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
);
157 * The default hierarchy, reserved for the subsystems that are otherwise
158 * unattached - it never has more than a single cgroup, and all tasks are
159 * part of that cgroup.
161 struct cgroup_root cgrp_dfl_root
= { .cgrp
.rstat_cpu
= &cgrp_dfl_root_rstat_cpu
};
162 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
165 * The default hierarchy always exists but is hidden until mounted for the
166 * first time. This is for backward compatibility.
168 static bool cgrp_dfl_visible
;
170 /* some controllers are not supported in the default hierarchy */
171 static u16 cgrp_dfl_inhibit_ss_mask
;
173 /* some controllers are implicitly enabled on the default hierarchy */
174 static u16 cgrp_dfl_implicit_ss_mask
;
176 /* some controllers can be threaded on the default hierarchy */
177 static u16 cgrp_dfl_threaded_ss_mask
;
179 /* The list of hierarchy roots */
180 LIST_HEAD(cgroup_roots
);
181 static int cgroup_root_count
;
183 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
184 static DEFINE_IDR(cgroup_hierarchy_idr
);
187 * Assign a monotonically increasing serial number to csses. It guarantees
188 * cgroups with bigger numbers are newer than those with smaller numbers.
189 * Also, as csses are always appended to the parent's ->children list, it
190 * guarantees that sibling csses are always sorted in the ascending serial
191 * number order on the list. Protected by cgroup_mutex.
193 static u64 css_serial_nr_next
= 1;
196 * These bitmasks identify subsystems with specific features to avoid
197 * having to do iterative checks repeatedly.
199 static u16 have_fork_callback __read_mostly
;
200 static u16 have_exit_callback __read_mostly
;
201 static u16 have_release_callback __read_mostly
;
202 static u16 have_canfork_callback __read_mostly
;
204 /* cgroup namespace for init task */
205 struct cgroup_namespace init_cgroup_ns
= {
206 .count
= REFCOUNT_INIT(2),
207 .user_ns
= &init_user_ns
,
208 .ns
.ops
= &cgroupns_operations
,
209 .ns
.inum
= PROC_CGROUP_INIT_INO
,
210 .root_cset
= &init_css_set
,
213 static struct file_system_type cgroup2_fs_type
;
214 static struct cftype cgroup_base_files
[];
216 static int cgroup_apply_control(struct cgroup
*cgrp
);
217 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
218 static void css_task_iter_skip(struct css_task_iter
*it
,
219 struct task_struct
*task
);
220 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
221 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
222 struct cgroup_subsys
*ss
);
223 static void css_release(struct percpu_ref
*ref
);
224 static void kill_css(struct cgroup_subsys_state
*css
);
225 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
226 struct cgroup
*cgrp
, struct cftype cfts
[],
230 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
231 * @ssid: subsys ID of interest
233 * cgroup_subsys_enabled() can only be used with literal subsys names which
234 * is fine for individual subsystems but unsuitable for cgroup core. This
235 * is slower static_key_enabled() based test indexed by @ssid.
237 bool cgroup_ssid_enabled(int ssid
)
239 if (CGROUP_SUBSYS_COUNT
== 0)
242 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
246 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
247 * @cgrp: the cgroup of interest
249 * The default hierarchy is the v2 interface of cgroup and this function
250 * can be used to test whether a cgroup is on the default hierarchy for
251 * cases where a subsystem should behave differnetly depending on the
254 * The set of behaviors which change on the default hierarchy are still
255 * being determined and the mount option is prefixed with __DEVEL__.
257 * List of changed behaviors:
259 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
260 * and "name" are disallowed.
262 * - When mounting an existing superblock, mount options should match.
264 * - Remount is disallowed.
266 * - rename(2) is disallowed.
268 * - "tasks" is removed. Everything should be at process granularity. Use
269 * "cgroup.procs" instead.
271 * - "cgroup.procs" is not sorted. pids will be unique unless they got
272 * recycled inbetween reads.
274 * - "release_agent" and "notify_on_release" are removed. Replacement
275 * notification mechanism will be implemented.
277 * - "cgroup.clone_children" is removed.
279 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
280 * and its descendants contain no task; otherwise, 1. The file also
281 * generates kernfs notification which can be monitored through poll and
282 * [di]notify when the value of the file changes.
284 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
285 * take masks of ancestors with non-empty cpus/mems, instead of being
286 * moved to an ancestor.
288 * - cpuset: a task can be moved into an empty cpuset, and again it takes
289 * masks of ancestors.
291 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
294 * - blkcg: blk-throttle becomes properly hierarchical.
296 * - debug: disallowed on the default hierarchy.
298 bool cgroup_on_dfl(const struct cgroup
*cgrp
)
300 return cgrp
->root
== &cgrp_dfl_root
;
303 /* IDR wrappers which synchronize using cgroup_idr_lock */
304 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
309 idr_preload(gfp_mask
);
310 spin_lock_bh(&cgroup_idr_lock
);
311 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
312 spin_unlock_bh(&cgroup_idr_lock
);
317 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
321 spin_lock_bh(&cgroup_idr_lock
);
322 ret
= idr_replace(idr
, ptr
, id
);
323 spin_unlock_bh(&cgroup_idr_lock
);
327 static void cgroup_idr_remove(struct idr
*idr
, int id
)
329 spin_lock_bh(&cgroup_idr_lock
);
331 spin_unlock_bh(&cgroup_idr_lock
);
334 static bool cgroup_has_tasks(struct cgroup
*cgrp
)
336 return cgrp
->nr_populated_csets
;
339 bool cgroup_is_threaded(struct cgroup
*cgrp
)
341 return cgrp
->dom_cgrp
!= cgrp
;
344 /* can @cgrp host both domain and threaded children? */
345 static bool cgroup_is_mixable(struct cgroup
*cgrp
)
348 * Root isn't under domain level resource control exempting it from
349 * the no-internal-process constraint, so it can serve as a thread
350 * root and a parent of resource domains at the same time.
352 return !cgroup_parent(cgrp
);
355 /* can @cgrp become a thread root? should always be true for a thread root */
356 static bool cgroup_can_be_thread_root(struct cgroup
*cgrp
)
358 /* mixables don't care */
359 if (cgroup_is_mixable(cgrp
))
362 /* domain roots can't be nested under threaded */
363 if (cgroup_is_threaded(cgrp
))
366 /* can only have either domain or threaded children */
367 if (cgrp
->nr_populated_domain_children
)
370 /* and no domain controllers can be enabled */
371 if (cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
377 /* is @cgrp root of a threaded subtree? */
378 bool cgroup_is_thread_root(struct cgroup
*cgrp
)
380 /* thread root should be a domain */
381 if (cgroup_is_threaded(cgrp
))
384 /* a domain w/ threaded children is a thread root */
385 if (cgrp
->nr_threaded_children
)
389 * A domain which has tasks and explicit threaded controllers
390 * enabled is a thread root.
392 if (cgroup_has_tasks(cgrp
) &&
393 (cgrp
->subtree_control
& cgrp_dfl_threaded_ss_mask
))
399 /* a domain which isn't connected to the root w/o brekage can't be used */
400 static bool cgroup_is_valid_domain(struct cgroup
*cgrp
)
402 /* the cgroup itself can be a thread root */
403 if (cgroup_is_threaded(cgrp
))
406 /* but the ancestors can't be unless mixable */
407 while ((cgrp
= cgroup_parent(cgrp
))) {
408 if (!cgroup_is_mixable(cgrp
) && cgroup_is_thread_root(cgrp
))
410 if (cgroup_is_threaded(cgrp
))
417 /* subsystems visibly enabled on a cgroup */
418 static u16
cgroup_control(struct cgroup
*cgrp
)
420 struct cgroup
*parent
= cgroup_parent(cgrp
);
421 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
424 u16 ss_mask
= parent
->subtree_control
;
426 /* threaded cgroups can only have threaded controllers */
427 if (cgroup_is_threaded(cgrp
))
428 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
432 if (cgroup_on_dfl(cgrp
))
433 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
434 cgrp_dfl_implicit_ss_mask
);
438 /* subsystems enabled on a cgroup */
439 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
441 struct cgroup
*parent
= cgroup_parent(cgrp
);
444 u16 ss_mask
= parent
->subtree_ss_mask
;
446 /* threaded cgroups can only have threaded controllers */
447 if (cgroup_is_threaded(cgrp
))
448 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
452 return cgrp
->root
->subsys_mask
;
456 * cgroup_css - obtain a cgroup's css for the specified subsystem
457 * @cgrp: the cgroup of interest
458 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
460 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
461 * function must be called either under cgroup_mutex or rcu_read_lock() and
462 * the caller is responsible for pinning the returned css if it wants to
463 * keep accessing it outside the said locks. This function may return
464 * %NULL if @cgrp doesn't have @subsys_id enabled.
466 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
467 struct cgroup_subsys
*ss
)
470 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
471 lockdep_is_held(&cgroup_mutex
));
477 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
478 * @cgrp: the cgroup of interest
479 * @ss: the subsystem of interest
481 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
482 * or is offline, %NULL is returned.
484 static struct cgroup_subsys_state
*cgroup_tryget_css(struct cgroup
*cgrp
,
485 struct cgroup_subsys
*ss
)
487 struct cgroup_subsys_state
*css
;
490 css
= cgroup_css(cgrp
, ss
);
491 if (!css
|| !css_tryget_online(css
))
499 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
500 * @cgrp: the cgroup of interest
501 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
503 * Similar to cgroup_css() but returns the effective css, which is defined
504 * as the matching css of the nearest ancestor including self which has @ss
505 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
506 * function is guaranteed to return non-NULL css.
508 static struct cgroup_subsys_state
*cgroup_e_css_by_mask(struct cgroup
*cgrp
,
509 struct cgroup_subsys
*ss
)
511 lockdep_assert_held(&cgroup_mutex
);
517 * This function is used while updating css associations and thus
518 * can't test the csses directly. Test ss_mask.
520 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
521 cgrp
= cgroup_parent(cgrp
);
526 return cgroup_css(cgrp
, ss
);
530 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
531 * @cgrp: the cgroup of interest
532 * @ss: the subsystem of interest
534 * Find and get the effective css of @cgrp for @ss. The effective css is
535 * defined as the matching css of the nearest ancestor including self which
536 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
537 * the root css is returned, so this function always returns a valid css.
539 * The returned css is not guaranteed to be online, and therefore it is the
540 * callers responsiblity to tryget a reference for it.
542 struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
543 struct cgroup_subsys
*ss
)
545 struct cgroup_subsys_state
*css
;
548 css
= cgroup_css(cgrp
, ss
);
552 cgrp
= cgroup_parent(cgrp
);
555 return init_css_set
.subsys
[ss
->id
];
559 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
560 * @cgrp: the cgroup of interest
561 * @ss: the subsystem of interest
563 * Find and get the effective css of @cgrp for @ss. The effective css is
564 * defined as the matching css of the nearest ancestor including self which
565 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
566 * the root css is returned, so this function always returns a valid css.
567 * The returned css must be put using css_put().
569 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
570 struct cgroup_subsys
*ss
)
572 struct cgroup_subsys_state
*css
;
577 css
= cgroup_css(cgrp
, ss
);
579 if (css
&& css_tryget_online(css
))
581 cgrp
= cgroup_parent(cgrp
);
584 css
= init_css_set
.subsys
[ss
->id
];
591 static void cgroup_get_live(struct cgroup
*cgrp
)
593 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
594 css_get(&cgrp
->self
);
598 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
599 * is responsible for taking the css_set_lock.
600 * @cgrp: the cgroup in question
602 int __cgroup_task_count(const struct cgroup
*cgrp
)
605 struct cgrp_cset_link
*link
;
607 lockdep_assert_held(&css_set_lock
);
609 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
610 count
+= link
->cset
->nr_tasks
;
616 * cgroup_task_count - count the number of tasks in a cgroup.
617 * @cgrp: the cgroup in question
619 int cgroup_task_count(const struct cgroup
*cgrp
)
623 spin_lock_irq(&css_set_lock
);
624 count
= __cgroup_task_count(cgrp
);
625 spin_unlock_irq(&css_set_lock
);
630 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
632 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
633 struct cftype
*cft
= of_cft(of
);
636 * This is open and unprotected implementation of cgroup_css().
637 * seq_css() is only called from a kernfs file operation which has
638 * an active reference on the file. Because all the subsystem
639 * files are drained before a css is disassociated with a cgroup,
640 * the matching css from the cgroup's subsys table is guaranteed to
641 * be and stay valid until the enclosing operation is complete.
644 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
648 EXPORT_SYMBOL_GPL(of_css
);
651 * for_each_css - iterate all css's of a cgroup
652 * @css: the iteration cursor
653 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
654 * @cgrp: the target cgroup to iterate css's of
656 * Should be called under cgroup_[tree_]mutex.
658 #define for_each_css(css, ssid, cgrp) \
659 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
660 if (!((css) = rcu_dereference_check( \
661 (cgrp)->subsys[(ssid)], \
662 lockdep_is_held(&cgroup_mutex)))) { } \
666 * for_each_e_css - iterate all effective css's of a cgroup
667 * @css: the iteration cursor
668 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
669 * @cgrp: the target cgroup to iterate css's of
671 * Should be called under cgroup_[tree_]mutex.
673 #define for_each_e_css(css, ssid, cgrp) \
674 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
675 if (!((css) = cgroup_e_css_by_mask(cgrp, \
676 cgroup_subsys[(ssid)]))) \
681 * do_each_subsys_mask - filter for_each_subsys with a bitmask
682 * @ss: the iteration cursor
683 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
684 * @ss_mask: the bitmask
686 * The block will only run for cases where the ssid-th bit (1 << ssid) of
689 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
690 unsigned long __ss_mask = (ss_mask); \
691 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
695 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
696 (ss) = cgroup_subsys[ssid]; \
699 #define while_each_subsys_mask() \
704 /* iterate over child cgrps, lock should be held throughout iteration */
705 #define cgroup_for_each_live_child(child, cgrp) \
706 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
707 if (({ lockdep_assert_held(&cgroup_mutex); \
708 cgroup_is_dead(child); })) \
712 /* walk live descendants in preorder */
713 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
714 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
715 if (({ lockdep_assert_held(&cgroup_mutex); \
716 (dsct) = (d_css)->cgroup; \
717 cgroup_is_dead(dsct); })) \
721 /* walk live descendants in postorder */
722 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
723 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
724 if (({ lockdep_assert_held(&cgroup_mutex); \
725 (dsct) = (d_css)->cgroup; \
726 cgroup_is_dead(dsct); })) \
731 * The default css_set - used by init and its children prior to any
732 * hierarchies being mounted. It contains a pointer to the root state
733 * for each subsystem. Also used to anchor the list of css_sets. Not
734 * reference-counted, to improve performance when child cgroups
735 * haven't been created.
737 struct css_set init_css_set
= {
738 .refcount
= REFCOUNT_INIT(1),
739 .dom_cset
= &init_css_set
,
740 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
741 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
742 .dying_tasks
= LIST_HEAD_INIT(init_css_set
.dying_tasks
),
743 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
744 .threaded_csets
= LIST_HEAD_INIT(init_css_set
.threaded_csets
),
745 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
746 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
747 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
750 * The following field is re-initialized when this cset gets linked
751 * in cgroup_init(). However, let's initialize the field
752 * statically too so that the default cgroup can be accessed safely
755 .dfl_cgrp
= &cgrp_dfl_root
.cgrp
,
758 static int css_set_count
= 1; /* 1 for init_css_set */
760 static bool css_set_threaded(struct css_set
*cset
)
762 return cset
->dom_cset
!= cset
;
766 * css_set_populated - does a css_set contain any tasks?
767 * @cset: target css_set
769 * css_set_populated() should be the same as !!cset->nr_tasks at steady
770 * state. However, css_set_populated() can be called while a task is being
771 * added to or removed from the linked list before the nr_tasks is
772 * properly updated. Hence, we can't just look at ->nr_tasks here.
774 static bool css_set_populated(struct css_set
*cset
)
776 lockdep_assert_held(&css_set_lock
);
778 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
782 * cgroup_update_populated - update the populated count of a cgroup
783 * @cgrp: the target cgroup
784 * @populated: inc or dec populated count
786 * One of the css_sets associated with @cgrp is either getting its first
787 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
788 * count is propagated towards root so that a given cgroup's
789 * nr_populated_children is zero iff none of its descendants contain any
792 * @cgrp's interface file "cgroup.populated" is zero if both
793 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
794 * 1 otherwise. When the sum changes from or to zero, userland is notified
795 * that the content of the interface file has changed. This can be used to
796 * detect when @cgrp and its descendants become populated or empty.
798 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
800 struct cgroup
*child
= NULL
;
801 int adj
= populated
? 1 : -1;
803 lockdep_assert_held(&css_set_lock
);
806 bool was_populated
= cgroup_is_populated(cgrp
);
809 cgrp
->nr_populated_csets
+= adj
;
811 if (cgroup_is_threaded(child
))
812 cgrp
->nr_populated_threaded_children
+= adj
;
814 cgrp
->nr_populated_domain_children
+= adj
;
817 if (was_populated
== cgroup_is_populated(cgrp
))
820 cgroup1_check_for_release(cgrp
);
821 TRACE_CGROUP_PATH(notify_populated
, cgrp
,
822 cgroup_is_populated(cgrp
));
823 cgroup_file_notify(&cgrp
->events_file
);
826 cgrp
= cgroup_parent(cgrp
);
831 * css_set_update_populated - update populated state of a css_set
832 * @cset: target css_set
833 * @populated: whether @cset is populated or depopulated
835 * @cset is either getting the first task or losing the last. Update the
836 * populated counters of all associated cgroups accordingly.
838 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
840 struct cgrp_cset_link
*link
;
842 lockdep_assert_held(&css_set_lock
);
844 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
845 cgroup_update_populated(link
->cgrp
, populated
);
849 * @task is leaving, advance task iterators which are pointing to it so
850 * that they can resume at the next position. Advancing an iterator might
851 * remove it from the list, use safe walk. See css_task_iter_skip() for
854 static void css_set_skip_task_iters(struct css_set
*cset
,
855 struct task_struct
*task
)
857 struct css_task_iter
*it
, *pos
;
859 list_for_each_entry_safe(it
, pos
, &cset
->task_iters
, iters_node
)
860 css_task_iter_skip(it
, task
);
864 * css_set_move_task - move a task from one css_set to another
865 * @task: task being moved
866 * @from_cset: css_set @task currently belongs to (may be NULL)
867 * @to_cset: new css_set @task is being moved to (may be NULL)
868 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
870 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
871 * css_set, @from_cset can be NULL. If @task is being disassociated
872 * instead of moved, @to_cset can be NULL.
874 * This function automatically handles populated counter updates and
875 * css_task_iter adjustments but the caller is responsible for managing
876 * @from_cset and @to_cset's reference counts.
