyned825712016-01-29 02:54:09 -06002244
2245 /*
2246 * In non-init cgroup namespace, instead of root cgroup's
2247 * dentry, we return the dentry corresponding to the
2248 * cgroupns->root_cgrp.
2249 */
2250 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
2251 struct dentry *nsdentry;
2252 struct cgroup *cgrp;
2253
2254 mutex_lock(&cgroup_mutex);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002255 spin_lock_irq(&css_set_lock);
Serge Hallyned825712016-01-29 02:54:09 -06002256
2257 cgrp = cset_cgroup_from_root(ns->root_cset, root);
2258
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002259 spin_unlock_irq(&css_set_lock);
Serge Hallyned825712016-01-29 02:54:09 -06002260 mutex_unlock(&cgroup_mutex);
2261
2262 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
2263 dput(dentry);
2264 dentry = nsdentry;
2265 }
2266
Li Zefanc6b3d5b2014-04-04 17:14:41 +08002267 if (IS_ERR(dentry) || !new_sb)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002268 cgroup_put(&root->cgrp);
Li Zefan3a32bd72014-06-30 11:50:59 +08002269
2270 /*
2271 * If @pinned_sb, we're reusing an existing root and holding an
2272 * extra ref on its sb. Mount is complete. Put the extra ref.
2273 */
2274 if (pinned_sb) {
2275 WARN_ON(new_sb);
2276 deactivate_super(pinned_sb);
2277 }
2278
Serge Hallyned825712016-01-29 02:54:09 -06002279 put_cgroup_ns(ns);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002280 return dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002281}
2282
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09002283static void cgroup_kill_sb(struct super_block *sb)
2284{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002285 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002286 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002287
Tejun Heo9d755d32014-05-14 09:15:02 -04002288 /*
2289 * If @root doesn't have any mounts or children, start killing it.
2290 * This prevents new mounts by disabling percpu_ref_tryget_live().
2291 * cgroup_mount() may wait for @root's release.
Li Zefan1f779fb2014-06-04 16:48:15 +08002292 *
2293 * And don't kill the default root.
Tejun Heo9d755d32014-05-14 09:15:02 -04002294 */
Johannes Weiner3c606d32015-01-22 10:19:43 -05002295 if (!list_empty(&root->cgrp.self.children) ||
Li Zefan1f779fb2014-06-04 16:48:15 +08002296 root == &cgrp_dfl_root)
Tejun Heo9d755d32014-05-14 09:15:02 -04002297 cgroup_put(&root->cgrp);
2298 else
2299 percpu_ref_kill(&root->cgrp.self.refcnt);
2300
Tejun Heo2bd59d42014-02-11 11:52:49 -05002301 kernfs_kill_sb(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002302}
2303
2304static struct file_system_type cgroup_fs_type = {
2305 .name = "cgroup",
Al Virof7e83572010-07-26 13:23:11 +04002306 .mount = cgroup_mount,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002307 .kill_sb = cgroup_kill_sb,
Serge Hallyn1c537532016-01-29 02:54:11 -06002308 .fs_flags = FS_USERNS_MOUNT,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002309};
2310
Tejun Heo67e9c742015-11-16 11:13:34 -05002311static struct file_system_type cgroup2_fs_type = {
2312 .name = "cgroup2",
2313 .mount = cgroup_mount,
2314 .kill_sb = cgroup_kill_sb,
Serge Hallyn1c537532016-01-29 02:54:11 -06002315 .fs_flags = FS_USERNS_MOUNT,
Tejun Heo67e9c742015-11-16 11:13:34 -05002316};
2317
Aditya Kalia79a9082016-01-29 02:54:06 -06002318static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2319 struct cgroup_namespace *ns)
2320{
2321 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2322 int ret;
2323
2324 ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2325 if (ret < 0 || ret >= buflen)
2326 return NULL;
2327 return buf;
2328}
2329
2330char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2331 struct cgroup_namespace *ns)
2332{
2333 char *ret;
2334
2335 mutex_lock(&cgroup_mutex);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002336 spin_lock_irq(&css_set_lock);
Aditya Kalia79a9082016-01-29 02:54:06 -06002337
2338 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2339
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002340 spin_unlock_irq(&css_set_lock);
Aditya Kalia79a9082016-01-29 02:54:06 -06002341 mutex_unlock(&cgroup_mutex);
2342
2343 return ret;
2344}
2345EXPORT_SYMBOL_GPL(cgroup_path_ns);
2346
Li Zefana043e3b2008-02-23 15:24:09 -08002347/**
Tejun Heo913ffdb2013-07-11 16:34:48 -07002348 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
Tejun Heo857a2be2013-04-14 20:50:08 -07002349 * @task: target task
Tejun Heo857a2be2013-04-14 20:50:08 -07002350 * @buf: the buffer to write the path into
2351 * @buflen: the length of the buffer
2352 *
Tejun Heo913ffdb2013-07-11 16:34:48 -07002353 * Determine @task's cgroup on the first (the one with the lowest non-zero
2354 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2355 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2356 * cgroup controller callbacks.
2357 *
Tejun Heoe61734c2014-02-12 09:29:50 -05002358 * Return value is the same as kernfs_path().
Tejun Heo857a2be2013-04-14 20:50:08 -07002359 */
Tejun Heoe61734c2014-02-12 09:29:50 -05002360char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
Tejun Heo857a2be2013-04-14 20:50:08 -07002361{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002362 struct cgroup_root *root;
Tejun Heo913ffdb2013-07-11 16:34:48 -07002363 struct cgroup *cgrp;
Tejun Heoe61734c2014-02-12 09:29:50 -05002364 int hierarchy_id = 1;
2365 char *path = NULL;
Tejun Heo857a2be2013-04-14 20:50:08 -07002366
2367 mutex_lock(&cgroup_mutex);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002368 spin_lock_irq(&css_set_lock);
Tejun Heo857a2be2013-04-14 20:50:08 -07002369
Tejun Heo913ffdb2013-07-11 16:34:48 -07002370 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2371
Tejun Heo857a2be2013-04-14 20:50:08 -07002372 if (root) {
2373 cgrp = task_cgroup_from_root(task, root);
Aditya Kalia79a9082016-01-29 02:54:06 -06002374 path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
Tejun Heo913ffdb2013-07-11 16:34:48 -07002375 } else {
2376 /* if no hierarchy exists, everyone is in "/" */
Tejun Heoe61734c2014-02-12 09:29:50 -05002377 if (strlcpy(buf, "/", buflen) < buflen)
2378 path = buf;
Tejun Heo857a2be2013-04-14 20:50:08 -07002379 }
2380
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002381 spin_unlock_irq(&css_set_lock);
Tejun Heo857a2be2013-04-14 20:50:08 -07002382 mutex_unlock(&cgroup_mutex);
Tejun Heoe61734c2014-02-12 09:29:50 -05002383 return path;
Tejun Heo857a2be2013-04-14 20:50:08 -07002384}
Tejun Heo913ffdb2013-07-11 16:34:48 -07002385EXPORT_SYMBOL_GPL(task_cgroup_path);
Tejun Heo857a2be2013-04-14 20:50:08 -07002386
Tejun Heob3dc0942014-02-25 10:04:01 -05002387/* used to track tasks and other necessary states during migration */
Tejun Heo2f7ee562011-12-12 18:12:21 -08002388struct cgroup_taskset {
Tejun Heob3dc0942014-02-25 10:04:01 -05002389 /* the src and dst cset list running through cset->mg_node */
2390 struct list_head src_csets;
2391 struct list_head dst_csets;
2392
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002393 /* the subsys currently being processed */
2394 int ssid;
2395
Tejun Heob3dc0942014-02-25 10:04:01 -05002396 /*
2397 * Fields for cgroup_taskset_*() iteration.
2398 *
2399 * Before migration is committed, the target migration tasks are on
2400 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2401 * the csets on ->dst_csets. ->csets point to either ->src_csets
2402 * or ->dst_csets depending on whether migration is committed.
2403 *
2404 * ->cur_csets and ->cur_task point to the current task position
2405 * during iteration.
2406 */
2407 struct list_head *csets;
2408 struct css_set *cur_cset;
2409 struct task_struct *cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002410};
2411
Tejun Heoadaae5d2015-09-11 15:00:21 -04002412#define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2413 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2414 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2415 .csets = &tset.src_csets, \
2416}
2417
2418/**
2419 * cgroup_taskset_add - try to add a migration target task to a taskset
2420 * @task: target task
2421 * @tset: target taskset
2422 *
2423 * Add @task, which is a migration target, to @tset. This function becomes
2424 * noop if @task doesn't need to be migrated. @task's css_set should have
2425 * been added as a migration source and @task->cg_list will be moved from
2426 * the css_set's tasks list to mg_tasks one.
2427 */
2428static void cgroup_taskset_add(struct task_struct *task,
2429 struct cgroup_taskset *tset)
2430{
2431 struct css_set *cset;
2432
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002433 lockdep_assert_held(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002434
2435 /* @task either already exited or can't exit until the end */
2436 if (task->flags & PF_EXITING)
2437 return;
2438
2439 /* leave @task alone if post_fork() hasn't linked it yet */
2440 if (list_empty(&task->cg_list))
2441 return;
2442
2443 cset = task_css_set(task);
2444 if (!cset->mg_src_cgrp)
2445 return;
2446
2447 list_move_tail(&task->cg_list, &cset->mg_tasks);
2448 if (list_empty(&cset->mg_node))
2449 list_add_tail(&cset->mg_node, &tset->src_csets);
2450 if (list_empty(&cset->mg_dst_cset->mg_node))
2451 list_move_tail(&cset->mg_dst_cset->mg_node,
2452 &tset->dst_csets);
2453}
2454
Tejun Heo2f7ee562011-12-12 18:12:21 -08002455/**
2456 * cgroup_taskset_first - reset taskset and return the first task
2457 * @tset: taskset of interest
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002458 * @dst_cssp: output variable for the destination css
Tejun Heo2f7ee562011-12-12 18:12:21 -08002459 *
2460 * @tset iteration is initialized and the first task is returned.
2461 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002462struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2463 struct cgroup_subsys_state **dst_cssp)
Tejun Heo2f7ee562011-12-12 18:12:21 -08002464{
Tejun Heob3dc0942014-02-25 10:04:01 -05002465 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2466 tset->cur_task = NULL;
2467
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002468 return cgroup_taskset_next(tset, dst_cssp);
Tejun Heo2f7ee562011-12-12 18:12:21 -08002469}
Tejun Heo2f7ee562011-12-12 18:12:21 -08002470
2471/**
2472 * cgroup_taskset_next - iterate to the next task in taskset
2473 * @tset: taskset of interest
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002474 * @dst_cssp: output variable for the destination css
Tejun Heo2f7ee562011-12-12 18:12:21 -08002475 *
2476 * Return the next task in @tset. Iteration must have been initialized
2477 * with cgroup_taskset_first().
2478 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002479struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2480 struct cgroup_subsys_state **dst_cssp)
Tejun Heo2f7ee562011-12-12 18:12:21 -08002481{
Tejun Heob3dc0942014-02-25 10:04:01 -05002482 struct css_set *cset = tset->cur_cset;
2483 struct task_struct *task = tset->cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002484
Tejun Heob3dc0942014-02-25 10:04:01 -05002485 while (&cset->mg_node != tset->csets) {
2486 if (!task)
2487 task = list_first_entry(&cset->mg_tasks,
2488 struct task_struct, cg_list);
2489 else
2490 task = list_next_entry(task, cg_list);
Tejun Heo2f7ee562011-12-12 18:12:21 -08002491
Tejun Heob3dc0942014-02-25 10:04:01 -05002492 if (&task->cg_list != &cset->mg_tasks) {
2493 tset->cur_cset = cset;
2494 tset->cur_task = task;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002495
2496 /*
2497 * This function may be called both before and
2498 * after cgroup_taskset_migrate(). The two cases
2499 * can be distinguished by looking at whether @cset
2500 * has its ->mg_dst_cset set.
2501 */
2502 if (cset->mg_dst_cset)
2503 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2504 else
2505 *dst_cssp = cset->subsys[tset->ssid];
2506
Tejun Heob3dc0942014-02-25 10:04:01 -05002507 return task;
2508 }
2509
2510 cset = list_next_entry(cset, mg_node);
2511 task = NULL;
2512 }
2513
2514 return NULL;
Tejun Heo2f7ee562011-12-12 18:12:21 -08002515}
Tejun Heo2f7ee562011-12-12 18:12:21 -08002516
2517/**
Tejun Heo37ff9f82016-03-08 11:51:26 -05002518 * cgroup_taskset_migrate - migrate a taskset
Tejun Heoadaae5d2015-09-11 15:00:21 -04002519 * @tset: taget taskset
Tejun Heo37ff9f82016-03-08 11:51:26 -05002520 * @root: cgroup root the migration is taking place on
Tejun Heoadaae5d2015-09-11 15:00:21 -04002521 *
Tejun Heo37ff9f82016-03-08 11:51:26 -05002522 * Migrate tasks in @tset as setup by migration preparation functions.
2523 * This function fails iff one of the ->can_attach callbacks fails and
2524 * guarantees that either all or none of the tasks in @tset are migrated.
2525 * @tset is consumed regardless of success.
Tejun Heoadaae5d2015-09-11 15:00:21 -04002526 */
2527static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
Tejun Heo37ff9f82016-03-08 11:51:26 -05002528 struct cgroup_root *root)
Tejun Heoadaae5d2015-09-11 15:00:21 -04002529{
Tejun Heo37ff9f82016-03-08 11:51:26 -05002530 struct cgroup_subsys *ss;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002531 struct task_struct *task, *tmp_task;
2532 struct css_set *cset, *tmp_cset;
Tejun Heo37ff9f82016-03-08 11:51:26 -05002533 int ssid, failed_ssid, ret;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002534
2535 /* methods shouldn't be called if no task is actually migrating */
2536 if (list_empty(&tset->src_csets))
2537 return 0;
2538
2539 /* check that we can legitimately attach to the cgroup */
Tejun Heo37ff9f82016-03-08 11:51:26 -05002540 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2541 if (ss->can_attach) {
2542 tset->ssid = ssid;
2543 ret = ss->can_attach(tset);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002544 if (ret) {
Tejun Heo37ff9f82016-03-08 11:51:26 -05002545 failed_ssid = ssid;
Tejun Heoadaae5d2015-09-11 15:00:21 -04002546 goto out_cancel_attach;
2547 }
2548 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002549 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002550
2551 /*
2552 * Now that we're guaranteed success, proceed to move all tasks to
2553 * the new cgroup. There are no failure cases after here, so this
2554 * is the commit point.
2555 */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002556 spin_lock_irq(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002557 list_for_each_entry(cset, &tset->src_csets, mg_node) {
Tejun Heof6d7d042015-10-15 16:41:52 -04002558 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2559 struct css_set *from_cset = task_css_set(task);
2560 struct css_set *to_cset = cset->mg_dst_cset;
2561
2562 get_css_set(to_cset);
2563 css_set_move_task(task, from_cset, to_cset, true);
2564 put_css_set_locked(from_cset);
2565 }
Tejun Heoadaae5d2015-09-11 15:00:21 -04002566 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002567 spin_unlock_irq(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002568
2569 /*
2570 * Migration is committed, all target tasks are now on dst_csets.
2571 * Nothing is sensitive to fork() after this point. Notify
2572 * controllers that migration is complete.
2573 */
2574 tset->csets = &tset->dst_csets;
2575
Tejun Heo37ff9f82016-03-08 11:51:26 -05002576 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2577 if (ss->attach) {
2578 tset->ssid = ssid;
2579 ss->attach(tset);
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002580 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002581 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002582
2583 ret = 0;
2584 goto out_release_tset;
2585
2586out_cancel_attach:
Tejun Heo37ff9f82016-03-08 11:51:26 -05002587 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2588 if (ssid == failed_ssid)
Tejun Heoadaae5d2015-09-11 15:00:21 -04002589 break;
Tejun Heo37ff9f82016-03-08 11:51:26 -05002590 if (ss->cancel_attach) {
2591 tset->ssid = ssid;
2592 ss->cancel_attach(tset);
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05002593 }
Tejun Heo37ff9f82016-03-08 11:51:26 -05002594 } while_each_subsys_mask();
Tejun Heoadaae5d2015-09-11 15:00:21 -04002595out_release_tset:
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002596 spin_lock_irq(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002597 list_splice_init(&tset->dst_csets, &tset->src_csets);
2598 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2599 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2600 list_del_init(&cset->mg_node);
2601 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002602 spin_unlock_irq(&css_set_lock);
Tejun Heoadaae5d2015-09-11 15:00:21 -04002603 return ret;
2604}
2605
2606/**
Tejun Heo6c694c82016-03-08 11:51:25 -05002607 * cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2608 * @dst_cgrp: destination cgroup to test
2609 *
2610 * On the default hierarchy, except for the root, subtree_control must be
2611 * zero for migration destination cgroups with tasks so that child cgroups
2612 * don't compete against tasks.
2613 */
2614static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2615{
2616 return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) ||
2617 !dst_cgrp->subtree_control;
2618}
2619
2620/**
Tejun Heo1958d2d2014-02-25 10:04:03 -05002621 * cgroup_migrate_finish - cleanup after attach
2622 * @preloaded_csets: list of preloaded css_sets
Ben Blum74a11662011-05-26 16:25:20 -07002623 *
Tejun Heo1958d2d2014-02-25 10:04:03 -05002624 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2625 * those functions for details.
Ben Blum74a11662011-05-26 16:25:20 -07002626 */
Tejun Heo1958d2d2014-02-25 10:04:03 -05002627static void cgroup_migrate_finish(struct list_head *preloaded_csets)
Ben Blum74a11662011-05-26 16:25:20 -07002628{
Tejun Heo1958d2d2014-02-25 10:04:03 -05002629 struct css_set *cset, *tmp_cset;
2630
2631 lockdep_assert_held(&cgroup_mutex);
2632
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002633 spin_lock_irq(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002634 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2635 cset->mg_src_cgrp = NULL;
Tejun Heoe4857982016-03-08 11:51:26 -05002636 cset->mg_dst_cgrp = NULL;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002637 cset->mg_dst_cset = NULL;
2638 list_del_init(&cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002639 put_css_set_locked(cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002640 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002641 spin_unlock_irq(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002642}
2643
2644/**
2645 * cgroup_migrate_add_src - add a migration source css_set
2646 * @src_cset: the source css_set to add
2647 * @dst_cgrp: the destination cgroup
2648 * @preloaded_csets: list of preloaded css_sets
2649 *
2650 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2651 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2652 * up by cgroup_migrate_finish().
2653 *
Tejun Heo1ed13282015-09-16 12:53:17 -04002654 * This function may be called without holding cgroup_threadgroup_rwsem
2655 * even if the target is a process. Threads may be created and destroyed
2656 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2657 * into play and the preloaded css_sets are guaranteed to cover all
2658 * migrations.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002659 */
2660static void cgroup_migrate_add_src(struct css_set *src_cset,
2661 struct cgroup *dst_cgrp,
2662 struct list_head *preloaded_csets)
2663{
2664 struct cgroup *src_cgrp;
2665
2666 lockdep_assert_held(&cgroup_mutex);
Tejun Heof0d9a5f2015-10-15 16:41:53 -04002667 lockdep_assert_held(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002668
Tejun Heo2b021cb2016-03-15 20:43:04 -04002669 /*
2670 * If ->dead, @src_set is associated with one or more dead cgroups
2671 * and doesn't contain any migratable tasks. Ignore it early so
2672 * that the rest of migration path doesn't get confused by it.
2673 */
2674 if (src_cset->dead)
2675 return;
2676
Tejun Heo1958d2d2014-02-25 10:04:03 -05002677 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2678
Tejun Heo1958d2d2014-02-25 10:04:03 -05002679 if (!list_empty(&src_cset->mg_preload_node))
2680 return;
2681
2682 WARN_ON(src_cset->mg_src_cgrp);
Tejun Heoe4857982016-03-08 11:51:26 -05002683 WARN_ON(src_cset->mg_dst_cgrp);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002684 WARN_ON(!list_empty(&src_cset->mg_tasks));
2685 WARN_ON(!list_empty(&src_cset->mg_node));
2686
2687 src_cset->mg_src_cgrp = src_cgrp;
Tejun Heoe4857982016-03-08 11:51:26 -05002688 src_cset->mg_dst_cgrp = dst_cgrp;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002689 get_css_set(src_cset);
2690 list_add(&src_cset->mg_preload_node, preloaded_csets);
2691}
2692
2693/**
2694 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
Tejun Heo1958d2d2014-02-25 10:04:03 -05002695 * @preloaded_csets: list of preloaded source css_sets
2696 *
Tejun Heoe4857982016-03-08 11:51:26 -05002697 * Tasks are about to be moved and all the source css_sets have been
2698 * preloaded to @preloaded_csets. This function looks up and pins all
2699 * destination css_sets, links each to its source, and append them to
2700 * @preloaded_csets.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002701 *
2702 * This function must be called after cgroup_migrate_add_src() has been
2703 * called on each migration source css_set. After migration is performed
2704 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2705 * @preloaded_csets.
2706 */
Tejun Heoe4857982016-03-08 11:51:26 -05002707static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets)
Tejun Heo1958d2d2014-02-25 10:04:03 -05002708{
2709 LIST_HEAD(csets);
Tejun Heof817de92014-04-23 11:13:16 -04002710 struct css_set *src_cset, *tmp_cset;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002711
2712 lockdep_assert_held(&cgroup_mutex);
2713
2714 /* look up the dst cset for each src cset and link it to src */
Tejun Heof817de92014-04-23 11:13:16 -04002715 list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
Tejun Heo1958d2d2014-02-25 10:04:03 -05002716 struct css_set *dst_cset;
2717
Tejun Heoe4857982016-03-08 11:51:26 -05002718 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002719 if (!dst_cset)
2720 goto err;
2721
2722 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002723
2724 /*
2725 * If src cset equals dst, it's noop. Drop the src.
2726 * cgroup_migrate() will skip the cset too. Note that we
2727 * can't handle src == dst as some nodes are used by both.
2728 */
2729 if (src_cset == dst_cset) {
2730 src_cset->mg_src_cgrp = NULL;
Tejun Heoe4857982016-03-08 11:51:26 -05002731 src_cset->mg_dst_cgrp = NULL;
Tejun Heof817de92014-04-23 11:13:16 -04002732 list_del_init(&src_cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002733 put_css_set(src_cset);
2734 put_css_set(dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002735 continue;
2736 }
2737
Tejun Heo1958d2d2014-02-25 10:04:03 -05002738 src_cset->mg_dst_cset = dst_cset;
2739
2740 if (list_empty(&dst_cset->mg_preload_node))
2741 list_add(&dst_cset->mg_preload_node, &csets);
2742 else
Zefan Lia25eb522014-09-19 16:51:00 +08002743 put_css_set(dst_cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002744 }
2745
Tejun Heof817de92014-04-23 11:13:16 -04002746 list_splice_tail(&csets, preloaded_csets);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002747 return 0;
2748err:
2749 cgroup_migrate_finish(&csets);
2750 return -ENOMEM;
2751}
2752
2753/**
2754 * cgroup_migrate - migrate a process or task to a cgroup
Tejun Heo1958d2d2014-02-25 10:04:03 -05002755 * @leader: the leader of the process or the task to migrate
2756 * @threadgroup: whether @leader points to the whole process or a single task
Tejun Heo37ff9f82016-03-08 11:51:26 -05002757 * @root: cgroup root migration is taking place on
Tejun Heo1958d2d2014-02-25 10:04:03 -05002758 *
Tejun Heo37ff9f82016-03-08 11:51:26 -05002759 * Migrate a process or task denoted by @leader. If migrating a process,
2760 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2761 * responsible for invoking cgroup_migrate_add_src() and
Tejun Heo1958d2d2014-02-25 10:04:03 -05002762 * cgroup_migrate_prepare_dst() on the targets before invoking this
2763 * function and following up with cgroup_migrate_finish().
2764 *
2765 * As long as a controller's ->can_attach() doesn't fail, this function is
2766 * guaranteed to succeed. This means that, excluding ->can_attach()
2767 * failure, when migrating multiple targets, the success or failure can be
2768 * decided for all targets by invoking group_migrate_prepare_dst() before
2769 * actually starting migrating.
2770 */
Tejun Heo9af2ec42015-09-11 15:00:20 -04002771static int cgroup_migrate(struct task_struct *leader, bool threadgroup,
Tejun Heo37ff9f82016-03-08 11:51:26 -05002772 struct cgroup_root *root)
Ben Blum74a11662011-05-26 16:25:20 -07002773{
Tejun Heoadaae5d2015-09-11 15:00:21 -04002774 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2775 struct task_struct *task;
Ben Blum74a11662011-05-26 16:25:20 -07002776
2777 /*
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002778 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2779 * already PF_EXITING could be freed from underneath us unless we
2780 * take an rcu_read_lock.
2781 */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002782 spin_lock_irq(&css_set_lock);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002783 rcu_read_lock();
Tejun Heo9db8de32014-02-13 06:58:43 -05002784 task = leader;
Ben Blum74a11662011-05-26 16:25:20 -07002785 do {
Tejun Heoadaae5d2015-09-11 15:00:21 -04002786 cgroup_taskset_add(task, &tset);
Li Zefan081aa452013-03-13 09:17:09 +08002787 if (!threadgroup)
2788 break;
Tejun Heo9db8de32014-02-13 06:58:43 -05002789 } while_each_thread(leader, task);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002790 rcu_read_unlock();
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002791 spin_unlock_irq(&css_set_lock);
Ben Blum74a11662011-05-26 16:25:20 -07002792
Tejun Heo37ff9f82016-03-08 11:51:26 -05002793 return cgroup_taskset_migrate(&tset, root);
Ben Blum74a11662011-05-26 16:25:20 -07002794}
2795
Tejun Heo1958d2d2014-02-25 10:04:03 -05002796/**
2797 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2798 * @dst_cgrp: the cgroup to attach to
2799 * @leader: the task or the leader of the threadgroup to be attached
2800 * @threadgroup: attach the whole threadgroup?
2801 *
Tejun Heo1ed13282015-09-16 12:53:17 -04002802 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002803 */
2804static int cgroup_attach_task(struct cgroup *dst_cgrp,
2805 struct task_struct *leader, bool threadgroup)
2806{
2807 LIST_HEAD(preloaded_csets);
2808 struct task_struct *task;
2809 int ret;
2810
Tejun Heo6c694c82016-03-08 11:51:25 -05002811 if (!cgroup_may_migrate_to(dst_cgrp))
2812 return -EBUSY;
2813
Tejun Heo1958d2d2014-02-25 10:04:03 -05002814 /* look up all src csets */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002815 spin_lock_irq(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002816 rcu_read_lock();
2817 task = leader;
2818 do {
2819 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2820 &preloaded_csets);
2821 if (!threadgroup)
2822 break;
2823 } while_each_thread(leader, task);
2824 rcu_read_unlock();
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002825 spin_unlock_irq(&css_set_lock);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002826
2827 /* prepare dst csets and commit */
Tejun Heoe4857982016-03-08 11:51:26 -05002828 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002829 if (!ret)
Tejun Heo37ff9f82016-03-08 11:51:26 -05002830 ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002831
2832 cgroup_migrate_finish(&preloaded_csets);
2833 return ret;
Ben Blum74a11662011-05-26 16:25:20 -07002834}
2835
Tejun Heo187fe842015-06-18 16:54:28 -04002836static int cgroup_procs_write_permission(struct task_struct *task,
2837 struct cgroup *dst_cgrp,
2838 struct kernfs_open_file *of)
Tejun Heodedf22e2015-06-18 16:54:28 -04002839{
2840 const struct cred *cred = current_cred();
2841 const struct cred *tcred = get_task_cred(task);
2842 int ret = 0;
2843
2844 /*
2845 * even if we're attaching all tasks in the thread group, we only
2846 * need to check permissions on one of them.
