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/bpf-cgroup.h>
34 #include <linux/cred.h>
35 #include <linux/errno.h>
36 #include <linux/init_task.h>
37 #include <linux/kernel.h>
38 #include <linux/magic.h>
39 #include <linux/mutex.h>
40 #include <linux/mount.h>
41 #include <linux/pagemap.h>
42 #include <linux/proc_fs.h>
43 #include <linux/rcupdate.h>
44 #include <linux/sched.h>
45 #include <linux/sched/task.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/percpu-rwsem.h>
49 #include <linux/string.h>
50 #include <linux/hashtable.h>
51 #include <linux/idr.h>
52 #include <linux/kthread.h>
53 #include <linux/atomic.h>
54 #include <linux/cpuset.h>
55 #include <linux/proc_ns.h>
56 #include <linux/nsproxy.h>
57 #include <linux/file.h>
58 #include <linux/fs_parser.h>
59 #include <linux/sched/cputime.h>
60 #include <linux/psi.h>
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/cgroup.h>
66 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
68 /* let's not notify more than 100 times per second */
69 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
72 * To avoid confusing the compiler (and generating warnings) with code
73 * that attempts to access what would be a 0-element array (i.e. sized
74 * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
75 * constant expression can be added.
77 #define CGROUP_HAS_SUBSYS_CONFIG (CGROUP_SUBSYS_COUNT > 0)
80 * cgroup_mutex is the master lock. Any modification to cgroup or its
81 * hierarchy must be performed while holding it.
83 * css_set_lock protects task->cgroups pointer, the list of css_set
84 * objects, and the chain of tasks off each css_set.
86 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
87 * cgroup.h can use them for lockdep annotations.
89 DEFINE_MUTEX(cgroup_mutex
);
90 DEFINE_SPINLOCK(css_set_lock
);
92 #ifdef CONFIG_PROVE_RCU
93 EXPORT_SYMBOL_GPL(cgroup_mutex
);
94 EXPORT_SYMBOL_GPL(css_set_lock
);
97 DEFINE_SPINLOCK(trace_cgroup_path_lock
);
98 char trace_cgroup_path
[TRACE_CGROUP_PATH_LEN
];
99 static bool cgroup_debug __read_mostly
;
102 * Protects cgroup_idr and css_idr so that IDs can be released without
103 * grabbing cgroup_mutex.
105 static DEFINE_SPINLOCK(cgroup_idr_lock
);
108 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
109 * against file removal/re-creation across css hiding.
111 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
113 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem
);
115 #define cgroup_assert_mutex_or_rcu_locked() \
116 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
117 !lockdep_is_held(&cgroup_mutex), \
118 "cgroup_mutex or RCU read lock required");
121 * cgroup destruction makes heavy use of work items and there can be a lot
122 * of concurrent destructions. Use a separate workqueue so that cgroup
123 * destruction work items don't end up filling up max_active of system_wq
124 * which may lead to deadlock.
126 static struct workqueue_struct
*cgroup_destroy_wq
;
128 /* generate an array of cgroup subsystem pointers */
129 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
130 struct cgroup_subsys
*cgroup_subsys
[] = {
131 #include <linux/cgroup_subsys.h>
135 /* array of cgroup subsystem names */
136 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
137 static const char *cgroup_subsys_name
[] = {
138 #include <linux/cgroup_subsys.h>
142 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
144 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
145 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
146 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
147 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
148 #include <linux/cgroup_subsys.h>
151 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
152 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
153 #include <linux/cgroup_subsys.h>
157 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
158 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
159 #include <linux/cgroup_subsys.h>
163 static DEFINE_PER_CPU(struct cgroup_rstat_cpu
, cgrp_dfl_root_rstat_cpu
);
165 /* the default hierarchy */
166 struct cgroup_root cgrp_dfl_root
= { .cgrp
.rstat_cpu
= &cgrp_dfl_root_rstat_cpu
};
167 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
170 * The default hierarchy always exists but is hidden until mounted for the
171 * first time. This is for backward compatibility.
173 static bool cgrp_dfl_visible
;
175 /* some controllers are not supported in the default hierarchy */
176 static u16 cgrp_dfl_inhibit_ss_mask
;
178 /* some controllers are implicitly enabled on the default hierarchy */
179 static u16 cgrp_dfl_implicit_ss_mask
;
181 /* some controllers can be threaded on the default hierarchy */
182 static u16 cgrp_dfl_threaded_ss_mask
;
184 /* The list of hierarchy roots */
185 LIST_HEAD(cgroup_roots
);
186 static int cgroup_root_count
;
188 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
189 static DEFINE_IDR(cgroup_hierarchy_idr
);
192 * Assign a monotonically increasing serial number to csses. It guarantees
193 * cgroups with bigger numbers are newer than those with smaller numbers.
194 * Also, as csses are always appended to the parent's ->children list, it
195 * guarantees that sibling csses are always sorted in the ascending serial
196 * number order on the list. Protected by cgroup_mutex.
198 static u64 css_serial_nr_next
= 1;
201 * These bitmasks identify subsystems with specific features to avoid
202 * having to do iterative checks repeatedly.
204 static u16 have_fork_callback __read_mostly
;
205 static u16 have_exit_callback __read_mostly
;
206 static u16 have_release_callback __read_mostly
;
207 static u16 have_canfork_callback __read_mostly
;
209 /* cgroup namespace for init task */
210 struct cgroup_namespace init_cgroup_ns
= {
211 .ns
.count
= REFCOUNT_INIT(2),
212 .user_ns
= &init_user_ns
,
213 .ns
.ops
= &cgroupns_operations
,
214 .ns
.inum
= PROC_CGROUP_INIT_INO
,
215 .root_cset
= &init_css_set
,
218 static struct file_system_type cgroup2_fs_type
;
219 static struct cftype cgroup_base_files
[];
220 static struct cftype cgroup_psi_files
[];
222 /* cgroup optional features */
223 enum cgroup_opt_features
{
225 OPT_FEATURE_PRESSURE
,
230 static const char *cgroup_opt_feature_names
[OPT_FEATURE_COUNT
] = {
236 static u16 cgroup_feature_disable_mask __read_mostly
;
238 static int cgroup_apply_control(struct cgroup
*cgrp
);
239 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
240 static void css_task_iter_skip(struct css_task_iter
*it
,
241 struct task_struct
*task
);
242 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
243 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
244 struct cgroup_subsys
*ss
);
245 static void css_release(struct percpu_ref
*ref
);
246 static void kill_css(struct cgroup_subsys_state
*css
);
247 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
248 struct cgroup
*cgrp
, struct cftype cfts
[],
252 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
253 * @ssid: subsys ID of interest
255 * cgroup_subsys_enabled() can only be used with literal subsys names which
256 * is fine for individual subsystems but unsuitable for cgroup core. This
257 * is slower static_key_enabled() based test indexed by @ssid.
259 bool cgroup_ssid_enabled(int ssid
)
261 if (!CGROUP_HAS_SUBSYS_CONFIG
)
264 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
268 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
269 * @cgrp: the cgroup of interest
271 * The default hierarchy is the v2 interface of cgroup and this function
272 * can be used to test whether a cgroup is on the default hierarchy for
273 * cases where a subsystem should behave differently depending on the
276 * List of changed behaviors:
278 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
279 * and "name" are disallowed.
281 * - When mounting an existing superblock, mount options should match.
283 * - rename(2) is disallowed.
285 * - "tasks" is removed. Everything should be at process granularity. Use
286 * "cgroup.procs" instead.
288 * - "cgroup.procs" is not sorted. pids will be unique unless they got
289 * recycled in-between reads.
291 * - "release_agent" and "notify_on_release" are removed. Replacement
292 * notification mechanism will be implemented.
294 * - "cgroup.clone_children" is removed.
296 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
297 * and its descendants contain no task; otherwise, 1. The file also
298 * generates kernfs notification which can be monitored through poll and
299 * [di]notify when the value of the file changes.
301 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
302 * take masks of ancestors with non-empty cpus/mems, instead of being
303 * moved to an ancestor.
305 * - cpuset: a task can be moved into an empty cpuset, and again it takes
306 * masks of ancestors.
308 * - blkcg: blk-throttle becomes properly hierarchical.
310 * - debug: disallowed on the default hierarchy.
312 bool cgroup_on_dfl(const struct cgroup
*cgrp
)
314 return cgrp
->root
== &cgrp_dfl_root
;
317 /* IDR wrappers which synchronize using cgroup_idr_lock */
318 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
323 idr_preload(gfp_mask
);
324 spin_lock_bh(&cgroup_idr_lock
);
325 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
326 spin_unlock_bh(&cgroup_idr_lock
);
331 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
335 spin_lock_bh(&cgroup_idr_lock
);
336 ret
= idr_replace(idr
, ptr
, id
);
337 spin_unlock_bh(&cgroup_idr_lock
);
341 static void cgroup_idr_remove(struct idr
*idr
, int id
)
343 spin_lock_bh(&cgroup_idr_lock
);
345 spin_unlock_bh(&cgroup_idr_lock
);
348 static bool cgroup_has_tasks(struct cgroup
*cgrp
)
350 return cgrp
->nr_populated_csets
;
353 bool cgroup_is_threaded(struct cgroup
*cgrp
)
355 return cgrp
->dom_cgrp
!= cgrp
;
358 /* can @cgrp host both domain and threaded children? */
359 static bool cgroup_is_mixable(struct cgroup
*cgrp
)
362 * Root isn't under domain level resource control exempting it from
363 * the no-internal-process constraint, so it can serve as a thread
364 * root and a parent of resource domains at the same time.
366 return !cgroup_parent(cgrp
);
369 /* can @cgrp become a thread root? Should always be true for a thread root */
370 static bool cgroup_can_be_thread_root(struct cgroup
*cgrp
)
372 /* mixables don't care */
373 if (cgroup_is_mixable(cgrp
))
376 /* domain roots can't be nested under threaded */
377 if (cgroup_is_threaded(cgrp
))
380 /* can only have either domain or threaded children */
381 if (cgrp
->nr_populated_domain_children
)
384 /* and no domain controllers can be enabled */
385 if (cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
391 /* is @cgrp root of a threaded subtree? */
392 bool cgroup_is_thread_root(struct cgroup
*cgrp
)
394 /* thread root should be a domain */
395 if (cgroup_is_threaded(cgrp
))
398 /* a domain w/ threaded children is a thread root */
399 if (cgrp
->nr_threaded_children
)
403 * A domain which has tasks and explicit threaded controllers
404 * enabled is a thread root.
406 if (cgroup_has_tasks(cgrp
) &&
407 (cgrp
->subtree_control
& cgrp_dfl_threaded_ss_mask
))
413 /* a domain which isn't connected to the root w/o brekage can't be used */
414 static bool cgroup_is_valid_domain(struct cgroup
*cgrp
)
416 /* the cgroup itself can be a thread root */
417 if (cgroup_is_threaded(cgrp
))
420 /* but the ancestors can't be unless mixable */
421 while ((cgrp
= cgroup_parent(cgrp
))) {
422 if (!cgroup_is_mixable(cgrp
) && cgroup_is_thread_root(cgrp
))
424 if (cgroup_is_threaded(cgrp
))
431 /* subsystems visibly enabled on a cgroup */
432 static u16
cgroup_control(struct cgroup
*cgrp
)
434 struct cgroup
*parent
= cgroup_parent(cgrp
);
435 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
438 u16 ss_mask
= parent
->subtree_control
;
440 /* threaded cgroups can only have threaded controllers */
441 if (cgroup_is_threaded(cgrp
))
442 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
446 if (cgroup_on_dfl(cgrp
))
447 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
448 cgrp_dfl_implicit_ss_mask
);
452 /* subsystems enabled on a cgroup */
453 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
455 struct cgroup
*parent
= cgroup_parent(cgrp
);
458 u16 ss_mask
= parent
->subtree_ss_mask
;
460 /* threaded cgroups can only have threaded controllers */
461 if (cgroup_is_threaded(cgrp
))
462 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
466 return cgrp
->root
->subsys_mask
;
470 * cgroup_css - obtain a cgroup's css for the specified subsystem
471 * @cgrp: the cgroup of interest
472 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
474 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
475 * function must be called either under cgroup_mutex or rcu_read_lock() and
476 * the caller is responsible for pinning the returned css if it wants to
477 * keep accessing it outside the said locks. This function may return
478 * %NULL if @cgrp doesn't have @subsys_id enabled.
480 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
481 struct cgroup_subsys
*ss
)
483 if (CGROUP_HAS_SUBSYS_CONFIG
&& ss
)
484 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
485 lockdep_is_held(&cgroup_mutex
));
491 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
492 * @cgrp: the cgroup of interest
493 * @ss: the subsystem of interest
495 * Find and get @cgrp's css associated with @ss. If the css doesn't exist
496 * or is offline, %NULL is returned.
498 static struct cgroup_subsys_state
*cgroup_tryget_css(struct cgroup
*cgrp
,
499 struct cgroup_subsys
*ss
)
501 struct cgroup_subsys_state
*css
;
504 css
= cgroup_css(cgrp
, ss
);
505 if (css
&& !css_tryget_online(css
))
513 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
514 * @cgrp: the cgroup of interest
515 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
517 * Similar to cgroup_css() but returns the effective css, which is defined
518 * as the matching css of the nearest ancestor including self which has @ss
519 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
520 * function is guaranteed to return non-NULL css.