878 static void css_set_move_task(struct task_struct
*task
,
879 struct css_set
*from_cset
, struct css_set
*to_cset
,
882 lockdep_assert_held(&css_set_lock
);
884 if (to_cset
&& !css_set_populated(to_cset
))
885 css_set_update_populated(to_cset
, true);
888 WARN_ON_ONCE(list_empty(&task
->cg_list
));
890 css_set_skip_task_iters(from_cset
, task
);
891 list_del_init(&task
->cg_list
);
892 if (!css_set_populated(from_cset
))
893 css_set_update_populated(from_cset
, false);
895 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
900 * We are synchronized through cgroup_threadgroup_rwsem
901 * against PF_EXITING setting such that we can't race
902 * against cgroup_exit() changing the css_set to
903 * init_css_set and dropping the old one.
905 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
907 cgroup_move_task(task
, to_cset
);
908 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
914 * hash table for cgroup groups. This improves the performance to find
915 * an existing css_set. This hash doesn't (currently) take into
916 * account cgroups in empty hierarchies.
918 #define CSS_SET_HASH_BITS 7
919 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
921 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
923 unsigned long key
= 0UL;
924 struct cgroup_subsys
*ss
;
927 for_each_subsys(ss
, i
)
928 key
+= (unsigned long)css
[i
];
929 key
= (key
>> 16) ^ key
;
934 void put_css_set_locked(struct css_set
*cset
)
936 struct cgrp_cset_link
*link
, *tmp_link
;
937 struct cgroup_subsys
*ss
;
940 lockdep_assert_held(&css_set_lock
);
942 if (!refcount_dec_and_test(&cset
->refcount
))
945 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
947 /* This css_set is dead. unlink it and release cgroup and css refs */
948 for_each_subsys(ss
, ssid
) {
949 list_del(&cset
->e_cset_node
[ssid
]);
950 css_put(cset
->subsys
[ssid
]);
952 hash_del(&cset
->hlist
);
955 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
956 list_del(&link
->cset_link
);
957 list_del(&link
->cgrp_link
);
958 if (cgroup_parent(link
->cgrp
))
959 cgroup_put(link
->cgrp
);
963 if (css_set_threaded(cset
)) {
964 list_del(&cset
->threaded_csets_node
);
965 put_css_set_locked(cset
->dom_cset
);
968 kfree_rcu(cset
, rcu_head
);
972 * compare_css_sets - helper function for find_existing_css_set().
973 * @cset: candidate css_set being tested
974 * @old_cset: existing css_set for a task
975 * @new_cgrp: cgroup that's being entered by the task
976 * @template: desired set of css pointers in css_set (pre-calculated)
978 * Returns true if "cset" matches "old_cset" except for the hierarchy
979 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
981 static bool compare_css_sets(struct css_set
*cset
,
982 struct css_set
*old_cset
,
983 struct cgroup
*new_cgrp
,
984 struct cgroup_subsys_state
*template[])
986 struct cgroup
*new_dfl_cgrp
;
987 struct list_head
*l1
, *l2
;
990 * On the default hierarchy, there can be csets which are
991 * associated with the same set of cgroups but different csses.
992 * Let's first ensure that csses match.
994 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
998 /* @cset's domain should match the default cgroup's */
999 if (cgroup_on_dfl(new_cgrp
))
1000 new_dfl_cgrp
= new_cgrp
;
1002 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
1004 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
1008 * Compare cgroup pointers in order to distinguish between
1009 * different cgroups in hierarchies. As different cgroups may
1010 * share the same effective css, this comparison is always
1013 l1
= &cset
->cgrp_links
;
1014 l2
= &old_cset
->cgrp_links
;
1016 struct cgrp_cset_link
*link1
, *link2
;
1017 struct cgroup
*cgrp1
, *cgrp2
;
1021 /* See if we reached the end - both lists are equal length. */
1022 if (l1
== &cset
->cgrp_links
) {
1023 BUG_ON(l2
!= &old_cset
->cgrp_links
);
1026 BUG_ON(l2
== &old_cset
->cgrp_links
);
1028 /* Locate the cgroups associated with these links. */
1029 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
1030 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
1031 cgrp1
= link1
->cgrp
;
1032 cgrp2
= link2
->cgrp
;
1033 /* Hierarchies should be linked in the same order. */
1034 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
1037 * If this hierarchy is the hierarchy of the cgroup
1038 * that's changing, then we need to check that this
1039 * css_set points to the new cgroup; if it's any other
1040 * hierarchy, then this css_set should point to the
1041 * same cgroup as the old css_set.
1043 if (cgrp1
->root
== new_cgrp
->root
) {
1044 if (cgrp1
!= new_cgrp
)
1055 * find_existing_css_set - init css array and find the matching css_set
1056 * @old_cset: the css_set that we're using before the cgroup transition
1057 * @cgrp: the cgroup that we're moving into
1058 * @template: out param for the new set of csses, should be clear on entry
1060 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
1061 struct cgroup
*cgrp
,
1062 struct cgroup_subsys_state
*template[])
1064 struct cgroup_root
*root
= cgrp
->root
;
1065 struct cgroup_subsys
*ss
;
1066 struct css_set
*cset
;
1071 * Build the set of subsystem state objects that we want to see in the
1072 * new css_set. while subsystems can change globally, the entries here
1073 * won't change, so no need for locking.
1075 for_each_subsys(ss
, i
) {
1076 if (root
->subsys_mask
& (1UL << i
)) {
1078 * @ss is in this hierarchy, so we want the
1079 * effective css from @cgrp.
1081 template[i
] = cgroup_e_css_by_mask(cgrp
, ss
);
1084 * @ss is not in this hierarchy, so we don't want
1085 * to change the css.
1087 template[i
] = old_cset
->subsys
[i
];
1091 key
= css_set_hash(template);
1092 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
1093 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
1096 /* This css_set matches what we need */
1100 /* No existing cgroup group matched */
1104 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
1106 struct cgrp_cset_link
*link
, *tmp_link
;
1108 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
1109 list_del(&link
->cset_link
);
1115 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1116 * @count: the number of links to allocate
1117 * @tmp_links: list_head the allocated links are put on
1119 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1120 * through ->cset_link. Returns 0 on success or -errno.
1122 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1124 struct cgrp_cset_link
*link
;
1127 INIT_LIST_HEAD(tmp_links
);
1129 for (i
= 0; i
< count
; i
++) {
1130 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1132 free_cgrp_cset_links(tmp_links
);
1135 list_add(&link
->cset_link
, tmp_links
);
1141 * link_css_set - a helper function to link a css_set to a cgroup
1142 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1143 * @cset: the css_set to be linked
1144 * @cgrp: the destination cgroup
1146 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1147 struct cgroup
*cgrp
)
1149 struct cgrp_cset_link
*link
;
1151 BUG_ON(list_empty(tmp_links
));
1153 if (cgroup_on_dfl(cgrp
))
1154 cset
->dfl_cgrp
= cgrp
;
1156 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1161 * Always add links to the tail of the lists so that the lists are
1162 * in choronological order.
1164 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1165 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1167 if (cgroup_parent(cgrp
))
1168 cgroup_get_live(cgrp
);
1172 * find_css_set - return a new css_set with one cgroup updated
1173 * @old_cset: the baseline css_set
1174 * @cgrp: the cgroup to be updated
1176 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1177 * substituted into the appropriate hierarchy.
1179 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1180 struct cgroup
*cgrp
)
1182 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1183 struct css_set
*cset
;
1184 struct list_head tmp_links
;
1185 struct cgrp_cset_link
*link
;
1186 struct cgroup_subsys
*ss
;
1190 lockdep_assert_held(&cgroup_mutex
);
1192 /* First see if we already have a cgroup group that matches
1193 * the desired set */
1194 spin_lock_irq(&css_set_lock
);
1195 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1198 spin_unlock_irq(&css_set_lock
);
1203 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1207 /* Allocate all the cgrp_cset_link objects that we'll need */
1208 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1213 refcount_set(&cset
->refcount
, 1);
1214 cset
->dom_cset
= cset
;
1215 INIT_LIST_HEAD(&cset
->tasks
);
1216 INIT_LIST_HEAD(&cset
->mg_tasks
);
1217 INIT_LIST_HEAD(&cset
->dying_tasks
);
1218 INIT_LIST_HEAD(&cset
->task_iters
);
1219 INIT_LIST_HEAD(&cset
->threaded_csets
);
1220 INIT_HLIST_NODE(&cset
->hlist
);
1221 INIT_LIST_HEAD(&cset
->cgrp_links
);
1222 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1223 INIT_LIST_HEAD(&cset
->mg_node
);
1225 /* Copy the set of subsystem state objects generated in
1226 * find_existing_css_set() */
1227 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1229 spin_lock_irq(&css_set_lock
);
1230 /* Add reference counts and links from the new css_set. */
1231 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1232 struct cgroup
*c
= link
->cgrp
;
1234 if (c
->root
== cgrp
->root
)
1236 link_css_set(&tmp_links
, cset
, c
);
1239 BUG_ON(!list_empty(&tmp_links
));
1243 /* Add @cset to the hash table */
1244 key
= css_set_hash(cset
->subsys
);
1245 hash_add(css_set_table
, &cset
->hlist
, key
);
1247 for_each_subsys(ss
, ssid
) {
1248 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1250 list_add_tail(&cset
->e_cset_node
[ssid
],
1251 &css
->cgroup
->e_csets
[ssid
]);
1255 spin_unlock_irq(&css_set_lock
);
1258 * If @cset should be threaded, look up the matching dom_cset and
1259 * link them up. We first fully initialize @cset then look for the
1260 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1261 * to stay empty until we return.
1263 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1264 struct css_set
*dcset
;
1266 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1272 spin_lock_irq(&css_set_lock
);
1273 cset
->dom_cset
= dcset
;
1274 list_add_tail(&cset
->threaded_csets_node
,
1275 &dcset
->threaded_csets
);
1276 spin_unlock_irq(&css_set_lock
);
1282 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1284 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1286 return root_cgrp
->root
;
1289 static int cgroup_init_root_id(struct cgroup_root
*root
)
1293 lockdep_assert_held(&cgroup_mutex
);
1295 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1299 root
->hierarchy_id
= id
;
1303 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1305 lockdep_assert_held(&cgroup_mutex
);
1307 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1310 void cgroup_free_root(struct cgroup_root
*root
)
1313 idr_destroy(&root
->cgroup_idr
);
1318 static void cgroup_destroy_root(struct cgroup_root
*root
)
1320 struct cgroup
*cgrp
= &root
->cgrp
;
1321 struct cgrp_cset_link
*link
, *tmp_link
;
1323 trace_cgroup_destroy_root(root
);
1325 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1327 BUG_ON(atomic_read(&root
->nr_cgrps
));
1328 BUG_ON(!list_empty(&cgrp
->self
.children
));
1330 /* Rebind all subsystems back to the default hierarchy */
1331 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1334 * Release all the links from cset_links to this hierarchy's
1337 spin_lock_irq(&css_set_lock
);
1339 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1340 list_del(&link
->cset_link
);
1341 list_del(&link
->cgrp_link
);
1345 spin_unlock_irq(&css_set_lock
);
1347 if (!list_empty(&root
->root_list
)) {
1348 list_del(&root
->root_list
);
1349 cgroup_root_count
--;
1352 cgroup_exit_root_id(root
);
1354 mutex_unlock(&cgroup_mutex
);
1356 kernfs_destroy_root(root
->kf_root
);
1357 cgroup_free_root(root
);
1361 * look up cgroup associated with current task's cgroup namespace on the
1362 * specified hierarchy
1364 static struct cgroup
*
1365 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1367 struct cgroup
*res
= NULL
;
1368 struct css_set
*cset
;
1370 lockdep_assert_held(&css_set_lock
);
1374 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1375 if (cset
== &init_css_set
) {
1378 struct cgrp_cset_link
*link
;
1380 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1381 struct cgroup
*c
= link
->cgrp
;
1383 if (c
->root
== root
) {
1395 /* look up cgroup associated with given css_set on the specified hierarchy */
1396 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1397 struct cgroup_root
*root
)
1399 struct cgroup
*res
= NULL
;
1401 lockdep_assert_held(&cgroup_mutex
);
1402 lockdep_assert_held(&css_set_lock
);
1404 if (cset
== &init_css_set
) {
1406 } else if (root
== &cgrp_dfl_root
) {
1407 res
= cset
->dfl_cgrp
;
1409 struct cgrp_cset_link
*link
;
1411 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1412 struct cgroup
*c
= link
->cgrp
;
1414 if (c
->root
== root
) {
1426 * Return the cgroup for "task" from the given hierarchy. Must be
1427 * called with cgroup_mutex and css_set_lock held.
1429 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1430 struct cgroup_root
*root
)
1433 * No need to lock the task - since we hold cgroup_mutex the
1434 * task can't change groups, so the only thing that can happen
1435 * is that it exits and its css is set back to init_css_set.
1437 return cset_cgroup_from_root(task_css_set(task
), root
);
1441 * A task must hold cgroup_mutex to modify cgroups.
1443 * Any task can increment and decrement the count field without lock.
1444 * So in general, code holding cgroup_mutex can't rely on the count
1445 * field not changing. However, if the count goes to zero, then only
1446 * cgroup_attach_task() can increment it again. Because a count of zero
1447 * means that no tasks are currently attached, therefore there is no
1448 * way a task attached to that cgroup can fork (the other way to
1449 * increment the count). So code holding cgroup_mutex can safely
1450 * assume that if the count is zero, it will stay zero. Similarly, if
1451 * a task holds cgroup_mutex on a cgroup with zero count, it
1452 * knows that the cgroup won't be removed, as cgroup_rmdir()
1455 * A cgroup can only be deleted if both its 'count' of using tasks
1456 * is zero, and its list of 'children' cgroups is empty. Since all
1457 * tasks in the system use _some_ cgroup, and since there is always at
1458 * least one task in the system (init, pid == 1), therefore, root cgroup
1459 * always has either children cgroups and/or using tasks. So we don't
1460 * need a special hack to ensure that root cgroup cannot be deleted.
1462 * P.S. One more locking exception. RCU is used to guard the
1463 * update of a tasks cgroup pointer by cgroup_attach_task()
1466 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1468 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1471 struct cgroup_subsys
*ss
= cft
->ss
;
1473 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1474 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
1475 const char *dbg
= (cft
->flags
& CFTYPE_DEBUG
) ? ".__DEBUG__." : "";
1477 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s%s.%s",
1478 dbg
, cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1481 strscpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1487 * cgroup_file_mode - deduce file mode of a control file
1488 * @cft: the control file in question
1490 * S_IRUGO for read, S_IWUSR for write.
1492 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1496 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1499 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1500 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1510 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1511 * @subtree_control: the new subtree_control mask to consider
1512 * @this_ss_mask: available subsystems
1514 * On the default hierarchy, a subsystem may request other subsystems to be
1515 * enabled together through its ->depends_on mask. In such cases, more
1516 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1518 * This function calculates which subsystems need to be enabled if
1519 * @subtree_control is to be applied while restricted to @this_ss_mask.
1521 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1523 u16 cur_ss_mask
= subtree_control
;
1524 struct cgroup_subsys
*ss
;
1527 lockdep_assert_held(&cgroup_mutex
);
1529 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1532 u16 new_ss_mask
= cur_ss_mask
;
1534 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1535 new_ss_mask
|= ss
->depends_on
;
1536 } while_each_subsys_mask();
1539 * Mask out subsystems which aren't available. This can
1540 * happen only if some depended-upon subsystems were bound
1541 * to non-default hierarchies.
1543 new_ss_mask
&= this_ss_mask
;
1545 if (new_ss_mask
== cur_ss_mask
)
1547 cur_ss_mask
= new_ss_mask
;
1554 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1555 * @kn: the kernfs_node being serviced
1557 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1558 * the method finishes if locking succeeded. Note that once this function
1559 * returns the cgroup returned by cgroup_kn_lock_live() may become
1560 * inaccessible any time. If the caller intends to continue to access the
1561 * cgroup, it should pin it before invoking this function.
1563 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1565 struct cgroup
*cgrp
;
1567 if (kernfs_type(kn
) == KERNFS_DIR
)
1570 cgrp
= kn
->parent
->priv
;
1572 mutex_unlock(&cgroup_mutex
);
1574 kernfs_unbreak_active_protection(kn
);
1579 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1580 * @kn: the kernfs_node being serviced
1581 * @drain_offline: perform offline draining on the cgroup
1583 * This helper is to be used by a cgroup kernfs method currently servicing
1584 * @kn. It breaks the active protection, performs cgroup locking and
1585 * verifies that the associated cgroup is alive. Returns the cgroup if
1586 * alive; otherwise, %NULL. A successful return should be undone by a
1587 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1588 * cgroup is drained of offlining csses before return.
1590 * Any cgroup kernfs method implementation which requires locking the
1591 * associated cgroup should use this helper. It avoids nesting cgroup
1592 * locking under kernfs active protection and allows all kernfs operations
1593 * including self-removal.
1595 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1597 struct cgroup
*cgrp
;
1599 if (kernfs_type(kn
) == KERNFS_DIR
)
1602 cgrp
= kn
->parent
->priv
;
1605 * We're gonna grab cgroup_mutex which nests outside kernfs
1606 * active_ref. cgroup liveliness check alone provides enough
1607 * protection against removal. Ensure @cgrp stays accessible and
1608 * break the active_ref protection.
1610 if (!cgroup_tryget(cgrp
))
1612 kernfs_break_active_protection(kn
);
1615 cgroup_lock_and_drain_offline(cgrp
);
1617 mutex_lock(&cgroup_mutex
);
1619 if (!cgroup_is_dead(cgrp
))
1622 cgroup_kn_unlock(kn
);
1626 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1628 char name
[CGROUP_FILE_NAME_MAX
];
1630 lockdep_assert_held(&cgroup_mutex
);
1632 if (cft
->file_offset
) {
1633 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1634 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1636 spin_lock_irq(&cgroup_file_kn_lock
);
1638 spin_unlock_irq(&cgroup_file_kn_lock
);
1640 del_timer_sync(&cfile
->notify_timer
);
1643 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1647 * css_clear_dir - remove subsys files in a cgroup directory
1650 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1652 struct cgroup
*cgrp
= css
->cgroup
;
1653 struct cftype
*cfts
;
1655 if (!(css
->flags
& CSS_VISIBLE
))
1658 css
->flags
&= ~CSS_VISIBLE
;
1661 if (cgroup_on_dfl(cgrp
))
1662 cfts
= cgroup_base_files
;
1664 cfts
= cgroup1_base_files
;
1666 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1668 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1669 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1674 * css_populate_dir - create subsys files in a cgroup directory
1677 * On failure, no file is added.
1679 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1681 struct cgroup
*cgrp
= css
->cgroup
;
1682 struct cftype
*cfts
, *failed_cfts
;
1685 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1689 if (cgroup_on_dfl(cgrp
))
1690 cfts
= cgroup_base_files
;
1692 cfts
= cgroup1_base_files
;
1694 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1698 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1699 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1707 css
->flags
|= CSS_VISIBLE
;
1711 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1712 if (cfts
== failed_cfts
)
1714 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1719 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1721 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1722 struct cgroup_subsys
*ss
;
1725 lockdep_assert_held(&cgroup_mutex
);
1727 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1729 * If @ss has non-root csses attached to it, can't move.