2847 */
2848 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2849 !uid_eq(cred->euid, tcred->uid) &&
2850 !uid_eq(cred->euid, tcred->suid))
2851 ret = -EACCES;
2852
Tejun Heo187fe842015-06-18 16:54:28 -04002853 if (!ret && cgroup_on_dfl(dst_cgrp)) {
2854 struct super_block *sb = of->file->f_path.dentry->d_sb;
2855 struct cgroup *cgrp;
2856 struct inode *inode;
2857
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002858 spin_lock_irq(&css_set_lock);
Tejun Heo187fe842015-06-18 16:54:28 -04002859 cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002860 spin_unlock_irq(&css_set_lock);
Tejun Heo187fe842015-06-18 16:54:28 -04002861
2862 while (!cgroup_is_descendant(dst_cgrp, cgrp))
2863 cgrp = cgroup_parent(cgrp);
2864
2865 ret = -ENOMEM;
Tejun Heo6f60ead2015-09-18 17:54:23 -04002866 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
Tejun Heo187fe842015-06-18 16:54:28 -04002867 if (inode) {
2868 ret = inode_permission(inode, MAY_WRITE);
2869 iput(inode);
2870 }
2871 }
2872
Tejun Heodedf22e2015-06-18 16:54:28 -04002873 put_cred(tcred);
2874 return ret;
2875}
2876
Ben Blum74a11662011-05-26 16:25:20 -07002877/*
2878 * Find the task_struct of the task to attach by vpid and pass it along to the
Tejun Heocd3d0952011-12-12 18:12:21 -08002879 * function to attach either it or all tasks in its threadgroup. Will lock
Tejun Heo0e1d7682014-02-25 10:04:03 -05002880 * cgroup_mutex and threadgroup.
Ben Blum74a11662011-05-26 16:25:20 -07002881 */
Tejun Heoacbef752014-05-13 12:16:22 -04002882static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2883 size_t nbytes, loff_t off, bool threadgroup)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002884{
Paul Menagebbcb81d2007-10-18 23:39:32 -07002885 struct task_struct *tsk;
Tejun Heo5cf1cac2016-04-21 19:06:48 -04002886 struct cgroup_subsys *ss;
Tejun Heoe76ecae2014-05-13 12:19:23 -04002887 struct cgroup *cgrp;
Tejun Heoacbef752014-05-13 12:16:22 -04002888 pid_t pid;
Tejun Heo5cf1cac2016-04-21 19:06:48 -04002889 int ssid, ret;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002890
Tejun Heoacbef752014-05-13 12:16:22 -04002891 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2892 return -EINVAL;
2893
Tejun Heo945ba192016-03-03 09:58:00 -05002894 cgrp = cgroup_kn_lock_live(of->kn, false);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002895 if (!cgrp)
Ben Blum74a11662011-05-26 16:25:20 -07002896 return -ENODEV;
2897
Tejun Heo3014dde2015-09-16 13:03:02 -04002898 percpu_down_write(&cgroup_threadgroup_rwsem);
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002899 rcu_read_lock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07002900 if (pid) {
Pavel Emelyanov73507f32008-02-07 00:14:47 -08002901 tsk = find_task_by_vpid(pid);
Ben Blum74a11662011-05-26 16:25:20 -07002902 if (!tsk) {
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09002903 ret = -ESRCH;
Tejun Heo3014dde2015-09-16 13:03:02 -04002904 goto out_unlock_rcu;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002905 }
Tejun Heodedf22e2015-06-18 16:54:28 -04002906 } else {
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002907 tsk = current;
Tejun Heodedf22e2015-06-18 16:54:28 -04002908 }
Tejun Heocd3d0952011-12-12 18:12:21 -08002909
2910 if (threadgroup)
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002911 tsk = tsk->group_leader;
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002912
2913 /*
Tejun Heo14a40ff2013-03-19 13:45:20 -07002914 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002915 * trapped in a cpuset, or RT worker may be born in a cgroup
2916 * with no rt_runtime allocated. Just say no.
2917 */
Tejun Heo14a40ff2013-03-19 13:45:20 -07002918 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002919 ret = -EINVAL;
Tejun Heo3014dde2015-09-16 13:03:02 -04002920 goto out_unlock_rcu;
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002921 }
2922
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002923 get_task_struct(tsk);
2924 rcu_read_unlock();
Tejun Heocd3d0952011-12-12 18:12:21 -08002925
Tejun Heo187fe842015-06-18 16:54:28 -04002926 ret = cgroup_procs_write_permission(tsk, cgrp, of);
Tejun Heodedf22e2015-06-18 16:54:28 -04002927 if (!ret)
2928 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
Li Zefan081aa452013-03-13 09:17:09 +08002929
Tejun Heof9f9e7b2015-09-16 11:51:12 -04002930 put_task_struct(tsk);
Tejun Heo3014dde2015-09-16 13:03:02 -04002931 goto out_unlock_threadgroup;
2932
2933out_unlock_rcu:
2934 rcu_read_unlock();
2935out_unlock_threadgroup:
2936 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heo5cf1cac2016-04-21 19:06:48 -04002937 for_each_subsys(ss, ssid)
2938 if (ss->post_attach)
2939 ss->post_attach();
Tejun Heoe76ecae2014-05-13 12:19:23 -04002940 cgroup_kn_unlock(of->kn);
Tejun Heoacbef752014-05-13 12:16:22 -04002941 return ret ?: nbytes;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002942}
2943
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002944/**
2945 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2946 * @from: attach to all cgroups of a given task
2947 * @tsk: the task to be attached
2948 */
2949int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2950{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002951 struct cgroup_root *root;
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002952 int retval = 0;
2953
Tejun Heo47cfcd02013-04-07 09:29:51 -07002954 mutex_lock(&cgroup_mutex);
Eric W. Biedermaneedd0f42016-07-15 06:35:51 -05002955 percpu_down_write(&cgroup_threadgroup_rwsem);
Tejun Heo985ed672014-03-19 10:23:53 -04002956 for_each_root(root) {
Tejun Heo96d365e2014-02-13 06:58:40 -05002957 struct cgroup *from_cgrp;
2958
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002959 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04002960 continue;
2961
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002962 spin_lock_irq(&css_set_lock);
Tejun Heo96d365e2014-02-13 06:58:40 -05002963 from_cgrp = task_cgroup_from_root(from, root);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03002964 spin_unlock_irq(&css_set_lock);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002965
Li Zefan6f4b7e62013-07-31 16:18:36 +08002966 retval = cgroup_attach_task(from_cgrp, tsk, false);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002967 if (retval)
2968 break;
2969 }
Eric W. Biedermaneedd0f42016-07-15 06:35:51 -05002970 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heo47cfcd02013-04-07 09:29:51 -07002971 mutex_unlock(&cgroup_mutex);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002972
2973 return retval;
2974}
2975EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2976
Tejun Heoacbef752014-05-13 12:16:22 -04002977static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2978 char *buf, size_t nbytes, loff_t off)
Paul Menageaf351022008-07-25 01:47:01 -07002979{
Tejun Heoacbef752014-05-13 12:16:22 -04002980 return __cgroup_procs_write(of, buf, nbytes, off, false);
Ben Blum74a11662011-05-26 16:25:20 -07002981}
2982
Tejun Heoacbef752014-05-13 12:16:22 -04002983static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2984 char *buf, size_t nbytes, loff_t off)
Ben Blum74a11662011-05-26 16:25:20 -07002985{
Tejun Heoacbef752014-05-13 12:16:22 -04002986 return __cgroup_procs_write(of, buf, nbytes, off, true);
Paul Menageaf351022008-07-25 01:47:01 -07002987}
2988
Tejun Heo451af5042014-05-13 12:16:21 -04002989static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2990 char *buf, size_t nbytes, loff_t off)
Paul Menagee788e0662008-07-25 01:46:59 -07002991{
Tejun Heoe76ecae2014-05-13 12:19:23 -04002992 struct cgroup *cgrp;
Tejun Heo5f469902014-02-11 11:52:48 -05002993
Tejun Heoe76ecae2014-05-13 12:19:23 -04002994 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2995
Tejun Heo945ba192016-03-03 09:58:00 -05002996 cgrp = cgroup_kn_lock_live(of->kn, false);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002997 if (!cgrp)
Paul Menagee788e0662008-07-25 01:46:59 -07002998 return -ENODEV;
Tejun Heo69e943b2014-02-08 10:36:58 -05002999 spin_lock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04003000 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
3001 sizeof(cgrp->root->release_agent_path));
Tejun Heo69e943b2014-02-08 10:36:58 -05003002 spin_unlock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04003003 cgroup_kn_unlock(of->kn);
Tejun Heo451af5042014-05-13 12:16:21 -04003004 return nbytes;
Paul Menagee788e0662008-07-25 01:46:59 -07003005}
3006
Tejun Heo2da8ca82013-12-05 12:28:04 -05003007static int cgroup_release_agent_show(struct seq_file *seq, void *v)
Paul Menagee788e0662008-07-25 01:46:59 -07003008{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003009 struct cgroup *cgrp = seq_css(seq)->cgroup;
Tejun Heo182446d2013-08-08 20:11:24 -04003010
Tejun Heo46cfeb02014-05-13 12:11:00 -04003011 spin_lock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07003012 seq_puts(seq, cgrp->root->release_agent_path);
Tejun Heo46cfeb02014-05-13 12:11:00 -04003013 spin_unlock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07003014 seq_putc(seq, '\n');
Paul Menagee788e0662008-07-25 01:46:59 -07003015 return 0;
3016}
3017
Tejun Heo2da8ca82013-12-05 12:28:04 -05003018static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
Tejun Heo873fe092013-04-14 20:15:26 -07003019{
Tejun Heoc1d5d422014-07-09 10:08:08 -04003020 seq_puts(seq, "0\n");
Paul Menage81a6a5c2007-10-18 23:39:38 -07003021 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003022}
3023
Tejun Heo6e5c8302016-02-22 22:25:47 -05003024static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
Tejun Heof8f22e52014-04-23 11:13:16 -04003025{
3026 struct cgroup_subsys *ss;
3027 bool printed = false;
3028 int ssid;
3029
Tejun Heob4e0eea2016-02-22 22:25:46 -05003030 do_each_subsys_mask(ss, ssid, ss_mask) {
Aleksa Saraia966a4e2015-06-06 10:02:15 +10003031 if (printed)
3032 seq_putc(seq, ' ');
3033 seq_printf(seq, "%s", ss->name);
3034 printed = true;
Tejun Heob4e0eea2016-02-22 22:25:46 -05003035 } while_each_subsys_mask();
Tejun Heof8f22e52014-04-23 11:13:16 -04003036 if (printed)
3037 seq_putc(seq, '\n');
3038}
3039
Tejun Heof8f22e52014-04-23 11:13:16 -04003040/* show controllers which are enabled from the parent */
3041static int cgroup_controllers_show(struct seq_file *seq, void *v)
3042{
3043 struct cgroup *cgrp = seq_css(seq)->cgroup;
3044
Tejun Heo5531dc92016-03-03 09:57:58 -05003045 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
Tejun Heof8f22e52014-04-23 11:13:16 -04003046 return 0;
3047}
3048
3049/* show controllers which are enabled for a given cgroup's children */
3050static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
3051{
3052 struct cgroup *cgrp = seq_css(seq)->cgroup;
3053
Tejun Heo667c2492014-07-08 18:02:56 -04003054 cgroup_print_ss_mask(seq, cgrp->subtree_control);
Tejun Heof8f22e52014-04-23 11:13:16 -04003055 return 0;
3056}
3057
3058/**
3059 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3060 * @cgrp: root of the subtree to update csses for
3061 *
Tejun Heo54962602016-03-03 09:58:01 -05003062 * @cgrp's control masks have changed and its subtree's css associations
3063 * need to be updated accordingly. This function looks up all css_sets
3064 * which are attached to the subtree, creates the matching updated css_sets
3065 * and migrates the tasks to the new ones.
Tejun Heof8f22e52014-04-23 11:13:16 -04003066 */
3067static int cgroup_update_dfl_csses(struct cgroup *cgrp)
3068{
3069 LIST_HEAD(preloaded_csets);
Tejun Heo10265072015-09-11 15:00:22 -04003070 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
Tejun Heo54962602016-03-03 09:58:01 -05003071 struct cgroup_subsys_state *d_css;
3072 struct cgroup *dsct;
Tejun Heof8f22e52014-04-23 11:13:16 -04003073 struct css_set *src_cset;
3074 int ret;
3075
Tejun Heof8f22e52014-04-23 11:13:16 -04003076 lockdep_assert_held(&cgroup_mutex);
3077
Tejun Heo3014dde2015-09-16 13:03:02 -04003078 percpu_down_write(&cgroup_threadgroup_rwsem);
3079
Tejun Heof8f22e52014-04-23 11:13:16 -04003080 /* look up all csses currently attached to @cgrp's subtree */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003081 spin_lock_irq(&css_set_lock);
Tejun Heo54962602016-03-03 09:58:01 -05003082 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003083 struct cgrp_cset_link *link;
3084
Tejun Heo54962602016-03-03 09:58:01 -05003085 list_for_each_entry(link, &dsct->cset_links, cset_link)
Tejun Heo58cdb1c2016-03-08 11:51:25 -05003086 cgroup_migrate_add_src(link->cset, dsct,
Tejun Heof8f22e52014-04-23 11:13:16 -04003087 &preloaded_csets);
3088 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003089 spin_unlock_irq(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003090
3091 /* NULL dst indicates self on default hierarchy */
Tejun Heoe4857982016-03-08 11:51:26 -05003092 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heof8f22e52014-04-23 11:13:16 -04003093 if (ret)
3094 goto out_finish;
3095
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003096 spin_lock_irq(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003097 list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
Tejun Heo10265072015-09-11 15:00:22 -04003098 struct task_struct *task, *ntask;
Tejun Heof8f22e52014-04-23 11:13:16 -04003099
3100 /* src_csets precede dst_csets, break on the first dst_cset */
3101 if (!src_cset->mg_src_cgrp)
3102 break;
3103
Tejun Heo10265072015-09-11 15:00:22 -04003104 /* all tasks in src_csets need to be migrated */
3105 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3106 cgroup_taskset_add(task, &tset);
Tejun Heof8f22e52014-04-23 11:13:16 -04003107 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003108 spin_unlock_irq(&css_set_lock);
Tejun Heof8f22e52014-04-23 11:13:16 -04003109
Tejun Heo37ff9f82016-03-08 11:51:26 -05003110 ret = cgroup_taskset_migrate(&tset, cgrp->root);
Tejun Heof8f22e52014-04-23 11:13:16 -04003111out_finish:
3112 cgroup_migrate_finish(&preloaded_csets);
Tejun Heo3014dde2015-09-16 13:03:02 -04003113 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heof8f22e52014-04-23 11:13:16 -04003114 return ret;
3115}
3116
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003117/**
Tejun Heo945ba192016-03-03 09:58:00 -05003118 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
Tejun Heoce3f1d92016-03-03 09:57:59 -05003119 * @cgrp: root of the target subtree
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003120 *
3121 * Because css offlining is asynchronous, userland may try to re-enable a
Tejun Heo945ba192016-03-03 09:58:00 -05003122 * controller while the previous css is still around. This function grabs
3123 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003124 */
Tejun Heo945ba192016-03-03 09:58:00 -05003125static void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3126 __acquires(&cgroup_mutex)
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003127{
3128 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003129 struct cgroup_subsys_state *d_css;
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003130 struct cgroup_subsys *ss;
3131 int ssid;
3132
Tejun Heo945ba192016-03-03 09:58:00 -05003133restart:
3134 mutex_lock(&cgroup_mutex);
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003135
Tejun Heoce3f1d92016-03-03 09:57:59 -05003136 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003137 for_each_subsys(ss, ssid) {
3138 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3139 DEFINE_WAIT(wait);
3140
Tejun Heoce3f1d92016-03-03 09:57:59 -05003141 if (!css || !percpu_ref_is_dying(&css->refcnt))
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003142 continue;
3143
3144 cgroup_get(dsct);
3145 prepare_to_wait(&dsct->offline_waitq, &wait,
3146 TASK_UNINTERRUPTIBLE);
3147
3148 mutex_unlock(&cgroup_mutex);
3149 schedule();
3150 finish_wait(&dsct->offline_waitq, &wait);
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003151
3152 cgroup_put(dsct);
Tejun Heo945ba192016-03-03 09:58:00 -05003153 goto restart;
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003154 }
3155 }
Tejun Heo1b9b96a2016-03-03 09:57:59 -05003156}
3157
Tejun Heo12b3bb62016-03-03 09:57:59 -05003158/**
Tejun Heo15a27c32016-03-03 09:57:59 -05003159 * cgroup_save_control - save control masks of a subtree
3160 * @cgrp: root of the target subtree
3161 *
3162 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
3163 * prefixed fields for @cgrp's subtree including @cgrp itself.
3164 */
3165static void cgroup_save_control(struct cgroup *cgrp)
3166{
3167 struct cgroup *dsct;
3168 struct cgroup_subsys_state *d_css;
3169
3170 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3171 dsct->old_subtree_control = dsct->subtree_control;
3172 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3173 }
3174}
3175
3176/**
3177 * cgroup_propagate_control - refresh control masks of a subtree
3178 * @cgrp: root of the target subtree
3179 *
3180 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3181 * ->subtree_control and propagate controller availability through the
3182 * subtree so that descendants don't have unavailable controllers enabled.
3183 */
3184static void cgroup_propagate_control(struct cgroup *cgrp)
3185{
3186 struct cgroup *dsct;
3187 struct cgroup_subsys_state *d_css;
3188
3189 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3190 dsct->subtree_control &= cgroup_control(dsct);
Tejun Heo5ced2512016-03-03 09:58:01 -05003191 dsct->subtree_ss_mask =
3192 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3193 cgroup_ss_mask(dsct));
Tejun Heo15a27c32016-03-03 09:57:59 -05003194 }
3195}
3196
3197/**
3198 * cgroup_restore_control - restore control masks of a subtree
3199 * @cgrp: root of the target subtree
3200 *
3201 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
3202 * prefixed fields for @cgrp's subtree including @cgrp itself.
3203 */
3204static void cgroup_restore_control(struct cgroup *cgrp)
3205{
3206 struct cgroup *dsct;
3207 struct cgroup_subsys_state *d_css;
3208
3209 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3210 dsct->subtree_control = dsct->old_subtree_control;
3211 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3212 }
3213}
3214
Tejun Heof6d635ad2016-03-08 11:51:26 -05003215static bool css_visible(struct cgroup_subsys_state *css)
3216{
3217 struct cgroup_subsys *ss = css->ss;
3218 struct cgroup *cgrp = css->cgroup;
3219
3220 if (cgroup_control(cgrp) & (1 << ss->id))
3221 return true;
3222 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3223 return false;
3224 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3225}
3226
Tejun Heo15a27c32016-03-03 09:57:59 -05003227/**
Tejun Heobdb53bd2016-03-03 09:57:59 -05003228 * cgroup_apply_control_enable - enable or show csses according to control
Tejun Heoce3f1d92016-03-03 09:57:59 -05003229 * @cgrp: root of the target subtree
Tejun Heobdb53bd2016-03-03 09:57:59 -05003230 *
Tejun Heoce3f1d92016-03-03 09:57:59 -05003231 * Walk @cgrp's subtree and create new csses or make the existing ones
Tejun Heobdb53bd2016-03-03 09:57:59 -05003232 * visible. A css is created invisible if it's being implicitly enabled
3233 * through dependency. An invisible css is made visible when the userland
3234 * explicitly enables it.
3235 *
3236 * Returns 0 on success, -errno on failure. On failure, csses which have
3237 * been processed already aren't cleaned up. The caller is responsible for
3238 * cleaning up with cgroup_apply_control_disble().
3239 */
3240static int cgroup_apply_control_enable(struct cgroup *cgrp)
3241{
3242 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003243 struct cgroup_subsys_state *d_css;
Tejun Heobdb53bd2016-03-03 09:57:59 -05003244 struct cgroup_subsys *ss;
3245 int ssid, ret;
3246
Tejun Heoce3f1d92016-03-03 09:57:59 -05003247 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
Tejun Heobdb53bd2016-03-03 09:57:59 -05003248 for_each_subsys(ss, ssid) {
3249 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3250
Tejun Heo945ba192016-03-03 09:58:00 -05003251 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3252
Tejun Heobdb53bd2016-03-03 09:57:59 -05003253 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3254 continue;
3255
3256 if (!css) {
3257 css = css_create(dsct, ss);
3258 if (IS_ERR(css))
3259 return PTR_ERR(css);
3260 }
3261
Tejun Heof6d635ad2016-03-08 11:51:26 -05003262 if (css_visible(css)) {
Tejun Heo334c3672016-03-03 09:58:01 -05003263 ret = css_populate_dir(css);
Tejun Heobdb53bd2016-03-03 09:57:59 -05003264 if (ret)
3265 return ret;
3266 }
3267 }
3268 }
3269
3270 return 0;
3271}
3272
3273/**
Tejun Heo12b3bb62016-03-03 09:57:59 -05003274 * cgroup_apply_control_disable - kill or hide csses according to control
Tejun Heoce3f1d92016-03-03 09:57:59 -05003275 * @cgrp: root of the target subtree
Tejun Heo12b3bb62016-03-03 09:57:59 -05003276 *
Tejun Heoce3f1d92016-03-03 09:57:59 -05003277 * Walk @cgrp's subtree and kill and hide csses so that they match
Tejun Heo12b3bb62016-03-03 09:57:59 -05003278 * cgroup_ss_mask() and cgroup_visible_mask().
3279 *
3280 * A css is hidden when the userland requests it to be disabled while other
3281 * subsystems are still depending on it. The css must not actively control
3282 * resources and be in the vanilla state if it's made visible again later.
3283 * Controllers which may be depended upon should provide ->css_reset() for
3284 * this purpose.
3285 */
3286static void cgroup_apply_control_disable(struct cgroup *cgrp)
3287{
3288 struct cgroup *dsct;
Tejun Heoce3f1d92016-03-03 09:57:59 -05003289 struct cgroup_subsys_state *d_css;
Tejun Heo12b3bb62016-03-03 09:57:59 -05003290 struct cgroup_subsys *ss;
3291 int ssid;
3292
Tejun Heoce3f1d92016-03-03 09:57:59 -05003293 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
Tejun Heo12b3bb62016-03-03 09:57:59 -05003294 for_each_subsys(ss, ssid) {
3295 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3296
Tejun Heo945ba192016-03-03 09:58:00 -05003297 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3298
Tejun Heo12b3bb62016-03-03 09:57:59 -05003299 if (!css)
3300 continue;
3301
Tejun Heo334c3672016-03-03 09:58:01 -05003302 if (css->parent &&
3303 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
Tejun Heo12b3bb62016-03-03 09:57:59 -05003304 kill_css(css);
Tejun Heof6d635ad2016-03-08 11:51:26 -05003305 } else if (!css_visible(css)) {
Tejun Heo334c3672016-03-03 09:58:01 -05003306 css_clear_dir(css);
Tejun Heo12b3bb62016-03-03 09:57:59 -05003307 if (ss->css_reset)
3308 ss->css_reset(css);
3309 }
3310 }
3311 }
3312}
3313
Tejun Heof7b28142016-03-03 09:58:00 -05003314/**
3315 * cgroup_apply_control - apply control mask updates to the subtree
3316 * @cgrp: root of the target subtree
3317 *
3318 * subsystems can be enabled and disabled in a subtree using the following
3319 * steps.
3320 *
3321 * 1. Call cgroup_save_control() to stash the current state.
3322 * 2. Update ->subtree_control masks in the subtree as desired.
3323 * 3. Call cgroup_apply_control() to apply the changes.
3324 * 4. Optionally perform other related operations.
3325 * 5. Call cgroup_finalize_control() to finish up.
3326 *
3327 * This function implements step 3 and propagates the mask changes
3328 * throughout @cgrp's subtree, updates csses accordingly and perform
3329 * process migrations.
3330 */
3331static int cgroup_apply_control(struct cgroup *cgrp)
3332{
3333 int ret;
3334
3335 cgroup_propagate_control(cgrp);
3336
3337 ret = cgroup_apply_control_enable(cgrp);
3338 if (ret)
3339 return ret;
3340
3341 /*
3342 * At this point, cgroup_e_css() results reflect the new csses
3343 * making the following cgroup_update_dfl_csses() properly update
3344 * css associations of all tasks in the subtree.
3345 */
3346 ret = cgroup_update_dfl_csses(cgrp);
3347 if (ret)
3348 return ret;
3349
3350 return 0;
3351}
3352
3353/**
3354 * cgroup_finalize_control - finalize control mask update
3355 * @cgrp: root of the target subtree
3356 * @ret: the result of the update
3357 *
3358 * Finalize control mask update. See cgroup_apply_control() for more info.
3359 */
3360static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3361{
3362 if (ret) {
3363 cgroup_restore_control(cgrp);
3364 cgroup_propagate_control(cgrp);
3365 }
3366
3367 cgroup_apply_control_disable(cgrp);
3368}
3369
Tejun Heof8f22e52014-04-23 11:13:16 -04003370/* change the enabled child controllers for a cgroup in the default hierarchy */
Tejun Heo451af5042014-05-13 12:16:21 -04003371static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3372 char *buf, size_t nbytes,
3373 loff_t off)
Tejun Heof8f22e52014-04-23 11:13:16 -04003374{
Tejun Heo6e5c8302016-02-22 22:25:47 -05003375 u16 enable = 0, disable = 0;
Tejun Heoa9746d82014-05-13 12:19:22 -04003376 struct cgroup *cgrp, *child;
Tejun Heof8f22e52014-04-23 11:13:16 -04003377 struct cgroup_subsys *ss;
Tejun Heo451af5042014-05-13 12:16:21 -04003378 char *tok;
Tejun Heof8f22e52014-04-23 11:13:16 -04003379 int ssid, ret;
3380
3381 /*
Tejun Heod37167a2014-05-13 12:10:59 -04003382 * Parse input - space separated list of subsystem names prefixed
3383 * with either + or -.