522 static struct cgroup_subsys_state
*cgroup_e_css_by_mask(struct cgroup
*cgrp
,
523 struct cgroup_subsys
*ss
)
525 lockdep_assert_held(&cgroup_mutex
);
531 * This function is used while updating css associations and thus
532 * can't test the csses directly. Test ss_mask.
534 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
535 cgrp
= cgroup_parent(cgrp
);
540 return cgroup_css(cgrp
, ss
);
544 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
545 * @cgrp: the cgroup of interest
546 * @ss: the subsystem of interest
548 * Find and get the effective css of @cgrp for @ss. The effective css is
549 * defined as the matching css of the nearest ancestor including self which
550 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
551 * the root css is returned, so this function always returns a valid css.
553 * The returned css is not guaranteed to be online, and therefore it is the
554 * callers responsibility to try get a reference for it.
556 struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
557 struct cgroup_subsys
*ss
)
559 struct cgroup_subsys_state
*css
;
561 if (!CGROUP_HAS_SUBSYS_CONFIG
)
565 css
= cgroup_css(cgrp
, ss
);
569 cgrp
= cgroup_parent(cgrp
);
572 return init_css_set
.subsys
[ss
->id
];
576 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
577 * @cgrp: the cgroup of interest
578 * @ss: the subsystem of interest
580 * Find and get the effective css of @cgrp for @ss. The effective css is
581 * defined as the matching css of the nearest ancestor including self which
582 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
583 * the root css is returned, so this function always returns a valid css.
584 * The returned css must be put using css_put().
586 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
587 struct cgroup_subsys
*ss
)
589 struct cgroup_subsys_state
*css
;
591 if (!CGROUP_HAS_SUBSYS_CONFIG
)
597 css
= cgroup_css(cgrp
, ss
);
599 if (css
&& css_tryget_online(css
))
601 cgrp
= cgroup_parent(cgrp
);
604 css
= init_css_set
.subsys
[ss
->id
];
610 EXPORT_SYMBOL_GPL(cgroup_get_e_css
);
612 static void cgroup_get_live(struct cgroup
*cgrp
)
614 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
615 css_get(&cgrp
->self
);
619 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
620 * is responsible for taking the css_set_lock.
621 * @cgrp: the cgroup in question
623 int __cgroup_task_count(const struct cgroup
*cgrp
)
626 struct cgrp_cset_link
*link
;
628 lockdep_assert_held(&css_set_lock
);
630 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
631 count
+= link
->cset
->nr_tasks
;
637 * cgroup_task_count - count the number of tasks in a cgroup.
638 * @cgrp: the cgroup in question
640 int cgroup_task_count(const struct cgroup
*cgrp
)
644 spin_lock_irq(&css_set_lock
);
645 count
= __cgroup_task_count(cgrp
);
646 spin_unlock_irq(&css_set_lock
);
651 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
653 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
654 struct cftype
*cft
= of_cft(of
);
657 * This is open and unprotected implementation of cgroup_css().
658 * seq_css() is only called from a kernfs file operation which has
659 * an active reference on the file. Because all the subsystem
660 * files are drained before a css is disassociated with a cgroup,
661 * the matching css from the cgroup's subsys table is guaranteed to
662 * be and stay valid until the enclosing operation is complete.
664 if (CGROUP_HAS_SUBSYS_CONFIG
&& cft
->ss
)
665 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
669 EXPORT_SYMBOL_GPL(of_css
);
672 * for_each_css - iterate all css's of a cgroup
673 * @css: the iteration cursor
674 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
675 * @cgrp: the target cgroup to iterate css's of
677 * Should be called under cgroup_[tree_]mutex.
679 #define for_each_css(css, ssid, cgrp) \
680 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
681 if (!((css) = rcu_dereference_check( \
682 (cgrp)->subsys[(ssid)], \
683 lockdep_is_held(&cgroup_mutex)))) { } \
687 * for_each_e_css - iterate all effective css's of a cgroup
688 * @css: the iteration cursor
689 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
690 * @cgrp: the target cgroup to iterate css's of
692 * Should be called under cgroup_[tree_]mutex.
694 #define for_each_e_css(css, ssid, cgrp) \
695 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
696 if (!((css) = cgroup_e_css_by_mask(cgrp, \
697 cgroup_subsys[(ssid)]))) \
702 * do_each_subsys_mask - filter for_each_subsys with a bitmask
703 * @ss: the iteration cursor
704 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
705 * @ss_mask: the bitmask
707 * The block will only run for cases where the ssid-th bit (1 << ssid) of
710 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
711 unsigned long __ss_mask = (ss_mask); \
712 if (!CGROUP_HAS_SUBSYS_CONFIG) { \
716 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
717 (ss) = cgroup_subsys[ssid]; \
720 #define while_each_subsys_mask() \
725 /* iterate over child cgrps, lock should be held throughout iteration */
726 #define cgroup_for_each_live_child(child, cgrp) \
727 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
728 if (({ lockdep_assert_held(&cgroup_mutex); \
729 cgroup_is_dead(child); })) \
733 /* walk live descendants in pre order */
734 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
735 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
736 if (({ lockdep_assert_held(&cgroup_mutex); \
737 (dsct) = (d_css)->cgroup; \
738 cgroup_is_dead(dsct); })) \
742 /* walk live descendants in postorder */
743 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
744 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
745 if (({ lockdep_assert_held(&cgroup_mutex); \
746 (dsct) = (d_css)->cgroup; \
747 cgroup_is_dead(dsct); })) \
752 * The default css_set - used by init and its children prior to any
753 * hierarchies being mounted. It contains a pointer to the root state
754 * for each subsystem. Also used to anchor the list of css_sets. Not
755 * reference-counted, to improve performance when child cgroups
756 * haven't been created.
758 struct css_set init_css_set
= {
759 .refcount
= REFCOUNT_INIT(1),
760 .dom_cset
= &init_css_set
,
761 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
762 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
763 .dying_tasks
= LIST_HEAD_INIT(init_css_set
.dying_tasks
),
764 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
765 .threaded_csets
= LIST_HEAD_INIT(init_css_set
.threaded_csets
),
766 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
767 .mg_src_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_src_preload_node
),
768 .mg_dst_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_dst_preload_node
),
769 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
772 * The following field is re-initialized when this cset gets linked
773 * in cgroup_init(). However, let's initialize the field
774 * statically too so that the default cgroup can be accessed safely
777 .dfl_cgrp
= &cgrp_dfl_root
.cgrp
,
780 static int css_set_count
= 1; /* 1 for init_css_set */
782 static bool css_set_threaded(struct css_set
*cset
)
784 return cset
->dom_cset
!= cset
;
788 * css_set_populated - does a css_set contain any tasks?
789 * @cset: target css_set
791 * css_set_populated() should be the same as !!cset->nr_tasks at steady
792 * state. However, css_set_populated() can be called while a task is being
793 * added to or removed from the linked list before the nr_tasks is
794 * properly updated. Hence, we can't just look at ->nr_tasks here.
796 static bool css_set_populated(struct css_set
*cset
)
798 lockdep_assert_held(&css_set_lock
);
800 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
804 * cgroup_update_populated - update the populated count of a cgroup
805 * @cgrp: the target cgroup
806 * @populated: inc or dec populated count
808 * One of the css_sets associated with @cgrp is either getting its first
809 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
810 * count is propagated towards root so that a given cgroup's
811 * nr_populated_children is zero iff none of its descendants contain any
814 * @cgrp's interface file "cgroup.populated" is zero if both
815 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
816 * 1 otherwise. When the sum changes from or to zero, userland is notified
817 * that the content of the interface file has changed. This can be used to
818 * detect when @cgrp and its descendants become populated or empty.
820 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
822 struct cgroup
*child
= NULL
;
823 int adj
= populated
? 1 : -1;
825 lockdep_assert_held(&css_set_lock
);
828 bool was_populated
= cgroup_is_populated(cgrp
);
831 cgrp
->nr_populated_csets
+= adj
;
833 if (cgroup_is_threaded(child
))
834 cgrp
->nr_populated_threaded_children
+= adj
;
836 cgrp
->nr_populated_domain_children
+= adj
;
839 if (was_populated
== cgroup_is_populated(cgrp
))
842 cgroup1_check_for_release(cgrp
);
843 TRACE_CGROUP_PATH(notify_populated
, cgrp
,
844 cgroup_is_populated(cgrp
));
845 cgroup_file_notify(&cgrp
->events_file
);
848 cgrp
= cgroup_parent(cgrp
);
853 * css_set_update_populated - update populated state of a css_set
854 * @cset: target css_set
855 * @populated: whether @cset is populated or depopulated
857 * @cset is either getting the first task or losing the last. Update the
858 * populated counters of all associated cgroups accordingly.
860 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
862 struct cgrp_cset_link
*link
;
864 lockdep_assert_held(&css_set_lock
);
866 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
867 cgroup_update_populated(link
->cgrp
, populated
);
871 * @task is leaving, advance task iterators which are pointing to it so
872 * that they can resume at the next position. Advancing an iterator might
873 * remove it from the list, use safe walk. See css_task_iter_skip() for
876 static void css_set_skip_task_iters(struct css_set
*cset
,
877 struct task_struct
*task
)
879 struct css_task_iter
*it
, *pos
;
881 list_for_each_entry_safe(it
, pos
, &cset
->task_iters
, iters_node
)
882 css_task_iter_skip(it
, task
);
886 * css_set_move_task - move a task from one css_set to another
887 * @task: task being moved
888 * @from_cset: css_set @task currently belongs to (may be NULL)
889 * @to_cset: new css_set @task is being moved to (may be NULL)
890 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
892 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
893 * css_set, @from_cset can be NULL. If @task is being disassociated
894 * instead of moved, @to_cset can be NULL.
896 * This function automatically handles populated counter updates and
897 * css_task_iter adjustments but the caller is responsible for managing
898 * @from_cset and @to_cset's reference counts.
900 static void css_set_move_task(struct task_struct
*task
,
901 struct css_set
*from_cset
, struct css_set
*to_cset
,
904 lockdep_assert_held(&css_set_lock
);
906 if (to_cset
&& !css_set_populated(to_cset
))
907 css_set_update_populated(to_cset
, true);
910 WARN_ON_ONCE(list_empty(&task
->cg_list
));
912 css_set_skip_task_iters(from_cset
, task
);
913 list_del_init(&task
->cg_list
);
914 if (!css_set_populated(from_cset
))
915 css_set_update_populated(from_cset
, false);
917 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
922 * We are synchronized through cgroup_threadgroup_rwsem
923 * against PF_EXITING setting such that we can't race
924 * against cgroup_exit()/cgroup_free() dropping the css_set.
926 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
928 cgroup_move_task(task
, to_cset
);
929 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
935 * hash table for cgroup groups. This improves the performance to find
936 * an existing css_set. This hash doesn't (currently) take into
937 * account cgroups in empty hierarchies.
939 #define CSS_SET_HASH_BITS 7
940 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
942 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
944 unsigned long key
= 0UL;
945 struct cgroup_subsys
*ss
;
948 for_each_subsys(ss
, i
)
949 key
+= (unsigned long)css
[i
];
950 key
= (key
>> 16) ^ key
;
955 void put_css_set_locked(struct css_set
*cset
)
957 struct cgrp_cset_link
*link
, *tmp_link
;
958 struct cgroup_subsys
*ss
;
961 lockdep_assert_held(&css_set_lock
);
963 if (!refcount_dec_and_test(&cset
->refcount
))
966 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
968 /* This css_set is dead. Unlink it and release cgroup and css refs */
969 for_each_subsys(ss
, ssid
) {
970 list_del(&cset
->e_cset_node
[ssid
]);
971 css_put(cset
->subsys
[ssid
]);
973 hash_del(&cset
->hlist
);
976 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
977 list_del(&link
->cset_link
);
978 list_del(&link
->cgrp_link
);
979 if (cgroup_parent(link
->cgrp
))
980 cgroup_put(link
->cgrp
);
984 if (css_set_threaded(cset
)) {
985 list_del(&cset
->threaded_csets_node
);
986 put_css_set_locked(cset
->dom_cset
);
989 kfree_rcu(cset
, rcu_head
);
993 * compare_css_sets - helper function for find_existing_css_set().
994 * @cset: candidate css_set being tested
995 * @old_cset: existing css_set for a task
996 * @new_cgrp: cgroup that's being entered by the task
997 * @template: desired set of css pointers in css_set (pre-calculated)
999 * Returns true if "cset" matches "old_cset" except for the hierarchy
1000 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
1002 static bool compare_css_sets(struct css_set
*cset
,
1003 struct css_set
*old_cset
,
1004 struct cgroup
*new_cgrp
,
1005 struct cgroup_subsys_state
*template[])
1007 struct cgroup
*new_dfl_cgrp
;
1008 struct list_head
*l1
, *l2
;
1011 * On the default hierarchy, there can be csets which are
1012 * associated with the same set of cgroups but different csses.
1013 * Let's first ensure that csses match.
