2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
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 * ---------------------------------------------------
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.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
57 #include <linux/fs_parser.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/psi.h>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/cgroup.h>
65 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
67 /* let's not notify more than 100 times per second */
68 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * css_set_lock protects task->cgroups pointer, the list of css_set
75 * objects, and the chain of tasks off each css_set.
77 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
78 * cgroup.h can use them for lockdep annotations.
80 DEFINE_MUTEX(cgroup_mutex
);
81 DEFINE_SPINLOCK(css_set_lock
);
83 #ifdef CONFIG_PROVE_RCU
84 EXPORT_SYMBOL_GPL(cgroup_mutex
);
85 EXPORT_SYMBOL_GPL(css_set_lock
);
88 DEFINE_SPINLOCK(trace_cgroup_path_lock
);
89 char trace_cgroup_path
[TRACE_CGROUP_PATH_LEN
];
90 bool cgroup_debug __read_mostly
;
93 * Protects cgroup_idr and css_idr so that IDs can be released without
94 * grabbing cgroup_mutex.
96 static DEFINE_SPINLOCK(cgroup_idr_lock
);
99 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
100 * against file removal/re-creation across css hiding.
102 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
104 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem
);
106 #define cgroup_assert_mutex_or_rcu_locked() \
107 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
108 !lockdep_is_held(&cgroup_mutex), \
109 "cgroup_mutex or RCU read lock required");
112 * cgroup destruction makes heavy use of work items and there can be a lot
113 * of concurrent destructions. Use a separate workqueue so that cgroup
114 * destruction work items don't end up filling up max_active of system_wq
115 * which may lead to deadlock.
117 static struct workqueue_struct
*cgroup_destroy_wq
;
119 /* generate an array of cgroup subsystem pointers */
120 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
121 struct cgroup_subsys
*cgroup_subsys
[] = {
122 #include <linux/cgroup_subsys.h>
126 /* array of cgroup subsystem names */
127 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
128 static const char *cgroup_subsys_name
[] = {
129 #include <linux/cgroup_subsys.h>
133 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
135 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
136 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
137 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
138 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
139 #include <linux/cgroup_subsys.h>
142 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
143 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
144 #include <linux/cgroup_subsys.h>
148 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
149 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
150 #include <linux/cgroup_subsys.h>
154 static DEFINE_PER_CPU(struct cgroup_rstat_cpu
, cgrp_dfl_root_rstat_cpu
);
156 /* the default hierarchy */
157 struct cgroup_root cgrp_dfl_root
= { .cgrp
.rstat_cpu
= &cgrp_dfl_root_rstat_cpu
};
158 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
161 * The default hierarchy always exists but is hidden until mounted for the
162 * first time. This is for backward compatibility.
164 static bool cgrp_dfl_visible
;
166 /* some controllers are not supported in the default hierarchy */
167 static u16 cgrp_dfl_inhibit_ss_mask
;
169 /* some controllers are implicitly enabled on the default hierarchy */
170 static u16 cgrp_dfl_implicit_ss_mask
;
172 /* some controllers can be threaded on the default hierarchy */
173 static u16 cgrp_dfl_threaded_ss_mask
;
175 /* The list of hierarchy roots */
176 LIST_HEAD(cgroup_roots
);
177 static int cgroup_root_count
;
179 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
180 static DEFINE_IDR(cgroup_hierarchy_idr
);
183 * Assign a monotonically increasing serial number to csses. It guarantees
184 * cgroups with bigger numbers are newer than those with smaller numbers.
185 * Also, as csses are always appended to the parent's ->children list, it
186 * guarantees that sibling csses are always sorted in the ascending serial
187 * number order on the list. Protected by cgroup_mutex.
189 static u64 css_serial_nr_next
= 1;
192 * These bitmasks identify subsystems with specific features to avoid
193 * having to do iterative checks repeatedly.
195 static u16 have_fork_callback __read_mostly
;
196 static u16 have_exit_callback __read_mostly
;
197 static u16 have_release_callback __read_mostly
;
198 static u16 have_canfork_callback __read_mostly
;
200 /* cgroup namespace for init task */
201 struct cgroup_namespace init_cgroup_ns
= {
202 .ns
.count
= REFCOUNT_INIT(2),
203 .user_ns
= &init_user_ns
,
204 .ns
.ops
= &cgroupns_operations
,
205 .ns
.inum
= PROC_CGROUP_INIT_INO
,
206 .root_cset
= &init_css_set
,
209 static struct file_system_type cgroup2_fs_type
;
210 static struct cftype cgroup_base_files
[];
212 static int cgroup_apply_control(struct cgroup
*cgrp
);
213 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
214 static void css_task_iter_skip(struct css_task_iter
*it
,
215 struct task_struct
*task
);
216 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
217 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
218 struct cgroup_subsys
*ss
);
219 static void css_release(struct percpu_ref
*ref
);
220 static void kill_css(struct cgroup_subsys_state
*css
);
221 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
222 struct cgroup
*cgrp
, struct cftype cfts
[],
226 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
227 * @ssid: subsys ID of interest
229 * cgroup_subsys_enabled() can only be used with literal subsys names which
230 * is fine for individual subsystems but unsuitable for cgroup core. This
231 * is slower static_key_enabled() based test indexed by @ssid.
233 bool cgroup_ssid_enabled(int ssid
)
235 if (CGROUP_SUBSYS_COUNT
== 0)
238 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
242 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
243 * @cgrp: the cgroup of interest
245 * The default hierarchy is the v2 interface of cgroup and this function
246 * can be used to test whether a cgroup is on the default hierarchy for
247 * cases where a subsystem should behave differently depending on the
250 * List of changed behaviors:
252 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
253 * and "name" are disallowed.
255 * - When mounting an existing superblock, mount options should match.
257 * - Remount is disallowed.
259 * - rename(2) is disallowed.
261 * - "tasks" is removed. Everything should be at process granularity. Use
262 * "cgroup.procs" instead.
264 * - "cgroup.procs" is not sorted. pids will be unique unless they got
265 * recycled in-between reads.
267 * - "release_agent" and "notify_on_release" are removed. Replacement
268 * notification mechanism will be implemented.
270 * - "cgroup.clone_children" is removed.
272 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
273 * and its descendants contain no task; otherwise, 1. The file also
274 * generates kernfs notification which can be monitored through poll and
275 * [di]notify when the value of the file changes.
277 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
278 * take masks of ancestors with non-empty cpus/mems, instead of being
279 * moved to an ancestor.
281 * - cpuset: a task can be moved into an empty cpuset, and again it takes
282 * masks of ancestors.
284 * - blkcg: blk-throttle becomes properly hierarchical.
286 * - debug: disallowed on the default hierarchy.
288 bool cgroup_on_dfl(const struct cgroup
*cgrp
)
290 return cgrp
->root
== &cgrp_dfl_root
;
293 /* IDR wrappers which synchronize using cgroup_idr_lock */
294 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
299 idr_preload(gfp_mask
);
300 spin_lock_bh(&cgroup_idr_lock
);
301 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
302 spin_unlock_bh(&cgroup_idr_lock
);
307 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
311 spin_lock_bh(&cgroup_idr_lock
);
312 ret
= idr_replace(idr
, ptr
, id
);
313 spin_unlock_bh(&cgroup_idr_lock
);
317 static void cgroup_idr_remove(struct idr
*idr
, int id
)
319 spin_lock_bh(&cgroup_idr_lock
);
321 spin_unlock_bh(&cgroup_idr_lock
);
324 static bool cgroup_has_tasks(struct cgroup
*cgrp
)
326 return cgrp
->nr_populated_csets
;
329 bool cgroup_is_threaded(struct cgroup
*cgrp
)
331 return cgrp
->dom_cgrp
!= cgrp
;
334 /* can @cgrp host both domain and threaded children? */
335 static bool cgroup_is_mixable(struct cgroup
*cgrp
)
338 * Root isn't under domain level resource control exempting it from
339 * the no-internal-process constraint, so it can serve as a thread
340 * root and a parent of resource domains at the same time.
342 return !cgroup_parent(cgrp
);
345 /* can @cgrp become a thread root? Should always be true for a thread root */
346 static bool cgroup_can_be_thread_root(struct cgroup
*cgrp
)
348 /* mixables don't care */
349 if (cgroup_is_mixable(cgrp
))
352 /* domain roots can't be nested under threaded */
353 if (cgroup_is_threaded(cgrp
))
356 /* can only have either domain or threaded children */
357 if (cgrp
->nr_populated_domain_children
)
360 /* and no domain controllers can be enabled */
361 if (cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
367 /* is @cgrp root of a threaded subtree? */
368 bool cgroup_is_thread_root(struct cgroup
*cgrp
)
370 /* thread root should be a domain */
371 if (cgroup_is_threaded(cgrp
))
374 /* a domain w/ threaded children is a thread root */
375 if (cgrp
->nr_threaded_children
)
379 * A domain which has tasks and explicit threaded controllers
380 * enabled is a thread root.
382 if (cgroup_has_tasks(cgrp
) &&
383 (cgrp
->subtree_control
& cgrp_dfl_threaded_ss_mask
))
389 /* a domain which isn't connected to the root w/o brekage can't be used */
390 static bool cgroup_is_valid_domain(struct cgroup
*cgrp
)
392 /* the cgroup itself can be a thread root */
393 if (cgroup_is_threaded(cgrp
))
396 /* but the ancestors can't be unless mixable */
397 while ((cgrp
= cgroup_parent(cgrp
))) {
398 if (!cgroup_is_mixable(cgrp
) && cgroup_is_thread_root(cgrp
))
400 if (cgroup_is_threaded(cgrp
))
407 /* subsystems visibly enabled on a cgroup */
408 static u16
cgroup_control(struct cgroup
*cgrp
)
410 struct cgroup
*parent
= cgroup_parent(cgrp
);
411 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
414 u16 ss_mask
= parent
->subtree_control
;
416 /* threaded cgroups can only have threaded controllers */
417 if (cgroup_is_threaded(cgrp
))
418 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
422 if (cgroup_on_dfl(cgrp
))
423 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
424 cgrp_dfl_implicit_ss_mask
);
428 /* subsystems enabled on a cgroup */
429 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
431 struct cgroup
*parent
= cgroup_parent(cgrp
);
434 u16 ss_mask
= parent
->subtree_ss_mask
;
436 /* threaded cgroups can only have threaded controllers */
437 if (cgroup_is_threaded(cgrp
))
438 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
442 return cgrp
->root
->subsys_mask
;
446 * cgroup_css - obtain a cgroup's css for the specified subsystem
447 * @cgrp: the cgroup of interest
448 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
450 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
451 * function must be called either under cgroup_mutex or rcu_read_lock() and
452 * the caller is responsible for pinning the returned css if it wants to
453 * keep accessing it outside the said locks. This function may return
454 * %NULL if @cgrp doesn't have @subsys_id enabled.
456 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
457 struct cgroup_subsys
*ss
)
460 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
461 lockdep_is_held(&cgroup_mutex
));
467 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
468 * @cgrp: the cgroup of interest
469 * @ss: the subsystem of interest
471 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
472 * or is offline, %NULL is returned.
474 static struct cgroup_subsys_state
*cgroup_tryget_css(struct cgroup
*cgrp
,
475 struct cgroup_subsys
*ss
)
477 struct cgroup_subsys_state
*css
;
480 css
= cgroup_css(cgrp
, ss
);
481 if (css
&& !css_tryget_online(css
))
489 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
490 * @cgrp: the cgroup of interest
491 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
493 * Similar to cgroup_css() but returns the effective css, which is defined
494 * as the matching css of the nearest ancestor including self which has @ss
495 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
496 * function is guaranteed to return non-NULL css.
498 static struct cgroup_subsys_state
*cgroup_e_css_by_mask(struct cgroup
*cgrp
,
499 struct cgroup_subsys
*ss
)
501 lockdep_assert_held(&cgroup_mutex
);
507 * This function is used while updating css associations and thus
508 * can't test the csses directly. Test ss_mask.
510 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
511 cgrp
= cgroup_parent(cgrp
);
516 return cgroup_css(cgrp
, ss
);
520 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
521 * @cgrp: the cgroup of interest
522 * @ss: the subsystem of interest
524 * Find and get the effective css of @cgrp for @ss. The effective css is
525 * defined as the matching css of the nearest ancestor including self which
526 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
527 * the root css is returned, so this function always returns a valid css.
529 * The returned css is not guaranteed to be online, and therefore it is the
530 * callers responsibility to try get a reference for it.
532 struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
533 struct cgroup_subsys
*ss
)
535 struct cgroup_subsys_state
*css
;
538 css
= cgroup_css(cgrp
, ss
);
542 cgrp
= cgroup_parent(cgrp
);
545 return init_css_set
.subsys
[ss
->id
];
549 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
550 * @cgrp: the cgroup of interest
551 * @ss: the subsystem of interest
553 * Find and get the effective css of @cgrp for @ss. The effective css is
554 * defined as the matching css of the nearest ancestor including self which
555 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
556 * the root css is returned, so this function always returns a valid css.
557 * The returned css must be put using css_put().
559 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
560 struct cgroup_subsys
*ss
)
562 struct cgroup_subsys_state
*css
;
567 css
= cgroup_css(cgrp
, ss
);
569 if (css
&& css_tryget_online(css
))
571 cgrp
= cgroup_parent(cgrp
);
574 css
= init_css_set
.subsys
[ss
->id
];
581 static void cgroup_get_live(struct cgroup
*cgrp
)
583 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
584 css_get(&cgrp
->self
);
588 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
589 * is responsible for taking the css_set_lock.
590 * @cgrp: the cgroup in question
592 int __cgroup_task_count(const struct cgroup
*cgrp
)
595 struct cgrp_cset_link
*link
;
597 lockdep_assert_held(&css_set_lock
);
599 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
600 count
+= link
->cset
->nr_tasks
;
606 * cgroup_task_count - count the number of tasks in a cgroup.
607 * @cgrp: the cgroup in question
609 int cgroup_task_count(const struct cgroup
*cgrp
)
613 spin_lock_irq(&css_set_lock
);
614 count
= __cgroup_task_count(cgrp
);
615 spin_unlock_irq(&css_set_lock
);
620 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
622 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
623 struct cftype
*cft
= of_cft(of
);
626 * This is open and unprotected implementation of cgroup_css().
627 * seq_css() is only called from a kernfs file operation which has
628 * an active reference on the file. Because all the subsystem
629 * files are drained before a css is disassociated with a cgroup,
630 * the matching css from the cgroup's subsys table is guaranteed to
631 * be and stay valid until the enclosing operation is complete.
634 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
638 EXPORT_SYMBOL_GPL(of_css
);
641 * for_each_css - iterate all css's of a cgroup
642 * @css: the iteration cursor
643 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
644 * @cgrp: the target cgroup to iterate css's of
646 * Should be called under cgroup_[tree_]mutex.
648 #define for_each_css(css, ssid, cgrp) \
649 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
650 if (!((css) = rcu_dereference_check( \
651 (cgrp)->subsys[(ssid)], \
652 lockdep_is_held(&cgroup_mutex)))) { } \
656 * for_each_e_css - iterate all effective css's of a cgroup
657 * @css: the iteration cursor
658 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
659 * @cgrp: the target cgroup to iterate css's of
661 * Should be called under cgroup_[tree_]mutex.
663 #define for_each_e_css(css, ssid, cgrp) \
664 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
665 if (!((css) = cgroup_e_css_by_mask(cgrp, \
666 cgroup_subsys[(ssid)]))) \
671 * do_each_subsys_mask - filter for_each_subsys with a bitmask
672 * @ss: the iteration cursor
673 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
674 * @ss_mask: the bitmask
676 * The block will only run for cases where the ssid-th bit (1 << ssid) of
679 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
680 unsigned long __ss_mask = (ss_mask); \
681 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
685 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
686 (ss) = cgroup_subsys[ssid]; \
689 #define while_each_subsys_mask() \
694 /* iterate over child cgrps, lock should be held throughout iteration */
695 #define cgroup_for_each_live_child(child, cgrp) \
696 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
697 if (({ lockdep_assert_held(&cgroup_mutex); \
698 cgroup_is_dead(child); })) \
702 /* walk live descendants in pre order */
703 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
704 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
705 if (({ lockdep_assert_held(&cgroup_mutex); \
706 (dsct) = (d_css)->cgroup; \
707 cgroup_is_dead(dsct); })) \
711 /* walk live descendants in postorder */
712 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
713 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
714 if (({ lockdep_assert_held(&cgroup_mutex); \
715 (dsct) = (d_css)->cgroup; \
716 cgroup_is_dead(dsct); })) \
721 * The default css_set - used by init and its children prior to any
722 * hierarchies being mounted. It contains a pointer to the root state
723 * for each subsystem. Also used to anchor the list of css_sets. Not
724 * reference-counted, to improve performance when child cgroups
725 * haven't been created.
727 struct css_set init_css_set
= {
728 .refcount
= REFCOUNT_INIT(1),
729 .dom_cset
= &init_css_set
,
730 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
731 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
732 .dying_tasks
= LIST_HEAD_INIT(init_css_set
.dying_tasks
),
733 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
734 .threaded_csets
= LIST_HEAD_INIT(init_css_set
.threaded_csets
),
735 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
736 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
737 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
740 * The following field is re-initialized when this cset gets linked
741 * in cgroup_init(). However, let's initialize the field
742 * statically too so that the default cgroup can be accessed safely
745 .dfl_cgrp
= &cgrp_dfl_root
.cgrp
,
748 static int css_set_count
= 1; /* 1 for init_css_set */
750 static bool css_set_threaded(struct css_set
*cset
)
752 return cset
->dom_cset
!= cset
;
756 * css_set_populated - does a css_set contain any tasks?
757 * @cset: target css_set
759 * css_set_populated() should be the same as !!cset->nr_tasks at steady
760 * state. However, css_set_populated() can be called while a task is being
761 * added to or removed from the linked list before the nr_tasks is
762 * properly updated. Hence, we can't just look at ->nr_tasks here.
764 static bool css_set_populated(struct css_set
*cset
)
766 lockdep_assert_held(&css_set_lock
);
768 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
772 * cgroup_update_populated - update the populated count of a cgroup
773 * @cgrp: the target cgroup
774 * @populated: inc or dec populated count
776 * One of the css_sets associated with @cgrp is either getting its first
777 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
778 * count is propagated towards root so that a given cgroup's
779 * nr_populated_children is zero iff none of its descendants contain any
782 * @cgrp's interface file "cgroup.populated" is zero if both
783 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
784 * 1 otherwise. When the sum changes from or to zero, userland is notified
785 * that the content of the interface file has changed. This can be used to
786 * detect when @cgrp and its descendants become populated or empty.
788 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
790 struct cgroup
*child
= NULL
;
791 int adj
= populated
? 1 : -1;
793 lockdep_assert_held(&css_set_lock
);
796 bool was_populated
= cgroup_is_populated(cgrp
);
799 cgrp
->nr_populated_csets
+= adj
;
801 if (cgroup_is_threaded(child
))
802 cgrp
->nr_populated_threaded_children
+= adj
;
804 cgrp
->nr_populated_domain_children
+= adj
;
807 if (was_populated
== cgroup_is_populated(cgrp
))
810 cgroup1_check_for_release(cgrp
);
811 TRACE_CGROUP_PATH(notify_populated
, cgrp
,
812 cgroup_is_populated(cgrp
));
813 cgroup_file_notify(&cgrp
->events_file
);
816 cgrp
= cgroup_parent(cgrp
);
821 * css_set_update_populated - update populated state of a css_set
822 * @cset: target css_set
823 * @populated: whether @cset is populated or depopulated
825 * @cset is either getting the first task or losing the last. Update the
826 * populated counters of all associated cgroups accordingly.