Tejun Heof8f22e52014-04-23 11:13:16 -04003384 */
Tejun Heo451af5042014-05-13 12:16:21 -04003385 buf = strstrip(buf);
3386 while ((tok = strsep(&buf, " "))) {
Tejun Heod37167a2014-05-13 12:10:59 -04003387 if (tok[0] == '\0')
3388 continue;
Tejun Heoa7165262016-02-23 10:00:50 -05003389 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
Tejun Heofc5ed1e2015-09-18 11:56:28 -04003390 if (!cgroup_ssid_enabled(ssid) ||
3391 strcmp(tok + 1, ss->name))
Tejun Heof8f22e52014-04-23 11:13:16 -04003392 continue;
3393
3394 if (*tok == '+') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04003395 enable |= 1 << ssid;
3396 disable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04003397 } else if (*tok == '-') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04003398 disable |= 1 << ssid;
3399 enable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04003400 } else {
3401 return -EINVAL;
3402 }
3403 break;
Tejun Heob4e0eea2016-02-22 22:25:46 -05003404 } while_each_subsys_mask();
Tejun Heof8f22e52014-04-23 11:13:16 -04003405 if (ssid == CGROUP_SUBSYS_COUNT)
3406 return -EINVAL;
3407 }
3408
Tejun Heo945ba192016-03-03 09:58:00 -05003409 cgrp = cgroup_kn_lock_live(of->kn, true);
Tejun Heoa9746d82014-05-13 12:19:22 -04003410 if (!cgrp)
3411 return -ENODEV;
Tejun Heof8f22e52014-04-23 11:13:16 -04003412
3413 for_each_subsys(ss, ssid) {
3414 if (enable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04003415 if (cgrp->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003416 enable &= ~(1 << ssid);
3417 continue;
3418 }
3419
Tejun Heo5531dc92016-03-03 09:57:58 -05003420 if (!(cgroup_control(cgrp) & (1 << ssid))) {
Tejun Heoc29adf22014-07-08 18:02:56 -04003421 ret = -ENOENT;
3422 goto out_unlock;
3423 }
Tejun Heof8f22e52014-04-23 11:13:16 -04003424 } else if (disable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04003425 if (!(cgrp->subtree_control & (1 << ssid))) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003426 disable &= ~(1 << ssid);
3427 continue;
3428 }
3429
3430 /* a child has it enabled? */
3431 cgroup_for_each_live_child(child, cgrp) {
Tejun Heo667c2492014-07-08 18:02:56 -04003432 if (child->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003433 ret = -EBUSY;
Tejun Heoddab2b62014-05-13 12:19:22 -04003434 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04003435 }
3436 }
3437 }
3438 }
3439
3440 if (!enable && !disable) {
3441 ret = 0;
Tejun Heoddab2b62014-05-13 12:19:22 -04003442 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04003443 }
3444
3445 /*
Tejun Heo667c2492014-07-08 18:02:56 -04003446 * Except for the root, subtree_control must be zero for a cgroup
Tejun Heof8f22e52014-04-23 11:13:16 -04003447 * with tasks so that child cgroups don't compete against tasks.
3448 */
Tejun Heod51f39b2014-05-16 13:22:48 -04003449 if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04003450 ret = -EBUSY;
3451 goto out_unlock;
3452 }
3453
Tejun Heo15a27c32016-03-03 09:57:59 -05003454 /* save and update control masks and prepare csses */
3455 cgroup_save_control(cgrp);
Tejun Heoc29adf22014-07-08 18:02:56 -04003456
Tejun Heo15a27c32016-03-03 09:57:59 -05003457 cgrp->subtree_control |= enable;
3458 cgrp->subtree_control &= ~disable;
Tejun Heof63070d2014-07-08 18:02:57 -04003459
Tejun Heof7b28142016-03-03 09:58:00 -05003460 ret = cgroup_apply_control(cgrp);
Tejun Heo755bf5e2014-11-18 02:49:50 -05003461
Tejun Heof7b28142016-03-03 09:58:00 -05003462 cgroup_finalize_control(cgrp, ret);
Tejun Heof8f22e52014-04-23 11:13:16 -04003463
3464 kernfs_activate(cgrp->kn);
3465 ret = 0;
3466out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04003467 cgroup_kn_unlock(of->kn);
Tejun Heo451af5042014-05-13 12:16:21 -04003468 return ret ?: nbytes;
Tejun Heof8f22e52014-04-23 11:13:16 -04003469}
3470
Tejun Heo4a07c222015-09-18 17:54:22 -04003471static int cgroup_events_show(struct seq_file *seq, void *v)
Tejun Heo842b5972014-04-25 18:28:02 -04003472{
Tejun Heo4a07c222015-09-18 17:54:22 -04003473 seq_printf(seq, "populated %d\n",
Tejun Heo27bd4db2015-10-15 16:41:50 -04003474 cgroup_is_populated(seq_css(seq)->cgroup));
Tejun Heo842b5972014-04-25 18:28:02 -04003475 return 0;
3476}
3477
Tejun Heo2bd59d42014-02-11 11:52:49 -05003478static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3479 size_t nbytes, loff_t off)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003480{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003481 struct cgroup *cgrp = of->kn->parent->priv;
3482 struct cftype *cft = of->kn->priv;
3483 struct cgroup_subsys_state *css;
Tejun Heoa742c592013-12-05 12:28:03 -05003484 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003485
Tejun Heob4168642014-05-13 12:16:21 -04003486 if (cft->write)
3487 return cft->write(of, buf, nbytes, off);
3488
Tejun Heo2bd59d42014-02-11 11:52:49 -05003489 /*
3490 * kernfs guarantees that a file isn't deleted with operations in
3491 * flight, which means that the matching css is and stays alive and
3492 * doesn't need to be pinned. The RCU locking is not necessary
3493 * either. It's just for the convenience of using cgroup_css().
3494 */
3495 rcu_read_lock();
3496 css = cgroup_css(cgrp, cft->ss);
3497 rcu_read_unlock();
Paul Menageddbcc7e2007-10-18 23:39:30 -07003498
Tejun Heo451af5042014-05-13 12:16:21 -04003499 if (cft->write_u64) {
Tejun Heoa742c592013-12-05 12:28:03 -05003500 unsigned long long v;
3501 ret = kstrtoull(buf, 0, &v);
3502 if (!ret)
3503 ret = cft->write_u64(css, cft, v);
3504 } else if (cft->write_s64) {
3505 long long v;
3506 ret = kstrtoll(buf, 0, &v);
3507 if (!ret)
3508 ret = cft->write_s64(css, cft, v);
Tejun Heoa742c592013-12-05 12:28:03 -05003509 } else {
3510 ret = -EINVAL;
3511 }
Tejun Heo2bd59d42014-02-11 11:52:49 -05003512
Tejun Heoa742c592013-12-05 12:28:03 -05003513 return ret ?: nbytes;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003514}
3515
Tejun Heo6612f052013-12-05 12:28:04 -05003516static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
Paul Menage91796562008-04-29 01:00:01 -07003517{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003518 return seq_cft(seq)->seq_start(seq, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05003519}
3520
3521static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3522{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003523 return seq_cft(seq)->seq_next(seq, v, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05003524}
3525
3526static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3527{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003528 seq_cft(seq)->seq_stop(seq, v);
Paul Menage91796562008-04-29 01:00:01 -07003529}
3530
3531static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3532{
Tejun Heo7da11272013-12-05 12:28:04 -05003533 struct cftype *cft = seq_cft(m);
3534 struct cgroup_subsys_state *css = seq_css(m);
Li Zefane0798ce2013-07-31 17:36:25 +08003535
Tejun Heo2da8ca82013-12-05 12:28:04 -05003536 if (cft->seq_show)
3537 return cft->seq_show(m, arg);
Paul Menage91796562008-04-29 01:00:01 -07003538
Tejun Heo896f5192013-12-05 12:28:04 -05003539 if (cft->read_u64)
3540 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3541 else if (cft->read_s64)
3542 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3543 else
3544 return -EINVAL;
3545 return 0;
Paul Menage91796562008-04-29 01:00:01 -07003546}
3547
Tejun Heo2bd59d42014-02-11 11:52:49 -05003548static struct kernfs_ops cgroup_kf_single_ops = {
3549 .atomic_write_len = PAGE_SIZE,
3550 .write = cgroup_file_write,
3551 .seq_show = cgroup_seqfile_show,
Paul Menage91796562008-04-29 01:00:01 -07003552};
3553
Tejun Heo2bd59d42014-02-11 11:52:49 -05003554static struct kernfs_ops cgroup_kf_ops = {
3555 .atomic_write_len = PAGE_SIZE,
3556 .write = cgroup_file_write,
3557 .seq_start = cgroup_seqfile_start,
3558 .seq_next = cgroup_seqfile_next,
3559 .seq_stop = cgroup_seqfile_stop,
3560 .seq_show = cgroup_seqfile_show,
3561};
Paul Menageddbcc7e2007-10-18 23:39:30 -07003562
3563/*
3564 * cgroup_rename - Only allow simple rename of directories in place.
3565 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05003566static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
3567 const char *new_name_str)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003568{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003569 struct cgroup *cgrp = kn->priv;
Li Zefan65dff752013-03-01 15:01:56 +08003570 int ret;
Li Zefan65dff752013-03-01 15:01:56 +08003571
Tejun Heo2bd59d42014-02-11 11:52:49 -05003572 if (kernfs_type(kn) != KERNFS_DIR)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003573 return -ENOTDIR;
Tejun Heo2bd59d42014-02-11 11:52:49 -05003574 if (kn->parent != new_parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003575 return -EIO;
Li Zefan65dff752013-03-01 15:01:56 +08003576
Tejun Heo6db8e852013-06-14 11:18:22 -07003577 /*
3578 * This isn't a proper migration and its usefulness is very
Tejun Heoaa6ec292014-07-09 10:08:08 -04003579 * limited. Disallow on the default hierarchy.
Tejun Heo6db8e852013-06-14 11:18:22 -07003580 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04003581 if (cgroup_on_dfl(cgrp))
Tejun Heo6db8e852013-06-14 11:18:22 -07003582 return -EPERM;
3583
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003584 /*
Tejun Heo8353da12014-05-13 12:19:23 -04003585 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003586 * active_ref. kernfs_rename() doesn't require active_ref
Tejun Heo8353da12014-05-13 12:19:23 -04003587 * protection. Break them before grabbing cgroup_mutex.
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003588 */
3589 kernfs_break_active_protection(new_parent);
3590 kernfs_break_active_protection(kn);
Li Zefan65dff752013-03-01 15:01:56 +08003591
Tejun Heo2bd59d42014-02-11 11:52:49 -05003592 mutex_lock(&cgroup_mutex);
Li Zefan65dff752013-03-01 15:01:56 +08003593
Tejun Heo2bd59d42014-02-11 11:52:49 -05003594 ret = kernfs_rename(kn, new_parent, new_name_str);
Li Zefan65dff752013-03-01 15:01:56 +08003595
Tejun Heo2bd59d42014-02-11 11:52:49 -05003596 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003597
Tejun Heoe1b2dc12014-03-20 11:10:15 -04003598 kernfs_unbreak_active_protection(kn);
3599 kernfs_unbreak_active_protection(new_parent);
Tejun Heo2bd59d42014-02-11 11:52:49 -05003600 return ret;
Li Zefan099fca32009-04-02 16:57:29 -07003601}
3602
Tejun Heo49957f82014-04-07 16:44:47 -04003603/* set uid and gid of cgroup dirs and files to that of the creator */
3604static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3605{
3606 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3607 .ia_uid = current_fsuid(),
3608 .ia_gid = current_fsgid(), };
3609
3610 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3611 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3612 return 0;
3613
3614 return kernfs_setattr(kn, &iattr);
3615}
3616
Tejun Heo4df8dc92015-09-18 17:54:23 -04003617static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3618 struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003619{
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05003620 char name[CGROUP_FILE_NAME_MAX];
Tejun Heo2bd59d42014-02-11 11:52:49 -05003621 struct kernfs_node *kn;
3622 struct lock_class_key *key = NULL;
Tejun Heo49957f82014-04-07 16:44:47 -04003623 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003624
Tejun Heo2bd59d42014-02-11 11:52:49 -05003625#ifdef CONFIG_DEBUG_LOCK_ALLOC
3626 key = &cft->lockdep_key;
3627#endif
3628 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3629 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
Tejun Heodfeb07502015-02-13 14:36:31 -08003630 NULL, key);
Tejun Heo49957f82014-04-07 16:44:47 -04003631 if (IS_ERR(kn))
3632 return PTR_ERR(kn);
3633
3634 ret = cgroup_kn_set_ugid(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003635 if (ret) {
Tejun Heo49957f82014-04-07 16:44:47 -04003636 kernfs_remove(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003637 return ret;
3638 }
3639
Tejun Heo6f60ead2015-09-18 17:54:23 -04003640 if (cft->file_offset) {
3641 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3642
Tejun Heo34c06252015-11-05 00:12:24 -05003643 spin_lock_irq(&cgroup_file_kn_lock);
Tejun Heo6f60ead2015-09-18 17:54:23 -04003644 cfile->kn = kn;
Tejun Heo34c06252015-11-05 00:12:24 -05003645 spin_unlock_irq(&cgroup_file_kn_lock);
Tejun Heo6f60ead2015-09-18 17:54:23 -04003646 }
3647
Tejun Heof8f22e52014-04-23 11:13:16 -04003648 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003649}
3650
Tejun Heob1f28d32013-06-28 16:24:10 -07003651/**
3652 * cgroup_addrm_files - add or remove files to a cgroup directory
Tejun Heo4df8dc92015-09-18 17:54:23 -04003653 * @css: the target css
3654 * @cgrp: the target cgroup (usually css->cgroup)
Tejun Heob1f28d32013-06-28 16:24:10 -07003655 * @cfts: array of cftypes to be added
3656 * @is_add: whether to add or remove
3657 *
3658 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
Tejun Heo6732ed82015-09-18 17:54:23 -04003659 * For removals, this function never fails.
Tejun Heob1f28d32013-06-28 16:24:10 -07003660 */
Tejun Heo4df8dc92015-09-18 17:54:23 -04003661static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3662 struct cgroup *cgrp, struct cftype cfts[],
Tejun Heo2bb566c2013-08-08 20:11:23 -04003663 bool is_add)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003664{
Tejun Heo6732ed82015-09-18 17:54:23 -04003665 struct cftype *cft, *cft_end = NULL;
Tejun Heob598dde2016-02-22 22:25:45 -05003666 int ret = 0;
Tejun Heob1f28d32013-06-28 16:24:10 -07003667
Tejun Heo01f64742014-05-13 12:19:23 -04003668 lockdep_assert_held(&cgroup_mutex);
Tejun Heodb0416b2012-04-01 12:09:55 -07003669
Tejun Heo6732ed82015-09-18 17:54:23 -04003670restart:
3671 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
Gao fengf33fddc2012-12-06 14:38:57 +08003672 /* does cft->flags tell us to skip this file on @cgrp? */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003673 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
Tejun Heo8cbbf2c2014-03-19 10:23:55 -04003674 continue;
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003675 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
Tejun Heo873fe092013-04-14 20:15:26 -07003676 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003677 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003678 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003679 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003680 continue;
3681
Li Zefan2739d3c2013-01-21 18:18:33 +08003682 if (is_add) {
Tejun Heo4df8dc92015-09-18 17:54:23 -04003683 ret = cgroup_add_file(css, cgrp, cft);
Tejun Heob1f28d32013-06-28 16:24:10 -07003684 if (ret) {
Joe Perchesed3d2612014-04-25 18:28:03 -04003685 pr_warn("%s: failed to add %s, err=%d\n",
3686 __func__, cft->name, ret);
Tejun Heo6732ed82015-09-18 17:54:23 -04003687 cft_end = cft;
3688 is_add = false;
3689 goto restart;
Tejun Heob1f28d32013-06-28 16:24:10 -07003690 }
Li Zefan2739d3c2013-01-21 18:18:33 +08003691 } else {
3692 cgroup_rm_file(cgrp, cft);
Tejun Heodb0416b2012-04-01 12:09:55 -07003693 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003694 }
Tejun Heob598dde2016-02-22 22:25:45 -05003695 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003696}
3697
Tejun Heo21a2d342014-02-12 09:29:49 -05003698static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003699{
3700 LIST_HEAD(pending);
Tejun Heo2bb566c2013-08-08 20:11:23 -04003701 struct cgroup_subsys *ss = cfts[0].ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04003702 struct cgroup *root = &ss->root->cgrp;
Tejun Heo492eb212013-08-08 20:11:25 -04003703 struct cgroup_subsys_state *css;
Tejun Heo9ccece82013-06-28 16:24:11 -07003704 int ret = 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003705
Tejun Heo01f64742014-05-13 12:19:23 -04003706 lockdep_assert_held(&cgroup_mutex);
Li Zefane8c82d22013-06-18 18:48:37 +08003707
Li Zefane8c82d22013-06-18 18:48:37 +08003708 /* add/rm files for all cgroups created before */
Tejun Heoca8bdca2013-08-26 18:40:56 -04003709 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
Tejun Heo492eb212013-08-08 20:11:25 -04003710 struct cgroup *cgrp = css->cgroup;
3711
Tejun Heo88cb04b2016-03-03 09:57:58 -05003712 if (!(css->flags & CSS_VISIBLE))
Li Zefane8c82d22013-06-18 18:48:37 +08003713 continue;
3714
Tejun Heo4df8dc92015-09-18 17:54:23 -04003715 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
Tejun Heo9ccece82013-06-28 16:24:11 -07003716 if (ret)
3717 break;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003718 }
Tejun Heo21a2d342014-02-12 09:29:49 -05003719
3720 if (is_add && !ret)
3721 kernfs_activate(root->kn);
Tejun Heo9ccece82013-06-28 16:24:11 -07003722 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003723}
3724
Tejun Heo2da440a2014-02-11 11:52:48 -05003725static void cgroup_exit_cftypes(struct cftype *cfts)
3726{
3727 struct cftype *cft;
3728
Tejun Heo2bd59d42014-02-11 11:52:49 -05003729 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3730 /* free copy for custom atomic_write_len, see init_cftypes() */
3731 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3732 kfree(cft->kf_ops);
3733 cft->kf_ops = NULL;
Tejun Heo2da440a2014-02-11 11:52:48 -05003734 cft->ss = NULL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003735
3736 /* revert flags set by cgroup core while adding @cfts */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003737 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
Tejun Heo2bd59d42014-02-11 11:52:49 -05003738 }
Tejun Heo2da440a2014-02-11 11:52:48 -05003739}
3740
Tejun Heo2bd59d42014-02-11 11:52:49 -05003741static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo2da440a2014-02-11 11:52:48 -05003742{
3743 struct cftype *cft;
3744
Tejun Heo2bd59d42014-02-11 11:52:49 -05003745 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3746 struct kernfs_ops *kf_ops;
3747
Tejun Heo0adb0702014-02-12 09:29:48 -05003748 WARN_ON(cft->ss || cft->kf_ops);
3749
Tejun Heo2bd59d42014-02-11 11:52:49 -05003750 if (cft->seq_start)
3751 kf_ops = &cgroup_kf_ops;
3752 else
3753 kf_ops = &cgroup_kf_single_ops;
3754
3755 /*
3756 * Ugh... if @cft wants a custom max_write_len, we need to
3757 * make a copy of kf_ops to set its atomic_write_len.
3758 */
3759 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3760 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3761 if (!kf_ops) {
3762 cgroup_exit_cftypes(cfts);
3763 return -ENOMEM;
3764 }
3765 kf_ops->atomic_write_len = cft->max_write_len;
3766 }
3767
3768 cft->kf_ops = kf_ops;
Tejun Heo2da440a2014-02-11 11:52:48 -05003769 cft->ss = ss;
Tejun Heo2bd59d42014-02-11 11:52:49 -05003770 }
3771
3772 return 0;
Tejun Heo2da440a2014-02-11 11:52:48 -05003773}
3774
Tejun Heo21a2d342014-02-12 09:29:49 -05003775static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3776{
Tejun Heo01f64742014-05-13 12:19:23 -04003777 lockdep_assert_held(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003778
3779 if (!cfts || !cfts[0].ss)
3780 return -ENOENT;
3781
3782 list_del(&cfts->node);
3783 cgroup_apply_cftypes(cfts, false);
3784 cgroup_exit_cftypes(cfts);
3785 return 0;
3786}
3787
Tejun Heo8e3f6542012-04-01 12:09:55 -07003788/**
Tejun Heo80b13582014-02-12 09:29:48 -05003789 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3790 * @cfts: zero-length name terminated array of cftypes
3791 *
3792 * Unregister @cfts. Files described by @cfts are removed from all
3793 * existing cgroups and all future cgroups won't have them either. This
3794 * function can be called anytime whether @cfts' subsys is attached or not.
3795 *
3796 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3797 * registered.
3798 */
3799int cgroup_rm_cftypes(struct cftype *cfts)
3800{
Tejun Heo21a2d342014-02-12 09:29:49 -05003801 int ret;
Tejun Heo80b13582014-02-12 09:29:48 -05003802
Tejun Heo01f64742014-05-13 12:19:23 -04003803 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003804 ret = cgroup_rm_cftypes_locked(cfts);
Tejun Heo01f64742014-05-13 12:19:23 -04003805 mutex_unlock(&cgroup_mutex);
Tejun Heo8e3f6542012-04-01 12:09:55 -07003806 return ret;
3807}
3808
3809/**
3810 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3811 * @ss: target cgroup subsystem
3812 * @cfts: zero-length name terminated array of cftypes
3813 *
3814 * Register @cfts to @ss. Files described by @cfts are created for all
3815 * existing cgroups to which @ss is attached and all future cgroups will
3816 * have them too. This function can be called anytime whether @ss is
3817 * attached or not.
3818 *
3819 * Returns 0 on successful registration, -errno on failure. Note that this
3820 * function currently returns 0 as long as @cfts registration is successful
3821 * even if some file creation attempts on existing cgroups fail.
3822 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003823static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003824{
Tejun Heo9ccece82013-06-28 16:24:11 -07003825 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003826
Tejun Heofc5ed1e2015-09-18 11:56:28 -04003827 if (!cgroup_ssid_enabled(ss->id))
Li Zefanc731ae12014-06-05 17:16:30 +08003828 return 0;
3829
Li Zefandc5736e2014-02-17 10:41:50 +08003830 if (!cfts || cfts[0].name[0] == '\0')
3831 return 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003832
Tejun Heo2bd59d42014-02-11 11:52:49 -05003833 ret = cgroup_init_cftypes(ss, cfts);
Tejun Heo9ccece82013-06-28 16:24:11 -07003834 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05003835 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003836
Tejun Heo01f64742014-05-13 12:19:23 -04003837 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003838
Tejun Heo0adb0702014-02-12 09:29:48 -05003839 list_add_tail(&cfts->node, &ss->cfts);
Tejun Heo21a2d342014-02-12 09:29:49 -05003840 ret = cgroup_apply_cftypes(cfts, true);
Tejun Heo9ccece82013-06-28 16:24:11 -07003841 if (ret)
Tejun Heo21a2d342014-02-12 09:29:49 -05003842 cgroup_rm_cftypes_locked(cfts);
3843
Tejun Heo01f64742014-05-13 12:19:23 -04003844 mutex_unlock(&cgroup_mutex);
Tejun Heo9ccece82013-06-28 16:24:11 -07003845 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003846}
Tejun Heo79578622012-04-01 12:09:56 -07003847
3848/**
Tejun Heoa8ddc822014-07-15 11:05:10 -04003849 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3850 * @ss: target cgroup subsystem
3851 * @cfts: zero-length name terminated array of cftypes
3852 *
3853 * Similar to cgroup_add_cftypes() but the added files are only used for
3854 * the default hierarchy.
3855 */
3856int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3857{
3858 struct cftype *cft;
3859
3860 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003861 cft->flags |= __CFTYPE_ONLY_ON_DFL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003862 return cgroup_add_cftypes(ss, cfts);
3863}
3864
3865/**
3866 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3867 * @ss: target cgroup subsystem
3868 * @cfts: zero-length name terminated array of cftypes
3869 *
3870 * Similar to cgroup_add_cftypes() but the added files are only used for
3871 * the legacy hierarchies.
3872 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003873int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3874{
Tejun Heoa8ddc822014-07-15 11:05:10 -04003875 struct cftype *cft;
3876
Tejun Heoe4b70372015-10-15 17:00:43 -04003877 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3878 cft->flags |= __CFTYPE_NOT_ON_DFL;
Tejun Heo2cf669a2014-07-15 11:05:09 -04003879 return cgroup_add_cftypes(ss, cfts);
3880}
3881
Li Zefana043e3b2008-02-23 15:24:09 -08003882/**
Tejun Heo34c06252015-11-05 00:12:24 -05003883 * cgroup_file_notify - generate a file modified event for a cgroup_file
3884 * @cfile: target cgroup_file
3885 *
3886 * @cfile must have been obtained by setting cftype->file_offset.
3887 */
3888void cgroup_file_notify(struct cgroup_file *cfile)
3889{
3890 unsigned long flags;
3891
3892 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3893 if (cfile->kn)
3894 kernfs_notify(cfile->kn);
3895 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3896}
3897
3898/**
Li Zefana043e3b2008-02-23 15:24:09 -08003899 * cgroup_task_count - count the number of tasks in a cgroup.
3900 * @cgrp: the cgroup in question
3901 *
3902 * Return the number of tasks in the cgroup.
3903 */
Tejun Heo07bc3562014-02-13 06:58:39 -05003904static int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07003905{
3906 int count = 0;
Tejun Heo69d02062013-06-12 21:04:50 -07003907 struct cgrp_cset_link *link;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003908
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003909 spin_lock_irq(&css_set_lock);
Tejun Heo69d02062013-06-12 21:04:50 -07003910 list_for_each_entry(link, &cgrp->cset_links, cset_link)
3911 count += atomic_read(&link->cset->refcount);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03003912 spin_unlock_irq(&css_set_lock);
Paul Menagebbcb81d2007-10-18 23:39:32 -07003913 return count;
3914}
3915
Tejun Heo574bd9f2012-11-09 09:12:29 -08003916/**
Tejun Heo492eb212013-08-08 20:11:25 -04003917 * css_next_child - find the next child of a given css
Tejun Heoc2931b72014-05-16 13:22:51 -04003918 * @pos: the current position (%NULL to initiate traversal)
3919 * @parent: css whose children to walk
Tejun Heo53fa5262013-05-24 10:55:38 +09003920 *
Tejun Heoc2931b72014-05-16 13:22:51 -04003921 * This function returns the next child of @parent and should be called
Tejun Heo87fb54f2013-12-06 15:11:55 -05003922 * under either cgroup_mutex or RCU read lock. The only requirement is
Tejun Heoc2931b72014-05-16 13:22:51 -04003923 * that @parent and @pos are accessible. The next sibling is guaranteed to
3924 * be returned regardless of their states.
3925 *
3926 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3927 * css which finished ->css_online() is guaranteed to be visible in the
3928 * future iterations and will stay visible until the last reference is put.
3929 * A css which hasn't finished ->css_online() or already finished
3930 * ->css_offline() may show up during traversal. It's each subsystem's
3931 * responsibility to synchronize against on/offlining.
Tejun Heo53fa5262013-05-24 10:55:38 +09003932 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003933struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3934 struct cgroup_subsys_state *parent)
Tejun Heo53fa5262013-05-24 10:55:38 +09003935{
Tejun Heoc2931b72014-05-16 13:22:51 -04003936 struct cgroup_subsys_state *next;
Tejun Heo53fa5262013-05-24 10:55:38 +09003937
Tejun Heo8353da12014-05-13 12:19:23 -04003938 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo53fa5262013-05-24 10:55:38 +09003939
3940 /*
Tejun Heode3f0342014-05-16 13:22:49 -04003941 * @pos could already have been unlinked from the sibling list.