1015 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
1019 /* @cset's domain should match the default cgroup's */
1020 if (cgroup_on_dfl(new_cgrp
))
1021 new_dfl_cgrp
= new_cgrp
;
1023 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
1025 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
1029 * Compare cgroup pointers in order to distinguish between
1030 * different cgroups in hierarchies. As different cgroups may
1031 * share the same effective css, this comparison is always
1034 l1
= &cset
->cgrp_links
;
1035 l2
= &old_cset
->cgrp_links
;
1037 struct cgrp_cset_link
*link1
, *link2
;
1038 struct cgroup
*cgrp1
, *cgrp2
;
1042 /* See if we reached the end - both lists are equal length. */
1043 if (l1
== &cset
->cgrp_links
) {
1044 BUG_ON(l2
!= &old_cset
->cgrp_links
);
1047 BUG_ON(l2
== &old_cset
->cgrp_links
);
1049 /* Locate the cgroups associated with these links. */
1050 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
1051 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
1052 cgrp1
= link1
->cgrp
;
1053 cgrp2
= link2
->cgrp
;
1054 /* Hierarchies should be linked in the same order. */
1055 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
1058 * If this hierarchy is the hierarchy of the cgroup
1059 * that's changing, then we need to check that this
1060 * css_set points to the new cgroup; if it's any other
1061 * hierarchy, then this css_set should point to the
1062 * same cgroup as the old css_set.
1064 if (cgrp1
->root
== new_cgrp
->root
) {
1065 if (cgrp1
!= new_cgrp
)
1076 * find_existing_css_set - init css array and find the matching css_set
1077 * @old_cset: the css_set that we're using before the cgroup transition
1078 * @cgrp: the cgroup that we're moving into
1079 * @template: out param for the new set of csses, should be clear on entry
1081 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
1082 struct cgroup
*cgrp
,
1083 struct cgroup_subsys_state
*template[])
1085 struct cgroup_root
*root
= cgrp
->root
;
1086 struct cgroup_subsys
*ss
;
1087 struct css_set
*cset
;
1092 * Build the set of subsystem state objects that we want to see in the
1093 * new css_set. While subsystems can change globally, the entries here
1094 * won't change, so no need for locking.
1096 for_each_subsys(ss
, i
) {
1097 if (root
->subsys_mask
& (1UL << i
)) {
1099 * @ss is in this hierarchy, so we want the
1100 * effective css from @cgrp.
1102 template[i
] = cgroup_e_css_by_mask(cgrp
, ss
);
1105 * @ss is not in this hierarchy, so we don't want
1106 * to change the css.
1108 template[i
] = old_cset
->subsys
[i
];
1112 key
= css_set_hash(template);
1113 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
1114 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
1117 /* This css_set matches what we need */
1121 /* No existing cgroup group matched */
1125 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
1127 struct cgrp_cset_link
*link
, *tmp_link
;
1129 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
1130 list_del(&link
->cset_link
);
1136 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1137 * @count: the number of links to allocate
1138 * @tmp_links: list_head the allocated links are put on
1140 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1141 * through ->cset_link. Returns 0 on success or -errno.
1143 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1145 struct cgrp_cset_link
*link
;
1148 INIT_LIST_HEAD(tmp_links
);
1150 for (i
= 0; i
< count
; i
++) {
1151 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1153 free_cgrp_cset_links(tmp_links
);
1156 list_add(&link
->cset_link
, tmp_links
);
1162 * link_css_set - a helper function to link a css_set to a cgroup
1163 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1164 * @cset: the css_set to be linked
1165 * @cgrp: the destination cgroup
1167 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1168 struct cgroup
*cgrp
)
1170 struct cgrp_cset_link
*link
;
1172 BUG_ON(list_empty(tmp_links
));
1174 if (cgroup_on_dfl(cgrp
))
1175 cset
->dfl_cgrp
= cgrp
;
1177 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1182 * Always add links to the tail of the lists so that the lists are
1183 * in chronological order.
1185 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1186 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1188 if (cgroup_parent(cgrp
))
1189 cgroup_get_live(cgrp
);
1193 * find_css_set - return a new css_set with one cgroup updated
1194 * @old_cset: the baseline css_set
1195 * @cgrp: the cgroup to be updated
1197 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1198 * substituted into the appropriate hierarchy.
1200 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1201 struct cgroup
*cgrp
)
1203 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1204 struct css_set
*cset
;
1205 struct list_head tmp_links
;
1206 struct cgrp_cset_link
*link
;
1207 struct cgroup_subsys
*ss
;
1211 lockdep_assert_held(&cgroup_mutex
);
1213 /* First see if we already have a cgroup group that matches
1214 * the desired set */
1215 spin_lock_irq(&css_set_lock
);
1216 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1219 spin_unlock_irq(&css_set_lock
);
1224 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1228 /* Allocate all the cgrp_cset_link objects that we'll need */
1229 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1234 refcount_set(&cset
->refcount
, 1);
1235 cset
->dom_cset
= cset
;
1236 INIT_LIST_HEAD(&cset
->tasks
);
1237 INIT_LIST_HEAD(&cset
->mg_tasks
);
1238 INIT_LIST_HEAD(&cset
->dying_tasks
);
1239 INIT_LIST_HEAD(&cset
->task_iters
);
1240 INIT_LIST_HEAD(&cset
->threaded_csets
);
1241 INIT_HLIST_NODE(&cset
->hlist
);
1242 INIT_LIST_HEAD(&cset
->cgrp_links
);
1243 INIT_LIST_HEAD(&cset
->mg_src_preload_node
);
1244 INIT_LIST_HEAD(&cset
->mg_dst_preload_node
);
1245 INIT_LIST_HEAD(&cset
->mg_node
);
1247 /* Copy the set of subsystem state objects generated in
1248 * find_existing_css_set() */
1249 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1251 spin_lock_irq(&css_set_lock
);
1252 /* Add reference counts and links from the new css_set. */
1253 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1254 struct cgroup
*c
= link
->cgrp
;
1256 if (c
->root
== cgrp
->root
)
1258 link_css_set(&tmp_links
, cset
, c
);
1261 BUG_ON(!list_empty(&tmp_links
));
1265 /* Add @cset to the hash table */
1266 key
= css_set_hash(cset
->subsys
);
1267 hash_add(css_set_table
, &cset
->hlist
, key
);
1269 for_each_subsys(ss
, ssid
) {
1270 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1272 list_add_tail(&cset
->e_cset_node
[ssid
],
1273 &css
->cgroup
->e_csets
[ssid
]);
1277 spin_unlock_irq(&css_set_lock
);
1280 * If @cset should be threaded, look up the matching dom_cset and
1281 * link them up. We first fully initialize @cset then look for the
1282 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1283 * to stay empty until we return.
1285 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1286 struct css_set
*dcset
;
1288 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1294 spin_lock_irq(&css_set_lock
);
1295 cset
->dom_cset
= dcset
;
1296 list_add_tail(&cset
->threaded_csets_node
,
1297 &dcset
->threaded_csets
);
1298 spin_unlock_irq(&css_set_lock
);
1304 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1306 struct cgroup
*root_cgrp
= kernfs_root_to_node(kf_root
)->priv
;
1308 return root_cgrp
->root
;
1311 void cgroup_favor_dynmods(struct cgroup_root
*root
, bool favor
)
1313 bool favoring
= root
->flags
& CGRP_ROOT_FAVOR_DYNMODS
;
1315 /* see the comment above CGRP_ROOT_FAVOR_DYNMODS definition */
1316 if (favor
&& !favoring
) {
1317 rcu_sync_enter(&cgroup_threadgroup_rwsem
.rss
);
1318 root
->flags
|= CGRP_ROOT_FAVOR_DYNMODS
;
1319 } else if (!favor
&& favoring
) {
1320 rcu_sync_exit(&cgroup_threadgroup_rwsem
.rss
);
1321 root
->flags
&= ~CGRP_ROOT_FAVOR_DYNMODS
;
1325 static int cgroup_init_root_id(struct cgroup_root
*root
)
1329 lockdep_assert_held(&cgroup_mutex
);
1331 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1335 root
->hierarchy_id
= id
;
1339 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1341 lockdep_assert_held(&cgroup_mutex
);
1343 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1346 void cgroup_free_root(struct cgroup_root
*root
)
1351 static void cgroup_destroy_root(struct cgroup_root
*root
)
1353 struct cgroup
*cgrp
= &root
->cgrp
;
1354 struct cgrp_cset_link
*link
, *tmp_link
;
1356 trace_cgroup_destroy_root(root
);
1358 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1360 BUG_ON(atomic_read(&root
->nr_cgrps
));
1361 BUG_ON(!list_empty(&cgrp
->self
.children
));
1363 /* Rebind all subsystems back to the default hierarchy */
1364 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1367 * Release all the links from cset_links to this hierarchy's
1370 spin_lock_irq(&css_set_lock
);
1372 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1373 list_del(&link
->cset_link
);
1374 list_del(&link
->cgrp_link
);
1378 spin_unlock_irq(&css_set_lock
);
1380 if (!list_empty(&root
->root_list
)) {
1381 list_del(&root
->root_list
);
1382 cgroup_root_count
--;
1385 cgroup_favor_dynmods(root
, false);
1386 cgroup_exit_root_id(root
);
1388 mutex_unlock(&cgroup_mutex
);
1390 cgroup_rstat_exit(cgrp
);
1391 kernfs_destroy_root(root
->kf_root
);
1392 cgroup_free_root(root
);
1395 static inline struct cgroup
*__cset_cgroup_from_root(struct css_set
*cset
,
1396 struct cgroup_root
*root
)
1398 struct cgroup
*res_cgroup
= NULL
;
1400 if (cset
== &init_css_set
) {
1401 res_cgroup
= &root
->cgrp
;
1402 } else if (root
== &cgrp_dfl_root
) {
1403 res_cgroup
= cset
->dfl_cgrp
;
1405 struct cgrp_cset_link
*link
;
1407 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1408 struct cgroup
*c
= link
->cgrp
;
1410 if (c
->root
== root
) {
1421 * look up cgroup associated with current task's cgroup namespace on the
1422 * specified hierarchy
1424 static struct cgroup
*
1425 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1427 struct cgroup
*res
= NULL
;
1428 struct css_set
*cset
;
1430 lockdep_assert_held(&css_set_lock
);
1434 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1435 res
= __cset_cgroup_from_root(cset
, root
);
1443 /* look up cgroup associated with given css_set on the specified hierarchy */
1444 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1445 struct cgroup_root
*root
)
1447 struct cgroup
*res
= NULL
;
1449 lockdep_assert_held(&cgroup_mutex
);
1450 lockdep_assert_held(&css_set_lock
);
1452 res
= __cset_cgroup_from_root(cset
, root
);
1459 * Return the cgroup for "task" from the given hierarchy. Must be
1460 * called with cgroup_mutex and css_set_lock held.
1462 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1463 struct cgroup_root
*root
)
1466 * No need to lock the task - since we hold css_set_lock the
1467 * task can't change groups.
1469 return cset_cgroup_from_root(task_css_set(task
), root
);
1473 * A task must hold cgroup_mutex to modify cgroups.
1475 * Any task can increment and decrement the count field without lock.
1476 * So in general, code holding cgroup_mutex can't rely on the count
1477 * field not changing. However, if the count goes to zero, then only
1478 * cgroup_attach_task() can increment it again. Because a count of zero
1479 * means that no tasks are currently attached, therefore there is no
1480 * way a task attached to that cgroup can fork (the other way to
1481 * increment the count). So code holding cgroup_mutex can safely
1482 * assume that if the count is zero, it will stay zero. Similarly, if
1483 * a task holds cgroup_mutex on a cgroup with zero count, it
1484 * knows that the cgroup won't be removed, as cgroup_rmdir()
1487 * A cgroup can only be deleted if both its 'count' of using tasks
1488 * is zero, and its list of 'children' cgroups is empty. Since all
1489 * tasks in the system use _some_ cgroup, and since there is always at
1490 * least one task in the system (init, pid == 1), therefore, root cgroup
1491 * always has either children cgroups and/or using tasks. So we don't
1492 * need a special hack to ensure that root cgroup cannot be deleted.
1494 * P.S. One more locking exception. RCU is used to guard the
1495 * update of a tasks cgroup pointer by cgroup_attach_task()
1498 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1500 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1503 struct cgroup_subsys
*ss
= cft
->ss
;
1505 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1506 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
1507 const char *dbg
= (cft
->flags
& CFTYPE_DEBUG
) ? ".__DEBUG__." : "";
1509 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s%s.%s",
1510 dbg
, cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1513 strscpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1519 * cgroup_file_mode - deduce file mode of a control file
1520 * @cft: the control file in question
1522 * S_IRUGO for read, S_IWUSR for write.
1524 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1528 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1531 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1532 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1542 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1543 * @subtree_control: the new subtree_control mask to consider
1544 * @this_ss_mask: available subsystems
1546 * On the default hierarchy, a subsystem may request other subsystems to be
1547 * enabled together through its ->depends_on mask. In such cases, more
1548 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1550 * This function calculates which subsystems need to be enabled if
1551 * @subtree_control is to be applied while restricted to @this_ss_mask.
