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
;
1724 u16 dfl_disable_ss_mask
= 0;
1726 lockdep_assert_held(&cgroup_mutex
);
1728 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1730 * If @ss has non-root csses attached to it, can't move.
1731 * If @ss is an implicit controller, it is exempt from this
1732 * rule and can be stolen.
1734 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1735 !ss
->implicit_on_dfl
)
1738 /* can't move between two non-dummy roots either */
1739 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1743 * Collect ssid's that need to be disabled from default
1746 if (ss
->root
== &cgrp_dfl_root
)
1747 dfl_disable_ss_mask
|= 1 << ssid
;
1749 } while_each_subsys_mask();
1751 if (dfl_disable_ss_mask
) {
1752 struct cgroup
*scgrp
= &cgrp_dfl_root
.cgrp
;
1755 * Controllers from default hierarchy that need to be rebound
1756 * are all disabled together in one go.
1758 cgrp_dfl_root
.subsys_mask
&= ~dfl_disable_ss_mask
;
1759 WARN_ON(cgroup_apply_control(scgrp
));
1760 cgroup_finalize_control(scgrp
, 0);
1763 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1764 struct cgroup_root
*src_root
= ss
->root
;
1765 struct cgroup
*scgrp
= &src_root
->cgrp
;
1766 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1767 struct css_set
*cset
;
1769 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1771 if (src_root
!= &cgrp_dfl_root
) {
1772 /* disable from the source */
1773 src_root
->subsys_mask
&= ~(1 << ssid
);
1774 WARN_ON(cgroup_apply_control(scgrp
));
1775 cgroup_finalize_control(scgrp
, 0);
1779 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1780 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1781 ss
->root
= dst_root
;
1782 css
->cgroup
= dcgrp
;
1784 spin_lock_irq(&css_set_lock
);
1785 hash_for_each(css_set_table
, i
, cset
, hlist
)
1786 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1787 &dcgrp
->e_csets
[ss
->id
]);
1788 spin_unlock_irq(&css_set_lock
);
1790 /* default hierarchy doesn't enable controllers by default */
1791 dst_root
->subsys_mask
|= 1 << ssid
;
1792 if (dst_root
== &cgrp_dfl_root
) {
1793 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1795 dcgrp
->subtree_control
|= 1 << ssid
;
1796 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1799 ret
= cgroup_apply_control(dcgrp
);
1801 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1806 } while_each_subsys_mask();
1808 kernfs_activate(dcgrp
->kn
);
1812 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1813 struct kernfs_root
*kf_root
)
1817 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1818 struct cgroup
*ns_cgroup
;
1820 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1824 spin_lock_irq(&css_set_lock
);
1825 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1826 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1827 spin_unlock_irq(&css_set_lock
);
1829 if (len
>= PATH_MAX
)
1832 seq_escape(sf
, buf
, " \t\n\\");
1839 enum cgroup2_param
{
1841 Opt_memory_localevents
,
1845 static const struct fs_parameter_spec cgroup2_param_specs
[] = {
1846 fsparam_flag("nsdelegate", Opt_nsdelegate
),
1847 fsparam_flag("memory_localevents", Opt_memory_localevents
),
1851 static const struct fs_parameter_description cgroup2_fs_parameters
= {
1853 .specs
= cgroup2_param_specs
,
1856 static int cgroup2_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
1858 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1859 struct fs_parse_result result
;
1862 opt
= fs_parse(fc
, &cgroup2_fs_parameters
, param
, &result
);
1867 case Opt_nsdelegate
:
1868 ctx
->flags
|= CGRP_ROOT_NS_DELEGATE
;
1870 case Opt_memory_localevents
:
1871 ctx
->flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1877 static void apply_cgroup_root_flags(unsigned int root_flags
)
1879 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1880 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1881 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1883 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1885 if (root_flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1886 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1888 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1892 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1894 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1895 seq_puts(seq
, ",nsdelegate");
1896 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1897 seq_puts(seq
, ",memory_localevents");
1901 static int cgroup_reconfigure(struct fs_context
*fc
)
1903 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1905 apply_cgroup_root_flags(ctx
->flags
);
1910 * To reduce the fork() overhead for systems that are not actually using
1911 * their cgroups capability, we don't maintain the lists running through
1912 * each css_set to its tasks until we see the list actually used - in other
1913 * words after the first mount.
1915 static bool use_task_css_set_links __read_mostly
;
1917 void cgroup_enable_task_cg_lists(void)
1919 struct task_struct
*p
, *g
;
1922 * We need tasklist_lock because RCU is not safe against
1923 * while_each_thread(). Besides, a forking task that has passed
1924 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1925 * is not guaranteed to have its child immediately visible in the
1926 * tasklist if we walk through it with RCU.
1928 read_lock(&tasklist_lock
);
1929 spin_lock_irq(&css_set_lock
);
1931 if (use_task_css_set_links
)
1934 use_task_css_set_links
= true;
1936 do_each_thread(g
, p
) {
1937 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1938 task_css_set(p
) != &init_css_set
);
1941 * We should check if the process is exiting, otherwise
1942 * it will race with cgroup_exit() in that the list
1943 * entry won't be deleted though the process has exited.
1944 * Do it while holding siglock so that we don't end up
1945 * racing against cgroup_exit().
1947 * Interrupts were already disabled while acquiring
1948 * the css_set_lock, so we do not need to disable it
1949 * again when acquiring the sighand->siglock here.
1951 spin_lock(&p
->sighand
->siglock
);
1952 if (!(p
->flags
& PF_EXITING
)) {
1953 struct css_set
*cset
= task_css_set(p
);
1955 if (!css_set_populated(cset
))
1956 css_set_update_populated(cset
, true);
1957 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1961 spin_unlock(&p
->sighand
->siglock
);
1962 } while_each_thread(g
, p
);
1964 spin_unlock_irq(&css_set_lock
);
1965 read_unlock(&tasklist_lock
);
1968 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1970 struct cgroup_subsys
*ss
;
1973 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1974 INIT_LIST_HEAD(&cgrp
->self
.children
);
1975 INIT_LIST_HEAD(&cgrp
->cset_links
);
1976 INIT_LIST_HEAD(&cgrp
->pidlists
);
1977 mutex_init(&cgrp
->pidlist_mutex
);
1978 cgrp
->self
.cgroup
= cgrp
;
1979 cgrp
->self
.flags
|= CSS_ONLINE
;
1980 cgrp
->dom_cgrp
= cgrp
;
1981 cgrp
->max_descendants
= INT_MAX
;
1982 cgrp
->max_depth
= INT_MAX
;
1983 INIT_LIST_HEAD(&cgrp
->rstat_css_list
);
1984 prev_cputime_init(&cgrp
->prev_cputime
);
1986 for_each_subsys(ss
, ssid
)
1987 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1989 init_waitqueue_head(&cgrp
->offline_waitq
);
1990 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1993 void init_cgroup_root(struct cgroup_fs_context
*ctx
)
1995 struct cgroup_root
*root
= ctx
->root
;
1996 struct cgroup
*cgrp
= &root
->cgrp
;
1998 INIT_LIST_HEAD(&root
->root_list
);
1999 atomic_set(&root
->nr_cgrps
, 1);
2001 init_cgroup_housekeeping(cgrp
);
2002 idr_init(&root
->cgroup_idr
);
2004 root
->flags
= ctx
->flags
;
2005 if (ctx
->release_agent
)
2006 strscpy(root
->release_agent_path
, ctx
->release_agent
, PATH_MAX
);
2008 strscpy(root
->name
, ctx
->name
, MAX_CGROUP_ROOT_NAMELEN
);
2009 if (ctx
->cpuset_clone_children
)
2010 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
2013 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
2015 LIST_HEAD(tmp_links
);
2016 struct cgroup
*root_cgrp
= &root
->cgrp
;
2017 struct kernfs_syscall_ops
*kf_sops
;
2018 struct css_set
*cset
;
2021 lockdep_assert_held(&cgroup_mutex
);
2023 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
2026 root_cgrp
->id
= ret
;
2027 root_cgrp
->ancestor_ids
[0] = ret
;
2029 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
2035 * We're accessing css_set_count without locking css_set_lock here,
2036 * but that's OK - it can only be increased by someone holding
2037 * cgroup_lock, and that's us. Later rebinding may disable
2038 * controllers on the default hierarchy and thus create new csets,
2039 * which can't be more than the existing ones. Allocate 2x.
2041 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
2045 ret
= cgroup_init_root_id(root
);
2049 kf_sops
= root
== &cgrp_dfl_root
?
2050 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
2052 root
->kf_root
= kernfs_create_root(kf_sops
,
2053 KERNFS_ROOT_CREATE_DEACTIVATED
|
2054 KERNFS_ROOT_SUPPORT_EXPORTOP
,
2056 if (IS_ERR(root
->kf_root
)) {
2057 ret
= PTR_ERR(root
->kf_root
);
2060 root_cgrp
->kn
= root
->kf_root
->kn
;
2062 ret
= css_populate_dir(&root_cgrp
->self
);
2066 ret
= rebind_subsystems(root
, ss_mask
);
2070 ret
= cgroup_bpf_inherit(root_cgrp
);
2073 trace_cgroup_setup_root(root
);
2076 * There must be no failure case after here, since rebinding takes
2077 * care of subsystems' refcounts, which are explicitly dropped in
2078 * the failure exit path.
2080 list_add(&root
->root_list
, &cgroup_roots
);
2081 cgroup_root_count
++;
2084 * Link the root cgroup in this hierarchy into all the css_set
2087 spin_lock_irq(&css_set_lock
);
2088 hash_for_each(css_set_table
, i
, cset
, hlist
) {
2089 link_css_set(&tmp_links
, cset
, root_cgrp
);
2090 if (css_set_populated(cset
))
2091 cgroup_update_populated(root_cgrp
, true);
2093 spin_unlock_irq(&css_set_lock
);
2095 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2096 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2098 kernfs_activate(root_cgrp
->kn
);
2103 kernfs_destroy_root(root
->kf_root
);
2104 root
->kf_root
= NULL
;
2106 cgroup_exit_root_id(root
);
2108 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2110 free_cgrp_cset_links(&tmp_links
);
2114 int cgroup_do_get_tree(struct fs_context
*fc
)
2116 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2119 ctx
->kfc
.root
= ctx
->root
->kf_root
;
2120 if (fc
->fs_type
== &cgroup2_fs_type
)
2121 ctx
->kfc
.magic
= CGROUP2_SUPER_MAGIC
;
2123 ctx
->kfc
.magic
= CGROUP_SUPER_MAGIC
;
2124 ret
= kernfs_get_tree(fc
);
2127 * In non-init cgroup namespace, instead of root cgroup's dentry,
2128 * we return the dentry corresponding to the cgroupns->root_cgrp.
2130 if (!ret
&& ctx
->ns
!= &init_cgroup_ns
) {
2131 struct dentry
*nsdentry
;
2132 struct super_block
*sb
= fc
->root
->d_sb
;
2133 struct cgroup
*cgrp
;
2135 mutex_lock(&cgroup_mutex
);
2136 spin_lock_irq(&css_set_lock
);
2138 cgrp
= cset_cgroup_from_root(ctx
->ns
->root_cset
, ctx
->root
);
2140 spin_unlock_irq(&css_set_lock
);
2141 mutex_unlock(&cgroup_mutex
);
2143 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
2145 if (IS_ERR(nsdentry
)) {
2146 deactivate_locked_super(sb
);
2147 ret
= PTR_ERR(nsdentry
);
2150 fc
->root
= nsdentry
;
2153 if (!ctx
->kfc
.new_sb_created
)
2154 cgroup_put(&ctx
->root
->cgrp
);
2160 * Destroy a cgroup filesystem context.
2162 static void cgroup_fs_context_free(struct fs_context
*fc
)
2164 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2167 kfree(ctx
->release_agent
);
2168 put_cgroup_ns(ctx
->ns
);
2169 kernfs_free_fs_context(fc
);
2173 static int cgroup_get_tree(struct fs_context
*fc
)
2175 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2178 cgrp_dfl_visible
= true;
2179 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2180 ctx
->root
= &cgrp_dfl_root
;
2182 ret
= cgroup_do_get_tree(fc
);
2184 apply_cgroup_root_flags(ctx
->flags
);
2188 static const struct fs_context_operations cgroup_fs_context_ops
= {
2189 .free
= cgroup_fs_context_free
,
2190 .parse_param
= cgroup2_parse_param
,
2191 .get_tree
= cgroup_get_tree
,
2192 .reconfigure
= cgroup_reconfigure
,
2195 static const struct fs_context_operations cgroup1_fs_context_ops
= {
2196 .free
= cgroup_fs_context_free
,
2197 .parse_param
= cgroup1_parse_param
,
2198 .get_tree
= cgroup1_get_tree
,
2199 .reconfigure
= cgroup1_reconfigure
,
2203 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2204 * we select the namespace we're going to use.
2206 static int cgroup_init_fs_context(struct fs_context
*fc
)
2208 struct cgroup_fs_context
*ctx
;
2210 ctx
= kzalloc(sizeof(struct cgroup_fs_context
), GFP_KERNEL
);
2215 * The first time anyone tries to mount a cgroup, enable the list
2216 * linking each css_set to its tasks and fix up all existing tasks.
2218 if (!use_task_css_set_links
)
2219 cgroup_enable_task_cg_lists();
2221 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
2222 get_cgroup_ns(ctx
->ns
);
2223 fc
->fs_private
= &ctx
->kfc
;
2224 if (fc
->fs_type
== &cgroup2_fs_type
)
2225 fc
->ops
= &cgroup_fs_context_ops
;
2227 fc
->ops
= &cgroup1_fs_context_ops
;
2228 put_user_ns(fc
->user_ns
);
2229 fc
->user_ns
= get_user_ns(ctx
->ns
->user_ns
);
2234 static void cgroup_kill_sb(struct super_block
*sb
)
2236 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2237 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2240 * If @root doesn't have any children, start killing it.
2241 * This prevents new mounts by disabling percpu_ref_tryget_live().
2242 * cgroup_mount() may wait for @root's release.
2244 * And don't kill the default root.