3942 * Once a cgroup is removed, its ->sibling.next is no longer
3943 * updated when its next sibling changes. CSS_RELEASED is set when
3944 * @pos is taken off list, at which time its next pointer is valid,
3945 * and, as releases are serialized, the one pointed to by the next
3946 * pointer is guaranteed to not have started release yet. This
3947 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3948 * critical section, the one pointed to by its next pointer is
3949 * guaranteed to not have finished its RCU grace period even if we
3950 * have dropped rcu_read_lock() inbetween iterations.
Tejun Heo3b287a52013-08-08 20:11:24 -04003951 *
Tejun Heode3f0342014-05-16 13:22:49 -04003952 * If @pos has CSS_RELEASED set, its next pointer can't be
3953 * dereferenced; however, as each css is given a monotonically
3954 * increasing unique serial number and always appended to the
3955 * sibling list, the next one can be found by walking the parent's
3956 * children until the first css with higher serial number than
3957 * @pos's. While this path can be slower, it happens iff iteration
3958 * races against release and the race window is very small.
Tejun Heo53fa5262013-05-24 10:55:38 +09003959 */
Tejun Heo3b287a52013-08-08 20:11:24 -04003960 if (!pos) {
Tejun Heoc2931b72014-05-16 13:22:51 -04003961 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3962 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3963 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
Tejun Heo3b287a52013-08-08 20:11:24 -04003964 } else {
Tejun Heoc2931b72014-05-16 13:22:51 -04003965 list_for_each_entry_rcu(next, &parent->children, sibling)
Tejun Heo3b287a52013-08-08 20:11:24 -04003966 if (next->serial_nr > pos->serial_nr)
3967 break;
Tejun Heo53fa5262013-05-24 10:55:38 +09003968 }
3969
Tejun Heo3b281af2014-04-23 11:13:15 -04003970 /*
3971 * @next, if not pointing to the head, can be dereferenced and is
Tejun Heoc2931b72014-05-16 13:22:51 -04003972 * the next sibling.
Tejun Heo3b281af2014-04-23 11:13:15 -04003973 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003974 if (&next->sibling != &parent->children)
3975 return next;
Tejun Heo3b281af2014-04-23 11:13:15 -04003976 return NULL;
Tejun Heo53fa5262013-05-24 10:55:38 +09003977}
Tejun Heo53fa5262013-05-24 10:55:38 +09003978
3979/**
Tejun Heo492eb212013-08-08 20:11:25 -04003980 * css_next_descendant_pre - find the next descendant for pre-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003981 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04003982 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003983 *
Tejun Heo492eb212013-08-08 20:11:25 -04003984 * To be used by css_for_each_descendant_pre(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04003985 * to visit for pre-order traversal of @root's descendants. @root is
3986 * included in the iteration and the first node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09003987 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003988 * While this function requires cgroup_mutex or RCU read locking, it
3989 * doesn't require the whole traversal to be contained in a single critical
3990 * section. This function will return the correct next descendant as long
3991 * as both @pos and @root are accessible and @pos is a descendant of @root.
Tejun Heoc2931b72014-05-16 13:22:51 -04003992 *
3993 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3994 * css which finished ->css_online() is guaranteed to be visible in the
3995 * future iterations and will stay visible until the last reference is put.
3996 * A css which hasn't finished ->css_online() or already finished
3997 * ->css_offline() may show up during traversal. It's each subsystem's
3998 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08003999 */
Tejun Heo492eb212013-08-08 20:11:25 -04004000struct cgroup_subsys_state *
4001css_next_descendant_pre(struct cgroup_subsys_state *pos,
4002 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08004003{
Tejun Heo492eb212013-08-08 20:11:25 -04004004 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004005
Tejun Heo8353da12014-05-13 12:19:23 -04004006 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08004007
Tejun Heobd8815a2013-08-08 20:11:27 -04004008 /* if first iteration, visit @root */
Tejun Heo7805d002013-05-24 10:50:24 +09004009 if (!pos)
Tejun Heobd8815a2013-08-08 20:11:27 -04004010 return root;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004011
4012 /* visit the first child if exists */
Tejun Heo492eb212013-08-08 20:11:25 -04004013 next = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004014 if (next)
4015 return next;
4016
4017 /* no child, visit my or the closest ancestor's next sibling */
Tejun Heo492eb212013-08-08 20:11:25 -04004018 while (pos != root) {
Tejun Heo5c9d5352014-05-16 13:22:48 -04004019 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09004020 if (next)
Tejun Heo574bd9f2012-11-09 09:12:29 -08004021 return next;
Tejun Heo5c9d5352014-05-16 13:22:48 -04004022 pos = pos->parent;
Tejun Heo7805d002013-05-24 10:50:24 +09004023 }
Tejun Heo574bd9f2012-11-09 09:12:29 -08004024
4025 return NULL;
4026}
Tejun Heo574bd9f2012-11-09 09:12:29 -08004027
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004028/**
Tejun Heo492eb212013-08-08 20:11:25 -04004029 * css_rightmost_descendant - return the rightmost descendant of a css
4030 * @pos: css of interest
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004031 *
Tejun Heo492eb212013-08-08 20:11:25 -04004032 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4033 * is returned. This can be used during pre-order traversal to skip
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004034 * subtree of @pos.
Tejun Heo75501a62013-05-24 10:55:38 +09004035 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05004036 * While this function requires cgroup_mutex or RCU read locking, it
4037 * doesn't require the whole traversal to be contained in a single critical
4038 * section. This function will return the correct rightmost descendant as
4039 * long as @pos is accessible.
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004040 */
Tejun Heo492eb212013-08-08 20:11:25 -04004041struct cgroup_subsys_state *
4042css_rightmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004043{
Tejun Heo492eb212013-08-08 20:11:25 -04004044 struct cgroup_subsys_state *last, *tmp;
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004045
Tejun Heo8353da12014-05-13 12:19:23 -04004046 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004047
4048 do {
4049 last = pos;
4050 /* ->prev isn't RCU safe, walk ->next till the end */
4051 pos = NULL;
Tejun Heo492eb212013-08-08 20:11:25 -04004052 css_for_each_child(tmp, last)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004053 pos = tmp;
4054 } while (pos);
4055
4056 return last;
4057}
Tejun Heo12a9d2f2013-01-07 08:49:33 -08004058
Tejun Heo492eb212013-08-08 20:11:25 -04004059static struct cgroup_subsys_state *
4060css_leftmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo574bd9f2012-11-09 09:12:29 -08004061{
Tejun Heo492eb212013-08-08 20:11:25 -04004062 struct cgroup_subsys_state *last;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004063
4064 do {
4065 last = pos;
Tejun Heo492eb212013-08-08 20:11:25 -04004066 pos = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004067 } while (pos);
4068
4069 return last;
4070}
4071
4072/**
Tejun Heo492eb212013-08-08 20:11:25 -04004073 * css_next_descendant_post - find the next descendant for post-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08004074 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04004075 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08004076 *
Tejun Heo492eb212013-08-08 20:11:25 -04004077 * To be used by css_for_each_descendant_post(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04004078 * to visit for post-order traversal of @root's descendants. @root is
4079 * included in the iteration and the last node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09004080 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05004081 * While this function requires cgroup_mutex or RCU read locking, it
4082 * doesn't require the whole traversal to be contained in a single critical
4083 * section. This function will return the correct next descendant as long
4084 * as both @pos and @cgroup are accessible and @pos is a descendant of
4085 * @cgroup.
Tejun Heoc2931b72014-05-16 13:22:51 -04004086 *
4087 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4088 * css which finished ->css_online() is guaranteed to be visible in the
4089 * future iterations and will stay visible until the last reference is put.
4090 * A css which hasn't finished ->css_online() or already finished
4091 * ->css_offline() may show up during traversal. It's each subsystem's
4092 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08004093 */
Tejun Heo492eb212013-08-08 20:11:25 -04004094struct cgroup_subsys_state *
4095css_next_descendant_post(struct cgroup_subsys_state *pos,
4096 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08004097{
Tejun Heo492eb212013-08-08 20:11:25 -04004098 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004099
Tejun Heo8353da12014-05-13 12:19:23 -04004100 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08004101
Tejun Heo58b79a92013-09-06 15:31:08 -04004102 /* if first iteration, visit leftmost descendant which may be @root */
4103 if (!pos)
4104 return css_leftmost_descendant(root);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004105
Tejun Heobd8815a2013-08-08 20:11:27 -04004106 /* if we visited @root, we're done */
4107 if (pos == root)
4108 return NULL;
4109
Tejun Heo574bd9f2012-11-09 09:12:29 -08004110 /* if there's an unvisited sibling, visit its leftmost descendant */
Tejun Heo5c9d5352014-05-16 13:22:48 -04004111 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09004112 if (next)
Tejun Heo492eb212013-08-08 20:11:25 -04004113 return css_leftmost_descendant(next);
Tejun Heo574bd9f2012-11-09 09:12:29 -08004114
4115 /* no sibling left, visit parent */
Tejun Heo5c9d5352014-05-16 13:22:48 -04004116 return pos->parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -08004117}
Tejun Heo574bd9f2012-11-09 09:12:29 -08004118
Tejun Heof3d46502014-05-16 13:22:52 -04004119/**
4120 * css_has_online_children - does a css have online children
4121 * @css: the target css
4122 *
4123 * Returns %true if @css has any online children; otherwise, %false. This
4124 * function can be called from any context but the caller is responsible
4125 * for synchronizing against on/offlining as necessary.
4126 */
4127bool css_has_online_children(struct cgroup_subsys_state *css)
Tejun Heocbc125e2014-05-14 09:15:01 -04004128{
Tejun Heof3d46502014-05-16 13:22:52 -04004129 struct cgroup_subsys_state *child;
4130 bool ret = false;
Tejun Heocbc125e2014-05-14 09:15:01 -04004131
4132 rcu_read_lock();
Tejun Heof3d46502014-05-16 13:22:52 -04004133 css_for_each_child(child, css) {
Li Zefan99bae5f2014-06-12 14:31:31 +08004134 if (child->flags & CSS_ONLINE) {
Tejun Heof3d46502014-05-16 13:22:52 -04004135 ret = true;
4136 break;
Tejun Heocbc125e2014-05-14 09:15:01 -04004137 }
4138 }
4139 rcu_read_unlock();
Tejun Heof3d46502014-05-16 13:22:52 -04004140 return ret;
Cliff Wickman31a7df02008-02-07 00:14:42 -08004141}
4142
Tejun Heo0942eee2013-08-08 20:11:26 -04004143/**
Tejun Heoecb9d532015-10-15 16:41:52 -04004144 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
Tejun Heo0942eee2013-08-08 20:11:26 -04004145 * @it: the iterator to advance
4146 *
4147 * Advance @it to the next css_set to walk.
Tejun Heod5158762013-08-08 20:11:26 -04004148 */
Tejun Heoecb9d532015-10-15 16:41:52 -04004149static void css_task_iter_advance_css_set(struct css_task_iter *it)
Tejun Heod5158762013-08-08 20:11:26 -04004150{
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004151 struct list_head *l = it->cset_pos;
Tejun Heod5158762013-08-08 20:11:26 -04004152 struct cgrp_cset_link *link;
4153 struct css_set *cset;
4154
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004155 lockdep_assert_held(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004156
Tejun Heod5158762013-08-08 20:11:26 -04004157 /* Advance to the next non-empty css_set */
4158 do {
4159 l = l->next;
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004160 if (l == it->cset_head) {
4161 it->cset_pos = NULL;
Tejun Heoecb9d532015-10-15 16:41:52 -04004162 it->task_pos = NULL;
Tejun Heod5158762013-08-08 20:11:26 -04004163 return;
4164 }
Tejun Heo3ebb2b62014-04-23 11:13:15 -04004165
4166 if (it->ss) {
4167 cset = container_of(l, struct css_set,
4168 e_cset_node[it->ss->id]);
4169 } else {
4170 link = list_entry(l, struct cgrp_cset_link, cset_link);
4171 cset = link->cset;
4172 }
Tejun Heo0de09422015-10-15 16:41:49 -04004173 } while (!css_set_populated(cset));
Tejun Heoc7561122014-02-25 10:04:01 -05004174
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004175 it->cset_pos = l;
Tejun Heoc7561122014-02-25 10:04:01 -05004176
4177 if (!list_empty(&cset->tasks))
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004178 it->task_pos = cset->tasks.next;
Tejun Heoc7561122014-02-25 10:04:01 -05004179 else
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004180 it->task_pos = cset->mg_tasks.next;
4181
4182 it->tasks_head = &cset->tasks;
4183 it->mg_tasks_head = &cset->mg_tasks;
Tejun Heoed27b9f2015-10-15 16:41:52 -04004184
4185 /*
4186 * We don't keep css_sets locked across iteration steps and thus
4187 * need to take steps to ensure that iteration can be resumed after
4188 * the lock is re-acquired. Iteration is performed at two levels -
4189 * css_sets and tasks in them.
4190 *
4191 * Once created, a css_set never leaves its cgroup lists, so a
4192 * pinned css_set is guaranteed to stay put and we can resume
4193 * iteration afterwards.
4194 *
4195 * Tasks may leave @cset across iteration steps. This is resolved
4196 * by registering each iterator with the css_set currently being
4197 * walked and making css_set_move_task() advance iterators whose
4198 * next task is leaving.
4199 */
4200 if (it->cur_cset) {
4201 list_del(&it->iters_node);
4202 put_css_set_locked(it->cur_cset);
4203 }
4204 get_css_set(cset);
4205 it->cur_cset = cset;
4206 list_add(&it->iters_node, &cset->task_iters);
Tejun Heod5158762013-08-08 20:11:26 -04004207}
4208
Tejun Heoecb9d532015-10-15 16:41:52 -04004209static void css_task_iter_advance(struct css_task_iter *it)
4210{
4211 struct list_head *l = it->task_pos;
4212
Tejun Heof0d9a5f2015-10-15 16:41:53 -04004213 lockdep_assert_held(&css_set_lock);
Tejun Heoecb9d532015-10-15 16:41:52 -04004214 WARN_ON_ONCE(!l);
4215
4216 /*
4217 * Advance iterator to find next entry. cset->tasks is consumed
4218 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4219 * next cset.
4220 */
4221 l = l->next;
4222
4223 if (l == it->tasks_head)
4224 l = it->mg_tasks_head->next;
4225
4226 if (l == it->mg_tasks_head)
4227 css_task_iter_advance_css_set(it);
4228 else
4229 it->task_pos = l;
4230}
4231
Tejun Heo0942eee2013-08-08 20:11:26 -04004232/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004233 * css_task_iter_start - initiate task iteration
4234 * @css: the css to walk tasks of
Tejun Heo0942eee2013-08-08 20:11:26 -04004235 * @it: the task iterator to use
4236 *
Tejun Heo72ec7022013-08-08 20:11:26 -04004237 * Initiate iteration through the tasks of @css. The caller can call
4238 * css_task_iter_next() to walk through the tasks until the function
4239 * returns NULL. On completion of iteration, css_task_iter_end() must be
4240 * called.
Tejun Heo0942eee2013-08-08 20:11:26 -04004241 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004242void css_task_iter_start(struct cgroup_subsys_state *css,
4243 struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004244{
Tejun Heo56fde9e2014-02-13 06:58:38 -05004245 /* no one should try to iterate before mounting cgroups */
4246 WARN_ON_ONCE(!use_task_css_set_links);
Paul Menage817929e2007-10-18 23:39:36 -07004247
Tejun Heoed27b9f2015-10-15 16:41:52 -04004248 memset(it, 0, sizeof(*it));
4249
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004250 spin_lock_irq(&css_set_lock);
Tejun Heoc59cd3d2013-08-08 20:11:26 -04004251
Tejun Heo3ebb2b62014-04-23 11:13:15 -04004252 it->ss = css->ss;
4253
4254 if (it->ss)
4255 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4256 else
4257 it->cset_pos = &css->cgroup->cset_links;
4258
Tejun Heo0f0a2b42014-04-23 11:13:15 -04004259 it->cset_head = it->cset_pos;
Tejun Heoc59cd3d2013-08-08 20:11:26 -04004260
Tejun Heoecb9d532015-10-15 16:41:52 -04004261 css_task_iter_advance_css_set(it);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004262
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004263 spin_unlock_irq(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07004264}
Paul Menage817929e2007-10-18 23:39:36 -07004265
Tejun Heo0942eee2013-08-08 20:11:26 -04004266/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004267 * css_task_iter_next - return the next task for the iterator
Tejun Heo0942eee2013-08-08 20:11:26 -04004268 * @it: the task iterator being iterated
4269 *
4270 * The "next" function for task iteration. @it should have been
Tejun Heo72ec7022013-08-08 20:11:26 -04004271 * initialized via css_task_iter_start(). Returns NULL when the iteration
4272 * reaches the end.
Tejun Heo0942eee2013-08-08 20:11:26 -04004273 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004274struct task_struct *css_task_iter_next(struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004275{
Tejun Heod5745672015-10-29 11:43:05 +09004276 if (it->cur_task) {
Tejun Heoed27b9f2015-10-15 16:41:52 -04004277 put_task_struct(it->cur_task);
Tejun Heod5745672015-10-29 11:43:05 +09004278 it->cur_task = NULL;
4279 }
Tejun Heoed27b9f2015-10-15 16:41:52 -04004280
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004281 spin_lock_irq(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004282
Tejun Heod5745672015-10-29 11:43:05 +09004283 if (it->task_pos) {
4284 it->cur_task = list_entry(it->task_pos, struct task_struct,
4285 cg_list);
4286 get_task_struct(it->cur_task);
4287 css_task_iter_advance(it);
4288 }
Tejun Heoed27b9f2015-10-15 16:41:52 -04004289
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004290 spin_unlock_irq(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004291
4292 return it->cur_task;
Paul Menage817929e2007-10-18 23:39:36 -07004293}
4294
Tejun Heo0942eee2013-08-08 20:11:26 -04004295/**
Tejun Heo72ec7022013-08-08 20:11:26 -04004296 * css_task_iter_end - finish task iteration
Tejun Heo0942eee2013-08-08 20:11:26 -04004297 * @it: the task iterator to finish
4298 *
Tejun Heo72ec7022013-08-08 20:11:26 -04004299 * Finish task iteration started by css_task_iter_start().
Tejun Heo0942eee2013-08-08 20:11:26 -04004300 */
Tejun Heo72ec7022013-08-08 20:11:26 -04004301void css_task_iter_end(struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07004302{
Tejun Heoed27b9f2015-10-15 16:41:52 -04004303 if (it->cur_cset) {
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004304 spin_lock_irq(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004305 list_del(&it->iters_node);
4306 put_css_set_locked(it->cur_cset);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004307 spin_unlock_irq(&css_set_lock);
Tejun Heoed27b9f2015-10-15 16:41:52 -04004308 }
4309
4310 if (it->cur_task)
4311 put_task_struct(it->cur_task);
Tejun Heo8cc99342013-04-07 09:29:50 -07004312}
4313
4314/**
4315 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4316 * @to: cgroup to which the tasks will be moved
4317 * @from: cgroup in which the tasks currently reside
Tejun Heoeaf797a2014-02-25 10:04:03 -05004318 *
4319 * Locking rules between cgroup_post_fork() and the migration path
4320 * guarantee that, if a task is forking while being migrated, the new child
4321 * is guaranteed to be either visible in the source cgroup after the
4322 * parent's migration is complete or put into the target cgroup. No task
4323 * can slip out of migration through forking.
Tejun Heo8cc99342013-04-07 09:29:50 -07004324 */
4325int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
4326{
Tejun Heo952aaa12014-02-25 10:04:03 -05004327 LIST_HEAD(preloaded_csets);
4328 struct cgrp_cset_link *link;
Tejun Heoe406d1c2014-02-13 06:58:39 -05004329 struct css_task_iter it;
4330 struct task_struct *task;
Tejun Heo952aaa12014-02-25 10:04:03 -05004331 int ret;
Tejun Heoe406d1c2014-02-13 06:58:39 -05004332
Tejun Heo6c694c82016-03-08 11:51:25 -05004333 if (!cgroup_may_migrate_to(to))
4334 return -EBUSY;
4335
Tejun Heo952aaa12014-02-25 10:04:03 -05004336 mutex_lock(&cgroup_mutex);
4337
Eric W. Biedermaneedd0f42016-07-15 06:35:51 -05004338 percpu_down_write(&cgroup_threadgroup_rwsem);
4339
Tejun Heo952aaa12014-02-25 10:04:03 -05004340 /* all tasks in @from are being moved, all csets are source */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004341 spin_lock_irq(&css_set_lock);
Tejun Heo952aaa12014-02-25 10:04:03 -05004342 list_for_each_entry(link, &from->cset_links, cset_link)
4343 cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03004344 spin_unlock_irq(&css_set_lock);
Tejun Heo952aaa12014-02-25 10:04:03 -05004345
Tejun Heoe4857982016-03-08 11:51:26 -05004346 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
Tejun Heo952aaa12014-02-25 10:04:03 -05004347 if (ret)
4348 goto out_err;
4349
4350 /*
Rami Rosen2cfa2b12016-01-09 23:33:06 +02004351 * Migrate tasks one-by-one until @from is empty. This fails iff
Tejun Heo952aaa12014-02-25 10:04:03 -05004352 * ->can_attach() fails.
4353 */
Tejun Heoe406d1c2014-02-13 06:58:39 -05004354 do {
Tejun Heo9d800df2014-05-14 09:15:00 -04004355 css_task_iter_start(&from->self, &it);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004356 task = css_task_iter_next(&it);
4357 if (task)
4358 get_task_struct(task);
4359 css_task_iter_end(&it);
4360
4361 if (task) {
Tejun Heo37ff9f82016-03-08 11:51:26 -05004362 ret = cgroup_migrate(task, false, to->root);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004363 put_task_struct(task);
4364 }
4365 } while (task && !ret);
Tejun Heo952aaa12014-02-25 10:04:03 -05004366out_err:
4367 cgroup_migrate_finish(&preloaded_csets);
Eric W. Biedermaneedd0f42016-07-15 06:35:51 -05004368 percpu_up_write(&cgroup_threadgroup_rwsem);
Tejun Heo952aaa12014-02-25 10:04:03 -05004369 mutex_unlock(&cgroup_mutex);
Tejun Heoe406d1c2014-02-13 06:58:39 -05004370 return ret;
Tejun Heo8cc99342013-04-07 09:29:50 -07004371}
4372
Paul Menage817929e2007-10-18 23:39:36 -07004373/*
Ben Blum102a7752009-09-23 15:56:26 -07004374 * Stuff for reading the 'tasks'/'procs' files.
Paul Menagebbcb81d2007-10-18 23:39:32 -07004375 *
4376 * Reading this file can return large amounts of data if a cgroup has
4377 * *lots* of attached tasks. So it may need several calls to read(),
4378 * but we cannot guarantee that the information we produce is correct
4379 * unless we produce it entirely atomically.
4380 *
Paul Menagebbcb81d2007-10-18 23:39:32 -07004381 */
Paul Menagebbcb81d2007-10-18 23:39:32 -07004382
Li Zefan24528252012-01-20 11:58:43 +08004383/* which pidlist file are we talking about? */
4384enum cgroup_filetype {
4385 CGROUP_FILE_PROCS,
4386 CGROUP_FILE_TASKS,
4387};
4388
4389/*
4390 * A pidlist is a list of pids that virtually represents the contents of one
4391 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4392 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4393 * to the cgroup.
4394 */
4395struct cgroup_pidlist {
4396 /*
4397 * used to find which pidlist is wanted. doesn't change as long as
4398 * this particular list stays in the list.
4399 */
4400 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
4401 /* array of xids */
4402 pid_t *list;
4403 /* how many elements the above list has */
4404 int length;
Li Zefan24528252012-01-20 11:58:43 +08004405 /* each of these stored in a list by its cgroup */
4406 struct list_head links;
4407 /* pointer to the cgroup we belong to, for list removal purposes */
4408 struct cgroup *owner;
Tejun Heob1a21362013-11-29 10:42:58 -05004409 /* for delayed destruction */
4410 struct delayed_work destroy_dwork;
Li Zefan24528252012-01-20 11:58:43 +08004411};
4412
Paul Menagebbcb81d2007-10-18 23:39:32 -07004413/*
Ben Blumd1d9fd32009-09-23 15:56:28 -07004414 * The following two functions "fix" the issue where there are more pids
4415 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4416 * TODO: replace with a kernel-wide solution to this problem
4417 */
4418#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4419static void *pidlist_allocate(int count)
4420{
4421 if (PIDLIST_TOO_LARGE(count))
4422 return vmalloc(count * sizeof(pid_t));
4423 else
4424 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
4425}
Tejun Heob1a21362013-11-29 10:42:58 -05004426
Ben Blumd1d9fd32009-09-23 15:56:28 -07004427static void pidlist_free(void *p)
4428{
Bandan Das58794512015-03-02 17:51:10 -05004429 kvfree(p);
Ben Blumd1d9fd32009-09-23 15:56:28 -07004430}
Ben Blumd1d9fd32009-09-23 15:56:28 -07004431
4432/*
Tejun Heob1a21362013-11-29 10:42:58 -05004433 * Used to destroy all pidlists lingering waiting for destroy timer. None
4434 * should be left afterwards.
4435 */
4436static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
4437{
4438 struct cgroup_pidlist *l, *tmp_l;
4439
4440 mutex_lock(&cgrp->pidlist_mutex);
4441 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
4442 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
4443 mutex_unlock(&cgrp->pidlist_mutex);
4444
4445 flush_workqueue(cgroup_pidlist_destroy_wq);
4446 BUG_ON(!list_empty(&cgrp->pidlists));
4447}
4448
4449static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
4450{
4451 struct delayed_work *dwork = to_delayed_work(work);
4452 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
4453 destroy_dwork);
4454 struct cgroup_pidlist *tofree = NULL;
4455
4456 mutex_lock(&l->owner->pidlist_mutex);
Tejun Heob1a21362013-11-29 10:42:58 -05004457
4458 /*
Tejun Heo04502362013-11-29 10:42:59 -05004459 * Destroy iff we didn't get queued again. The state won't change
4460 * as destroy_dwork can only be queued while locked.
Tejun Heob1a21362013-11-29 10:42:58 -05004461 */
Tejun Heo04502362013-11-29 10:42:59 -05004462 if (!delayed_work_pending(dwork)) {
Tejun Heob1a21362013-11-29 10:42:58 -05004463 list_del(&l->links);
4464 pidlist_free(l->list);
4465 put_pid_ns(l->key.ns);
4466 tofree = l;
4467 }
4468
Tejun Heob1a21362013-11-29 10:42:58 -05004469 mutex_unlock(&l->owner->pidlist_mutex);
4470 kfree(tofree);
4471}
4472
4473/*
Ben Blum102a7752009-09-23 15:56:26 -07004474 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
Li Zefan6ee211a2013-03-12 15:36:00 -07004475 * Returns the number of unique elements.