828 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
830 struct cgrp_cset_link
*link
;
832 lockdep_assert_held(&css_set_lock
);
834 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
835 cgroup_update_populated(link
->cgrp
, populated
);
839 * @task is leaving, advance task iterators which are pointing to it so
840 * that they can resume at the next position. Advancing an iterator might
841 * remove it from the list, use safe walk. See css_task_iter_skip() for
844 static void css_set_skip_task_iters(struct css_set
*cset
,
845 struct task_struct
*task
)
847 struct css_task_iter
*it
, *pos
;
849 list_for_each_entry_safe(it
, pos
, &cset
->task_iters
, iters_node
)
850 css_task_iter_skip(it
, task
);
854 * css_set_move_task - move a task from one css_set to another
855 * @task: task being moved
856 * @from_cset: css_set @task currently belongs to (may be NULL)
857 * @to_cset: new css_set @task is being moved to (may be NULL)
858 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
860 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
861 * css_set, @from_cset can be NULL. If @task is being disassociated
862 * instead of moved, @to_cset can be NULL.
864 * This function automatically handles populated counter updates and
865 * css_task_iter adjustments but the caller is responsible for managing
866 * @from_cset and @to_cset's reference counts.
868 static void css_set_move_task(struct task_struct
*task
,
869 struct css_set
*from_cset
, struct css_set
*to_cset
,
872 lockdep_assert_held(&css_set_lock
);
874 if (to_cset
&& !css_set_populated(to_cset
))
875 css_set_update_populated(to_cset
, true);
878 WARN_ON_ONCE(list_empty(&task
->cg_list
));
880 css_set_skip_task_iters(from_cset
, task
);
881 list_del_init(&task
->cg_list
);
882 if (!css_set_populated(from_cset
))
883 css_set_update_populated(from_cset
, false);
885 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
890 * We are synchronized through cgroup_threadgroup_rwsem
891 * against PF_EXITING setting such that we can't race
892 * against cgroup_exit()/cgroup_free() dropping the css_set.
894 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
896 cgroup_move_task(task
, to_cset
);
897 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
903 * hash table for cgroup groups. This improves the performance to find
904 * an existing css_set. This hash doesn't (currently) take into
905 * account cgroups in empty hierarchies.
907 #define CSS_SET_HASH_BITS 7
908 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
910 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
912 unsigned long key
= 0UL;
913 struct cgroup_subsys
*ss
;
916 for_each_subsys(ss
, i
)
917 key
+= (unsigned long)css
[i
];
918 key
= (key
>> 16) ^ key
;
923 void put_css_set_locked(struct css_set
*cset
)
925 struct cgrp_cset_link
*link
, *tmp_link
;
926 struct cgroup_subsys
*ss
;
929 lockdep_assert_held(&css_set_lock
);
931 if (!refcount_dec_and_test(&cset
->refcount
))
934 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
936 /* This css_set is dead. Unlink it and release cgroup and css refs */
937 for_each_subsys(ss
, ssid
) {
938 list_del(&cset
->e_cset_node
[ssid
]);
939 css_put(cset
->subsys
[ssid
]);
941 hash_del(&cset
->hlist
);
944 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
945 list_del(&link
->cset_link
);
946 list_del(&link
->cgrp_link
);
947 if (cgroup_parent(link
->cgrp
))
948 cgroup_put(link
->cgrp
);
952 if (css_set_threaded(cset
)) {
953 list_del(&cset
->threaded_csets_node
);
954 put_css_set_locked(cset
->dom_cset
);
957 kfree_rcu(cset
, rcu_head
);
961 * compare_css_sets - helper function for find_existing_css_set().
962 * @cset: candidate css_set being tested
963 * @old_cset: existing css_set for a task
964 * @new_cgrp: cgroup that's being entered by the task
965 * @template: desired set of css pointers in css_set (pre-calculated)
967 * Returns true if "cset" matches "old_cset" except for the hierarchy
968 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
970 static bool compare_css_sets(struct css_set
*cset
,
971 struct css_set
*old_cset
,
972 struct cgroup
*new_cgrp
,
973 struct cgroup_subsys_state
*template[])
975 struct cgroup
*new_dfl_cgrp
;
976 struct list_head
*l1
, *l2
;
979 * On the default hierarchy, there can be csets which are
980 * associated with the same set of cgroups but different csses.
981 * Let's first ensure that csses match.
983 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
987 /* @cset's domain should match the default cgroup's */
988 if (cgroup_on_dfl(new_cgrp
))
989 new_dfl_cgrp
= new_cgrp
;
991 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
993 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
997 * Compare cgroup pointers in order to distinguish between
998 * different cgroups in hierarchies. As different cgroups may
999 * share the same effective css, this comparison is always
1002 l1
= &cset
->cgrp_links
;
1003 l2
= &old_cset
->cgrp_links
;
1005 struct cgrp_cset_link
*link1
, *link2
;
1006 struct cgroup
*cgrp1
, *cgrp2
;
1010 /* See if we reached the end - both lists are equal length. */
1011 if (l1
== &cset
->cgrp_links
) {
1012 BUG_ON(l2
!= &old_cset
->cgrp_links
);
1015 BUG_ON(l2
== &old_cset
->cgrp_links
);
1017 /* Locate the cgroups associated with these links. */
1018 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
1019 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
1020 cgrp1
= link1
->cgrp
;
1021 cgrp2
= link2
->cgrp
;
1022 /* Hierarchies should be linked in the same order. */
1023 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
1026 * If this hierarchy is the hierarchy of the cgroup
1027 * that's changing, then we need to check that this
1028 * css_set points to the new cgroup; if it's any other
1029 * hierarchy, then this css_set should point to the
1030 * same cgroup as the old css_set.
1032 if (cgrp1
->root
== new_cgrp
->root
) {
1033 if (cgrp1
!= new_cgrp
)
1044 * find_existing_css_set - init css array and find the matching css_set
1045 * @old_cset: the css_set that we're using before the cgroup transition
1046 * @cgrp: the cgroup that we're moving into
1047 * @template: out param for the new set of csses, should be clear on entry
1049 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
1050 struct cgroup
*cgrp
,
1051 struct cgroup_subsys_state
*template[])
1053 struct cgroup_root
*root
= cgrp
->root
;
1054 struct cgroup_subsys
*ss
;
1055 struct css_set
*cset
;
1060 * Build the set of subsystem state objects that we want to see in the
1061 * new css_set. While subsystems can change globally, the entries here
1062 * won't change, so no need for locking.
1064 for_each_subsys(ss
, i
) {
1065 if (root
->subsys_mask
& (1UL << i
)) {
1067 * @ss is in this hierarchy, so we want the
1068 * effective css from @cgrp.
1070 template[i
] = cgroup_e_css_by_mask(cgrp
, ss
);
1073 * @ss is not in this hierarchy, so we don't want
1074 * to change the css.
1076 template[i
] = old_cset
->subsys
[i
];
1080 key
= css_set_hash(template);
1081 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
1082 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
1085 /* This css_set matches what we need */
1089 /* No existing cgroup group matched */
1093 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
1095 struct cgrp_cset_link
*link
, *tmp_link
;
1097 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
1098 list_del(&link
->cset_link
);
1104 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1105 * @count: the number of links to allocate
1106 * @tmp_links: list_head the allocated links are put on
1108 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1109 * through ->cset_link. Returns 0 on success or -errno.
1111 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1113 struct cgrp_cset_link
*link
;
1116 INIT_LIST_HEAD(tmp_links
);
1118 for (i
= 0; i
< count
; i
++) {
1119 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1121 free_cgrp_cset_links(tmp_links
);
1124 list_add(&link
->cset_link
, tmp_links
);
1130 * link_css_set - a helper function to link a css_set to a cgroup
1131 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1132 * @cset: the css_set to be linked
1133 * @cgrp: the destination cgroup
1135 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1136 struct cgroup
*cgrp
)
1138 struct cgrp_cset_link
*link
;
1140 BUG_ON(list_empty(tmp_links
));
1142 if (cgroup_on_dfl(cgrp
))
1143 cset
->dfl_cgrp
= cgrp
;
1145 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1150 * Always add links to the tail of the lists so that the lists are
1151 * in chronological order.
1153 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1154 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1156 if (cgroup_parent(cgrp
))
1157 cgroup_get_live(cgrp
);
1161 * find_css_set - return a new css_set with one cgroup updated
1162 * @old_cset: the baseline css_set
1163 * @cgrp: the cgroup to be updated
1165 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1166 * substituted into the appropriate hierarchy.
1168 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1169 struct cgroup
*cgrp
)
1171 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1172 struct css_set
*cset
;
1173 struct list_head tmp_links
;
1174 struct cgrp_cset_link
*link
;
1175 struct cgroup_subsys
*ss
;
1179 lockdep_assert_held(&cgroup_mutex
);
1181 /* First see if we already have a cgroup group that matches
1182 * the desired set */
1183 spin_lock_irq(&css_set_lock
);
1184 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1187 spin_unlock_irq(&css_set_lock
);
1192 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1196 /* Allocate all the cgrp_cset_link objects that we'll need */
1197 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1202 refcount_set(&cset
->refcount
, 1);
1203 cset
->dom_cset
= cset
;
1204 INIT_LIST_HEAD(&cset
->tasks
);
1205 INIT_LIST_HEAD(&cset
->mg_tasks
);
1206 INIT_LIST_HEAD(&cset
->dying_tasks
);
1207 INIT_LIST_HEAD(&cset
->task_iters
);
1208 INIT_LIST_HEAD(&cset
->threaded_csets
);
1209 INIT_HLIST_NODE(&cset
->hlist
);
1210 INIT_LIST_HEAD(&cset
->cgrp_links
);
1211 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1212 INIT_LIST_HEAD(&cset
->mg_node
);
1214 /* Copy the set of subsystem state objects generated in
1215 * find_existing_css_set() */
1216 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1218 spin_lock_irq(&css_set_lock
);
1219 /* Add reference counts and links from the new css_set. */
1220 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1221 struct cgroup
*c
= link
->cgrp
;
1223 if (c
->root
== cgrp
->root
)
1225 link_css_set(&tmp_links
, cset
, c
);
1228 BUG_ON(!list_empty(&tmp_links
));
1232 /* Add @cset to the hash table */
1233 key
= css_set_hash(cset
->subsys
);
1234 hash_add(css_set_table
, &cset
->hlist
, key
);
1236 for_each_subsys(ss
, ssid
) {
1237 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1239 list_add_tail(&cset
->e_cset_node
[ssid
],
1240 &css
->cgroup
->e_csets
[ssid
]);
1244 spin_unlock_irq(&css_set_lock
);
1247 * If @cset should be threaded, look up the matching dom_cset and
1248 * link them up. We first fully initialize @cset then look for the
1249 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1250 * to stay empty until we return.
1252 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1253 struct css_set
*dcset
;
1255 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1261 spin_lock_irq(&css_set_lock
);
1262 cset
->dom_cset
= dcset
;
1263 list_add_tail(&cset
->threaded_csets_node
,
1264 &dcset
->threaded_csets
);
1265 spin_unlock_irq(&css_set_lock
);
1271 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1273 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1275 return root_cgrp
->root
;
1278 static int cgroup_init_root_id(struct cgroup_root
*root
)
1282 lockdep_assert_held(&cgroup_mutex
);
1284 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1288 root
->hierarchy_id
= id
;
1292 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1294 lockdep_assert_held(&cgroup_mutex
);
1296 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1299 void cgroup_free_root(struct cgroup_root
*root
)
1304 static void cgroup_destroy_root(struct cgroup_root
*root
)
1306 struct cgroup
*cgrp
= &root
->cgrp
;
1307 struct cgrp_cset_link
*link
, *tmp_link
;
1309 trace_cgroup_destroy_root(root
);
1311 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1313 BUG_ON(atomic_read(&root
->nr_cgrps
));
1314 BUG_ON(!list_empty(&cgrp
->self
.children
));
1316 /* Rebind all subsystems back to the default hierarchy */
1317 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1320 * Release all the links from cset_links to this hierarchy's
1323 spin_lock_irq(&css_set_lock
);
1325 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1326 list_del(&link
->cset_link
);
1327 list_del(&link
->cgrp_link
);
1331 spin_unlock_irq(&css_set_lock
);
1333 if (!list_empty(&root
->root_list
)) {
1334 list_del(&root
->root_list
);
1335 cgroup_root_count
--;
1338 cgroup_exit_root_id(root
);
1340 mutex_unlock(&cgroup_mutex
);
1342 kernfs_destroy_root(root
->kf_root
);
1343 cgroup_free_root(root
);
1347 * look up cgroup associated with current task's cgroup namespace on the
1348 * specified hierarchy
1350 static struct cgroup
*
1351 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1353 struct cgroup
*res
= NULL
;
1354 struct css_set
*cset
;
1356 lockdep_assert_held(&css_set_lock
);
1360 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1361 if (cset
== &init_css_set
) {
1363 } else if (root
== &cgrp_dfl_root
) {
1364 res
= cset
->dfl_cgrp
;
1366 struct cgrp_cset_link
*link
;
1368 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1369 struct cgroup
*c
= link
->cgrp
;
1371 if (c
->root
== root
) {
1383 /* look up cgroup associated with given css_set on the specified hierarchy */
1384 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1385 struct cgroup_root
*root
)
1387 struct cgroup
*res
= NULL
;
1389 lockdep_assert_held(&cgroup_mutex
);
1390 lockdep_assert_held(&css_set_lock
);
1392 if (cset
== &init_css_set
) {
1394 } else if (root
== &cgrp_dfl_root
) {
1395 res
= cset
->dfl_cgrp
;
1397 struct cgrp_cset_link
*link
;
1399 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1400 struct cgroup
*c
= link
->cgrp
;
1402 if (c
->root
== root
) {
1414 * Return the cgroup for "task" from the given hierarchy. Must be
1415 * called with cgroup_mutex and css_set_lock held.
1417 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1418 struct cgroup_root
*root
)
1421 * No need to lock the task - since we hold css_set_lock the
1422 * task can't change groups.
1424 return cset_cgroup_from_root(task_css_set(task
), root
);
1428 * A task must hold cgroup_mutex to modify cgroups.
1430 * Any task can increment and decrement the count field without lock.
1431 * So in general, code holding cgroup_mutex can't rely on the count
1432 * field not changing. However, if the count goes to zero, then only
1433 * cgroup_attach_task() can increment it again. Because a count of zero
1434 * means that no tasks are currently attached, therefore there is no
1435 * way a task attached to that cgroup can fork (the other way to
1436 * increment the count). So code holding cgroup_mutex can safely
1437 * assume that if the count is zero, it will stay zero. Similarly, if
1438 * a task holds cgroup_mutex on a cgroup with zero count, it
1439 * knows that the cgroup won't be removed, as cgroup_rmdir()
1442 * A cgroup can only be deleted if both its 'count' of using tasks
1443 * is zero, and its list of 'children' cgroups is empty. Since all
1444 * tasks in the system use _some_ cgroup, and since there is always at
1445 * least one task in the system (init, pid == 1), therefore, root cgroup
1446 * always has either children cgroups and/or using tasks. So we don't
1447 * need a special hack to ensure that root cgroup cannot be deleted.
1449 * P.S. One more locking exception. RCU is used to guard the
1450 * update of a tasks cgroup pointer by cgroup_attach_task()
1453 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1455 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1458 struct cgroup_subsys
*ss
= cft
->ss
;
1460 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1461 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
1462 const char *dbg
= (cft
->flags
& CFTYPE_DEBUG
) ? ".__DEBUG__." : "";
1464 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s%s.%s",
1465 dbg
, cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1468 strscpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1474 * cgroup_file_mode - deduce file mode of a control file
1475 * @cft: the control file in question
1477 * S_IRUGO for read, S_IWUSR for write.
1479 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1483 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1486 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1487 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1497 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1498 * @subtree_control: the new subtree_control mask to consider
1499 * @this_ss_mask: available subsystems
1501 * On the default hierarchy, a subsystem may request other subsystems to be
1502 * enabled together through its ->depends_on mask. In such cases, more
1503 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1505 * This function calculates which subsystems need to be enabled if
1506 * @subtree_control is to be applied while restricted to @this_ss_mask.
1508 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1510 u16 cur_ss_mask
= subtree_control
;
1511 struct cgroup_subsys
*ss
;
1514 lockdep_assert_held(&cgroup_mutex
);
1516 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1519 u16 new_ss_mask
= cur_ss_mask
;
1521 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1522 new_ss_mask
|= ss
->depends_on
;
1523 } while_each_subsys_mask();
1526 * Mask out subsystems which aren't available. This can
1527 * happen only if some depended-upon subsystems were bound
1528 * to non-default hierarchies.
1530 new_ss_mask
&= this_ss_mask
;
1532 if (new_ss_mask
== cur_ss_mask
)
1534 cur_ss_mask
= new_ss_mask
;
1541 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1542 * @kn: the kernfs_node being serviced
1544 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1545 * the method finishes if locking succeeded. Note that once this function
1546 * returns the cgroup returned by cgroup_kn_lock_live() may become
1547 * inaccessible any time. If the caller intends to continue to access the
1548 * cgroup, it should pin it before invoking this function.
1550 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1552 struct cgroup
*cgrp
;
1554 if (kernfs_type(kn
) == KERNFS_DIR
)
1557 cgrp
= kn
->parent
->priv
;
1559 mutex_unlock(&cgroup_mutex
);
1561 kernfs_unbreak_active_protection(kn
);
1566 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1567 * @kn: the kernfs_node being serviced
1568 * @drain_offline: perform offline draining on the cgroup
1570 * This helper is to be used by a cgroup kernfs method currently servicing
1571 * @kn. It breaks the active protection, performs cgroup locking and
1572 * verifies that the associated cgroup is alive. Returns the cgroup if
1573 * alive; otherwise, %NULL. A successful return should be undone by a
1574 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1575 * cgroup is drained of offlining csses before return.
1577 * Any cgroup kernfs method implementation which requires locking the
1578 * associated cgroup should use this helper. It avoids nesting cgroup
1579 * locking under kernfs active protection and allows all kernfs operations
1580 * including self-removal.
1582 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1584 struct cgroup
*cgrp
;
1586 if (kernfs_type(kn
) == KERNFS_DIR
)
1589 cgrp
= kn
->parent
->priv
;
1592 * We're gonna grab cgroup_mutex which nests outside kernfs
1593 * active_ref. cgroup liveliness check alone provides enough
1594 * protection against removal. Ensure @cgrp stays accessible and
1595 * break the active_ref protection.
1597 if (!cgroup_tryget(cgrp
))
1599 kernfs_break_active_protection(kn
);
1602 cgroup_lock_and_drain_offline(cgrp
);
1604 mutex_lock(&cgroup_mutex
);
1606 if (!cgroup_is_dead(cgrp
))
1609 cgroup_kn_unlock(kn
);
1613 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1615 char name
[CGROUP_FILE_NAME_MAX
];
1617 lockdep_assert_held(&cgroup_mutex
);
1619 if (cft
->file_offset
) {
1620 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1621 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1623 spin_lock_irq(&cgroup_file_kn_lock
);
1625 spin_unlock_irq(&cgroup_file_kn_lock
);
1627 del_timer_sync(&cfile
->notify_timer
);
1630 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1634 * css_clear_dir - remove subsys files in a cgroup directory
1637 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1639 struct cgroup
*cgrp
= css
->cgroup
;
1640 struct cftype
*cfts
;
1642 if (!(css
->flags
& CSS_VISIBLE
))
1645 css
->flags
&= ~CSS_VISIBLE
;
1648 if (cgroup_on_dfl(cgrp
))
1649 cfts
= cgroup_base_files
;
1651 cfts
= cgroup1_base_files
;
1653 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1655 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1656 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1661 * css_populate_dir - create subsys files in a cgroup directory
1664 * On failure, no file is added.