2246 if (list_empty(&root
->cgrp
.self
.children
) && root
!= &cgrp_dfl_root
&&
2247 !percpu_ref_is_dying(&root
->cgrp
.self
.refcnt
))
2248 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2249 cgroup_put(&root
->cgrp
);
2253 struct file_system_type cgroup_fs_type
= {
2255 .init_fs_context
= cgroup_init_fs_context
,
2256 .parameters
= &cgroup1_fs_parameters
,
2257 .kill_sb
= cgroup_kill_sb
,
2258 .fs_flags
= FS_USERNS_MOUNT
,
2261 static struct file_system_type cgroup2_fs_type
= {
2263 .init_fs_context
= cgroup_init_fs_context
,
2264 .parameters
= &cgroup2_fs_parameters
,
2265 .kill_sb
= cgroup_kill_sb
,
2266 .fs_flags
= FS_USERNS_MOUNT
,
2269 #ifdef CONFIG_CPUSETS
2270 static const struct fs_context_operations cpuset_fs_context_ops
= {
2271 .get_tree
= cgroup1_get_tree
,
2272 .free
= cgroup_fs_context_free
,
2276 * This is ugly, but preserves the userspace API for existing cpuset
2277 * users. If someone tries to mount the "cpuset" filesystem, we
2278 * silently switch it to mount "cgroup" instead
2280 static int cpuset_init_fs_context(struct fs_context
*fc
)
2282 char *agent
= kstrdup("/sbin/cpuset_release_agent", GFP_USER
);
2283 struct cgroup_fs_context
*ctx
;
2286 err
= cgroup_init_fs_context(fc
);
2292 fc
->ops
= &cpuset_fs_context_ops
;
2294 ctx
= cgroup_fc2context(fc
);
2295 ctx
->subsys_mask
= 1 << cpuset_cgrp_id
;
2296 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
2297 ctx
->release_agent
= agent
;
2299 get_filesystem(&cgroup_fs_type
);
2300 put_filesystem(fc
->fs_type
);
2301 fc
->fs_type
= &cgroup_fs_type
;
2306 static struct file_system_type cpuset_fs_type
= {
2308 .init_fs_context
= cpuset_init_fs_context
,
2309 .fs_flags
= FS_USERNS_MOUNT
,
2313 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2314 struct cgroup_namespace
*ns
)
2316 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2318 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2321 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2322 struct cgroup_namespace
*ns
)
2326 mutex_lock(&cgroup_mutex
);
2327 spin_lock_irq(&css_set_lock
);
2329 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2331 spin_unlock_irq(&css_set_lock
);
2332 mutex_unlock(&cgroup_mutex
);
2336 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2339 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2340 * @task: target task
2341 * @buf: the buffer to write the path into
2342 * @buflen: the length of the buffer
2344 * Determine @task's cgroup on the first (the one with the lowest non-zero
2345 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2346 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2347 * cgroup controller callbacks.
2349 * Return value is the same as kernfs_path().
2351 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2353 struct cgroup_root
*root
;
2354 struct cgroup
*cgrp
;
2355 int hierarchy_id
= 1;
2358 mutex_lock(&cgroup_mutex
);
2359 spin_lock_irq(&css_set_lock
);
2361 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2364 cgrp
= task_cgroup_from_root(task
, root
);
2365 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2367 /* if no hierarchy exists, everyone is in "/" */
2368 ret
= strlcpy(buf
, "/", buflen
);
2371 spin_unlock_irq(&css_set_lock
);
2372 mutex_unlock(&cgroup_mutex
);
2375 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2378 * cgroup_migrate_add_task - add a migration target task to a migration context
2379 * @task: target task
2380 * @mgctx: target migration context
2382 * Add @task, which is a migration target, to @mgctx->tset. This function
2383 * becomes noop if @task doesn't need to be migrated. @task's css_set
2384 * should have been added as a migration source and @task->cg_list will be
2385 * moved from the css_set's tasks list to mg_tasks one.
2387 static void cgroup_migrate_add_task(struct task_struct
*task
,
2388 struct cgroup_mgctx
*mgctx
)
2390 struct css_set
*cset
;
2392 lockdep_assert_held(&css_set_lock
);
2394 /* @task either already exited or can't exit until the end */
2395 if (task
->flags
& PF_EXITING
)
2398 /* leave @task alone if post_fork() hasn't linked it yet */
2399 if (list_empty(&task
->cg_list
))
2402 cset
= task_css_set(task
);
2403 if (!cset
->mg_src_cgrp
)
2406 mgctx
->tset
.nr_tasks
++;
2408 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2409 if (list_empty(&cset
->mg_node
))
2410 list_add_tail(&cset
->mg_node
,
2411 &mgctx
->tset
.src_csets
);
2412 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2413 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2414 &mgctx
->tset
.dst_csets
);
2418 * cgroup_taskset_first - reset taskset and return the first task
2419 * @tset: taskset of interest
2420 * @dst_cssp: output variable for the destination css
2422 * @tset iteration is initialized and the first task is returned.
2424 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2425 struct cgroup_subsys_state
**dst_cssp
)
2427 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2428 tset
->cur_task
= NULL
;
2430 return cgroup_taskset_next(tset
, dst_cssp
);
2434 * cgroup_taskset_next - iterate to the next task in taskset
2435 * @tset: taskset of interest
2436 * @dst_cssp: output variable for the destination css
2438 * Return the next task in @tset. Iteration must have been initialized
2439 * with cgroup_taskset_first().
2441 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2442 struct cgroup_subsys_state
**dst_cssp
)
2444 struct css_set
*cset
= tset
->cur_cset
;
2445 struct task_struct
*task
= tset
->cur_task
;
2447 while (&cset
->mg_node
!= tset
->csets
) {
2449 task
= list_first_entry(&cset
->mg_tasks
,
2450 struct task_struct
, cg_list
);
2452 task
= list_next_entry(task
, cg_list
);
2454 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2455 tset
->cur_cset
= cset
;
2456 tset
->cur_task
= task
;
2459 * This function may be called both before and
2460 * after cgroup_taskset_migrate(). The two cases
2461 * can be distinguished by looking at whether @cset
2462 * has its ->mg_dst_cset set.
2464 if (cset
->mg_dst_cset
)
2465 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2467 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2472 cset
= list_next_entry(cset
, mg_node
);
2480 * cgroup_taskset_migrate - migrate a taskset
2481 * @mgctx: migration context
2483 * Migrate tasks in @mgctx as setup by migration preparation functions.
2484 * This function fails iff one of the ->can_attach callbacks fails and
2485 * guarantees that either all or none of the tasks in @mgctx are migrated.
2486 * @mgctx is consumed regardless of success.
2488 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2490 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2491 struct cgroup_subsys
*ss
;
2492 struct task_struct
*task
, *tmp_task
;
2493 struct css_set
*cset
, *tmp_cset
;
2494 int ssid
, failed_ssid
, ret
;
2496 /* check that we can legitimately attach to the cgroup */
2497 if (tset
->nr_tasks
) {
2498 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2499 if (ss
->can_attach
) {
2501 ret
= ss
->can_attach(tset
);
2504 goto out_cancel_attach
;
2507 } while_each_subsys_mask();
2511 * Now that we're guaranteed success, proceed to move all tasks to
2512 * the new cgroup. There are no failure cases after here, so this
2513 * is the commit point.
2515 spin_lock_irq(&css_set_lock
);
2516 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2517 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2518 struct css_set
*from_cset
= task_css_set(task
);
2519 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2521 get_css_set(to_cset
);
2522 to_cset
->nr_tasks
++;
2523 css_set_move_task(task
, from_cset
, to_cset
, true);
2524 from_cset
->nr_tasks
--;
2526 * If the source or destination cgroup is frozen,
2527 * the task might require to change its state.
2529 cgroup_freezer_migrate_task(task
, from_cset
->dfl_cgrp
,
2531 put_css_set_locked(from_cset
);
2535 spin_unlock_irq(&css_set_lock
);
2538 * Migration is committed, all target tasks are now on dst_csets.
2539 * Nothing is sensitive to fork() after this point. Notify
2540 * controllers that migration is complete.
2542 tset
->csets
= &tset
->dst_csets
;
2544 if (tset
->nr_tasks
) {
2545 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2550 } while_each_subsys_mask();
2554 goto out_release_tset
;
2557 if (tset
->nr_tasks
) {
2558 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2559 if (ssid
== failed_ssid
)
2561 if (ss
->cancel_attach
) {
2563 ss
->cancel_attach(tset
);
2565 } while_each_subsys_mask();
2568 spin_lock_irq(&css_set_lock
);
2569 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2570 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2571 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2572 list_del_init(&cset
->mg_node
);
2574 spin_unlock_irq(&css_set_lock
);
2577 * Re-initialize the cgroup_taskset structure in case it is reused
2578 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2582 tset
->csets
= &tset
->src_csets
;
2587 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2588 * @dst_cgrp: destination cgroup to test
2590 * On the default hierarchy, except for the mixable, (possible) thread root
2591 * and threaded cgroups, subtree_control must be zero for migration
2592 * destination cgroups with tasks so that child cgroups don't compete
2595 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2597 /* v1 doesn't have any restriction */
2598 if (!cgroup_on_dfl(dst_cgrp
))
2601 /* verify @dst_cgrp can host resources */
2602 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2605 /* mixables don't care */
2606 if (cgroup_is_mixable(dst_cgrp
))
2610 * If @dst_cgrp is already or can become a thread root or is
2611 * threaded, it doesn't matter.
2613 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2616 /* apply no-internal-process constraint */
2617 if (dst_cgrp
->subtree_control
)
2624 * cgroup_migrate_finish - cleanup after attach
2625 * @mgctx: migration context
2627 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2628 * those functions for details.
2630 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2632 LIST_HEAD(preloaded
);
2633 struct css_set
*cset
, *tmp_cset
;
2635 lockdep_assert_held(&cgroup_mutex
);
2637 spin_lock_irq(&css_set_lock
);
2639 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2640 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2642 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2643 cset
->mg_src_cgrp
= NULL
;
2644 cset
->mg_dst_cgrp
= NULL
;
2645 cset
->mg_dst_cset
= NULL
;
2646 list_del_init(&cset
->mg_preload_node
);
2647 put_css_set_locked(cset
);
2650 spin_unlock_irq(&css_set_lock
);
2654 * cgroup_migrate_add_src - add a migration source css_set
2655 * @src_cset: the source css_set to add
2656 * @dst_cgrp: the destination cgroup
2657 * @mgctx: migration context
2659 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2660 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2661 * up by cgroup_migrate_finish().
2663 * This function may be called without holding cgroup_threadgroup_rwsem
2664 * even if the target is a process. Threads may be created and destroyed
2665 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2666 * into play and the preloaded css_sets are guaranteed to cover all
2669 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2670 struct cgroup
*dst_cgrp
,
2671 struct cgroup_mgctx
*mgctx
)
2673 struct cgroup
*src_cgrp
;
2675 lockdep_assert_held(&cgroup_mutex
);
2676 lockdep_assert_held(&css_set_lock
);
2679 * If ->dead, @src_set is associated with one or more dead cgroups
2680 * and doesn't contain any migratable tasks. Ignore it early so
2681 * that the rest of migration path doesn't get confused by it.
2686 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2688 if (!list_empty(&src_cset
->mg_preload_node
))
2691 WARN_ON(src_cset
->mg_src_cgrp
);
2692 WARN_ON(src_cset
->mg_dst_cgrp
);
2693 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2694 WARN_ON(!list_empty(&src_cset
->mg_node
));
2696 src_cset
->mg_src_cgrp
= src_cgrp
;
2697 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2698 get_css_set(src_cset
);
2699 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2703 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2704 * @mgctx: migration context
2706 * Tasks are about to be moved and all the source css_sets have been
2707 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2708 * pins all destination css_sets, links each to its source, and append them
2709 * to @mgctx->preloaded_dst_csets.
2711 * This function must be called after cgroup_migrate_add_src() has been
2712 * called on each migration source css_set. After migration is performed
2713 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2716 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2718 struct css_set
*src_cset
, *tmp_cset
;
2720 lockdep_assert_held(&cgroup_mutex
);
2722 /* look up the dst cset for each src cset and link it to src */
2723 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2725 struct css_set
*dst_cset
;
2726 struct cgroup_subsys
*ss
;
2729 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2733 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2736 * If src cset equals dst, it's noop. Drop the src.
2737 * cgroup_migrate() will skip the cset too. Note that we
2738 * can't handle src == dst as some nodes are used by both.
2740 if (src_cset
== dst_cset
) {
2741 src_cset
->mg_src_cgrp
= NULL
;
2742 src_cset
->mg_dst_cgrp
= NULL
;
2743 list_del_init(&src_cset
->mg_preload_node
);
2744 put_css_set(src_cset
);
2745 put_css_set(dst_cset
);
2749 src_cset
->mg_dst_cset
= dst_cset
;
2751 if (list_empty(&dst_cset
->mg_preload_node
))
2752 list_add_tail(&dst_cset
->mg_preload_node
,
2753 &mgctx
->preloaded_dst_csets
);
2755 put_css_set(dst_cset
);
2757 for_each_subsys(ss
, ssid
)
2758 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2759 mgctx
->ss_mask
|= 1 << ssid
;
2766 * cgroup_migrate - migrate a process or task to a cgroup
2767 * @leader: the leader of the process or the task to migrate
2768 * @threadgroup: whether @leader points to the whole process or a single task
2769 * @mgctx: migration context
2771 * Migrate a process or task denoted by @leader. If migrating a process,
2772 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2773 * responsible for invoking cgroup_migrate_add_src() and
2774 * cgroup_migrate_prepare_dst() on the targets before invoking this
2775 * function and following up with cgroup_migrate_finish().
2777 * As long as a controller's ->can_attach() doesn't fail, this function is
2778 * guaranteed to succeed. This means that, excluding ->can_attach()
2779 * failure, when migrating multiple targets, the success or failure can be
2780 * decided for all targets by invoking group_migrate_prepare_dst() before
2781 * actually starting migrating.
2783 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2784 struct cgroup_mgctx
*mgctx
)
2786 struct task_struct
*task
;
2789 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2790 * already PF_EXITING could be freed from underneath us unless we
2791 * take an rcu_read_lock.
2793 spin_lock_irq(&css_set_lock
);
2797 cgroup_migrate_add_task(task
, mgctx
);
2800 } while_each_thread(leader
, task
);
2802 spin_unlock_irq(&css_set_lock
);
2804 return cgroup_migrate_execute(mgctx
);
2808 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2809 * @dst_cgrp: the cgroup to attach to
2810 * @leader: the task or the leader of the threadgroup to be attached
2811 * @threadgroup: attach the whole threadgroup?
2813 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2815 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2818 DEFINE_CGROUP_MGCTX(mgctx
);
2819 struct task_struct
*task
;
2822 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
2826 /* look up all src csets */
2827 spin_lock_irq(&css_set_lock
);
2831 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2834 } while_each_thread(leader
, task
);
2836 spin_unlock_irq(&css_set_lock
);
2838 /* prepare dst csets and commit */
2839 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2841 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2843 cgroup_migrate_finish(&mgctx
);
2846 TRACE_CGROUP_PATH(attach_task
, dst_cgrp
, leader
, threadgroup
);
2851 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
)
2852 __acquires(&cgroup_threadgroup_rwsem
)
2854 struct task_struct
*tsk
;
2857 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2858 return ERR_PTR(-EINVAL
);
2860 percpu_down_write(&cgroup_threadgroup_rwsem
);
2864 tsk
= find_task_by_vpid(pid
);
2866 tsk
= ERR_PTR(-ESRCH
);
2867 goto out_unlock_threadgroup
;
2874 tsk
= tsk
->group_leader
;
2877 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2878 * If userland migrates such a kthread to a non-root cgroup, it can
2879 * become trapped in a cpuset, or RT kthread may be born in a
2880 * cgroup with no rt_runtime allocated. Just say no.