1553 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1555 u16 cur_ss_mask
= subtree_control
;
1556 struct cgroup_subsys
*ss
;
1559 lockdep_assert_held(&cgroup_mutex
);
1561 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1564 u16 new_ss_mask
= cur_ss_mask
;
1566 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1567 new_ss_mask
|= ss
->depends_on
;
1568 } while_each_subsys_mask();
1571 * Mask out subsystems which aren't available. This can
1572 * happen only if some depended-upon subsystems were bound
1573 * to non-default hierarchies.
1575 new_ss_mask
&= this_ss_mask
;
1577 if (new_ss_mask
== cur_ss_mask
)
1579 cur_ss_mask
= new_ss_mask
;
1586 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1587 * @kn: the kernfs_node being serviced
1589 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1590 * the method finishes if locking succeeded. Note that once this function
1591 * returns the cgroup returned by cgroup_kn_lock_live() may become
1592 * inaccessible any time. If the caller intends to continue to access the
1593 * cgroup, it should pin it before invoking this function.
1595 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1597 struct cgroup
*cgrp
;
1599 if (kernfs_type(kn
) == KERNFS_DIR
)
1602 cgrp
= kn
->parent
->priv
;
1604 mutex_unlock(&cgroup_mutex
);
1606 kernfs_unbreak_active_protection(kn
);
1611 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1612 * @kn: the kernfs_node being serviced
1613 * @drain_offline: perform offline draining on the cgroup
1615 * This helper is to be used by a cgroup kernfs method currently servicing
1616 * @kn. It breaks the active protection, performs cgroup locking and
1617 * verifies that the associated cgroup is alive. Returns the cgroup if
1618 * alive; otherwise, %NULL. A successful return should be undone by a
1619 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1620 * cgroup is drained of offlining csses before return.
1622 * Any cgroup kernfs method implementation which requires locking the
1623 * associated cgroup should use this helper. It avoids nesting cgroup
1624 * locking under kernfs active protection and allows all kernfs operations
1625 * including self-removal.
1627 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1629 struct cgroup
*cgrp
;
1631 if (kernfs_type(kn
) == KERNFS_DIR
)
1634 cgrp
= kn
->parent
->priv
;
1637 * We're gonna grab cgroup_mutex which nests outside kernfs
1638 * active_ref. cgroup liveliness check alone provides enough
1639 * protection against removal. Ensure @cgrp stays accessible and
1640 * break the active_ref protection.
1642 if (!cgroup_tryget(cgrp
))
1644 kernfs_break_active_protection(kn
);
1647 cgroup_lock_and_drain_offline(cgrp
);
1649 mutex_lock(&cgroup_mutex
);
1651 if (!cgroup_is_dead(cgrp
))
1654 cgroup_kn_unlock(kn
);
1658 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1660 char name
[CGROUP_FILE_NAME_MAX
];
1662 lockdep_assert_held(&cgroup_mutex
);
1664 if (cft
->file_offset
) {
1665 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1666 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1668 spin_lock_irq(&cgroup_file_kn_lock
);
1670 spin_unlock_irq(&cgroup_file_kn_lock
);
1672 del_timer_sync(&cfile
->notify_timer
);
1675 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1679 * css_clear_dir - remove subsys files in a cgroup directory
1682 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1684 struct cgroup
*cgrp
= css
->cgroup
;
1685 struct cftype
*cfts
;
1687 if (!(css
->flags
& CSS_VISIBLE
))
1690 css
->flags
&= ~CSS_VISIBLE
;
1693 if (cgroup_on_dfl(cgrp
)) {
1694 cgroup_addrm_files(css
, cgrp
,
1695 cgroup_base_files
, false);
1696 if (cgroup_psi_enabled())
1697 cgroup_addrm_files(css
, cgrp
,
1698 cgroup_psi_files
, false);
1700 cgroup_addrm_files(css
, cgrp
,
1701 cgroup1_base_files
, false);
1704 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1705 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1710 * css_populate_dir - create subsys files in a cgroup directory
1713 * On failure, no file is added.
1715 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1717 struct cgroup
*cgrp
= css
->cgroup
;
1718 struct cftype
*cfts
, *failed_cfts
;
1721 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1725 if (cgroup_on_dfl(cgrp
)) {
1726 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
,
1727 cgroup_base_files
, true);
1731 if (cgroup_psi_enabled()) {
1732 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
,
1733 cgroup_psi_files
, true);
1738 cgroup_addrm_files(css
, cgrp
,
1739 cgroup1_base_files
, true);
1742 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1743 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1751 css
->flags
|= CSS_VISIBLE
;
1755 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1756 if (cfts
== failed_cfts
)
1758 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1763 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1765 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1766 struct cgroup_subsys
*ss
;
1768 u16 dfl_disable_ss_mask
= 0;
1770 lockdep_assert_held(&cgroup_mutex
);
1772 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1774 * If @ss has non-root csses attached to it, can't move.
1775 * If @ss is an implicit controller, it is exempt from this
1776 * rule and can be stolen.
1778 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1779 !ss
->implicit_on_dfl
)
1782 /* can't move between two non-dummy roots either */
1783 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1787 * Collect ssid's that need to be disabled from default
1790 if (ss
->root
== &cgrp_dfl_root
)
1791 dfl_disable_ss_mask
|= 1 << ssid
;
1793 } while_each_subsys_mask();
1795 if (dfl_disable_ss_mask
) {
1796 struct cgroup
*scgrp
= &cgrp_dfl_root
.cgrp
;
1799 * Controllers from default hierarchy that need to be rebound
1800 * are all disabled together in one go.
1802 cgrp_dfl_root
.subsys_mask
&= ~dfl_disable_ss_mask
;
1803 WARN_ON(cgroup_apply_control(scgrp
));
1804 cgroup_finalize_control(scgrp
, 0);
1807 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1808 struct cgroup_root
*src_root
= ss
->root
;
1809 struct cgroup
*scgrp
= &src_root
->cgrp
;
1810 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1811 struct css_set
*cset
;
1813 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1815 if (src_root
!= &cgrp_dfl_root
) {
1816 /* disable from the source */
1817 src_root
->subsys_mask
&= ~(1 << ssid
);
1818 WARN_ON(cgroup_apply_control(scgrp
));
1819 cgroup_finalize_control(scgrp
, 0);
1823 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1824 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1825 ss
->root
= dst_root
;
1826 css
->cgroup
= dcgrp
;
1828 spin_lock_irq(&css_set_lock
);
1829 hash_for_each(css_set_table
, i
, cset
, hlist
)
1830 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1831 &dcgrp
->e_csets
[ss
->id
]);
1832 spin_unlock_irq(&css_set_lock
);
1834 if (ss
->css_rstat_flush
) {
1835 list_del_rcu(&css
->rstat_css_node
);
1837 list_add_rcu(&css
->rstat_css_node
,
1838 &dcgrp
->rstat_css_list
);
1841 /* default hierarchy doesn't enable controllers by default */
1842 dst_root
->subsys_mask
|= 1 << ssid
;
1843 if (dst_root
== &cgrp_dfl_root
) {
1844 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1846 dcgrp
->subtree_control
|= 1 << ssid
;
1847 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1850 ret
= cgroup_apply_control(dcgrp
);
1852 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1857 } while_each_subsys_mask();
1859 kernfs_activate(dcgrp
->kn
);
1863 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1864 struct kernfs_root
*kf_root
)
1868 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1869 struct cgroup
*ns_cgroup
;
1871 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1875 spin_lock_irq(&css_set_lock
);
1876 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1877 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1878 spin_unlock_irq(&css_set_lock
);
1880 if (len
>= PATH_MAX
)
1883 seq_escape(sf
, buf
, " \t\n\\");
1890 enum cgroup2_param
{
1893 Opt_memory_localevents
,
1894 Opt_memory_recursiveprot
,
1898 static const struct fs_parameter_spec cgroup2_fs_parameters
[] = {
1899 fsparam_flag("nsdelegate", Opt_nsdelegate
),
1900 fsparam_flag("favordynmods", Opt_favordynmods
),
1901 fsparam_flag("memory_localevents", Opt_memory_localevents
),
1902 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot
),
1906 static int cgroup2_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
1908 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1909 struct fs_parse_result result
;
1912 opt
= fs_parse(fc
, cgroup2_fs_parameters
, param
, &result
);
1917 case Opt_nsdelegate
:
1918 ctx
->flags
|= CGRP_ROOT_NS_DELEGATE
;
1920 case Opt_favordynmods
:
1921 ctx
->flags
|= CGRP_ROOT_FAVOR_DYNMODS
;
1923 case Opt_memory_localevents
:
1924 ctx
->flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1926 case Opt_memory_recursiveprot
:
1927 ctx
->flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1933 static void apply_cgroup_root_flags(unsigned int root_flags
)
1935 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1936 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1937 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1939 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1941 cgroup_favor_dynmods(&cgrp_dfl_root
,
1942 root_flags
& CGRP_ROOT_FAVOR_DYNMODS
);
1944 if (root_flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1945 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1947 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1949 if (root_flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1950 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1952 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1956 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1958 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1959 seq_puts(seq
, ",nsdelegate");
1960 if (cgrp_dfl_root
.flags
& CGRP_ROOT_FAVOR_DYNMODS
)
1961 seq_puts(seq
, ",favordynmods");
1962 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1963 seq_puts(seq
, ",memory_localevents");
1964 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1965 seq_puts(seq
, ",memory_recursiveprot");
1969 static int cgroup_reconfigure(struct fs_context
*fc
)
1971 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1973 apply_cgroup_root_flags(ctx
->flags
);
1977 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1979 struct cgroup_subsys
*ss
;
1982 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1983 INIT_LIST_HEAD(&cgrp
->self
.children
);
1984 INIT_LIST_HEAD(&cgrp
->cset_links
);
1985 INIT_LIST_HEAD(&cgrp
->pidlists
);
1986 mutex_init(&cgrp
->pidlist_mutex
);
1987 cgrp
->self
.cgroup
= cgrp
;
1988 cgrp
->self
.flags
|= CSS_ONLINE
;
1989 cgrp
->dom_cgrp
= cgrp
;
1990 cgrp
->max_descendants
= INT_MAX
;
1991 cgrp
->max_depth
= INT_MAX
;
1992 INIT_LIST_HEAD(&cgrp
->rstat_css_list
);
1993 prev_cputime_init(&cgrp
->prev_cputime
);
1995 for_each_subsys(ss
, ssid
)
1996 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1998 init_waitqueue_head(&cgrp
->offline_waitq
);
1999 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
2002 void init_cgroup_root(struct cgroup_fs_context
*ctx
)
2004 struct cgroup_root
*root
= ctx
->root
;
2005 struct cgroup
*cgrp
= &root
->cgrp
;
2007 INIT_LIST_HEAD(&root
->root_list
);
2008 atomic_set(&root
->nr_cgrps
, 1);
2010 init_cgroup_housekeeping(cgrp
);
2012 /* DYNMODS must be modified through cgroup_favor_dynmods() */
2013 root
->flags
= ctx
->flags
& ~CGRP_ROOT_FAVOR_DYNMODS
;
2014 if (ctx
->release_agent
)
2015 strscpy(root
->release_agent_path
, ctx
->release_agent
, PATH_MAX
);
2017 strscpy(root
->name
, ctx
->name
, MAX_CGROUP_ROOT_NAMELEN
);
2018 if (ctx
->cpuset_clone_children
)
2019 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
2022 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
2024 LIST_HEAD(tmp_links
);
2025 struct cgroup
*root_cgrp
= &root
->cgrp
;
2026 struct kernfs_syscall_ops
*kf_sops
;
2027 struct css_set
*cset
;
2030 lockdep_assert_held(&cgroup_mutex
);
2032 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
2038 * We're accessing css_set_count without locking css_set_lock here,
2039 * but that's OK - it can only be increased by someone holding
2040 * cgroup_lock, and that's us. Later rebinding may disable
2041 * controllers on the default hierarchy and thus create new csets,
2042 * which can't be more than the existing ones. Allocate 2x.
2044 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
2048 ret
= cgroup_init_root_id(root
);
2052 kf_sops
= root
== &cgrp_dfl_root
?
2053 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
2055 root
->kf_root
= kernfs_create_root(kf_sops
,
2056 KERNFS_ROOT_CREATE_DEACTIVATED
|
2057 KERNFS_ROOT_SUPPORT_EXPORTOP
|
2058 KERNFS_ROOT_SUPPORT_USER_XATTR
,
2060 if (IS_ERR(root
->kf_root
)) {
2061 ret
= PTR_ERR(root
->kf_root
);
2064 root_cgrp
->kn
= kernfs_root_to_node(root
->kf_root
);
2065 WARN_ON_ONCE(cgroup_ino(root_cgrp
) != 1);
2066 root_cgrp
->ancestors
[0] = root_cgrp
;
2068 ret
= css_populate_dir(&root_cgrp
->self
);
2072 ret
= cgroup_rstat_init(root_cgrp
);
2076 ret
= rebind_subsystems(root
, ss_mask
);
2080 ret
= cgroup_bpf_inherit(root_cgrp
);
2083 trace_cgroup_setup_root(root
);
2086 * There must be no failure case after here, since rebinding takes
2087 * care of subsystems' refcounts, which are explicitly dropped in
2088 * the failure exit path.