1666 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1668 struct cgroup
*cgrp
= css
->cgroup
;
1669 struct cftype
*cfts
, *failed_cfts
;
1672 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1676 if (cgroup_on_dfl(cgrp
))
1677 cfts
= cgroup_base_files
;
1679 cfts
= cgroup1_base_files
;
1681 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1685 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1686 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1694 css
->flags
|= CSS_VISIBLE
;
1698 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1699 if (cfts
== failed_cfts
)
1701 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1706 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1708 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1709 struct cgroup_subsys
*ss
;
1712 lockdep_assert_held(&cgroup_mutex
);
1714 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1716 * If @ss has non-root csses attached to it, can't move.
1717 * If @ss is an implicit controller, it is exempt from this
1718 * rule and can be stolen.
1720 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1721 !ss
->implicit_on_dfl
)
1724 /* can't move between two non-dummy roots either */
1725 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1727 } while_each_subsys_mask();
1729 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1730 struct cgroup_root
*src_root
= ss
->root
;
1731 struct cgroup
*scgrp
= &src_root
->cgrp
;
1732 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1733 struct css_set
*cset
;
1735 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1737 /* disable from the source */
1738 src_root
->subsys_mask
&= ~(1 << ssid
);
1739 WARN_ON(cgroup_apply_control(scgrp
));
1740 cgroup_finalize_control(scgrp
, 0);
1743 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1744 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1745 ss
->root
= dst_root
;
1746 css
->cgroup
= dcgrp
;
1748 spin_lock_irq(&css_set_lock
);
1749 hash_for_each(css_set_table
, i
, cset
, hlist
)
1750 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1751 &dcgrp
->e_csets
[ss
->id
]);
1752 spin_unlock_irq(&css_set_lock
);
1754 /* default hierarchy doesn't enable controllers by default */
1755 dst_root
->subsys_mask
|= 1 << ssid
;
1756 if (dst_root
== &cgrp_dfl_root
) {
1757 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1759 dcgrp
->subtree_control
|= 1 << ssid
;
1760 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1763 ret
= cgroup_apply_control(dcgrp
);
1765 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1770 } while_each_subsys_mask();
1772 kernfs_activate(dcgrp
->kn
);
1776 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1777 struct kernfs_root
*kf_root
)
1781 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1782 struct cgroup
*ns_cgroup
;
1784 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1788 spin_lock_irq(&css_set_lock
);
1789 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1790 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1791 spin_unlock_irq(&css_set_lock
);
1793 if (len
>= PATH_MAX
)
1796 seq_escape(sf
, buf
, " \t\n\\");
1803 enum cgroup2_param
{
1805 Opt_memory_localevents
,
1806 Opt_memory_recursiveprot
,
1810 static const struct fs_parameter_spec cgroup2_fs_parameters
[] = {
1811 fsparam_flag("nsdelegate", Opt_nsdelegate
),
1812 fsparam_flag("memory_localevents", Opt_memory_localevents
),
1813 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot
),
1817 static int cgroup2_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
1819 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1820 struct fs_parse_result result
;
1823 opt
= fs_parse(fc
, cgroup2_fs_parameters
, param
, &result
);
1828 case Opt_nsdelegate
:
1829 ctx
->flags
|= CGRP_ROOT_NS_DELEGATE
;
1831 case Opt_memory_localevents
:
1832 ctx
->flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1834 case Opt_memory_recursiveprot
:
1835 ctx
->flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1841 static void apply_cgroup_root_flags(unsigned int root_flags
)
1843 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1844 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1845 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1847 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1849 if (root_flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1850 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1852 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1854 if (root_flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1855 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1857 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1861 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1863 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1864 seq_puts(seq
, ",nsdelegate");
1865 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1866 seq_puts(seq
, ",memory_localevents");
1867 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1868 seq_puts(seq
, ",memory_recursiveprot");
1872 static int cgroup_reconfigure(struct fs_context
*fc
)
1874 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1876 apply_cgroup_root_flags(ctx
->flags
);
1880 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1882 struct cgroup_subsys
*ss
;
1885 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1886 INIT_LIST_HEAD(&cgrp
->self
.children
);
1887 INIT_LIST_HEAD(&cgrp
->cset_links
);
1888 INIT_LIST_HEAD(&cgrp
->pidlists
);
1889 mutex_init(&cgrp
->pidlist_mutex
);
1890 cgrp
->self
.cgroup
= cgrp
;
1891 cgrp
->self
.flags
|= CSS_ONLINE
;
1892 cgrp
->dom_cgrp
= cgrp
;
1893 cgrp
->max_descendants
= INT_MAX
;
1894 cgrp
->max_depth
= INT_MAX
;
1895 INIT_LIST_HEAD(&cgrp
->rstat_css_list
);
1896 prev_cputime_init(&cgrp
->prev_cputime
);
1898 for_each_subsys(ss
, ssid
)
1899 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1901 init_waitqueue_head(&cgrp
->offline_waitq
);
1902 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1905 void init_cgroup_root(struct cgroup_fs_context
*ctx
)
1907 struct cgroup_root
*root
= ctx
->root
;
1908 struct cgroup
*cgrp
= &root
->cgrp
;
1910 INIT_LIST_HEAD(&root
->root_list
);
1911 atomic_set(&root
->nr_cgrps
, 1);
1913 init_cgroup_housekeeping(cgrp
);
1915 root
->flags
= ctx
->flags
;
1916 if (ctx
->release_agent
)
1917 strscpy(root
->release_agent_path
, ctx
->release_agent
, PATH_MAX
);
1919 strscpy(root
->name
, ctx
->name
, MAX_CGROUP_ROOT_NAMELEN
);
1920 if (ctx
->cpuset_clone_children
)
1921 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1924 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
1926 LIST_HEAD(tmp_links
);
1927 struct cgroup
*root_cgrp
= &root
->cgrp
;
1928 struct kernfs_syscall_ops
*kf_sops
;
1929 struct css_set
*cset
;
1932 lockdep_assert_held(&cgroup_mutex
);
1934 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
1940 * We're accessing css_set_count without locking css_set_lock here,
1941 * but that's OK - it can only be increased by someone holding
1942 * cgroup_lock, and that's us. Later rebinding may disable
1943 * controllers on the default hierarchy and thus create new csets,
1944 * which can't be more than the existing ones. Allocate 2x.
1946 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
1950 ret
= cgroup_init_root_id(root
);
1954 kf_sops
= root
== &cgrp_dfl_root
?
1955 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
1957 root
->kf_root
= kernfs_create_root(kf_sops
,
1958 KERNFS_ROOT_CREATE_DEACTIVATED
|
1959 KERNFS_ROOT_SUPPORT_EXPORTOP
|
1960 KERNFS_ROOT_SUPPORT_USER_XATTR
,
1962 if (IS_ERR(root
->kf_root
)) {
1963 ret
= PTR_ERR(root
->kf_root
);
1966 root_cgrp
->kn
= root
->kf_root
->kn
;
1967 WARN_ON_ONCE(cgroup_ino(root_cgrp
) != 1);
1968 root_cgrp
->ancestor_ids
[0] = cgroup_id(root_cgrp
);
1970 ret
= css_populate_dir(&root_cgrp
->self
);
1974 ret
= rebind_subsystems(root
, ss_mask
);
1978 ret
= cgroup_bpf_inherit(root_cgrp
);
1981 trace_cgroup_setup_root(root
);
1984 * There must be no failure case after here, since rebinding takes
1985 * care of subsystems' refcounts, which are explicitly dropped in
1986 * the failure exit path.
1988 list_add(&root
->root_list
, &cgroup_roots
);
1989 cgroup_root_count
++;
1992 * Link the root cgroup in this hierarchy into all the css_set
1995 spin_lock_irq(&css_set_lock
);
1996 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1997 link_css_set(&tmp_links
, cset
, root_cgrp
);
1998 if (css_set_populated(cset
))
1999 cgroup_update_populated(root_cgrp
, true);
2001 spin_unlock_irq(&css_set_lock
);
2003 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2004 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2010 kernfs_destroy_root(root
->kf_root
);
2011 root
->kf_root
= NULL
;
2013 cgroup_exit_root_id(root
);
2015 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2017 free_cgrp_cset_links(&tmp_links
);
2021 int cgroup_do_get_tree(struct fs_context
*fc
)
2023 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2026 ctx
->kfc
.root
= ctx
->root
->kf_root
;
2027 if (fc
->fs_type
== &cgroup2_fs_type
)
2028 ctx
->kfc
.magic
= CGROUP2_SUPER_MAGIC
;
2030 ctx
->kfc
.magic
= CGROUP_SUPER_MAGIC
;
2031 ret
= kernfs_get_tree(fc
);
2034 * In non-init cgroup namespace, instead of root cgroup's dentry,
2035 * we return the dentry corresponding to the cgroupns->root_cgrp.
2037 if (!ret
&& ctx
->ns
!= &init_cgroup_ns
) {
2038 struct dentry
*nsdentry
;
2039 struct super_block
*sb
= fc
->root
->d_sb
;
2040 struct cgroup
*cgrp
;
2042 mutex_lock(&cgroup_mutex
);
2043 spin_lock_irq(&css_set_lock
);
2045 cgrp
= cset_cgroup_from_root(ctx
->ns
->root_cset
, ctx
->root
);
2047 spin_unlock_irq(&css_set_lock
);
2048 mutex_unlock(&cgroup_mutex
);
2050 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
2052 if (IS_ERR(nsdentry
)) {
2053 deactivate_locked_super(sb
);
2054 ret
= PTR_ERR(nsdentry
);
2057 fc
->root
= nsdentry
;
2060 if (!ctx
->kfc
.new_sb_created
)
2061 cgroup_put(&ctx
->root
->cgrp
);
2067 * Destroy a cgroup filesystem context.
2069 static void cgroup_fs_context_free(struct fs_context
*fc
)
2071 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2074 kfree(ctx
->release_agent
);
2075 put_cgroup_ns(ctx
->ns
);
2076 kernfs_free_fs_context(fc
);
2080 static int cgroup_get_tree(struct fs_context
*fc
)
2082 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2085 cgrp_dfl_visible
= true;
2086 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2087 ctx
->root
= &cgrp_dfl_root
;
2089 ret
= cgroup_do_get_tree(fc
);
2091 apply_cgroup_root_flags(ctx
->flags
);
2095 static const struct fs_context_operations cgroup_fs_context_ops
= {
2096 .free
= cgroup_fs_context_free
,
2097 .parse_param
= cgroup2_parse_param
,
2098 .get_tree
= cgroup_get_tree
,
2099 .reconfigure
= cgroup_reconfigure
,
2102 static const struct fs_context_operations cgroup1_fs_context_ops
= {
2103 .free
= cgroup_fs_context_free
,
2104 .parse_param
= cgroup1_parse_param
,
2105 .get_tree
= cgroup1_get_tree
,
2106 .reconfigure
= cgroup1_reconfigure
,
2110 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2111 * we select the namespace we're going to use.
2113 static int cgroup_init_fs_context(struct fs_context
*fc
)
2115 struct cgroup_fs_context
*ctx
;
2117 ctx
= kzalloc(sizeof(struct cgroup_fs_context
), GFP_KERNEL
);
2121 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
2122 get_cgroup_ns(ctx
->ns
);
2123 fc
->fs_private
= &ctx
->kfc
;
2124 if (fc
->fs_type
== &cgroup2_fs_type
)
2125 fc
->ops
= &cgroup_fs_context_ops
;
2127 fc
->ops
= &cgroup1_fs_context_ops
;
2128 put_user_ns(fc
->user_ns
);
2129 fc
->user_ns
= get_user_ns(ctx
->ns
->user_ns
);
2134 static void cgroup_kill_sb(struct super_block
*sb
)
2136 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2137 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2140 * If @root doesn't have any children, start killing it.
2141 * This prevents new mounts by disabling percpu_ref_tryget_live().
2142 * cgroup_mount() may wait for @root's release.
2144 * And don't kill the default root.
2146 if (list_empty(&root
->cgrp
.self
.children
) && root
!= &cgrp_dfl_root
&&
2147 !percpu_ref_is_dying(&root
->cgrp
.self
.refcnt
))
2148 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2149 cgroup_put(&root
->cgrp
);
2153 struct file_system_type cgroup_fs_type
= {
2155 .init_fs_context
= cgroup_init_fs_context
,
2156 .parameters
= cgroup1_fs_parameters
,
2157 .kill_sb
= cgroup_kill_sb
,
2158 .fs_flags
= FS_USERNS_MOUNT
,
2161 static struct file_system_type cgroup2_fs_type
= {
2163 .init_fs_context
= cgroup_init_fs_context
,
2164 .parameters
= cgroup2_fs_parameters
,
2165 .kill_sb
= cgroup_kill_sb
,
2166 .fs_flags
= FS_USERNS_MOUNT
,
2169 #ifdef CONFIG_CPUSETS
2170 static const struct fs_context_operations cpuset_fs_context_ops
= {
2171 .get_tree
= cgroup1_get_tree
,
2172 .free
= cgroup_fs_context_free
,
2176 * This is ugly, but preserves the userspace API for existing cpuset
2177 * users. If someone tries to mount the "cpuset" filesystem, we
2178 * silently switch it to mount "cgroup" instead
2180 static int cpuset_init_fs_context(struct fs_context
*fc
)
2182 char *agent
= kstrdup("/sbin/cpuset_release_agent", GFP_USER
);
2183 struct cgroup_fs_context
*ctx
;
2186 err
= cgroup_init_fs_context(fc
);
2192 fc
->ops
= &cpuset_fs_context_ops
;
2194 ctx
= cgroup_fc2context(fc
);
2195 ctx
->subsys_mask
= 1 << cpuset_cgrp_id
;
2196 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
2197 ctx
->release_agent
= agent
;
2199 get_filesystem(&cgroup_fs_type
);
2200 put_filesystem(fc
->fs_type
);
2201 fc
->fs_type
= &cgroup_fs_type
;
2206 static struct file_system_type cpuset_fs_type
= {
2208 .init_fs_context
= cpuset_init_fs_context
,
2209 .fs_flags
= FS_USERNS_MOUNT
,
2213 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2214 struct cgroup_namespace
*ns
)
2216 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2218 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2221 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2222 struct cgroup_namespace
*ns
)
2226 mutex_lock(&cgroup_mutex
);
2227 spin_lock_irq(&css_set_lock
);
2229 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2231 spin_unlock_irq(&css_set_lock
);
2232 mutex_unlock(&cgroup_mutex
);
2236 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2239 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2240 * @task: target task
2241 * @buf: the buffer to write the path into
2242 * @buflen: the length of the buffer
2244 * Determine @task's cgroup on the first (the one with the lowest non-zero
2245 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2246 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2247 * cgroup controller callbacks.
2249 * Return value is the same as kernfs_path().
2251 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2253 struct cgroup_root
*root
;
2254 struct cgroup
*cgrp
;
2255 int hierarchy_id
= 1;
2258 mutex_lock(&cgroup_mutex
);
2259 spin_lock_irq(&css_set_lock
);
2261 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2264 cgrp
= task_cgroup_from_root(task
, root
);
2265 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2267 /* if no hierarchy exists, everyone is in "/" */
2268 ret
= strlcpy(buf
, "/", buflen
);
2271 spin_unlock_irq(&css_set_lock
);
2272 mutex_unlock(&cgroup_mutex
);
2275 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2278 * cgroup_migrate_add_task - add a migration target task to a migration context
2279 * @task: target task
2280 * @mgctx: target migration context
2282 * Add @task, which is a migration target, to @mgctx->tset. This function
2283 * becomes noop if @task doesn't need to be migrated. @task's css_set
2284 * should have been added as a migration source and @task->cg_list will be
2285 * moved from the css_set's tasks list to mg_tasks one.
2287 static void cgroup_migrate_add_task(struct task_struct
*task
,
2288 struct cgroup_mgctx
*mgctx
)
2290 struct css_set
*cset
;
2292 lockdep_assert_held(&css_set_lock
);
2294 /* @task either already exited or can't exit until the end */
2295 if (task
->flags
& PF_EXITING
)
2298 /* cgroup_threadgroup_rwsem protects racing against forks */
2299 WARN_ON_ONCE(list_empty(&task
->cg_list
));
2301 cset
= task_css_set(task
);
2302 if (!cset
->mg_src_cgrp
)
2305 mgctx
->tset
.nr_tasks
++;
2307 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2308 if (list_empty(&cset
->mg_node
))
2309 list_add_tail(&cset
->mg_node
,
2310 &mgctx
->tset
.src_csets
);
2311 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2312 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2313 &mgctx
->tset
.dst_csets
);
2317 * cgroup_taskset_first - reset taskset and return the first task
2318 * @tset: taskset of interest
2319 * @dst_cssp: output variable for the destination css
2321 * @tset iteration is initialized and the first task is returned.
2323 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2324 struct cgroup_subsys_state
**dst_cssp
)
2326 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2327 tset
->cur_task
= NULL
;
2329 return cgroup_taskset_next(tset
, dst_cssp
);
2333 * cgroup_taskset_next - iterate to the next task in taskset
2334 * @tset: taskset of interest
2335 * @dst_cssp: output variable for the destination css
2337 * Return the next task in @tset. Iteration must have been initialized
2338 * with cgroup_taskset_first().
2340 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2341 struct cgroup_subsys_state
**dst_cssp
)
2343 struct css_set
*cset
= tset
->cur_cset
;
2344 struct task_struct
*task
= tset
->cur_task
;
2346 while (&cset
->mg_node
!= tset
->csets
) {
2348 task
= list_first_entry(&cset
->mg_tasks
,
2349 struct task_struct
, cg_list
);
2351 task
= list_next_entry(task
, cg_list
);
2353 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2354 tset
->cur_cset
= cset
;
2355 tset
->cur_task
= task
;
2358 * This function may be called both before and
2359 * after cgroup_taskset_migrate(). The two cases
2360 * can be distinguished by looking at whether @cset
2361 * has its ->mg_dst_cset set.
2363 if (cset
->mg_dst_cset
)
2364 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2366 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2371 cset
= list_next_entry(cset
, mg_node
);
2379 * cgroup_taskset_migrate - migrate a taskset
2380 * @mgctx: migration context
2382 * Migrate tasks in @mgctx as setup by migration preparation functions.
2383 * This function fails iff one of the ->can_attach callbacks fails and
2384 * guarantees that either all or none of the tasks in @mgctx are migrated.
2385 * @mgctx is consumed regardless of success.
2387 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2389 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2390 struct cgroup_subsys
*ss
;
2391 struct task_struct
*task
, *tmp_task
;
2392 struct css_set
*cset
, *tmp_cset
;
2393 int ssid
, failed_ssid
, ret
;
2395 /* check that we can legitimately attach to the cgroup */
2396 if (tset
->nr_tasks
) {
2397 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2398 if (ss
->can_attach
) {
2400 ret
= ss
->can_attach(tset
);
2403 goto out_cancel_attach
;
2406 } while_each_subsys_mask();
2410 * Now that we're guaranteed success, proceed to move all tasks to
2411 * the new cgroup. There are no failure cases after here, so this
2412 * is the commit point.