1730 * If @ss is an implicit controller, it is exempt from this
1731 * rule and can be stolen.
1733 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1734 !ss
->implicit_on_dfl
)
1737 /* can't move between two non-dummy roots either */
1738 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1740 } while_each_subsys_mask();
1742 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1743 struct cgroup_root
*src_root
= ss
->root
;
1744 struct cgroup
*scgrp
= &src_root
->cgrp
;
1745 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1746 struct css_set
*cset
;
1748 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1750 /* disable from the source */
1751 src_root
->subsys_mask
&= ~(1 << ssid
);
1752 WARN_ON(cgroup_apply_control(scgrp
));
1753 cgroup_finalize_control(scgrp
, 0);
1756 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1757 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1758 ss
->root
= dst_root
;
1759 css
->cgroup
= dcgrp
;
1761 spin_lock_irq(&css_set_lock
);
1762 hash_for_each(css_set_table
, i
, cset
, hlist
)
1763 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1764 &dcgrp
->e_csets
[ss
->id
]);
1765 spin_unlock_irq(&css_set_lock
);
1767 /* default hierarchy doesn't enable controllers by default */
1768 dst_root
->subsys_mask
|= 1 << ssid
;
1769 if (dst_root
== &cgrp_dfl_root
) {
1770 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1772 dcgrp
->subtree_control
|= 1 << ssid
;
1773 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1776 ret
= cgroup_apply_control(dcgrp
);
1778 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1783 } while_each_subsys_mask();
1785 kernfs_activate(dcgrp
->kn
);
1789 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1790 struct kernfs_root
*kf_root
)
1794 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1795 struct cgroup
*ns_cgroup
;
1797 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1801 spin_lock_irq(&css_set_lock
);
1802 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1803 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1804 spin_unlock_irq(&css_set_lock
);
1806 if (len
>= PATH_MAX
)
1809 seq_escape(sf
, buf
, " \t\n\\");
1816 enum cgroup2_param
{
1818 Opt_memory_localevents
,
1822 static const struct fs_parameter_spec cgroup2_param_specs
[] = {
1823 fsparam_flag("nsdelegate", Opt_nsdelegate
),
1824 fsparam_flag("memory_localevents", Opt_memory_localevents
),
1828 static const struct fs_parameter_description cgroup2_fs_parameters
= {
1830 .specs
= cgroup2_param_specs
,
1833 static int cgroup2_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
1835 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1836 struct fs_parse_result result
;
1839 opt
= fs_parse(fc
, &cgroup2_fs_parameters
, param
, &result
);
1844 case Opt_nsdelegate
:
1845 ctx
->flags
|= CGRP_ROOT_NS_DELEGATE
;
1847 case Opt_memory_localevents
:
1848 ctx
->flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1854 static void apply_cgroup_root_flags(unsigned int root_flags
)
1856 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1857 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1858 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1860 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1862 if (root_flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1863 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1865 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1869 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1871 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1872 seq_puts(seq
, ",nsdelegate");
1873 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1874 seq_puts(seq
, ",memory_localevents");
1878 static int cgroup_reconfigure(struct fs_context
*fc
)
1880 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1882 apply_cgroup_root_flags(ctx
->flags
);
1887 * To reduce the fork() overhead for systems that are not actually using
1888 * their cgroups capability, we don't maintain the lists running through
1889 * each css_set to its tasks until we see the list actually used - in other
1890 * words after the first mount.
1892 static bool use_task_css_set_links __read_mostly
;
1894 static void cgroup_enable_task_cg_lists(void)
1896 struct task_struct
*p
, *g
;
1899 * We need tasklist_lock because RCU is not safe against
1900 * while_each_thread(). Besides, a forking task that has passed
1901 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1902 * is not guaranteed to have its child immediately visible in the
1903 * tasklist if we walk through it with RCU.
1905 read_lock(&tasklist_lock
);
1906 spin_lock_irq(&css_set_lock
);
1908 if (use_task_css_set_links
)
1911 use_task_css_set_links
= true;
1913 do_each_thread(g
, p
) {
1914 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1915 task_css_set(p
) != &init_css_set
);
1918 * We should check if the process is exiting, otherwise
1919 * it will race with cgroup_exit() in that the list
1920 * entry won't be deleted though the process has exited.
1921 * Do it while holding siglock so that we don't end up
1922 * racing against cgroup_exit().
1924 * Interrupts were already disabled while acquiring
1925 * the css_set_lock, so we do not need to disable it
1926 * again when acquiring the sighand->siglock here.
1928 spin_lock(&p
->sighand
->siglock
);
1929 if (!(p
->flags
& PF_EXITING
)) {
1930 struct css_set
*cset
= task_css_set(p
);
1932 if (!css_set_populated(cset
))
1933 css_set_update_populated(cset
, true);
1934 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1938 spin_unlock(&p
->sighand
->siglock
);
1939 } while_each_thread(g
, p
);
1941 spin_unlock_irq(&css_set_lock
);
1942 read_unlock(&tasklist_lock
);
1945 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1947 struct cgroup_subsys
*ss
;
1950 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1951 INIT_LIST_HEAD(&cgrp
->self
.children
);
1952 INIT_LIST_HEAD(&cgrp
->cset_links
);
1953 INIT_LIST_HEAD(&cgrp
->pidlists
);
1954 mutex_init(&cgrp
->pidlist_mutex
);
1955 cgrp
->self
.cgroup
= cgrp
;
1956 cgrp
->self
.flags
|= CSS_ONLINE
;
1957 cgrp
->dom_cgrp
= cgrp
;
1958 cgrp
->max_descendants
= INT_MAX
;
1959 cgrp
->max_depth
= INT_MAX
;
1960 INIT_LIST_HEAD(&cgrp
->rstat_css_list
);
1961 prev_cputime_init(&cgrp
->prev_cputime
);
1963 for_each_subsys(ss
, ssid
)
1964 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1966 init_waitqueue_head(&cgrp
->offline_waitq
);
1967 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1970 void init_cgroup_root(struct cgroup_fs_context
*ctx
)
1972 struct cgroup_root
*root
= ctx
->root
;
1973 struct cgroup
*cgrp
= &root
->cgrp
;
1975 INIT_LIST_HEAD(&root
->root_list
);
1976 atomic_set(&root
->nr_cgrps
, 1);
1978 init_cgroup_housekeeping(cgrp
);
1979 idr_init(&root
->cgroup_idr
);
1981 root
->flags
= ctx
->flags
;
1982 if (ctx
->release_agent
)
1983 strscpy(root
->release_agent_path
, ctx
->release_agent
, PATH_MAX
);
1985 strscpy(root
->name
, ctx
->name
, MAX_CGROUP_ROOT_NAMELEN
);
1986 if (ctx
->cpuset_clone_children
)
1987 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1990 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
1992 LIST_HEAD(tmp_links
);
1993 struct cgroup
*root_cgrp
= &root
->cgrp
;
1994 struct kernfs_syscall_ops
*kf_sops
;
1995 struct css_set
*cset
;
1998 lockdep_assert_held(&cgroup_mutex
);
2000 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
2003 root_cgrp
->id
= ret
;
2004 root_cgrp
->ancestor_ids
[0] = ret
;
2006 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
2012 * We're accessing css_set_count without locking css_set_lock here,
2013 * but that's OK - it can only be increased by someone holding
2014 * cgroup_lock, and that's us. Later rebinding may disable
2015 * controllers on the default hierarchy and thus create new csets,
2016 * which can't be more than the existing ones. Allocate 2x.
2018 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
2022 ret
= cgroup_init_root_id(root
);
2026 kf_sops
= root
== &cgrp_dfl_root
?
2027 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
2029 root
->kf_root
= kernfs_create_root(kf_sops
,
2030 KERNFS_ROOT_CREATE_DEACTIVATED
|
2031 KERNFS_ROOT_SUPPORT_EXPORTOP
,
2033 if (IS_ERR(root
->kf_root
)) {
2034 ret
= PTR_ERR(root
->kf_root
);
2037 root_cgrp
->kn
= root
->kf_root
->kn
;
2039 ret
= css_populate_dir(&root_cgrp
->self
);
2043 ret
= rebind_subsystems(root
, ss_mask
);
2047 ret
= cgroup_bpf_inherit(root_cgrp
);
2050 trace_cgroup_setup_root(root
);
2053 * There must be no failure case after here, since rebinding takes
2054 * care of subsystems' refcounts, which are explicitly dropped in
2055 * the failure exit path.
2057 list_add(&root
->root_list
, &cgroup_roots
);
2058 cgroup_root_count
++;
2061 * Link the root cgroup in this hierarchy into all the css_set
2064 spin_lock_irq(&css_set_lock
);
2065 hash_for_each(css_set_table
, i
, cset
, hlist
) {
2066 link_css_set(&tmp_links
, cset
, root_cgrp
);
2067 if (css_set_populated(cset
))
2068 cgroup_update_populated(root_cgrp
, true);
2070 spin_unlock_irq(&css_set_lock
);
2072 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2073 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2075 kernfs_activate(root_cgrp
->kn
);
2080 kernfs_destroy_root(root
->kf_root
);
2081 root
->kf_root
= NULL
;
2083 cgroup_exit_root_id(root
);
2085 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2087 free_cgrp_cset_links(&tmp_links
);
2091 int cgroup_do_get_tree(struct fs_context
*fc
)
2093 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2096 ctx
->kfc
.root
= ctx
->root
->kf_root
;
2097 if (fc
->fs_type
== &cgroup2_fs_type
)
2098 ctx
->kfc
.magic
= CGROUP2_SUPER_MAGIC
;
2100 ctx
->kfc
.magic
= CGROUP_SUPER_MAGIC
;
2101 ret
= kernfs_get_tree(fc
);
2104 * In non-init cgroup namespace, instead of root cgroup's dentry,
2105 * we return the dentry corresponding to the cgroupns->root_cgrp.
2107 if (!ret
&& ctx
->ns
!= &init_cgroup_ns
) {
2108 struct dentry
*nsdentry
;
2109 struct super_block
*sb
= fc
->root
->d_sb
;
2110 struct cgroup
*cgrp
;
2112 mutex_lock(&cgroup_mutex
);
2113 spin_lock_irq(&css_set_lock
);
2115 cgrp
= cset_cgroup_from_root(ctx
->ns
->root_cset
, ctx
->root
);
2117 spin_unlock_irq(&css_set_lock
);
2118 mutex_unlock(&cgroup_mutex
);
2120 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
2122 fc
->root
= nsdentry
;
2123 if (IS_ERR(nsdentry
)) {
2124 ret
= PTR_ERR(nsdentry
);
2125 deactivate_locked_super(sb
);
2129 if (!ctx
->kfc
.new_sb_created
)
2130 cgroup_put(&ctx
->root
->cgrp
);
2136 * Destroy a cgroup filesystem context.
2138 static void cgroup_fs_context_free(struct fs_context
*fc
)
2140 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2143 kfree(ctx
->release_agent
);
2144 put_cgroup_ns(ctx
->ns
);
2145 kernfs_free_fs_context(fc
);
2149 static int cgroup_get_tree(struct fs_context
*fc
)
2151 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2154 cgrp_dfl_visible
= true;
2155 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2156 ctx
->root
= &cgrp_dfl_root
;
2158 ret
= cgroup_do_get_tree(fc
);
2160 apply_cgroup_root_flags(ctx
->flags
);
2164 static const struct fs_context_operations cgroup_fs_context_ops
= {
2165 .free
= cgroup_fs_context_free
,
2166 .parse_param
= cgroup2_parse_param
,
2167 .get_tree
= cgroup_get_tree
,
2168 .reconfigure
= cgroup_reconfigure
,
2171 static const struct fs_context_operations cgroup1_fs_context_ops
= {
2172 .free
= cgroup_fs_context_free
,
2173 .parse_param
= cgroup1_parse_param
,
2174 .get_tree
= cgroup1_get_tree
,
2175 .reconfigure
= cgroup1_reconfigure
,
2179 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2180 * we select the namespace we're going to use.
2182 static int cgroup_init_fs_context(struct fs_context
*fc
)
2184 struct cgroup_fs_context
*ctx
;
2186 ctx
= kzalloc(sizeof(struct cgroup_fs_context
), GFP_KERNEL
);
2191 * The first time anyone tries to mount a cgroup, enable the list
2192 * linking each css_set to its tasks and fix up all existing tasks.
2194 if (!use_task_css_set_links
)
2195 cgroup_enable_task_cg_lists();
2197 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
2198 get_cgroup_ns(ctx
->ns
);
2199 fc
->fs_private
= &ctx
->kfc
;
2200 if (fc
->fs_type
== &cgroup2_fs_type
)
2201 fc
->ops
= &cgroup_fs_context_ops
;
2203 fc
->ops
= &cgroup1_fs_context_ops
;
2204 put_user_ns(fc
->user_ns
);
2205 fc
->user_ns
= get_user_ns(ctx
->ns
->user_ns
);
2210 static void cgroup_kill_sb(struct super_block
*sb
)
2212 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2213 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2216 * If @root doesn't have any children, start killing it.
2217 * This prevents new mounts by disabling percpu_ref_tryget_live().
2218 * cgroup_mount() may wait for @root's release.
2220 * And don't kill the default root.
2222 if (list_empty(&root
->cgrp
.self
.children
) && root
!= &cgrp_dfl_root
&&
2223 !percpu_ref_is_dying(&root
->cgrp
.self
.refcnt
))
2224 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2225 cgroup_put(&root
->cgrp
);
2229 struct file_system_type cgroup_fs_type
= {
2231 .init_fs_context
= cgroup_init_fs_context
,
2232 .parameters
= &cgroup1_fs_parameters
,
2233 .kill_sb
= cgroup_kill_sb
,
2234 .fs_flags
= FS_USERNS_MOUNT
,
2237 static struct file_system_type cgroup2_fs_type
= {
2239 .init_fs_context
= cgroup_init_fs_context
,
2240 .parameters
= &cgroup2_fs_parameters
,
2241 .kill_sb
= cgroup_kill_sb
,
2242 .fs_flags
= FS_USERNS_MOUNT
,
2245 #ifdef CONFIG_CPUSETS
2246 static const struct fs_context_operations cpuset_fs_context_ops
= {
2247 .get_tree
= cgroup1_get_tree
,
2248 .free
= cgroup_fs_context_free
,
2252 * This is ugly, but preserves the userspace API for existing cpuset
2253 * users. If someone tries to mount the "cpuset" filesystem, we
2254 * silently switch it to mount "cgroup" instead
2256 static int cpuset_init_fs_context(struct fs_context
*fc
)
2258 char *agent
= kstrdup("/sbin/cpuset_release_agent", GFP_USER
);
2259 struct cgroup_fs_context
*ctx
;
2262 err
= cgroup_init_fs_context(fc
);
2268 fc
->ops
= &cpuset_fs_context_ops
;
2270 ctx
= cgroup_fc2context(fc
);
2271 ctx
->subsys_mask
= 1 << cpuset_cgrp_id
;
2272 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
2273 ctx
->release_agent
= agent
;
2275 get_filesystem(&cgroup_fs_type
);
2276 put_filesystem(fc
->fs_type
);
2277 fc
->fs_type
= &cgroup_fs_type
;
2282 static struct file_system_type cpuset_fs_type
= {
2284 .init_fs_context
= cpuset_init_fs_context
,
2285 .fs_flags
= FS_USERNS_MOUNT
,
2289 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2290 struct cgroup_namespace
*ns
)
2292 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2294 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2297 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2298 struct cgroup_namespace
*ns
)
2302 mutex_lock(&cgroup_mutex
);
2303 spin_lock_irq(&css_set_lock
);
2305 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2307 spin_unlock_irq(&css_set_lock
);
2308 mutex_unlock(&cgroup_mutex
);
2312 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2315 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2316 * @task: target task
2317 * @buf: the buffer to write the path into
2318 * @buflen: the length of the buffer
2320 * Determine @task's cgroup on the first (the one with the lowest non-zero
2321 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2322 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2323 * cgroup controller callbacks.
2325 * Return value is the same as kernfs_path().
2327 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2329 struct cgroup_root
*root
;
2330 struct cgroup
*cgrp
;
2331 int hierarchy_id
= 1;
2334 mutex_lock(&cgroup_mutex
);
2335 spin_lock_irq(&css_set_lock
);
2337 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2340 cgrp
= task_cgroup_from_root(task
, root
);
2341 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2343 /* if no hierarchy exists, everyone is in "/" */
2344 ret
= strlcpy(buf
, "/", buflen
);
2347 spin_unlock_irq(&css_set_lock
);
2348 mutex_unlock(&cgroup_mutex
);
2351 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2354 * cgroup_migrate_add_task - add a migration target task to a migration context
2355 * @task: target task
2356 * @mgctx: target migration context
2358 * Add @task, which is a migration target, to @mgctx->tset. This function
2359 * becomes noop if @task doesn't need to be migrated. @task's css_set
2360 * should have been added as a migration source and @task->cg_list will be
2361 * moved from the css_set's tasks list to mg_tasks one.
2363 static void cgroup_migrate_add_task(struct task_struct
*task
,
2364 struct cgroup_mgctx
*mgctx
)
2366 struct css_set
*cset
;
2368 lockdep_assert_held(&css_set_lock
);
2370 /* @task either already exited or can't exit until the end */
2371 if (task
->flags
& PF_EXITING
)
2374 /* leave @task alone if post_fork() hasn't linked it yet */
2375 if (list_empty(&task
->cg_list
))
2378 cset
= task_css_set(task
);
2379 if (!cset
->mg_src_cgrp
)
2382 mgctx
->tset
.nr_tasks
++;
2384 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2385 if (list_empty(&cset
->mg_node
))
2386 list_add_tail(&cset
->mg_node
,
2387 &mgctx
->tset
.src_csets
);
2388 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2389 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2390 &mgctx
->tset
.dst_csets
);
2394 * cgroup_taskset_first - reset taskset and return the first task
2395 * @tset: taskset of interest
2396 * @dst_cssp: output variable for the destination css
2398 * @tset iteration is initialized and the first task is returned.
2400 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2401 struct cgroup_subsys_state
**dst_cssp
)
2403 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2404 tset
->cur_task
= NULL
;
2406 return cgroup_taskset_next(tset
, dst_cssp
);
2410 * cgroup_taskset_next - iterate to the next task in taskset
2411 * @tset: taskset of interest
2412 * @dst_cssp: output variable for the destination css
2414 * Return the next task in @tset. Iteration must have been initialized
2415 * with cgroup_taskset_first().
2417 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2418 struct cgroup_subsys_state
**dst_cssp
)
2420 struct css_set
*cset
= tset
->cur_cset
;
2421 struct task_struct
*task
= tset
->cur_task
;
2423 while (&cset
->mg_node
!= tset
->csets
) {
2425 task
= list_first_entry(&cset
->mg_tasks
,
2426 struct task_struct
, cg_list
);
2428 task
= list_next_entry(task
, cg_list
);
2430 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2431 tset
->cur_cset
= cset
;
2432 tset
->cur_task
= task
;
2435 * This function may be called both before and
2436 * after cgroup_taskset_migrate(). The two cases
2437 * can be distinguished by looking at whether @cset
2438 * has its ->mg_dst_cset set.