2090 list_add(&root
->root_list
, &cgroup_roots
);
2091 cgroup_root_count
++;
2094 * Link the root cgroup in this hierarchy into all the css_set
2097 spin_lock_irq(&css_set_lock
);
2098 hash_for_each(css_set_table
, i
, cset
, hlist
) {
2099 link_css_set(&tmp_links
, cset
, root_cgrp
);
2100 if (css_set_populated(cset
))
2101 cgroup_update_populated(root_cgrp
, true);
2103 spin_unlock_irq(&css_set_lock
);
2105 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2106 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2112 cgroup_rstat_exit(root_cgrp
);
2114 kernfs_destroy_root(root
->kf_root
);
2115 root
->kf_root
= NULL
;
2117 cgroup_exit_root_id(root
);
2119 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2121 free_cgrp_cset_links(&tmp_links
);
2125 int cgroup_do_get_tree(struct fs_context
*fc
)
2127 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2130 ctx
->kfc
.root
= ctx
->root
->kf_root
;
2131 if (fc
->fs_type
== &cgroup2_fs_type
)
2132 ctx
->kfc
.magic
= CGROUP2_SUPER_MAGIC
;
2134 ctx
->kfc
.magic
= CGROUP_SUPER_MAGIC
;
2135 ret
= kernfs_get_tree(fc
);
2138 * In non-init cgroup namespace, instead of root cgroup's dentry,
2139 * we return the dentry corresponding to the cgroupns->root_cgrp.
2141 if (!ret
&& ctx
->ns
!= &init_cgroup_ns
) {
2142 struct dentry
*nsdentry
;
2143 struct super_block
*sb
= fc
->root
->d_sb
;
2144 struct cgroup
*cgrp
;
2146 mutex_lock(&cgroup_mutex
);
2147 spin_lock_irq(&css_set_lock
);
2149 cgrp
= cset_cgroup_from_root(ctx
->ns
->root_cset
, ctx
->root
);
2151 spin_unlock_irq(&css_set_lock
);
2152 mutex_unlock(&cgroup_mutex
);
2154 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
2156 if (IS_ERR(nsdentry
)) {
2157 deactivate_locked_super(sb
);
2158 ret
= PTR_ERR(nsdentry
);
2161 fc
->root
= nsdentry
;
2164 if (!ctx
->kfc
.new_sb_created
)
2165 cgroup_put(&ctx
->root
->cgrp
);
2171 * Destroy a cgroup filesystem context.
2173 static void cgroup_fs_context_free(struct fs_context
*fc
)
2175 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2178 kfree(ctx
->release_agent
);
2179 put_cgroup_ns(ctx
->ns
);
2180 kernfs_free_fs_context(fc
);
2184 static int cgroup_get_tree(struct fs_context
*fc
)
2186 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2189 WRITE_ONCE(cgrp_dfl_visible
, true);
2190 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2191 ctx
->root
= &cgrp_dfl_root
;
2193 ret
= cgroup_do_get_tree(fc
);
2195 apply_cgroup_root_flags(ctx
->flags
);
2199 static const struct fs_context_operations cgroup_fs_context_ops
= {
2200 .free
= cgroup_fs_context_free
,
2201 .parse_param
= cgroup2_parse_param
,
2202 .get_tree
= cgroup_get_tree
,
2203 .reconfigure
= cgroup_reconfigure
,
2206 static const struct fs_context_operations cgroup1_fs_context_ops
= {
2207 .free
= cgroup_fs_context_free
,
2208 .parse_param
= cgroup1_parse_param
,
2209 .get_tree
= cgroup1_get_tree
,
2210 .reconfigure
= cgroup1_reconfigure
,
2214 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2215 * we select the namespace we're going to use.
2217 static int cgroup_init_fs_context(struct fs_context
*fc
)
2219 struct cgroup_fs_context
*ctx
;
2221 ctx
= kzalloc(sizeof(struct cgroup_fs_context
), GFP_KERNEL
);
2225 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
2226 get_cgroup_ns(ctx
->ns
);
2227 fc
->fs_private
= &ctx
->kfc
;
2228 if (fc
->fs_type
== &cgroup2_fs_type
)
2229 fc
->ops
= &cgroup_fs_context_ops
;
2231 fc
->ops
= &cgroup1_fs_context_ops
;
2232 put_user_ns(fc
->user_ns
);
2233 fc
->user_ns
= get_user_ns(ctx
->ns
->user_ns
);
2236 #ifdef CONFIG_CGROUP_FAVOR_DYNMODS
2237 ctx
->flags
|= CGRP_ROOT_FAVOR_DYNMODS
;
2242 static void cgroup_kill_sb(struct super_block
*sb
)
2244 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2245 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2248 * If @root doesn't have any children, start killing it.
2249 * This prevents new mounts by disabling percpu_ref_tryget_live().
2251 * And don't kill the default root.
2253 if (list_empty(&root
->cgrp
.self
.children
) && root
!= &cgrp_dfl_root
&&
2254 !percpu_ref_is_dying(&root
->cgrp
.self
.refcnt
)) {
2255 cgroup_bpf_offline(&root
->cgrp
);
2256 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2258 cgroup_put(&root
->cgrp
);
2262 struct file_system_type cgroup_fs_type
= {
2264 .init_fs_context
= cgroup_init_fs_context
,
2265 .parameters
= cgroup1_fs_parameters
,
2266 .kill_sb
= cgroup_kill_sb
,
2267 .fs_flags
= FS_USERNS_MOUNT
,
2270 static struct file_system_type cgroup2_fs_type
= {
2272 .init_fs_context
= cgroup_init_fs_context
,
2273 .parameters
= cgroup2_fs_parameters
,
2274 .kill_sb
= cgroup_kill_sb
,
2275 .fs_flags
= FS_USERNS_MOUNT
,
2278 #ifdef CONFIG_CPUSETS
2279 static const struct fs_context_operations cpuset_fs_context_ops
= {
2280 .get_tree
= cgroup1_get_tree
,
2281 .free
= cgroup_fs_context_free
,
2285 * This is ugly, but preserves the userspace API for existing cpuset
2286 * users. If someone tries to mount the "cpuset" filesystem, we
2287 * silently switch it to mount "cgroup" instead
2289 static int cpuset_init_fs_context(struct fs_context
*fc
)
2291 char *agent
= kstrdup("/sbin/cpuset_release_agent", GFP_USER
);
2292 struct cgroup_fs_context
*ctx
;
2295 err
= cgroup_init_fs_context(fc
);
2301 fc
->ops
= &cpuset_fs_context_ops
;
2303 ctx
= cgroup_fc2context(fc
);
2304 ctx
->subsys_mask
= 1 << cpuset_cgrp_id
;
2305 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
2306 ctx
->release_agent
= agent
;
2308 get_filesystem(&cgroup_fs_type
);
2309 put_filesystem(fc
->fs_type
);
2310 fc
->fs_type
= &cgroup_fs_type
;
2315 static struct file_system_type cpuset_fs_type
= {
2317 .init_fs_context
= cpuset_init_fs_context
,
2318 .fs_flags
= FS_USERNS_MOUNT
,
2322 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2323 struct cgroup_namespace
*ns
)
2325 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2327 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2330 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2331 struct cgroup_namespace
*ns
)
2335 mutex_lock(&cgroup_mutex
);
2336 spin_lock_irq(&css_set_lock
);
2338 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2340 spin_unlock_irq(&css_set_lock
);
2341 mutex_unlock(&cgroup_mutex
);
2345 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2348 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2349 * @task: target task
2350 * @buf: the buffer to write the path into
2351 * @buflen: the length of the buffer
2353 * Determine @task's cgroup on the first (the one with the lowest non-zero
2354 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2355 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2356 * cgroup controller callbacks.
2358 * Return value is the same as kernfs_path().
2360 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2362 struct cgroup_root
*root
;
2363 struct cgroup
*cgrp
;
2364 int hierarchy_id
= 1;
2367 mutex_lock(&cgroup_mutex
);
2368 spin_lock_irq(&css_set_lock
);
2370 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2373 cgrp
= task_cgroup_from_root(task
, root
);
2374 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2376 /* if no hierarchy exists, everyone is in "/" */
2377 ret
= strscpy(buf
, "/", buflen
);
2380 spin_unlock_irq(&css_set_lock
);
2381 mutex_unlock(&cgroup_mutex
);
2384 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2387 * cgroup_attach_lock - Lock for ->attach()
2388 * @lock_threadgroup: whether to down_write cgroup_threadgroup_rwsem
2390 * cgroup migration sometimes needs to stabilize threadgroups against forks and
2391 * exits by write-locking cgroup_threadgroup_rwsem. However, some ->attach()
2392 * implementations (e.g. cpuset), also need to disable CPU hotplug.
2393 * Unfortunately, letting ->attach() operations acquire cpus_read_lock() can
2394 * lead to deadlocks.
2396 * Bringing up a CPU may involve creating and destroying tasks which requires
2397 * read-locking threadgroup_rwsem, so threadgroup_rwsem nests inside
2398 * cpus_read_lock(). If we call an ->attach() which acquires the cpus lock while
2399 * write-locking threadgroup_rwsem, the locking order is reversed and we end up
2400 * waiting for an on-going CPU hotplug operation which in turn is waiting for
2401 * the threadgroup_rwsem to be released to create new tasks. For more details:
2403 * http://lkml.kernel.org/r/20220711174629.uehfmqegcwn2lqzu@wubuntu
2405 * Resolve the situation by always acquiring cpus_read_lock() before optionally
2406 * write-locking cgroup_threadgroup_rwsem. This allows ->attach() to assume that
2407 * CPU hotplug is disabled on entry.
2409 void cgroup_attach_lock(bool lock_threadgroup
)
2412 if (lock_threadgroup
)
2413 percpu_down_write(&cgroup_threadgroup_rwsem
);
2417 * cgroup_attach_unlock - Undo cgroup_attach_lock()
2418 * @lock_threadgroup: whether to up_write cgroup_threadgroup_rwsem
2420 void cgroup_attach_unlock(bool lock_threadgroup
)
2422 if (lock_threadgroup
)
2423 percpu_up_write(&cgroup_threadgroup_rwsem
);
2428 * cgroup_migrate_add_task - add a migration target task to a migration context
2429 * @task: target task
2430 * @mgctx: target migration context
2432 * Add @task, which is a migration target, to @mgctx->tset. This function
2433 * becomes noop if @task doesn't need to be migrated. @task's css_set
2434 * should have been added as a migration source and @task->cg_list will be
2435 * moved from the css_set's tasks list to mg_tasks one.
2437 static void cgroup_migrate_add_task(struct task_struct
*task
,
2438 struct cgroup_mgctx
*mgctx
)
2440 struct css_set
*cset
;
2442 lockdep_assert_held(&css_set_lock
);
2444 /* @task either already exited or can't exit until the end */
2445 if (task
->flags
& PF_EXITING
)
2448 /* cgroup_threadgroup_rwsem protects racing against forks */
2449 WARN_ON_ONCE(list_empty(&task
->cg_list
));
2451 cset
= task_css_set(task
);
2452 if (!cset
->mg_src_cgrp
)
2455 mgctx
->tset
.nr_tasks
++;
2457 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2458 if (list_empty(&cset
->mg_node
))
2459 list_add_tail(&cset
->mg_node
,
2460 &mgctx
->tset
.src_csets
);
2461 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2462 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2463 &mgctx
->tset
.dst_csets
);
2467 * cgroup_taskset_first - reset taskset and return the first task
2468 * @tset: taskset of interest
2469 * @dst_cssp: output variable for the destination css
2471 * @tset iteration is initialized and the first task is returned.
2473 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2474 struct cgroup_subsys_state
**dst_cssp
)
2476 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2477 tset
->cur_task
= NULL
;
2479 return cgroup_taskset_next(tset
, dst_cssp
);
2483 * cgroup_taskset_next - iterate to the next task in taskset
2484 * @tset: taskset of interest
2485 * @dst_cssp: output variable for the destination css
2487 * Return the next task in @tset. Iteration must have been initialized
2488 * with cgroup_taskset_first().
2490 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2491 struct cgroup_subsys_state
**dst_cssp
)
2493 struct css_set
*cset
= tset
->cur_cset
;
2494 struct task_struct
*task
= tset
->cur_task
;
2496 while (CGROUP_HAS_SUBSYS_CONFIG
&& &cset
->mg_node
!= tset
->csets
) {
2498 task
= list_first_entry(&cset
->mg_tasks
,
2499 struct task_struct
, cg_list
);
2501 task
= list_next_entry(task
, cg_list
);
2503 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2504 tset
->cur_cset
= cset
;
2505 tset
->cur_task
= task
;
2508 * This function may be called both before and
2509 * after cgroup_taskset_migrate(). The two cases
2510 * can be distinguished by looking at whether @cset
2511 * has its ->mg_dst_cset set.
2513 if (cset
->mg_dst_cset
)
2514 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2516 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2521 cset
= list_next_entry(cset
, mg_node
);
2529 * cgroup_migrate_execute - migrate a taskset
2530 * @mgctx: migration context
2532 * Migrate tasks in @mgctx as setup by migration preparation functions.
2533 * This function fails iff one of the ->can_attach callbacks fails and
2534 * guarantees that either all or none of the tasks in @mgctx are migrated.
2535 * @mgctx is consumed regardless of success.