2414 spin_lock_irq(&css_set_lock
);
2415 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2416 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2417 struct css_set
*from_cset
= task_css_set(task
);
2418 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2420 get_css_set(to_cset
);
2421 to_cset
->nr_tasks
++;
2422 css_set_move_task(task
, from_cset
, to_cset
, true);
2423 from_cset
->nr_tasks
--;
2425 * If the source or destination cgroup is frozen,
2426 * the task might require to change its state.
2428 cgroup_freezer_migrate_task(task
, from_cset
->dfl_cgrp
,
2430 put_css_set_locked(from_cset
);
2434 spin_unlock_irq(&css_set_lock
);
2437 * Migration is committed, all target tasks are now on dst_csets.
2438 * Nothing is sensitive to fork() after this point. Notify
2439 * controllers that migration is complete.
2441 tset
->csets
= &tset
->dst_csets
;
2443 if (tset
->nr_tasks
) {
2444 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2449 } while_each_subsys_mask();
2453 goto out_release_tset
;
2456 if (tset
->nr_tasks
) {
2457 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2458 if (ssid
== failed_ssid
)
2460 if (ss
->cancel_attach
) {
2462 ss
->cancel_attach(tset
);
2464 } while_each_subsys_mask();
2467 spin_lock_irq(&css_set_lock
);
2468 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2469 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2470 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2471 list_del_init(&cset
->mg_node
);
2473 spin_unlock_irq(&css_set_lock
);
2476 * Re-initialize the cgroup_taskset structure in case it is reused
2477 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2481 tset
->csets
= &tset
->src_csets
;
2486 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2487 * @dst_cgrp: destination cgroup to test
2489 * On the default hierarchy, except for the mixable, (possible) thread root
2490 * and threaded cgroups, subtree_control must be zero for migration
2491 * destination cgroups with tasks so that child cgroups don't compete
2494 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2496 /* v1 doesn't have any restriction */
2497 if (!cgroup_on_dfl(dst_cgrp
))
2500 /* verify @dst_cgrp can host resources */
2501 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2504 /* mixables don't care */
2505 if (cgroup_is_mixable(dst_cgrp
))
2509 * If @dst_cgrp is already or can become a thread root or is
2510 * threaded, it doesn't matter.
2512 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2515 /* apply no-internal-process constraint */
2516 if (dst_cgrp
->subtree_control
)
2523 * cgroup_migrate_finish - cleanup after attach
2524 * @mgctx: migration context
2526 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2527 * those functions for details.
2529 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2531 LIST_HEAD(preloaded
);
2532 struct css_set
*cset
, *tmp_cset
;
2534 lockdep_assert_held(&cgroup_mutex
);
2536 spin_lock_irq(&css_set_lock
);
2538 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2539 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2541 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2542 cset
->mg_src_cgrp
= NULL
;
2543 cset
->mg_dst_cgrp
= NULL
;
2544 cset
->mg_dst_cset
= NULL
;
2545 list_del_init(&cset
->mg_preload_node
);
2546 put_css_set_locked(cset
);
2549 spin_unlock_irq(&css_set_lock
);
2553 * cgroup_migrate_add_src - add a migration source css_set
2554 * @src_cset: the source css_set to add
2555 * @dst_cgrp: the destination cgroup
2556 * @mgctx: migration context
2558 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2559 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2560 * up by cgroup_migrate_finish().
2562 * This function may be called without holding cgroup_threadgroup_rwsem
2563 * even if the target is a process. Threads may be created and destroyed
2564 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2565 * into play and the preloaded css_sets are guaranteed to cover all
2568 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2569 struct cgroup
*dst_cgrp
,
2570 struct cgroup_mgctx
*mgctx
)
2572 struct cgroup
*src_cgrp
;
2574 lockdep_assert_held(&cgroup_mutex
);
2575 lockdep_assert_held(&css_set_lock
);
2578 * If ->dead, @src_set is associated with one or more dead cgroups
2579 * and doesn't contain any migratable tasks. Ignore it early so
2580 * that the rest of migration path doesn't get confused by it.
2585 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2587 if (!list_empty(&src_cset
->mg_preload_node
))
2590 WARN_ON(src_cset
->mg_src_cgrp
);
2591 WARN_ON(src_cset
->mg_dst_cgrp
);
2592 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2593 WARN_ON(!list_empty(&src_cset
->mg_node
));
2595 src_cset
->mg_src_cgrp
= src_cgrp
;
2596 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2597 get_css_set(src_cset
);
2598 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2602 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2603 * @mgctx: migration context
2605 * Tasks are about to be moved and all the source css_sets have been
2606 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2607 * pins all destination css_sets, links each to its source, and append them
2608 * to @mgctx->preloaded_dst_csets.
2610 * This function must be called after cgroup_migrate_add_src() has been
2611 * called on each migration source css_set. After migration is performed
2612 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2615 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2617 struct css_set
*src_cset
, *tmp_cset
;
2619 lockdep_assert_held(&cgroup_mutex
);
2621 /* look up the dst cset for each src cset and link it to src */
2622 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2624 struct css_set
*dst_cset
;
2625 struct cgroup_subsys
*ss
;
2628 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2632 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2635 * If src cset equals dst, it's noop. Drop the src.
2636 * cgroup_migrate() will skip the cset too. Note that we
2637 * can't handle src == dst as some nodes are used by both.
2639 if (src_cset
== dst_cset
) {
2640 src_cset
->mg_src_cgrp
= NULL
;
2641 src_cset
->mg_dst_cgrp
= NULL
;
2642 list_del_init(&src_cset
->mg_preload_node
);
2643 put_css_set(src_cset
);
2644 put_css_set(dst_cset
);
2648 src_cset
->mg_dst_cset
= dst_cset
;
2650 if (list_empty(&dst_cset
->mg_preload_node
))
2651 list_add_tail(&dst_cset
->mg_preload_node
,
2652 &mgctx
->preloaded_dst_csets
);
2654 put_css_set(dst_cset
);
2656 for_each_subsys(ss
, ssid
)
2657 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2658 mgctx
->ss_mask
|= 1 << ssid
;
2665 * cgroup_migrate - migrate a process or task to a cgroup
2666 * @leader: the leader of the process or the task to migrate
2667 * @threadgroup: whether @leader points to the whole process or a single task
2668 * @mgctx: migration context
2670 * Migrate a process or task denoted by @leader. If migrating a process,
2671 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2672 * responsible for invoking cgroup_migrate_add_src() and
2673 * cgroup_migrate_prepare_dst() on the targets before invoking this
2674 * function and following up with cgroup_migrate_finish().
2676 * As long as a controller's ->can_attach() doesn't fail, this function is
2677 * guaranteed to succeed. This means that, excluding ->can_attach()
2678 * failure, when migrating multiple targets, the success or failure can be
2679 * decided for all targets by invoking group_migrate_prepare_dst() before
2680 * actually starting migrating.
2682 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2683 struct cgroup_mgctx
*mgctx
)
2685 struct task_struct
*task
;
2688 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2689 * already PF_EXITING could be freed from underneath us unless we
2690 * take an rcu_read_lock.
2692 spin_lock_irq(&css_set_lock
);
2696 cgroup_migrate_add_task(task
, mgctx
);
2699 } while_each_thread(leader
, task
);
2701 spin_unlock_irq(&css_set_lock
);
2703 return cgroup_migrate_execute(mgctx
);
2707 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2708 * @dst_cgrp: the cgroup to attach to
2709 * @leader: the task or the leader of the threadgroup to be attached
2710 * @threadgroup: attach the whole threadgroup?
2712 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2714 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2717 DEFINE_CGROUP_MGCTX(mgctx
);
2718 struct task_struct
*task
;
2721 /* look up all src csets */
2722 spin_lock_irq(&css_set_lock
);
2726 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2729 } while_each_thread(leader
, task
);
2731 spin_unlock_irq(&css_set_lock
);
2733 /* prepare dst csets and commit */
2734 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2736 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2738 cgroup_migrate_finish(&mgctx
);
2741 TRACE_CGROUP_PATH(attach_task
, dst_cgrp
, leader
, threadgroup
);
2746 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
,
2748 __acquires(&cgroup_threadgroup_rwsem
)
2750 struct task_struct
*tsk
;
2753 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2754 return ERR_PTR(-EINVAL
);
2757 * If we migrate a single thread, we don't care about threadgroup
2758 * stability. If the thread is `current`, it won't exit(2) under our
2759 * hands or change PID through exec(2). We exclude
2760 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2761 * callers by cgroup_mutex.
2762 * Therefore, we can skip the global lock.
2764 lockdep_assert_held(&cgroup_mutex
);
2765 if (pid
|| threadgroup
) {
2766 percpu_down_write(&cgroup_threadgroup_rwsem
);
2774 tsk
= find_task_by_vpid(pid
);
2776 tsk
= ERR_PTR(-ESRCH
);
2777 goto out_unlock_threadgroup
;
2784 tsk
= tsk
->group_leader
;
2787 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2788 * If userland migrates such a kthread to a non-root cgroup, it can
2789 * become trapped in a cpuset, or RT kthread may be born in a
2790 * cgroup with no rt_runtime allocated. Just say no.
2792 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2793 tsk
= ERR_PTR(-EINVAL
);
2794 goto out_unlock_threadgroup
;
2797 get_task_struct(tsk
);
2798 goto out_unlock_rcu
;
2800 out_unlock_threadgroup
:
2802 percpu_up_write(&cgroup_threadgroup_rwsem
);
2810 void cgroup_procs_write_finish(struct task_struct
*task
, bool locked
)
2811 __releases(&cgroup_threadgroup_rwsem
)
2813 struct cgroup_subsys
*ss
;
2816 /* release reference from cgroup_procs_write_start() */
2817 put_task_struct(task
);
2820 percpu_up_write(&cgroup_threadgroup_rwsem
);
2821 for_each_subsys(ss
, ssid
)
2822 if (ss
->post_attach
)
2826 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2828 struct cgroup_subsys
*ss
;
2829 bool printed
= false;
2832 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2835 seq_puts(seq
, ss
->name
);
2837 } while_each_subsys_mask();
2839 seq_putc(seq
, '\n');
2842 /* show controllers which are enabled from the parent */
2843 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2845 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2847 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2851 /* show controllers which are enabled for a given cgroup's children */
2852 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2854 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2856 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2861 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2862 * @cgrp: root of the subtree to update csses for
2864 * @cgrp's control masks have changed and its subtree's css associations
2865 * need to be updated accordingly. This function looks up all css_sets
2866 * which are attached to the subtree, creates the matching updated css_sets
2867 * and migrates the tasks to the new ones.
2869 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2871 DEFINE_CGROUP_MGCTX(mgctx
);
2872 struct cgroup_subsys_state
*d_css
;
2873 struct cgroup
*dsct
;
2874 struct css_set
*src_cset
;
2877 lockdep_assert_held(&cgroup_mutex
);
2879 percpu_down_write(&cgroup_threadgroup_rwsem
);
2881 /* look up all csses currently attached to @cgrp's subtree */
2882 spin_lock_irq(&css_set_lock
);
2883 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2884 struct cgrp_cset_link
*link
;
2886 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2887 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2889 spin_unlock_irq(&css_set_lock
);
2891 /* NULL dst indicates self on default hierarchy */
2892 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2896 spin_lock_irq(&css_set_lock
);
2897 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2898 struct task_struct
*task
, *ntask
;
2900 /* all tasks in src_csets need to be migrated */
2901 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2902 cgroup_migrate_add_task(task
, &mgctx
);
2904 spin_unlock_irq(&css_set_lock
);
2906 ret
= cgroup_migrate_execute(&mgctx
);
2908 cgroup_migrate_finish(&mgctx
);
2909 percpu_up_write(&cgroup_threadgroup_rwsem
);
2914 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2915 * @cgrp: root of the target subtree
2917 * Because css offlining is asynchronous, userland may try to re-enable a
2918 * controller while the previous css is still around. This function grabs
2919 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2921 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
2922 __acquires(&cgroup_mutex
)
2924 struct cgroup
*dsct
;
2925 struct cgroup_subsys_state
*d_css
;
2926 struct cgroup_subsys
*ss
;
2930 mutex_lock(&cgroup_mutex
);
2932 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2933 for_each_subsys(ss
, ssid
) {
2934 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2937 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
2940 cgroup_get_live(dsct
);
2941 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
2942 TASK_UNINTERRUPTIBLE
);
2944 mutex_unlock(&cgroup_mutex
);
2946 finish_wait(&dsct
->offline_waitq
, &wait
);
2955 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2956 * @cgrp: root of the target subtree
2958 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2959 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2962 static void cgroup_save_control(struct cgroup
*cgrp
)
2964 struct cgroup
*dsct
;
2965 struct cgroup_subsys_state
*d_css
;
2967 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2968 dsct
->old_subtree_control
= dsct
->subtree_control
;
2969 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
2970 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
2975 * cgroup_propagate_control - refresh control masks of a subtree
2976 * @cgrp: root of the target subtree
2978 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2979 * ->subtree_control and propagate controller availability through the
2980 * subtree so that descendants don't have unavailable controllers enabled.
2982 static void cgroup_propagate_control(struct cgroup
*cgrp
)
2984 struct cgroup
*dsct
;
2985 struct cgroup_subsys_state
*d_css
;
2987 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2988 dsct
->subtree_control
&= cgroup_control(dsct
);
2989 dsct
->subtree_ss_mask
=
2990 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
2991 cgroup_ss_mask(dsct
));
2996 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
2997 * @cgrp: root of the target subtree
2999 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3000 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3003 static void cgroup_restore_control(struct cgroup
*cgrp
)
3005 struct cgroup
*dsct
;
3006 struct cgroup_subsys_state
*d_css
;
3008 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3009 dsct
->subtree_control
= dsct
->old_subtree_control
;
3010 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3011 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
3015 static bool css_visible(struct cgroup_subsys_state
*css
)
3017 struct cgroup_subsys
*ss
= css
->ss
;
3018 struct cgroup
*cgrp
= css
->cgroup
;
3020 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3022 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3024 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3028 * cgroup_apply_control_enable - enable or show csses according to control
3029 * @cgrp: root of the target subtree
3031 * Walk @cgrp's subtree and create new csses or make the existing ones
3032 * visible. A css is created invisible if it's being implicitly enabled
3033 * through dependency. An invisible css is made visible when the userland
3034 * explicitly enables it.
3036 * Returns 0 on success, -errno on failure. On failure, csses which have
3037 * been processed already aren't cleaned up. The caller is responsible for
3038 * cleaning up with cgroup_apply_control_disable().
3040 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3042 struct cgroup
*dsct
;
3043 struct cgroup_subsys_state
*d_css
;
3044 struct cgroup_subsys
*ss
;
3047 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3048 for_each_subsys(ss
, ssid
) {
3049 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3051 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3055 css
= css_create(dsct
, ss
);
3057 return PTR_ERR(css
);
3060 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3062 if (css_visible(css
)) {
3063 ret
= css_populate_dir(css
);
3074 * cgroup_apply_control_disable - kill or hide csses according to control
3075 * @cgrp: root of the target subtree
3077 * Walk @cgrp's subtree and kill and hide csses so that they match
3078 * cgroup_ss_mask() and cgroup_visible_mask().
3080 * A css is hidden when the userland requests it to be disabled while other
3081 * subsystems are still depending on it. The css must not actively control
3082 * resources and be in the vanilla state if it's made visible again later.
3083 * Controllers which may be depended upon should provide ->css_reset() for
3086 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3088 struct cgroup
*dsct
;
3089 struct cgroup_subsys_state
*d_css
;
3090 struct cgroup_subsys
*ss
;
3093 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3094 for_each_subsys(ss
, ssid
) {
3095 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3100 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3103 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3105 } else if (!css_visible(css
)) {
3115 * cgroup_apply_control - apply control mask updates to the subtree
3116 * @cgrp: root of the target subtree
3118 * subsystems can be enabled and disabled in a subtree using the following
3121 * 1. Call cgroup_save_control() to stash the current state.
3122 * 2. Update ->subtree_control masks in the subtree as desired.
3123 * 3. Call cgroup_apply_control() to apply the changes.
3124 * 4. Optionally perform other related operations.
3125 * 5. Call cgroup_finalize_control() to finish up.
3127 * This function implements step 3 and propagates the mask changes
3128 * throughout @cgrp's subtree, updates csses accordingly and perform
3129 * process migrations.
3131 static int cgroup_apply_control(struct cgroup
*cgrp
)
3135 cgroup_propagate_control(cgrp
);
3137 ret
= cgroup_apply_control_enable(cgrp
);
3142 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3143 * making the following cgroup_update_dfl_csses() properly update
3144 * css associations of all tasks in the subtree.
3146 ret
= cgroup_update_dfl_csses(cgrp
);
3154 * cgroup_finalize_control - finalize control mask update
3155 * @cgrp: root of the target subtree
3156 * @ret: the result of the update
3158 * Finalize control mask update. See cgroup_apply_control() for more info.
3160 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3163 cgroup_restore_control(cgrp
);
3164 cgroup_propagate_control(cgrp
);
3167 cgroup_apply_control_disable(cgrp
);
3170 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3172 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3174 /* if nothing is getting enabled, nothing to worry about */
3178 /* can @cgrp host any resources? */
3179 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3182 /* mixables don't care */
3183 if (cgroup_is_mixable(cgrp
))
3186 if (domain_enable
) {
3187 /* can't enable domain controllers inside a thread subtree */
3188 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3192 * Threaded controllers can handle internal competitions
3193 * and are always allowed inside a (prospective) thread
3196 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3201 * Controllers can't be enabled for a cgroup with tasks to avoid
3202 * child cgroups competing against tasks.
3204 if (cgroup_has_tasks(cgrp
))
3210 /* change the enabled child controllers for a cgroup in the default hierarchy */
3211 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3212 char *buf
, size_t nbytes
,
3215 u16 enable
= 0, disable
= 0;
3216 struct cgroup
*cgrp
, *child
;
3217 struct cgroup_subsys
*ss
;
3222 * Parse input - space separated list of subsystem names prefixed
3223 * with either + or -.
3225 buf
= strstrip(buf
);
3226 while ((tok
= strsep(&buf
, " "))) {
3229 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3230 if (!cgroup_ssid_enabled(ssid
) ||
3231 strcmp(tok
+ 1, ss
->name
))
3235 enable
|= 1 << ssid
;
3236 disable
&= ~(1 << ssid
);
3237 } else if (*tok
== '-') {
3238 disable
|= 1 << ssid
;
3239 enable
&= ~(1 << ssid
);
3244 } while_each_subsys_mask();
3245 if (ssid
== CGROUP_SUBSYS_COUNT
)
3249 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3253 for_each_subsys(ss
, ssid
) {
3254 if (enable
& (1 << ssid
)) {
3255 if (cgrp
->subtree_control
& (1 << ssid
)) {
3256 enable
&= ~(1 << ssid
);
3260 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3264 } else if (disable
& (1 << ssid
)) {
3265 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3266 disable
&= ~(1 << ssid
);
3270 /* a child has it enabled? */
3271 cgroup_for_each_live_child(child
, cgrp
) {
3272 if (child
->subtree_control
& (1 << ssid
)) {
3280 if (!enable
&& !disable
) {
3285 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3289 /* save and update control masks and prepare csses */
3290 cgroup_save_control(cgrp
);
3292 cgrp
->subtree_control
|= enable
;
3293 cgrp
->subtree_control
&= ~disable
;
3295 ret
= cgroup_apply_control(cgrp
);
3296 cgroup_finalize_control(cgrp
, ret
);
3300 kernfs_activate(cgrp
->kn
);
3302 cgroup_kn_unlock(of
->kn
);
3303 return ret
?: nbytes
;
3307 * cgroup_enable_threaded - make @cgrp threaded
3308 * @cgrp: the target cgroup
3310 * Called when "threaded" is written to the cgroup.type interface file and
3311 * tries to make @cgrp threaded and join the parent's resource domain.