2440 if (cset
->mg_dst_cset
)
2441 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2443 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2448 cset
= list_next_entry(cset
, mg_node
);
2456 * cgroup_taskset_migrate - migrate a taskset
2457 * @mgctx: migration context
2459 * Migrate tasks in @mgctx as setup by migration preparation functions.
2460 * This function fails iff one of the ->can_attach callbacks fails and
2461 * guarantees that either all or none of the tasks in @mgctx are migrated.
2462 * @mgctx is consumed regardless of success.
2464 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2466 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2467 struct cgroup_subsys
*ss
;
2468 struct task_struct
*task
, *tmp_task
;
2469 struct css_set
*cset
, *tmp_cset
;
2470 int ssid
, failed_ssid
, ret
;
2472 /* check that we can legitimately attach to the cgroup */
2473 if (tset
->nr_tasks
) {
2474 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2475 if (ss
->can_attach
) {
2477 ret
= ss
->can_attach(tset
);
2480 goto out_cancel_attach
;
2483 } while_each_subsys_mask();
2487 * Now that we're guaranteed success, proceed to move all tasks to
2488 * the new cgroup. There are no failure cases after here, so this
2489 * is the commit point.
2491 spin_lock_irq(&css_set_lock
);
2492 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2493 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2494 struct css_set
*from_cset
= task_css_set(task
);
2495 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2497 get_css_set(to_cset
);
2498 to_cset
->nr_tasks
++;
2499 css_set_move_task(task
, from_cset
, to_cset
, true);
2500 from_cset
->nr_tasks
--;
2502 * If the source or destination cgroup is frozen,
2503 * the task might require to change its state.
2505 cgroup_freezer_migrate_task(task
, from_cset
->dfl_cgrp
,
2507 put_css_set_locked(from_cset
);
2511 spin_unlock_irq(&css_set_lock
);
2514 * Migration is committed, all target tasks are now on dst_csets.
2515 * Nothing is sensitive to fork() after this point. Notify
2516 * controllers that migration is complete.
2518 tset
->csets
= &tset
->dst_csets
;
2520 if (tset
->nr_tasks
) {
2521 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2526 } while_each_subsys_mask();
2530 goto out_release_tset
;
2533 if (tset
->nr_tasks
) {
2534 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2535 if (ssid
== failed_ssid
)
2537 if (ss
->cancel_attach
) {
2539 ss
->cancel_attach(tset
);
2541 } while_each_subsys_mask();
2544 spin_lock_irq(&css_set_lock
);
2545 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2546 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2547 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2548 list_del_init(&cset
->mg_node
);
2550 spin_unlock_irq(&css_set_lock
);
2553 * Re-initialize the cgroup_taskset structure in case it is reused
2554 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2558 tset
->csets
= &tset
->src_csets
;
2563 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2564 * @dst_cgrp: destination cgroup to test
2566 * On the default hierarchy, except for the mixable, (possible) thread root
2567 * and threaded cgroups, subtree_control must be zero for migration
2568 * destination cgroups with tasks so that child cgroups don't compete
2571 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2573 /* v1 doesn't have any restriction */
2574 if (!cgroup_on_dfl(dst_cgrp
))
2577 /* verify @dst_cgrp can host resources */
2578 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2581 /* mixables don't care */
2582 if (cgroup_is_mixable(dst_cgrp
))
2586 * If @dst_cgrp is already or can become a thread root or is
2587 * threaded, it doesn't matter.
2589 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2592 /* apply no-internal-process constraint */
2593 if (dst_cgrp
->subtree_control
)
2600 * cgroup_migrate_finish - cleanup after attach
2601 * @mgctx: migration context
2603 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2604 * those functions for details.
2606 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2608 LIST_HEAD(preloaded
);
2609 struct css_set
*cset
, *tmp_cset
;
2611 lockdep_assert_held(&cgroup_mutex
);
2613 spin_lock_irq(&css_set_lock
);
2615 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2616 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2618 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2619 cset
->mg_src_cgrp
= NULL
;
2620 cset
->mg_dst_cgrp
= NULL
;
2621 cset
->mg_dst_cset
= NULL
;
2622 list_del_init(&cset
->mg_preload_node
);
2623 put_css_set_locked(cset
);
2626 spin_unlock_irq(&css_set_lock
);
2630 * cgroup_migrate_add_src - add a migration source css_set
2631 * @src_cset: the source css_set to add
2632 * @dst_cgrp: the destination cgroup
2633 * @mgctx: migration context
2635 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2636 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2637 * up by cgroup_migrate_finish().
2639 * This function may be called without holding cgroup_threadgroup_rwsem
2640 * even if the target is a process. Threads may be created and destroyed
2641 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2642 * into play and the preloaded css_sets are guaranteed to cover all
2645 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2646 struct cgroup
*dst_cgrp
,
2647 struct cgroup_mgctx
*mgctx
)
2649 struct cgroup
*src_cgrp
;
2651 lockdep_assert_held(&cgroup_mutex
);
2652 lockdep_assert_held(&css_set_lock
);
2655 * If ->dead, @src_set is associated with one or more dead cgroups
2656 * and doesn't contain any migratable tasks. Ignore it early so
2657 * that the rest of migration path doesn't get confused by it.
2662 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2664 if (!list_empty(&src_cset
->mg_preload_node
))
2667 WARN_ON(src_cset
->mg_src_cgrp
);
2668 WARN_ON(src_cset
->mg_dst_cgrp
);
2669 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2670 WARN_ON(!list_empty(&src_cset
->mg_node
));
2672 src_cset
->mg_src_cgrp
= src_cgrp
;
2673 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2674 get_css_set(src_cset
);
2675 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2679 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2680 * @mgctx: migration context
2682 * Tasks are about to be moved and all the source css_sets have been
2683 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2684 * pins all destination css_sets, links each to its source, and append them
2685 * to @mgctx->preloaded_dst_csets.
2687 * This function must be called after cgroup_migrate_add_src() has been
2688 * called on each migration source css_set. After migration is performed
2689 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2692 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2694 struct css_set
*src_cset
, *tmp_cset
;
2696 lockdep_assert_held(&cgroup_mutex
);
2698 /* look up the dst cset for each src cset and link it to src */
2699 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2701 struct css_set
*dst_cset
;
2702 struct cgroup_subsys
*ss
;
2705 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2709 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2712 * If src cset equals dst, it's noop. Drop the src.
2713 * cgroup_migrate() will skip the cset too. Note that we
2714 * can't handle src == dst as some nodes are used by both.
2716 if (src_cset
== dst_cset
) {
2717 src_cset
->mg_src_cgrp
= NULL
;
2718 src_cset
->mg_dst_cgrp
= NULL
;
2719 list_del_init(&src_cset
->mg_preload_node
);
2720 put_css_set(src_cset
);
2721 put_css_set(dst_cset
);
2725 src_cset
->mg_dst_cset
= dst_cset
;
2727 if (list_empty(&dst_cset
->mg_preload_node
))
2728 list_add_tail(&dst_cset
->mg_preload_node
,
2729 &mgctx
->preloaded_dst_csets
);
2731 put_css_set(dst_cset
);
2733 for_each_subsys(ss
, ssid
)
2734 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2735 mgctx
->ss_mask
|= 1 << ssid
;
2742 * cgroup_migrate - migrate a process or task to a cgroup
2743 * @leader: the leader of the process or the task to migrate
2744 * @threadgroup: whether @leader points to the whole process or a single task
2745 * @mgctx: migration context
2747 * Migrate a process or task denoted by @leader. If migrating a process,
2748 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2749 * responsible for invoking cgroup_migrate_add_src() and
2750 * cgroup_migrate_prepare_dst() on the targets before invoking this
2751 * function and following up with cgroup_migrate_finish().
2753 * As long as a controller's ->can_attach() doesn't fail, this function is
2754 * guaranteed to succeed. This means that, excluding ->can_attach()
2755 * failure, when migrating multiple targets, the success or failure can be
2756 * decided for all targets by invoking group_migrate_prepare_dst() before
2757 * actually starting migrating.
2759 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2760 struct cgroup_mgctx
*mgctx
)
2762 struct task_struct
*task
;
2765 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2766 * already PF_EXITING could be freed from underneath us unless we
2767 * take an rcu_read_lock.
2769 spin_lock_irq(&css_set_lock
);
2773 cgroup_migrate_add_task(task
, mgctx
);
2776 } while_each_thread(leader
, task
);
2778 spin_unlock_irq(&css_set_lock
);
2780 return cgroup_migrate_execute(mgctx
);
2784 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2785 * @dst_cgrp: the cgroup to attach to
2786 * @leader: the task or the leader of the threadgroup to be attached
2787 * @threadgroup: attach the whole threadgroup?
2789 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2791 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2794 DEFINE_CGROUP_MGCTX(mgctx
);
2795 struct task_struct
*task
;
2798 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
2802 /* look up all src csets */
2803 spin_lock_irq(&css_set_lock
);
2807 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2810 } while_each_thread(leader
, task
);
2812 spin_unlock_irq(&css_set_lock
);
2814 /* prepare dst csets and commit */
2815 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2817 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2819 cgroup_migrate_finish(&mgctx
);
2822 TRACE_CGROUP_PATH(attach_task
, dst_cgrp
, leader
, threadgroup
);
2827 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
)
2828 __acquires(&cgroup_threadgroup_rwsem
)
2830 struct task_struct
*tsk
;
2833 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2834 return ERR_PTR(-EINVAL
);
2836 percpu_down_write(&cgroup_threadgroup_rwsem
);
2840 tsk
= find_task_by_vpid(pid
);
2842 tsk
= ERR_PTR(-ESRCH
);
2843 goto out_unlock_threadgroup
;
2850 tsk
= tsk
->group_leader
;
2853 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2854 * If userland migrates such a kthread to a non-root cgroup, it can
2855 * become trapped in a cpuset, or RT kthread may be born in a
2856 * cgroup with no rt_runtime allocated. Just say no.
2858 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2859 tsk
= ERR_PTR(-EINVAL
);
2860 goto out_unlock_threadgroup
;
2863 get_task_struct(tsk
);
2864 goto out_unlock_rcu
;
2866 out_unlock_threadgroup
:
2867 percpu_up_write(&cgroup_threadgroup_rwsem
);
2873 void cgroup_procs_write_finish(struct task_struct
*task
)
2874 __releases(&cgroup_threadgroup_rwsem
)
2876 struct cgroup_subsys
*ss
;
2879 /* release reference from cgroup_procs_write_start() */
2880 put_task_struct(task
);
2882 percpu_up_write(&cgroup_threadgroup_rwsem
);
2883 for_each_subsys(ss
, ssid
)
2884 if (ss
->post_attach
)
2888 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2890 struct cgroup_subsys
*ss
;
2891 bool printed
= false;
2894 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2897 seq_printf(seq
, "%s", ss
->name
);
2899 } while_each_subsys_mask();
2901 seq_putc(seq
, '\n');
2904 /* show controllers which are enabled from the parent */
2905 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2907 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2909 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2913 /* show controllers which are enabled for a given cgroup's children */
2914 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2916 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2918 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2923 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2924 * @cgrp: root of the subtree to update csses for
2926 * @cgrp's control masks have changed and its subtree's css associations
2927 * need to be updated accordingly. This function looks up all css_sets
2928 * which are attached to the subtree, creates the matching updated css_sets
2929 * and migrates the tasks to the new ones.
2931 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2933 DEFINE_CGROUP_MGCTX(mgctx
);
2934 struct cgroup_subsys_state
*d_css
;
2935 struct cgroup
*dsct
;
2936 struct css_set
*src_cset
;
2939 lockdep_assert_held(&cgroup_mutex
);
2941 percpu_down_write(&cgroup_threadgroup_rwsem
);
2943 /* look up all csses currently attached to @cgrp's subtree */
2944 spin_lock_irq(&css_set_lock
);
2945 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2946 struct cgrp_cset_link
*link
;
2948 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2949 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2951 spin_unlock_irq(&css_set_lock
);
2953 /* NULL dst indicates self on default hierarchy */
2954 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2958 spin_lock_irq(&css_set_lock
);
2959 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2960 struct task_struct
*task
, *ntask
;
2962 /* all tasks in src_csets need to be migrated */
2963 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2964 cgroup_migrate_add_task(task
, &mgctx
);
2966 spin_unlock_irq(&css_set_lock
);
2968 ret
= cgroup_migrate_execute(&mgctx
);
2970 cgroup_migrate_finish(&mgctx
);
2971 percpu_up_write(&cgroup_threadgroup_rwsem
);
2976 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2977 * @cgrp: root of the target subtree
2979 * Because css offlining is asynchronous, userland may try to re-enable a
2980 * controller while the previous css is still around. This function grabs
2981 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2983 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
2984 __acquires(&cgroup_mutex
)
2986 struct cgroup
*dsct
;
2987 struct cgroup_subsys_state
*d_css
;
2988 struct cgroup_subsys
*ss
;
2992 mutex_lock(&cgroup_mutex
);
2994 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2995 for_each_subsys(ss
, ssid
) {
2996 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2999 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
3002 cgroup_get_live(dsct
);
3003 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
3004 TASK_UNINTERRUPTIBLE
);
3006 mutex_unlock(&cgroup_mutex
);
3008 finish_wait(&dsct
->offline_waitq
, &wait
);
3017 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3018 * @cgrp: root of the target subtree
3020 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3021 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3024 static void cgroup_save_control(struct cgroup
*cgrp
)
3026 struct cgroup
*dsct
;
3027 struct cgroup_subsys_state
*d_css
;
3029 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3030 dsct
->old_subtree_control
= dsct
->subtree_control
;
3031 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
3032 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
3037 * cgroup_propagate_control - refresh control masks of a subtree
3038 * @cgrp: root of the target subtree
3040 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3041 * ->subtree_control and propagate controller availability through the
3042 * subtree so that descendants don't have unavailable controllers enabled.
3044 static void cgroup_propagate_control(struct cgroup
*cgrp
)
3046 struct cgroup
*dsct
;
3047 struct cgroup_subsys_state
*d_css
;
3049 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3050 dsct
->subtree_control
&= cgroup_control(dsct
);
3051 dsct
->subtree_ss_mask
=
3052 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
3053 cgroup_ss_mask(dsct
));
3058 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3059 * @cgrp: root of the target subtree
3061 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3062 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3065 static void cgroup_restore_control(struct cgroup
*cgrp
)
3067 struct cgroup
*dsct
;
3068 struct cgroup_subsys_state
*d_css
;
3070 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3071 dsct
->subtree_control
= dsct
->old_subtree_control
;
3072 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3073 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
3077 static bool css_visible(struct cgroup_subsys_state
*css
)
3079 struct cgroup_subsys
*ss
= css
->ss
;
3080 struct cgroup
*cgrp
= css
->cgroup
;
3082 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3084 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3086 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3090 * cgroup_apply_control_enable - enable or show csses according to control
3091 * @cgrp: root of the target subtree
3093 * Walk @cgrp's subtree and create new csses or make the existing ones
3094 * visible. A css is created invisible if it's being implicitly enabled
3095 * through dependency. An invisible css is made visible when the userland
3096 * explicitly enables it.
3098 * Returns 0 on success, -errno on failure. On failure, csses which have
3099 * been processed already aren't cleaned up. The caller is responsible for
3100 * cleaning up with cgroup_apply_control_disable().
3102 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3104 struct cgroup
*dsct
;
3105 struct cgroup_subsys_state
*d_css
;
3106 struct cgroup_subsys
*ss
;
3109 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3110 for_each_subsys(ss
, ssid
) {
3111 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3113 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
3115 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3119 css
= css_create(dsct
, ss
);
3121 return PTR_ERR(css
);
3124 if (css_visible(css
)) {
3125 ret
= css_populate_dir(css
);
3136 * cgroup_apply_control_disable - kill or hide csses according to control
3137 * @cgrp: root of the target subtree
3139 * Walk @cgrp's subtree and kill and hide csses so that they match
3140 * cgroup_ss_mask() and cgroup_visible_mask().
3142 * A css is hidden when the userland requests it to be disabled while other
3143 * subsystems are still depending on it. The css must not actively control
3144 * resources and be in the vanilla state if it's made visible again later.
3145 * Controllers which may be depended upon should provide ->css_reset() for
3148 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3150 struct cgroup
*dsct
;
3151 struct cgroup_subsys_state
*d_css
;
3152 struct cgroup_subsys
*ss
;
3155 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3156 for_each_subsys(ss
, ssid
) {
3157 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3159 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
3165 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3167 } else if (!css_visible(css
)) {
3177 * cgroup_apply_control - apply control mask updates to the subtree
3178 * @cgrp: root of the target subtree
3180 * subsystems can be enabled and disabled in a subtree using the following
3183 * 1. Call cgroup_save_control() to stash the current state.
3184 * 2. Update ->subtree_control masks in the subtree as desired.
3185 * 3. Call cgroup_apply_control() to apply the changes.
3186 * 4. Optionally perform other related operations.
3187 * 5. Call cgroup_finalize_control() to finish up.
3189 * This function implements step 3 and propagates the mask changes
3190 * throughout @cgrp's subtree, updates csses accordingly and perform
3191 * process migrations.
3193 static int cgroup_apply_control(struct cgroup
*cgrp
)
3197 cgroup_propagate_control(cgrp
);
3199 ret
= cgroup_apply_control_enable(cgrp
);
3204 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3205 * making the following cgroup_update_dfl_csses() properly update
3206 * css associations of all tasks in the subtree.
3208 ret
= cgroup_update_dfl_csses(cgrp
);
3216 * cgroup_finalize_control - finalize control mask update
3217 * @cgrp: root of the target subtree
3218 * @ret: the result of the update
3220 * Finalize control mask update. See cgroup_apply_control() for more info.
3222 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3225 cgroup_restore_control(cgrp
);
3226 cgroup_propagate_control(cgrp
);
3229 cgroup_apply_control_disable(cgrp
);
3232 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3234 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3236 /* if nothing is getting enabled, nothing to worry about */
3240 /* can @cgrp host any resources? */
3241 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3244 /* mixables don't care */
3245 if (cgroup_is_mixable(cgrp
))
3248 if (domain_enable
) {
3249 /* can't enable domain controllers inside a thread subtree */
3250 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3254 * Threaded controllers can handle internal competitions
3255 * and are always allowed inside a (prospective) thread
3258 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3263 * Controllers can't be enabled for a cgroup with tasks to avoid
3264 * child cgroups competing against tasks.
3266 if (cgroup_has_tasks(cgrp
))
3272 /* change the enabled child controllers for a cgroup in the default hierarchy */
3273 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3274 char *buf
, size_t nbytes
,
3277 u16 enable
= 0, disable
= 0;
3278 struct cgroup
*cgrp
, *child
;
3279 struct cgroup_subsys
*ss
;
3284 * Parse input - space separated list of subsystem names prefixed
3285 * with either + or -.