2537 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2539 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2540 struct cgroup_subsys
*ss
;
2541 struct task_struct
*task
, *tmp_task
;
2542 struct css_set
*cset
, *tmp_cset
;
2543 int ssid
, failed_ssid
, ret
;
2545 /* check that we can legitimately attach to the cgroup */
2546 if (tset
->nr_tasks
) {
2547 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2548 if (ss
->can_attach
) {
2550 ret
= ss
->can_attach(tset
);
2553 goto out_cancel_attach
;
2556 } while_each_subsys_mask();
2560 * Now that we're guaranteed success, proceed to move all tasks to
2561 * the new cgroup. There are no failure cases after here, so this
2562 * is the commit point.
2564 spin_lock_irq(&css_set_lock
);
2565 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2566 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2567 struct css_set
*from_cset
= task_css_set(task
);
2568 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2570 get_css_set(to_cset
);
2571 to_cset
->nr_tasks
++;
2572 css_set_move_task(task
, from_cset
, to_cset
, true);
2573 from_cset
->nr_tasks
--;
2575 * If the source or destination cgroup is frozen,
2576 * the task might require to change its state.
2578 cgroup_freezer_migrate_task(task
, from_cset
->dfl_cgrp
,
2580 put_css_set_locked(from_cset
);
2584 spin_unlock_irq(&css_set_lock
);
2587 * Migration is committed, all target tasks are now on dst_csets.
2588 * Nothing is sensitive to fork() after this point. Notify
2589 * controllers that migration is complete.
2591 tset
->csets
= &tset
->dst_csets
;
2593 if (tset
->nr_tasks
) {
2594 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2599 } while_each_subsys_mask();
2603 goto out_release_tset
;
2606 if (tset
->nr_tasks
) {
2607 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2608 if (ssid
== failed_ssid
)
2610 if (ss
->cancel_attach
) {
2612 ss
->cancel_attach(tset
);
2614 } while_each_subsys_mask();
2617 spin_lock_irq(&css_set_lock
);
2618 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2619 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2620 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2621 list_del_init(&cset
->mg_node
);
2623 spin_unlock_irq(&css_set_lock
);
2626 * Re-initialize the cgroup_taskset structure in case it is reused
2627 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2631 tset
->csets
= &tset
->src_csets
;
2636 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2637 * @dst_cgrp: destination cgroup to test
2639 * On the default hierarchy, except for the mixable, (possible) thread root
2640 * and threaded cgroups, subtree_control must be zero for migration
2641 * destination cgroups with tasks so that child cgroups don't compete
2644 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2646 /* v1 doesn't have any restriction */
2647 if (!cgroup_on_dfl(dst_cgrp
))
2650 /* verify @dst_cgrp can host resources */
2651 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2655 * If @dst_cgrp is already or can become a thread root or is
2656 * threaded, it doesn't matter.
2658 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2661 /* apply no-internal-process constraint */
2662 if (dst_cgrp
->subtree_control
)
2669 * cgroup_migrate_finish - cleanup after attach
2670 * @mgctx: migration context
2672 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2673 * those functions for details.
2675 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2677 struct css_set
*cset
, *tmp_cset
;
2679 lockdep_assert_held(&cgroup_mutex
);
2681 spin_lock_irq(&css_set_lock
);
2683 list_for_each_entry_safe(cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2684 mg_src_preload_node
) {
2685 cset
->mg_src_cgrp
= NULL
;
2686 cset
->mg_dst_cgrp
= NULL
;
2687 cset
->mg_dst_cset
= NULL
;
2688 list_del_init(&cset
->mg_src_preload_node
);
2689 put_css_set_locked(cset
);
2692 list_for_each_entry_safe(cset
, tmp_cset
, &mgctx
->preloaded_dst_csets
,
2693 mg_dst_preload_node
) {
2694 cset
->mg_src_cgrp
= NULL
;
2695 cset
->mg_dst_cgrp
= NULL
;
2696 cset
->mg_dst_cset
= NULL
;
2697 list_del_init(&cset
->mg_dst_preload_node
);
2698 put_css_set_locked(cset
);
2701 spin_unlock_irq(&css_set_lock
);
2705 * cgroup_migrate_add_src - add a migration source css_set
2706 * @src_cset: the source css_set to add
2707 * @dst_cgrp: the destination cgroup
2708 * @mgctx: migration context
2710 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2711 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2712 * up by cgroup_migrate_finish().
2714 * This function may be called without holding cgroup_threadgroup_rwsem
2715 * even if the target is a process. Threads may be created and destroyed
2716 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2717 * into play and the preloaded css_sets are guaranteed to cover all
2720 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2721 struct cgroup
*dst_cgrp
,
2722 struct cgroup_mgctx
*mgctx
)
2724 struct cgroup
*src_cgrp
;
2726 lockdep_assert_held(&cgroup_mutex
);
2727 lockdep_assert_held(&css_set_lock
);
2730 * If ->dead, @src_set is associated with one or more dead cgroups
2731 * and doesn't contain any migratable tasks. Ignore it early so
2732 * that the rest of migration path doesn't get confused by it.
2737 if (!list_empty(&src_cset
->mg_src_preload_node
))
2740 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2742 WARN_ON(src_cset
->mg_src_cgrp
);
2743 WARN_ON(src_cset
->mg_dst_cgrp
);
2744 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2745 WARN_ON(!list_empty(&src_cset
->mg_node
));
2747 src_cset
->mg_src_cgrp
= src_cgrp
;
2748 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2749 get_css_set(src_cset
);
2750 list_add_tail(&src_cset
->mg_src_preload_node
, &mgctx
->preloaded_src_csets
);
2754 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2755 * @mgctx: migration context
2757 * Tasks are about to be moved and all the source css_sets have been
2758 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2759 * pins all destination css_sets, links each to its source, and append them
2760 * to @mgctx->preloaded_dst_csets.
2762 * This function must be called after cgroup_migrate_add_src() has been
2763 * called on each migration source css_set. After migration is performed
2764 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2767 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2769 struct css_set
*src_cset
, *tmp_cset
;
2771 lockdep_assert_held(&cgroup_mutex
);
2773 /* look up the dst cset for each src cset and link it to src */
2774 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2775 mg_src_preload_node
) {
2776 struct css_set
*dst_cset
;
2777 struct cgroup_subsys
*ss
;
2780 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2784 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2787 * If src cset equals dst, it's noop. Drop the src.
2788 * cgroup_migrate() will skip the cset too. Note that we
2789 * can't handle src == dst as some nodes are used by both.
2791 if (src_cset
== dst_cset
) {
2792 src_cset
->mg_src_cgrp
= NULL
;
2793 src_cset
->mg_dst_cgrp
= NULL
;
2794 list_del_init(&src_cset
->mg_src_preload_node
);
2795 put_css_set(src_cset
);
2796 put_css_set(dst_cset
);
2800 src_cset
->mg_dst_cset
= dst_cset
;
2802 if (list_empty(&dst_cset
->mg_dst_preload_node
))
2803 list_add_tail(&dst_cset
->mg_dst_preload_node
,
2804 &mgctx
->preloaded_dst_csets
);
2806 put_css_set(dst_cset
);
2808 for_each_subsys(ss
, ssid
)
2809 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2810 mgctx
->ss_mask
|= 1 << ssid
;
2817 * cgroup_migrate - migrate a process or task to a cgroup
2818 * @leader: the leader of the process or the task to migrate
2819 * @threadgroup: whether @leader points to the whole process or a single task
2820 * @mgctx: migration context
2822 * Migrate a process or task denoted by @leader. If migrating a process,
2823 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2824 * responsible for invoking cgroup_migrate_add_src() and
2825 * cgroup_migrate_prepare_dst() on the targets before invoking this
2826 * function and following up with cgroup_migrate_finish().
2828 * As long as a controller's ->can_attach() doesn't fail, this function is
2829 * guaranteed to succeed. This means that, excluding ->can_attach()
2830 * failure, when migrating multiple targets, the success or failure can be
2831 * decided for all targets by invoking group_migrate_prepare_dst() before
2832 * actually starting migrating.
2834 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2835 struct cgroup_mgctx
*mgctx
)
2837 struct task_struct
*task
;
2840 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2841 * already PF_EXITING could be freed from underneath us unless we
2842 * take an rcu_read_lock.
2844 spin_lock_irq(&css_set_lock
);
2848 cgroup_migrate_add_task(task
, mgctx
);
2851 } while_each_thread(leader
, task
);
2853 spin_unlock_irq(&css_set_lock
);
2855 return cgroup_migrate_execute(mgctx
);
2859 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2860 * @dst_cgrp: the cgroup to attach to
2861 * @leader: the task or the leader of the threadgroup to be attached
2862 * @threadgroup: attach the whole threadgroup?
2864 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2866 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2869 DEFINE_CGROUP_MGCTX(mgctx
);
2870 struct task_struct
*task
;
2873 /* look up all src csets */
2874 spin_lock_irq(&css_set_lock
);
2878 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2881 } while_each_thread(leader
, task
);
2883 spin_unlock_irq(&css_set_lock
);
2885 /* prepare dst csets and commit */
2886 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2888 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2890 cgroup_migrate_finish(&mgctx
);
2893 TRACE_CGROUP_PATH(attach_task
, dst_cgrp
, leader
, threadgroup
);
2898 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
,
2899 bool *threadgroup_locked
)
2901 struct task_struct
*tsk
;
2904 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2905 return ERR_PTR(-EINVAL
);
2908 * If we migrate a single thread, we don't care about threadgroup
2909 * stability. If the thread is `current`, it won't exit(2) under our
2910 * hands or change PID through exec(2). We exclude
2911 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2912 * callers by cgroup_mutex.
2913 * Therefore, we can skip the global lock.
2915 lockdep_assert_held(&cgroup_mutex
);
2916 *threadgroup_locked
= pid
|| threadgroup
;
2917 cgroup_attach_lock(*threadgroup_locked
);
2921 tsk
= find_task_by_vpid(pid
);
2923 tsk
= ERR_PTR(-ESRCH
);
2924 goto out_unlock_threadgroup
;
2931 tsk
= tsk
->group_leader
;
2934 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2935 * If userland migrates such a kthread to a non-root cgroup, it can
2936 * become trapped in a cpuset, or RT kthread may be born in a
2937 * cgroup with no rt_runtime allocated. Just say no.
2939 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2940 tsk
= ERR_PTR(-EINVAL
);
2941 goto out_unlock_threadgroup
;
2944 get_task_struct(tsk
);
2945 goto out_unlock_rcu
;
2947 out_unlock_threadgroup
:
2948 cgroup_attach_unlock(*threadgroup_locked
);
2949 *threadgroup_locked
= false;
2955 void cgroup_procs_write_finish(struct task_struct
*task
, bool threadgroup_locked
)
2957 struct cgroup_subsys
*ss
;
2960 /* release reference from cgroup_procs_write_start() */
2961 put_task_struct(task
);
2963 cgroup_attach_unlock(threadgroup_locked
);
2965 for_each_subsys(ss
, ssid
)
2966 if (ss
->post_attach
)
2970 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2972 struct cgroup_subsys
*ss
;
2973 bool printed
= false;
2976 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2979 seq_puts(seq
, ss
->name
);
2981 } while_each_subsys_mask();
2983 seq_putc(seq
, '\n');
2986 /* show controllers which are enabled from the parent */
2987 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2989 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2991 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2995 /* show controllers which are enabled for a given cgroup's children */
2996 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2998 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3000 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
3005 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3006 * @cgrp: root of the subtree to update csses for
3008 * @cgrp's control masks have changed and its subtree's css associations
3009 * need to be updated accordingly. This function looks up all css_sets
3010 * which are attached to the subtree, creates the matching updated css_sets
3011 * and migrates the tasks to the new ones.
3013 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
3015 DEFINE_CGROUP_MGCTX(mgctx
);
3016 struct cgroup_subsys_state
*d_css
;
3017 struct cgroup
*dsct
;
3018 struct css_set
*src_cset
;
3022 lockdep_assert_held(&cgroup_mutex
);
3024 /* look up all csses currently attached to @cgrp's subtree */
3025 spin_lock_irq(&css_set_lock
);
3026 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3027 struct cgrp_cset_link
*link
;
3030 * As cgroup_update_dfl_csses() is only called by
3031 * cgroup_apply_control(). The csses associated with the
3032 * given cgrp will not be affected by changes made to
3033 * its subtree_control file. We can skip them.
3038 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
3039 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
3041 spin_unlock_irq(&css_set_lock
);
3044 * We need to write-lock threadgroup_rwsem while migrating tasks.
3045 * However, if there are no source csets for @cgrp, changing its
3046 * controllers isn't gonna produce any task migrations and the
3047 * write-locking can be skipped safely.