3312 * This function is never called on the root cgroup as cgroup.type doesn't
3315 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3317 struct cgroup
*parent
= cgroup_parent(cgrp
);
3318 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3319 struct cgroup
*dsct
;
3320 struct cgroup_subsys_state
*d_css
;
3323 lockdep_assert_held(&cgroup_mutex
);
3325 /* noop if already threaded */
3326 if (cgroup_is_threaded(cgrp
))
3330 * If @cgroup is populated or has domain controllers enabled, it
3331 * can't be switched. While the below cgroup_can_be_thread_root()
3332 * test can catch the same conditions, that's only when @parent is
3333 * not mixable, so let's check it explicitly.
3335 if (cgroup_is_populated(cgrp
) ||
3336 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3339 /* we're joining the parent's domain, ensure its validity */
3340 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3341 !cgroup_can_be_thread_root(dom_cgrp
))
3345 * The following shouldn't cause actual migrations and should
3348 cgroup_save_control(cgrp
);
3350 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3351 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3352 dsct
->dom_cgrp
= dom_cgrp
;
3354 ret
= cgroup_apply_control(cgrp
);
3356 parent
->nr_threaded_children
++;
3358 cgroup_finalize_control(cgrp
, ret
);
3362 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3364 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3366 if (cgroup_is_threaded(cgrp
))
3367 seq_puts(seq
, "threaded\n");
3368 else if (!cgroup_is_valid_domain(cgrp
))
3369 seq_puts(seq
, "domain invalid\n");
3370 else if (cgroup_is_thread_root(cgrp
))
3371 seq_puts(seq
, "domain threaded\n");
3373 seq_puts(seq
, "domain\n");
3378 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3379 size_t nbytes
, loff_t off
)
3381 struct cgroup
*cgrp
;
3384 /* only switching to threaded mode is supported */
3385 if (strcmp(strstrip(buf
), "threaded"))
3388 /* drain dying csses before we re-apply (threaded) subtree control */
3389 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3393 /* threaded can only be enabled */
3394 ret
= cgroup_enable_threaded(cgrp
);
3396 cgroup_kn_unlock(of
->kn
);
3397 return ret
?: nbytes
;
3400 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3402 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3403 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3405 if (descendants
== INT_MAX
)
3406 seq_puts(seq
, "max\n");
3408 seq_printf(seq
, "%d\n", descendants
);
3413 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3414 char *buf
, size_t nbytes
, loff_t off
)
3416 struct cgroup
*cgrp
;
3420 buf
= strstrip(buf
);
3421 if (!strcmp(buf
, "max")) {
3422 descendants
= INT_MAX
;
3424 ret
= kstrtoint(buf
, 0, &descendants
);
3429 if (descendants
< 0)
3432 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3436 cgrp
->max_descendants
= descendants
;
3438 cgroup_kn_unlock(of
->kn
);
3443 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3445 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3446 int depth
= READ_ONCE(cgrp
->max_depth
);
3448 if (depth
== INT_MAX
)
3449 seq_puts(seq
, "max\n");
3451 seq_printf(seq
, "%d\n", depth
);
3456 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3457 char *buf
, size_t nbytes
, loff_t off
)
3459 struct cgroup
*cgrp
;
3463 buf
= strstrip(buf
);
3464 if (!strcmp(buf
, "max")) {
3467 ret
= kstrtoint(buf
, 0, &depth
);
3475 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3479 cgrp
->max_depth
= depth
;
3481 cgroup_kn_unlock(of
->kn
);
3486 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3488 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3490 seq_printf(seq
, "populated %d\n", cgroup_is_populated(cgrp
));
3491 seq_printf(seq
, "frozen %d\n", test_bit(CGRP_FROZEN
, &cgrp
->flags
));
3496 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3498 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3500 seq_printf(seq
, "nr_descendants %d\n",
3501 cgroup
->nr_descendants
);
3502 seq_printf(seq
, "nr_dying_descendants %d\n",
3503 cgroup
->nr_dying_descendants
);
3508 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3509 struct cgroup
*cgrp
, int ssid
)
3511 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3512 struct cgroup_subsys_state
*css
;
3515 if (!ss
->css_extra_stat_show
)
3518 css
= cgroup_tryget_css(cgrp
, ss
);
3522 ret
= ss
->css_extra_stat_show(seq
, css
);
3527 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3529 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3532 cgroup_base_stat_cputime_show(seq
);
3533 #ifdef CONFIG_CGROUP_SCHED
3534 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3540 static int cgroup_io_pressure_show(struct seq_file
*seq
, void *v
)
3542 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3543 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3545 return psi_show(seq
, psi
, PSI_IO
);
3547 static int cgroup_memory_pressure_show(struct seq_file
*seq
, void *v
)
3549 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3550 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3552 return psi_show(seq
, psi
, PSI_MEM
);
3554 static int cgroup_cpu_pressure_show(struct seq_file
*seq
, void *v
)
3556 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3557 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3559 return psi_show(seq
, psi
, PSI_CPU
);
3562 static ssize_t
cgroup_pressure_write(struct kernfs_open_file
*of
, char *buf
,
3563 size_t nbytes
, enum psi_res res
)
3565 struct psi_trigger
*new;
3566 struct cgroup
*cgrp
;
3567 struct psi_group
*psi
;
3569 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3574 cgroup_kn_unlock(of
->kn
);
3576 psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3577 new = psi_trigger_create(psi
, buf
, nbytes
, res
);
3580 return PTR_ERR(new);
3583 psi_trigger_replace(&of
->priv
, new);
3590 static ssize_t
cgroup_io_pressure_write(struct kernfs_open_file
*of
,
3591 char *buf
, size_t nbytes
,
3594 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_IO
);
3597 static ssize_t
cgroup_memory_pressure_write(struct kernfs_open_file
*of
,
3598 char *buf
, size_t nbytes
,
3601 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_MEM
);
3604 static ssize_t
cgroup_cpu_pressure_write(struct kernfs_open_file
*of
,
3605 char *buf
, size_t nbytes
,
3608 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_CPU
);
3611 static __poll_t
cgroup_pressure_poll(struct kernfs_open_file
*of
,
3614 return psi_trigger_poll(&of
->priv
, of
->file
, pt
);
3617 static void cgroup_pressure_release(struct kernfs_open_file
*of
)
3619 psi_trigger_replace(&of
->priv
, NULL
);
3621 #endif /* CONFIG_PSI */
3623 static int cgroup_freeze_show(struct seq_file
*seq
, void *v
)
3625 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3627 seq_printf(seq
, "%d\n", cgrp
->freezer
.freeze
);
3632 static ssize_t
cgroup_freeze_write(struct kernfs_open_file
*of
,
3633 char *buf
, size_t nbytes
, loff_t off
)
3635 struct cgroup
*cgrp
;
3639 ret
= kstrtoint(strstrip(buf
), 0, &freeze
);
3643 if (freeze
< 0 || freeze
> 1)
3646 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3650 cgroup_freeze(cgrp
, freeze
);
3652 cgroup_kn_unlock(of
->kn
);
3657 static int cgroup_file_open(struct kernfs_open_file
*of
)
3659 struct cftype
*cft
= of_cft(of
);
3662 return cft
->open(of
);
3666 static void cgroup_file_release(struct kernfs_open_file
*of
)
3668 struct cftype
*cft
= of_cft(of
);
3674 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3675 size_t nbytes
, loff_t off
)
3677 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3678 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3679 struct cftype
*cft
= of_cft(of
);
3680 struct cgroup_subsys_state
*css
;
3687 * If namespaces are delegation boundaries, disallow writes to
3688 * files in an non-init namespace root from inside the namespace
3689 * except for the files explicitly marked delegatable -
3690 * cgroup.procs and cgroup.subtree_control.
3692 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3693 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3694 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3698 return cft
->write(of
, buf
, nbytes
, off
);
3701 * kernfs guarantees that a file isn't deleted with operations in
3702 * flight, which means that the matching css is and stays alive and
3703 * doesn't need to be pinned. The RCU locking is not necessary
3704 * either. It's just for the convenience of using cgroup_css().
3707 css
= cgroup_css(cgrp
, cft
->ss
);
3710 if (cft
->write_u64
) {
3711 unsigned long long v
;
3712 ret
= kstrtoull(buf
, 0, &v
);
3714 ret
= cft
->write_u64(css
, cft
, v
);
3715 } else if (cft
->write_s64
) {
3717 ret
= kstrtoll(buf
, 0, &v
);
3719 ret
= cft
->write_s64(css
, cft
, v
);
3724 return ret
?: nbytes
;
3727 static __poll_t
cgroup_file_poll(struct kernfs_open_file
*of
, poll_table
*pt
)
3729 struct cftype
*cft
= of_cft(of
);
3732 return cft
->poll(of
, pt
);
3734 return kernfs_generic_poll(of
, pt
);
3737 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3739 return seq_cft(seq
)->seq_start(seq
, ppos
);
3742 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3744 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3747 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3749 if (seq_cft(seq
)->seq_stop
)
3750 seq_cft(seq
)->seq_stop(seq
, v
);
3753 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3755 struct cftype
*cft
= seq_cft(m
);
3756 struct cgroup_subsys_state
*css
= seq_css(m
);
3759 return cft
->seq_show(m
, arg
);
3762 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3763 else if (cft
->read_s64
)
3764 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3770 static struct kernfs_ops cgroup_kf_single_ops
= {
3771 .atomic_write_len
= PAGE_SIZE
,
3772 .open
= cgroup_file_open
,
3773 .release
= cgroup_file_release
,
3774 .write
= cgroup_file_write
,
3775 .poll
= cgroup_file_poll
,
3776 .seq_show
= cgroup_seqfile_show
,
3779 static struct kernfs_ops cgroup_kf_ops
= {
3780 .atomic_write_len
= PAGE_SIZE
,
3781 .open
= cgroup_file_open
,
3782 .release
= cgroup_file_release
,
3783 .write
= cgroup_file_write
,
3784 .poll
= cgroup_file_poll
,
3785 .seq_start
= cgroup_seqfile_start
,
3786 .seq_next
= cgroup_seqfile_next
,
3787 .seq_stop
= cgroup_seqfile_stop
,
3788 .seq_show
= cgroup_seqfile_show
,
3791 /* set uid and gid of cgroup dirs and files to that of the creator */
3792 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3794 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3795 .ia_uid
= current_fsuid(),
3796 .ia_gid
= current_fsgid(), };
3798 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3799 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3802 return kernfs_setattr(kn
, &iattr
);
3805 static void cgroup_file_notify_timer(struct timer_list
*timer
)
3807 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
3811 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3814 char name
[CGROUP_FILE_NAME_MAX
];
3815 struct kernfs_node
*kn
;
3816 struct lock_class_key
*key
= NULL
;
3819 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3820 key
= &cft
->lockdep_key
;
3822 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3823 cgroup_file_mode(cft
),
3824 GLOBAL_ROOT_UID
, GLOBAL_ROOT_GID
,
3825 0, cft
->kf_ops
, cft
,
3830 ret
= cgroup_kn_set_ugid(kn
);
3836 if (cft
->file_offset
) {
3837 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3839 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
3841 spin_lock_irq(&cgroup_file_kn_lock
);
3843 spin_unlock_irq(&cgroup_file_kn_lock
);
3850 * cgroup_addrm_files - add or remove files to a cgroup directory
3851 * @css: the target css
3852 * @cgrp: the target cgroup (usually css->cgroup)
3853 * @cfts: array of cftypes to be added
3854 * @is_add: whether to add or remove
3856 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3857 * For removals, this function never fails.
3859 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3860 struct cgroup
*cgrp
, struct cftype cfts
[],
3863 struct cftype
*cft
, *cft_end
= NULL
;
3866 lockdep_assert_held(&cgroup_mutex
);
3869 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3870 /* does cft->flags tell us to skip this file on @cgrp? */
3871 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3873 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3875 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3877 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3879 if ((cft
->flags
& CFTYPE_DEBUG
) && !cgroup_debug
)
3882 ret
= cgroup_add_file(css
, cgrp
, cft
);
3884 pr_warn("%s: failed to add %s, err=%d\n",
3885 __func__
, cft
->name
, ret
);
3891 cgroup_rm_file(cgrp
, cft
);
3897 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3899 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3900 struct cgroup
*root
= &ss
->root
->cgrp
;
3901 struct cgroup_subsys_state
*css
;
3904 lockdep_assert_held(&cgroup_mutex
);
3906 /* add/rm files for all cgroups created before */
3907 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3908 struct cgroup
*cgrp
= css
->cgroup
;
3910 if (!(css
->flags
& CSS_VISIBLE
))
3913 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3919 kernfs_activate(root
->kn
);
3923 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3927 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3928 /* free copy for custom atomic_write_len, see init_cftypes() */
3929 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3934 /* revert flags set by cgroup core while adding @cfts */
3935 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3939 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3943 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3944 struct kernfs_ops
*kf_ops
;
3946 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3949 kf_ops
= &cgroup_kf_ops
;
3951 kf_ops
= &cgroup_kf_single_ops
;
3954 * Ugh... if @cft wants a custom max_write_len, we need to
3955 * make a copy of kf_ops to set its atomic_write_len.
3957 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3958 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3960 cgroup_exit_cftypes(cfts
);
3963 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3966 cft
->kf_ops
= kf_ops
;
3973 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3975 lockdep_assert_held(&cgroup_mutex
);
3977 if (!cfts
|| !cfts
[0].ss
)
3980 list_del(&cfts
->node
);
3981 cgroup_apply_cftypes(cfts
, false);
3982 cgroup_exit_cftypes(cfts
);
3987 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3988 * @cfts: zero-length name terminated array of cftypes
3990 * Unregister @cfts. Files described by @cfts are removed from all
3991 * existing cgroups and all future cgroups won't have them either. This
3992 * function can be called anytime whether @cfts' subsys is attached or not.
3994 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3997 int cgroup_rm_cftypes(struct cftype
*cfts
)
4001 mutex_lock(&cgroup_mutex
);
4002 ret
= cgroup_rm_cftypes_locked(cfts
);
4003 mutex_unlock(&cgroup_mutex
);
4008 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4009 * @ss: target cgroup subsystem
4010 * @cfts: zero-length name terminated array of cftypes
4012 * Register @cfts to @ss. Files described by @cfts are created for all
4013 * existing cgroups to which @ss is attached and all future cgroups will
4014 * have them too. This function can be called anytime whether @ss is
4017 * Returns 0 on successful registration, -errno on failure. Note that this
4018 * function currently returns 0 as long as @cfts registration is successful
4019 * even if some file creation attempts on existing cgroups fail.
4021 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4025 if (!cgroup_ssid_enabled(ss
->id
))
4028 if (!cfts
|| cfts
[0].name
[0] == '\0')
4031 ret
= cgroup_init_cftypes(ss
, cfts
);
4035 mutex_lock(&cgroup_mutex
);
4037 list_add_tail(&cfts
->node
, &ss
->cfts
);
4038 ret
= cgroup_apply_cftypes(cfts
, true);
4040 cgroup_rm_cftypes_locked(cfts
);
4042 mutex_unlock(&cgroup_mutex
);
4047 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4048 * @ss: target cgroup subsystem
4049 * @cfts: zero-length name terminated array of cftypes
4051 * Similar to cgroup_add_cftypes() but the added files are only used for
4052 * the default hierarchy.
4054 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4058 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4059 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
4060 return cgroup_add_cftypes(ss
, cfts
);
4064 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4065 * @ss: target cgroup subsystem
4066 * @cfts: zero-length name terminated array of cftypes
4068 * Similar to cgroup_add_cftypes() but the added files are only used for
4069 * the legacy hierarchies.
4071 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4075 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4076 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
4077 return cgroup_add_cftypes(ss
, cfts
);
4081 * cgroup_file_notify - generate a file modified event for a cgroup_file
4082 * @cfile: target cgroup_file
4084 * @cfile must have been obtained by setting cftype->file_offset.
4086 void cgroup_file_notify(struct cgroup_file
*cfile
)
4088 unsigned long flags
;
4090 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
4092 unsigned long last
= cfile
->notified_at
;
4093 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
4095 if (time_in_range(jiffies
, last
, next
)) {
4096 timer_reduce(&cfile
->notify_timer
, next
);
4098 kernfs_notify(cfile
->kn
);
4099 cfile
->notified_at
= jiffies
;
4102 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
4106 * css_next_child - find the next child of a given css
4107 * @pos: the current position (%NULL to initiate traversal)
4108 * @parent: css whose children to walk
4110 * This function returns the next child of @parent and should be called
4111 * under either cgroup_mutex or RCU read lock. The only requirement is
4112 * that @parent and @pos are accessible. The next sibling is guaranteed to
4113 * be returned regardless of their states.
4115 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4116 * css which finished ->css_online() is guaranteed to be visible in the
4117 * future iterations and will stay visible until the last reference is put.
4118 * A css which hasn't finished ->css_online() or already finished
4119 * ->css_offline() may show up during traversal. It's each subsystem's
4120 * responsibility to synchronize against on/offlining.
4122 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
4123 struct cgroup_subsys_state
*parent
)
4125 struct cgroup_subsys_state
*next
;
4127 cgroup_assert_mutex_or_rcu_locked();
4130 * @pos could already have been unlinked from the sibling list.
4131 * Once a cgroup is removed, its ->sibling.next is no longer
4132 * updated when its next sibling changes. CSS_RELEASED is set when
4133 * @pos is taken off list, at which time its next pointer is valid,
4134 * and, as releases are serialized, the one pointed to by the next
4135 * pointer is guaranteed to not have started release yet. This
4136 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4137 * critical section, the one pointed to by its next pointer is
4138 * guaranteed to not have finished its RCU grace period even if we
4139 * have dropped rcu_read_lock() in-between iterations.
4141 * If @pos has CSS_RELEASED set, its next pointer can't be
4142 * dereferenced; however, as each css is given a monotonically
4143 * increasing unique serial number and always appended to the
4144 * sibling list, the next one can be found by walking the parent's
4145 * children until the first css with higher serial number than
4146 * @pos's. While this path can be slower, it happens iff iteration
4147 * races against release and the race window is very small.
4150 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
4151 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
4152 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
4154 list_for_each_entry_rcu(next
, &parent
->children
, sibling
,
4155 lockdep_is_held(&cgroup_mutex
))
4156 if (next
->serial_nr
> pos
->serial_nr
)
4161 * @next, if not pointing to the head, can be dereferenced and is
4164 if (&next
->sibling
!= &parent
->children
)
4170 * css_next_descendant_pre - find the next descendant for pre-order walk
4171 * @pos: the current position (%NULL to initiate traversal)
4172 * @root: css whose descendants to walk
4174 * To be used by css_for_each_descendant_pre(). Find the next descendant
4175 * to visit for pre-order traversal of @root's descendants. @root is
4176 * included in the iteration and the first node to be visited.