3287 buf
= strstrip(buf
);
3288 while ((tok
= strsep(&buf
, " "))) {
3291 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3292 if (!cgroup_ssid_enabled(ssid
) ||
3293 strcmp(tok
+ 1, ss
->name
))
3297 enable
|= 1 << ssid
;
3298 disable
&= ~(1 << ssid
);
3299 } else if (*tok
== '-') {
3300 disable
|= 1 << ssid
;
3301 enable
&= ~(1 << ssid
);
3306 } while_each_subsys_mask();
3307 if (ssid
== CGROUP_SUBSYS_COUNT
)
3311 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3315 for_each_subsys(ss
, ssid
) {
3316 if (enable
& (1 << ssid
)) {
3317 if (cgrp
->subtree_control
& (1 << ssid
)) {
3318 enable
&= ~(1 << ssid
);
3322 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3326 } else if (disable
& (1 << ssid
)) {
3327 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3328 disable
&= ~(1 << ssid
);
3332 /* a child has it enabled? */
3333 cgroup_for_each_live_child(child
, cgrp
) {
3334 if (child
->subtree_control
& (1 << ssid
)) {
3342 if (!enable
&& !disable
) {
3347 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3351 /* save and update control masks and prepare csses */
3352 cgroup_save_control(cgrp
);
3354 cgrp
->subtree_control
|= enable
;
3355 cgrp
->subtree_control
&= ~disable
;
3357 ret
= cgroup_apply_control(cgrp
);
3358 cgroup_finalize_control(cgrp
, ret
);
3362 kernfs_activate(cgrp
->kn
);
3364 cgroup_kn_unlock(of
->kn
);
3365 return ret
?: nbytes
;
3369 * cgroup_enable_threaded - make @cgrp threaded
3370 * @cgrp: the target cgroup
3372 * Called when "threaded" is written to the cgroup.type interface file and
3373 * tries to make @cgrp threaded and join the parent's resource domain.
3374 * This function is never called on the root cgroup as cgroup.type doesn't
3377 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3379 struct cgroup
*parent
= cgroup_parent(cgrp
);
3380 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3381 struct cgroup
*dsct
;
3382 struct cgroup_subsys_state
*d_css
;
3385 lockdep_assert_held(&cgroup_mutex
);
3387 /* noop if already threaded */
3388 if (cgroup_is_threaded(cgrp
))
3392 * If @cgroup is populated or has domain controllers enabled, it
3393 * can't be switched. While the below cgroup_can_be_thread_root()
3394 * test can catch the same conditions, that's only when @parent is
3395 * not mixable, so let's check it explicitly.
3397 if (cgroup_is_populated(cgrp
) ||
3398 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3401 /* we're joining the parent's domain, ensure its validity */
3402 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3403 !cgroup_can_be_thread_root(dom_cgrp
))
3407 * The following shouldn't cause actual migrations and should
3410 cgroup_save_control(cgrp
);
3412 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3413 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3414 dsct
->dom_cgrp
= dom_cgrp
;
3416 ret
= cgroup_apply_control(cgrp
);
3418 parent
->nr_threaded_children
++;
3420 cgroup_finalize_control(cgrp
, ret
);
3424 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3426 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3428 if (cgroup_is_threaded(cgrp
))
3429 seq_puts(seq
, "threaded\n");
3430 else if (!cgroup_is_valid_domain(cgrp
))
3431 seq_puts(seq
, "domain invalid\n");
3432 else if (cgroup_is_thread_root(cgrp
))
3433 seq_puts(seq
, "domain threaded\n");
3435 seq_puts(seq
, "domain\n");
3440 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3441 size_t nbytes
, loff_t off
)
3443 struct cgroup
*cgrp
;
3446 /* only switching to threaded mode is supported */
3447 if (strcmp(strstrip(buf
), "threaded"))
3450 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3454 /* threaded can only be enabled */
3455 ret
= cgroup_enable_threaded(cgrp
);
3457 cgroup_kn_unlock(of
->kn
);
3458 return ret
?: nbytes
;
3461 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3463 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3464 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3466 if (descendants
== INT_MAX
)
3467 seq_puts(seq
, "max\n");
3469 seq_printf(seq
, "%d\n", descendants
);
3474 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3475 char *buf
, size_t nbytes
, loff_t off
)
3477 struct cgroup
*cgrp
;
3481 buf
= strstrip(buf
);
3482 if (!strcmp(buf
, "max")) {
3483 descendants
= INT_MAX
;
3485 ret
= kstrtoint(buf
, 0, &descendants
);
3490 if (descendants
< 0)
3493 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3497 cgrp
->max_descendants
= descendants
;
3499 cgroup_kn_unlock(of
->kn
);
3504 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3506 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3507 int depth
= READ_ONCE(cgrp
->max_depth
);
3509 if (depth
== INT_MAX
)
3510 seq_puts(seq
, "max\n");
3512 seq_printf(seq
, "%d\n", depth
);
3517 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3518 char *buf
, size_t nbytes
, loff_t off
)
3520 struct cgroup
*cgrp
;
3524 buf
= strstrip(buf
);
3525 if (!strcmp(buf
, "max")) {
3528 ret
= kstrtoint(buf
, 0, &depth
);
3536 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3540 cgrp
->max_depth
= depth
;
3542 cgroup_kn_unlock(of
->kn
);
3547 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3549 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3551 seq_printf(seq
, "populated %d\n", cgroup_is_populated(cgrp
));
3552 seq_printf(seq
, "frozen %d\n", test_bit(CGRP_FROZEN
, &cgrp
->flags
));
3557 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3559 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3561 seq_printf(seq
, "nr_descendants %d\n",
3562 cgroup
->nr_descendants
);
3563 seq_printf(seq
, "nr_dying_descendants %d\n",
3564 cgroup
->nr_dying_descendants
);
3569 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3570 struct cgroup
*cgrp
, int ssid
)
3572 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3573 struct cgroup_subsys_state
*css
;
3576 if (!ss
->css_extra_stat_show
)
3579 css
= cgroup_tryget_css(cgrp
, ss
);
3583 ret
= ss
->css_extra_stat_show(seq
, css
);
3588 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3590 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3593 cgroup_base_stat_cputime_show(seq
);
3594 #ifdef CONFIG_CGROUP_SCHED
3595 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3601 static int cgroup_io_pressure_show(struct seq_file
*seq
, void *v
)
3603 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3604 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3606 return psi_show(seq
, psi
, PSI_IO
);
3608 static int cgroup_memory_pressure_show(struct seq_file
*seq
, void *v
)
3610 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3611 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3613 return psi_show(seq
, psi
, PSI_MEM
);
3615 static int cgroup_cpu_pressure_show(struct seq_file
*seq
, void *v
)
3617 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3618 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3620 return psi_show(seq
, psi
, PSI_CPU
);
3623 static ssize_t
cgroup_pressure_write(struct kernfs_open_file
*of
, char *buf
,
3624 size_t nbytes
, enum psi_res res
)
3626 struct psi_trigger
*new;
3627 struct cgroup
*cgrp
;
3629 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3634 cgroup_kn_unlock(of
->kn
);
3636 new = psi_trigger_create(&cgrp
->psi
, buf
, nbytes
, res
);
3639 return PTR_ERR(new);
3642 psi_trigger_replace(&of
->priv
, new);
3649 static ssize_t
cgroup_io_pressure_write(struct kernfs_open_file
*of
,
3650 char *buf
, size_t nbytes
,
3653 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_IO
);
3656 static ssize_t
cgroup_memory_pressure_write(struct kernfs_open_file
*of
,
3657 char *buf
, size_t nbytes
,
3660 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_MEM
);
3663 static ssize_t
cgroup_cpu_pressure_write(struct kernfs_open_file
*of
,
3664 char *buf
, size_t nbytes
,
3667 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_CPU
);
3670 static __poll_t
cgroup_pressure_poll(struct kernfs_open_file
*of
,
3673 return psi_trigger_poll(&of
->priv
, of
->file
, pt
);
3676 static void cgroup_pressure_release(struct kernfs_open_file
*of
)
3678 psi_trigger_replace(&of
->priv
, NULL
);
3680 #endif /* CONFIG_PSI */
3682 static int cgroup_freeze_show(struct seq_file
*seq
, void *v
)
3684 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3686 seq_printf(seq
, "%d\n", cgrp
->freezer
.freeze
);
3691 static ssize_t
cgroup_freeze_write(struct kernfs_open_file
*of
,
3692 char *buf
, size_t nbytes
, loff_t off
)
3694 struct cgroup
*cgrp
;
3698 ret
= kstrtoint(strstrip(buf
), 0, &freeze
);
3702 if (freeze
< 0 || freeze
> 1)
3705 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3709 cgroup_freeze(cgrp
, freeze
);
3711 cgroup_kn_unlock(of
->kn
);
3716 static int cgroup_file_open(struct kernfs_open_file
*of
)
3718 struct cftype
*cft
= of
->kn
->priv
;
3721 return cft
->open(of
);
3725 static void cgroup_file_release(struct kernfs_open_file
*of
)
3727 struct cftype
*cft
= of
->kn
->priv
;
3733 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3734 size_t nbytes
, loff_t off
)
3736 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3737 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3738 struct cftype
*cft
= of
->kn
->priv
;
3739 struct cgroup_subsys_state
*css
;
3743 * If namespaces are delegation boundaries, disallow writes to
3744 * files in an non-init namespace root from inside the namespace
3745 * except for the files explicitly marked delegatable -
3746 * cgroup.procs and cgroup.subtree_control.
3748 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3749 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3750 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3754 return cft
->write(of
, buf
, nbytes
, off
);
3757 * kernfs guarantees that a file isn't deleted with operations in
3758 * flight, which means that the matching css is and stays alive and
3759 * doesn't need to be pinned. The RCU locking is not necessary
3760 * either. It's just for the convenience of using cgroup_css().
3763 css
= cgroup_css(cgrp
, cft
->ss
);
3766 if (cft
->write_u64
) {
3767 unsigned long long v
;
3768 ret
= kstrtoull(buf
, 0, &v
);
3770 ret
= cft
->write_u64(css
, cft
, v
);
3771 } else if (cft
->write_s64
) {
3773 ret
= kstrtoll(buf
, 0, &v
);
3775 ret
= cft
->write_s64(css
, cft
, v
);
3780 return ret
?: nbytes
;
3783 static __poll_t
cgroup_file_poll(struct kernfs_open_file
*of
, poll_table
*pt
)
3785 struct cftype
*cft
= of
->kn
->priv
;
3788 return cft
->poll(of
, pt
);
3790 return kernfs_generic_poll(of
, pt
);
3793 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3795 return seq_cft(seq
)->seq_start(seq
, ppos
);
3798 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3800 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3803 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3805 if (seq_cft(seq
)->seq_stop
)
3806 seq_cft(seq
)->seq_stop(seq
, v
);
3809 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3811 struct cftype
*cft
= seq_cft(m
);
3812 struct cgroup_subsys_state
*css
= seq_css(m
);
3815 return cft
->seq_show(m
, arg
);
3818 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3819 else if (cft
->read_s64
)
3820 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3826 static struct kernfs_ops cgroup_kf_single_ops
= {
3827 .atomic_write_len
= PAGE_SIZE
,
3828 .open
= cgroup_file_open
,
3829 .release
= cgroup_file_release
,
3830 .write
= cgroup_file_write
,
3831 .poll
= cgroup_file_poll
,
3832 .seq_show
= cgroup_seqfile_show
,
3835 static struct kernfs_ops cgroup_kf_ops
= {
3836 .atomic_write_len
= PAGE_SIZE
,
3837 .open
= cgroup_file_open
,
3838 .release
= cgroup_file_release
,
3839 .write
= cgroup_file_write
,
3840 .poll
= cgroup_file_poll
,
3841 .seq_start
= cgroup_seqfile_start
,
3842 .seq_next
= cgroup_seqfile_next
,
3843 .seq_stop
= cgroup_seqfile_stop
,
3844 .seq_show
= cgroup_seqfile_show
,
3847 /* set uid and gid of cgroup dirs and files to that of the creator */
3848 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3850 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3851 .ia_uid
= current_fsuid(),
3852 .ia_gid
= current_fsgid(), };
3854 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3855 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3858 return kernfs_setattr(kn
, &iattr
);
3861 static void cgroup_file_notify_timer(struct timer_list
*timer
)
3863 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
3867 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3870 char name
[CGROUP_FILE_NAME_MAX
];
3871 struct kernfs_node
*kn
;
3872 struct lock_class_key
*key
= NULL
;
3875 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3876 key
= &cft
->lockdep_key
;
3878 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3879 cgroup_file_mode(cft
),
3880 GLOBAL_ROOT_UID
, GLOBAL_ROOT_GID
,
3881 0, cft
->kf_ops
, cft
,
3886 ret
= cgroup_kn_set_ugid(kn
);
3892 if (cft
->file_offset
) {
3893 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3895 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
3897 spin_lock_irq(&cgroup_file_kn_lock
);
3899 spin_unlock_irq(&cgroup_file_kn_lock
);
3906 * cgroup_addrm_files - add or remove files to a cgroup directory
3907 * @css: the target css
3908 * @cgrp: the target cgroup (usually css->cgroup)
3909 * @cfts: array of cftypes to be added
3910 * @is_add: whether to add or remove
3912 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3913 * For removals, this function never fails.
3915 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3916 struct cgroup
*cgrp
, struct cftype cfts
[],
3919 struct cftype
*cft
, *cft_end
= NULL
;
3922 lockdep_assert_held(&cgroup_mutex
);
3925 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3926 /* does cft->flags tell us to skip this file on @cgrp? */
3927 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3929 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3931 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3933 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3935 if ((cft
->flags
& CFTYPE_DEBUG
) && !cgroup_debug
)
3938 ret
= cgroup_add_file(css
, cgrp
, cft
);
3940 pr_warn("%s: failed to add %s, err=%d\n",
3941 __func__
, cft
->name
, ret
);
3947 cgroup_rm_file(cgrp
, cft
);
3953 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3955 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3956 struct cgroup
*root
= &ss
->root
->cgrp
;
3957 struct cgroup_subsys_state
*css
;
3960 lockdep_assert_held(&cgroup_mutex
);
3962 /* add/rm files for all cgroups created before */
3963 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3964 struct cgroup
*cgrp
= css
->cgroup
;
3966 if (!(css
->flags
& CSS_VISIBLE
))
3969 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3975 kernfs_activate(root
->kn
);
3979 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3983 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3984 /* free copy for custom atomic_write_len, see init_cftypes() */
3985 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3990 /* revert flags set by cgroup core while adding @cfts */
3991 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3995 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3999 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4000 struct kernfs_ops
*kf_ops
;
4002 WARN_ON(cft
->ss
|| cft
->kf_ops
);
4005 kf_ops
= &cgroup_kf_ops
;
4007 kf_ops
= &cgroup_kf_single_ops
;
4010 * Ugh... if @cft wants a custom max_write_len, we need to
4011 * make a copy of kf_ops to set its atomic_write_len.
4013 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
4014 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
4016 cgroup_exit_cftypes(cfts
);
4019 kf_ops
->atomic_write_len
= cft
->max_write_len
;
4022 cft
->kf_ops
= kf_ops
;
4029 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
4031 lockdep_assert_held(&cgroup_mutex
);
4033 if (!cfts
|| !cfts
[0].ss
)
4036 list_del(&cfts
->node
);
4037 cgroup_apply_cftypes(cfts
, false);
4038 cgroup_exit_cftypes(cfts
);
4043 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4044 * @cfts: zero-length name terminated array of cftypes
4046 * Unregister @cfts. Files described by @cfts are removed from all
4047 * existing cgroups and all future cgroups won't have them either. This
4048 * function can be called anytime whether @cfts' subsys is attached or not.
4050 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4053 int cgroup_rm_cftypes(struct cftype
*cfts
)
4057 mutex_lock(&cgroup_mutex
);
4058 ret
= cgroup_rm_cftypes_locked(cfts
);
4059 mutex_unlock(&cgroup_mutex
);
4064 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4065 * @ss: target cgroup subsystem
4066 * @cfts: zero-length name terminated array of cftypes
4068 * Register @cfts to @ss. Files described by @cfts are created for all
4069 * existing cgroups to which @ss is attached and all future cgroups will
4070 * have them too. This function can be called anytime whether @ss is
4073 * Returns 0 on successful registration, -errno on failure. Note that this
4074 * function currently returns 0 as long as @cfts registration is successful
4075 * even if some file creation attempts on existing cgroups fail.
4077 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4081 if (!cgroup_ssid_enabled(ss
->id
))
4084 if (!cfts
|| cfts
[0].name
[0] == '\0')
4087 ret
= cgroup_init_cftypes(ss
, cfts
);
4091 mutex_lock(&cgroup_mutex
);
4093 list_add_tail(&cfts
->node
, &ss
->cfts
);
4094 ret
= cgroup_apply_cftypes(cfts
, true);
4096 cgroup_rm_cftypes_locked(cfts
);
4098 mutex_unlock(&cgroup_mutex
);
4103 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4104 * @ss: target cgroup subsystem
4105 * @cfts: zero-length name terminated array of cftypes
4107 * Similar to cgroup_add_cftypes() but the added files are only used for
4108 * the default hierarchy.
4110 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4114 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4115 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
4116 return cgroup_add_cftypes(ss
, cfts
);
4120 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4121 * @ss: target cgroup subsystem
4122 * @cfts: zero-length name terminated array of cftypes
4124 * Similar to cgroup_add_cftypes() but the added files are only used for
4125 * the legacy hierarchies.
4127 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4131 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4132 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
4133 return cgroup_add_cftypes(ss
, cfts
);
4137 * cgroup_file_notify - generate a file modified event for a cgroup_file
4138 * @cfile: target cgroup_file
4140 * @cfile must have been obtained by setting cftype->file_offset.
4142 void cgroup_file_notify(struct cgroup_file
*cfile
)
4144 unsigned long flags
;
4146 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
4148 unsigned long last
= cfile
->notified_at
;
4149 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
4151 if (time_in_range(jiffies
, last
, next
)) {
4152 timer_reduce(&cfile
->notify_timer
, next
);
4154 kernfs_notify(cfile
->kn
);
4155 cfile
->notified_at
= jiffies
;
4158 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
4162 * css_next_child - find the next child of a given css
4163 * @pos: the current position (%NULL to initiate traversal)
4164 * @parent: css whose children to walk
4166 * This function returns the next child of @parent and should be called
4167 * under either cgroup_mutex or RCU read lock. The only requirement is
4168 * that @parent and @pos are accessible. The next sibling is guaranteed to
4169 * be returned regardless of their states.
4171 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4172 * css which finished ->css_online() is guaranteed to be visible in the
4173 * future iterations and will stay visible until the last reference is put.
4174 * A css which hasn't finished ->css_online() or already finished
4175 * ->css_offline() may show up during traversal. It's each subsystem's
4176 * responsibility to synchronize against on/offlining.
4178 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
4179 struct cgroup_subsys_state
*parent
)
4181 struct cgroup_subsys_state
*next
;
4183 cgroup_assert_mutex_or_rcu_locked();
4186 * @pos could already have been unlinked from the sibling list.