Paul Menagebbcb81d2007-10-18 23:39:32 -07004476 */
Li Zefan6ee211a2013-03-12 15:36:00 -07004477static int pidlist_uniq(pid_t *list, int length)
Paul Menagebbcb81d2007-10-18 23:39:32 -07004478{
Ben Blum102a7752009-09-23 15:56:26 -07004479 int src, dest = 1;
Ben Blum102a7752009-09-23 15:56:26 -07004480
4481 /*
4482 * we presume the 0th element is unique, so i starts at 1. trivial
4483 * edge cases first; no work needs to be done for either
4484 */
4485 if (length == 0 || length == 1)
4486 return length;
4487 /* src and dest walk down the list; dest counts unique elements */
4488 for (src = 1; src < length; src++) {
4489 /* find next unique element */
4490 while (list[src] == list[src-1]) {
4491 src++;
4492 if (src == length)
4493 goto after;
4494 }
4495 /* dest always points to where the next unique element goes */
4496 list[dest] = list[src];
4497 dest++;
4498 }
4499after:
Ben Blum102a7752009-09-23 15:56:26 -07004500 return dest;
4501}
4502
Tejun Heoafb2bc12013-11-29 10:42:59 -05004503/*
4504 * The two pid files - task and cgroup.procs - guaranteed that the result
4505 * is sorted, which forced this whole pidlist fiasco. As pid order is
4506 * different per namespace, each namespace needs differently sorted list,
4507 * making it impossible to use, for example, single rbtree of member tasks
4508 * sorted by task pointer. As pidlists can be fairly large, allocating one
4509 * per open file is dangerous, so cgroup had to implement shared pool of
4510 * pidlists keyed by cgroup and namespace.
4511 *
4512 * All this extra complexity was caused by the original implementation
4513 * committing to an entirely unnecessary property. In the long term, we
Tejun Heoaa6ec292014-07-09 10:08:08 -04004514 * want to do away with it. Explicitly scramble sort order if on the
4515 * default hierarchy so that no such expectation exists in the new
4516 * interface.
Tejun Heoafb2bc12013-11-29 10:42:59 -05004517 *
4518 * Scrambling is done by swapping every two consecutive bits, which is
4519 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4520 */
4521static pid_t pid_fry(pid_t pid)
4522{
4523 unsigned a = pid & 0x55555555;
4524 unsigned b = pid & 0xAAAAAAAA;
4525
4526 return (a << 1) | (b >> 1);
4527}
4528
4529static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
4530{
Tejun Heoaa6ec292014-07-09 10:08:08 -04004531 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05004532 return pid_fry(pid);
4533 else
4534 return pid;
4535}
4536
Ben Blum102a7752009-09-23 15:56:26 -07004537static int cmppid(const void *a, const void *b)
4538{
4539 return *(pid_t *)a - *(pid_t *)b;
4540}
4541
Tejun Heoafb2bc12013-11-29 10:42:59 -05004542static int fried_cmppid(const void *a, const void *b)
4543{
4544 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
4545}
4546
Ben Blum72a8cb32009-09-23 15:56:27 -07004547static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
4548 enum cgroup_filetype type)
4549{
4550 struct cgroup_pidlist *l;
4551 /* don't need task_nsproxy() if we're looking at ourself */
Eric W. Biederman17cf22c2010-03-02 14:51:53 -08004552 struct pid_namespace *ns = task_active_pid_ns(current);
Li Zefanb70cc5f2010-03-10 15:22:12 -08004553
Tejun Heoe6b81712013-11-29 10:42:59 -05004554 lockdep_assert_held(&cgrp->pidlist_mutex);
4555
4556 list_for_each_entry(l, &cgrp->pidlists, links)
4557 if (l->key.type == type && l->key.ns == ns)
Ben Blum72a8cb32009-09-23 15:56:27 -07004558 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05004559 return NULL;
4560}
4561
Ben Blum72a8cb32009-09-23 15:56:27 -07004562/*
4563 * find the appropriate pidlist for our purpose (given procs vs tasks)
4564 * returns with the lock on that pidlist already held, and takes care
4565 * of the use count, or returns NULL with no locks held if we're out of
4566 * memory.
4567 */
Tejun Heoe6b81712013-11-29 10:42:59 -05004568static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
4569 enum cgroup_filetype type)
Ben Blum72a8cb32009-09-23 15:56:27 -07004570{
4571 struct cgroup_pidlist *l;
Ben Blum72a8cb32009-09-23 15:56:27 -07004572
Tejun Heoe6b81712013-11-29 10:42:59 -05004573 lockdep_assert_held(&cgrp->pidlist_mutex);
4574
4575 l = cgroup_pidlist_find(cgrp, type);
4576 if (l)
4577 return l;
4578
Ben Blum72a8cb32009-09-23 15:56:27 -07004579 /* entry not found; create a new one */
Tejun Heof4f4be22013-06-12 21:04:51 -07004580 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
Tejun Heoe6b81712013-11-29 10:42:59 -05004581 if (!l)
Ben Blum72a8cb32009-09-23 15:56:27 -07004582 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05004583
Tejun Heob1a21362013-11-29 10:42:58 -05004584 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
Ben Blum72a8cb32009-09-23 15:56:27 -07004585 l->key.type = type;
Tejun Heoe6b81712013-11-29 10:42:59 -05004586 /* don't need task_nsproxy() if we're looking at ourself */
4587 l->key.ns = get_pid_ns(task_active_pid_ns(current));
Ben Blum72a8cb32009-09-23 15:56:27 -07004588 l->owner = cgrp;
4589 list_add(&l->links, &cgrp->pidlists);
Ben Blum72a8cb32009-09-23 15:56:27 -07004590 return l;
4591}
4592
4593/*
Ben Blum102a7752009-09-23 15:56:26 -07004594 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4595 */
Ben Blum72a8cb32009-09-23 15:56:27 -07004596static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
4597 struct cgroup_pidlist **lp)
Ben Blum102a7752009-09-23 15:56:26 -07004598{
4599 pid_t *array;
4600 int length;
4601 int pid, n = 0; /* used for populating the array */
Tejun Heo72ec7022013-08-08 20:11:26 -04004602 struct css_task_iter it;
Paul Menage817929e2007-10-18 23:39:36 -07004603 struct task_struct *tsk;
Ben Blum102a7752009-09-23 15:56:26 -07004604 struct cgroup_pidlist *l;
4605
Tejun Heo4bac00d2013-11-29 10:42:59 -05004606 lockdep_assert_held(&cgrp->pidlist_mutex);
4607
Ben Blum102a7752009-09-23 15:56:26 -07004608 /*
4609 * If cgroup gets more users after we read count, we won't have
4610 * enough space - tough. This race is indistinguishable to the
4611 * caller from the case that the additional cgroup users didn't
4612 * show up until sometime later on.
4613 */
4614 length = cgroup_task_count(cgrp);
Ben Blumd1d9fd32009-09-23 15:56:28 -07004615 array = pidlist_allocate(length);
Ben Blum102a7752009-09-23 15:56:26 -07004616 if (!array)
4617 return -ENOMEM;
4618 /* now, populate the array */
Tejun Heo9d800df2014-05-14 09:15:00 -04004619 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04004620 while ((tsk = css_task_iter_next(&it))) {
Ben Blum102a7752009-09-23 15:56:26 -07004621 if (unlikely(n == length))
Paul Menage817929e2007-10-18 23:39:36 -07004622 break;
Ben Blum102a7752009-09-23 15:56:26 -07004623 /* get tgid or pid for procs or tasks file respectively */
Ben Blum72a8cb32009-09-23 15:56:27 -07004624 if (type == CGROUP_FILE_PROCS)
4625 pid = task_tgid_vnr(tsk);
4626 else
4627 pid = task_pid_vnr(tsk);
Ben Blum102a7752009-09-23 15:56:26 -07004628 if (pid > 0) /* make sure to only use valid results */
4629 array[n++] = pid;
Paul Menage817929e2007-10-18 23:39:36 -07004630 }
Tejun Heo72ec7022013-08-08 20:11:26 -04004631 css_task_iter_end(&it);
Ben Blum102a7752009-09-23 15:56:26 -07004632 length = n;
4633 /* now sort & (if procs) strip out duplicates */
Tejun Heoaa6ec292014-07-09 10:08:08 -04004634 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05004635 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
4636 else
4637 sort(array, length, sizeof(pid_t), cmppid, NULL);
Ben Blum72a8cb32009-09-23 15:56:27 -07004638 if (type == CGROUP_FILE_PROCS)
Li Zefan6ee211a2013-03-12 15:36:00 -07004639 length = pidlist_uniq(array, length);
Tejun Heoe6b81712013-11-29 10:42:59 -05004640
Tejun Heoe6b81712013-11-29 10:42:59 -05004641 l = cgroup_pidlist_find_create(cgrp, type);
Ben Blum72a8cb32009-09-23 15:56:27 -07004642 if (!l) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07004643 pidlist_free(array);
Ben Blum72a8cb32009-09-23 15:56:27 -07004644 return -ENOMEM;
Ben Blum102a7752009-09-23 15:56:26 -07004645 }
Tejun Heoe6b81712013-11-29 10:42:59 -05004646
4647 /* store array, freeing old if necessary */
Ben Blumd1d9fd32009-09-23 15:56:28 -07004648 pidlist_free(l->list);
Ben Blum102a7752009-09-23 15:56:26 -07004649 l->list = array;
4650 l->length = length;
Ben Blum72a8cb32009-09-23 15:56:27 -07004651 *lp = l;
Ben Blum102a7752009-09-23 15:56:26 -07004652 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004653}
4654
Balbir Singh846c7bb2007-10-18 23:39:44 -07004655/**
Li Zefana043e3b2008-02-23 15:24:09 -08004656 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07004657 * @stats: cgroupstats to fill information into
4658 * @dentry: A dentry entry belonging to the cgroup for which stats have
4659 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08004660 *
4661 * Build and fill cgroupstats so that taskstats can export it to user
4662 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07004663 */
4664int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
4665{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004666 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07004667 struct cgroup *cgrp;
Tejun Heo72ec7022013-08-08 20:11:26 -04004668 struct css_task_iter it;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004669 struct task_struct *tsk;
Li Zefan33d283b2008-11-19 15:36:48 -08004670
Tejun Heo2bd59d42014-02-11 11:52:49 -05004671 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4672 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4673 kernfs_type(kn) != KERNFS_DIR)
4674 return -EINVAL;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004675
Li Zefanbad34662014-02-14 16:54:28 +08004676 mutex_lock(&cgroup_mutex);
4677
Tejun Heo2bd59d42014-02-11 11:52:49 -05004678 /*
4679 * We aren't being called from kernfs and there's no guarantee on
Tejun Heoec903c02014-05-13 12:11:01 -04004680 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
Tejun Heo2bd59d42014-02-11 11:52:49 -05004681 * @kn->priv is RCU safe. Let's do the RCU dancing.
4682 */
4683 rcu_read_lock();
4684 cgrp = rcu_dereference(kn->priv);
Li Zefanbad34662014-02-14 16:54:28 +08004685 if (!cgrp || cgroup_is_dead(cgrp)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05004686 rcu_read_unlock();
Li Zefanbad34662014-02-14 16:54:28 +08004687 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004688 return -ENOENT;
4689 }
Li Zefanbad34662014-02-14 16:54:28 +08004690 rcu_read_unlock();
Balbir Singh846c7bb2007-10-18 23:39:44 -07004691
Tejun Heo9d800df2014-05-14 09:15:00 -04004692 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04004693 while ((tsk = css_task_iter_next(&it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07004694 switch (tsk->state) {
4695 case TASK_RUNNING:
4696 stats->nr_running++;
4697 break;
4698 case TASK_INTERRUPTIBLE:
4699 stats->nr_sleeping++;
4700 break;
4701 case TASK_UNINTERRUPTIBLE:
4702 stats->nr_uninterruptible++;
4703 break;
4704 case TASK_STOPPED:
4705 stats->nr_stopped++;
4706 break;
4707 default:
4708 if (delayacct_is_task_waiting_on_io(tsk))
4709 stats->nr_io_wait++;
4710 break;
4711 }
4712 }
Tejun Heo72ec7022013-08-08 20:11:26 -04004713 css_task_iter_end(&it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07004714
Li Zefanbad34662014-02-14 16:54:28 +08004715 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004716 return 0;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004717}
4718
Paul Menage8f3ff202009-09-23 15:56:25 -07004719
Paul Menagecc31edc2008-10-18 20:28:04 -07004720/*
Ben Blum102a7752009-09-23 15:56:26 -07004721 * seq_file methods for the tasks/procs files. The seq_file position is the
Paul Menagecc31edc2008-10-18 20:28:04 -07004722 * next pid to display; the seq_file iterator is a pointer to the pid
Ben Blum102a7752009-09-23 15:56:26 -07004723 * in the cgroup->l->list array.
Paul Menagecc31edc2008-10-18 20:28:04 -07004724 */
4725
Ben Blum102a7752009-09-23 15:56:26 -07004726static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004727{
4728 /*
4729 * Initially we receive a position value that corresponds to
4730 * one more than the last pid shown (or 0 on the first call or
4731 * after a seek to the start). Use a binary-search to find the
4732 * next pid to display, if any
4733 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05004734 struct kernfs_open_file *of = s->private;
Tejun Heo7da11272013-12-05 12:28:04 -05004735 struct cgroup *cgrp = seq_css(s)->cgroup;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004736 struct cgroup_pidlist *l;
Tejun Heo7da11272013-12-05 12:28:04 -05004737 enum cgroup_filetype type = seq_cft(s)->private;
Paul Menagecc31edc2008-10-18 20:28:04 -07004738 int index = 0, pid = *pos;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004739 int *iter, ret;
Paul Menagecc31edc2008-10-18 20:28:04 -07004740
Tejun Heo4bac00d2013-11-29 10:42:59 -05004741 mutex_lock(&cgrp->pidlist_mutex);
4742
4743 /*
Tejun Heo5d224442013-12-05 12:28:04 -05004744 * !NULL @of->priv indicates that this isn't the first start()
Tejun Heo4bac00d2013-11-29 10:42:59 -05004745 * after open. If the matching pidlist is around, we can use that.
Tejun Heo5d224442013-12-05 12:28:04 -05004746 * Look for it. Note that @of->priv can't be used directly. It
Tejun Heo4bac00d2013-11-29 10:42:59 -05004747 * could already have been destroyed.
4748 */
Tejun Heo5d224442013-12-05 12:28:04 -05004749 if (of->priv)
4750 of->priv = cgroup_pidlist_find(cgrp, type);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004751
4752 /*
4753 * Either this is the first start() after open or the matching
4754 * pidlist has been destroyed inbetween. Create a new one.
4755 */
Tejun Heo5d224442013-12-05 12:28:04 -05004756 if (!of->priv) {
4757 ret = pidlist_array_load(cgrp, type,
4758 (struct cgroup_pidlist **)&of->priv);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004759 if (ret)
4760 return ERR_PTR(ret);
4761 }
Tejun Heo5d224442013-12-05 12:28:04 -05004762 l = of->priv;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004763
Paul Menagecc31edc2008-10-18 20:28:04 -07004764 if (pid) {
Ben Blum102a7752009-09-23 15:56:26 -07004765 int end = l->length;
Stephen Rothwell20777762008-10-21 16:11:20 +11004766
Paul Menagecc31edc2008-10-18 20:28:04 -07004767 while (index < end) {
4768 int mid = (index + end) / 2;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004769 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
Paul Menagecc31edc2008-10-18 20:28:04 -07004770 index = mid;
4771 break;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004772 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
Paul Menagecc31edc2008-10-18 20:28:04 -07004773 index = mid + 1;
4774 else
4775 end = mid;
4776 }
4777 }
4778 /* If we're off the end of the array, we're done */
Ben Blum102a7752009-09-23 15:56:26 -07004779 if (index >= l->length)
Paul Menagecc31edc2008-10-18 20:28:04 -07004780 return NULL;
4781 /* Update the abstract position to be the actual pid that we found */
Ben Blum102a7752009-09-23 15:56:26 -07004782 iter = l->list + index;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004783 *pos = cgroup_pid_fry(cgrp, *iter);
Paul Menagecc31edc2008-10-18 20:28:04 -07004784 return iter;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004785}
4786
Ben Blum102a7752009-09-23 15:56:26 -07004787static void cgroup_pidlist_stop(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004788{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004789 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004790 struct cgroup_pidlist *l = of->priv;
Tejun Heo62236852013-11-29 10:42:58 -05004791
Tejun Heo5d224442013-12-05 12:28:04 -05004792 if (l)
4793 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
Tejun Heo04502362013-11-29 10:42:59 -05004794 CGROUP_PIDLIST_DESTROY_DELAY);
Tejun Heo7da11272013-12-05 12:28:04 -05004795 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07004796}
4797
Ben Blum102a7752009-09-23 15:56:26 -07004798static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004799{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004800 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004801 struct cgroup_pidlist *l = of->priv;
Ben Blum102a7752009-09-23 15:56:26 -07004802 pid_t *p = v;
4803 pid_t *end = l->list + l->length;
Paul Menagecc31edc2008-10-18 20:28:04 -07004804 /*
4805 * Advance to the next pid in the array. If this goes off the
4806 * end, we're done
4807 */
4808 p++;
4809 if (p >= end) {
4810 return NULL;
4811 } else {
Tejun Heo7da11272013-12-05 12:28:04 -05004812 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
Paul Menagecc31edc2008-10-18 20:28:04 -07004813 return p;
4814 }
4815}
4816
Ben Blum102a7752009-09-23 15:56:26 -07004817static int cgroup_pidlist_show(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004818{
Joe Perches94ff2122015-04-15 16:18:20 -07004819 seq_printf(s, "%d\n", *(int *)v);
4820
4821 return 0;
Paul Menagecc31edc2008-10-18 20:28:04 -07004822}
4823
Tejun Heo182446d2013-08-08 20:11:24 -04004824static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4825 struct cftype *cft)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004826{
Tejun Heo182446d2013-08-08 20:11:24 -04004827 return notify_on_release(css->cgroup);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004828}
4829
Tejun Heo182446d2013-08-08 20:11:24 -04004830static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4831 struct cftype *cft, u64 val)
Paul Menage6379c102008-07-25 01:47:01 -07004832{
Paul Menage6379c102008-07-25 01:47:01 -07004833 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004834 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004835 else
Tejun Heo182446d2013-08-08 20:11:24 -04004836 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004837 return 0;
4838}
4839
Tejun Heo182446d2013-08-08 20:11:24 -04004840static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4841 struct cftype *cft)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004842{
Tejun Heo182446d2013-08-08 20:11:24 -04004843 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004844}
4845
Tejun Heo182446d2013-08-08 20:11:24 -04004846static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4847 struct cftype *cft, u64 val)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004848{
4849 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004850 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004851 else
Tejun Heo182446d2013-08-08 20:11:24 -04004852 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004853 return 0;
4854}
4855
Tejun Heoa14c6872014-07-15 11:05:09 -04004856/* cgroup core interface files for the default hierarchy */
4857static struct cftype cgroup_dfl_base_files[] = {
4858 {
4859 .name = "cgroup.procs",
Tejun Heo6f60ead2015-09-18 17:54:23 -04004860 .file_offset = offsetof(struct cgroup, procs_file),
Tejun Heoa14c6872014-07-15 11:05:09 -04004861 .seq_start = cgroup_pidlist_start,
4862 .seq_next = cgroup_pidlist_next,
4863 .seq_stop = cgroup_pidlist_stop,
4864 .seq_show = cgroup_pidlist_show,
4865 .private = CGROUP_FILE_PROCS,
4866 .write = cgroup_procs_write,
Tejun Heoa14c6872014-07-15 11:05:09 -04004867 },
4868 {
4869 .name = "cgroup.controllers",
Tejun Heoa14c6872014-07-15 11:05:09 -04004870 .seq_show = cgroup_controllers_show,
4871 },
4872 {
4873 .name = "cgroup.subtree_control",
4874 .seq_show = cgroup_subtree_control_show,
4875 .write = cgroup_subtree_control_write,
4876 },
4877 {
Tejun Heo4a07c222015-09-18 17:54:22 -04004878 .name = "cgroup.events",
Tejun Heoa14c6872014-07-15 11:05:09 -04004879 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo6f60ead2015-09-18 17:54:23 -04004880 .file_offset = offsetof(struct cgroup, events_file),
Tejun Heo4a07c222015-09-18 17:54:22 -04004881 .seq_show = cgroup_events_show,
Tejun Heoa14c6872014-07-15 11:05:09 -04004882 },
4883 { } /* terminate */
4884};
4885
4886/* cgroup core interface files for the legacy hierarchies */
4887static struct cftype cgroup_legacy_base_files[] = {
Paul Menage81a6a5c2007-10-18 23:39:38 -07004888 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004889 .name = "cgroup.procs",
Tejun Heo6612f052013-12-05 12:28:04 -05004890 .seq_start = cgroup_pidlist_start,
4891 .seq_next = cgroup_pidlist_next,
4892 .seq_stop = cgroup_pidlist_stop,
4893 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004894 .private = CGROUP_FILE_PROCS,
Tejun Heoacbef752014-05-13 12:16:22 -04004895 .write = cgroup_procs_write,
Ben Blum102a7752009-09-23 15:56:26 -07004896 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07004897 {
Daniel Lezcano97978e62010-10-27 15:33:35 -07004898 .name = "cgroup.clone_children",
4899 .read_u64 = cgroup_clone_children_read,
4900 .write_u64 = cgroup_clone_children_write,
4901 },
Tejun Heo6e6ff252012-04-01 12:09:55 -07004902 {
Tejun Heo873fe092013-04-14 20:15:26 -07004903 .name = "cgroup.sane_behavior",
4904 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004905 .seq_show = cgroup_sane_behavior_show,
Tejun Heo873fe092013-04-14 20:15:26 -07004906 },
Tejun Heof8f22e52014-04-23 11:13:16 -04004907 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004908 .name = "tasks",
Tejun Heo6612f052013-12-05 12:28:04 -05004909 .seq_start = cgroup_pidlist_start,
4910 .seq_next = cgroup_pidlist_next,
4911 .seq_stop = cgroup_pidlist_stop,
4912 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004913 .private = CGROUP_FILE_TASKS,
Tejun Heoacbef752014-05-13 12:16:22 -04004914 .write = cgroup_tasks_write,
Tejun Heod5c56ce2013-06-03 19:14:34 -07004915 },
4916 {
4917 .name = "notify_on_release",
Tejun Heod5c56ce2013-06-03 19:14:34 -07004918 .read_u64 = cgroup_read_notify_on_release,
4919 .write_u64 = cgroup_write_notify_on_release,
4920 },
Tejun Heo873fe092013-04-14 20:15:26 -07004921 {
Tejun Heo6e6ff252012-04-01 12:09:55 -07004922 .name = "release_agent",
Tejun Heoa14c6872014-07-15 11:05:09 -04004923 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004924 .seq_show = cgroup_release_agent_show,
Tejun Heo451af5042014-05-13 12:16:21 -04004925 .write = cgroup_release_agent_write,
Tejun Heo5f469902014-02-11 11:52:48 -05004926 .max_write_len = PATH_MAX - 1,
Tejun Heo6e6ff252012-04-01 12:09:55 -07004927 },
Tejun Heodb0416b2012-04-01 12:09:55 -07004928 { } /* terminate */
Paul Menagebbcb81d2007-10-18 23:39:32 -07004929};
4930
Tejun Heo0c21ead2013-08-13 20:22:51 -04004931/*
4932 * css destruction is four-stage process.
4933 *
4934 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4935 * Implemented in kill_css().
4936 *
4937 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
Tejun Heoec903c02014-05-13 12:11:01 -04004938 * and thus css_tryget_online() is guaranteed to fail, the css can be
4939 * offlined by invoking offline_css(). After offlining, the base ref is
4940 * put. Implemented in css_killed_work_fn().
Tejun Heo0c21ead2013-08-13 20:22:51 -04004941 *
4942 * 3. When the percpu_ref reaches zero, the only possible remaining
4943 * accessors are inside RCU read sections. css_release() schedules the
4944 * RCU callback.
4945 *
4946 * 4. After the grace period, the css can be freed. Implemented in
4947 * css_free_work_fn().
4948 *
4949 * It is actually hairier because both step 2 and 4 require process context
4950 * and thus involve punting to css->destroy_work adding two additional
4951 * steps to the already complex sequence.
4952 */
Tejun Heo35ef10d2013-08-13 11:01:54 -04004953static void css_free_work_fn(struct work_struct *work)
Tejun Heo48ddbe12012-04-01 12:09:56 -07004954{
4955 struct cgroup_subsys_state *css =
Tejun Heo35ef10d2013-08-13 11:01:54 -04004956 container_of(work, struct cgroup_subsys_state, destroy_work);
Vladimir Davydov01e58652015-02-12 14:59:26 -08004957 struct cgroup_subsys *ss = css->ss;
Tejun Heo0c21ead2013-08-13 20:22:51 -04004958 struct cgroup *cgrp = css->cgroup;
Tejun Heo48ddbe12012-04-01 12:09:56 -07004959
Tejun Heo9a1049d2014-06-28 08:10:14 -04004960 percpu_ref_exit(&css->refcnt);
4961
Vladimir Davydov01e58652015-02-12 14:59:26 -08004962 if (ss) {
Tejun Heo9d755d32014-05-14 09:15:02 -04004963 /* css free path */
Tejun Heo8bb5ef72016-01-21 15:32:15 -05004964 struct cgroup_subsys_state *parent = css->parent;
Vladimir Davydov01e58652015-02-12 14:59:26 -08004965 int id = css->id;
4966
Vladimir Davydov01e58652015-02-12 14:59:26 -08004967 ss->css_free(css);
4968 cgroup_idr_remove(&ss->css_idr, id);
Tejun Heo9d755d32014-05-14 09:15:02 -04004969 cgroup_put(cgrp);
Tejun Heo8bb5ef72016-01-21 15:32:15 -05004970
4971 if (parent)
4972 css_put(parent);
Tejun Heo9d755d32014-05-14 09:15:02 -04004973 } else {
4974 /* cgroup free path */
4975 atomic_dec(&cgrp->root->nr_cgrps);
4976 cgroup_pidlist_destroy_all(cgrp);
Zefan Li971ff492014-09-18 16:06:19 +08004977 cancel_work_sync(&cgrp->release_agent_work);
Tejun Heo9d755d32014-05-14 09:15:02 -04004978
Tejun Heod51f39b2014-05-16 13:22:48 -04004979 if (cgroup_parent(cgrp)) {
Tejun Heo9d755d32014-05-14 09:15:02 -04004980 /*
4981 * We get a ref to the parent, and put the ref when
4982 * this cgroup is being freed, so it's guaranteed
4983 * that the parent won't be destroyed before its
4984 * children.
4985 */
Tejun Heod51f39b2014-05-16 13:22:48 -04004986 cgroup_put(cgroup_parent(cgrp));
Tejun Heo9d755d32014-05-14 09:15:02 -04004987 kernfs_put(cgrp->kn);
4988 kfree(cgrp);
4989 } else {
4990 /*
4991 * This is root cgroup's refcnt reaching zero,
4992 * which indicates that the root should be
4993 * released.