4178 * While this function requires cgroup_mutex or RCU read locking, it
4179 * doesn't require the whole traversal to be contained in a single critical
4180 * section. This function will return the correct next descendant as long
4181 * as both @pos and @root are accessible and @pos is a descendant of @root.
4183 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4184 * css which finished ->css_online() is guaranteed to be visible in the
4185 * future iterations and will stay visible until the last reference is put.
4186 * A css which hasn't finished ->css_online() or already finished
4187 * ->css_offline() may show up during traversal. It's each subsystem's
4188 * responsibility to synchronize against on/offlining.
4190 struct cgroup_subsys_state
*
4191 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4192 struct cgroup_subsys_state
*root
)
4194 struct cgroup_subsys_state
*next
;
4196 cgroup_assert_mutex_or_rcu_locked();
4198 /* if first iteration, visit @root */
4202 /* visit the first child if exists */
4203 next
= css_next_child(NULL
, pos
);
4207 /* no child, visit my or the closest ancestor's next sibling */
4208 while (pos
!= root
) {
4209 next
= css_next_child(pos
, pos
->parent
);
4217 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
4220 * css_rightmost_descendant - return the rightmost descendant of a css
4221 * @pos: css of interest
4223 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4224 * is returned. This can be used during pre-order traversal to skip
4227 * While this function requires cgroup_mutex or RCU read locking, it
4228 * doesn't require the whole traversal to be contained in a single critical
4229 * section. This function will return the correct rightmost descendant as
4230 * long as @pos is accessible.
4232 struct cgroup_subsys_state
*
4233 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4235 struct cgroup_subsys_state
*last
, *tmp
;
4237 cgroup_assert_mutex_or_rcu_locked();
4241 /* ->prev isn't RCU safe, walk ->next till the end */
4243 css_for_each_child(tmp
, last
)
4250 static struct cgroup_subsys_state
*
4251 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4253 struct cgroup_subsys_state
*last
;
4257 pos
= css_next_child(NULL
, pos
);
4264 * css_next_descendant_post - find the next descendant for post-order walk
4265 * @pos: the current position (%NULL to initiate traversal)
4266 * @root: css whose descendants to walk
4268 * To be used by css_for_each_descendant_post(). Find the next descendant
4269 * to visit for post-order traversal of @root's descendants. @root is
4270 * included in the iteration and the last node to be visited.
4272 * While this function requires cgroup_mutex or RCU read locking, it
4273 * doesn't require the whole traversal to be contained in a single critical
4274 * section. This function will return the correct next descendant as long
4275 * as both @pos and @cgroup are accessible and @pos is a descendant of
4278 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4279 * css which finished ->css_online() is guaranteed to be visible in the
4280 * future iterations and will stay visible until the last reference is put.
4281 * A css which hasn't finished ->css_online() or already finished
4282 * ->css_offline() may show up during traversal. It's each subsystem's
4283 * responsibility to synchronize against on/offlining.
4285 struct cgroup_subsys_state
*
4286 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4287 struct cgroup_subsys_state
*root
)
4289 struct cgroup_subsys_state
*next
;
4291 cgroup_assert_mutex_or_rcu_locked();
4293 /* if first iteration, visit leftmost descendant which may be @root */
4295 return css_leftmost_descendant(root
);
4297 /* if we visited @root, we're done */
4301 /* if there's an unvisited sibling, visit its leftmost descendant */
4302 next
= css_next_child(pos
, pos
->parent
);
4304 return css_leftmost_descendant(next
);
4306 /* no sibling left, visit parent */
4311 * css_has_online_children - does a css have online children
4312 * @css: the target css
4314 * Returns %true if @css has any online children; otherwise, %false. This
4315 * function can be called from any context but the caller is responsible
4316 * for synchronizing against on/offlining as necessary.
4318 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4320 struct cgroup_subsys_state
*child
;
4324 css_for_each_child(child
, css
) {
4325 if (child
->flags
& CSS_ONLINE
) {
4334 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4336 struct list_head
*l
;
4337 struct cgrp_cset_link
*link
;
4338 struct css_set
*cset
;
4340 lockdep_assert_held(&css_set_lock
);
4342 /* find the next threaded cset */
4343 if (it
->tcset_pos
) {
4344 l
= it
->tcset_pos
->next
;
4346 if (l
!= it
->tcset_head
) {
4348 return container_of(l
, struct css_set
,
4349 threaded_csets_node
);
4352 it
->tcset_pos
= NULL
;
4355 /* find the next cset */
4358 if (l
== it
->cset_head
) {
4359 it
->cset_pos
= NULL
;
4364 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4366 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4372 /* initialize threaded css_set walking */
4373 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4375 put_css_set_locked(it
->cur_dcset
);
4376 it
->cur_dcset
= cset
;
4379 it
->tcset_head
= &cset
->threaded_csets
;
4380 it
->tcset_pos
= &cset
->threaded_csets
;
4387 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4388 * @it: the iterator to advance
4390 * Advance @it to the next css_set to walk.
4392 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4394 struct css_set
*cset
;
4396 lockdep_assert_held(&css_set_lock
);
4398 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4399 while ((cset
= css_task_iter_next_css_set(it
))) {
4400 if (!list_empty(&cset
->tasks
)) {
4401 it
->cur_tasks_head
= &cset
->tasks
;
4403 } else if (!list_empty(&cset
->mg_tasks
)) {
4404 it
->cur_tasks_head
= &cset
->mg_tasks
;
4406 } else if (!list_empty(&cset
->dying_tasks
)) {
4407 it
->cur_tasks_head
= &cset
->dying_tasks
;
4412 it
->task_pos
= NULL
;
4415 it
->task_pos
= it
->cur_tasks_head
->next
;
4418 * We don't keep css_sets locked across iteration steps and thus
4419 * need to take steps to ensure that iteration can be resumed after
4420 * the lock is re-acquired. Iteration is performed at two levels -
4421 * css_sets and tasks in them.
4423 * Once created, a css_set never leaves its cgroup lists, so a
4424 * pinned css_set is guaranteed to stay put and we can resume
4425 * iteration afterwards.
4427 * Tasks may leave @cset across iteration steps. This is resolved
4428 * by registering each iterator with the css_set currently being
4429 * walked and making css_set_move_task() advance iterators whose
4430 * next task is leaving.
4433 list_del(&it
->iters_node
);
4434 put_css_set_locked(it
->cur_cset
);
4437 it
->cur_cset
= cset
;
4438 list_add(&it
->iters_node
, &cset
->task_iters
);
4441 static void css_task_iter_skip(struct css_task_iter
*it
,
4442 struct task_struct
*task
)
4444 lockdep_assert_held(&css_set_lock
);
4446 if (it
->task_pos
== &task
->cg_list
) {
4447 it
->task_pos
= it
->task_pos
->next
;
4448 it
->flags
|= CSS_TASK_ITER_SKIPPED
;
4452 static void css_task_iter_advance(struct css_task_iter
*it
)
4454 struct task_struct
*task
;
4456 lockdep_assert_held(&css_set_lock
);
4460 * Advance iterator to find next entry. We go through cset
4461 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4464 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4465 it
->flags
&= ~CSS_TASK_ITER_SKIPPED
;
4467 it
->task_pos
= it
->task_pos
->next
;
4469 if (it
->task_pos
== &it
->cur_cset
->tasks
) {
4470 it
->cur_tasks_head
= &it
->cur_cset
->mg_tasks
;
4471 it
->task_pos
= it
->cur_tasks_head
->next
;
4473 if (it
->task_pos
== &it
->cur_cset
->mg_tasks
) {
4474 it
->cur_tasks_head
= &it
->cur_cset
->dying_tasks
;
4475 it
->task_pos
= it
->cur_tasks_head
->next
;
4477 if (it
->task_pos
== &it
->cur_cset
->dying_tasks
)
4478 css_task_iter_advance_css_set(it
);
4480 /* called from start, proceed to the first cset */
4481 css_task_iter_advance_css_set(it
);
4487 task
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
4489 if (it
->flags
& CSS_TASK_ITER_PROCS
) {
4490 /* if PROCS, skip over tasks which aren't group leaders */
4491 if (!thread_group_leader(task
))
4494 /* and dying leaders w/o live member threads */
4495 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
&&
4496 !atomic_read(&task
->signal
->live
))
4499 /* skip all dying ones */
4500 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
)
4506 * css_task_iter_start - initiate task iteration
4507 * @css: the css to walk tasks of
4508 * @flags: CSS_TASK_ITER_* flags
4509 * @it: the task iterator to use
4511 * Initiate iteration through the tasks of @css. The caller can call
4512 * css_task_iter_next() to walk through the tasks until the function
4513 * returns NULL. On completion of iteration, css_task_iter_end() must be
4516 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4517 struct css_task_iter
*it
)
4519 memset(it
, 0, sizeof(*it
));
4521 spin_lock_irq(&css_set_lock
);
4527 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4529 it
->cset_pos
= &css
->cgroup
->cset_links
;
4531 it
->cset_head
= it
->cset_pos
;
4533 css_task_iter_advance(it
);
4535 spin_unlock_irq(&css_set_lock
);
4539 * css_task_iter_next - return the next task for the iterator
4540 * @it: the task iterator being iterated
4542 * The "next" function for task iteration. @it should have been
4543 * initialized via css_task_iter_start(). Returns NULL when the iteration
4546 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4549 put_task_struct(it
->cur_task
);
4550 it
->cur_task
= NULL
;
4553 spin_lock_irq(&css_set_lock
);
4555 /* @it may be half-advanced by skips, finish advancing */
4556 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4557 css_task_iter_advance(it
);
4560 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4562 get_task_struct(it
->cur_task
);
4563 css_task_iter_advance(it
);
4566 spin_unlock_irq(&css_set_lock
);
4568 return it
->cur_task
;
4572 * css_task_iter_end - finish task iteration
4573 * @it: the task iterator to finish
4575 * Finish task iteration started by css_task_iter_start().
4577 void css_task_iter_end(struct css_task_iter
*it
)
4580 spin_lock_irq(&css_set_lock
);
4581 list_del(&it
->iters_node
);
4582 put_css_set_locked(it
->cur_cset
);
4583 spin_unlock_irq(&css_set_lock
);
4587 put_css_set(it
->cur_dcset
);
4590 put_task_struct(it
->cur_task
);
4593 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4596 css_task_iter_end(of
->priv
);
4601 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4603 struct kernfs_open_file
*of
= s
->private;
4604 struct css_task_iter
*it
= of
->priv
;
4609 return css_task_iter_next(it
);
4612 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4613 unsigned int iter_flags
)
4615 struct kernfs_open_file
*of
= s
->private;
4616 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4617 struct css_task_iter
*it
= of
->priv
;
4620 * When a seq_file is seeked, it's always traversed sequentially
4621 * from position 0, so we can simply keep iterating on !0 *pos.
4624 if (WARN_ON_ONCE((*pos
)))
4625 return ERR_PTR(-EINVAL
);
4627 it
= kzalloc(sizeof(*it
), GFP_KERNEL
);
4629 return ERR_PTR(-ENOMEM
);
4631 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4632 } else if (!(*pos
)) {
4633 css_task_iter_end(it
);
4634 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4636 return it
->cur_task
;
4638 return cgroup_procs_next(s
, NULL
, NULL
);
4641 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4643 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4646 * All processes of a threaded subtree belong to the domain cgroup
4647 * of the subtree. Only threads can be distributed across the
4648 * subtree. Reject reads on cgroup.procs in the subtree proper.
4649 * They're always empty anyway.
4651 if (cgroup_is_threaded(cgrp
))
4652 return ERR_PTR(-EOPNOTSUPP
);
4654 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4655 CSS_TASK_ITER_THREADED
);
4658 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4660 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4664 static int cgroup_may_write(const struct cgroup
*cgrp
, struct super_block
*sb
)
4667 struct inode
*inode
;
4669 lockdep_assert_held(&cgroup_mutex
);
4671 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
4675 ret
= inode_permission(inode
, MAY_WRITE
);
4680 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4681 struct cgroup
*dst_cgrp
,
4682 struct super_block
*sb
)
4684 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4685 struct cgroup
*com_cgrp
= src_cgrp
;
4688 lockdep_assert_held(&cgroup_mutex
);
4690 /* find the common ancestor */
4691 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4692 com_cgrp
= cgroup_parent(com_cgrp
);
4694 /* %current should be authorized to migrate to the common ancestor */
4695 ret
= cgroup_may_write(com_cgrp
, sb
);
4700 * If namespaces are delegation boundaries, %current must be able
4701 * to see both source and destination cgroups from its namespace.
4703 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4704 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4705 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4711 static int cgroup_attach_permissions(struct cgroup
*src_cgrp
,
4712 struct cgroup
*dst_cgrp
,
4713 struct super_block
*sb
, bool threadgroup
)
4717 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
, sb
);
4721 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
4725 if (!threadgroup
&& (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
))
4731 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4732 char *buf
, size_t nbytes
, loff_t off
)
4734 struct cgroup
*src_cgrp
, *dst_cgrp
;
4735 struct task_struct
*task
;
4739 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4743 task
= cgroup_procs_write_start(buf
, true, &locked
);
4744 ret
= PTR_ERR_OR_ZERO(task
);
4748 /* find the source cgroup */
4749 spin_lock_irq(&css_set_lock
);
4750 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4751 spin_unlock_irq(&css_set_lock
);
4753 ret
= cgroup_attach_permissions(src_cgrp
, dst_cgrp
,
4754 of
->file
->f_path
.dentry
->d_sb
, true);
4758 ret
= cgroup_attach_task(dst_cgrp
, task
, true);
4761 cgroup_procs_write_finish(task
, locked
);
4763 cgroup_kn_unlock(of
->kn
);
4765 return ret
?: nbytes
;
4768 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4770 return __cgroup_procs_start(s
, pos
, 0);
4773 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4774 char *buf
, size_t nbytes
, loff_t off
)
4776 struct cgroup
*src_cgrp
, *dst_cgrp
;
4777 struct task_struct
*task
;
4781 buf
= strstrip(buf
);
4783 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4787 task
= cgroup_procs_write_start(buf
, false, &locked
);
4788 ret
= PTR_ERR_OR_ZERO(task
);
4792 /* find the source cgroup */
4793 spin_lock_irq(&css_set_lock
);
4794 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4795 spin_unlock_irq(&css_set_lock
);
4797 /* thread migrations follow the cgroup.procs delegation rule */
4798 ret
= cgroup_attach_permissions(src_cgrp
, dst_cgrp
,
4799 of
->file
->f_path
.dentry
->d_sb
, false);
4803 ret
= cgroup_attach_task(dst_cgrp
, task
, false);
4806 cgroup_procs_write_finish(task
, locked
);
4808 cgroup_kn_unlock(of
->kn
);
4810 return ret
?: nbytes
;
4813 /* cgroup core interface files for the default hierarchy */
4814 static struct cftype cgroup_base_files
[] = {
4816 .name
= "cgroup.type",
4817 .flags
= CFTYPE_NOT_ON_ROOT
,
4818 .seq_show
= cgroup_type_show
,
4819 .write
= cgroup_type_write
,
4822 .name
= "cgroup.procs",
4823 .flags
= CFTYPE_NS_DELEGATABLE
,
4824 .file_offset
= offsetof(struct cgroup
, procs_file
),
4825 .release
= cgroup_procs_release
,
4826 .seq_start
= cgroup_procs_start
,
4827 .seq_next
= cgroup_procs_next
,
4828 .seq_show
= cgroup_procs_show
,
4829 .write
= cgroup_procs_write
,
4832 .name
= "cgroup.threads",
4833 .flags
= CFTYPE_NS_DELEGATABLE
,
4834 .release
= cgroup_procs_release
,
4835 .seq_start
= cgroup_threads_start
,
4836 .seq_next
= cgroup_procs_next
,
4837 .seq_show
= cgroup_procs_show
,
4838 .write
= cgroup_threads_write
,
4841 .name
= "cgroup.controllers",
4842 .seq_show
= cgroup_controllers_show
,
4845 .name
= "cgroup.subtree_control",
4846 .flags
= CFTYPE_NS_DELEGATABLE
,
4847 .seq_show
= cgroup_subtree_control_show
,
4848 .write
= cgroup_subtree_control_write
,
4851 .name
= "cgroup.events",
4852 .flags
= CFTYPE_NOT_ON_ROOT
,
4853 .file_offset
= offsetof(struct cgroup
, events_file
),
4854 .seq_show
= cgroup_events_show
,
4857 .name
= "cgroup.max.descendants",
4858 .seq_show
= cgroup_max_descendants_show
,
4859 .write
= cgroup_max_descendants_write
,
4862 .name
= "cgroup.max.depth",
4863 .seq_show
= cgroup_max_depth_show
,
4864 .write
= cgroup_max_depth_write
,
4867 .name
= "cgroup.stat",
4868 .seq_show
= cgroup_stat_show
,
4871 .name
= "cgroup.freeze",
4872 .flags
= CFTYPE_NOT_ON_ROOT
,
4873 .seq_show
= cgroup_freeze_show
,
4874 .write
= cgroup_freeze_write
,
4878 .seq_show
= cpu_stat_show
,
4882 .name
= "io.pressure",
4883 .seq_show
= cgroup_io_pressure_show
,
4884 .write
= cgroup_io_pressure_write
,
4885 .poll
= cgroup_pressure_poll
,
4886 .release
= cgroup_pressure_release
,
4889 .name
= "memory.pressure",
4890 .seq_show
= cgroup_memory_pressure_show
,
4891 .write
= cgroup_memory_pressure_write
,
4892 .poll
= cgroup_pressure_poll
,
4893 .release
= cgroup_pressure_release
,
4896 .name
= "cpu.pressure",
4897 .seq_show
= cgroup_cpu_pressure_show
,
4898 .write
= cgroup_cpu_pressure_write
,
4899 .poll
= cgroup_pressure_poll
,
4900 .release
= cgroup_pressure_release
,
4902 #endif /* CONFIG_PSI */
4907 * css destruction is four-stage process.
4909 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4910 * Implemented in kill_css().
4912 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4913 * and thus css_tryget_online() is guaranteed to fail, the css can be
4914 * offlined by invoking offline_css(). After offlining, the base ref is
4915 * put. Implemented in css_killed_work_fn().
4917 * 3. When the percpu_ref reaches zero, the only possible remaining
4918 * accessors are inside RCU read sections. css_release() schedules the
4921 * 4. After the grace period, the css can be freed. Implemented in
4922 * css_free_work_fn().
4924 * It is actually hairier because both step 2 and 4 require process context
4925 * and thus involve punting to css->destroy_work adding two additional
4926 * steps to the already complex sequence.