4187 * Once a cgroup is removed, its ->sibling.next is no longer
4188 * updated when its next sibling changes. CSS_RELEASED is set when
4189 * @pos is taken off list, at which time its next pointer is valid,
4190 * and, as releases are serialized, the one pointed to by the next
4191 * pointer is guaranteed to not have started release yet. This
4192 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4193 * critical section, the one pointed to by its next pointer is
4194 * guaranteed to not have finished its RCU grace period even if we
4195 * have dropped rcu_read_lock() inbetween iterations.
4197 * If @pos has CSS_RELEASED set, its next pointer can't be
4198 * dereferenced; however, as each css is given a monotonically
4199 * increasing unique serial number and always appended to the
4200 * sibling list, the next one can be found by walking the parent's
4201 * children until the first css with higher serial number than
4202 * @pos's. While this path can be slower, it happens iff iteration
4203 * races against release and the race window is very small.
4206 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
4207 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
4208 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
4210 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
4211 if (next
->serial_nr
> pos
->serial_nr
)
4216 * @next, if not pointing to the head, can be dereferenced and is
4219 if (&next
->sibling
!= &parent
->children
)
4225 * css_next_descendant_pre - find the next descendant for pre-order walk
4226 * @pos: the current position (%NULL to initiate traversal)
4227 * @root: css whose descendants to walk
4229 * To be used by css_for_each_descendant_pre(). Find the next descendant
4230 * to visit for pre-order traversal of @root's descendants. @root is
4231 * included in the iteration and the first node to be visited.
4233 * While this function requires cgroup_mutex or RCU read locking, it
4234 * doesn't require the whole traversal to be contained in a single critical
4235 * section. This function will return the correct next descendant as long
4236 * as both @pos and @root are accessible and @pos is a descendant of @root.
4238 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4239 * css which finished ->css_online() is guaranteed to be visible in the
4240 * future iterations and will stay visible until the last reference is put.
4241 * A css which hasn't finished ->css_online() or already finished
4242 * ->css_offline() may show up during traversal. It's each subsystem's
4243 * responsibility to synchronize against on/offlining.
4245 struct cgroup_subsys_state
*
4246 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4247 struct cgroup_subsys_state
*root
)
4249 struct cgroup_subsys_state
*next
;
4251 cgroup_assert_mutex_or_rcu_locked();
4253 /* if first iteration, visit @root */
4257 /* visit the first child if exists */
4258 next
= css_next_child(NULL
, pos
);
4262 /* no child, visit my or the closest ancestor's next sibling */
4263 while (pos
!= root
) {
4264 next
= css_next_child(pos
, pos
->parent
);
4272 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
4275 * css_rightmost_descendant - return the rightmost descendant of a css
4276 * @pos: css of interest
4278 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4279 * is returned. This can be used during pre-order traversal to skip
4282 * While this function requires cgroup_mutex or RCU read locking, it
4283 * doesn't require the whole traversal to be contained in a single critical
4284 * section. This function will return the correct rightmost descendant as
4285 * long as @pos is accessible.
4287 struct cgroup_subsys_state
*
4288 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4290 struct cgroup_subsys_state
*last
, *tmp
;
4292 cgroup_assert_mutex_or_rcu_locked();
4296 /* ->prev isn't RCU safe, walk ->next till the end */
4298 css_for_each_child(tmp
, last
)
4305 static struct cgroup_subsys_state
*
4306 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4308 struct cgroup_subsys_state
*last
;
4312 pos
= css_next_child(NULL
, pos
);
4319 * css_next_descendant_post - find the next descendant for post-order walk
4320 * @pos: the current position (%NULL to initiate traversal)
4321 * @root: css whose descendants to walk
4323 * To be used by css_for_each_descendant_post(). Find the next descendant
4324 * to visit for post-order traversal of @root's descendants. @root is
4325 * included in the iteration and the last node to be visited.
4327 * While this function requires cgroup_mutex or RCU read locking, it
4328 * doesn't require the whole traversal to be contained in a single critical
4329 * section. This function will return the correct next descendant as long
4330 * as both @pos and @cgroup are accessible and @pos is a descendant of
4333 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4334 * css which finished ->css_online() is guaranteed to be visible in the
4335 * future iterations and will stay visible until the last reference is put.
4336 * A css which hasn't finished ->css_online() or already finished
4337 * ->css_offline() may show up during traversal. It's each subsystem's
4338 * responsibility to synchronize against on/offlining.
4340 struct cgroup_subsys_state
*
4341 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4342 struct cgroup_subsys_state
*root
)
4344 struct cgroup_subsys_state
*next
;
4346 cgroup_assert_mutex_or_rcu_locked();
4348 /* if first iteration, visit leftmost descendant which may be @root */
4350 return css_leftmost_descendant(root
);
4352 /* if we visited @root, we're done */
4356 /* if there's an unvisited sibling, visit its leftmost descendant */
4357 next
= css_next_child(pos
, pos
->parent
);
4359 return css_leftmost_descendant(next
);
4361 /* no sibling left, visit parent */
4366 * css_has_online_children - does a css have online children
4367 * @css: the target css
4369 * Returns %true if @css has any online children; otherwise, %false. This
4370 * function can be called from any context but the caller is responsible
4371 * for synchronizing against on/offlining as necessary.
4373 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4375 struct cgroup_subsys_state
*child
;
4379 css_for_each_child(child
, css
) {
4380 if (child
->flags
& CSS_ONLINE
) {
4389 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4391 struct list_head
*l
;
4392 struct cgrp_cset_link
*link
;
4393 struct css_set
*cset
;
4395 lockdep_assert_held(&css_set_lock
);
4397 /* find the next threaded cset */
4398 if (it
->tcset_pos
) {
4399 l
= it
->tcset_pos
->next
;
4401 if (l
!= it
->tcset_head
) {
4403 return container_of(l
, struct css_set
,
4404 threaded_csets_node
);
4407 it
->tcset_pos
= NULL
;
4410 /* find the next cset */
4413 if (l
== it
->cset_head
) {
4414 it
->cset_pos
= NULL
;
4419 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4421 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4427 /* initialize threaded css_set walking */
4428 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4430 put_css_set_locked(it
->cur_dcset
);
4431 it
->cur_dcset
= cset
;
4434 it
->tcset_head
= &cset
->threaded_csets
;
4435 it
->tcset_pos
= &cset
->threaded_csets
;
4442 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4443 * @it: the iterator to advance
4445 * Advance @it to the next css_set to walk.
4447 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4449 struct css_set
*cset
;
4451 lockdep_assert_held(&css_set_lock
);
4453 /* Advance to the next non-empty css_set */
4455 cset
= css_task_iter_next_css_set(it
);
4457 it
->task_pos
= NULL
;
4460 } while (!css_set_populated(cset
) && list_empty(&cset
->dying_tasks
));
4462 if (!list_empty(&cset
->tasks
))
4463 it
->task_pos
= cset
->tasks
.next
;
4464 else if (!list_empty(&cset
->mg_tasks
))
4465 it
->task_pos
= cset
->mg_tasks
.next
;
4467 it
->task_pos
= cset
->dying_tasks
.next
;
4469 it
->tasks_head
= &cset
->tasks
;
4470 it
->mg_tasks_head
= &cset
->mg_tasks
;
4471 it
->dying_tasks_head
= &cset
->dying_tasks
;
4474 * We don't keep css_sets locked across iteration steps and thus
4475 * need to take steps to ensure that iteration can be resumed after
4476 * the lock is re-acquired. Iteration is performed at two levels -
4477 * css_sets and tasks in them.
4479 * Once created, a css_set never leaves its cgroup lists, so a
4480 * pinned css_set is guaranteed to stay put and we can resume
4481 * iteration afterwards.
4483 * Tasks may leave @cset across iteration steps. This is resolved
4484 * by registering each iterator with the css_set currently being
4485 * walked and making css_set_move_task() advance iterators whose
4486 * next task is leaving.
4489 list_del(&it
->iters_node
);
4490 put_css_set_locked(it
->cur_cset
);
4493 it
->cur_cset
= cset
;
4494 list_add(&it
->iters_node
, &cset
->task_iters
);
4497 static void css_task_iter_skip(struct css_task_iter
*it
,
4498 struct task_struct
*task
)
4500 lockdep_assert_held(&css_set_lock
);
4502 if (it
->task_pos
== &task
->cg_list
) {
4503 it
->task_pos
= it
->task_pos
->next
;
4504 it
->flags
|= CSS_TASK_ITER_SKIPPED
;
4508 static void css_task_iter_advance(struct css_task_iter
*it
)
4510 struct task_struct
*task
;
4512 lockdep_assert_held(&css_set_lock
);
4516 * Advance iterator to find next entry. cset->tasks is
4517 * consumed first and then ->mg_tasks. After ->mg_tasks,
4518 * we move onto the next cset.
4520 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4521 it
->flags
&= ~CSS_TASK_ITER_SKIPPED
;
4523 it
->task_pos
= it
->task_pos
->next
;
4525 if (it
->task_pos
== it
->tasks_head
)
4526 it
->task_pos
= it
->mg_tasks_head
->next
;
4527 if (it
->task_pos
== it
->mg_tasks_head
)
4528 it
->task_pos
= it
->dying_tasks_head
->next
;
4529 if (it
->task_pos
== it
->dying_tasks_head
)
4530 css_task_iter_advance_css_set(it
);
4532 /* called from start, proceed to the first cset */
4533 css_task_iter_advance_css_set(it
);
4539 task
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
4541 if (it
->flags
& CSS_TASK_ITER_PROCS
) {
4542 /* if PROCS, skip over tasks which aren't group leaders */
4543 if (!thread_group_leader(task
))
4546 /* and dying leaders w/o live member threads */
4547 if (!atomic_read(&task
->signal
->live
))
4550 /* skip all dying ones */
4551 if (task
->flags
& PF_EXITING
)
4557 * css_task_iter_start - initiate task iteration
4558 * @css: the css to walk tasks of
4559 * @flags: CSS_TASK_ITER_* flags
4560 * @it: the task iterator to use
4562 * Initiate iteration through the tasks of @css. The caller can call
4563 * css_task_iter_next() to walk through the tasks until the function
4564 * returns NULL. On completion of iteration, css_task_iter_end() must be
4567 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4568 struct css_task_iter
*it
)
4570 /* no one should try to iterate before mounting cgroups */
4571 WARN_ON_ONCE(!use_task_css_set_links
);
4573 memset(it
, 0, sizeof(*it
));
4575 spin_lock_irq(&css_set_lock
);
4581 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4583 it
->cset_pos
= &css
->cgroup
->cset_links
;
4585 it
->cset_head
= it
->cset_pos
;
4587 css_task_iter_advance(it
);
4589 spin_unlock_irq(&css_set_lock
);
4593 * css_task_iter_next - return the next task for the iterator
4594 * @it: the task iterator being iterated
4596 * The "next" function for task iteration. @it should have been
4597 * initialized via css_task_iter_start(). Returns NULL when the iteration
4600 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4603 put_task_struct(it
->cur_task
);
4604 it
->cur_task
= NULL
;
4607 spin_lock_irq(&css_set_lock
);
4609 /* @it may be half-advanced by skips, finish advancing */
4610 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4611 css_task_iter_advance(it
);
4614 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4616 get_task_struct(it
->cur_task
);
4617 css_task_iter_advance(it
);
4620 spin_unlock_irq(&css_set_lock
);
4622 return it
->cur_task
;
4626 * css_task_iter_end - finish task iteration
4627 * @it: the task iterator to finish
4629 * Finish task iteration started by css_task_iter_start().
4631 void css_task_iter_end(struct css_task_iter
*it
)
4634 spin_lock_irq(&css_set_lock
);
4635 list_del(&it
->iters_node
);
4636 put_css_set_locked(it
->cur_cset
);
4637 spin_unlock_irq(&css_set_lock
);
4641 put_css_set(it
->cur_dcset
);
4644 put_task_struct(it
->cur_task
);
4647 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4650 css_task_iter_end(of
->priv
);
4655 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4657 struct kernfs_open_file
*of
= s
->private;
4658 struct css_task_iter
*it
= of
->priv
;
4660 return css_task_iter_next(it
);
4663 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4664 unsigned int iter_flags
)
4666 struct kernfs_open_file
*of
= s
->private;
4667 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4668 struct css_task_iter
*it
= of
->priv
;
4671 * When a seq_file is seeked, it's always traversed sequentially
4672 * from position 0, so we can simply keep iterating on !0 *pos.
4675 if (WARN_ON_ONCE((*pos
)++))
4676 return ERR_PTR(-EINVAL
);
4678 it
= kzalloc(sizeof(*it
), GFP_KERNEL
);
4680 return ERR_PTR(-ENOMEM
);
4682 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4683 } else if (!(*pos
)++) {
4684 css_task_iter_end(it
);
4685 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4688 return cgroup_procs_next(s
, NULL
, NULL
);
4691 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4693 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4696 * All processes of a threaded subtree belong to the domain cgroup
4697 * of the subtree. Only threads can be distributed across the
4698 * subtree. Reject reads on cgroup.procs in the subtree proper.
4699 * They're always empty anyway.
4701 if (cgroup_is_threaded(cgrp
))
4702 return ERR_PTR(-EOPNOTSUPP
);
4704 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4705 CSS_TASK_ITER_THREADED
);
4708 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4710 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4714 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4715 struct cgroup
*dst_cgrp
,
4716 struct super_block
*sb
)
4718 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4719 struct cgroup
*com_cgrp
= src_cgrp
;
4720 struct inode
*inode
;
4723 lockdep_assert_held(&cgroup_mutex
);
4725 /* find the common ancestor */
4726 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4727 com_cgrp
= cgroup_parent(com_cgrp
);
4729 /* %current should be authorized to migrate to the common ancestor */
4730 inode
= kernfs_get_inode(sb
, com_cgrp
->procs_file
.kn
);
4734 ret
= inode_permission(inode
, MAY_WRITE
);
4740 * If namespaces are delegation boundaries, %current must be able
4741 * to see both source and destination cgroups from its namespace.
4743 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4744 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4745 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4751 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4752 char *buf
, size_t nbytes
, loff_t off
)
4754 struct cgroup
*src_cgrp
, *dst_cgrp
;
4755 struct task_struct
*task
;
4758 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4762 task
= cgroup_procs_write_start(buf
, true);
4763 ret
= PTR_ERR_OR_ZERO(task
);
4767 /* find the source cgroup */
4768 spin_lock_irq(&css_set_lock
);
4769 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4770 spin_unlock_irq(&css_set_lock
);
4772 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4773 of
->file
->f_path
.dentry
->d_sb
);
4777 ret
= cgroup_attach_task(dst_cgrp
, task
, true);
4780 cgroup_procs_write_finish(task
);
4782 cgroup_kn_unlock(of
->kn
);
4784 return ret
?: nbytes
;
4787 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4789 return __cgroup_procs_start(s
, pos
, 0);
4792 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4793 char *buf
, size_t nbytes
, loff_t off
)
4795 struct cgroup
*src_cgrp
, *dst_cgrp
;
4796 struct task_struct
*task
;
4799 buf
= strstrip(buf
);
4801 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4805 task
= cgroup_procs_write_start(buf
, false);
4806 ret
= PTR_ERR_OR_ZERO(task
);
4810 /* find the source cgroup */
4811 spin_lock_irq(&css_set_lock
);
4812 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4813 spin_unlock_irq(&css_set_lock
);
4815 /* thread migrations follow the cgroup.procs delegation rule */
4816 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4817 of
->file
->f_path
.dentry
->d_sb
);
4821 /* and must be contained in the same domain */
4823 if (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
)
4826 ret
= cgroup_attach_task(dst_cgrp
, task
, false);
4829 cgroup_procs_write_finish(task
);
4831 cgroup_kn_unlock(of
->kn
);
4833 return ret
?: nbytes
;
4836 /* cgroup core interface files for the default hierarchy */
4837 static struct cftype cgroup_base_files
[] = {
4839 .name
= "cgroup.type",
4840 .flags
= CFTYPE_NOT_ON_ROOT
,
4841 .seq_show
= cgroup_type_show
,
4842 .write
= cgroup_type_write
,
4845 .name
= "cgroup.procs",
4846 .flags
= CFTYPE_NS_DELEGATABLE
,
4847 .file_offset
= offsetof(struct cgroup
, procs_file
),
4848 .release
= cgroup_procs_release
,
4849 .seq_start
= cgroup_procs_start
,
4850 .seq_next
= cgroup_procs_next
,
4851 .seq_show
= cgroup_procs_show
,
4852 .write
= cgroup_procs_write
,
4855 .name
= "cgroup.threads",
4856 .flags
= CFTYPE_NS_DELEGATABLE
,
4857 .release
= cgroup_procs_release
,
4858 .seq_start
= cgroup_threads_start
,
4859 .seq_next
= cgroup_procs_next
,
4860 .seq_show
= cgroup_procs_show
,
4861 .write
= cgroup_threads_write
,
4864 .name
= "cgroup.controllers",
4865 .seq_show
= cgroup_controllers_show
,
4868 .name
= "cgroup.subtree_control",
4869 .flags
= CFTYPE_NS_DELEGATABLE
,
4870 .seq_show
= cgroup_subtree_control_show
,
4871 .write
= cgroup_subtree_control_write
,
4874 .name
= "cgroup.events",
4875 .flags
= CFTYPE_NOT_ON_ROOT
,
4876 .file_offset
= offsetof(struct cgroup
, events_file
),
4877 .seq_show
= cgroup_events_show
,
4880 .name
= "cgroup.max.descendants",
4881 .seq_show
= cgroup_max_descendants_show
,
4882 .write
= cgroup_max_descendants_write
,
4885 .name
= "cgroup.max.depth",
4886 .seq_show
= cgroup_max_depth_show
,
4887 .write
= cgroup_max_depth_write
,
4890 .name
= "cgroup.stat",
4891 .seq_show
= cgroup_stat_show
,
4894 .name
= "cgroup.freeze",
4895 .flags
= CFTYPE_NOT_ON_ROOT
,
4896 .seq_show
= cgroup_freeze_show
,
4897 .write
= cgroup_freeze_write
,
4901 .flags
= CFTYPE_NOT_ON_ROOT
,
4902 .seq_show
= cpu_stat_show
,
4906 .name
= "io.pressure",
4907 .seq_show
= cgroup_io_pressure_show
,
4908 .write
= cgroup_io_pressure_write
,
4909 .poll
= cgroup_pressure_poll
,
4910 .release
= cgroup_pressure_release
,
4913 .name
= "memory.pressure",
4914 .seq_show
= cgroup_memory_pressure_show
,
4915 .write
= cgroup_memory_pressure_write
,
4916 .poll
= cgroup_pressure_poll
,
4917 .release
= cgroup_pressure_release
,
4920 .name
= "cpu.pressure",
4921 .seq_show
= cgroup_cpu_pressure_show
,
4922 .write
= cgroup_cpu_pressure_write
,
4923 .poll
= cgroup_pressure_poll
,
4924 .release
= cgroup_pressure_release
,
4926 #endif /* CONFIG_PSI */
4931 * css destruction is four-stage process.