2882 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2883 tsk
= ERR_PTR(-EINVAL
);
2884 goto out_unlock_threadgroup
;
2887 get_task_struct(tsk
);
2888 goto out_unlock_rcu
;
2890 out_unlock_threadgroup
:
2891 percpu_up_write(&cgroup_threadgroup_rwsem
);
2897 void cgroup_procs_write_finish(struct task_struct
*task
)
2898 __releases(&cgroup_threadgroup_rwsem
)
2900 struct cgroup_subsys
*ss
;
2903 /* release reference from cgroup_procs_write_start() */
2904 put_task_struct(task
);
2906 percpu_up_write(&cgroup_threadgroup_rwsem
);
2907 for_each_subsys(ss
, ssid
)
2908 if (ss
->post_attach
)
2912 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2914 struct cgroup_subsys
*ss
;
2915 bool printed
= false;
2918 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2921 seq_puts(seq
, ss
->name
);
2923 } while_each_subsys_mask();
2925 seq_putc(seq
, '\n');
2928 /* show controllers which are enabled from the parent */
2929 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2931 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2933 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2937 /* show controllers which are enabled for a given cgroup's children */
2938 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2940 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2942 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2947 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2948 * @cgrp: root of the subtree to update csses for
2950 * @cgrp's control masks have changed and its subtree's css associations
2951 * need to be updated accordingly. This function looks up all css_sets
2952 * which are attached to the subtree, creates the matching updated css_sets
2953 * and migrates the tasks to the new ones.
2955 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2957 DEFINE_CGROUP_MGCTX(mgctx
);
2958 struct cgroup_subsys_state
*d_css
;
2959 struct cgroup
*dsct
;
2960 struct css_set
*src_cset
;
2963 lockdep_assert_held(&cgroup_mutex
);
2965 percpu_down_write(&cgroup_threadgroup_rwsem
);
2967 /* look up all csses currently attached to @cgrp's subtree */
2968 spin_lock_irq(&css_set_lock
);
2969 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2970 struct cgrp_cset_link
*link
;
2972 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2973 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2975 spin_unlock_irq(&css_set_lock
);
2977 /* NULL dst indicates self on default hierarchy */
2978 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2982 spin_lock_irq(&css_set_lock
);
2983 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2984 struct task_struct
*task
, *ntask
;
2986 /* all tasks in src_csets need to be migrated */
2987 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2988 cgroup_migrate_add_task(task
, &mgctx
);
2990 spin_unlock_irq(&css_set_lock
);
2992 ret
= cgroup_migrate_execute(&mgctx
);
2994 cgroup_migrate_finish(&mgctx
);
2995 percpu_up_write(&cgroup_threadgroup_rwsem
);
3000 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3001 * @cgrp: root of the target subtree
3003 * Because css offlining is asynchronous, userland may try to re-enable a
3004 * controller while the previous css is still around. This function grabs
3005 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3007 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
3008 __acquires(&cgroup_mutex
)
3010 struct cgroup
*dsct
;
3011 struct cgroup_subsys_state
*d_css
;
3012 struct cgroup_subsys
*ss
;
3016 mutex_lock(&cgroup_mutex
);
3018 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3019 for_each_subsys(ss
, ssid
) {
3020 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3023 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
3026 cgroup_get_live(dsct
);
3027 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
3028 TASK_UNINTERRUPTIBLE
);
3030 mutex_unlock(&cgroup_mutex
);
3032 finish_wait(&dsct
->offline_waitq
, &wait
);
3041 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3042 * @cgrp: root of the target subtree
3044 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3045 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3048 static void cgroup_save_control(struct cgroup
*cgrp
)
3050 struct cgroup
*dsct
;
3051 struct cgroup_subsys_state
*d_css
;
3053 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3054 dsct
->old_subtree_control
= dsct
->subtree_control
;
3055 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
3056 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
3061 * cgroup_propagate_control - refresh control masks of a subtree
3062 * @cgrp: root of the target subtree
3064 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3065 * ->subtree_control and propagate controller availability through the
3066 * subtree so that descendants don't have unavailable controllers enabled.
3068 static void cgroup_propagate_control(struct cgroup
*cgrp
)
3070 struct cgroup
*dsct
;
3071 struct cgroup_subsys_state
*d_css
;
3073 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3074 dsct
->subtree_control
&= cgroup_control(dsct
);
3075 dsct
->subtree_ss_mask
=
3076 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
3077 cgroup_ss_mask(dsct
));
3082 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3083 * @cgrp: root of the target subtree
3085 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3086 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3089 static void cgroup_restore_control(struct cgroup
*cgrp
)
3091 struct cgroup
*dsct
;
3092 struct cgroup_subsys_state
*d_css
;
3094 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3095 dsct
->subtree_control
= dsct
->old_subtree_control
;
3096 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3097 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
3101 static bool css_visible(struct cgroup_subsys_state
*css
)
3103 struct cgroup_subsys
*ss
= css
->ss
;
3104 struct cgroup
*cgrp
= css
->cgroup
;
3106 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3108 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3110 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3114 * cgroup_apply_control_enable - enable or show csses according to control
3115 * @cgrp: root of the target subtree
3117 * Walk @cgrp's subtree and create new csses or make the existing ones
3118 * visible. A css is created invisible if it's being implicitly enabled
3119 * through dependency. An invisible css is made visible when the userland
3120 * explicitly enables it.
3122 * Returns 0 on success, -errno on failure. On failure, csses which have
3123 * been processed already aren't cleaned up. The caller is responsible for
3124 * cleaning up with cgroup_apply_control_disable().
3126 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3128 struct cgroup
*dsct
;
3129 struct cgroup_subsys_state
*d_css
;
3130 struct cgroup_subsys
*ss
;
3133 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3134 for_each_subsys(ss
, ssid
) {
3135 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3137 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3141 css
= css_create(dsct
, ss
);
3143 return PTR_ERR(css
);
3146 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3148 if (css_visible(css
)) {
3149 ret
= css_populate_dir(css
);
3160 * cgroup_apply_control_disable - kill or hide csses according to control
3161 * @cgrp: root of the target subtree
3163 * Walk @cgrp's subtree and kill and hide csses so that they match
3164 * cgroup_ss_mask() and cgroup_visible_mask().
3166 * A css is hidden when the userland requests it to be disabled while other
3167 * subsystems are still depending on it. The css must not actively control
3168 * resources and be in the vanilla state if it's made visible again later.
3169 * Controllers which may be depended upon should provide ->css_reset() for
3172 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3174 struct cgroup
*dsct
;
3175 struct cgroup_subsys_state
*d_css
;
3176 struct cgroup_subsys
*ss
;
3179 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3180 for_each_subsys(ss
, ssid
) {
3181 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3186 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3189 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3191 } else if (!css_visible(css
)) {
3201 * cgroup_apply_control - apply control mask updates to the subtree
3202 * @cgrp: root of the target subtree
3204 * subsystems can be enabled and disabled in a subtree using the following
3207 * 1. Call cgroup_save_control() to stash the current state.
3208 * 2. Update ->subtree_control masks in the subtree as desired.
3209 * 3. Call cgroup_apply_control() to apply the changes.
3210 * 4. Optionally perform other related operations.
3211 * 5. Call cgroup_finalize_control() to finish up.
3213 * This function implements step 3 and propagates the mask changes
3214 * throughout @cgrp's subtree, updates csses accordingly and perform
3215 * process migrations.
3217 static int cgroup_apply_control(struct cgroup
*cgrp
)
3221 cgroup_propagate_control(cgrp
);
3223 ret
= cgroup_apply_control_enable(cgrp
);
3228 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3229 * making the following cgroup_update_dfl_csses() properly update
3230 * css associations of all tasks in the subtree.
3232 ret
= cgroup_update_dfl_csses(cgrp
);
3240 * cgroup_finalize_control - finalize control mask update
3241 * @cgrp: root of the target subtree
3242 * @ret: the result of the update
3244 * Finalize control mask update. See cgroup_apply_control() for more info.
3246 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3249 cgroup_restore_control(cgrp
);
3250 cgroup_propagate_control(cgrp
);
3253 cgroup_apply_control_disable(cgrp
);
3256 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3258 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3260 /* if nothing is getting enabled, nothing to worry about */
3264 /* can @cgrp host any resources? */
3265 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3268 /* mixables don't care */
3269 if (cgroup_is_mixable(cgrp
))
3272 if (domain_enable
) {
3273 /* can't enable domain controllers inside a thread subtree */
3274 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3278 * Threaded controllers can handle internal competitions
3279 * and are always allowed inside a (prospective) thread
3282 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3287 * Controllers can't be enabled for a cgroup with tasks to avoid
3288 * child cgroups competing against tasks.
3290 if (cgroup_has_tasks(cgrp
))
3296 /* change the enabled child controllers for a cgroup in the default hierarchy */
3297 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3298 char *buf
, size_t nbytes
,
3301 u16 enable
= 0, disable
= 0;
3302 struct cgroup
*cgrp
, *child
;
3303 struct cgroup_subsys
*ss
;
3308 * Parse input - space separated list of subsystem names prefixed
3309 * with either + or -.
3311 buf
= strstrip(buf
);
3312 while ((tok
= strsep(&buf
, " "))) {
3315 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3316 if (!cgroup_ssid_enabled(ssid
) ||
3317 strcmp(tok
+ 1, ss
->name
))
3321 enable
|= 1 << ssid
;
3322 disable
&= ~(1 << ssid
);
3323 } else if (*tok
== '-') {
3324 disable
|= 1 << ssid
;
3325 enable
&= ~(1 << ssid
);
3330 } while_each_subsys_mask();
3331 if (ssid
== CGROUP_SUBSYS_COUNT
)
3335 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3339 for_each_subsys(ss
, ssid
) {
3340 if (enable
& (1 << ssid
)) {
3341 if (cgrp
->subtree_control
& (1 << ssid
)) {
3342 enable
&= ~(1 << ssid
);
3346 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3350 } else if (disable
& (1 << ssid
)) {
3351 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3352 disable
&= ~(1 << ssid
);
3356 /* a child has it enabled? */
3357 cgroup_for_each_live_child(child
, cgrp
) {
3358 if (child
->subtree_control
& (1 << ssid
)) {
3366 if (!enable
&& !disable
) {
3371 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3375 /* save and update control masks and prepare csses */
3376 cgroup_save_control(cgrp
);
3378 cgrp
->subtree_control
|= enable
;
3379 cgrp
->subtree_control
&= ~disable
;
3381 ret
= cgroup_apply_control(cgrp
);
3382 cgroup_finalize_control(cgrp
, ret
);
3386 kernfs_activate(cgrp
->kn
);
3388 cgroup_kn_unlock(of
->kn
);
3389 return ret
?: nbytes
;
3393 * cgroup_enable_threaded - make @cgrp threaded
3394 * @cgrp: the target cgroup
3396 * Called when "threaded" is written to the cgroup.type interface file and
3397 * tries to make @cgrp threaded and join the parent's resource domain.
3398 * This function is never called on the root cgroup as cgroup.type doesn't
3401 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3403 struct cgroup
*parent
= cgroup_parent(cgrp
);
3404 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3405 struct cgroup
*dsct
;
3406 struct cgroup_subsys_state
*d_css
;
3409 lockdep_assert_held(&cgroup_mutex
);
3411 /* noop if already threaded */
3412 if (cgroup_is_threaded(cgrp
))
3416 * If @cgroup is populated or has domain controllers enabled, it
3417 * can't be switched. While the below cgroup_can_be_thread_root()
3418 * test can catch the same conditions, that's only when @parent is
3419 * not mixable, so let's check it explicitly.