4994 */
4995 cgroup_destroy_root(cgrp->root);
4996 }
4997 }
Tejun Heo0c21ead2013-08-13 20:22:51 -04004998}
4999
5000static void css_free_rcu_fn(struct rcu_head *rcu_head)
5001{
5002 struct cgroup_subsys_state *css =
5003 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
5004
Tejun Heo0c21ead2013-08-13 20:22:51 -04005005 INIT_WORK(&css->destroy_work, css_free_work_fn);
Tejun Heoe5fca242013-11-22 17:14:39 -05005006 queue_work(cgroup_destroy_wq, &css->destroy_work);
Tejun Heo48ddbe12012-04-01 12:09:56 -07005007}
5008
Tejun Heo25e15d82014-05-14 09:15:02 -04005009static void css_release_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07005010{
5011 struct cgroup_subsys_state *css =
Tejun Heo25e15d82014-05-14 09:15:02 -04005012 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heo15a4c832014-05-04 15:09:14 -04005013 struct cgroup_subsys *ss = css->ss;
Tejun Heo9d755d32014-05-14 09:15:02 -04005014 struct cgroup *cgrp = css->cgroup;
Tejun Heod3daf282013-06-13 19:39:16 -07005015
Tejun Heo1fed1b22014-05-16 13:22:49 -04005016 mutex_lock(&cgroup_mutex);
5017
Tejun Heode3f0342014-05-16 13:22:49 -04005018 css->flags |= CSS_RELEASED;
Tejun Heo1fed1b22014-05-16 13:22:49 -04005019 list_del_rcu(&css->sibling);
5020
Tejun Heo9d755d32014-05-14 09:15:02 -04005021 if (ss) {
5022 /* css release path */
Vladimir Davydov01e58652015-02-12 14:59:26 -08005023 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
Tejun Heo7d172cc2014-11-18 02:49:51 -05005024 if (ss->css_released)
5025 ss->css_released(css);
Tejun Heo9d755d32014-05-14 09:15:02 -04005026 } else {
5027 /* cgroup release path */
Tejun Heo9d755d32014-05-14 09:15:02 -04005028 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
5029 cgrp->id = -1;
Li Zefana4189482014-09-04 14:43:07 +08005030
5031 /*
5032 * There are two control paths which try to determine
5033 * cgroup from dentry without going through kernfs -
5034 * cgroupstats_build() and css_tryget_online_from_dir().
5035 * Those are supported by RCU protecting clearing of
5036 * cgrp->kn->priv backpointer.
5037 */
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005038 if (cgrp->kn)
5039 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5040 NULL);
Tejun Heo9d755d32014-05-14 09:15:02 -04005041 }
Tejun Heo15a4c832014-05-04 15:09:14 -04005042
Tejun Heo1fed1b22014-05-16 13:22:49 -04005043 mutex_unlock(&cgroup_mutex);
5044
Tejun Heo0c21ead2013-08-13 20:22:51 -04005045 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07005046}
5047
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07005048static void css_release(struct percpu_ref *ref)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005049{
5050 struct cgroup_subsys_state *css =
5051 container_of(ref, struct cgroup_subsys_state, refcnt);
5052
Tejun Heo25e15d82014-05-14 09:15:02 -04005053 INIT_WORK(&css->destroy_work, css_release_work_fn);
5054 queue_work(cgroup_destroy_wq, &css->destroy_work);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005055}
5056
Tejun Heoddfcada2014-05-04 15:09:14 -04005057static void init_and_link_css(struct cgroup_subsys_state *css,
5058 struct cgroup_subsys *ss, struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005059{
Tejun Heo0cb51d72014-05-16 13:22:49 -04005060 lockdep_assert_held(&cgroup_mutex);
5061
Tejun Heoddfcada2014-05-04 15:09:14 -04005062 cgroup_get(cgrp);
5063
Tejun Heod5c419b2014-05-16 13:22:48 -04005064 memset(css, 0, sizeof(*css));
Paul Menagebd89aab2007-10-18 23:40:44 -07005065 css->cgroup = cgrp;
Tejun Heo72c97e52013-08-08 20:11:22 -04005066 css->ss = ss;
Tejun Heo8fa3b8d2016-05-26 15:42:13 -04005067 css->id = -1;
Tejun Heod5c419b2014-05-16 13:22:48 -04005068 INIT_LIST_HEAD(&css->sibling);
5069 INIT_LIST_HEAD(&css->children);
Tejun Heo0cb51d72014-05-16 13:22:49 -04005070 css->serial_nr = css_serial_nr_next++;
Tejun Heoaa226ff2016-01-21 15:31:11 -05005071 atomic_set(&css->online_cnt, 0);
Tejun Heo48ddbe12012-04-01 12:09:56 -07005072
Tejun Heod51f39b2014-05-16 13:22:48 -04005073 if (cgroup_parent(cgrp)) {
5074 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
Tejun Heoddfcada2014-05-04 15:09:14 -04005075 css_get(css->parent);
Tejun Heoddfcada2014-05-04 15:09:14 -04005076 }
Tejun Heo0ae78e02013-08-13 11:01:54 -04005077
Tejun Heoca8bdca2013-08-26 18:40:56 -04005078 BUG_ON(cgroup_css(cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005079}
5080
Li Zefan2a4ac632013-07-31 16:16:40 +08005081/* invoke ->css_online() on a new CSS and mark it online if successful */
Tejun Heo623f9262013-08-13 11:01:55 -04005082static int online_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08005083{
Tejun Heo623f9262013-08-13 11:01:55 -04005084 struct cgroup_subsys *ss = css->ss;
Tejun Heob1929db2012-11-19 08:13:38 -08005085 int ret = 0;
5086
Tejun Heoa31f2d32012-11-19 08:13:37 -08005087 lockdep_assert_held(&cgroup_mutex);
5088
Tejun Heo92fb9742012-11-19 08:13:38 -08005089 if (ss->css_online)
Tejun Heoeb954192013-08-08 20:11:23 -04005090 ret = ss->css_online(css);
Tejun Heoae7f1642013-08-13 20:22:50 -04005091 if (!ret) {
Tejun Heoeb954192013-08-08 20:11:23 -04005092 css->flags |= CSS_ONLINE;
Tejun Heoaec25022014-02-08 10:36:58 -05005093 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
Tejun Heoaa226ff2016-01-21 15:31:11 -05005094
5095 atomic_inc(&css->online_cnt);
5096 if (css->parent)
5097 atomic_inc(&css->parent->online_cnt);
Tejun Heoae7f1642013-08-13 20:22:50 -04005098 }
Tejun Heob1929db2012-11-19 08:13:38 -08005099 return ret;
Tejun Heoa31f2d32012-11-19 08:13:37 -08005100}
5101
Li Zefan2a4ac632013-07-31 16:16:40 +08005102/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
Tejun Heo623f9262013-08-13 11:01:55 -04005103static void offline_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08005104{
Tejun Heo623f9262013-08-13 11:01:55 -04005105 struct cgroup_subsys *ss = css->ss;
Tejun Heoa31f2d32012-11-19 08:13:37 -08005106
5107 lockdep_assert_held(&cgroup_mutex);
5108
5109 if (!(css->flags & CSS_ONLINE))
5110 return;
5111
Vladimir Davydovfa062352016-03-01 19:56:30 +03005112 if (ss->css_reset)
5113 ss->css_reset(css);
5114
Li Zefand7eeac12013-03-12 15:35:59 -07005115 if (ss->css_offline)
Tejun Heoeb954192013-08-08 20:11:23 -04005116 ss->css_offline(css);
Tejun Heoa31f2d32012-11-19 08:13:37 -08005117
Tejun Heoeb954192013-08-08 20:11:23 -04005118 css->flags &= ~CSS_ONLINE;
Tejun Heoe3297802014-04-23 11:13:15 -04005119 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
Tejun Heof8f22e52014-04-23 11:13:16 -04005120
5121 wake_up_all(&css->cgroup->offline_waitq);
Tejun Heoa31f2d32012-11-19 08:13:37 -08005122}
5123
Tejun Heoc81c925a2013-12-06 15:11:56 -05005124/**
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005125 * css_create - create a cgroup_subsys_state
Tejun Heoc81c925a2013-12-06 15:11:56 -05005126 * @cgrp: the cgroup new css will be associated with
5127 * @ss: the subsys of new css
5128 *
5129 * Create a new css associated with @cgrp - @ss pair. On success, the new
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005130 * css is online and installed in @cgrp. This function doesn't create the
5131 * interface files. Returns 0 on success, -errno on failure.
Tejun Heoc81c925a2013-12-06 15:11:56 -05005132 */
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005133static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5134 struct cgroup_subsys *ss)
Tejun Heoc81c925a2013-12-06 15:11:56 -05005135{
Tejun Heod51f39b2014-05-16 13:22:48 -04005136 struct cgroup *parent = cgroup_parent(cgrp);
Tejun Heo1fed1b22014-05-16 13:22:49 -04005137 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005138 struct cgroup_subsys_state *css;
5139 int err;
5140
Tejun Heoc81c925a2013-12-06 15:11:56 -05005141 lockdep_assert_held(&cgroup_mutex);
5142
Tejun Heo1fed1b22014-05-16 13:22:49 -04005143 css = ss->css_alloc(parent_css);
Tejun Heoe7e15b82016-06-21 13:06:24 -04005144 if (!css)
5145 css = ERR_PTR(-ENOMEM);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005146 if (IS_ERR(css))
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005147 return css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005148
Tejun Heoddfcada2014-05-04 15:09:14 -04005149 init_and_link_css(css, ss, cgrp);
Tejun Heoa2bed822014-05-04 15:09:14 -04005150
Tejun Heo2aad2a82014-09-24 13:31:50 -04005151 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005152 if (err)
Li Zefan3eb59ec2014-03-18 17:02:36 +08005153 goto err_free_css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005154
Vladimir Davydovcf780b72015-08-03 15:32:26 +03005155 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
Tejun Heo15a4c832014-05-04 15:09:14 -04005156 if (err < 0)
Wenwei Taob00c52d2016-05-13 22:59:20 +08005157 goto err_free_css;
Tejun Heo15a4c832014-05-04 15:09:14 -04005158 css->id = err;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005159
Tejun Heo15a4c832014-05-04 15:09:14 -04005160 /* @css is ready to be brought online now, make it visible */
Tejun Heo1fed1b22014-05-16 13:22:49 -04005161 list_add_tail_rcu(&css->sibling, &parent_css->children);
Tejun Heo15a4c832014-05-04 15:09:14 -04005162 cgroup_idr_replace(&ss->css_idr, css, css->id);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005163
5164 err = online_css(css);
5165 if (err)
Tejun Heo1fed1b22014-05-16 13:22:49 -04005166 goto err_list_del;
Tejun Heo94419622014-03-19 10:23:54 -04005167
Tejun Heoc81c925a2013-12-06 15:11:56 -05005168 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
Tejun Heod51f39b2014-05-16 13:22:48 -04005169 cgroup_parent(parent)) {
Joe Perchesed3d2612014-04-25 18:28:03 -04005170 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 -04005171 current->comm, current->pid, ss->name);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005172 if (!strcmp(ss->name, "memory"))
Joe Perchesed3d2612014-04-25 18:28:03 -04005173 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
Tejun Heoc81c925a2013-12-06 15:11:56 -05005174 ss->warned_broken_hierarchy = true;
5175 }
5176
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005177 return css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05005178
Tejun Heo1fed1b22014-05-16 13:22:49 -04005179err_list_del:
5180 list_del_rcu(&css->sibling);
Li Zefan3eb59ec2014-03-18 17:02:36 +08005181err_free_css:
Tejun Heoa2bed822014-05-04 15:09:14 -04005182 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heo6cd0f5b2016-03-03 09:57:58 -05005183 return ERR_PTR(err);
Tejun Heoc81c925a2013-12-06 15:11:56 -05005184}
5185
Tejun Heoa5bca212016-03-03 09:57:58 -05005186static struct cgroup *cgroup_create(struct cgroup *parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005187{
Tejun Heoa5bca212016-03-03 09:57:58 -05005188 struct cgroup_root *root = parent->root;
Tejun Heoa5bca212016-03-03 09:57:58 -05005189 struct cgroup *cgrp, *tcgrp;
5190 int level = parent->level + 1;
Tejun Heo03970d32016-03-03 09:58:00 -05005191 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005192
Tejun Heo0a950f62012-11-19 09:02:12 -08005193 /* allocate the cgroup and its ID, 0 is reserved for the root */
Tejun Heob11cfb52015-11-20 15:55:52 -05005194 cgrp = kzalloc(sizeof(*cgrp) +
5195 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
Tejun Heoa5bca212016-03-03 09:57:58 -05005196 if (!cgrp)
5197 return ERR_PTR(-ENOMEM);
Li Zefan0ab02ca2014-02-11 16:05:46 +08005198
Tejun Heo2aad2a82014-09-24 13:31:50 -04005199 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04005200 if (ret)
5201 goto out_free_cgrp;
5202
Li Zefan0ab02ca2014-02-11 16:05:46 +08005203 /*
5204 * Temporarily set the pointer to NULL, so idr_find() won't return
5205 * a half-baked cgroup.
5206 */
Vladimir Davydovcf780b72015-08-03 15:32:26 +03005207 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
Li Zefan0ab02ca2014-02-11 16:05:46 +08005208 if (cgrp->id < 0) {
Tejun Heoba0f4d72014-05-13 12:19:22 -04005209 ret = -ENOMEM;
Tejun Heo9d755d32014-05-14 09:15:02 -04005210 goto out_cancel_ref;
Tejun Heo976c06b2012-11-05 09:16:59 -08005211 }
5212
Paul Menagecc31edc2008-10-18 20:28:04 -07005213 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005214
Tejun Heo9d800df2014-05-14 09:15:00 -04005215 cgrp->self.parent = &parent->self;
Tejun Heoba0f4d72014-05-13 12:19:22 -04005216 cgrp->root = root;
Tejun Heob11cfb52015-11-20 15:55:52 -05005217 cgrp->level = level;
5218
5219 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
5220 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005221
Li Zefanb6abdb02008-03-04 14:28:19 -08005222 if (notify_on_release(parent))
5223 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5224
Tejun Heo2260e7f2012-11-19 08:13:38 -08005225 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5226 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07005227
Tejun Heo0cb51d72014-05-16 13:22:49 -04005228 cgrp->self.serial_nr = css_serial_nr_next++;
Tejun Heo53fa5262013-05-24 10:55:38 +09005229
Tejun Heo4e139af2012-11-19 08:13:36 -08005230 /* allocation complete, commit to creation */
Tejun Heod5c419b2014-05-16 13:22:48 -04005231 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
Tejun Heo3c9c8252014-02-12 09:29:50 -05005232 atomic_inc(&root->nr_cgrps);
Tejun Heo59f52962014-02-11 11:52:49 -05005233 cgroup_get(parent);
Li Zefan415cf072013-04-08 14:35:02 +08005234
Tejun Heo0d802552013-12-06 15:11:56 -05005235 /*
5236 * @cgrp is now fully operational. If something fails after this
5237 * point, it'll be released via the normal destruction path.
5238 */
Tejun Heo6fa49182014-05-04 15:09:13 -04005239 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08005240
Tejun Heobd53d612014-04-23 11:13:16 -04005241 /*
5242 * On the default hierarchy, a child doesn't automatically inherit
Tejun Heo667c2492014-07-08 18:02:56 -04005243 * subtree_control from the parent. Each is configured manually.
Tejun Heobd53d612014-04-23 11:13:16 -04005244 */
Tejun Heo03970d32016-03-03 09:58:00 -05005245 if (!cgroup_on_dfl(cgrp))
Tejun Heo5531dc92016-03-03 09:57:58 -05005246 cgrp->subtree_control = cgroup_control(cgrp);
Tejun Heo03970d32016-03-03 09:58:00 -05005247
5248 cgroup_propagate_control(cgrp);
5249
5250 /* @cgrp doesn't have dir yet so the following will only create csses */
5251 ret = cgroup_apply_control_enable(cgrp);
5252 if (ret)
5253 goto out_destroy;
Tejun Heof392e512014-04-23 11:13:14 -04005254
Tejun Heoa5bca212016-03-03 09:57:58 -05005255 return cgrp;
5256
5257out_cancel_ref:
5258 percpu_ref_exit(&cgrp->self.refcnt);
5259out_free_cgrp:
5260 kfree(cgrp);
5261 return ERR_PTR(ret);
5262out_destroy:
5263 cgroup_destroy_locked(cgrp);
5264 return ERR_PTR(ret);
5265}
5266
5267static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
5268 umode_t mode)
5269{
5270 struct cgroup *parent, *cgrp;
Tejun Heoa5bca212016-03-03 09:57:58 -05005271 struct kernfs_node *kn;
Tejun Heo03970d32016-03-03 09:58:00 -05005272 int ret;
Tejun Heoa5bca212016-03-03 09:57:58 -05005273
5274 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5275 if (strchr(name, '\n'))
5276 return -EINVAL;
5277
Tejun Heo945ba192016-03-03 09:58:00 -05005278 parent = cgroup_kn_lock_live(parent_kn, false);
Tejun Heoa5bca212016-03-03 09:57:58 -05005279 if (!parent)
5280 return -ENODEV;
5281
5282 cgrp = cgroup_create(parent);
5283 if (IS_ERR(cgrp)) {
5284 ret = PTR_ERR(cgrp);
5285 goto out_unlock;
5286 }
5287
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005288 /* create the directory */
5289 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5290 if (IS_ERR(kn)) {
5291 ret = PTR_ERR(kn);
5292 goto out_destroy;
5293 }
5294 cgrp->kn = kn;
5295
5296 /*
5297 * This extra ref will be put in cgroup_free_fn() and guarantees
5298 * that @cgrp->kn is always accessible.
5299 */
5300 kernfs_get(kn);
5301
5302 ret = cgroup_kn_set_ugid(kn);
5303 if (ret)
5304 goto out_destroy;
5305
Tejun Heo334c3672016-03-03 09:58:01 -05005306 ret = css_populate_dir(&cgrp->self);
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005307 if (ret)
5308 goto out_destroy;
5309
Tejun Heo03970d32016-03-03 09:58:00 -05005310 ret = cgroup_apply_control_enable(cgrp);
5311 if (ret)
5312 goto out_destroy;
Tejun Heo195e9b6c2016-03-03 09:57:58 -05005313
5314 /* let's create and online css's */
Tejun Heo2bd59d42014-02-11 11:52:49 -05005315 kernfs_activate(kn);
5316
Tejun Heoba0f4d72014-05-13 12:19:22 -04005317 ret = 0;
5318 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005319
Tejun Heoa5bca212016-03-03 09:57:58 -05005320out_destroy:
5321 cgroup_destroy_locked(cgrp);
Tejun Heoba0f4d72014-05-13 12:19:22 -04005322out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04005323 cgroup_kn_unlock(parent_kn);
Tejun Heoe1b2dc12014-03-20 11:10:15 -04005324 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005325}
5326
Tejun Heo223dbc32013-08-13 20:22:50 -04005327/*
5328 * This is called when the refcnt of a css is confirmed to be killed.
Tejun Heo249f3462014-05-14 09:15:01 -04005329 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5330 * initate destruction and put the css ref from kill_css().
Tejun Heo223dbc32013-08-13 20:22:50 -04005331 */
5332static void css_killed_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07005333{
Tejun Heo223dbc32013-08-13 20:22:50 -04005334 struct cgroup_subsys_state *css =
5335 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heod3daf282013-06-13 19:39:16 -07005336
Tejun Heof20104d2013-08-13 20:22:50 -04005337 mutex_lock(&cgroup_mutex);
Tejun Heo09a503ea2013-08-13 20:22:50 -04005338
Tejun Heoaa226ff2016-01-21 15:31:11 -05005339 do {
5340 offline_css(css);
5341 css_put(css);
5342 /* @css can't go away while we're holding cgroup_mutex */
5343 css = css->parent;
5344 } while (css && atomic_dec_and_test(&css->online_cnt));
5345
5346 mutex_unlock(&cgroup_mutex);
Tejun Heod3daf282013-06-13 19:39:16 -07005347}
5348
Tejun Heo223dbc32013-08-13 20:22:50 -04005349/* css kill confirmation processing requires process context, bounce */
5350static void css_killed_ref_fn(struct percpu_ref *ref)
Tejun Heod3daf282013-06-13 19:39:16 -07005351{
5352 struct cgroup_subsys_state *css =
5353 container_of(ref, struct cgroup_subsys_state, refcnt);
5354
Tejun Heoaa226ff2016-01-21 15:31:11 -05005355 if (atomic_dec_and_test(&css->online_cnt)) {
5356 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5357 queue_work(cgroup_destroy_wq, &css->destroy_work);
5358 }
Tejun Heod3daf282013-06-13 19:39:16 -07005359}
5360
Tejun Heof392e512014-04-23 11:13:14 -04005361/**
5362 * kill_css - destroy a css
5363 * @css: css to destroy
5364 *
5365 * This function initiates destruction of @css by removing cgroup interface
5366 * files and putting its base reference. ->css_offline() will be invoked
Tejun Heoec903c02014-05-13 12:11:01 -04005367 * asynchronously once css_tryget_online() is guaranteed to fail and when
5368 * the reference count reaches zero, @css will be released.
Tejun Heof392e512014-04-23 11:13:14 -04005369 */
5370static void kill_css(struct cgroup_subsys_state *css)
Tejun Heoedae0c32013-08-13 20:22:51 -04005371{
Tejun Heo01f64742014-05-13 12:19:23 -04005372 lockdep_assert_held(&cgroup_mutex);
Tejun Heo94419622014-03-19 10:23:54 -04005373
Tejun Heo2bd59d42014-02-11 11:52:49 -05005374 /*
5375 * This must happen before css is disassociated with its cgroup.
5376 * See seq_css() for details.
5377 */
Tejun Heo334c3672016-03-03 09:58:01 -05005378 css_clear_dir(css);
Tejun Heo3c14f8b2013-08-13 20:22:51 -04005379
Tejun Heoedae0c32013-08-13 20:22:51 -04005380 /*
5381 * Killing would put the base ref, but we need to keep it alive
5382 * until after ->css_offline().
5383 */
5384 css_get(css);
5385
5386 /*
5387 * cgroup core guarantees that, by the time ->css_offline() is
5388 * invoked, no new css reference will be given out via
Tejun Heoec903c02014-05-13 12:11:01 -04005389 * css_tryget_online(). We can't simply call percpu_ref_kill() and
Tejun Heoedae0c32013-08-13 20:22:51 -04005390 * proceed to offlining css's because percpu_ref_kill() doesn't
5391 * guarantee that the ref is seen as killed on all CPUs on return.
5392 *
5393 * Use percpu_ref_kill_and_confirm() to get notifications as each
5394 * css is confirmed to be seen as killed on all CPUs.
5395 */
5396 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07005397}
5398
5399/**
5400 * cgroup_destroy_locked - the first stage of cgroup destruction
5401 * @cgrp: cgroup to be destroyed
5402 *
5403 * css's make use of percpu refcnts whose killing latency shouldn't be
5404 * exposed to userland and are RCU protected. Also, cgroup core needs to
Tejun Heoec903c02014-05-13 12:11:01 -04005405 * guarantee that css_tryget_online() won't succeed by the time
5406 * ->css_offline() is invoked. To satisfy all the requirements,
5407 * destruction is implemented in the following two steps.
Tejun Heod3daf282013-06-13 19:39:16 -07005408 *
5409 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5410 * userland visible parts and start killing the percpu refcnts of
5411 * css's. Set up so that the next stage will be kicked off once all
5412 * the percpu refcnts are confirmed to be killed.
5413 *
5414 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5415 * rest of destruction. Once all cgroup references are gone, the
5416 * cgroup is RCU-freed.
5417 *
5418 * This function implements s1. After this step, @cgrp is gone as far as
5419 * the userland is concerned and a new cgroup with the same name may be
5420 * created. As cgroup doesn't care about the names internally, this
5421 * doesn't cause any problem.
5422 */
Tejun Heo42809dd2012-11-19 08:13:37 -08005423static int cgroup_destroy_locked(struct cgroup *cgrp)
5424 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005425{
Tejun Heo2bd59d42014-02-11 11:52:49 -05005426 struct cgroup_subsys_state *css;
Tejun Heo2b021cb2016-03-15 20:43:04 -04005427 struct cgrp_cset_link *link;
Tejun Heo1c6727a2013-12-06 15:11:56 -05005428 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005429
Tejun Heo42809dd2012-11-19 08:13:37 -08005430 lockdep_assert_held(&cgroup_mutex);
5431
Tejun Heo91486f62015-10-15 16:41:51 -04005432 /*
5433 * Only migration can raise populated from zero and we're already
5434 * holding cgroup_mutex.
5435 */
5436 if (cgroup_is_populated(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07005437 return -EBUSY;
Tejun Heoed9577932012-11-05 09:16:58 -08005438
Tejun Heo1a90dd52012-11-05 09:16:59 -08005439 /*
Tejun Heod5c419b2014-05-16 13:22:48 -04005440 * Make sure there's no live children. We can't test emptiness of
5441 * ->self.children as dead children linger on it while being
5442 * drained; otherwise, "rmdir parent/child parent" may fail.
Hugh Dickinsbb78a922013-08-28 16:31:23 -07005443 */
Tejun Heof3d46502014-05-16 13:22:52 -04005444 if (css_has_online_children(&cgrp->self))
Hugh Dickinsbb78a922013-08-28 16:31:23 -07005445 return -EBUSY;
5446
5447 /*
Tejun Heo2b021cb2016-03-15 20:43:04 -04005448 * Mark @cgrp and the associated csets dead. The former prevents
5449 * further task migration and child creation by disabling
5450 * cgroup_lock_live_group(). The latter makes the csets ignored by
5451 * the migration path.
Tejun Heo455050d2013-06-13 19:27:41 -07005452 */
Tejun Heo184faf32014-05-16 13:22:51 -04005453 cgrp->self.flags &= ~CSS_ONLINE;
Tejun Heo1a90dd52012-11-05 09:16:59 -08005454
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005455 spin_lock_irq(&css_set_lock);
Tejun Heo2b021cb2016-03-15 20:43:04 -04005456 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5457 link->cset->dead = true;
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005458 spin_unlock_irq(&css_set_lock);
Tejun Heo2b021cb2016-03-15 20:43:04 -04005459
Tejun Heo249f3462014-05-14 09:15:01 -04005460 /* initiate massacre of all css's */
Tejun Heo1a90dd52012-11-05 09:16:59 -08005461 for_each_css(css, ssid, cgrp)
Tejun Heo455050d2013-06-13 19:27:41 -07005462 kill_css(css);
5463
Tejun Heo455050d2013-06-13 19:27:41 -07005464 /*
Tejun Heo01f64742014-05-13 12:19:23 -04005465 * Remove @cgrp directory along with the base files. @cgrp has an
5466 * extra ref on its kn.