3049 has_tasks
= !list_empty(&mgctx
.preloaded_src_csets
);
3050 cgroup_attach_lock(has_tasks
);
3052 /* NULL dst indicates self on default hierarchy */
3053 ret
= cgroup_migrate_prepare_dst(&mgctx
);
3057 spin_lock_irq(&css_set_lock
);
3058 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
,
3059 mg_src_preload_node
) {
3060 struct task_struct
*task
, *ntask
;
3062 /* all tasks in src_csets need to be migrated */
3063 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
3064 cgroup_migrate_add_task(task
, &mgctx
);
3066 spin_unlock_irq(&css_set_lock
);
3068 ret
= cgroup_migrate_execute(&mgctx
);
3070 cgroup_migrate_finish(&mgctx
);
3071 cgroup_attach_unlock(has_tasks
);
3076 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3077 * @cgrp: root of the target subtree
3079 * Because css offlining is asynchronous, userland may try to re-enable a
3080 * controller while the previous css is still around. This function grabs
3081 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3083 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
3084 __acquires(&cgroup_mutex
)
3086 struct cgroup
*dsct
;
3087 struct cgroup_subsys_state
*d_css
;
3088 struct cgroup_subsys
*ss
;
3092 mutex_lock(&cgroup_mutex
);
3094 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3095 for_each_subsys(ss
, ssid
) {
3096 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3099 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
3102 cgroup_get_live(dsct
);
3103 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
3104 TASK_UNINTERRUPTIBLE
);
3106 mutex_unlock(&cgroup_mutex
);
3108 finish_wait(&dsct
->offline_waitq
, &wait
);
3117 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3118 * @cgrp: root of the target subtree
3120 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3121 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3124 static void cgroup_save_control(struct cgroup
*cgrp
)
3126 struct cgroup
*dsct
;
3127 struct cgroup_subsys_state
*d_css
;
3129 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3130 dsct
->old_subtree_control
= dsct
->subtree_control
;
3131 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
3132 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
3137 * cgroup_propagate_control - refresh control masks of a subtree
3138 * @cgrp: root of the target subtree
3140 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3141 * ->subtree_control and propagate controller availability through the
3142 * subtree so that descendants don't have unavailable controllers enabled.
3144 static void cgroup_propagate_control(struct cgroup
*cgrp
)
3146 struct cgroup
*dsct
;
3147 struct cgroup_subsys_state
*d_css
;
3149 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3150 dsct
->subtree_control
&= cgroup_control(dsct
);
3151 dsct
->subtree_ss_mask
=
3152 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
3153 cgroup_ss_mask(dsct
));
3158 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3159 * @cgrp: root of the target subtree
3161 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3162 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3165 static void cgroup_restore_control(struct cgroup
*cgrp
)
3167 struct cgroup
*dsct
;
3168 struct cgroup_subsys_state
*d_css
;
3170 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3171 dsct
->subtree_control
= dsct
->old_subtree_control
;
3172 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3173 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
3177 static bool css_visible(struct cgroup_subsys_state
*css
)
3179 struct cgroup_subsys
*ss
= css
->ss
;
3180 struct cgroup
*cgrp
= css
->cgroup
;
3182 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3184 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3186 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3190 * cgroup_apply_control_enable - enable or show csses according to control
3191 * @cgrp: root of the target subtree
3193 * Walk @cgrp's subtree and create new csses or make the existing ones
3194 * visible. A css is created invisible if it's being implicitly enabled
3195 * through dependency. An invisible css is made visible when the userland
3196 * explicitly enables it.
3198 * Returns 0 on success, -errno on failure. On failure, csses which have
3199 * been processed already aren't cleaned up. The caller is responsible for
3200 * cleaning up with cgroup_apply_control_disable().
3202 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3204 struct cgroup
*dsct
;
3205 struct cgroup_subsys_state
*d_css
;
3206 struct cgroup_subsys
*ss
;
3209 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3210 for_each_subsys(ss
, ssid
) {
3211 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3213 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3217 css
= css_create(dsct
, ss
);
3219 return PTR_ERR(css
);
3222 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3224 if (css_visible(css
)) {
3225 ret
= css_populate_dir(css
);
3236 * cgroup_apply_control_disable - kill or hide csses according to control
3237 * @cgrp: root of the target subtree
3239 * Walk @cgrp's subtree and kill and hide csses so that they match
3240 * cgroup_ss_mask() and cgroup_visible_mask().
3242 * A css is hidden when the userland requests it to be disabled while other
3243 * subsystems are still depending on it. The css must not actively control
3244 * resources and be in the vanilla state if it's made visible again later.
3245 * Controllers which may be depended upon should provide ->css_reset() for
3248 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3250 struct cgroup
*dsct
;
3251 struct cgroup_subsys_state
*d_css
;
3252 struct cgroup_subsys
*ss
;
3255 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3256 for_each_subsys(ss
, ssid
) {
3257 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3262 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3265 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3267 } else if (!css_visible(css
)) {
3277 * cgroup_apply_control - apply control mask updates to the subtree
3278 * @cgrp: root of the target subtree
3280 * subsystems can be enabled and disabled in a subtree using the following
3283 * 1. Call cgroup_save_control() to stash the current state.
3284 * 2. Update ->subtree_control masks in the subtree as desired.
3285 * 3. Call cgroup_apply_control() to apply the changes.
3286 * 4. Optionally perform other related operations.
3287 * 5. Call cgroup_finalize_control() to finish up.
3289 * This function implements step 3 and propagates the mask changes
3290 * throughout @cgrp's subtree, updates csses accordingly and perform
3291 * process migrations.
3293 static int cgroup_apply_control(struct cgroup
*cgrp
)
3297 cgroup_propagate_control(cgrp
);
3299 ret
= cgroup_apply_control_enable(cgrp
);
3304 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3305 * making the following cgroup_update_dfl_csses() properly update
3306 * css associations of all tasks in the subtree.
3308 return cgroup_update_dfl_csses(cgrp
);
3312 * cgroup_finalize_control - finalize control mask update
3313 * @cgrp: root of the target subtree
3314 * @ret: the result of the update
3316 * Finalize control mask update. See cgroup_apply_control() for more info.
3318 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3321 cgroup_restore_control(cgrp
);
3322 cgroup_propagate_control(cgrp
);
3325 cgroup_apply_control_disable(cgrp
);
3328 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3330 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3332 /* if nothing is getting enabled, nothing to worry about */
3336 /* can @cgrp host any resources? */
3337 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3340 /* mixables don't care */
3341 if (cgroup_is_mixable(cgrp
))
3344 if (domain_enable
) {
3345 /* can't enable domain controllers inside a thread subtree */
3346 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3350 * Threaded controllers can handle internal competitions
3351 * and are always allowed inside a (prospective) thread
3354 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3359 * Controllers can't be enabled for a cgroup with tasks to avoid
3360 * child cgroups competing against tasks.
3362 if (cgroup_has_tasks(cgrp
))
3368 /* change the enabled child controllers for a cgroup in the default hierarchy */
3369 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3370 char *buf
, size_t nbytes
,
3373 u16 enable
= 0, disable
= 0;
3374 struct cgroup
*cgrp
, *child
;
3375 struct cgroup_subsys
*ss
;
3380 * Parse input - space separated list of subsystem names prefixed
3381 * with either + or -.
3383 buf
= strstrip(buf
);
3384 while ((tok
= strsep(&buf
, " "))) {
3387 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3388 if (!cgroup_ssid_enabled(ssid
) ||
3389 strcmp(tok
+ 1, ss
->name
))
3393 enable
|= 1 << ssid
;
3394 disable
&= ~(1 << ssid
);
3395 } else if (*tok
== '-') {
3396 disable
|= 1 << ssid
;
3397 enable
&= ~(1 << ssid
);
3402 } while_each_subsys_mask();
3403 if (ssid
== CGROUP_SUBSYS_COUNT
)
3407 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3411 for_each_subsys(ss
, ssid
) {
3412 if (enable
& (1 << ssid
)) {
3413 if (cgrp
->subtree_control
& (1 << ssid
)) {
3414 enable
&= ~(1 << ssid
);
3418 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3422 } else if (disable
& (1 << ssid
)) {
3423 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3424 disable
&= ~(1 << ssid
);
3428 /* a child has it enabled? */
3429 cgroup_for_each_live_child(child
, cgrp
) {
3430 if (child
->subtree_control
& (1 << ssid
)) {
3438 if (!enable
&& !disable
) {
3443 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3447 /* save and update control masks and prepare csses */
3448 cgroup_save_control(cgrp
);
3450 cgrp
->subtree_control
|= enable
;
3451 cgrp
->subtree_control
&= ~disable
;
3453 ret
= cgroup_apply_control(cgrp
);
3454 cgroup_finalize_control(cgrp
, ret
);
3458 kernfs_activate(cgrp
->kn
);
3460 cgroup_kn_unlock(of
->kn
);
3461 return ret
?: nbytes
;
3465 * cgroup_enable_threaded - make @cgrp threaded
3466 * @cgrp: the target cgroup
3468 * Called when "threaded" is written to the cgroup.type interface file and
3469 * tries to make @cgrp threaded and join the parent's resource domain.
3470 * This function is never called on the root cgroup as cgroup.type doesn't
3473 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3475 struct cgroup
*parent
= cgroup_parent(cgrp
);
3476 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3477 struct cgroup
*dsct
;
3478 struct cgroup_subsys_state
*d_css
;
3481 lockdep_assert_held(&cgroup_mutex
);
3483 /* noop if already threaded */
3484 if (cgroup_is_threaded(cgrp
))
3488 * If @cgroup is populated or has domain controllers enabled, it
3489 * can't be switched. While the below cgroup_can_be_thread_root()
3490 * test can catch the same conditions, that's only when @parent is
3491 * not mixable, so let's check it explicitly.
3493 if (cgroup_is_populated(cgrp
) ||
3494 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3497 /* we're joining the parent's domain, ensure its validity */
3498 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3499 !cgroup_can_be_thread_root(dom_cgrp
))
3503 * The following shouldn't cause actual migrations and should
3506 cgroup_save_control(cgrp
);
3508 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3509 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3510 dsct
->dom_cgrp
= dom_cgrp
;
3512 ret
= cgroup_apply_control(cgrp
);
3514 parent
->nr_threaded_children
++;
3516 cgroup_finalize_control(cgrp
, ret
);
3520 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3522 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3524 if (cgroup_is_threaded(cgrp
))
3525 seq_puts(seq
, "threaded\n");
3526 else if (!cgroup_is_valid_domain(cgrp
))
3527 seq_puts(seq
, "domain invalid\n");
3528 else if (cgroup_is_thread_root(cgrp
))
3529 seq_puts(seq
, "domain threaded\n");
3531 seq_puts(seq
, "domain\n");
3536 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3537 size_t nbytes
, loff_t off
)
3539 struct cgroup
*cgrp
;
3542 /* only switching to threaded mode is supported */
3543 if (strcmp(strstrip(buf
), "threaded"))
3546 /* drain dying csses before we re-apply (threaded) subtree control */
3547 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3551 /* threaded can only be enabled */
3552 ret
= cgroup_enable_threaded(cgrp
);
3554 cgroup_kn_unlock(of
->kn
);
3555 return ret
?: nbytes
;
3558 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3560 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3561 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3563 if (descendants
== INT_MAX
)
3564 seq_puts(seq
, "max\n");
3566 seq_printf(seq
, "%d\n", descendants
);
3571 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3572 char *buf
, size_t nbytes
, loff_t off
)
3574 struct cgroup
*cgrp
;
3578 buf
= strstrip(buf
);
3579 if (!strcmp(buf
, "max")) {
3580 descendants
= INT_MAX
;
3582 ret
= kstrtoint(buf
, 0, &descendants
);
3587 if (descendants
< 0)
3590 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3594 cgrp
->max_descendants
= descendants
;
3596 cgroup_kn_unlock(of
->kn
);
3601 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3603 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3604 int depth
= READ_ONCE(cgrp
->max_depth
);
3606 if (depth
== INT_MAX
)
3607 seq_puts(seq
, "max\n");
3609 seq_printf(seq
, "%d\n", depth
);
3614 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3615 char *buf
, size_t nbytes
, loff_t off
)
3617 struct cgroup
*cgrp
;
3621 buf
= strstrip(buf
);
3622 if (!strcmp(buf
, "max")) {
3625 ret
= kstrtoint(buf
, 0, &depth
);
3633 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3637 cgrp
->max_depth
= depth
;
3639 cgroup_kn_unlock(of
->kn
);