3421 if (cgroup_is_populated(cgrp
) ||
3422 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3425 /* we're joining the parent's domain, ensure its validity */
3426 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3427 !cgroup_can_be_thread_root(dom_cgrp
))
3431 * The following shouldn't cause actual migrations and should
3434 cgroup_save_control(cgrp
);
3436 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3437 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3438 dsct
->dom_cgrp
= dom_cgrp
;
3440 ret
= cgroup_apply_control(cgrp
);
3442 parent
->nr_threaded_children
++;
3444 cgroup_finalize_control(cgrp
, ret
);
3448 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3450 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3452 if (cgroup_is_threaded(cgrp
))
3453 seq_puts(seq
, "threaded\n");
3454 else if (!cgroup_is_valid_domain(cgrp
))
3455 seq_puts(seq
, "domain invalid\n");
3456 else if (cgroup_is_thread_root(cgrp
))
3457 seq_puts(seq
, "domain threaded\n");
3459 seq_puts(seq
, "domain\n");
3464 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3465 size_t nbytes
, loff_t off
)
3467 struct cgroup
*cgrp
;
3470 /* only switching to threaded mode is supported */
3471 if (strcmp(strstrip(buf
), "threaded"))
3474 /* drain dying csses before we re-apply (threaded) subtree control */
3475 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3479 /* threaded can only be enabled */
3480 ret
= cgroup_enable_threaded(cgrp
);
3482 cgroup_kn_unlock(of
->kn
);
3483 return ret
?: nbytes
;
3486 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3488 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3489 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3491 if (descendants
== INT_MAX
)
3492 seq_puts(seq
, "max\n");
3494 seq_printf(seq
, "%d\n", descendants
);
3499 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3500 char *buf
, size_t nbytes
, loff_t off
)
3502 struct cgroup
*cgrp
;
3506 buf
= strstrip(buf
);
3507 if (!strcmp(buf
, "max")) {
3508 descendants
= INT_MAX
;
3510 ret
= kstrtoint(buf
, 0, &descendants
);
3515 if (descendants
< 0)
3518 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3522 cgrp
->max_descendants
= descendants
;
3524 cgroup_kn_unlock(of
->kn
);
3529 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3531 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3532 int depth
= READ_ONCE(cgrp
->max_depth
);
3534 if (depth
== INT_MAX
)
3535 seq_puts(seq
, "max\n");
3537 seq_printf(seq
, "%d\n", depth
);
3542 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3543 char *buf
, size_t nbytes
, loff_t off
)
3545 struct cgroup
*cgrp
;
3549 buf
= strstrip(buf
);
3550 if (!strcmp(buf
, "max")) {
3553 ret
= kstrtoint(buf
, 0, &depth
);
3561 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3565 cgrp
->max_depth
= depth
;
3567 cgroup_kn_unlock(of
->kn
);
3572 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3574 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3576 seq_printf(seq
, "populated %d\n", cgroup_is_populated(cgrp
));
3577 seq_printf(seq
, "frozen %d\n", test_bit(CGRP_FROZEN
, &cgrp
->flags
));
3582 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3584 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3586 seq_printf(seq
, "nr_descendants %d\n",
3587 cgroup
->nr_descendants
);
3588 seq_printf(seq
, "nr_dying_descendants %d\n",
3589 cgroup
->nr_dying_descendants
);
3594 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3595 struct cgroup
*cgrp
, int ssid
)
3597 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3598 struct cgroup_subsys_state
*css
;
3601 if (!ss
->css_extra_stat_show
)
3604 css
= cgroup_tryget_css(cgrp
, ss
);
3608 ret
= ss
->css_extra_stat_show(seq
, css
);
3613 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3615 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3618 cgroup_base_stat_cputime_show(seq
);
3619 #ifdef CONFIG_CGROUP_SCHED
3620 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3626 static int cgroup_io_pressure_show(struct seq_file
*seq
, void *v
)
3628 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3629 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3631 return psi_show(seq
, psi
, PSI_IO
);
3633 static int cgroup_memory_pressure_show(struct seq_file
*seq
, void *v
)
3635 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3636 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3638 return psi_show(seq
, psi
, PSI_MEM
);
3640 static int cgroup_cpu_pressure_show(struct seq_file
*seq
, void *v
)
3642 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3643 struct psi_group
*psi
= cgroup
->id
== 1 ? &psi_system
: &cgroup
->psi
;
3645 return psi_show(seq
, psi
, PSI_CPU
);
3648 static ssize_t
cgroup_pressure_write(struct kernfs_open_file
*of
, char *buf
,
3649 size_t nbytes
, enum psi_res res
)
3651 struct cgroup_file_ctx
*ctx
= of
->priv
;
3652 struct psi_trigger
*new;
3653 struct cgroup
*cgrp
;
3654 struct psi_group
*psi
;
3656 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3661 cgroup_kn_unlock(of
->kn
);
3663 /* Allow only one trigger per file descriptor */
3664 if (ctx
->psi
.trigger
) {
3669 psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3670 new = psi_trigger_create(psi
, buf
, nbytes
, res
);
3673 return PTR_ERR(new);
3676 smp_store_release(&ctx
->psi
.trigger
, new);
3682 static ssize_t
cgroup_io_pressure_write(struct kernfs_open_file
*of
,
3683 char *buf
, size_t nbytes
,
3686 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_IO
);
3689 static ssize_t
cgroup_memory_pressure_write(struct kernfs_open_file
*of
,
3690 char *buf
, size_t nbytes
,
3693 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_MEM
);
3696 static ssize_t
cgroup_cpu_pressure_write(struct kernfs_open_file
*of
,
3697 char *buf
, size_t nbytes
,
3700 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_CPU
);
3703 static __poll_t
cgroup_pressure_poll(struct kernfs_open_file
*of
,
3706 struct cgroup_file_ctx
*ctx
= of
->priv
;
3707 return psi_trigger_poll(&ctx
->psi
.trigger
, of
->file
, pt
);
3710 static void cgroup_pressure_release(struct kernfs_open_file
*of
)
3712 struct cgroup_file_ctx
*ctx
= of
->priv
;
3714 psi_trigger_destroy(ctx
->psi
.trigger
);
3716 #endif /* CONFIG_PSI */
3718 static int cgroup_freeze_show(struct seq_file
*seq
, void *v
)
3720 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3722 seq_printf(seq
, "%d\n", cgrp
->freezer
.freeze
);
3727 static ssize_t
cgroup_freeze_write(struct kernfs_open_file
*of
,
3728 char *buf
, size_t nbytes
, loff_t off
)
3730 struct cgroup
*cgrp
;
3734 ret
= kstrtoint(strstrip(buf
), 0, &freeze
);
3738 if (freeze
< 0 || freeze
> 1)
3741 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3745 cgroup_freeze(cgrp
, freeze
);
3747 cgroup_kn_unlock(of
->kn
);
3752 static int cgroup_file_open(struct kernfs_open_file
*of
)
3754 struct cftype
*cft
= of
->kn
->priv
;
3755 struct cgroup_file_ctx
*ctx
;
3758 ctx
= kzalloc(sizeof(*ctx
), GFP_KERNEL
);
3766 ret
= cft
->open(of
);
3772 static void cgroup_file_release(struct kernfs_open_file
*of
)
3774 struct cftype
*cft
= of
->kn
->priv
;
3775 struct cgroup_file_ctx
*ctx
= of
->priv
;
3782 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3783 size_t nbytes
, loff_t off
)
3785 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3786 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3787 struct cftype
*cft
= of
->kn
->priv
;
3788 struct cgroup_subsys_state
*css
;
3792 * If namespaces are delegation boundaries, disallow writes to
3793 * files in an non-init namespace root from inside the namespace
3794 * except for the files explicitly marked delegatable -
3795 * cgroup.procs and cgroup.subtree_control.
3797 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3798 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3799 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3803 return cft
->write(of
, buf
, nbytes
, off
);
3806 * kernfs guarantees that a file isn't deleted with operations in
3807 * flight, which means that the matching css is and stays alive and
3808 * doesn't need to be pinned. The RCU locking is not necessary
3809 * either. It's just for the convenience of using cgroup_css().
3812 css
= cgroup_css(cgrp
, cft
->ss
);
3815 if (cft
->write_u64
) {
3816 unsigned long long v
;
3817 ret
= kstrtoull(buf
, 0, &v
);
3819 ret
= cft
->write_u64(css
, cft
, v
);
3820 } else if (cft
->write_s64
) {
3822 ret
= kstrtoll(buf
, 0, &v
);
3824 ret
= cft
->write_s64(css
, cft
, v
);
3829 return ret
?: nbytes
;
3832 static __poll_t
cgroup_file_poll(struct kernfs_open_file
*of
, poll_table
*pt
)
3834 struct cftype
*cft
= of
->kn
->priv
;
3837 return cft
->poll(of
, pt
);
3839 return kernfs_generic_poll(of
, pt
);
3842 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3844 return seq_cft(seq
)->seq_start(seq
, ppos
);
3847 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3849 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3852 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3854 if (seq_cft(seq
)->seq_stop
)
3855 seq_cft(seq
)->seq_stop(seq
, v
);
3858 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3860 struct cftype
*cft
= seq_cft(m
);
3861 struct cgroup_subsys_state
*css
= seq_css(m
);
3864 return cft
->seq_show(m
, arg
);
3867 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3868 else if (cft
->read_s64
)
3869 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3875 static struct kernfs_ops cgroup_kf_single_ops
= {
3876 .atomic_write_len
= PAGE_SIZE
,
3877 .open
= cgroup_file_open
,
3878 .release
= cgroup_file_release
,
3879 .write
= cgroup_file_write
,
3880 .poll
= cgroup_file_poll
,
3881 .seq_show
= cgroup_seqfile_show
,
3884 static struct kernfs_ops cgroup_kf_ops
= {
3885 .atomic_write_len
= PAGE_SIZE
,
3886 .open
= cgroup_file_open
,
3887 .release
= cgroup_file_release
,
3888 .write
= cgroup_file_write
,
3889 .poll
= cgroup_file_poll
,
3890 .seq_start
= cgroup_seqfile_start
,
3891 .seq_next
= cgroup_seqfile_next
,
3892 .seq_stop
= cgroup_seqfile_stop
,
3893 .seq_show
= cgroup_seqfile_show
,
3896 /* set uid and gid of cgroup dirs and files to that of the creator */
3897 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3899 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3900 .ia_uid
= current_fsuid(),
3901 .ia_gid
= current_fsgid(), };
3903 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3904 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3907 return kernfs_setattr(kn
, &iattr
);
3910 static void cgroup_file_notify_timer(struct timer_list
*timer
)
3912 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
3916 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3919 char name
[CGROUP_FILE_NAME_MAX
];
3920 struct kernfs_node
*kn
;
3921 struct lock_class_key
*key
= NULL
;
3924 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3925 key
= &cft
->lockdep_key
;
3927 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3928 cgroup_file_mode(cft
),
3929 GLOBAL_ROOT_UID
, GLOBAL_ROOT_GID
,
3930 0, cft
->kf_ops
, cft
,
3935 ret
= cgroup_kn_set_ugid(kn
);
3941 if (cft
->file_offset
) {
3942 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3944 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
3946 spin_lock_irq(&cgroup_file_kn_lock
);
3948 spin_unlock_irq(&cgroup_file_kn_lock
);
3955 * cgroup_addrm_files - add or remove files to a cgroup directory
3956 * @css: the target css
3957 * @cgrp: the target cgroup (usually css->cgroup)
3958 * @cfts: array of cftypes to be added
3959 * @is_add: whether to add or remove
3961 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3962 * For removals, this function never fails.
3964 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3965 struct cgroup
*cgrp
, struct cftype cfts
[],
3968 struct cftype
*cft
, *cft_end
= NULL
;
3971 lockdep_assert_held(&cgroup_mutex
);
3974 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3975 /* does cft->flags tell us to skip this file on @cgrp? */
3976 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3978 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3980 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3982 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3984 if ((cft
->flags
& CFTYPE_DEBUG
) && !cgroup_debug
)
3987 ret
= cgroup_add_file(css
, cgrp
, cft
);
3989 pr_warn("%s: failed to add %s, err=%d\n",
3990 __func__
, cft
->name
, ret
);
3996 cgroup_rm_file(cgrp
, cft
);
4002 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
4004 struct cgroup_subsys
*ss
= cfts
[0].ss
;
4005 struct cgroup
*root
= &ss
->root
->cgrp
;
4006 struct cgroup_subsys_state
*css
;
4009 lockdep_assert_held(&cgroup_mutex
);
4011 /* add/rm files for all cgroups created before */
4012 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
4013 struct cgroup
*cgrp
= css
->cgroup
;
4015 if (!(css
->flags
& CSS_VISIBLE
))
4018 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
4024 kernfs_activate(root
->kn
);
4028 static void cgroup_exit_cftypes(struct cftype
*cfts
)
4032 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4033 /* free copy for custom atomic_write_len, see init_cftypes() */
4034 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
4039 /* revert flags set by cgroup core while adding @cfts */
4040 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
4044 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4048 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4049 struct kernfs_ops
*kf_ops
;
4051 WARN_ON(cft
->ss
|| cft
->kf_ops
);
4054 kf_ops
= &cgroup_kf_ops
;
4056 kf_ops
= &cgroup_kf_single_ops
;
4059 * Ugh... if @cft wants a custom max_write_len, we need to
4060 * make a copy of kf_ops to set its atomic_write_len.
4062 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
4063 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
4065 cgroup_exit_cftypes(cfts
);
4068 kf_ops
->atomic_write_len
= cft
->max_write_len
;
4071 cft
->kf_ops
= kf_ops
;
4078 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
4080 lockdep_assert_held(&cgroup_mutex
);
4082 if (!cfts
|| !cfts
[0].ss
)
4085 list_del(&cfts
->node
);
4086 cgroup_apply_cftypes(cfts
, false);
4087 cgroup_exit_cftypes(cfts
);
4092 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4093 * @cfts: zero-length name terminated array of cftypes
4095 * Unregister @cfts. Files described by @cfts are removed from all
4096 * existing cgroups and all future cgroups won't have them either. This
4097 * function can be called anytime whether @cfts' subsys is attached or not.
4099 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4102 int cgroup_rm_cftypes(struct cftype
*cfts
)
4106 mutex_lock(&cgroup_mutex
);
4107 ret
= cgroup_rm_cftypes_locked(cfts
);
4108 mutex_unlock(&cgroup_mutex
);
4113 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4114 * @ss: target cgroup subsystem
4115 * @cfts: zero-length name terminated array of cftypes
4117 * Register @cfts to @ss. Files described by @cfts are created for all
4118 * existing cgroups to which @ss is attached and all future cgroups will
4119 * have them too. This function can be called anytime whether @ss is
4122 * Returns 0 on successful registration, -errno on failure. Note that this
4123 * function currently returns 0 as long as @cfts registration is successful
4124 * even if some file creation attempts on existing cgroups fail.
4126 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4130 if (!cgroup_ssid_enabled(ss
->id
))
4133 if (!cfts
|| cfts
[0].name
[0] == '\0')
4136 ret
= cgroup_init_cftypes(ss
, cfts
);
4140 mutex_lock(&cgroup_mutex
);
4142 list_add_tail(&cfts
->node
, &ss
->cfts
);
4143 ret
= cgroup_apply_cftypes(cfts
, true);
4145 cgroup_rm_cftypes_locked(cfts
);
4147 mutex_unlock(&cgroup_mutex
);
4152 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4153 * @ss: target cgroup subsystem
4154 * @cfts: zero-length name terminated array of cftypes
4156 * Similar to cgroup_add_cftypes() but the added files are only used for
4157 * the default hierarchy.
4159 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4163 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4164 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
4165 return cgroup_add_cftypes(ss
, cfts
);
4169 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4170 * @ss: target cgroup subsystem
4171 * @cfts: zero-length name terminated array of cftypes
4173 * Similar to cgroup_add_cftypes() but the added files are only used for
4174 * the legacy hierarchies.
4176 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4180 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4181 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
4182 return cgroup_add_cftypes(ss
, cfts
);
4186 * cgroup_file_notify - generate a file modified event for a cgroup_file
4187 * @cfile: target cgroup_file
4189 * @cfile must have been obtained by setting cftype->file_offset.
4191 void cgroup_file_notify(struct cgroup_file
*cfile
)
4193 unsigned long flags
;
4195 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
4197 unsigned long last
= cfile
->notified_at
;
4198 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
4200 if (time_in_range(jiffies
, last
, next
)) {
4201 timer_reduce(&cfile
->notify_timer
, next
);
4203 kernfs_notify(cfile
->kn
);
4204 cfile
->notified_at
= jiffies
;
4207 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
4211 * css_next_child - find the next child of a given css
4212 * @pos: the current position (%NULL to initiate traversal)
4213 * @parent: css whose children to walk
4215 * This function returns the next child of @parent and should be called
4216 * under either cgroup_mutex or RCU read lock. The only requirement is
4217 * that @parent and @pos are accessible. The next sibling is guaranteed to
4218 * be returned regardless of their states.
4220 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4221 * css which finished ->css_online() is guaranteed to be visible in the
4222 * future iterations and will stay visible until the last reference is put.