Tejun Heo455050d2013-06-13 19:27:41 -07005467 */
Tejun Heo01f64742014-05-13 12:19:23 -04005468 kernfs_remove(cgrp->kn);
Tejun Heof20104d2013-08-13 20:22:50 -04005469
Tejun Heod51f39b2014-05-16 13:22:48 -04005470 check_for_release(cgroup_parent(cgrp));
Tejun Heo2bd59d42014-02-11 11:52:49 -05005471
Tejun Heo249f3462014-05-14 09:15:01 -04005472 /* put the base reference */
Tejun Heo9d755d32014-05-14 09:15:02 -04005473 percpu_ref_kill(&cgrp->self.refcnt);
Tejun Heo455050d2013-06-13 19:27:41 -07005474
Tejun Heoea15f8c2013-06-13 19:27:42 -07005475 return 0;
5476};
5477
Tejun Heo2bd59d42014-02-11 11:52:49 -05005478static int cgroup_rmdir(struct kernfs_node *kn)
Tejun Heo42809dd2012-11-19 08:13:37 -08005479{
Tejun Heoa9746d82014-05-13 12:19:22 -04005480 struct cgroup *cgrp;
Tejun Heo2bd59d42014-02-11 11:52:49 -05005481 int ret = 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08005482
Tejun Heo945ba192016-03-03 09:58:00 -05005483 cgrp = cgroup_kn_lock_live(kn, false);
Tejun Heoa9746d82014-05-13 12:19:22 -04005484 if (!cgrp)
5485 return 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08005486
Tejun Heoa9746d82014-05-13 12:19:22 -04005487 ret = cgroup_destroy_locked(cgrp);
Tejun Heo42809dd2012-11-19 08:13:37 -08005488
Tejun Heoa9746d82014-05-13 12:19:22 -04005489 cgroup_kn_unlock(kn);
Tejun Heo42809dd2012-11-19 08:13:37 -08005490 return ret;
5491}
5492
Tejun Heo2bd59d42014-02-11 11:52:49 -05005493static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5494 .remount_fs = cgroup_remount,
5495 .show_options = cgroup_show_options,
5496 .mkdir = cgroup_mkdir,
5497 .rmdir = cgroup_rmdir,
5498 .rename = cgroup_rename,
Serge E. Hallyn4f41fc52016-05-09 09:59:55 -05005499 .show_path = cgroup_show_path,
Tejun Heo2bd59d42014-02-11 11:52:49 -05005500};
Tejun Heo8e3f6542012-04-01 12:09:55 -07005501
Tejun Heo15a4c832014-05-04 15:09:14 -04005502static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
Paul Menageddbcc7e2007-10-18 23:39:30 -07005503{
Paul Menageddbcc7e2007-10-18 23:39:30 -07005504 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08005505
Tejun Heoa5ae9892015-12-29 14:53:56 -05005506 pr_debug("Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005507
Tejun Heo648bb562012-11-19 08:13:36 -08005508 mutex_lock(&cgroup_mutex);
5509
Tejun Heo15a4c832014-05-04 15:09:14 -04005510 idr_init(&ss->css_idr);
Tejun Heo0adb0702014-02-12 09:29:48 -05005511 INIT_LIST_HEAD(&ss->cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -07005512
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005513 /* Create the root cgroup state for this subsystem */
5514 ss->root = &cgrp_dfl_root;
5515 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005516 /* We don't handle early failures gracefully */
5517 BUG_ON(IS_ERR(css));
Tejun Heoddfcada2014-05-04 15:09:14 -04005518 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
Tejun Heo3b514d22014-05-16 13:22:47 -04005519
5520 /*
5521 * Root csses are never destroyed and we can't initialize
5522 * percpu_ref during early init. Disable refcnting.
5523 */
5524 css->flags |= CSS_NO_REF;
5525
Tejun Heo15a4c832014-05-04 15:09:14 -04005526 if (early) {
Tejun Heo9395a452014-05-14 09:15:02 -04005527 /* allocation can't be done safely during early init */
Tejun Heo15a4c832014-05-04 15:09:14 -04005528 css->id = 1;
5529 } else {
5530 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5531 BUG_ON(css->id < 0);
5532 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07005533
Li Zefane8d55fd2008-04-29 01:00:13 -07005534 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07005535 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07005536 * newly registered, all tasks and hence the
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005537 * init_css_set is in the subsystem's root cgroup. */
Tejun Heoaec25022014-02-08 10:36:58 -05005538 init_css_set.subsys[ss->id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005539
Aleksa Saraicb4a3162015-06-06 10:02:14 +10005540 have_fork_callback |= (bool)ss->fork << ss->id;
5541 have_exit_callback |= (bool)ss->exit << ss->id;
Tejun Heoafcf6c82015-10-15 16:41:53 -04005542 have_free_callback |= (bool)ss->free << ss->id;
Aleksa Sarai7e476822015-06-09 21:32:09 +10005543 have_canfork_callback |= (bool)ss->can_fork << ss->id;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005544
Li Zefane8d55fd2008-04-29 01:00:13 -07005545 /* At system boot, before all subsystems have been
5546 * registered, no tasks have been forked, so we don't
5547 * need to invoke fork callbacks here. */
5548 BUG_ON(!list_empty(&init_task.tasks));
5549
Tejun Heoae7f1642013-08-13 20:22:50 -04005550 BUG_ON(online_css(css));
Tejun Heoa8638032012-11-09 09:12:29 -08005551
Tejun Heo648bb562012-11-19 08:13:36 -08005552 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005553}
5554
5555/**
Li Zefana043e3b2008-02-23 15:24:09 -08005556 * cgroup_init_early - cgroup initialization at system boot
5557 *
5558 * Initialize cgroups at system boot, and initialize any
5559 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07005560 */
5561int __init cgroup_init_early(void)
5562{
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04005563 static struct cgroup_sb_opts __initdata opts;
Tejun Heo30159ec2013-06-25 11:53:37 -07005564 struct cgroup_subsys *ss;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005565 int i;
Tejun Heo30159ec2013-06-25 11:53:37 -07005566
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005567 init_cgroup_root(&cgrp_dfl_root, &opts);
Tejun Heo3b514d22014-05-16 13:22:47 -04005568 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5569
Tejun Heoa4ea1cc2013-06-21 15:52:33 -07005570 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07005571
Tejun Heo3ed80a62014-02-08 10:36:58 -05005572 for_each_subsys(ss, i) {
Tejun Heoaec25022014-02-08 10:36:58 -05005573 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
Xiubo Li63253ad2016-02-26 13:07:38 +08005574 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
Tejun Heo073219e2014-02-08 10:36:58 -05005575 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
Tejun Heoaec25022014-02-08 10:36:58 -05005576 ss->id, ss->name);
Tejun Heo073219e2014-02-08 10:36:58 -05005577 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5578 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005579
Tejun Heoaec25022014-02-08 10:36:58 -05005580 ss->id = i;
Tejun Heo073219e2014-02-08 10:36:58 -05005581 ss->name = cgroup_subsys_name[i];
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005582 if (!ss->legacy_name)
5583 ss->legacy_name = cgroup_subsys_name[i];
Paul Menageddbcc7e2007-10-18 23:39:30 -07005584
5585 if (ss->early_init)
Tejun Heo15a4c832014-05-04 15:09:14 -04005586 cgroup_init_subsys(ss, true);
Paul Menageddbcc7e2007-10-18 23:39:30 -07005587 }
5588 return 0;
5589}
5590
Tejun Heo6e5c8302016-02-22 22:25:47 -05005591static u16 cgroup_disable_mask __initdata;
Tejun Heoa3e72732015-09-25 16:24:27 -04005592
Paul Menageddbcc7e2007-10-18 23:39:30 -07005593/**
Li Zefana043e3b2008-02-23 15:24:09 -08005594 * cgroup_init - cgroup initialization
5595 *
5596 * Register cgroup filesystem and /proc file, and initialize
5597 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07005598 */
5599int __init cgroup_init(void)
5600{
Tejun Heo30159ec2013-06-25 11:53:37 -07005601 struct cgroup_subsys *ss;
Tejun Heo035f4f52015-10-15 17:00:43 -04005602 int ssid;
Paul Menagea4243162007-10-18 23:39:35 -07005603
Tejun Heo6e5c8302016-02-22 22:25:47 -05005604 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
Tejun Heo1ed13282015-09-16 12:53:17 -04005605 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
Tejun Heoa14c6872014-07-15 11:05:09 -04005606 BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
5607 BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
Paul Menageddbcc7e2007-10-18 23:39:30 -07005608
Aditya Kalia79a9082016-01-29 02:54:06 -06005609 get_user_ns(init_cgroup_ns.user_ns);
5610
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005611 mutex_lock(&cgroup_mutex);
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005612
Tejun Heo2378d8b2016-03-03 09:57:57 -05005613 /*
5614 * Add init_css_set to the hash table so that dfl_root can link to
5615 * it during init.
5616 */
5617 hash_add(css_set_table, &init_css_set.hlist,
5618 css_set_hash(init_css_set.subsys));
Tejun Heo82fe9b02013-06-25 11:53:37 -07005619
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005620 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
Greg KH676db4a2010-08-05 13:53:35 -07005621
Tejun Heo54e7b4e2013-04-14 11:36:57 -07005622 mutex_unlock(&cgroup_mutex);
5623
Tejun Heo172a2c062014-03-19 10:23:53 -04005624 for_each_subsys(ss, ssid) {
Tejun Heo15a4c832014-05-04 15:09:14 -04005625 if (ss->early_init) {
5626 struct cgroup_subsys_state *css =
5627 init_css_set.subsys[ss->id];
5628
5629 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5630 GFP_KERNEL);
5631 BUG_ON(css->id < 0);
5632 } else {
5633 cgroup_init_subsys(ss, false);
5634 }
Tejun Heo172a2c062014-03-19 10:23:53 -04005635
Tejun Heo2d8f2432014-04-23 11:13:15 -04005636 list_add_tail(&init_css_set.e_cset_node[ssid],
5637 &cgrp_dfl_root.cgrp.e_csets[ssid]);
Tejun Heo172a2c062014-03-19 10:23:53 -04005638
5639 /*
Li Zefanc731ae12014-06-05 17:16:30 +08005640 * Setting dfl_root subsys_mask needs to consider the
5641 * disabled flag and cftype registration needs kmalloc,
5642 * both of which aren't available during early_init.
Tejun Heo172a2c062014-03-19 10:23:53 -04005643 */
Tejun Heoa3e72732015-09-25 16:24:27 -04005644 if (cgroup_disable_mask & (1 << ssid)) {
5645 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5646 printk(KERN_INFO "Disabling %s control group subsystem\n",
5647 ss->name);
Tejun Heoa8ddc822014-07-15 11:05:10 -04005648 continue;
Tejun Heoa3e72732015-09-25 16:24:27 -04005649 }
Tejun Heoa8ddc822014-07-15 11:05:10 -04005650
Johannes Weiner223ffb22016-02-11 13:34:49 -05005651 if (cgroup_ssid_no_v1(ssid))
5652 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5653 ss->name);
5654
Tejun Heoa8ddc822014-07-15 11:05:10 -04005655 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5656
Tejun Heof6d635ad2016-03-08 11:51:26 -05005657 if (ss->implicit_on_dfl)
5658 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5659 else if (!ss->dfl_cftypes)
Tejun Heoa7165262016-02-23 10:00:50 -05005660 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
Tejun Heo5de4fa12014-07-15 11:05:10 -04005661
Tejun Heoa8ddc822014-07-15 11:05:10 -04005662 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5663 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5664 } else {
5665 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5666 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
Li Zefanc731ae12014-06-05 17:16:30 +08005667 }
Vladimir Davydov295458e2015-02-19 17:34:46 +03005668
5669 if (ss->bind)
5670 ss->bind(init_css_set.subsys[ssid]);
Tejun Heo172a2c062014-03-19 10:23:53 -04005671 }
Greg KH676db4a2010-08-05 13:53:35 -07005672
Tejun Heo2378d8b2016-03-03 09:57:57 -05005673 /* init_css_set.subsys[] has been updated, re-hash */
5674 hash_del(&init_css_set.hlist);
5675 hash_add(css_set_table, &init_css_set.hlist,
5676 css_set_hash(init_css_set.subsys));
5677
Tejun Heo035f4f52015-10-15 17:00:43 -04005678 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5679 WARN_ON(register_filesystem(&cgroup_fs_type));
Tejun Heo67e9c742015-11-16 11:13:34 -05005680 WARN_ON(register_filesystem(&cgroup2_fs_type));
Tejun Heo035f4f52015-10-15 17:00:43 -04005681 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
Paul Menagea4243162007-10-18 23:39:35 -07005682
Tejun Heo2bd59d42014-02-11 11:52:49 -05005683 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07005684}
Paul Menageb4f48b62007-10-18 23:39:33 -07005685
Tejun Heoe5fca242013-11-22 17:14:39 -05005686static int __init cgroup_wq_init(void)
5687{
5688 /*
5689 * There isn't much point in executing destruction path in
5690 * parallel. Good chunk is serialized with cgroup_mutex anyway.
Tejun Heo1a115332014-02-12 19:06:19 -05005691 * Use 1 for @max_active.
Tejun Heoe5fca242013-11-22 17:14:39 -05005692 *
5693 * We would prefer to do this in cgroup_init() above, but that
5694 * is called before init_workqueues(): so leave this until after.
5695 */
Tejun Heo1a115332014-02-12 19:06:19 -05005696 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
Tejun Heoe5fca242013-11-22 17:14:39 -05005697 BUG_ON(!cgroup_destroy_wq);
Tejun Heob1a21362013-11-29 10:42:58 -05005698
5699 /*
5700 * Used to destroy pidlists and separate to serve as flush domain.
5701 * Cap @max_active to 1 too.
5702 */
5703 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5704 0, 1);
5705 BUG_ON(!cgroup_pidlist_destroy_wq);
5706
Tejun Heoe5fca242013-11-22 17:14:39 -05005707 return 0;
5708}
5709core_initcall(cgroup_wq_init);
5710
Paul Menagea4243162007-10-18 23:39:35 -07005711/*
5712 * proc_cgroup_show()
5713 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5714 * - Used for /proc/<pid>/cgroup.
Paul Menagea4243162007-10-18 23:39:35 -07005715 */
Zefan Li006f4ac2014-09-18 16:03:15 +08005716int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5717 struct pid *pid, struct task_struct *tsk)
Paul Menagea4243162007-10-18 23:39:35 -07005718{
Tejun Heoe61734c2014-02-12 09:29:50 -05005719 char *buf, *path;
Paul Menagea4243162007-10-18 23:39:35 -07005720 int retval;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005721 struct cgroup_root *root;
Paul Menagea4243162007-10-18 23:39:35 -07005722
5723 retval = -ENOMEM;
Tejun Heoe61734c2014-02-12 09:29:50 -05005724 buf = kmalloc(PATH_MAX, GFP_KERNEL);
Paul Menagea4243162007-10-18 23:39:35 -07005725 if (!buf)
5726 goto out;
5727
Paul Menagea4243162007-10-18 23:39:35 -07005728 mutex_lock(&cgroup_mutex);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005729 spin_lock_irq(&css_set_lock);
Paul Menagea4243162007-10-18 23:39:35 -07005730
Tejun Heo985ed672014-03-19 10:23:53 -04005731 for_each_root(root) {
Paul Menagea4243162007-10-18 23:39:35 -07005732 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07005733 struct cgroup *cgrp;
Tejun Heob85d2042013-12-06 15:11:57 -05005734 int ssid, count = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005735
Tejun Heoa7165262016-02-23 10:00:50 -05005736 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
Tejun Heo985ed672014-03-19 10:23:53 -04005737 continue;
5738
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005739 seq_printf(m, "%d:", root->hierarchy_id);
Tejun Heod98817d2015-08-18 13:58:16 -07005740 if (root != &cgrp_dfl_root)
5741 for_each_subsys(ss, ssid)
5742 if (root->subsys_mask & (1 << ssid))
5743 seq_printf(m, "%s%s", count++ ? "," : "",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005744 ss->legacy_name);
Paul Menagec6d57f332009-09-23 15:56:19 -07005745 if (strlen(root->name))
5746 seq_printf(m, "%sname=%s", count ? "," : "",
5747 root->name);
Paul Menagea4243162007-10-18 23:39:35 -07005748 seq_putc(m, ':');
Tejun Heo2e91fa72015-10-15 16:41:53 -04005749
Paul Menage7717f7b2009-09-23 15:56:22 -07005750 cgrp = task_cgroup_from_root(tsk, root);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005751
5752 /*
5753 * On traditional hierarchies, all zombie tasks show up as
5754 * belonging to the root cgroup. On the default hierarchy,
5755 * while a zombie doesn't show up in "cgroup.procs" and
5756 * thus can't be migrated, its /proc/PID/cgroup keeps
5757 * reporting the cgroup it belonged to before exiting. If
5758 * the cgroup is removed before the zombie is reaped,
5759 * " (deleted)" is appended to the cgroup path.
5760 */
5761 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
Aditya Kalia79a9082016-01-29 02:54:06 -06005762 path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5763 current->nsproxy->cgroup_ns);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005764 if (!path) {
5765 retval = -ENAMETOOLONG;
5766 goto out_unlock;
5767 }
5768 } else {
5769 path = "/";
Tejun Heoe61734c2014-02-12 09:29:50 -05005770 }
Tejun Heo2e91fa72015-10-15 16:41:53 -04005771
Tejun Heoe61734c2014-02-12 09:29:50 -05005772 seq_puts(m, path);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005773
5774 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5775 seq_puts(m, " (deleted)\n");
5776 else
5777 seq_putc(m, '\n');
Paul Menagea4243162007-10-18 23:39:35 -07005778 }
5779
Zefan Li006f4ac2014-09-18 16:03:15 +08005780 retval = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005781out_unlock:
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005782 spin_unlock_irq(&css_set_lock);
Paul Menagea4243162007-10-18 23:39:35 -07005783 mutex_unlock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07005784 kfree(buf);
5785out:
5786 return retval;
5787}
5788
Paul Menagea4243162007-10-18 23:39:35 -07005789/* Display information about each subsystem and each hierarchy */
5790static int proc_cgroupstats_show(struct seq_file *m, void *v)
5791{
Tejun Heo30159ec2013-06-25 11:53:37 -07005792 struct cgroup_subsys *ss;
Paul Menagea4243162007-10-18 23:39:35 -07005793 int i;
Paul Menagea4243162007-10-18 23:39:35 -07005794
Paul Menage8bab8dd2008-04-04 14:29:57 -07005795 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Ben Blumaae8aab2010-03-10 15:22:07 -08005796 /*
5797 * ideally we don't want subsystems moving around while we do this.
5798 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5799 * subsys/hierarchy state.
5800 */
Paul Menagea4243162007-10-18 23:39:35 -07005801 mutex_lock(&cgroup_mutex);
Tejun Heo30159ec2013-06-25 11:53:37 -07005802
5803 for_each_subsys(ss, i)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005804 seq_printf(m, "%s\t%d\t%d\t%d\n",
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07005805 ss->legacy_name, ss->root->hierarchy_id,
Tejun Heofc5ed1e2015-09-18 11:56:28 -04005806 atomic_read(&ss->root->nr_cgrps),
5807 cgroup_ssid_enabled(i));
Tejun Heo30159ec2013-06-25 11:53:37 -07005808
Paul Menagea4243162007-10-18 23:39:35 -07005809 mutex_unlock(&cgroup_mutex);
5810 return 0;
5811}
5812
5813static int cgroupstats_open(struct inode *inode, struct file *file)
5814{
Al Viro9dce07f2008-03-29 03:07:28 +00005815 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07005816}
5817
Alexey Dobriyan828c0952009-10-01 15:43:56 -07005818static const struct file_operations proc_cgroupstats_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07005819 .open = cgroupstats_open,
5820 .read = seq_read,
5821 .llseek = seq_lseek,
5822 .release = single_release,
5823};
5824
Paul Menageb4f48b62007-10-18 23:39:33 -07005825/**
Tejun Heoeaf797a2014-02-25 10:04:03 -05005826 * cgroup_fork - initialize cgroup related fields during copy_process()
Li Zefana043e3b2008-02-23 15:24:09 -08005827 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07005828 *
Tejun Heoeaf797a2014-02-25 10:04:03 -05005829 * A task is associated with the init_css_set until cgroup_post_fork()
5830 * attaches it to the parent's css_set. Empty cg_list indicates that
5831 * @child isn't holding reference to its css_set.
Paul Menageb4f48b62007-10-18 23:39:33 -07005832 */
5833void cgroup_fork(struct task_struct *child)
5834{
Tejun Heoeaf797a2014-02-25 10:04:03 -05005835 RCU_INIT_POINTER(child->cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07005836 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07005837}
5838
5839/**
Aleksa Sarai7e476822015-06-09 21:32:09 +10005840 * cgroup_can_fork - called on a new task before the process is exposed
5841 * @child: the task in question.
5842 *
5843 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5844 * returns an error, the fork aborts with that error code. This allows for
5845 * a cgroup subsystem to conditionally allow or deny new forks.
5846 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005847int cgroup_can_fork(struct task_struct *child)
Aleksa Sarai7e476822015-06-09 21:32:09 +10005848{
5849 struct cgroup_subsys *ss;
5850 int i, j, ret;
5851
Tejun Heob4e0eea2016-02-22 22:25:46 -05005852 do_each_subsys_mask(ss, i, have_canfork_callback) {
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005853 ret = ss->can_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005854 if (ret)
5855 goto out_revert;
Tejun Heob4e0eea2016-02-22 22:25:46 -05005856 } while_each_subsys_mask();
Aleksa Sarai7e476822015-06-09 21:32:09 +10005857
5858 return 0;
5859
5860out_revert:
5861 for_each_subsys(ss, j) {
5862 if (j >= i)
5863 break;
5864 if (ss->cancel_fork)
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005865 ss->cancel_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005866 }
5867
5868 return ret;
5869}
5870
5871/**
5872 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5873 * @child: the task in question
5874 *
5875 * This calls the cancel_fork() callbacks if a fork failed *after*
5876 * cgroup_can_fork() succeded.
5877 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005878void cgroup_cancel_fork(struct task_struct *child)
Aleksa Sarai7e476822015-06-09 21:32:09 +10005879{
5880 struct cgroup_subsys *ss;
5881 int i;
5882
5883 for_each_subsys(ss, i)
5884 if (ss->cancel_fork)
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005885 ss->cancel_fork(child);
Aleksa Sarai7e476822015-06-09 21:32:09 +10005886}
5887
5888/**
Li Zefana043e3b2008-02-23 15:24:09 -08005889 * cgroup_post_fork - called on a new task after adding it to the task list
5890 * @child: the task in question
5891 *
Tejun Heo5edee612012-10-16 15:03:14 -07005892 * Adds the task to the list running through its css_set if necessary and
5893 * call the subsystem fork() callbacks. Has to be after the task is
5894 * visible on the task list in case we race with the first call to
Tejun Heo0942eee2013-08-08 20:11:26 -04005895 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
Tejun Heo5edee612012-10-16 15:03:14 -07005896 * list.
Li Zefana043e3b2008-02-23 15:24:09 -08005897 */
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005898void cgroup_post_fork(struct task_struct *child)
Paul Menage817929e2007-10-18 23:39:36 -07005899{
Tejun Heo30159ec2013-06-25 11:53:37 -07005900 struct cgroup_subsys *ss;
Tejun Heo5edee612012-10-16 15:03:14 -07005901 int i;
5902
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005903 /*
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005904 * This may race against cgroup_enable_task_cg_lists(). As that
Tejun Heoeaf797a2014-02-25 10:04:03 -05005905 * function sets use_task_css_set_links before grabbing
5906 * tasklist_lock and we just went through tasklist_lock to add
5907 * @child, it's guaranteed that either we see the set
5908 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5909 * @child during its iteration.
5910 *
5911 * If we won the race, @child is associated with %current's
Tejun Heof0d9a5f2015-10-15 16:41:53 -04005912 * css_set. Grabbing css_set_lock guarantees both that the
Tejun Heoeaf797a2014-02-25 10:04:03 -05005913 * association is stable, and, on completion of the parent's
5914 * migration, @child is visible in the source of migration or
5915 * already in the destination cgroup. This guarantee is necessary
5916 * when implementing operations which need to migrate all tasks of
5917 * a cgroup to another.
5918 *
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005919 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
Tejun Heoeaf797a2014-02-25 10:04:03 -05005920 * will remain in init_css_set. This is safe because all tasks are
5921 * in the init_css_set before cg_links is enabled and there's no
5922 * operation which transfers all tasks out of init_css_set.
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005923 */
Paul Menage817929e2007-10-18 23:39:36 -07005924 if (use_task_css_set_links) {
Tejun Heoeaf797a2014-02-25 10:04:03 -05005925 struct css_set *cset;
5926
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005927 spin_lock_irq(&css_set_lock);
Tejun Heo0e1d7682014-02-25 10:04:03 -05005928 cset = task_css_set(current);
Tejun Heoeaf797a2014-02-25 10:04:03 -05005929 if (list_empty(&child->cg_list)) {
Tejun Heoeaf797a2014-02-25 10:04:03 -05005930 get_css_set(cset);
Tejun Heof6d7d042015-10-15 16:41:52 -04005931 css_set_move_task(child, NULL, cset, false);
Tejun Heoeaf797a2014-02-25 10:04:03 -05005932 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005933 spin_unlock_irq(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -07005934 }
Tejun Heo5edee612012-10-16 15:03:14 -07005935
5936 /*
5937 * Call ss->fork(). This must happen after @child is linked on
5938 * css_set; otherwise, @child might change state between ->fork()
5939 * and addition to css_set.
5940 */
Tejun Heob4e0eea2016-02-22 22:25:46 -05005941 do_each_subsys_mask(ss, i, have_fork_callback) {
Oleg Nesterovb53202e2015-12-03 10:24:08 -05005942 ss->fork(child);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005943 } while_each_subsys_mask();
Paul Menage817929e2007-10-18 23:39:36 -07005944}
Tejun Heo5edee612012-10-16 15:03:14 -07005945
Paul Menage817929e2007-10-18 23:39:36 -07005946/**
Paul Menageb4f48b62007-10-18 23:39:33 -07005947 * cgroup_exit - detach cgroup from exiting task
5948 * @tsk: pointer to task_struct of exiting process
5949 *
5950 * Description: Detach cgroup from @tsk and release it.
5951 *
5952 * Note that cgroups marked notify_on_release force every task in
5953 * them to take the global cgroup_mutex mutex when exiting.
5954 * This could impact scaling on very large systems. Be reluctant to
5955 * use notify_on_release cgroups where very high task exit scaling
5956 * is required on large systems.
5957 *
Tejun Heo0e1d7682014-02-25 10:04:03 -05005958 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5959 * call cgroup_exit() while the task is still competent to handle
5960 * notify_on_release(), then leave the task attached to the root cgroup in
5961 * each hierarchy for the remainder of its exit. No need to bother with
5962 * init_css_set refcnting. init_css_set never goes away and we can't race
Li Zefane8604cb2014-03-28 15:18:27 +08005963 * with migration path - PF_EXITING is visible to migration path.