4933 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4934 * Implemented in kill_css().
4936 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4937 * and thus css_tryget_online() is guaranteed to fail, the css can be
4938 * offlined by invoking offline_css(). After offlining, the base ref is
4939 * put. Implemented in css_killed_work_fn().
4941 * 3. When the percpu_ref reaches zero, the only possible remaining
4942 * accessors are inside RCU read sections. css_release() schedules the
4945 * 4. After the grace period, the css can be freed. Implemented in
4946 * css_free_work_fn().
4948 * It is actually hairier because both step 2 and 4 require process context
4949 * and thus involve punting to css->destroy_work adding two additional
4950 * steps to the already complex sequence.
4952 static void css_free_rwork_fn(struct work_struct
*work
)
4954 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
4955 struct cgroup_subsys_state
, destroy_rwork
);
4956 struct cgroup_subsys
*ss
= css
->ss
;
4957 struct cgroup
*cgrp
= css
->cgroup
;
4959 percpu_ref_exit(&css
->refcnt
);
4963 struct cgroup_subsys_state
*parent
= css
->parent
;
4967 cgroup_idr_remove(&ss
->css_idr
, id
);
4973 /* cgroup free path */
4974 atomic_dec(&cgrp
->root
->nr_cgrps
);
4975 cgroup1_pidlist_destroy_all(cgrp
);
4976 cancel_work_sync(&cgrp
->release_agent_work
);
4978 if (cgroup_parent(cgrp
)) {
4980 * We get a ref to the parent, and put the ref when
4981 * this cgroup is being freed, so it's guaranteed
4982 * that the parent won't be destroyed before its
4985 cgroup_put(cgroup_parent(cgrp
));
4986 kernfs_put(cgrp
->kn
);
4987 psi_cgroup_free(cgrp
);
4988 if (cgroup_on_dfl(cgrp
))
4989 cgroup_rstat_exit(cgrp
);
4993 * This is root cgroup's refcnt reaching zero,
4994 * which indicates that the root should be
4997 cgroup_destroy_root(cgrp
->root
);
5002 static void css_release_work_fn(struct work_struct
*work
)
5004 struct cgroup_subsys_state
*css
=
5005 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5006 struct cgroup_subsys
*ss
= css
->ss
;
5007 struct cgroup
*cgrp
= css
->cgroup
;
5009 mutex_lock(&cgroup_mutex
);
5011 css
->flags
|= CSS_RELEASED
;
5012 list_del_rcu(&css
->sibling
);
5015 /* css release path */
5016 if (!list_empty(&css
->rstat_css_node
)) {
5017 cgroup_rstat_flush(cgrp
);
5018 list_del_rcu(&css
->rstat_css_node
);
5021 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
5022 if (ss
->css_released
)
5023 ss
->css_released(css
);
5025 struct cgroup
*tcgrp
;
5027 /* cgroup release path */
5028 TRACE_CGROUP_PATH(release
, cgrp
);
5030 if (cgroup_on_dfl(cgrp
))
5031 cgroup_rstat_flush(cgrp
);
5033 spin_lock_irq(&css_set_lock
);
5034 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
5035 tcgrp
= cgroup_parent(tcgrp
))
5036 tcgrp
->nr_dying_descendants
--;
5037 spin_unlock_irq(&css_set_lock
);
5039 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
5043 * There are two control paths which try to determine
5044 * cgroup from dentry without going through kernfs -
5045 * cgroupstats_build() and css_tryget_online_from_dir().
5046 * Those are supported by RCU protecting clearing of
5047 * cgrp->kn->priv backpointer.
5050 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5054 mutex_unlock(&cgroup_mutex
);
5056 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5057 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5060 static void css_release(struct percpu_ref
*ref
)
5062 struct cgroup_subsys_state
*css
=
5063 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5065 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5066 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5069 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5070 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5072 lockdep_assert_held(&cgroup_mutex
);
5074 cgroup_get_live(cgrp
);
5076 memset(css
, 0, sizeof(*css
));
5080 INIT_LIST_HEAD(&css
->sibling
);
5081 INIT_LIST_HEAD(&css
->children
);
5082 INIT_LIST_HEAD(&css
->rstat_css_node
);
5083 css
->serial_nr
= css_serial_nr_next
++;
5084 atomic_set(&css
->online_cnt
, 0);
5086 if (cgroup_parent(cgrp
)) {
5087 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5088 css_get(css
->parent
);
5091 if (cgroup_on_dfl(cgrp
) && ss
->css_rstat_flush
)
5092 list_add_rcu(&css
->rstat_css_node
, &cgrp
->rstat_css_list
);
5094 BUG_ON(cgroup_css(cgrp
, ss
));
5097 /* invoke ->css_online() on a new CSS and mark it online if successful */
5098 static int online_css(struct cgroup_subsys_state
*css
)
5100 struct cgroup_subsys
*ss
= css
->ss
;
5103 lockdep_assert_held(&cgroup_mutex
);
5106 ret
= ss
->css_online(css
);
5108 css
->flags
|= CSS_ONLINE
;
5109 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5111 atomic_inc(&css
->online_cnt
);
5113 atomic_inc(&css
->parent
->online_cnt
);
5118 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5119 static void offline_css(struct cgroup_subsys_state
*css
)
5121 struct cgroup_subsys
*ss
= css
->ss
;
5123 lockdep_assert_held(&cgroup_mutex
);
5125 if (!(css
->flags
& CSS_ONLINE
))
5128 if (ss
->css_offline
)
5129 ss
->css_offline(css
);
5131 css
->flags
&= ~CSS_ONLINE
;
5132 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5134 wake_up_all(&css
->cgroup
->offline_waitq
);
5138 * css_create - create a cgroup_subsys_state
5139 * @cgrp: the cgroup new css will be associated with
5140 * @ss: the subsys of new css
5142 * Create a new css associated with @cgrp - @ss pair. On success, the new
5143 * css is online and installed in @cgrp. This function doesn't create the
5144 * interface files. Returns 0 on success, -errno on failure.
5146 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5147 struct cgroup_subsys
*ss
)
5149 struct cgroup
*parent
= cgroup_parent(cgrp
);
5150 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5151 struct cgroup_subsys_state
*css
;
5154 lockdep_assert_held(&cgroup_mutex
);
5156 css
= ss
->css_alloc(parent_css
);
5158 css
= ERR_PTR(-ENOMEM
);
5162 init_and_link_css(css
, ss
, cgrp
);
5164 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5168 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5173 /* @css is ready to be brought online now, make it visible */
5174 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5175 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5177 err
= online_css(css
);
5181 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
5182 cgroup_parent(parent
)) {
5183 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5184 current
->comm
, current
->pid
, ss
->name
);
5185 if (!strcmp(ss
->name
, "memory"))
5186 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5187 ss
->warned_broken_hierarchy
= true;
5193 list_del_rcu(&css
->sibling
);
5195 list_del_rcu(&css
->rstat_css_node
);
5196 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5197 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5198 return ERR_PTR(err
);
5202 * The returned cgroup is fully initialized including its control mask, but
5203 * it isn't associated with its kernfs_node and doesn't have the control
5206 static struct cgroup
*cgroup_create(struct cgroup
*parent
)
5208 struct cgroup_root
*root
= parent
->root
;
5209 struct cgroup
*cgrp
, *tcgrp
;
5210 int level
= parent
->level
+ 1;
5213 /* allocate the cgroup and its ID, 0 is reserved for the root */
5214 cgrp
= kzalloc(struct_size(cgrp
, ancestor_ids
, (level
+ 1)),
5217 return ERR_PTR(-ENOMEM
);
5219 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5223 if (cgroup_on_dfl(parent
)) {
5224 ret
= cgroup_rstat_init(cgrp
);
5226 goto out_cancel_ref
;
5230 * Temporarily set the pointer to NULL, so idr_find() won't return
5231 * a half-baked cgroup.
5233 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
5239 init_cgroup_housekeeping(cgrp
);
5241 cgrp
->self
.parent
= &parent
->self
;
5243 cgrp
->level
= level
;
5245 ret
= psi_cgroup_alloc(cgrp
);
5249 ret
= cgroup_bpf_inherit(cgrp
);
5254 * New cgroup inherits effective freeze counter, and
5255 * if the parent has to be frozen, the child has too.
5257 cgrp
->freezer
.e_freeze
= parent
->freezer
.e_freeze
;
5258 if (cgrp
->freezer
.e_freeze
) {
5260 * Set the CGRP_FREEZE flag, so when a process will be
5261 * attached to the child cgroup, it will become frozen.
5262 * At this point the new cgroup is unpopulated, so we can
5263 * consider it frozen immediately.
5265 set_bit(CGRP_FREEZE
, &cgrp
->flags
);
5266 set_bit(CGRP_FROZEN
, &cgrp
->flags
);
5269 spin_lock_irq(&css_set_lock
);
5270 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5271 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
5273 if (tcgrp
!= cgrp
) {
5274 tcgrp
->nr_descendants
++;
5277 * If the new cgroup is frozen, all ancestor cgroups
5278 * get a new frozen descendant, but their state can't
5279 * change because of this.
5281 if (cgrp
->freezer
.e_freeze
)
5282 tcgrp
->freezer
.nr_frozen_descendants
++;
5285 spin_unlock_irq(&css_set_lock
);
5287 if (notify_on_release(parent
))
5288 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5290 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5291 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5293 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5295 /* allocation complete, commit to creation */
5296 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5297 atomic_inc(&root
->nr_cgrps
);
5298 cgroup_get_live(parent
);
5301 * @cgrp is now fully operational. If something fails after this
5302 * point, it'll be released via the normal destruction path.
5304 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
5307 * On the default hierarchy, a child doesn't automatically inherit
5308 * subtree_control from the parent. Each is configured manually.
5310 if (!cgroup_on_dfl(cgrp
))
5311 cgrp
->subtree_control
= cgroup_control(cgrp
);
5313 cgroup_propagate_control(cgrp
);
5318 psi_cgroup_free(cgrp
);
5320 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
5322 if (cgroup_on_dfl(parent
))
5323 cgroup_rstat_exit(cgrp
);
5325 percpu_ref_exit(&cgrp
->self
.refcnt
);
5328 return ERR_PTR(ret
);
5331 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
5333 struct cgroup
*cgroup
;
5337 lockdep_assert_held(&cgroup_mutex
);
5339 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
5340 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
5343 if (level
> cgroup
->max_depth
)
5354 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
5356 struct cgroup
*parent
, *cgrp
;
5357 struct kernfs_node
*kn
;
5360 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5361 if (strchr(name
, '\n'))
5364 parent
= cgroup_kn_lock_live(parent_kn
, false);
5368 if (!cgroup_check_hierarchy_limits(parent
)) {
5373 cgrp
= cgroup_create(parent
);
5375 ret
= PTR_ERR(cgrp
);
5379 /* create the directory */
5380 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5388 * This extra ref will be put in cgroup_free_fn() and guarantees
5389 * that @cgrp->kn is always accessible.
5393 ret
= cgroup_kn_set_ugid(kn
);
5397 ret
= css_populate_dir(&cgrp
->self
);
5401 ret
= cgroup_apply_control_enable(cgrp
);
5405 TRACE_CGROUP_PATH(mkdir
, cgrp
);
5407 /* let's create and online css's */
5408 kernfs_activate(kn
);
5414 cgroup_destroy_locked(cgrp
);
5416 cgroup_kn_unlock(parent_kn
);
5421 * This is called when the refcnt of a css is confirmed to be killed.
5422 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5423 * initate destruction and put the css ref from kill_css().
5425 static void css_killed_work_fn(struct work_struct
*work
)
5427 struct cgroup_subsys_state
*css
=
5428 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5430 mutex_lock(&cgroup_mutex
);
5435 /* @css can't go away while we're holding cgroup_mutex */
5437 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5439 mutex_unlock(&cgroup_mutex
);
5442 /* css kill confirmation processing requires process context, bounce */
5443 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5445 struct cgroup_subsys_state
*css
=
5446 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5448 if (atomic_dec_and_test(&css
->online_cnt
)) {
5449 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5450 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5455 * kill_css - destroy a css
5456 * @css: css to destroy
5458 * This function initiates destruction of @css by removing cgroup interface
5459 * files and putting its base reference. ->css_offline() will be invoked
5460 * asynchronously once css_tryget_online() is guaranteed to fail and when
5461 * the reference count reaches zero, @css will be released.
5463 static void kill_css(struct cgroup_subsys_state
*css
)
5465 lockdep_assert_held(&cgroup_mutex
);
5467 if (css
->flags
& CSS_DYING
)
5470 css
->flags
|= CSS_DYING
;
5473 * This must happen before css is disassociated with its cgroup.
5474 * See seq_css() for details.
5479 * Killing would put the base ref, but we need to keep it alive
5480 * until after ->css_offline().
5485 * cgroup core guarantees that, by the time ->css_offline() is
5486 * invoked, no new css reference will be given out via
5487 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5488 * proceed to offlining css's because percpu_ref_kill() doesn't
5489 * guarantee that the ref is seen as killed on all CPUs on return.
5491 * Use percpu_ref_kill_and_confirm() to get notifications as each
5492 * css is confirmed to be seen as killed on all CPUs.
5494 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5498 * cgroup_destroy_locked - the first stage of cgroup destruction
5499 * @cgrp: cgroup to be destroyed
5501 * css's make use of percpu refcnts whose killing latency shouldn't be
5502 * exposed to userland and are RCU protected. Also, cgroup core needs to
5503 * guarantee that css_tryget_online() won't succeed by the time
5504 * ->css_offline() is invoked. To satisfy all the requirements,
5505 * destruction is implemented in the following two steps.
5507 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5508 * userland visible parts and start killing the percpu refcnts of
5509 * css's. Set up so that the next stage will be kicked off once all
5510 * the percpu refcnts are confirmed to be killed.
5512 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5513 * rest of destruction. Once all cgroup references are gone, the
5514 * cgroup is RCU-freed.
5516 * This function implements s1. After this step, @cgrp is gone as far as
5517 * the userland is concerned and a new cgroup with the same name may be
5518 * created. As cgroup doesn't care about the names internally, this
5519 * doesn't cause any problem.
5521 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5522 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5524 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5525 struct cgroup_subsys_state
*css
;
5526 struct cgrp_cset_link
*link
;
5529 lockdep_assert_held(&cgroup_mutex
);
5532 * Only migration can raise populated from zero and we're already
5533 * holding cgroup_mutex.
5535 if (cgroup_is_populated(cgrp
))
5539 * Make sure there's no live children. We can't test emptiness of
5540 * ->self.children as dead children linger on it while being
5541 * drained; otherwise, "rmdir parent/child parent" may fail.
5543 if (css_has_online_children(&cgrp
->self
))
5547 * Mark @cgrp and the associated csets dead. The former prevents
5548 * further task migration and child creation by disabling
5549 * cgroup_lock_live_group(). The latter makes the csets ignored by
5550 * the migration path.
5552 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5554 spin_lock_irq(&css_set_lock
);
5555 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5556 link
->cset
->dead
= true;
5557 spin_unlock_irq(&css_set_lock
);
5559 /* initiate massacre of all css's */
5560 for_each_css(css
, ssid
, cgrp
)
5563 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5564 css_clear_dir(&cgrp
->self
);
5565 kernfs_remove(cgrp
->kn
);
5567 if (parent
&& cgroup_is_threaded(cgrp
))
5568 parent
->nr_threaded_children
--;
5570 spin_lock_irq(&css_set_lock
);
5571 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5572 tcgrp
->nr_descendants
--;
5573 tcgrp
->nr_dying_descendants
++;
5575 * If the dying cgroup is frozen, decrease frozen descendants
5576 * counters of ancestor cgroups.
5578 if (test_bit(CGRP_FROZEN
, &cgrp
->flags
))
5579 tcgrp
->freezer
.nr_frozen_descendants
--;
5581 spin_unlock_irq(&css_set_lock
);
5583 cgroup1_check_for_release(parent
);
5585 cgroup_bpf_offline(cgrp
);
5587 /* put the base reference */
5588 percpu_ref_kill(&cgrp
->self
.refcnt
);
5593 int cgroup_rmdir(struct kernfs_node
*kn
)
5595 struct cgroup
*cgrp
;
5598 cgrp
= cgroup_kn_lock_live(kn
, false);
5602 ret
= cgroup_destroy_locked(cgrp
);
5604 TRACE_CGROUP_PATH(rmdir
, cgrp
);
5606 cgroup_kn_unlock(kn
);
5610 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5611 .show_options
= cgroup_show_options
,
5612 .mkdir
= cgroup_mkdir
,
5613 .rmdir
= cgroup_rmdir
,
5614 .show_path
= cgroup_show_path
,
5617 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5619 struct cgroup_subsys_state
*css
;
5621 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5623 mutex_lock(&cgroup_mutex
);
5625 idr_init(&ss
->css_idr
);
5626 INIT_LIST_HEAD(&ss
->cfts
);
5628 /* Create the root cgroup state for this subsystem */
5629 ss
->root
= &cgrp_dfl_root
;
5630 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5631 /* We don't handle early failures gracefully */
5632 BUG_ON(IS_ERR(css
));
5633 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5636 * Root csses are never destroyed and we can't initialize
5637 * percpu_ref during early init. Disable refcnting.
5639 css
->flags
|= CSS_NO_REF
;
5642 /* allocation can't be done safely during early init */
5645 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5646 BUG_ON(css
->id
< 0);
5649 /* Update the init_css_set to contain a subsys
5650 * pointer to this state - since the subsystem is
5651 * newly registered, all tasks and hence the
5652 * init_css_set is in the subsystem's root cgroup. */
5653 init_css_set
.subsys
[ss
->id
] = css
;
5655 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5656 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5657 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5658 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5660 /* At system boot, before all subsystems have been
5661 * registered, no tasks have been forked, so we don't
5662 * need to invoke fork callbacks here. */
5663 BUG_ON(!list_empty(&init_task
.tasks
));
5665 BUG_ON(online_css(css
));
5667 mutex_unlock(&cgroup_mutex
);
5671 * cgroup_init_early - cgroup initialization at system boot
5673 * Initialize cgroups at system boot, and initialize any
5674 * subsystems that request early init.
5676 int __init
cgroup_init_early(void)
5678 static struct cgroup_fs_context __initdata ctx
;
5679 struct cgroup_subsys
*ss
;
5682 ctx
.root
= &cgrp_dfl_root
;
5683 init_cgroup_root(&ctx
);
5684 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5686 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5688 for_each_subsys(ss
, i
) {
5689 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5690 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5691 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5693 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5694 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5697 ss
->name
= cgroup_subsys_name
[i
];
5698 if (!ss
->legacy_name
)
5699 ss
->legacy_name
= cgroup_subsys_name
[i
];
5702 cgroup_init_subsys(ss
, true);
5707 static u16 cgroup_disable_mask __initdata
;
5710 * cgroup_init - cgroup initialization
5712 * Register cgroup filesystem and /proc file, and initialize
5713 * any subsystems that didn't request early init.