4223 * A css which hasn't finished ->css_online() or already finished
4224 * ->css_offline() may show up during traversal. It's each subsystem's
4225 * responsibility to synchronize against on/offlining.
4227 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
4228 struct cgroup_subsys_state
*parent
)
4230 struct cgroup_subsys_state
*next
;
4232 cgroup_assert_mutex_or_rcu_locked();
4235 * @pos could already have been unlinked from the sibling list.
4236 * Once a cgroup is removed, its ->sibling.next is no longer
4237 * updated when its next sibling changes. CSS_RELEASED is set when
4238 * @pos is taken off list, at which time its next pointer is valid,
4239 * and, as releases are serialized, the one pointed to by the next
4240 * pointer is guaranteed to not have started release yet. This
4241 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4242 * critical section, the one pointed to by its next pointer is
4243 * guaranteed to not have finished its RCU grace period even if we
4244 * have dropped rcu_read_lock() inbetween iterations.
4246 * If @pos has CSS_RELEASED set, its next pointer can't be
4247 * dereferenced; however, as each css is given a monotonically
4248 * increasing unique serial number and always appended to the
4249 * sibling list, the next one can be found by walking the parent's
4250 * children until the first css with higher serial number than
4251 * @pos's. While this path can be slower, it happens iff iteration
4252 * races against release and the race window is very small.
4255 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
4256 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
4257 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
4259 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
4260 if (next
->serial_nr
> pos
->serial_nr
)
4265 * @next, if not pointing to the head, can be dereferenced and is
4268 if (&next
->sibling
!= &parent
->children
)
4274 * css_next_descendant_pre - find the next descendant for pre-order walk
4275 * @pos: the current position (%NULL to initiate traversal)
4276 * @root: css whose descendants to walk
4278 * To be used by css_for_each_descendant_pre(). Find the next descendant
4279 * to visit for pre-order traversal of @root's descendants. @root is
4280 * included in the iteration and the first node to be visited.
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 next descendant as long
4285 * as both @pos and @root are accessible and @pos is a descendant of @root.
4287 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4288 * css which finished ->css_online() is guaranteed to be visible in the
4289 * future iterations and will stay visible until the last reference is put.
4290 * A css which hasn't finished ->css_online() or already finished
4291 * ->css_offline() may show up during traversal. It's each subsystem's
4292 * responsibility to synchronize against on/offlining.
4294 struct cgroup_subsys_state
*
4295 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4296 struct cgroup_subsys_state
*root
)
4298 struct cgroup_subsys_state
*next
;
4300 cgroup_assert_mutex_or_rcu_locked();
4302 /* if first iteration, visit @root */
4306 /* visit the first child if exists */
4307 next
= css_next_child(NULL
, pos
);
4311 /* no child, visit my or the closest ancestor's next sibling */
4312 while (pos
!= root
) {
4313 next
= css_next_child(pos
, pos
->parent
);
4321 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
4324 * css_rightmost_descendant - return the rightmost descendant of a css
4325 * @pos: css of interest
4327 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4328 * is returned. This can be used during pre-order traversal to skip
4331 * While this function requires cgroup_mutex or RCU read locking, it
4332 * doesn't require the whole traversal to be contained in a single critical
4333 * section. This function will return the correct rightmost descendant as
4334 * long as @pos is accessible.
4336 struct cgroup_subsys_state
*
4337 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4339 struct cgroup_subsys_state
*last
, *tmp
;
4341 cgroup_assert_mutex_or_rcu_locked();
4345 /* ->prev isn't RCU safe, walk ->next till the end */
4347 css_for_each_child(tmp
, last
)
4354 static struct cgroup_subsys_state
*
4355 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4357 struct cgroup_subsys_state
*last
;
4361 pos
= css_next_child(NULL
, pos
);
4368 * css_next_descendant_post - find the next descendant for post-order walk
4369 * @pos: the current position (%NULL to initiate traversal)
4370 * @root: css whose descendants to walk
4372 * To be used by css_for_each_descendant_post(). Find the next descendant
4373 * to visit for post-order traversal of @root's descendants. @root is
4374 * included in the iteration and the last node to be visited.
4376 * While this function requires cgroup_mutex or RCU read locking, it
4377 * doesn't require the whole traversal to be contained in a single critical
4378 * section. This function will return the correct next descendant as long
4379 * as both @pos and @cgroup are accessible and @pos is a descendant of
4382 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4383 * css which finished ->css_online() is guaranteed to be visible in the
4384 * future iterations and will stay visible until the last reference is put.
4385 * A css which hasn't finished ->css_online() or already finished
4386 * ->css_offline() may show up during traversal. It's each subsystem's
4387 * responsibility to synchronize against on/offlining.
4389 struct cgroup_subsys_state
*
4390 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4391 struct cgroup_subsys_state
*root
)
4393 struct cgroup_subsys_state
*next
;
4395 cgroup_assert_mutex_or_rcu_locked();
4397 /* if first iteration, visit leftmost descendant which may be @root */
4399 return css_leftmost_descendant(root
);
4401 /* if we visited @root, we're done */
4405 /* if there's an unvisited sibling, visit its leftmost descendant */
4406 next
= css_next_child(pos
, pos
->parent
);
4408 return css_leftmost_descendant(next
);
4410 /* no sibling left, visit parent */
4415 * css_has_online_children - does a css have online children
4416 * @css: the target css
4418 * Returns %true if @css has any online children; otherwise, %false. This
4419 * function can be called from any context but the caller is responsible
4420 * for synchronizing against on/offlining as necessary.
4422 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4424 struct cgroup_subsys_state
*child
;
4428 css_for_each_child(child
, css
) {
4429 if (child
->flags
& CSS_ONLINE
) {
4438 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4440 struct list_head
*l
;
4441 struct cgrp_cset_link
*link
;
4442 struct css_set
*cset
;
4444 lockdep_assert_held(&css_set_lock
);
4446 /* find the next threaded cset */
4447 if (it
->tcset_pos
) {
4448 l
= it
->tcset_pos
->next
;
4450 if (l
!= it
->tcset_head
) {
4452 return container_of(l
, struct css_set
,
4453 threaded_csets_node
);
4456 it
->tcset_pos
= NULL
;
4459 /* find the next cset */
4462 if (l
== it
->cset_head
) {
4463 it
->cset_pos
= NULL
;
4468 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4470 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4476 /* initialize threaded css_set walking */
4477 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4479 put_css_set_locked(it
->cur_dcset
);
4480 it
->cur_dcset
= cset
;
4483 it
->tcset_head
= &cset
->threaded_csets
;
4484 it
->tcset_pos
= &cset
->threaded_csets
;
4491 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4492 * @it: the iterator to advance
4494 * Advance @it to the next css_set to walk.
4496 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4498 struct css_set
*cset
;
4500 lockdep_assert_held(&css_set_lock
);
4502 /* Advance to the next non-empty css_set */
4504 cset
= css_task_iter_next_css_set(it
);
4506 it
->task_pos
= NULL
;
4509 } while (!css_set_populated(cset
) && list_empty(&cset
->dying_tasks
));
4511 if (!list_empty(&cset
->tasks
)) {
4512 it
->task_pos
= cset
->tasks
.next
;
4513 it
->cur_tasks_head
= &cset
->tasks
;
4514 } else if (!list_empty(&cset
->mg_tasks
)) {
4515 it
->task_pos
= cset
->mg_tasks
.next
;
4516 it
->cur_tasks_head
= &cset
->mg_tasks
;
4518 it
->task_pos
= cset
->dying_tasks
.next
;
4519 it
->cur_tasks_head
= &cset
->dying_tasks
;
4522 it
->tasks_head
= &cset
->tasks
;
4523 it
->mg_tasks_head
= &cset
->mg_tasks
;
4524 it
->dying_tasks_head
= &cset
->dying_tasks
;
4527 * We don't keep css_sets locked across iteration steps and thus
4528 * need to take steps to ensure that iteration can be resumed after
4529 * the lock is re-acquired. Iteration is performed at two levels -
4530 * css_sets and tasks in them.
4532 * Once created, a css_set never leaves its cgroup lists, so a
4533 * pinned css_set is guaranteed to stay put and we can resume
4534 * iteration afterwards.
4536 * Tasks may leave @cset across iteration steps. This is resolved
4537 * by registering each iterator with the css_set currently being
4538 * walked and making css_set_move_task() advance iterators whose
4539 * next task is leaving.
4542 list_del(&it
->iters_node
);
4543 put_css_set_locked(it
->cur_cset
);
4546 it
->cur_cset
= cset
;
4547 list_add(&it
->iters_node
, &cset
->task_iters
);
4550 static void css_task_iter_skip(struct css_task_iter
*it
,
4551 struct task_struct
*task
)
4553 lockdep_assert_held(&css_set_lock
);
4555 if (it
->task_pos
== &task
->cg_list
) {
4556 it
->task_pos
= it
->task_pos
->next
;
4557 it
->flags
|= CSS_TASK_ITER_SKIPPED
;
4561 static void css_task_iter_advance(struct css_task_iter
*it
)
4563 struct task_struct
*task
;
4565 lockdep_assert_held(&css_set_lock
);
4569 * Advance iterator to find next entry. cset->tasks is
4570 * consumed first and then ->mg_tasks. After ->mg_tasks,
4571 * we move onto the next cset.
4573 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4574 it
->flags
&= ~CSS_TASK_ITER_SKIPPED
;
4576 it
->task_pos
= it
->task_pos
->next
;
4578 if (it
->task_pos
== it
->tasks_head
) {
4579 it
->task_pos
= it
->mg_tasks_head
->next
;
4580 it
->cur_tasks_head
= it
->mg_tasks_head
;
4582 if (it
->task_pos
== it
->mg_tasks_head
) {
4583 it
->task_pos
= it
->dying_tasks_head
->next
;
4584 it
->cur_tasks_head
= it
->dying_tasks_head
;
4586 if (it
->task_pos
== it
->dying_tasks_head
)
4587 css_task_iter_advance_css_set(it
);
4589 /* called from start, proceed to the first cset */
4590 css_task_iter_advance_css_set(it
);
4596 task
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
4598 if (it
->flags
& CSS_TASK_ITER_PROCS
) {
4599 /* if PROCS, skip over tasks which aren't group leaders */
4600 if (!thread_group_leader(task
))
4603 /* and dying leaders w/o live member threads */
4604 if (it
->cur_tasks_head
== it
->dying_tasks_head
&&
4605 !atomic_read(&task
->signal
->live
))
4608 /* skip all dying ones */
4609 if (it
->cur_tasks_head
== it
->dying_tasks_head
)
4615 * css_task_iter_start - initiate task iteration
4616 * @css: the css to walk tasks of
4617 * @flags: CSS_TASK_ITER_* flags
4618 * @it: the task iterator to use
4620 * Initiate iteration through the tasks of @css. The caller can call
4621 * css_task_iter_next() to walk through the tasks until the function
4622 * returns NULL. On completion of iteration, css_task_iter_end() must be
4625 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4626 struct css_task_iter
*it
)
4628 /* no one should try to iterate before mounting cgroups */
4629 WARN_ON_ONCE(!use_task_css_set_links
);
4631 memset(it
, 0, sizeof(*it
));
4633 spin_lock_irq(&css_set_lock
);
4639 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4641 it
->cset_pos
= &css
->cgroup
->cset_links
;
4643 it
->cset_head
= it
->cset_pos
;
4645 css_task_iter_advance(it
);
4647 spin_unlock_irq(&css_set_lock
);
4651 * css_task_iter_next - return the next task for the iterator
4652 * @it: the task iterator being iterated
4654 * The "next" function for task iteration. @it should have been
4655 * initialized via css_task_iter_start(). Returns NULL when the iteration
4658 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4661 put_task_struct(it
->cur_task
);
4662 it
->cur_task
= NULL
;
4665 spin_lock_irq(&css_set_lock
);
4667 /* @it may be half-advanced by skips, finish advancing */
4668 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4669 css_task_iter_advance(it
);
4672 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4674 get_task_struct(it
->cur_task
);
4675 css_task_iter_advance(it
);
4678 spin_unlock_irq(&css_set_lock
);
4680 return it
->cur_task
;
4684 * css_task_iter_end - finish task iteration
4685 * @it: the task iterator to finish
4687 * Finish task iteration started by css_task_iter_start().
4689 void css_task_iter_end(struct css_task_iter
*it
)
4692 spin_lock_irq(&css_set_lock
);
4693 list_del(&it
->iters_node
);
4694 put_css_set_locked(it
->cur_cset
);
4695 spin_unlock_irq(&css_set_lock
);
4699 put_css_set(it
->cur_dcset
);
4702 put_task_struct(it
->cur_task
);
4705 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4707 struct cgroup_file_ctx
*ctx
= of
->priv
;
4709 if (ctx
->procs
.started
)
4710 css_task_iter_end(&ctx
->procs
.iter
);
4713 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4715 struct kernfs_open_file
*of
= s
->private;
4716 struct cgroup_file_ctx
*ctx
= of
->priv
;
4721 return css_task_iter_next(&ctx
->procs
.iter
);
4724 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4725 unsigned int iter_flags
)
4727 struct kernfs_open_file
*of
= s
->private;
4728 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4729 struct cgroup_file_ctx
*ctx
= of
->priv
;
4730 struct css_task_iter
*it
= &ctx
->procs
.iter
;
4733 * When a seq_file is seeked, it's always traversed sequentially
4734 * from position 0, so we can simply keep iterating on !0 *pos.
4736 if (!ctx
->procs
.started
) {
4737 if (WARN_ON_ONCE((*pos
)))
4738 return ERR_PTR(-EINVAL
);
4739 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4740 ctx
->procs
.started
= true;
4741 } else if (!(*pos
)) {
4742 css_task_iter_end(it
);
4743 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4745 return it
->cur_task
;
4747 return cgroup_procs_next(s
, NULL
, NULL
);
4750 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4752 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4755 * All processes of a threaded subtree belong to the domain cgroup
4756 * of the subtree. Only threads can be distributed across the
4757 * subtree. Reject reads on cgroup.procs in the subtree proper.
4758 * They're always empty anyway.
4760 if (cgroup_is_threaded(cgrp
))
4761 return ERR_PTR(-EOPNOTSUPP
);
4763 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4764 CSS_TASK_ITER_THREADED
);
4767 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4769 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4773 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4774 struct cgroup
*dst_cgrp
,
4775 struct super_block
*sb
)
4777 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4778 struct cgroup
*com_cgrp
= src_cgrp
;
4779 struct inode
*inode
;
4782 lockdep_assert_held(&cgroup_mutex
);
4784 /* find the common ancestor */
4785 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4786 com_cgrp
= cgroup_parent(com_cgrp
);
4788 /* %current should be authorized to migrate to the common ancestor */
4789 inode
= kernfs_get_inode(sb
, com_cgrp
->procs_file
.kn
);
4793 ret
= inode_permission(inode
, MAY_WRITE
);
4799 * If namespaces are delegation boundaries, %current must be able
4800 * to see both source and destination cgroups from its namespace.