4928 static void css_free_rwork_fn(struct work_struct
*work
)
4930 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
4931 struct cgroup_subsys_state
, destroy_rwork
);
4932 struct cgroup_subsys
*ss
= css
->ss
;
4933 struct cgroup
*cgrp
= css
->cgroup
;
4935 percpu_ref_exit(&css
->refcnt
);
4939 struct cgroup_subsys_state
*parent
= css
->parent
;
4943 cgroup_idr_remove(&ss
->css_idr
, id
);
4949 /* cgroup free path */
4950 atomic_dec(&cgrp
->root
->nr_cgrps
);
4951 cgroup1_pidlist_destroy_all(cgrp
);
4952 cancel_work_sync(&cgrp
->release_agent_work
);
4954 if (cgroup_parent(cgrp
)) {
4956 * We get a ref to the parent, and put the ref when
4957 * this cgroup is being freed, so it's guaranteed
4958 * that the parent won't be destroyed before its
4961 cgroup_put(cgroup_parent(cgrp
));
4962 kernfs_put(cgrp
->kn
);
4963 psi_cgroup_free(cgrp
);
4964 if (cgroup_on_dfl(cgrp
))
4965 cgroup_rstat_exit(cgrp
);
4969 * This is root cgroup's refcnt reaching zero,
4970 * which indicates that the root should be
4973 cgroup_destroy_root(cgrp
->root
);
4978 static void css_release_work_fn(struct work_struct
*work
)
4980 struct cgroup_subsys_state
*css
=
4981 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4982 struct cgroup_subsys
*ss
= css
->ss
;
4983 struct cgroup
*cgrp
= css
->cgroup
;
4985 mutex_lock(&cgroup_mutex
);
4987 css
->flags
|= CSS_RELEASED
;
4988 list_del_rcu(&css
->sibling
);
4991 /* css release path */
4992 if (!list_empty(&css
->rstat_css_node
)) {
4993 cgroup_rstat_flush(cgrp
);
4994 list_del_rcu(&css
->rstat_css_node
);
4997 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4998 if (ss
->css_released
)
4999 ss
->css_released(css
);
5001 struct cgroup
*tcgrp
;
5003 /* cgroup release path */
5004 TRACE_CGROUP_PATH(release
, cgrp
);
5006 if (cgroup_on_dfl(cgrp
))
5007 cgroup_rstat_flush(cgrp
);
5009 spin_lock_irq(&css_set_lock
);
5010 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
5011 tcgrp
= cgroup_parent(tcgrp
))
5012 tcgrp
->nr_dying_descendants
--;
5013 spin_unlock_irq(&css_set_lock
);
5016 * There are two control paths which try to determine
5017 * cgroup from dentry without going through kernfs -
5018 * cgroupstats_build() and css_tryget_online_from_dir().
5019 * Those are supported by RCU protecting clearing of
5020 * cgrp->kn->priv backpointer.
5023 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5027 mutex_unlock(&cgroup_mutex
);
5029 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5030 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5033 static void css_release(struct percpu_ref
*ref
)
5035 struct cgroup_subsys_state
*css
=
5036 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5038 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5039 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5042 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5043 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5045 lockdep_assert_held(&cgroup_mutex
);
5047 cgroup_get_live(cgrp
);
5049 memset(css
, 0, sizeof(*css
));
5053 INIT_LIST_HEAD(&css
->sibling
);
5054 INIT_LIST_HEAD(&css
->children
);
5055 INIT_LIST_HEAD(&css
->rstat_css_node
);
5056 css
->serial_nr
= css_serial_nr_next
++;
5057 atomic_set(&css
->online_cnt
, 0);
5059 if (cgroup_parent(cgrp
)) {
5060 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5061 css_get(css
->parent
);
5064 if (cgroup_on_dfl(cgrp
) && ss
->css_rstat_flush
)
5065 list_add_rcu(&css
->rstat_css_node
, &cgrp
->rstat_css_list
);
5067 BUG_ON(cgroup_css(cgrp
, ss
));
5070 /* invoke ->css_online() on a new CSS and mark it online if successful */
5071 static int online_css(struct cgroup_subsys_state
*css
)
5073 struct cgroup_subsys
*ss
= css
->ss
;
5076 lockdep_assert_held(&cgroup_mutex
);
5079 ret
= ss
->css_online(css
);
5081 css
->flags
|= CSS_ONLINE
;
5082 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5084 atomic_inc(&css
->online_cnt
);
5086 atomic_inc(&css
->parent
->online_cnt
);
5091 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5092 static void offline_css(struct cgroup_subsys_state
*css
)
5094 struct cgroup_subsys
*ss
= css
->ss
;
5096 lockdep_assert_held(&cgroup_mutex
);
5098 if (!(css
->flags
& CSS_ONLINE
))
5101 if (ss
->css_offline
)
5102 ss
->css_offline(css
);
5104 css
->flags
&= ~CSS_ONLINE
;
5105 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5107 wake_up_all(&css
->cgroup
->offline_waitq
);
5111 * css_create - create a cgroup_subsys_state
5112 * @cgrp: the cgroup new css will be associated with
5113 * @ss: the subsys of new css
5115 * Create a new css associated with @cgrp - @ss pair. On success, the new
5116 * css is online and installed in @cgrp. This function doesn't create the
5117 * interface files. Returns 0 on success, -errno on failure.
5119 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5120 struct cgroup_subsys
*ss
)
5122 struct cgroup
*parent
= cgroup_parent(cgrp
);
5123 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5124 struct cgroup_subsys_state
*css
;
5127 lockdep_assert_held(&cgroup_mutex
);
5129 css
= ss
->css_alloc(parent_css
);
5131 css
= ERR_PTR(-ENOMEM
);
5135 init_and_link_css(css
, ss
, cgrp
);
5137 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5141 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5146 /* @css is ready to be brought online now, make it visible */
5147 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5148 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5150 err
= online_css(css
);
5157 list_del_rcu(&css
->sibling
);
5159 list_del_rcu(&css
->rstat_css_node
);
5160 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5161 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5162 return ERR_PTR(err
);
5166 * The returned cgroup is fully initialized including its control mask, but
5167 * it isn't associated with its kernfs_node and doesn't have the control
5170 static struct cgroup
*cgroup_create(struct cgroup
*parent
, const char *name
,
5173 struct cgroup_root
*root
= parent
->root
;
5174 struct cgroup
*cgrp
, *tcgrp
;
5175 struct kernfs_node
*kn
;
5176 int level
= parent
->level
+ 1;
5179 /* allocate the cgroup and its ID, 0 is reserved for the root */
5180 cgrp
= kzalloc(struct_size(cgrp
, ancestor_ids
, (level
+ 1)),
5183 return ERR_PTR(-ENOMEM
);
5185 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5189 if (cgroup_on_dfl(parent
)) {
5190 ret
= cgroup_rstat_init(cgrp
);
5192 goto out_cancel_ref
;
5195 /* create the directory */
5196 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5203 init_cgroup_housekeeping(cgrp
);
5205 cgrp
->self
.parent
= &parent
->self
;
5207 cgrp
->level
= level
;
5209 ret
= psi_cgroup_alloc(cgrp
);
5211 goto out_kernfs_remove
;
5213 ret
= cgroup_bpf_inherit(cgrp
);
5218 * New cgroup inherits effective freeze counter, and
5219 * if the parent has to be frozen, the child has too.
5221 cgrp
->freezer
.e_freeze
= parent
->freezer
.e_freeze
;
5222 if (cgrp
->freezer
.e_freeze
) {
5224 * Set the CGRP_FREEZE flag, so when a process will be
5225 * attached to the child cgroup, it will become frozen.
5226 * At this point the new cgroup is unpopulated, so we can
5227 * consider it frozen immediately.
5229 set_bit(CGRP_FREEZE
, &cgrp
->flags
);
5230 set_bit(CGRP_FROZEN
, &cgrp
->flags
);
5233 spin_lock_irq(&css_set_lock
);
5234 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5235 cgrp
->ancestor_ids
[tcgrp
->level
] = cgroup_id(tcgrp
);
5237 if (tcgrp
!= cgrp
) {
5238 tcgrp
->nr_descendants
++;
5241 * If the new cgroup is frozen, all ancestor cgroups
5242 * get a new frozen descendant, but their state can't
5243 * change because of this.
5245 if (cgrp
->freezer
.e_freeze
)
5246 tcgrp
->freezer
.nr_frozen_descendants
++;
5249 spin_unlock_irq(&css_set_lock
);
5251 if (notify_on_release(parent
))
5252 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5254 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5255 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5257 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5259 /* allocation complete, commit to creation */
5260 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5261 atomic_inc(&root
->nr_cgrps
);
5262 cgroup_get_live(parent
);
5265 * On the default hierarchy, a child doesn't automatically inherit
5266 * subtree_control from the parent. Each is configured manually.
5268 if (!cgroup_on_dfl(cgrp
))
5269 cgrp
->subtree_control
= cgroup_control(cgrp
);
5271 cgroup_propagate_control(cgrp
);
5276 psi_cgroup_free(cgrp
);
5278 kernfs_remove(cgrp
->kn
);
5280 if (cgroup_on_dfl(parent
))
5281 cgroup_rstat_exit(cgrp
);
5283 percpu_ref_exit(&cgrp
->self
.refcnt
);
5286 return ERR_PTR(ret
);
5289 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
5291 struct cgroup
*cgroup
;
5295 lockdep_assert_held(&cgroup_mutex
);
5297 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
5298 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
5301 if (level
> cgroup
->max_depth
)
5312 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
5314 struct cgroup
*parent
, *cgrp
;
5317 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5318 if (strchr(name
, '\n'))
5321 parent
= cgroup_kn_lock_live(parent_kn
, false);
5325 if (!cgroup_check_hierarchy_limits(parent
)) {
5330 cgrp
= cgroup_create(parent
, name
, mode
);
5332 ret
= PTR_ERR(cgrp
);
5337 * This extra ref will be put in cgroup_free_fn() and guarantees
5338 * that @cgrp->kn is always accessible.
5340 kernfs_get(cgrp
->kn
);
5342 ret
= cgroup_kn_set_ugid(cgrp
->kn
);
5346 ret
= css_populate_dir(&cgrp
->self
);
5350 ret
= cgroup_apply_control_enable(cgrp
);
5354 TRACE_CGROUP_PATH(mkdir
, cgrp
);
5356 /* let's create and online css's */
5357 kernfs_activate(cgrp
->kn
);
5363 cgroup_destroy_locked(cgrp
);
5365 cgroup_kn_unlock(parent_kn
);
5370 * This is called when the refcnt of a css is confirmed to be killed.
5371 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5372 * initate destruction and put the css ref from kill_css().
5374 static void css_killed_work_fn(struct work_struct
*work
)
5376 struct cgroup_subsys_state
*css
=
5377 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5379 mutex_lock(&cgroup_mutex
);
5384 /* @css can't go away while we're holding cgroup_mutex */
5386 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5388 mutex_unlock(&cgroup_mutex
);
5391 /* css kill confirmation processing requires process context, bounce */
5392 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5394 struct cgroup_subsys_state
*css
=
5395 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5397 if (atomic_dec_and_test(&css
->online_cnt
)) {
5398 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5399 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5404 * kill_css - destroy a css
5405 * @css: css to destroy
5407 * This function initiates destruction of @css by removing cgroup interface
5408 * files and putting its base reference. ->css_offline() will be invoked
5409 * asynchronously once css_tryget_online() is guaranteed to fail and when
5410 * the reference count reaches zero, @css will be released.
5412 static void kill_css(struct cgroup_subsys_state
*css
)
5414 lockdep_assert_held(&cgroup_mutex
);
5416 if (css
->flags
& CSS_DYING
)
5419 css
->flags
|= CSS_DYING
;
5422 * This must happen before css is disassociated with its cgroup.
5423 * See seq_css() for details.
5428 * Killing would put the base ref, but we need to keep it alive
5429 * until after ->css_offline().
5434 * cgroup core guarantees that, by the time ->css_offline() is
5435 * invoked, no new css reference will be given out via
5436 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5437 * proceed to offlining css's because percpu_ref_kill() doesn't
5438 * guarantee that the ref is seen as killed on all CPUs on return.
5440 * Use percpu_ref_kill_and_confirm() to get notifications as each
5441 * css is confirmed to be seen as killed on all CPUs.
5443 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5447 * cgroup_destroy_locked - the first stage of cgroup destruction
5448 * @cgrp: cgroup to be destroyed
5450 * css's make use of percpu refcnts whose killing latency shouldn't be
5451 * exposed to userland and are RCU protected. Also, cgroup core needs to
5452 * guarantee that css_tryget_online() won't succeed by the time
5453 * ->css_offline() is invoked. To satisfy all the requirements,
5454 * destruction is implemented in the following two steps.
5456 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5457 * userland visible parts and start killing the percpu refcnts of
5458 * css's. Set up so that the next stage will be kicked off once all
5459 * the percpu refcnts are confirmed to be killed.
5461 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5462 * rest of destruction. Once all cgroup references are gone, the
5463 * cgroup is RCU-freed.
5465 * This function implements s1. After this step, @cgrp is gone as far as
5466 * the userland is concerned and a new cgroup with the same name may be
5467 * created. As cgroup doesn't care about the names internally, this
5468 * doesn't cause any problem.
5470 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5471 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5473 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5474 struct cgroup_subsys_state
*css
;
5475 struct cgrp_cset_link
*link
;
5478 lockdep_assert_held(&cgroup_mutex
);
5481 * Only migration can raise populated from zero and we're already
5482 * holding cgroup_mutex.
5484 if (cgroup_is_populated(cgrp
))
5488 * Make sure there's no live children. We can't test emptiness of
5489 * ->self.children as dead children linger on it while being
5490 * drained; otherwise, "rmdir parent/child parent" may fail.
5492 if (css_has_online_children(&cgrp
->self
))
5496 * Mark @cgrp and the associated csets dead. The former prevents
5497 * further task migration and child creation by disabling
5498 * cgroup_lock_live_group(). The latter makes the csets ignored by
5499 * the migration path.
5501 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5503 spin_lock_irq(&css_set_lock
);
5504 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5505 link
->cset
->dead
= true;
5506 spin_unlock_irq(&css_set_lock
);
5508 /* initiate massacre of all css's */
5509 for_each_css(css
, ssid
, cgrp
)
5512 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5513 css_clear_dir(&cgrp
->self
);
5514 kernfs_remove(cgrp
->kn
);
5516 if (parent
&& cgroup_is_threaded(cgrp
))
5517 parent
->nr_threaded_children
--;
5519 spin_lock_irq(&css_set_lock
);
5520 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5521 tcgrp
->nr_descendants
--;
5522 tcgrp
->nr_dying_descendants
++;
5524 * If the dying cgroup is frozen, decrease frozen descendants
5525 * counters of ancestor cgroups.
5527 if (test_bit(CGRP_FROZEN
, &cgrp
->flags
))
5528 tcgrp
->freezer
.nr_frozen_descendants
--;
5530 spin_unlock_irq(&css_set_lock
);
5532 cgroup1_check_for_release(parent
);
5534 cgroup_bpf_offline(cgrp
);
5536 /* put the base reference */
5537 percpu_ref_kill(&cgrp
->self
.refcnt
);
5542 int cgroup_rmdir(struct kernfs_node
*kn
)
5544 struct cgroup
*cgrp
;
5547 cgrp
= cgroup_kn_lock_live(kn
, false);
5551 ret
= cgroup_destroy_locked(cgrp
);
5553 TRACE_CGROUP_PATH(rmdir
, cgrp
);
5555 cgroup_kn_unlock(kn
);
5559 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5560 .show_options
= cgroup_show_options
,
5561 .mkdir
= cgroup_mkdir
,
5562 .rmdir
= cgroup_rmdir
,
5563 .show_path
= cgroup_show_path
,
5566 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5568 struct cgroup_subsys_state
*css
;
5570 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5572 mutex_lock(&cgroup_mutex
);
5574 idr_init(&ss
->css_idr
);
5575 INIT_LIST_HEAD(&ss
->cfts
);
5577 /* Create the root cgroup state for this subsystem */
5578 ss
->root
= &cgrp_dfl_root
;
5579 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5580 /* We don't handle early failures gracefully */
5581 BUG_ON(IS_ERR(css
));
5582 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5585 * Root csses are never destroyed and we can't initialize
5586 * percpu_ref during early init. Disable refcnting.
5588 css
->flags
|= CSS_NO_REF
;
5591 /* allocation can't be done safely during early init */
5594 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5595 BUG_ON(css
->id
< 0);
5598 /* Update the init_css_set to contain a subsys
5599 * pointer to this state - since the subsystem is
5600 * newly registered, all tasks and hence the
5601 * init_css_set is in the subsystem's root cgroup. */
5602 init_css_set
.subsys
[ss
->id
] = css
;
5604 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5605 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5606 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5607 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5609 /* At system boot, before all subsystems have been
5610 * registered, no tasks have been forked, so we don't
5611 * need to invoke fork callbacks here. */
5612 BUG_ON(!list_empty(&init_task
.tasks
));
5614 BUG_ON(online_css(css
));
5616 mutex_unlock(&cgroup_mutex
);
5620 * cgroup_init_early - cgroup initialization at system boot
5622 * Initialize cgroups at system boot, and initialize any
5623 * subsystems that request early init.
5625 int __init
cgroup_init_early(void)
5627 static struct cgroup_fs_context __initdata ctx
;
5628 struct cgroup_subsys
*ss
;
5631 ctx
.root
= &cgrp_dfl_root
;
5632 init_cgroup_root(&ctx
);
5633 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5635 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5637 for_each_subsys(ss
, i
) {
5638 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5639 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5640 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5642 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5643 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5646 ss
->name
= cgroup_subsys_name
[i
];
5647 if (!ss
->legacy_name
)
5648 ss
->legacy_name
= cgroup_subsys_name
[i
];
5651 cgroup_init_subsys(ss
, true);
5656 static u16 cgroup_disable_mask __initdata
;
5659 * cgroup_init - cgroup initialization
5661 * Register cgroup filesystem and /proc file, and initialize
5662 * any subsystems that didn't request early init.
5664 int __init
cgroup_init(void)
5666 struct cgroup_subsys
*ss
;
5669 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5670 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5671 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5673 cgroup_rstat_boot();
5676 * The latency of the synchronize_rcu() is too high for cgroups,
5677 * avoid it at the cost of forcing all readers into the slow path.
5679 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5681 get_user_ns(init_cgroup_ns
.user_ns
);
5683 mutex_lock(&cgroup_mutex
);
5686 * Add init_css_set to the hash table so that dfl_root can link to
5689 hash_add(css_set_table
, &init_css_set
.hlist
,
5690 css_set_hash(init_css_set
.subsys
));
5692 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5694 mutex_unlock(&cgroup_mutex
);
5696 for_each_subsys(ss
, ssid
) {
5697 if (ss
->early_init
) {
5698 struct cgroup_subsys_state
*css
=
5699 init_css_set
.subsys
[ss
->id
];
5701 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5703 BUG_ON(css
->id
< 0);
5705 cgroup_init_subsys(ss
, false);
5708 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5709 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5712 * Setting dfl_root subsys_mask needs to consider the
5713 * disabled flag and cftype registration needs kmalloc,
5714 * both of which aren't available during early_init.
5716 if (cgroup_disable_mask
& (1 << ssid
)) {
5717 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5718 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5723 if (cgroup1_ssid_disabled(ssid
))
5724 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5727 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5729 /* implicit controllers must be threaded too */
5730 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5732 if (ss
->implicit_on_dfl
)
5733 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5734 else if (!ss
->dfl_cftypes
)
5735 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5738 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5740 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5741 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5743 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5744 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5748 ss
->bind(init_css_set
.subsys
[ssid
]);
5750 mutex_lock(&cgroup_mutex
);
5751 css_populate_dir(init_css_set
.subsys
[ssid
]);
5752 mutex_unlock(&cgroup_mutex
);
5755 /* init_css_set.subsys[] has been updated, re-hash */
5756 hash_del(&init_css_set
.hlist
);
5757 hash_add(css_set_table
, &init_css_set
.hlist
,
5758 css_set_hash(init_css_set
.subsys
));
5760 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5761 WARN_ON(register_filesystem(&cgroup_fs_type
));
5762 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5763 WARN_ON(!proc_create_single("cgroups", 0, NULL
, proc_cgroupstats_show
));
5764 #ifdef CONFIG_CPUSETS
5765 WARN_ON(register_filesystem(&cpuset_fs_type
));
5771 static int __init
cgroup_wq_init(void)
5774 * There isn't much point in executing destruction path in
5775 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5776 * Use 1 for @max_active.