5715 int __init
cgroup_init(void)
5717 struct cgroup_subsys
*ss
;
5720 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5721 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5722 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5724 cgroup_rstat_boot();
5727 * The latency of the synchronize_rcu() is too high for cgroups,
5728 * avoid it at the cost of forcing all readers into the slow path.
5730 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5732 get_user_ns(init_cgroup_ns
.user_ns
);
5734 mutex_lock(&cgroup_mutex
);
5737 * Add init_css_set to the hash table so that dfl_root can link to
5740 hash_add(css_set_table
, &init_css_set
.hlist
,
5741 css_set_hash(init_css_set
.subsys
));
5743 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5745 mutex_unlock(&cgroup_mutex
);
5747 for_each_subsys(ss
, ssid
) {
5748 if (ss
->early_init
) {
5749 struct cgroup_subsys_state
*css
=
5750 init_css_set
.subsys
[ss
->id
];
5752 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5754 BUG_ON(css
->id
< 0);
5756 cgroup_init_subsys(ss
, false);
5759 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5760 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5763 * Setting dfl_root subsys_mask needs to consider the
5764 * disabled flag and cftype registration needs kmalloc,
5765 * both of which aren't available during early_init.
5767 if (cgroup_disable_mask
& (1 << ssid
)) {
5768 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5769 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5774 if (cgroup1_ssid_disabled(ssid
))
5775 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5778 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5780 /* implicit controllers must be threaded too */
5781 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5783 if (ss
->implicit_on_dfl
)
5784 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5785 else if (!ss
->dfl_cftypes
)
5786 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5789 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5791 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5792 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5794 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5795 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5799 ss
->bind(init_css_set
.subsys
[ssid
]);
5801 mutex_lock(&cgroup_mutex
);
5802 css_populate_dir(init_css_set
.subsys
[ssid
]);
5803 mutex_unlock(&cgroup_mutex
);
5806 /* init_css_set.subsys[] has been updated, re-hash */
5807 hash_del(&init_css_set
.hlist
);
5808 hash_add(css_set_table
, &init_css_set
.hlist
,
5809 css_set_hash(init_css_set
.subsys
));
5811 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5812 WARN_ON(register_filesystem(&cgroup_fs_type
));
5813 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5814 WARN_ON(!proc_create_single("cgroups", 0, NULL
, proc_cgroupstats_show
));
5815 #ifdef CONFIG_CPUSETS
5816 WARN_ON(register_filesystem(&cpuset_fs_type
));
5822 static int __init
cgroup_wq_init(void)
5825 * There isn't much point in executing destruction path in
5826 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5827 * Use 1 for @max_active.
5829 * We would prefer to do this in cgroup_init() above, but that
5830 * is called before init_workqueues(): so leave this until after.
5832 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5833 BUG_ON(!cgroup_destroy_wq
);
5836 core_initcall(cgroup_wq_init
);
5838 void cgroup_path_from_kernfs_id(const union kernfs_node_id
*id
,
5839 char *buf
, size_t buflen
)
5841 struct kernfs_node
*kn
;
5843 kn
= kernfs_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5846 kernfs_path(kn
, buf
, buflen
);
5851 * proc_cgroup_show()
5852 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5853 * - Used for /proc/<pid>/cgroup.
5855 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5856 struct pid
*pid
, struct task_struct
*tsk
)
5860 struct cgroup_root
*root
;
5863 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5867 mutex_lock(&cgroup_mutex
);
5868 spin_lock_irq(&css_set_lock
);
5870 for_each_root(root
) {
5871 struct cgroup_subsys
*ss
;
5872 struct cgroup
*cgrp
;
5873 int ssid
, count
= 0;
5875 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5878 seq_printf(m
, "%d:", root
->hierarchy_id
);
5879 if (root
!= &cgrp_dfl_root
)
5880 for_each_subsys(ss
, ssid
)
5881 if (root
->subsys_mask
& (1 << ssid
))
5882 seq_printf(m
, "%s%s", count
++ ? "," : "",
5884 if (strlen(root
->name
))
5885 seq_printf(m
, "%sname=%s", count
? "," : "",
5889 cgrp
= task_cgroup_from_root(tsk
, root
);
5892 * On traditional hierarchies, all zombie tasks show up as
5893 * belonging to the root cgroup. On the default hierarchy,
5894 * while a zombie doesn't show up in "cgroup.procs" and
5895 * thus can't be migrated, its /proc/PID/cgroup keeps
5896 * reporting the cgroup it belonged to before exiting. If
5897 * the cgroup is removed before the zombie is reaped,
5898 * " (deleted)" is appended to the cgroup path.
5900 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5901 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5902 current
->nsproxy
->cgroup_ns
);
5903 if (retval
>= PATH_MAX
)
5904 retval
= -ENAMETOOLONG
;
5913 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5914 seq_puts(m
, " (deleted)\n");
5921 spin_unlock_irq(&css_set_lock
);
5922 mutex_unlock(&cgroup_mutex
);
5929 * cgroup_fork - initialize cgroup related fields during copy_process()
5930 * @child: pointer to task_struct of forking parent process.
5932 * A task is associated with the init_css_set until cgroup_post_fork()
5933 * attaches it to the parent's css_set. Empty cg_list indicates that
5934 * @child isn't holding reference to its css_set.
5936 void cgroup_fork(struct task_struct
*child
)
5938 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5939 INIT_LIST_HEAD(&child
->cg_list
);
5943 * cgroup_can_fork - called on a new task before the process is exposed
5944 * @child: the task in question.
5946 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5947 * returns an error, the fork aborts with that error code. This allows for
5948 * a cgroup subsystem to conditionally allow or deny new forks.
5950 int cgroup_can_fork(struct task_struct
*child
)
5952 struct cgroup_subsys
*ss
;
5955 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
5956 ret
= ss
->can_fork(child
);
5959 } while_each_subsys_mask();
5964 for_each_subsys(ss
, j
) {
5967 if (ss
->cancel_fork
)
5968 ss
->cancel_fork(child
);
5975 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5976 * @child: the task in question
5978 * This calls the cancel_fork() callbacks if a fork failed *after*
5979 * cgroup_can_fork() succeded.
5981 void cgroup_cancel_fork(struct task_struct
*child
)
5983 struct cgroup_subsys
*ss
;
5986 for_each_subsys(ss
, i
)
5987 if (ss
->cancel_fork
)
5988 ss
->cancel_fork(child
);
5992 * cgroup_post_fork - called on a new task after adding it to the task list
5993 * @child: the task in question
5995 * Adds the task to the list running through its css_set if necessary and
5996 * call the subsystem fork() callbacks. Has to be after the task is
5997 * visible on the task list in case we race with the first call to
5998 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
6001 void cgroup_post_fork(struct task_struct
*child
)
6003 struct cgroup_subsys
*ss
;
6007 * This may race against cgroup_enable_task_cg_lists(). As that
6008 * function sets use_task_css_set_links before grabbing
6009 * tasklist_lock and we just went through tasklist_lock to add
6010 * @child, it's guaranteed that either we see the set
6011 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
6012 * @child during its iteration.
6014 * If we won the race, @child is associated with %current's
6015 * css_set. Grabbing css_set_lock guarantees both that the
6016 * association is stable, and, on completion of the parent's
6017 * migration, @child is visible in the source of migration or
6018 * already in the destination cgroup. This guarantee is necessary
6019 * when implementing operations which need to migrate all tasks of
6020 * a cgroup to another.
6022 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
6023 * will remain in init_css_set. This is safe because all tasks are
6024 * in the init_css_set before cg_links is enabled and there's no
6025 * operation which transfers all tasks out of init_css_set.
6027 if (use_task_css_set_links
) {
6028 struct css_set
*cset
;
6030 spin_lock_irq(&css_set_lock
);
6031 cset
= task_css_set(current
);
6032 if (list_empty(&child
->cg_list
)) {
6035 css_set_move_task(child
, NULL
, cset
, false);
6039 * If the cgroup has to be frozen, the new task has too.
6040 * Let's set the JOBCTL_TRAP_FREEZE jobctl bit to get
6041 * the task into the frozen state.
6043 if (unlikely(cgroup_task_freeze(child
))) {
6044 spin_lock(&child
->sighand
->siglock
);
6045 WARN_ON_ONCE(child
->frozen
);
6046 child
->jobctl
|= JOBCTL_TRAP_FREEZE
;
6047 spin_unlock(&child
->sighand
->siglock
);
6050 * Calling cgroup_update_frozen() isn't required here,
6051 * because it will be called anyway a bit later
6052 * from do_freezer_trap(). So we avoid cgroup's
6053 * transient switch from the frozen state and back.
6057 spin_unlock_irq(&css_set_lock
);
6061 * Call ss->fork(). This must happen after @child is linked on
6062 * css_set; otherwise, @child might change state between ->fork()
6063 * and addition to css_set.
6065 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
6067 } while_each_subsys_mask();
6071 * cgroup_exit - detach cgroup from exiting task
6072 * @tsk: pointer to task_struct of exiting process
6074 * Description: Detach cgroup from @tsk and release it.
6076 * Note that cgroups marked notify_on_release force every task in
6077 * them to take the global cgroup_mutex mutex when exiting.
6078 * This could impact scaling on very large systems. Be reluctant to
6079 * use notify_on_release cgroups where very high task exit scaling
6080 * is required on large systems.
6082 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
6083 * call cgroup_exit() while the task is still competent to handle
6084 * notify_on_release(), then leave the task attached to the root cgroup in
6085 * each hierarchy for the remainder of its exit. No need to bother with
6086 * init_css_set refcnting. init_css_set never goes away and we can't race
6087 * with migration path - PF_EXITING is visible to migration path.
6089 void cgroup_exit(struct task_struct
*tsk
)
6091 struct cgroup_subsys
*ss
;
6092 struct css_set
*cset
;
6096 * Unlink from @tsk from its css_set. As migration path can't race
6097 * with us, we can check css_set and cg_list without synchronization.
6099 cset
= task_css_set(tsk
);
6101 if (!list_empty(&tsk
->cg_list
)) {
6102 spin_lock_irq(&css_set_lock
);
6103 css_set_move_task(tsk
, cset
, NULL
, false);
6104 list_add_tail(&tsk
->cg_list
, &cset
->dying_tasks
);
6107 WARN_ON_ONCE(cgroup_task_frozen(tsk
));
6108 if (unlikely(cgroup_task_freeze(tsk
)))
6109 cgroup_update_frozen(task_dfl_cgroup(tsk
));
6111 spin_unlock_irq(&css_set_lock
);
6116 /* see cgroup_post_fork() for details */
6117 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
6119 } while_each_subsys_mask();
6122 void cgroup_release(struct task_struct
*task
)
6124 struct cgroup_subsys
*ss
;
6127 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
6129 } while_each_subsys_mask();
6131 if (use_task_css_set_links
) {
6132 spin_lock_irq(&css_set_lock
);
6133 css_set_skip_task_iters(task_css_set(task
), task
);
6134 list_del_init(&task
->cg_list
);
6135 spin_unlock_irq(&css_set_lock
);
6139 void cgroup_free(struct task_struct
*task
)
6141 struct css_set
*cset
= task_css_set(task
);
6145 static int __init
cgroup_disable(char *str
)
6147 struct cgroup_subsys
*ss
;
6151 while ((token
= strsep(&str
, ",")) != NULL
) {
6155 for_each_subsys(ss
, i
) {
6156 if (strcmp(token
, ss
->name
) &&
6157 strcmp(token
, ss
->legacy_name
))
6159 cgroup_disable_mask
|= 1 << i
;
6164 __setup("cgroup_disable=", cgroup_disable
);
6166 void __init __weak
enable_debug_cgroup(void) { }
6168 static int __init
enable_cgroup_debug(char *str
)
6170 cgroup_debug
= true;
6171 enable_debug_cgroup();
6174 __setup("cgroup_debug", enable_cgroup_debug
);
6177 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6178 * @dentry: directory dentry of interest
6179 * @ss: subsystem of interest
6181 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6182 * to get the corresponding css and return it. If such css doesn't exist
6183 * or can't be pinned, an ERR_PTR value is returned.
6185 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6186 struct cgroup_subsys
*ss
)
6188 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6189 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6190 struct cgroup_subsys_state
*css
= NULL
;
6191 struct cgroup
*cgrp
;
6193 /* is @dentry a cgroup dir? */
6194 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6195 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6196 return ERR_PTR(-EBADF
);
6201 * This path doesn't originate from kernfs and @kn could already
6202 * have been or be removed at any point. @kn->priv is RCU
6203 * protected for this access. See css_release_work_fn() for details.
6205 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6207 css
= cgroup_css(cgrp
, ss
);
6209 if (!css
|| !css_tryget_online(css
))
6210 css
= ERR_PTR(-ENOENT
);
6217 * css_from_id - lookup css by id
6218 * @id: the cgroup id
6219 * @ss: cgroup subsys to be looked into
6221 * Returns the css if there's valid one with @id, otherwise returns NULL.
6222 * Should be called under rcu_read_lock().
6224 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6226 WARN_ON_ONCE(!rcu_read_lock_held());
6227 return idr_find(&ss
->css_idr
, id
);
6231 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6232 * @path: path on the default hierarchy
6234 * Find the cgroup at @path on the default hierarchy, increment its
6235 * reference count and return it. Returns pointer to the found cgroup on
6236 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6237 * if @path points to a non-directory.
6239 struct cgroup
*cgroup_get_from_path(const char *path
)
6241 struct kernfs_node
*kn
;
6242 struct cgroup
*cgrp
;
6244 mutex_lock(&cgroup_mutex
);
6246 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
6248 if (kernfs_type(kn
) == KERNFS_DIR
) {
6250 cgroup_get_live(cgrp
);
6252 cgrp
= ERR_PTR(-ENOTDIR
);
6256 cgrp
= ERR_PTR(-ENOENT
);
6259 mutex_unlock(&cgroup_mutex
);
6262 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6265 * cgroup_get_from_fd - get a cgroup pointer from a fd
6266 * @fd: fd obtained by open(cgroup2_dir)
6268 * Find the cgroup from a fd which should be obtained
6269 * by opening a cgroup directory. Returns a pointer to the
6270 * cgroup on success. ERR_PTR is returned if the cgroup
6273 struct cgroup
*cgroup_get_from_fd(int fd
)
6275 struct cgroup_subsys_state
*css
;
6276 struct cgroup
*cgrp
;
6281 return ERR_PTR(-EBADF
);
6283 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
6286 return ERR_CAST(css
);
6289 if (!cgroup_on_dfl(cgrp
)) {
6291 return ERR_PTR(-EBADF
);
6296 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6298 static u64
power_of_ten(int power
)
6307 * cgroup_parse_float - parse a floating number
6308 * @input: input string
6309 * @dec_shift: number of decimal digits to shift
6312 * Parse a decimal floating point number in @input and store the result in
6313 * @v with decimal point right shifted @dec_shift times. For example, if
6314 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6315 * Returns 0 on success, -errno otherwise.
6317 * There's nothing cgroup specific about this function except that it's
6318 * currently the only user.
6320 int cgroup_parse_float(const char *input
, unsigned dec_shift
, s64
*v
)
6322 s64 whole
, frac
= 0;
6323 int fstart
= 0, fend
= 0, flen
;
6325 if (!sscanf(input
, "%lld.%n%lld%n", &whole
, &fstart
, &frac
, &fend
))
6330 flen
= fend
> fstart
? fend
- fstart
: 0;
6331 if (flen
< dec_shift
)
6332 frac
*= power_of_ten(dec_shift
- flen
);
6334 frac
= DIV_ROUND_CLOSEST_ULL(frac
, power_of_ten(flen
- dec_shift
));
6336 *v
= whole
* power_of_ten(dec_shift
) + frac
;
6341 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6342 * definition in cgroup-defs.h.
6344 #ifdef CONFIG_SOCK_CGROUP_DATA
6346 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6348 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
6349 static bool cgroup_sk_alloc_disabled __read_mostly
;
6351 void cgroup_sk_alloc_disable(void)
6353 if (cgroup_sk_alloc_disabled
)
6355 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6356 cgroup_sk_alloc_disabled
= true;
6361 #define cgroup_sk_alloc_disabled false
6365 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6367 if (cgroup_sk_alloc_disabled
)
6370 /* Socket clone path */
6373 * We might be cloning a socket which is left in an empty
6374 * cgroup and the cgroup might have already been rmdir'd.
6375 * Don't use cgroup_get_live().
6377 cgroup_get(sock_cgroup_ptr(skcd
));
6378 cgroup_bpf_get(sock_cgroup_ptr(skcd
));
6385 struct css_set
*cset
;
6387 cset
= task_css_set(current
);
6388 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6389 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
6390 cgroup_bpf_get(cset
->dfl_cgrp
);
6399 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6401 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6403 cgroup_bpf_put(cgrp
);
6407 #endif /* CONFIG_SOCK_CGROUP_DATA */
6409 #ifdef CONFIG_CGROUP_BPF
6410 int cgroup_bpf_attach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6411 enum bpf_attach_type type
, u32 flags
)
6415 mutex_lock(&cgroup_mutex
);
6416 ret
= __cgroup_bpf_attach(cgrp
, prog
, type
, flags
);
6417 mutex_unlock(&cgroup_mutex
);
6420 int cgroup_bpf_detach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6421 enum bpf_attach_type type
, u32 flags
)
6425 mutex_lock(&cgroup_mutex
);
6426 ret
= __cgroup_bpf_detach(cgrp
, prog
, type
);
6427 mutex_unlock(&cgroup_mutex
);
6430 int cgroup_bpf_query(struct cgroup
*cgrp
, const union bpf_attr
*attr
,
6431 union bpf_attr __user
*uattr
)
6435 mutex_lock(&cgroup_mutex
);
6436 ret
= __cgroup_bpf_query(cgrp
, attr
, uattr
);
6437 mutex_unlock(&cgroup_mutex
);
6440 #endif /* CONFIG_CGROUP_BPF */
6443 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
6444 ssize_t size
, const char *prefix
)
6449 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
6450 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
6454 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
6456 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
6458 if (WARN_ON(ret
>= size
))
6465 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6468 struct cgroup_subsys
*ss
;
6472 ret
= show_delegatable_files(cgroup_base_files
, buf
, PAGE_SIZE
- ret
,
6475 for_each_subsys(ss
, ssid
)
6476 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
6478 cgroup_subsys_name
[ssid
]);
6482 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
6484 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6487 return snprintf(buf
, PAGE_SIZE
, "nsdelegate\nmemory_localevents\n");
6489 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
6491 static struct attribute
*cgroup_sysfs_attrs
[] = {
6492 &cgroup_delegate_attr
.attr
,
6493 &cgroup_features_attr
.attr
,
6497 static const struct attribute_group cgroup_sysfs_attr_group
= {
6498 .attrs
= cgroup_sysfs_attrs
,
6502 static int __init
cgroup_sysfs_init(void)
6504 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
6506 subsys_initcall(cgroup_sysfs_init
);
6508 #endif /* CONFIG_SYSFS */