4802 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4803 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4804 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4810 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4811 char *buf
, size_t nbytes
, loff_t off
)
4813 struct cgroup
*src_cgrp
, *dst_cgrp
;
4814 struct task_struct
*task
;
4815 const struct cred
*saved_cred
;
4818 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4822 task
= cgroup_procs_write_start(buf
, true);
4823 ret
= PTR_ERR_OR_ZERO(task
);
4827 /* find the source cgroup */
4828 spin_lock_irq(&css_set_lock
);
4829 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4830 spin_unlock_irq(&css_set_lock
);
4833 * Process and thread migrations follow same delegation rule. Check
4834 * permissions using the credentials from file open to protect against
4835 * inherited fd attacks.
4837 saved_cred
= override_creds(of
->file
->f_cred
);
4838 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4839 of
->file
->f_path
.dentry
->d_sb
);
4840 revert_creds(saved_cred
);
4844 ret
= cgroup_attach_task(dst_cgrp
, task
, true);
4847 cgroup_procs_write_finish(task
);
4849 cgroup_kn_unlock(of
->kn
);
4851 return ret
?: nbytes
;
4854 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4856 return __cgroup_procs_start(s
, pos
, 0);
4859 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4860 char *buf
, size_t nbytes
, loff_t off
)
4862 struct cgroup
*src_cgrp
, *dst_cgrp
;
4863 struct task_struct
*task
;
4864 const struct cred
*saved_cred
;
4867 buf
= strstrip(buf
);
4869 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4873 task
= cgroup_procs_write_start(buf
, false);
4874 ret
= PTR_ERR_OR_ZERO(task
);
4878 /* find the source cgroup */
4879 spin_lock_irq(&css_set_lock
);
4880 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4881 spin_unlock_irq(&css_set_lock
);
4884 * Process and thread migrations follow same delegation rule. Check
4885 * permissions using the credentials from file open to protect against
4886 * inherited fd attacks.
4888 saved_cred
= override_creds(of
->file
->f_cred
);
4889 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4890 of
->file
->f_path
.dentry
->d_sb
);
4891 revert_creds(saved_cred
);
4895 /* and must be contained in the same domain */
4897 if (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
)
4900 ret
= cgroup_attach_task(dst_cgrp
, task
, false);
4903 cgroup_procs_write_finish(task
);
4905 cgroup_kn_unlock(of
->kn
);
4907 return ret
?: nbytes
;
4910 /* cgroup core interface files for the default hierarchy */
4911 static struct cftype cgroup_base_files
[] = {
4913 .name
= "cgroup.type",
4914 .flags
= CFTYPE_NOT_ON_ROOT
,
4915 .seq_show
= cgroup_type_show
,
4916 .write
= cgroup_type_write
,
4919 .name
= "cgroup.procs",
4920 .flags
= CFTYPE_NS_DELEGATABLE
,
4921 .file_offset
= offsetof(struct cgroup
, procs_file
),
4922 .release
= cgroup_procs_release
,
4923 .seq_start
= cgroup_procs_start
,
4924 .seq_next
= cgroup_procs_next
,
4925 .seq_show
= cgroup_procs_show
,
4926 .write
= cgroup_procs_write
,
4929 .name
= "cgroup.threads",
4930 .flags
= CFTYPE_NS_DELEGATABLE
,
4931 .release
= cgroup_procs_release
,
4932 .seq_start
= cgroup_threads_start
,
4933 .seq_next
= cgroup_procs_next
,
4934 .seq_show
= cgroup_procs_show
,
4935 .write
= cgroup_threads_write
,
4938 .name
= "cgroup.controllers",
4939 .seq_show
= cgroup_controllers_show
,
4942 .name
= "cgroup.subtree_control",
4943 .flags
= CFTYPE_NS_DELEGATABLE
,
4944 .seq_show
= cgroup_subtree_control_show
,
4945 .write
= cgroup_subtree_control_write
,
4948 .name
= "cgroup.events",
4949 .flags
= CFTYPE_NOT_ON_ROOT
,
4950 .file_offset
= offsetof(struct cgroup
, events_file
),
4951 .seq_show
= cgroup_events_show
,
4954 .name
= "cgroup.max.descendants",
4955 .seq_show
= cgroup_max_descendants_show
,
4956 .write
= cgroup_max_descendants_write
,
4959 .name
= "cgroup.max.depth",
4960 .seq_show
= cgroup_max_depth_show
,
4961 .write
= cgroup_max_depth_write
,
4964 .name
= "cgroup.stat",
4965 .seq_show
= cgroup_stat_show
,
4968 .name
= "cgroup.freeze",
4969 .flags
= CFTYPE_NOT_ON_ROOT
,
4970 .seq_show
= cgroup_freeze_show
,
4971 .write
= cgroup_freeze_write
,
4975 .flags
= CFTYPE_NOT_ON_ROOT
,
4976 .seq_show
= cpu_stat_show
,
4980 .name
= "io.pressure",
4981 .seq_show
= cgroup_io_pressure_show
,
4982 .write
= cgroup_io_pressure_write
,
4983 .poll
= cgroup_pressure_poll
,
4984 .release
= cgroup_pressure_release
,
4987 .name
= "memory.pressure",
4988 .seq_show
= cgroup_memory_pressure_show
,
4989 .write
= cgroup_memory_pressure_write
,
4990 .poll
= cgroup_pressure_poll
,
4991 .release
= cgroup_pressure_release
,
4994 .name
= "cpu.pressure",
4995 .seq_show
= cgroup_cpu_pressure_show
,
4996 .write
= cgroup_cpu_pressure_write
,
4997 .poll
= cgroup_pressure_poll
,
4998 .release
= cgroup_pressure_release
,
5000 #endif /* CONFIG_PSI */
5005 * css destruction is four-stage process.
5007 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5008 * Implemented in kill_css().
5010 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5011 * and thus css_tryget_online() is guaranteed to fail, the css can be
5012 * offlined by invoking offline_css(). After offlining, the base ref is
5013 * put. Implemented in css_killed_work_fn().
5015 * 3. When the percpu_ref reaches zero, the only possible remaining
5016 * accessors are inside RCU read sections. css_release() schedules the
5019 * 4. After the grace period, the css can be freed. Implemented in
5020 * css_free_work_fn().
5022 * It is actually hairier because both step 2 and 4 require process context
5023 * and thus involve punting to css->destroy_work adding two additional
5024 * steps to the already complex sequence.
5026 static void css_free_rwork_fn(struct work_struct
*work
)
5028 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
5029 struct cgroup_subsys_state
, destroy_rwork
);
5030 struct cgroup_subsys
*ss
= css
->ss
;
5031 struct cgroup
*cgrp
= css
->cgroup
;
5033 percpu_ref_exit(&css
->refcnt
);
5037 struct cgroup_subsys_state
*parent
= css
->parent
;
5041 cgroup_idr_remove(&ss
->css_idr
, id
);
5047 /* cgroup free path */
5048 atomic_dec(&cgrp
->root
->nr_cgrps
);
5049 cgroup1_pidlist_destroy_all(cgrp
);
5050 cancel_work_sync(&cgrp
->release_agent_work
);
5052 if (cgroup_parent(cgrp
)) {
5054 * We get a ref to the parent, and put the ref when
5055 * this cgroup is being freed, so it's guaranteed
5056 * that the parent won't be destroyed before its
5059 cgroup_put(cgroup_parent(cgrp
));
5060 kernfs_put(cgrp
->kn
);
5061 psi_cgroup_free(cgrp
);
5062 if (cgroup_on_dfl(cgrp
))
5063 cgroup_rstat_exit(cgrp
);
5067 * This is root cgroup's refcnt reaching zero,
5068 * which indicates that the root should be
5071 cgroup_destroy_root(cgrp
->root
);
5076 static void css_release_work_fn(struct work_struct
*work
)
5078 struct cgroup_subsys_state
*css
=
5079 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5080 struct cgroup_subsys
*ss
= css
->ss
;
5081 struct cgroup
*cgrp
= css
->cgroup
;
5083 mutex_lock(&cgroup_mutex
);
5085 css
->flags
|= CSS_RELEASED
;
5086 list_del_rcu(&css
->sibling
);
5089 /* css release path */
5090 if (!list_empty(&css
->rstat_css_node
)) {
5091 cgroup_rstat_flush(cgrp
);
5092 list_del_rcu(&css
->rstat_css_node
);
5095 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
5096 if (ss
->css_released
)
5097 ss
->css_released(css
);
5099 struct cgroup
*tcgrp
;
5101 /* cgroup release path */
5102 TRACE_CGROUP_PATH(release
, cgrp
);
5104 if (cgroup_on_dfl(cgrp
))
5105 cgroup_rstat_flush(cgrp
);
5107 spin_lock_irq(&css_set_lock
);
5108 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
5109 tcgrp
= cgroup_parent(tcgrp
))
5110 tcgrp
->nr_dying_descendants
--;
5111 spin_unlock_irq(&css_set_lock
);
5113 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
5117 * There are two control paths which try to determine
5118 * cgroup from dentry without going through kernfs -
5119 * cgroupstats_build() and css_tryget_online_from_dir().
5120 * Those are supported by RCU protecting clearing of
5121 * cgrp->kn->priv backpointer.
5124 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5128 mutex_unlock(&cgroup_mutex
);
5130 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5131 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5134 static void css_release(struct percpu_ref
*ref
)
5136 struct cgroup_subsys_state
*css
=
5137 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5139 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5140 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5143 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5144 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5146 lockdep_assert_held(&cgroup_mutex
);
5148 cgroup_get_live(cgrp
);
5150 memset(css
, 0, sizeof(*css
));
5154 INIT_LIST_HEAD(&css
->sibling
);
5155 INIT_LIST_HEAD(&css
->children
);
5156 INIT_LIST_HEAD(&css
->rstat_css_node
);
5157 css
->serial_nr
= css_serial_nr_next
++;
5158 atomic_set(&css
->online_cnt
, 0);
5160 if (cgroup_parent(cgrp
)) {
5161 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5162 css_get(css
->parent
);
5165 if (cgroup_on_dfl(cgrp
) && ss
->css_rstat_flush
)
5166 list_add_rcu(&css
->rstat_css_node
, &cgrp
->rstat_css_list
);
5168 BUG_ON(cgroup_css(cgrp
, ss
));
5171 /* invoke ->css_online() on a new CSS and mark it online if successful */
5172 static int online_css(struct cgroup_subsys_state
*css
)
5174 struct cgroup_subsys
*ss
= css
->ss
;
5177 lockdep_assert_held(&cgroup_mutex
);
5180 ret
= ss
->css_online(css
);
5182 css
->flags
|= CSS_ONLINE
;
5183 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5185 atomic_inc(&css
->online_cnt
);
5187 atomic_inc(&css
->parent
->online_cnt
);
5192 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5193 static void offline_css(struct cgroup_subsys_state
*css
)
5195 struct cgroup_subsys
*ss
= css
->ss
;
5197 lockdep_assert_held(&cgroup_mutex
);
5199 if (!(css
->flags
& CSS_ONLINE
))
5202 if (ss
->css_offline
)
5203 ss
->css_offline(css
);
5205 css
->flags
&= ~CSS_ONLINE
;
5206 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5208 wake_up_all(&css
->cgroup
->offline_waitq
);
5212 * css_create - create a cgroup_subsys_state
5213 * @cgrp: the cgroup new css will be associated with
5214 * @ss: the subsys of new css
5216 * Create a new css associated with @cgrp - @ss pair. On success, the new
5217 * css is online and installed in @cgrp. This function doesn't create the
5218 * interface files. Returns 0 on success, -errno on failure.
5220 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5221 struct cgroup_subsys
*ss
)
5223 struct cgroup
*parent
= cgroup_parent(cgrp
);
5224 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5225 struct cgroup_subsys_state
*css
;
5228 lockdep_assert_held(&cgroup_mutex
);
5230 css
= ss
->css_alloc(parent_css
);
5232 css
= ERR_PTR(-ENOMEM
);
5236 init_and_link_css(css
, ss
, cgrp
);
5238 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5242 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5247 /* @css is ready to be brought online now, make it visible */
5248 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5249 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5251 err
= online_css(css
);
5255 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
5256 cgroup_parent(parent
)) {
5257 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5258 current
->comm
, current
->pid
, ss
->name
);
5259 if (!strcmp(ss
->name
, "memory"))
5260 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5261 ss
->warned_broken_hierarchy
= true;
5267 list_del_rcu(&css
->sibling
);
5269 list_del_rcu(&css
->rstat_css_node
);
5270 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5271 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5272 return ERR_PTR(err
);
5276 * The returned cgroup is fully initialized including its control mask, but
5277 * it isn't associated with its kernfs_node and doesn't have the control
5280 static struct cgroup
*cgroup_create(struct cgroup
*parent
)
5282 struct cgroup_root
*root
= parent
->root
;
5283 struct cgroup
*cgrp
, *tcgrp
;
5284 int level
= parent
->level
+ 1;
5287 /* allocate the cgroup and its ID, 0 is reserved for the root */
5288 cgrp
= kzalloc(struct_size(cgrp
, ancestor_ids
, (level
+ 1)),
5291 return ERR_PTR(-ENOMEM
);
5293 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5297 if (cgroup_on_dfl(parent
)) {
5298 ret
= cgroup_rstat_init(cgrp
);
5300 goto out_cancel_ref
;
5304 * Temporarily set the pointer to NULL, so idr_find() won't return
5305 * a half-baked cgroup.
5307 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
5313 init_cgroup_housekeeping(cgrp
);
5315 cgrp
->self
.parent
= &parent
->self
;
5317 cgrp
->level
= level
;
5319 ret
= psi_cgroup_alloc(cgrp
);
5323 ret
= cgroup_bpf_inherit(cgrp
);
5328 * New cgroup inherits effective freeze counter, and
5329 * if the parent has to be frozen, the child has too.
5331 cgrp
->freezer
.e_freeze
= parent
->freezer
.e_freeze
;
5332 if (cgrp
->freezer
.e_freeze
) {
5334 * Set the CGRP_FREEZE flag, so when a process will be
5335 * attached to the child cgroup, it will become frozen.
5336 * At this point the new cgroup is unpopulated, so we can
5337 * consider it frozen immediately.
5339 set_bit(CGRP_FREEZE
, &cgrp
->flags
);
5340 set_bit(CGRP_FROZEN
, &cgrp
->flags
);
5343 spin_lock_irq(&css_set_lock
);
5344 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5345 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
5347 if (tcgrp
!= cgrp
) {
5348 tcgrp
->nr_descendants
++;
5351 * If the new cgroup is frozen, all ancestor cgroups
5352 * get a new frozen descendant, but their state can't
5353 * change because of this.