5778 * We would prefer to do this in cgroup_init() above, but that
5779 * is called before init_workqueues(): so leave this until after.
5781 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5782 BUG_ON(!cgroup_destroy_wq
);
5785 core_initcall(cgroup_wq_init
);
5787 void cgroup_path_from_kernfs_id(u64 id
, char *buf
, size_t buflen
)
5789 struct kernfs_node
*kn
;
5791 kn
= kernfs_find_and_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5794 kernfs_path(kn
, buf
, buflen
);
5799 * proc_cgroup_show()
5800 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5801 * - Used for /proc/<pid>/cgroup.
5803 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5804 struct pid
*pid
, struct task_struct
*tsk
)
5808 struct cgroup_root
*root
;
5811 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5815 mutex_lock(&cgroup_mutex
);
5816 spin_lock_irq(&css_set_lock
);
5818 for_each_root(root
) {
5819 struct cgroup_subsys
*ss
;
5820 struct cgroup
*cgrp
;
5821 int ssid
, count
= 0;
5823 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5826 seq_printf(m
, "%d:", root
->hierarchy_id
);
5827 if (root
!= &cgrp_dfl_root
)
5828 for_each_subsys(ss
, ssid
)
5829 if (root
->subsys_mask
& (1 << ssid
))
5830 seq_printf(m
, "%s%s", count
++ ? "," : "",
5832 if (strlen(root
->name
))
5833 seq_printf(m
, "%sname=%s", count
? "," : "",
5837 cgrp
= task_cgroup_from_root(tsk
, root
);
5840 * On traditional hierarchies, all zombie tasks show up as
5841 * belonging to the root cgroup. On the default hierarchy,
5842 * while a zombie doesn't show up in "cgroup.procs" and
5843 * thus can't be migrated, its /proc/PID/cgroup keeps
5844 * reporting the cgroup it belonged to before exiting. If
5845 * the cgroup is removed before the zombie is reaped,
5846 * " (deleted)" is appended to the cgroup path.
5848 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5849 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5850 current
->nsproxy
->cgroup_ns
);
5851 if (retval
>= PATH_MAX
)
5852 retval
= -ENAMETOOLONG
;
5861 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5862 seq_puts(m
, " (deleted)\n");
5869 spin_unlock_irq(&css_set_lock
);
5870 mutex_unlock(&cgroup_mutex
);
5877 * cgroup_fork - initialize cgroup related fields during copy_process()
5878 * @child: pointer to task_struct of forking parent process.
5880 * A task is associated with the init_css_set until cgroup_post_fork()
5881 * attaches it to the target css_set.
5883 void cgroup_fork(struct task_struct
*child
)
5885 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5886 INIT_LIST_HEAD(&child
->cg_list
);
5889 static struct cgroup
*cgroup_get_from_file(struct file
*f
)
5891 struct cgroup_subsys_state
*css
;
5892 struct cgroup
*cgrp
;
5894 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
5896 return ERR_CAST(css
);
5899 if (!cgroup_on_dfl(cgrp
)) {
5901 return ERR_PTR(-EBADF
);
5908 * cgroup_css_set_fork - find or create a css_set for a child process
5909 * @kargs: the arguments passed to create the child process
5911 * This functions finds or creates a new css_set which the child
5912 * process will be attached to in cgroup_post_fork(). By default,
5913 * the child process will be given the same css_set as its parent.
5915 * If CLONE_INTO_CGROUP is specified this function will try to find an
5916 * existing css_set which includes the requested cgroup and if not create
5917 * a new css_set that the child will be attached to later. If this function
5918 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
5919 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
5920 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
5921 * to the target cgroup.
5923 static int cgroup_css_set_fork(struct kernel_clone_args
*kargs
)
5924 __acquires(&cgroup_mutex
) __acquires(&cgroup_threadgroup_rwsem
)
5927 struct cgroup
*dst_cgrp
= NULL
;
5928 struct css_set
*cset
;
5929 struct super_block
*sb
;
5932 if (kargs
->flags
& CLONE_INTO_CGROUP
)
5933 mutex_lock(&cgroup_mutex
);
5935 cgroup_threadgroup_change_begin(current
);
5937 spin_lock_irq(&css_set_lock
);
5938 cset
= task_css_set(current
);
5940 spin_unlock_irq(&css_set_lock
);
5942 if (!(kargs
->flags
& CLONE_INTO_CGROUP
)) {
5947 f
= fget_raw(kargs
->cgroup
);
5952 sb
= f
->f_path
.dentry
->d_sb
;
5954 dst_cgrp
= cgroup_get_from_file(f
);
5955 if (IS_ERR(dst_cgrp
)) {
5956 ret
= PTR_ERR(dst_cgrp
);
5961 if (cgroup_is_dead(dst_cgrp
)) {
5967 * Verify that we the target cgroup is writable for us. This is
5968 * usually done by the vfs layer but since we're not going through
5969 * the vfs layer here we need to do it "manually".
5971 ret
= cgroup_may_write(dst_cgrp
, sb
);
5975 ret
= cgroup_attach_permissions(cset
->dfl_cgrp
, dst_cgrp
, sb
,
5976 !(kargs
->flags
& CLONE_THREAD
));
5980 kargs
->cset
= find_css_set(cset
, dst_cgrp
);
5988 kargs
->cgrp
= dst_cgrp
;
5992 cgroup_threadgroup_change_end(current
);
5993 mutex_unlock(&cgroup_mutex
);
5997 cgroup_put(dst_cgrp
);
6000 put_css_set(kargs
->cset
);
6005 * cgroup_css_set_put_fork - drop references we took during fork
6006 * @kargs: the arguments passed to create the child process
6008 * Drop references to the prepared css_set and target cgroup if
6009 * CLONE_INTO_CGROUP was requested.
6011 static void cgroup_css_set_put_fork(struct kernel_clone_args
*kargs
)
6012 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6014 cgroup_threadgroup_change_end(current
);
6016 if (kargs
->flags
& CLONE_INTO_CGROUP
) {
6017 struct cgroup
*cgrp
= kargs
->cgrp
;
6018 struct css_set
*cset
= kargs
->cset
;
6020 mutex_unlock(&cgroup_mutex
);
6035 * cgroup_can_fork - called on a new task before the process is exposed
6036 * @child: the child process
6038 * This prepares a new css_set for the child process which the child will
6039 * be attached to in cgroup_post_fork().
6040 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6041 * callback returns an error, the fork aborts with that error code. This
6042 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6044 int cgroup_can_fork(struct task_struct
*child
, struct kernel_clone_args
*kargs
)
6046 struct cgroup_subsys
*ss
;
6049 ret
= cgroup_css_set_fork(kargs
);
6053 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
6054 ret
= ss
->can_fork(child
, kargs
->cset
);
6057 } while_each_subsys_mask();
6062 for_each_subsys(ss
, j
) {
6065 if (ss
->cancel_fork
)
6066 ss
->cancel_fork(child
, kargs
->cset
);
6069 cgroup_css_set_put_fork(kargs
);
6075 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6076 * @child: the child process
6077 * @kargs: the arguments passed to create the child process
6079 * This calls the cancel_fork() callbacks if a fork failed *after*
6080 * cgroup_can_fork() succeded and cleans up references we took to
6081 * prepare a new css_set for the child process in cgroup_can_fork().
6083 void cgroup_cancel_fork(struct task_struct
*child
,
6084 struct kernel_clone_args
*kargs
)
6086 struct cgroup_subsys
*ss
;
6089 for_each_subsys(ss
, i
)
6090 if (ss
->cancel_fork
)
6091 ss
->cancel_fork(child
, kargs
->cset
);
6093 cgroup_css_set_put_fork(kargs
);
6097 * cgroup_post_fork - finalize cgroup setup for the child process
6098 * @child: the child process
6100 * Attach the child process to its css_set calling the subsystem fork()
6103 void cgroup_post_fork(struct task_struct
*child
,
6104 struct kernel_clone_args
*kargs
)
6105 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6107 struct cgroup_subsys
*ss
;
6108 struct css_set
*cset
;
6114 spin_lock_irq(&css_set_lock
);
6116 /* init tasks are special, only link regular threads */
6117 if (likely(child
->pid
)) {
6118 WARN_ON_ONCE(!list_empty(&child
->cg_list
));
6120 css_set_move_task(child
, NULL
, cset
, false);
6127 * If the cgroup has to be frozen, the new task has too. Let's set
6128 * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
6131 if (unlikely(cgroup_task_freeze(child
))) {
6132 spin_lock(&child
->sighand
->siglock
);
6133 WARN_ON_ONCE(child
->frozen
);
6134 child
->jobctl
|= JOBCTL_TRAP_FREEZE
;
6135 spin_unlock(&child
->sighand
->siglock
);
6138 * Calling cgroup_update_frozen() isn't required here,
6139 * because it will be called anyway a bit later from
6140 * do_freezer_trap(). So we avoid cgroup's transient switch
6141 * from the frozen state and back.
6145 spin_unlock_irq(&css_set_lock
);
6148 * Call ss->fork(). This must happen after @child is linked on
6149 * css_set; otherwise, @child might change state between ->fork()
6150 * and addition to css_set.
6152 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
6154 } while_each_subsys_mask();
6156 /* Make the new cset the root_cset of the new cgroup namespace. */
6157 if (kargs
->flags
& CLONE_NEWCGROUP
) {
6158 struct css_set
*rcset
= child
->nsproxy
->cgroup_ns
->root_cset
;
6161 child
->nsproxy
->cgroup_ns
->root_cset
= cset
;
6165 cgroup_css_set_put_fork(kargs
);
6169 * cgroup_exit - detach cgroup from exiting task
6170 * @tsk: pointer to task_struct of exiting process
6172 * Description: Detach cgroup from @tsk.
6175 void cgroup_exit(struct task_struct
*tsk
)
6177 struct cgroup_subsys
*ss
;
6178 struct css_set
*cset
;
6181 spin_lock_irq(&css_set_lock
);
6183 WARN_ON_ONCE(list_empty(&tsk
->cg_list
));
6184 cset
= task_css_set(tsk
);
6185 css_set_move_task(tsk
, cset
, NULL
, false);
6186 list_add_tail(&tsk
->cg_list
, &cset
->dying_tasks
);
6189 WARN_ON_ONCE(cgroup_task_frozen(tsk
));
6190 if (unlikely(cgroup_task_freeze(tsk
)))
6191 cgroup_update_frozen(task_dfl_cgroup(tsk
));
6193 spin_unlock_irq(&css_set_lock
);
6195 /* see cgroup_post_fork() for details */
6196 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
6198 } while_each_subsys_mask();
6201 void cgroup_release(struct task_struct
*task
)
6203 struct cgroup_subsys
*ss
;
6206 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
6208 } while_each_subsys_mask();
6210 spin_lock_irq(&css_set_lock
);
6211 css_set_skip_task_iters(task_css_set(task
), task
);
6212 list_del_init(&task
->cg_list
);
6213 spin_unlock_irq(&css_set_lock
);
6216 void cgroup_free(struct task_struct
*task
)
6218 struct css_set
*cset
= task_css_set(task
);
6222 static int __init
cgroup_disable(char *str
)
6224 struct cgroup_subsys
*ss
;
6228 while ((token
= strsep(&str
, ",")) != NULL
) {
6232 for_each_subsys(ss
, i
) {
6233 if (strcmp(token
, ss
->name
) &&
6234 strcmp(token
, ss
->legacy_name
))
6236 cgroup_disable_mask
|= 1 << i
;
6241 __setup("cgroup_disable=", cgroup_disable
);
6243 void __init __weak
enable_debug_cgroup(void) { }
6245 static int __init
enable_cgroup_debug(char *str
)
6247 cgroup_debug
= true;
6248 enable_debug_cgroup();
6251 __setup("cgroup_debug", enable_cgroup_debug
);
6254 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6255 * @dentry: directory dentry of interest
6256 * @ss: subsystem of interest
6258 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6259 * to get the corresponding css and return it. If such css doesn't exist
6260 * or can't be pinned, an ERR_PTR value is returned.
6262 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6263 struct cgroup_subsys
*ss
)
6265 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6266 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6267 struct cgroup_subsys_state
*css
= NULL
;
6268 struct cgroup
*cgrp
;
6270 /* is @dentry a cgroup dir? */
6271 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6272 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6273 return ERR_PTR(-EBADF
);
6278 * This path doesn't originate from kernfs and @kn could already
6279 * have been or be removed at any point. @kn->priv is RCU
6280 * protected for this access. See css_release_work_fn() for details.
6282 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6284 css
= cgroup_css(cgrp
, ss
);
6286 if (!css
|| !css_tryget_online(css
))
6287 css
= ERR_PTR(-ENOENT
);
6294 * css_from_id - lookup css by id
6295 * @id: the cgroup id
6296 * @ss: cgroup subsys to be looked into
6298 * Returns the css if there's valid one with @id, otherwise returns NULL.
6299 * Should be called under rcu_read_lock().
6301 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6303 WARN_ON_ONCE(!rcu_read_lock_held());
6304 return idr_find(&ss
->css_idr
, id
);
6308 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6309 * @path: path on the default hierarchy
6311 * Find the cgroup at @path on the default hierarchy, increment its
6312 * reference count and return it. Returns pointer to the found cgroup on
6313 * success, ERR_PTR(-ENOENT) if @path doesn't exist and ERR_PTR(-ENOTDIR)
6314 * if @path points to a non-directory.
6316 struct cgroup
*cgroup_get_from_path(const char *path
)
6318 struct kernfs_node
*kn
;
6319 struct cgroup
*cgrp
;
6321 mutex_lock(&cgroup_mutex
);
6323 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
6325 if (kernfs_type(kn
) == KERNFS_DIR
) {
6327 cgroup_get_live(cgrp
);
6329 cgrp
= ERR_PTR(-ENOTDIR
);
6333 cgrp
= ERR_PTR(-ENOENT
);
6336 mutex_unlock(&cgroup_mutex
);
6339 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6342 * cgroup_get_from_fd - get a cgroup pointer from a fd
6343 * @fd: fd obtained by open(cgroup2_dir)
6345 * Find the cgroup from a fd which should be obtained
6346 * by opening a cgroup directory. Returns a pointer to the
6347 * cgroup on success. ERR_PTR is returned if the cgroup
6350 struct cgroup
*cgroup_get_from_fd(int fd
)
6352 struct cgroup
*cgrp
;
6357 return ERR_PTR(-EBADF
);
6359 cgrp
= cgroup_get_from_file(f
);
6363 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6365 static u64
power_of_ten(int power
)
6374 * cgroup_parse_float - parse a floating number
6375 * @input: input string
6376 * @dec_shift: number of decimal digits to shift
6379 * Parse a decimal floating point number in @input and store the result in
6380 * @v with decimal point right shifted @dec_shift times. For example, if
6381 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6382 * Returns 0 on success, -errno otherwise.
6384 * There's nothing cgroup specific about this function except that it's
6385 * currently the only user.
6387 int cgroup_parse_float(const char *input
, unsigned dec_shift
, s64
*v
)
6389 s64 whole
, frac
= 0;
6390 int fstart
= 0, fend
= 0, flen
;
6392 if (!sscanf(input
, "%lld.%n%lld%n", &whole
, &fstart
, &frac
, &fend
))
6397 flen
= fend
> fstart
? fend
- fstart
: 0;
6398 if (flen
< dec_shift
)
6399 frac
*= power_of_ten(dec_shift
- flen
);
6401 frac
= DIV_ROUND_CLOSEST_ULL(frac
, power_of_ten(flen
- dec_shift
));
6403 *v
= whole
* power_of_ten(dec_shift
) + frac
;
6408 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6409 * definition in cgroup-defs.h.
6411 #ifdef CONFIG_SOCK_CGROUP_DATA
6413 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6415 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
6416 static bool cgroup_sk_alloc_disabled __read_mostly
;
6418 void cgroup_sk_alloc_disable(void)
6420 if (cgroup_sk_alloc_disabled
)
6422 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6423 cgroup_sk_alloc_disabled
= true;
6428 #define cgroup_sk_alloc_disabled false
6432 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6434 if (cgroup_sk_alloc_disabled
) {
6435 skcd
->no_refcnt
= 1;
6439 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6446 struct css_set
*cset
;
6448 cset
= task_css_set(current
);
6449 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6450 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
6451 cgroup_bpf_get(cset
->dfl_cgrp
);
6460 void cgroup_sk_clone(struct sock_cgroup_data
*skcd
)
6463 if (skcd
->no_refcnt
)
6466 * We might be cloning a socket which is left in an empty
6467 * cgroup and the cgroup might have already been rmdir'd.
6468 * Don't use cgroup_get_live().
6470 cgroup_get(sock_cgroup_ptr(skcd
));
6471 cgroup_bpf_get(sock_cgroup_ptr(skcd
));
6475 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6477 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6479 if (skcd
->no_refcnt
)
6481 cgroup_bpf_put(cgrp
);
6485 #endif /* CONFIG_SOCK_CGROUP_DATA */
6487 #ifdef CONFIG_CGROUP_BPF
6488 int cgroup_bpf_attach(struct cgroup
*cgrp
,
6489 struct bpf_prog
*prog
, struct bpf_prog
*replace_prog
,
6490 struct bpf_cgroup_link
*link
,
6491 enum bpf_attach_type type
,
6496 mutex_lock(&cgroup_mutex
);
6497 ret
= __cgroup_bpf_attach(cgrp
, prog
, replace_prog
, link
, type
, flags
);
6498 mutex_unlock(&cgroup_mutex
);
6502 int cgroup_bpf_detach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6503 enum bpf_attach_type type
)
6507 mutex_lock(&cgroup_mutex
);
6508 ret
= __cgroup_bpf_detach(cgrp
, prog
, NULL
, type
);
6509 mutex_unlock(&cgroup_mutex
);
6513 int cgroup_bpf_query(struct cgroup
*cgrp
, const union bpf_attr
*attr
,
6514 union bpf_attr __user
*uattr
)
6518 mutex_lock(&cgroup_mutex
);
6519 ret
= __cgroup_bpf_query(cgrp
, attr
, uattr
);
6520 mutex_unlock(&cgroup_mutex
);
6523 #endif /* CONFIG_CGROUP_BPF */
6526 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
6527 ssize_t size
, const char *prefix
)
6532 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
6533 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
6537 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
6539 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
6541 if (WARN_ON(ret
>= size
))
6548 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6551 struct cgroup_subsys
*ss
;
6555 ret
= show_delegatable_files(cgroup_base_files
, buf
, PAGE_SIZE
- ret
,
6558 for_each_subsys(ss
, ssid
)
6559 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
6561 cgroup_subsys_name
[ssid
]);
6565 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
6567 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6570 return snprintf(buf
, PAGE_SIZE
,
6572 "memory_localevents\n"
6573 "memory_recursiveprot\n");
6575 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
6577 static struct attribute
*cgroup_sysfs_attrs
[] = {
6578 &cgroup_delegate_attr
.attr
,
6579 &cgroup_features_attr
.attr
,
6583 static const struct attribute_group cgroup_sysfs_attr_group
= {
6584 .attrs
= cgroup_sysfs_attrs
,
6588 static int __init
cgroup_sysfs_init(void)
6590 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
6592 subsys_initcall(cgroup_sysfs_init
);
6594 #endif /* CONFIG_SYSFS */