Paul Menageb4f48b62007-10-18 23:39:33 -07005964 */
Li Zefan1ec41832014-03-28 15:22:19 +08005965void cgroup_exit(struct task_struct *tsk)
Paul Menageb4f48b62007-10-18 23:39:33 -07005966{
Tejun Heo30159ec2013-06-25 11:53:37 -07005967 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -07005968 struct css_set *cset;
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005969 int i;
Paul Menage817929e2007-10-18 23:39:36 -07005970
5971 /*
Tejun Heo0e1d7682014-02-25 10:04:03 -05005972 * Unlink from @tsk from its css_set. As migration path can't race
Tejun Heo0de09422015-10-15 16:41:49 -04005973 * with us, we can check css_set and cg_list without synchronization.
Paul Menage817929e2007-10-18 23:39:36 -07005974 */
Tejun Heo0de09422015-10-15 16:41:49 -04005975 cset = task_css_set(tsk);
5976
Paul Menage817929e2007-10-18 23:39:36 -07005977 if (!list_empty(&tsk->cg_list)) {
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005978 spin_lock_irq(&css_set_lock);
Tejun Heof6d7d042015-10-15 16:41:52 -04005979 css_set_move_task(tsk, cset, NULL, false);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03005980 spin_unlock_irq(&css_set_lock);
Tejun Heo2e91fa72015-10-15 16:41:53 -04005981 } else {
5982 get_css_set(cset);
Paul Menage817929e2007-10-18 23:39:36 -07005983 }
5984
Aleksa Saraicb4a3162015-06-06 10:02:14 +10005985 /* see cgroup_post_fork() for details */
Tejun Heob4e0eea2016-02-22 22:25:46 -05005986 do_each_subsys_mask(ss, i, have_exit_callback) {
Tejun Heo2e91fa72015-10-15 16:41:53 -04005987 ss->exit(tsk);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005988 } while_each_subsys_mask();
Tejun Heo2e91fa72015-10-15 16:41:53 -04005989}
Tejun Heo30159ec2013-06-25 11:53:37 -07005990
Tejun Heo2e91fa72015-10-15 16:41:53 -04005991void cgroup_free(struct task_struct *task)
5992{
5993 struct css_set *cset = task_css_set(task);
Tejun Heoafcf6c82015-10-15 16:41:53 -04005994 struct cgroup_subsys *ss;
5995 int ssid;
5996
Tejun Heob4e0eea2016-02-22 22:25:46 -05005997 do_each_subsys_mask(ss, ssid, have_free_callback) {
Tejun Heoafcf6c82015-10-15 16:41:53 -04005998 ss->free(task);
Tejun Heob4e0eea2016-02-22 22:25:46 -05005999 } while_each_subsys_mask();
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01006000
Tejun Heo2e91fa72015-10-15 16:41:53 -04006001 put_css_set(cset);
Paul Menageb4f48b62007-10-18 23:39:33 -07006002}
Paul Menage697f4162007-10-18 23:39:34 -07006003
Paul Menagebd89aab2007-10-18 23:40:44 -07006004static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07006005{
Tejun Heo27bd4db2015-10-15 16:41:50 -04006006 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
Zefan Li971ff492014-09-18 16:06:19 +08006007 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
6008 schedule_work(&cgrp->release_agent_work);
Paul Menage81a6a5c2007-10-18 23:39:38 -07006009}
6010
Paul Menage81a6a5c2007-10-18 23:39:38 -07006011/*
6012 * Notify userspace when a cgroup is released, by running the
6013 * configured release agent with the name of the cgroup (path
6014 * relative to the root of cgroup file system) as the argument.
6015 *
6016 * Most likely, this user command will try to rmdir this cgroup.
6017 *
6018 * This races with the possibility that some other task will be
6019 * attached to this cgroup before it is removed, or that some other
6020 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
6021 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
6022 * unused, and this cgroup will be reprieved from its death sentence,
6023 * to continue to serve a useful existence. Next time it's released,
6024 * we will get notified again, if it still has 'notify_on_release' set.
6025 *
6026 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
6027 * means only wait until the task is successfully execve()'d. The
6028 * separate release agent task is forked by call_usermodehelper(),
6029 * then control in this thread returns here, without waiting for the
6030 * release agent task. We don't bother to wait because the caller of
6031 * this routine has no use for the exit status of the release agent
6032 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07006033 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07006034static void cgroup_release_agent(struct work_struct *work)
6035{
Zefan Li971ff492014-09-18 16:06:19 +08006036 struct cgroup *cgrp =
6037 container_of(work, struct cgroup, release_agent_work);
6038 char *pathbuf = NULL, *agentbuf = NULL, *path;
6039 char *argv[3], *envp[3];
6040
Paul Menage81a6a5c2007-10-18 23:39:38 -07006041 mutex_lock(&cgroup_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07006042
Zefan Li971ff492014-09-18 16:06:19 +08006043 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
6044 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
6045 if (!pathbuf || !agentbuf)
6046 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07006047
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006048 spin_lock_irq(&css_set_lock);
Aditya Kalia79a9082016-01-29 02:54:06 -06006049 path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006050 spin_unlock_irq(&css_set_lock);
Zefan Li971ff492014-09-18 16:06:19 +08006051 if (!path)
6052 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07006053
Zefan Li971ff492014-09-18 16:06:19 +08006054 argv[0] = agentbuf;
6055 argv[1] = path;
6056 argv[2] = NULL;
6057
6058 /* minimal command environment */
6059 envp[0] = "HOME=/";
6060 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
6061 envp[2] = NULL;
6062
Paul Menage81a6a5c2007-10-18 23:39:38 -07006063 mutex_unlock(&cgroup_mutex);
Zefan Li971ff492014-09-18 16:06:19 +08006064 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
Zefan Li3e2cd912014-09-20 14:35:43 +08006065 goto out_free;
Zefan Li971ff492014-09-18 16:06:19 +08006066out:
Zefan Li3e2cd912014-09-20 14:35:43 +08006067 mutex_unlock(&cgroup_mutex);
6068out_free:
Zefan Li971ff492014-09-18 16:06:19 +08006069 kfree(agentbuf);
6070 kfree(pathbuf);
Paul Menage81a6a5c2007-10-18 23:39:38 -07006071}
Paul Menage8bab8dd2008-04-04 14:29:57 -07006072
6073static int __init cgroup_disable(char *str)
6074{
Tejun Heo30159ec2013-06-25 11:53:37 -07006075 struct cgroup_subsys *ss;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006076 char *token;
Tejun Heo30159ec2013-06-25 11:53:37 -07006077 int i;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006078
6079 while ((token = strsep(&str, ",")) != NULL) {
6080 if (!*token)
6081 continue;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006082
Tejun Heo3ed80a62014-02-08 10:36:58 -05006083 for_each_subsys(ss, i) {
Tejun Heo3e1d2ee2015-08-18 13:58:16 -07006084 if (strcmp(token, ss->name) &&
6085 strcmp(token, ss->legacy_name))
6086 continue;
Tejun Heoa3e72732015-09-25 16:24:27 -04006087 cgroup_disable_mask |= 1 << i;
Paul Menage8bab8dd2008-04-04 14:29:57 -07006088 }
6089 }
6090 return 1;
6091}
6092__setup("cgroup_disable=", cgroup_disable);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07006093
Johannes Weiner223ffb22016-02-11 13:34:49 -05006094static int __init cgroup_no_v1(char *str)
6095{
6096 struct cgroup_subsys *ss;
6097 char *token;
6098 int i;
6099
6100 while ((token = strsep(&str, ",")) != NULL) {
6101 if (!*token)
6102 continue;
6103
6104 if (!strcmp(token, "all")) {
Tejun Heo6e5c8302016-02-22 22:25:47 -05006105 cgroup_no_v1_mask = U16_MAX;
Johannes Weiner223ffb22016-02-11 13:34:49 -05006106 break;
6107 }
6108
6109 for_each_subsys(ss, i) {
6110 if (strcmp(token, ss->name) &&
6111 strcmp(token, ss->legacy_name))
6112 continue;
6113
6114 cgroup_no_v1_mask |= 1 << i;
6115 }
6116 }
6117 return 1;
6118}
6119__setup("cgroup_no_v1=", cgroup_no_v1);
6120
Tejun Heob77d7b62013-08-13 11:01:54 -04006121/**
Tejun Heoec903c02014-05-13 12:11:01 -04006122 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
Tejun Heo35cf0832013-08-26 18:40:56 -04006123 * @dentry: directory dentry of interest
6124 * @ss: subsystem of interest
Tejun Heob77d7b62013-08-13 11:01:54 -04006125 *
Tejun Heo5a17f542014-02-11 11:52:47 -05006126 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6127 * to get the corresponding css and return it. If such css doesn't exist
6128 * or can't be pinned, an ERR_PTR value is returned.
Stephane Eraniane5d13672011-02-14 11:20:01 +02006129 */
Tejun Heoec903c02014-05-13 12:11:01 -04006130struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6131 struct cgroup_subsys *ss)
Stephane Eraniane5d13672011-02-14 11:20:01 +02006132{
Tejun Heo2bd59d42014-02-11 11:52:49 -05006133 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
Tejun Heof17fc252016-02-23 10:00:51 -05006134 struct file_system_type *s_type = dentry->d_sb->s_type;
Tejun Heo2bd59d42014-02-11 11:52:49 -05006135 struct cgroup_subsys_state *css = NULL;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006136 struct cgroup *cgrp;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006137
Tejun Heo35cf0832013-08-26 18:40:56 -04006138 /* is @dentry a cgroup dir? */
Tejun Heof17fc252016-02-23 10:00:51 -05006139 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6140 !kn || kernfs_type(kn) != KERNFS_DIR)
Stephane Eraniane5d13672011-02-14 11:20:01 +02006141 return ERR_PTR(-EBADF);
6142
Tejun Heo5a17f542014-02-11 11:52:47 -05006143 rcu_read_lock();
6144
Tejun Heo2bd59d42014-02-11 11:52:49 -05006145 /*
6146 * This path doesn't originate from kernfs and @kn could already
6147 * have been or be removed at any point. @kn->priv is RCU
Li Zefana4189482014-09-04 14:43:07 +08006148 * protected for this access. See css_release_work_fn() for details.
Tejun Heo2bd59d42014-02-11 11:52:49 -05006149 */
6150 cgrp = rcu_dereference(kn->priv);
6151 if (cgrp)
6152 css = cgroup_css(cgrp, ss);
Tejun Heo5a17f542014-02-11 11:52:47 -05006153
Tejun Heoec903c02014-05-13 12:11:01 -04006154 if (!css || !css_tryget_online(css))
Tejun Heo5a17f542014-02-11 11:52:47 -05006155 css = ERR_PTR(-ENOENT);
6156
6157 rcu_read_unlock();
6158 return css;
Stephane Eraniane5d13672011-02-14 11:20:01 +02006159}
Stephane Eraniane5d13672011-02-14 11:20:01 +02006160
Li Zefan1cb650b2013-08-19 10:05:24 +08006161/**
6162 * css_from_id - lookup css by id
6163 * @id: the cgroup id
6164 * @ss: cgroup subsys to be looked into
6165 *
6166 * Returns the css if there's valid one with @id, otherwise returns NULL.
6167 * Should be called under rcu_read_lock().
6168 */
6169struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6170{
Tejun Heo6fa49182014-05-04 15:09:13 -04006171 WARN_ON_ONCE(!rcu_read_lock_held());
Johannes Weinerd6ccc552016-06-17 12:24:27 -04006172 return idr_find(&ss->css_idr, id);
Stephane Eraniane5d13672011-02-14 11:20:01 +02006173}
6174
Tejun Heo16af4392015-11-20 15:55:52 -05006175/**
6176 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6177 * @path: path on the default hierarchy
6178 *
6179 * Find the cgroup at @path on the default hierarchy, increment its
6180 * reference count and return it. Returns pointer to the found cgroup on
6181 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6182 * if @path points to a non-directory.
6183 */
6184struct cgroup *cgroup_get_from_path(const char *path)
6185{
6186 struct kernfs_node *kn;
6187 struct cgroup *cgrp;
6188
6189 mutex_lock(&cgroup_mutex);
6190
6191 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6192 if (kn) {
6193 if (kernfs_type(kn) == KERNFS_DIR) {
6194 cgrp = kn->priv;
6195 cgroup_get(cgrp);
6196 } else {
6197 cgrp = ERR_PTR(-ENOTDIR);
6198 }
6199 kernfs_put(kn);
6200 } else {
6201 cgrp = ERR_PTR(-ENOENT);
6202 }
6203
6204 mutex_unlock(&cgroup_mutex);
6205 return cgrp;
6206}
6207EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6208
Martin KaFai Lau1f3fe7e2016-06-30 10:28:42 -07006209/**
6210 * cgroup_get_from_fd - get a cgroup pointer from a fd
6211 * @fd: fd obtained by open(cgroup2_dir)
6212 *
6213 * Find the cgroup from a fd which should be obtained
6214 * by opening a cgroup directory. Returns a pointer to the
6215 * cgroup on success. ERR_PTR is returned if the cgroup
6216 * cannot be found.
6217 */
6218struct cgroup *cgroup_get_from_fd(int fd)
6219{
6220 struct cgroup_subsys_state *css;
6221 struct cgroup *cgrp;
6222 struct file *f;
6223
6224 f = fget_raw(fd);
6225 if (!f)
6226 return ERR_PTR(-EBADF);
6227
6228 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6229 fput(f);
6230 if (IS_ERR(css))
6231 return ERR_CAST(css);
6232
6233 cgrp = css->cgroup;
6234 if (!cgroup_on_dfl(cgrp)) {
6235 cgroup_put(cgrp);
6236 return ERR_PTR(-EBADF);
6237 }
6238
6239 return cgrp;
6240}
6241EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6242
Tejun Heobd1060a2015-12-07 17:38:53 -05006243/*
6244 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6245 * definition in cgroup-defs.h.
6246 */
6247#ifdef CONFIG_SOCK_CGROUP_DATA
6248
6249#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6250
Tejun Heo3fa4cc92015-12-14 11:24:06 -05006251DEFINE_SPINLOCK(cgroup_sk_update_lock);
Tejun Heobd1060a2015-12-07 17:38:53 -05006252static bool cgroup_sk_alloc_disabled __read_mostly;
6253
6254void cgroup_sk_alloc_disable(void)
6255{
6256 if (cgroup_sk_alloc_disabled)
6257 return;
6258 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6259 cgroup_sk_alloc_disabled = true;
6260}
6261
6262#else
6263
6264#define cgroup_sk_alloc_disabled false
6265
6266#endif
6267
6268void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6269{
6270 if (cgroup_sk_alloc_disabled)
6271 return;
6272
6273 rcu_read_lock();
6274
6275 while (true) {
6276 struct css_set *cset;
6277
6278 cset = task_css_set(current);
6279 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6280 skcd->val = (unsigned long)cset->dfl_cgrp;
6281 break;
6282 }
6283 cpu_relax();
6284 }
6285
6286 rcu_read_unlock();
6287}
6288
6289void cgroup_sk_free(struct sock_cgroup_data *skcd)
6290{
6291 cgroup_put(sock_cgroup_ptr(skcd));
6292}
6293
6294#endif /* CONFIG_SOCK_CGROUP_DATA */
6295
Aditya Kalia79a9082016-01-29 02:54:06 -06006296/* cgroup namespaces */
6297
6298static struct cgroup_namespace *alloc_cgroup_ns(void)
6299{
6300 struct cgroup_namespace *new_ns;
6301 int ret;
6302
6303 new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL);
6304 if (!new_ns)
6305 return ERR_PTR(-ENOMEM);
6306 ret = ns_alloc_inum(&new_ns->ns);
6307 if (ret) {
6308 kfree(new_ns);
6309 return ERR_PTR(ret);
6310 }
6311 atomic_set(&new_ns->count, 1);
6312 new_ns->ns.ops = &cgroupns_operations;
6313 return new_ns;
6314}
6315
6316void free_cgroup_ns(struct cgroup_namespace *ns)
6317{
6318 put_css_set(ns->root_cset);
6319 put_user_ns(ns->user_ns);
6320 ns_free_inum(&ns->ns);
6321 kfree(ns);
6322}
6323EXPORT_SYMBOL(free_cgroup_ns);
6324
6325struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
6326 struct user_namespace *user_ns,
6327 struct cgroup_namespace *old_ns)
6328{
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006329 struct cgroup_namespace *new_ns;
6330 struct css_set *cset;
Aditya Kalia79a9082016-01-29 02:54:06 -06006331
6332 BUG_ON(!old_ns);
6333
6334 if (!(flags & CLONE_NEWCGROUP)) {
6335 get_cgroup_ns(old_ns);
6336 return old_ns;
6337 }
6338
6339 /* Allow only sysadmin to create cgroup namespace. */
Aditya Kalia79a9082016-01-29 02:54:06 -06006340 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006341 return ERR_PTR(-EPERM);
Aditya Kalia79a9082016-01-29 02:54:06 -06006342
Eric W. Biederman7bd88302016-07-15 06:35:24 -05006343 /* It is not safe to take cgroup_mutex here */
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006344 spin_lock_irq(&css_set_lock);
Aditya Kalia79a9082016-01-29 02:54:06 -06006345 cset = task_css_set(current);
6346 get_css_set(cset);
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006347 spin_unlock_irq(&css_set_lock);
Aditya Kalia79a9082016-01-29 02:54:06 -06006348
Aditya Kalia79a9082016-01-29 02:54:06 -06006349 new_ns = alloc_cgroup_ns();
Tejun Heod2202552016-02-18 11:44:24 -05006350 if (IS_ERR(new_ns)) {
Tejun Heofa5ff8a2016-02-28 08:59:33 -05006351 put_css_set(cset);
6352 return new_ns;
Tejun Heod2202552016-02-18 11:44:24 -05006353 }
Aditya Kalia79a9082016-01-29 02:54:06 -06006354
6355 new_ns->user_ns = get_user_ns(user_ns);
6356 new_ns->root_cset = cset;
6357
6358 return new_ns;
Aditya Kalia79a9082016-01-29 02:54:06 -06006359}
6360
6361static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns)
6362{
6363 return container_of(ns, struct cgroup_namespace, ns);
6364}
6365
Aditya Kalia0530e02016-01-29 02:54:07 -06006366static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Aditya Kalia79a9082016-01-29 02:54:06 -06006367{
Aditya Kalia0530e02016-01-29 02:54:07 -06006368 struct cgroup_namespace *cgroup_ns = to_cg_ns(ns);
6369
6370 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) ||
6371 !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN))
6372 return -EPERM;
6373
6374 /* Don't need to do anything if we are attaching to our own cgroupns. */
6375 if (cgroup_ns == nsproxy->cgroup_ns)
6376 return 0;
6377
6378 get_cgroup_ns(cgroup_ns);
6379 put_cgroup_ns(nsproxy->cgroup_ns);
6380 nsproxy->cgroup_ns = cgroup_ns;
6381
6382 return 0;
Aditya Kalia79a9082016-01-29 02:54:06 -06006383}
6384
6385static struct ns_common *cgroupns_get(struct task_struct *task)
6386{
6387 struct cgroup_namespace *ns = NULL;
6388 struct nsproxy *nsproxy;
6389
6390 task_lock(task);
6391 nsproxy = task->nsproxy;
6392 if (nsproxy) {
6393 ns = nsproxy->cgroup_ns;
6394 get_cgroup_ns(ns);
6395 }
6396 task_unlock(task);
6397
6398 return ns ? &ns->ns : NULL;
6399}
6400
6401static void cgroupns_put(struct ns_common *ns)
6402{
6403 put_cgroup_ns(to_cg_ns(ns));
6404}
6405
6406const struct proc_ns_operations cgroupns_operations = {
6407 .name = "cgroup",
6408 .type = CLONE_NEWCGROUP,
6409 .get = cgroupns_get,
6410 .put = cgroupns_put,
6411 .install = cgroupns_install,
6412};
6413
6414static __init int cgroup_namespaces_init(void)
6415{
6416 return 0;
6417}
6418subsys_initcall(cgroup_namespaces_init);
6419
Paul Menagefe693432009-09-23 15:56:20 -07006420#ifdef CONFIG_CGROUP_DEBUG
Tejun Heoeb954192013-08-08 20:11:23 -04006421static struct cgroup_subsys_state *
6422debug_css_alloc(struct cgroup_subsys_state *parent_css)
Paul Menagefe693432009-09-23 15:56:20 -07006423{
6424 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
6425
6426 if (!css)
6427 return ERR_PTR(-ENOMEM);
6428
6429 return css;
6430}
6431
Tejun Heoeb954192013-08-08 20:11:23 -04006432static void debug_css_free(struct cgroup_subsys_state *css)
Paul Menagefe693432009-09-23 15:56:20 -07006433{
Tejun Heoeb954192013-08-08 20:11:23 -04006434 kfree(css);
Paul Menagefe693432009-09-23 15:56:20 -07006435}
6436
Tejun Heo182446d2013-08-08 20:11:24 -04006437static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
6438 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006439{
Tejun Heo182446d2013-08-08 20:11:24 -04006440 return cgroup_task_count(css->cgroup);
Paul Menagefe693432009-09-23 15:56:20 -07006441}
6442
Tejun Heo182446d2013-08-08 20:11:24 -04006443static u64 current_css_set_read(struct cgroup_subsys_state *css,
6444 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006445{
6446 return (u64)(unsigned long)current->cgroups;
6447}
6448
Tejun Heo182446d2013-08-08 20:11:24 -04006449static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
Li Zefan03c78cb2013-06-14 11:17:19 +08006450 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006451{
6452 u64 count;
6453
6454 rcu_read_lock();
Tejun Heoa8ad8052013-06-21 15:52:04 -07006455 count = atomic_read(&task_css_set(current)->refcount);
Paul Menagefe693432009-09-23 15:56:20 -07006456 rcu_read_unlock();
6457 return count;
6458}
6459
Tejun Heo2da8ca82013-12-05 12:28:04 -05006460static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07006461{
Tejun Heo69d02062013-06-12 21:04:50 -07006462 struct cgrp_cset_link *link;
Tejun Heo5abb8852013-06-12 21:04:49 -07006463 struct css_set *cset;
Tejun Heoe61734c2014-02-12 09:29:50 -05006464 char *name_buf;
Paul Menage7717f7b2009-09-23 15:56:22 -07006465
Tejun Heoe61734c2014-02-12 09:29:50 -05006466 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
6467 if (!name_buf)
6468 return -ENOMEM;
Paul Menage7717f7b2009-09-23 15:56:22 -07006469
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006470 spin_lock_irq(&css_set_lock);
Paul Menage7717f7b2009-09-23 15:56:22 -07006471 rcu_read_lock();
Tejun Heo5abb8852013-06-12 21:04:49 -07006472 cset = rcu_dereference(current->cgroups);
Tejun Heo69d02062013-06-12 21:04:50 -07006473 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07006474 struct cgroup *c = link->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -07006475
Tejun Heoa2dd4242014-03-19 10:23:55 -04006476 cgroup_name(c, name_buf, NAME_MAX + 1);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07006477 seq_printf(seq, "Root %d group %s\n",
Tejun Heoa2dd4242014-03-19 10:23:55 -04006478 c->root->hierarchy_id, name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07006479 }
6480 rcu_read_unlock();
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006481 spin_unlock_irq(&css_set_lock);
Tejun Heoe61734c2014-02-12 09:29:50 -05006482 kfree(name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07006483 return 0;
6484}
6485
6486#define MAX_TASKS_SHOWN_PER_CSS 25
Tejun Heo2da8ca82013-12-05 12:28:04 -05006487static int cgroup_css_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07006488{
Tejun Heo2da8ca82013-12-05 12:28:04 -05006489 struct cgroup_subsys_state *css = seq_css(seq);
Tejun Heo69d02062013-06-12 21:04:50 -07006490 struct cgrp_cset_link *link;
Paul Menage7717f7b2009-09-23 15:56:22 -07006491
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006492 spin_lock_irq(&css_set_lock);
Tejun Heo182446d2013-08-08 20:11:24 -04006493 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
Tejun Heo69d02062013-06-12 21:04:50 -07006494 struct css_set *cset = link->cset;
Paul Menage7717f7b2009-09-23 15:56:22 -07006495 struct task_struct *task;
6496 int count = 0;
Tejun Heoc7561122014-02-25 10:04:01 -05006497
Tejun Heo5abb8852013-06-12 21:04:49 -07006498 seq_printf(seq, "css_set %p\n", cset);
Tejun Heoc7561122014-02-25 10:04:01 -05006499
Tejun Heo5abb8852013-06-12 21:04:49 -07006500 list_for_each_entry(task, &cset->tasks, cg_list) {
Tejun Heoc7561122014-02-25 10:04:01 -05006501 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6502 goto overflow;
6503 seq_printf(seq, " task %d\n", task_pid_vnr(task));
Paul Menage7717f7b2009-09-23 15:56:22 -07006504 }
Tejun Heoc7561122014-02-25 10:04:01 -05006505
6506 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
6507 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6508 goto overflow;
6509 seq_printf(seq, " task %d\n", task_pid_vnr(task));
6510 }
6511 continue;
6512 overflow:
6513 seq_puts(seq, " ...\n");
Paul Menage7717f7b2009-09-23 15:56:22 -07006514 }
Daniel Bristot de Oliveira82d64892016-06-22 17:28:41 -03006515 spin_unlock_irq(&css_set_lock);
Paul Menage7717f7b2009-09-23 15:56:22 -07006516 return 0;
6517}
6518
Tejun Heo182446d2013-08-08 20:11:24 -04006519static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07006520{
Tejun Heo27bd4db2015-10-15 16:41:50 -04006521 return (!cgroup_is_populated(css->cgroup) &&
Zefan Lia25eb522014-09-19 16:51:00 +08006522 !css_has_online_children(&css->cgroup->self));
Paul Menagefe693432009-09-23 15:56:20 -07006523}
6524
6525static struct cftype debug_files[] = {
6526 {
Paul Menagefe693432009-09-23 15:56:20 -07006527 .name = "taskcount",
6528 .read_u64 = debug_taskcount_read,
6529 },
6530
6531 {
6532 .name = "current_css_set",
6533 .read_u64 = current_css_set_read,
6534 },
6535
6536 {
6537 .name = "current_css_set_refcount",
6538 .read_u64 = current_css_set_refcount_read,
6539 },
6540
6541 {
Paul Menage7717f7b2009-09-23 15:56:22 -07006542 .name = "current_css_set_cg_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006543 .seq_show = current_css_set_cg_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07006544 },
6545
6546 {
6547 .name = "cgroup_css_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006548 .seq_show = cgroup_css_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07006549 },
6550
6551 {
Paul Menagefe693432009-09-23 15:56:20 -07006552 .name = "releasable",
6553 .read_u64 = releasable_read,
6554 },
Paul Menagefe693432009-09-23 15:56:20 -07006555
Tejun Heo4baf6e32012-04-01 12:09:55 -07006556 { } /* terminate */
6557};
Paul Menagefe693432009-09-23 15:56:20 -07006558
Tejun Heo073219e2014-02-08 10:36:58 -05006559struct cgroup_subsys debug_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006560 .css_alloc = debug_css_alloc,
6561 .css_free = debug_css_free,
Tejun Heo55779642014-07-15 11:05:09 -04006562 .legacy_cftypes = debug_files,
Paul Menagefe693432009-09-23 15:56:20 -07006563};
6564#endif /* CONFIG_CGROUP_DEBUG */