5355 if (cgrp
->freezer
.e_freeze
)
5356 tcgrp
->freezer
.nr_frozen_descendants
++;
5359 spin_unlock_irq(&css_set_lock
);
5361 if (notify_on_release(parent
))
5362 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5364 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5365 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5367 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5369 /* allocation complete, commit to creation */
5370 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5371 atomic_inc(&root
->nr_cgrps
);
5372 cgroup_get_live(parent
);
5375 * @cgrp is now fully operational. If something fails after this
5376 * point, it'll be released via the normal destruction path.
5378 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
5381 * On the default hierarchy, a child doesn't automatically inherit
5382 * subtree_control from the parent. Each is configured manually.
5384 if (!cgroup_on_dfl(cgrp
))
5385 cgrp
->subtree_control
= cgroup_control(cgrp
);
5387 cgroup_propagate_control(cgrp
);
5392 psi_cgroup_free(cgrp
);
5394 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
5396 if (cgroup_on_dfl(parent
))
5397 cgroup_rstat_exit(cgrp
);
5399 percpu_ref_exit(&cgrp
->self
.refcnt
);
5402 return ERR_PTR(ret
);
5405 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
5407 struct cgroup
*cgroup
;
5411 lockdep_assert_held(&cgroup_mutex
);
5413 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
5414 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
5417 if (level
> cgroup
->max_depth
)
5428 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
5430 struct cgroup
*parent
, *cgrp
;
5431 struct kernfs_node
*kn
;
5434 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5435 if (strchr(name
, '\n'))
5438 parent
= cgroup_kn_lock_live(parent_kn
, false);
5442 if (!cgroup_check_hierarchy_limits(parent
)) {
5447 cgrp
= cgroup_create(parent
);
5449 ret
= PTR_ERR(cgrp
);
5453 /* create the directory */
5454 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5462 * This extra ref will be put in cgroup_free_fn() and guarantees
5463 * that @cgrp->kn is always accessible.
5467 ret
= cgroup_kn_set_ugid(kn
);
5471 ret
= css_populate_dir(&cgrp
->self
);
5475 ret
= cgroup_apply_control_enable(cgrp
);
5479 TRACE_CGROUP_PATH(mkdir
, cgrp
);
5481 /* let's create and online css's */
5482 kernfs_activate(kn
);
5488 cgroup_destroy_locked(cgrp
);
5490 cgroup_kn_unlock(parent_kn
);
5495 * This is called when the refcnt of a css is confirmed to be killed.
5496 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5497 * initate destruction and put the css ref from kill_css().
5499 static void css_killed_work_fn(struct work_struct
*work
)
5501 struct cgroup_subsys_state
*css
=
5502 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5504 mutex_lock(&cgroup_mutex
);
5509 /* @css can't go away while we're holding cgroup_mutex */
5511 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5513 mutex_unlock(&cgroup_mutex
);
5516 /* css kill confirmation processing requires process context, bounce */
5517 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5519 struct cgroup_subsys_state
*css
=
5520 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5522 if (atomic_dec_and_test(&css
->online_cnt
)) {
5523 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5524 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5529 * kill_css - destroy a css
5530 * @css: css to destroy
5532 * This function initiates destruction of @css by removing cgroup interface
5533 * files and putting its base reference. ->css_offline() will be invoked
5534 * asynchronously once css_tryget_online() is guaranteed to fail and when
5535 * the reference count reaches zero, @css will be released.
5537 static void kill_css(struct cgroup_subsys_state
*css
)
5539 lockdep_assert_held(&cgroup_mutex
);
5541 if (css
->flags
& CSS_DYING
)
5544 css
->flags
|= CSS_DYING
;
5547 * This must happen before css is disassociated with its cgroup.
5548 * See seq_css() for details.
5553 * Killing would put the base ref, but we need to keep it alive
5554 * until after ->css_offline().
5559 * cgroup core guarantees that, by the time ->css_offline() is
5560 * invoked, no new css reference will be given out via
5561 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5562 * proceed to offlining css's because percpu_ref_kill() doesn't
5563 * guarantee that the ref is seen as killed on all CPUs on return.
5565 * Use percpu_ref_kill_and_confirm() to get notifications as each
5566 * css is confirmed to be seen as killed on all CPUs.
5568 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5572 * cgroup_destroy_locked - the first stage of cgroup destruction
5573 * @cgrp: cgroup to be destroyed
5575 * css's make use of percpu refcnts whose killing latency shouldn't be
5576 * exposed to userland and are RCU protected. Also, cgroup core needs to
5577 * guarantee that css_tryget_online() won't succeed by the time
5578 * ->css_offline() is invoked. To satisfy all the requirements,
5579 * destruction is implemented in the following two steps.
5581 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5582 * userland visible parts and start killing the percpu refcnts of
5583 * css's. Set up so that the next stage will be kicked off once all
5584 * the percpu refcnts are confirmed to be killed.
5586 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5587 * rest of destruction. Once all cgroup references are gone, the
5588 * cgroup is RCU-freed.
5590 * This function implements s1. After this step, @cgrp is gone as far as
5591 * the userland is concerned and a new cgroup with the same name may be
5592 * created. As cgroup doesn't care about the names internally, this
5593 * doesn't cause any problem.
5595 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5596 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5598 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5599 struct cgroup_subsys_state
*css
;
5600 struct cgrp_cset_link
*link
;
5603 lockdep_assert_held(&cgroup_mutex
);
5606 * Only migration can raise populated from zero and we're already
5607 * holding cgroup_mutex.
5609 if (cgroup_is_populated(cgrp
))
5613 * Make sure there's no live children. We can't test emptiness of
5614 * ->self.children as dead children linger on it while being
5615 * drained; otherwise, "rmdir parent/child parent" may fail.
5617 if (css_has_online_children(&cgrp
->self
))
5621 * Mark @cgrp and the associated csets dead. The former prevents
5622 * further task migration and child creation by disabling
5623 * cgroup_lock_live_group(). The latter makes the csets ignored by
5624 * the migration path.
5626 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5628 spin_lock_irq(&css_set_lock
);
5629 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5630 link
->cset
->dead
= true;
5631 spin_unlock_irq(&css_set_lock
);
5633 /* initiate massacre of all css's */
5634 for_each_css(css
, ssid
, cgrp
)
5637 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5638 css_clear_dir(&cgrp
->self
);
5639 kernfs_remove(cgrp
->kn
);
5641 if (parent
&& cgroup_is_threaded(cgrp
))
5642 parent
->nr_threaded_children
--;
5644 spin_lock_irq(&css_set_lock
);
5645 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5646 tcgrp
->nr_descendants
--;
5647 tcgrp
->nr_dying_descendants
++;
5649 * If the dying cgroup is frozen, decrease frozen descendants
5650 * counters of ancestor cgroups.
5652 if (test_bit(CGRP_FROZEN
, &cgrp
->flags
))
5653 tcgrp
->freezer
.nr_frozen_descendants
--;
5655 spin_unlock_irq(&css_set_lock
);
5657 cgroup1_check_for_release(parent
);
5659 cgroup_bpf_offline(cgrp
);
5661 /* put the base reference */
5662 percpu_ref_kill(&cgrp
->self
.refcnt
);
5667 int cgroup_rmdir(struct kernfs_node
*kn
)
5669 struct cgroup
*cgrp
;
5672 cgrp
= cgroup_kn_lock_live(kn
, false);
5676 ret
= cgroup_destroy_locked(cgrp
);
5678 TRACE_CGROUP_PATH(rmdir
, cgrp
);
5680 cgroup_kn_unlock(kn
);
5684 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5685 .show_options
= cgroup_show_options
,
5686 .mkdir
= cgroup_mkdir
,
5687 .rmdir
= cgroup_rmdir
,
5688 .show_path
= cgroup_show_path
,
5691 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5693 struct cgroup_subsys_state
*css
;
5695 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5697 mutex_lock(&cgroup_mutex
);
5699 idr_init(&ss
->css_idr
);
5700 INIT_LIST_HEAD(&ss
->cfts
);
5702 /* Create the root cgroup state for this subsystem */
5703 ss
->root
= &cgrp_dfl_root
;
5704 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5705 /* We don't handle early failures gracefully */
5706 BUG_ON(IS_ERR(css
));
5707 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5710 * Root csses are never destroyed and we can't initialize
5711 * percpu_ref during early init. Disable refcnting.
5713 css
->flags
|= CSS_NO_REF
;
5716 /* allocation can't be done safely during early init */
5719 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5720 BUG_ON(css
->id
< 0);
5723 /* Update the init_css_set to contain a subsys
5724 * pointer to this state - since the subsystem is
5725 * newly registered, all tasks and hence the
5726 * init_css_set is in the subsystem's root cgroup. */
5727 init_css_set
.subsys
[ss
->id
] = css
;
5729 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5730 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5731 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5732 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5734 /* At system boot, before all subsystems have been
5735 * registered, no tasks have been forked, so we don't
5736 * need to invoke fork callbacks here. */
5737 BUG_ON(!list_empty(&init_task
.tasks
));
5739 BUG_ON(online_css(css
));
5741 mutex_unlock(&cgroup_mutex
);
5745 * cgroup_init_early - cgroup initialization at system boot
5747 * Initialize cgroups at system boot, and initialize any
5748 * subsystems that request early init.
5750 int __init
cgroup_init_early(void)
5752 static struct cgroup_fs_context __initdata ctx
;
5753 struct cgroup_subsys
*ss
;
5756 ctx
.root
= &cgrp_dfl_root
;
5757 init_cgroup_root(&ctx
);
5758 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5760 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5762 for_each_subsys(ss
, i
) {
5763 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5764 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5765 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5767 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5768 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5771 ss
->name
= cgroup_subsys_name
[i
];
5772 if (!ss
->legacy_name
)
5773 ss
->legacy_name
= cgroup_subsys_name
[i
];
5776 cgroup_init_subsys(ss
, true);
5782 * cgroup_init - cgroup initialization
5784 * Register cgroup filesystem and /proc file, and initialize
5785 * any subsystems that didn't request early init.
5787 int __init
cgroup_init(void)
5789 struct cgroup_subsys
*ss
;
5792 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5793 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5794 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5796 cgroup_rstat_boot();
5799 * The latency of the synchronize_rcu() is too high for cgroups,
5800 * avoid it at the cost of forcing all readers into the slow path.
5802 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5804 get_user_ns(init_cgroup_ns
.user_ns
);
5806 mutex_lock(&cgroup_mutex
);
5809 * Add init_css_set to the hash table so that dfl_root can link to
5812 hash_add(css_set_table
, &init_css_set
.hlist
,
5813 css_set_hash(init_css_set
.subsys
));
5815 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5817 mutex_unlock(&cgroup_mutex
);
5819 for_each_subsys(ss
, ssid
) {
5820 if (ss
->early_init
) {
5821 struct cgroup_subsys_state
*css
=
5822 init_css_set
.subsys
[ss
->id
];
5824 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5826 BUG_ON(css
->id
< 0);
5828 cgroup_init_subsys(ss
, false);
5831 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5832 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5835 * Setting dfl_root subsys_mask needs to consider the
5836 * disabled flag and cftype registration needs kmalloc,
5837 * both of which aren't available during early_init.
5839 if (!cgroup_ssid_enabled(ssid
))
5842 if (cgroup1_ssid_disabled(ssid
))
5843 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5846 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5848 /* implicit controllers must be threaded too */
5849 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5851 if (ss
->implicit_on_dfl
)
5852 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5853 else if (!ss
->dfl_cftypes
)
5854 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5857 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5859 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5860 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5862 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5863 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5867 ss
->bind(init_css_set
.subsys
[ssid
]);
5869 mutex_lock(&cgroup_mutex
);
5870 css_populate_dir(init_css_set
.subsys
[ssid
]);
5871 mutex_unlock(&cgroup_mutex
);
5874 /* init_css_set.subsys[] has been updated, re-hash */
5875 hash_del(&init_css_set
.hlist
);
5876 hash_add(css_set_table
, &init_css_set
.hlist
,
5877 css_set_hash(init_css_set
.subsys
));
5879 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5880 WARN_ON(register_filesystem(&cgroup_fs_type
));
5881 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5882 WARN_ON(!proc_create_single("cgroups", 0, NULL
, proc_cgroupstats_show
));
5883 #ifdef CONFIG_CPUSETS
5884 WARN_ON(register_filesystem(&cpuset_fs_type
));
5890 static int __init
cgroup_wq_init(void)
5893 * There isn't much point in executing destruction path in
5894 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5895 * Use 1 for @max_active.
5897 * We would prefer to do this in cgroup_init() above, but that
5898 * is called before init_workqueues(): so leave this until after.
5900 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5901 BUG_ON(!cgroup_destroy_wq
);
5904 core_initcall(cgroup_wq_init
);
5906 void cgroup_path_from_kernfs_id(const union kernfs_node_id
*id
,
5907 char *buf
, size_t buflen
)
5909 struct kernfs_node
*kn
;
5911 kn
= kernfs_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5914 kernfs_path(kn
, buf
, buflen
);
5919 * proc_cgroup_show()
5920 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5921 * - Used for /proc/<pid>/cgroup.
5923 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5924 struct pid
*pid
, struct task_struct
*tsk
)
5928 struct cgroup_root
*root
;
5931 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5935 mutex_lock(&cgroup_mutex
);
5936 spin_lock_irq(&css_set_lock
);
5938 for_each_root(root
) {
5939 struct cgroup_subsys
*ss
;
5940 struct cgroup
*cgrp
;
5941 int ssid
, count
= 0;
5943 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5946 seq_printf(m
, "%d:", root
->hierarchy_id
);
5947 if (root
!= &cgrp_dfl_root
)
5948 for_each_subsys(ss
, ssid
)
5949 if (root
->subsys_mask
& (1 << ssid
))
5950 seq_printf(m
, "%s%s", count
++ ? "," : "",
5952 if (strlen(root
->name
))
5953 seq_printf(m
, "%sname=%s", count
? "," : "",
5957 cgrp
= task_cgroup_from_root(tsk
, root
);
5960 * On traditional hierarchies, all zombie tasks show up as
5961 * belonging to the root cgroup. On the default hierarchy,
5962 * while a zombie doesn't show up in "cgroup.procs" and
5963 * thus can't be migrated, its /proc/PID/cgroup keeps
5964 * reporting the cgroup it belonged to before exiting. If
5965 * the cgroup is removed before the zombie is reaped,
5966 * " (deleted)" is appended to the cgroup path.
5968 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5969 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5970 current
->nsproxy
->cgroup_ns
);
5971 if (retval
>= PATH_MAX
)
5972 retval
= -ENAMETOOLONG
;
5981 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5982 seq_puts(m
, " (deleted)\n");
5989 spin_unlock_irq(&css_set_lock
);
5990 mutex_unlock(&cgroup_mutex
);
5997 * cgroup_fork - initialize cgroup related fields during copy_process()
5998 * @child: pointer to task_struct of forking parent process.
6000 * A task is associated with the init_css_set until cgroup_post_fork()
6001 * attaches it to the parent's css_set. Empty cg_list indicates that
6002 * @child isn't holding reference to its css_set.
6004 void cgroup_fork(struct task_struct
*child
)
6006 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
6007 INIT_LIST_HEAD(&child
->cg_list
);
6011 * cgroup_can_fork - called on a new task before the process is exposed
6012 * @child: the task in question.
6014 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
6015 * returns an error, the fork aborts with that error code. This allows for
6016 * a cgroup subsystem to conditionally allow or deny new forks.
6018 int cgroup_can_fork(struct task_struct
*child
)
6020 struct cgroup_subsys
*ss
;
6023 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
6024 ret
= ss
->can_fork(child
);
6027 } while_each_subsys_mask();
6032 for_each_subsys(ss
, j
) {
6035 if (ss
->cancel_fork
)
6036 ss
->cancel_fork(child
);
6043 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6044 * @child: the task in question
6046 * This calls the cancel_fork() callbacks if a fork failed *after*
6047 * cgroup_can_fork() succeded.
6049 void cgroup_cancel_fork(struct task_struct
*child
)
6051 struct cgroup_subsys
*ss
;
6054 for_each_subsys(ss
, i
)
6055 if (ss
->cancel_fork
)
6056 ss
->cancel_fork(child
);
6060 * cgroup_post_fork - called on a new task after adding it to the task list
6061 * @child: the task in question
6063 * Adds the task to the list running through its css_set if necessary and
6064 * call the subsystem fork() callbacks. Has to be after the task is
6065 * visible on the task list in case we race with the first call to
6066 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
6069 void cgroup_post_fork(struct task_struct
*child
)
6071 struct cgroup_subsys
*ss
;
6075 * This may race against cgroup_enable_task_cg_lists(). As that
6076 * function sets use_task_css_set_links before grabbing
6077 * tasklist_lock and we just went through tasklist_lock to add
6078 * @child, it's guaranteed that either we see the set
6079 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
6080 * @child during its iteration.
6082 * If we won the race, @child is associated with %current's
6083 * css_set. Grabbing css_set_lock guarantees both that the
6084 * association is stable, and, on completion of the parent's
6085 * migration, @child is visible in the source of migration or
6086 * already in the destination cgroup. This guarantee is necessary
6087 * when implementing operations which need to migrate all tasks of
6088 * a cgroup to another.
6090 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
6091 * will remain in init_css_set. This is safe because all tasks are
6092 * in the init_css_set before cg_links is enabled and there's no
6093 * operation which transfers all tasks out of init_css_set.
6095 if (use_task_css_set_links
) {
6096 struct css_set
*cset
;
6098 spin_lock_irq(&css_set_lock
);
6099 cset
= task_css_set(current
);
6100 if (list_empty(&child
->cg_list
)) {
6103 css_set_move_task(child
, NULL
, cset
, false);
6107 * If the cgroup has to be frozen, the new task has too.
6108 * Let's set the JOBCTL_TRAP_FREEZE jobctl bit to get
6109 * the task into the frozen state.
6111 if (unlikely(cgroup_task_freeze(child
))) {
6112 spin_lock(&child
->sighand
->siglock
);
6113 WARN_ON_ONCE(child
->frozen
);
6114 child
->jobctl
|= JOBCTL_TRAP_FREEZE
;
6115 spin_unlock(&child
->sighand
->siglock
);
6118 * Calling cgroup_update_frozen() isn't required here,
6119 * because it will be called anyway a bit later
6120 * from do_freezer_trap(). So we avoid cgroup's
6121 * transient switch from the frozen state and back.
6125 spin_unlock_irq(&css_set_lock
);
6129 * Call ss->fork(). This must happen after @child is linked on
6130 * css_set; otherwise, @child might change state between ->fork()
6131 * and addition to css_set.
6133 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
6135 } while_each_subsys_mask();
6139 * cgroup_exit - detach cgroup from exiting task
6140 * @tsk: pointer to task_struct of exiting process
6142 * Description: Detach cgroup from @tsk and release it.
6144 * Note that cgroups marked notify_on_release force every task in
6145 * them to take the global cgroup_mutex mutex when exiting.
6146 * This could impact scaling on very large systems. Be reluctant to
6147 * use notify_on_release cgroups where very high task exit scaling
6148 * is required on large systems.
6150 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
6151 * call cgroup_exit() while the task is still competent to handle
6152 * notify_on_release(), then leave the task attached to the root cgroup in
6153 * each hierarchy for the remainder of its exit. No need to bother with
6154 * init_css_set refcnting. init_css_set never goes away and we can't race
6155 * with migration path - PF_EXITING is visible to migration path.
6157 void cgroup_exit(struct task_struct
*tsk
)
6159 struct cgroup_subsys
*ss
;
6160 struct css_set
*cset
;
6164 * Unlink from @tsk from its css_set. As migration path can't race
6165 * with us, we can check css_set and cg_list without synchronization.
6167 cset
= task_css_set(tsk
);
6169 if (!list_empty(&tsk
->cg_list
)) {
6170 spin_lock_irq(&css_set_lock
);
6171 css_set_move_task(tsk
, cset
, NULL
, false);
6172 list_add_tail(&tsk
->cg_list
, &cset
->dying_tasks
);
6175 WARN_ON_ONCE(cgroup_task_frozen(tsk
));
6176 if (unlikely(cgroup_task_freeze(tsk
)))
6177 cgroup_update_frozen(task_dfl_cgroup(tsk
));
6179 spin_unlock_irq(&css_set_lock
);
6184 /* see cgroup_post_fork() for details */
6185 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
6187 } while_each_subsys_mask();
6190 void cgroup_release(struct task_struct
*task
)
6192 struct cgroup_subsys
*ss
;
6195 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
6197 } while_each_subsys_mask();
6199 if (use_task_css_set_links
) {
6200 spin_lock_irq(&css_set_lock
);
6201 css_set_skip_task_iters(task_css_set(task
), task
);
6202 list_del_init(&task
->cg_list
);
6203 spin_unlock_irq(&css_set_lock
);
6207 void cgroup_free(struct task_struct
*task
)
6209 struct css_set
*cset
= task_css_set(task
);
6213 static int __init
cgroup_disable(char *str
)
6215 struct cgroup_subsys
*ss
;
6219 while ((token
= strsep(&str
, ",")) != NULL
) {
6223 for_each_subsys(ss
, i
) {
6224 if (strcmp(token
, ss
->name
) &&
6225 strcmp(token
, ss
->legacy_name
))
6228 static_branch_disable(cgroup_subsys_enabled_key
[i
]);
6229 pr_info("Disabling %s control group subsystem\n",
6235 __setup("cgroup_disable=", cgroup_disable
);
6237 void __init __weak
enable_debug_cgroup(void) { }
6239 static int __init
enable_cgroup_debug(char *str
)
6241 cgroup_debug
= true;
6242 enable_debug_cgroup();
6245 __setup("cgroup_debug", enable_cgroup_debug
);
6248 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6249 * @dentry: directory dentry of interest
6250 * @ss: subsystem of interest
6252 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6253 * to get the corresponding css and return it. If such css doesn't exist
6254 * or can't be pinned, an ERR_PTR value is returned.
6256 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6257 struct cgroup_subsys
*ss
)
6259 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6260 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6261 struct cgroup_subsys_state
*css
= NULL
;
6262 struct cgroup
*cgrp
;
6264 /* is @dentry a cgroup dir? */
6265 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6266 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6267 return ERR_PTR(-EBADF
);
6272 * This path doesn't originate from kernfs and @kn could already
6273 * have been or be removed at any point. @kn->priv is RCU
6274 * protected for this access. See css_release_work_fn() for details.
6276 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6278 css
= cgroup_css(cgrp
, ss
);
6280 if (!css
|| !css_tryget_online(css
))
6281 css
= ERR_PTR(-ENOENT
);
6288 * css_from_id - lookup css by id
6289 * @id: the cgroup id
6290 * @ss: cgroup subsys to be looked into
6292 * Returns the css if there's valid one with @id, otherwise returns NULL.
6293 * Should be called under rcu_read_lock().
6295 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6297 WARN_ON_ONCE(!rcu_read_lock_held());
6298 return idr_find(&ss
->css_idr
, id
);
6302 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6303 * @path: path on the default hierarchy
6305 * Find the cgroup at @path on the default hierarchy, increment its
6306 * reference count and return it. Returns pointer to the found cgroup on
6307 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6308 * if @path points to a non-directory.
6310 struct cgroup
*cgroup_get_from_path(const char *path
)
6312 struct kernfs_node
*kn
;
6313 struct cgroup
*cgrp
;
6315 mutex_lock(&cgroup_mutex
);
6317 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
6319 if (kernfs_type(kn
) == KERNFS_DIR
) {
6321 cgroup_get_live(cgrp
);
6323 cgrp
= ERR_PTR(-ENOTDIR
);
6327 cgrp
= ERR_PTR(-ENOENT
);
6330 mutex_unlock(&cgroup_mutex
);
6333 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6336 * cgroup_get_from_fd - get a cgroup pointer from a fd
6337 * @fd: fd obtained by open(cgroup2_dir)
6339 * Find the cgroup from a fd which should be obtained
6340 * by opening a cgroup directory. Returns a pointer to the
6341 * cgroup on success. ERR_PTR is returned if the cgroup
6344 struct cgroup
*cgroup_get_from_fd(int fd
)
6346 struct cgroup_subsys_state
*css
;
6347 struct cgroup
*cgrp
;
6352 return ERR_PTR(-EBADF
);
6354 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
6357 return ERR_CAST(css
);
6360 if (!cgroup_on_dfl(cgrp
)) {
6362 return ERR_PTR(-EBADF
);
6367 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6369 static u64
power_of_ten(int power
)
6378 * cgroup_parse_float - parse a floating number
6379 * @input: input string
6380 * @dec_shift: number of decimal digits to shift
6383 * Parse a decimal floating point number in @input and store the result in
6384 * @v with decimal point right shifted @dec_shift times. For example, if
6385 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6386 * Returns 0 on success, -errno otherwise.
6388 * There's nothing cgroup specific about this function except that it's
6389 * currently the only user.
6391 int cgroup_parse_float(const char *input
, unsigned dec_shift
, s64
*v
)
6393 s64 whole
, frac
= 0;
6394 int fstart
= 0, fend
= 0, flen
;
6396 if (!sscanf(input
, "%lld.%n%lld%n", &whole
, &fstart
, &frac
, &fend
))
6401 flen
= fend
> fstart
? fend
- fstart
: 0;
6402 if (flen
< dec_shift
)
6403 frac
*= power_of_ten(dec_shift
- flen
);
6405 frac
= DIV_ROUND_CLOSEST_ULL(frac
, power_of_ten(flen
- dec_shift
));
6407 *v
= whole
* power_of_ten(dec_shift
) + frac
;
6412 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6413 * definition in cgroup-defs.h.
6415 #ifdef CONFIG_SOCK_CGROUP_DATA
6417 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6419 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
6420 static bool cgroup_sk_alloc_disabled __read_mostly
;
6422 void cgroup_sk_alloc_disable(void)
6424 if (cgroup_sk_alloc_disabled
)
6426 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6427 cgroup_sk_alloc_disabled
= true;
6432 #define cgroup_sk_alloc_disabled false
6436 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6438 if (cgroup_sk_alloc_disabled
) {
6439 skcd
->no_refcnt
= 1;
6443 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6450 struct css_set
*cset
;
6452 cset
= task_css_set(current
);
6453 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6454 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
6455 cgroup_bpf_get(cset
->dfl_cgrp
);
6464 void cgroup_sk_clone(struct sock_cgroup_data
*skcd
)
6467 if (skcd
->no_refcnt
)
6470 * We might be cloning a socket which is left in an empty
6471 * cgroup and the cgroup might have already been rmdir'd.
6472 * Don't use cgroup_get_live().
6474 cgroup_get(sock_cgroup_ptr(skcd
));
6475 cgroup_bpf_get(sock_cgroup_ptr(skcd
));
6479 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6481 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6483 if (skcd
->no_refcnt
)
6485 cgroup_bpf_put(cgrp
);
6489 #endif /* CONFIG_SOCK_CGROUP_DATA */
6491 #ifdef CONFIG_CGROUP_BPF
6492 int cgroup_bpf_attach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6493 enum bpf_attach_type type
, u32 flags
)
6497 mutex_lock(&cgroup_mutex
);
6498 ret
= __cgroup_bpf_attach(cgrp
, prog
, type
, flags
);
6499 mutex_unlock(&cgroup_mutex
);
6502 int cgroup_bpf_detach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6503 enum bpf_attach_type type
, u32 flags
)
6507 mutex_lock(&cgroup_mutex
);
6508 ret
= __cgroup_bpf_detach(cgrp
, prog
, type
);
6509 mutex_unlock(&cgroup_mutex
);
6512 int cgroup_bpf_query(struct cgroup
*cgrp
, const union bpf_attr
*attr
,
6513 union bpf_attr __user
*uattr
)
6517 mutex_lock(&cgroup_mutex
);
6518 ret
= __cgroup_bpf_query(cgrp
, attr
, uattr
);
6519 mutex_unlock(&cgroup_mutex
);
6522 #endif /* CONFIG_CGROUP_BPF */
6525 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
6526 ssize_t size
, const char *prefix
)
6531 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
6532 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
6536 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
6538 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
6540 if (WARN_ON(ret
>= size
))
6547 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6550 struct cgroup_subsys
*ss
;
6554 ret
= show_delegatable_files(cgroup_base_files
, buf
, PAGE_SIZE
- ret
,
6557 for_each_subsys(ss
, ssid
)
6558 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
6560 cgroup_subsys_name
[ssid
]);
6564 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
6566 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6569 return snprintf(buf
, PAGE_SIZE
, "nsdelegate\nmemory_localevents\n");
6571 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
6573 static struct attribute
*cgroup_sysfs_attrs
[] = {
6574 &cgroup_delegate_attr
.attr
,
6575 &cgroup_features_attr
.attr
,
6579 static const struct attribute_group cgroup_sysfs_attr_group
= {
6580 .attrs
= cgroup_sysfs_attrs
,
6584 static int __init
cgroup_sysfs_init(void)
6586 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
6588 subsys_initcall(cgroup_sysfs_init
);
6590 #endif /* CONFIG_SYSFS */