3644 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3646 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3648 seq_printf(seq
, "populated %d\n", cgroup_is_populated(cgrp
));
3649 seq_printf(seq
, "frozen %d\n", test_bit(CGRP_FROZEN
, &cgrp
->flags
));
3654 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3656 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3658 seq_printf(seq
, "nr_descendants %d\n",
3659 cgroup
->nr_descendants
);
3660 seq_printf(seq
, "nr_dying_descendants %d\n",
3661 cgroup
->nr_dying_descendants
);
3666 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3667 struct cgroup
*cgrp
, int ssid
)
3669 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3670 struct cgroup_subsys_state
*css
;
3673 if (!ss
->css_extra_stat_show
)
3676 css
= cgroup_tryget_css(cgrp
, ss
);
3680 ret
= ss
->css_extra_stat_show(seq
, css
);
3685 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3687 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3690 cgroup_base_stat_cputime_show(seq
);
3691 #ifdef CONFIG_CGROUP_SCHED
3692 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3698 static int cgroup_io_pressure_show(struct seq_file
*seq
, void *v
)
3700 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3701 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: cgrp
->psi
;
3703 return psi_show(seq
, psi
, PSI_IO
);
3705 static int cgroup_memory_pressure_show(struct seq_file
*seq
, void *v
)
3707 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3708 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: cgrp
->psi
;
3710 return psi_show(seq
, psi
, PSI_MEM
);
3712 static int cgroup_cpu_pressure_show(struct seq_file
*seq
, void *v
)
3714 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3715 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: cgrp
->psi
;
3717 return psi_show(seq
, psi
, PSI_CPU
);
3720 static ssize_t
cgroup_pressure_write(struct kernfs_open_file
*of
, char *buf
,
3721 size_t nbytes
, enum psi_res res
)
3723 struct cgroup_file_ctx
*ctx
= of
->priv
;
3724 struct psi_trigger
*new;
3725 struct cgroup
*cgrp
;
3726 struct psi_group
*psi
;
3728 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3733 cgroup_kn_unlock(of
->kn
);
3735 /* Allow only one trigger per file descriptor */
3736 if (ctx
->psi
.trigger
) {
3741 psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: cgrp
->psi
;
3742 new = psi_trigger_create(psi
, buf
, res
);
3745 return PTR_ERR(new);
3748 smp_store_release(&ctx
->psi
.trigger
, new);
3754 static ssize_t
cgroup_io_pressure_write(struct kernfs_open_file
*of
,
3755 char *buf
, size_t nbytes
,
3758 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_IO
);
3761 static ssize_t
cgroup_memory_pressure_write(struct kernfs_open_file
*of
,
3762 char *buf
, size_t nbytes
,
3765 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_MEM
);
3768 static ssize_t
cgroup_cpu_pressure_write(struct kernfs_open_file
*of
,
3769 char *buf
, size_t nbytes
,
3772 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_CPU
);
3775 static __poll_t
cgroup_pressure_poll(struct kernfs_open_file
*of
,
3778 struct cgroup_file_ctx
*ctx
= of
->priv
;
3780 return psi_trigger_poll(&ctx
->psi
.trigger
, of
->file
, pt
);
3783 static void cgroup_pressure_release(struct kernfs_open_file
*of
)
3785 struct cgroup_file_ctx
*ctx
= of
->priv
;
3787 psi_trigger_destroy(ctx
->psi
.trigger
);
3790 bool cgroup_psi_enabled(void)
3792 return (cgroup_feature_disable_mask
& (1 << OPT_FEATURE_PRESSURE
)) == 0;
3795 #else /* CONFIG_PSI */
3796 bool cgroup_psi_enabled(void)
3801 #endif /* CONFIG_PSI */
3803 static int cgroup_freeze_show(struct seq_file
*seq
, void *v
)
3805 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3807 seq_printf(seq
, "%d\n", cgrp
->freezer
.freeze
);
3812 static ssize_t
cgroup_freeze_write(struct kernfs_open_file
*of
,
3813 char *buf
, size_t nbytes
, loff_t off
)
3815 struct cgroup
*cgrp
;
3819 ret
= kstrtoint(strstrip(buf
), 0, &freeze
);
3823 if (freeze
< 0 || freeze
> 1)
3826 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3830 cgroup_freeze(cgrp
, freeze
);
3832 cgroup_kn_unlock(of
->kn
);
3837 static void __cgroup_kill(struct cgroup
*cgrp
)
3839 struct css_task_iter it
;
3840 struct task_struct
*task
;
3842 lockdep_assert_held(&cgroup_mutex
);
3844 spin_lock_irq(&css_set_lock
);
3845 set_bit(CGRP_KILL
, &cgrp
->flags
);
3846 spin_unlock_irq(&css_set_lock
);
3848 css_task_iter_start(&cgrp
->self
, CSS_TASK_ITER_PROCS
| CSS_TASK_ITER_THREADED
, &it
);
3849 while ((task
= css_task_iter_next(&it
))) {
3850 /* Ignore kernel threads here. */
3851 if (task
->flags
& PF_KTHREAD
)
3854 /* Skip tasks that are already dying. */
3855 if (__fatal_signal_pending(task
))
3858 send_sig(SIGKILL
, task
, 0);
3860 css_task_iter_end(&it
);
3862 spin_lock_irq(&css_set_lock
);
3863 clear_bit(CGRP_KILL
, &cgrp
->flags
);
3864 spin_unlock_irq(&css_set_lock
);
3867 static void cgroup_kill(struct cgroup
*cgrp
)
3869 struct cgroup_subsys_state
*css
;
3870 struct cgroup
*dsct
;
3872 lockdep_assert_held(&cgroup_mutex
);
3874 cgroup_for_each_live_descendant_pre(dsct
, css
, cgrp
)
3875 __cgroup_kill(dsct
);
3878 static ssize_t
cgroup_kill_write(struct kernfs_open_file
*of
, char *buf
,
3879 size_t nbytes
, loff_t off
)
3883 struct cgroup
*cgrp
;
3885 ret
= kstrtoint(strstrip(buf
), 0, &kill
);
3892 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3897 * Killing is a process directed operation, i.e. the whole thread-group
3898 * is taken down so act like we do for cgroup.procs and only make this
3899 * writable in non-threaded cgroups.
3901 if (cgroup_is_threaded(cgrp
))
3906 cgroup_kn_unlock(of
->kn
);
3908 return ret
?: nbytes
;
3911 static int cgroup_file_open(struct kernfs_open_file
*of
)
3913 struct cftype
*cft
= of_cft(of
);
3914 struct cgroup_file_ctx
*ctx
;
3917 ctx
= kzalloc(sizeof(*ctx
), GFP_KERNEL
);
3921 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
3922 get_cgroup_ns(ctx
->ns
);
3928 ret
= cft
->open(of
);
3930 put_cgroup_ns(ctx
->ns
);
3936 static void cgroup_file_release(struct kernfs_open_file
*of
)
3938 struct cftype
*cft
= of_cft(of
);
3939 struct cgroup_file_ctx
*ctx
= of
->priv
;
3943 put_cgroup_ns(ctx
->ns
);
3947 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3948 size_t nbytes
, loff_t off
)
3950 struct cgroup_file_ctx
*ctx
= of
->priv
;
3951 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3952 struct cftype
*cft
= of_cft(of
);
3953 struct cgroup_subsys_state
*css
;
3960 * If namespaces are delegation boundaries, disallow writes to
3961 * files in an non-init namespace root from inside the namespace
3962 * except for the files explicitly marked delegatable -
3963 * cgroup.procs and cgroup.subtree_control.
3965 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3966 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3967 ctx
->ns
!= &init_cgroup_ns
&& ctx
->ns
->root_cset
->dfl_cgrp
== cgrp
)
3971 return cft
->write(of
, buf
, nbytes
, off
);
3974 * kernfs guarantees that a file isn't deleted with operations in
3975 * flight, which means that the matching css is and stays alive and
3976 * doesn't need to be pinned. The RCU locking is not necessary
3977 * either. It's just for the convenience of using cgroup_css().
3980 css
= cgroup_css(cgrp
, cft
->ss
);
3983 if (cft
->write_u64
) {
3984 unsigned long long v
;
3985 ret
= kstrtoull(buf
, 0, &v
);
3987 ret
= cft
->write_u64(css
, cft
, v
);
3988 } else if (cft
->write_s64
) {
3990 ret
= kstrtoll(buf
, 0, &v
);
3992 ret
= cft
->write_s64(css
, cft
, v
);
3997 return ret
?: nbytes
;
4000 static __poll_t
cgroup_file_poll(struct kernfs_open_file
*of
, poll_table
*pt
)
4002 struct cftype
*cft
= of_cft(of
);
4005 return cft
->poll(of
, pt
);
4007 return kernfs_generic_poll(of
, pt
);
4010 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
4012 return seq_cft(seq
)->seq_start(seq
, ppos
);
4015 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
4017 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
4020 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
4022 if (seq_cft(seq
)->seq_stop
)
4023 seq_cft(seq
)->seq_stop(seq
, v
);
4026 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
4028 struct cftype
*cft
= seq_cft(m
);
4029 struct cgroup_subsys_state
*css
= seq_css(m
);
4032 return cft
->seq_show(m
, arg
);
4035 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
4036 else if (cft
->read_s64
)
4037 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
4043 static struct kernfs_ops cgroup_kf_single_ops
= {
4044 .atomic_write_len
= PAGE_SIZE
,
4045 .open
= cgroup_file_open
,
4046 .release
= cgroup_file_release
,
4047 .write
= cgroup_file_write
,
4048 .poll
= cgroup_file_poll
,
4049 .seq_show
= cgroup_seqfile_show
,
4052 static struct kernfs_ops cgroup_kf_ops
= {
4053 .atomic_write_len
= PAGE_SIZE
,
4054 .open
= cgroup_file_open
,
4055 .release
= cgroup_file_release
,
4056 .write
= cgroup_file_write
,
4057 .poll
= cgroup_file_poll
,
4058 .seq_start
= cgroup_seqfile_start
,
4059 .seq_next
= cgroup_seqfile_next
,
4060 .seq_stop
= cgroup_seqfile_stop
,
4061 .seq_show
= cgroup_seqfile_show
,
4064 /* set uid and gid of cgroup dirs and files to that of the creator */
4065 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
4067 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
4068 .ia_uid
= current_fsuid(),
4069 .ia_gid
= current_fsgid(), };
4071 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
4072 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
4075 return kernfs_setattr(kn
, &iattr
);
4078 static void cgroup_file_notify_timer(struct timer_list
*timer
)
4080 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
4084 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
4087 char name
[CGROUP_FILE_NAME_MAX
];
4088 struct kernfs_node
*kn
;
4089 struct lock_class_key
*key
= NULL
;
4092 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4093 key
= &cft
->lockdep_key
;
4095 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
4096 cgroup_file_mode(cft
),
4097 GLOBAL_ROOT_UID
, GLOBAL_ROOT_GID
,
4098 0, cft
->kf_ops
, cft
,
4103 ret
= cgroup_kn_set_ugid(kn
);
4109 if (cft
->file_offset
) {
4110 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
4112 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
4114 spin_lock_irq(&cgroup_file_kn_lock
);
4116 spin_unlock_irq(&cgroup_file_kn_lock
);
4123 * cgroup_addrm_files - add or remove files to a cgroup directory
4124 * @css: the target css
4125 * @cgrp: the target cgroup (usually css->cgroup)
4126 * @cfts: array of cftypes to be added
4127 * @is_add: whether to add or remove
4129 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
4130 * For removals, this function never fails.
4132 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
4133 struct cgroup
*cgrp
, struct cftype cfts
[],
4136 struct cftype
*cft
, *cft_end
= NULL
;
4139 lockdep_assert_held(&cgroup_mutex
);
4142 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
4143 /* does cft->flags tell us to skip this file on @cgrp? */
4144 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
4146 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
4148 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
4150 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
4152 if ((cft
->flags
& CFTYPE_DEBUG
) && !cgroup_debug
)
4155 ret
= cgroup_add_file(css
, cgrp
, cft
);
4157 pr_warn("%s: failed to add %s, err=%d\n",
4158 __func__
, cft
->name
, ret
);
4164 cgroup_rm_file(cgrp
, cft
);
4170 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
4172 struct cgroup_subsys
*ss
= cfts
[0].ss
;
4173 struct cgroup
*root
= &ss
->root
->cgrp
;
4174 struct cgroup_subsys_state
*css
;
4177 lockdep_assert_held(&cgroup_mutex
);
4179 /* add/rm files for all cgroups created before */
4180 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
4181 struct cgroup
*cgrp
= css
->cgroup
;
4183 if (!(css
->flags
& CSS_VISIBLE
))
4186 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
4192 kernfs_activate(root
->kn
);
4196 static void cgroup_exit_cftypes(struct cftype
*cfts
)
4200 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4201 /* free copy for custom atomic_write_len, see init_cftypes() */
4202 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
4207 /* revert flags set by cgroup core while adding @cfts */
4208 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
|
4213 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4218 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4219 struct kernfs_ops
*kf_ops
;
4221 WARN_ON(cft
->ss
|| cft
->kf_ops
);
4223 if (cft
->flags
& __CFTYPE_ADDED
) {
4229 kf_ops
= &cgroup_kf_ops
;
4231 kf_ops
= &cgroup_kf_single_ops
;
4234 * Ugh... if @cft wants a custom max_write_len, we need to
4235 * make a copy of kf_ops to set its atomic_write_len.
4237 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
4238 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
4243 kf_ops
->atomic_write_len
= cft
->max_write_len
;
4246 cft
->kf_ops
= kf_ops
;
4248 cft
->flags
|= __CFTYPE_ADDED
;
4252 cgroup_exit_cftypes(cfts
);
4256 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
4258 lockdep_assert_held(&cgroup_mutex
);
4260 list_del(&cfts
->node
);
4261 cgroup_apply_cftypes(cfts
, false);
4262 cgroup_exit_cftypes(cfts
);
4267 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4268 * @cfts: zero-length name terminated array of cftypes
4270 * Unregister @cfts. Files described by @cfts are removed from all
4271 * existing cgroups and all future cgroups won't have them either. This
4272 * function can be called anytime whether @cfts' subsys is attached or not.
4274 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4277 int cgroup_rm_cftypes(struct cftype
*cfts
)
4281 if (!cfts
|| cfts
[0].name
[0] == '\0')
4284 if (!(cfts
[0].flags
& __CFTYPE_ADDED
))
4287 mutex_lock(&cgroup_mutex
);
4288 ret
= cgroup_rm_cftypes_locked(cfts
);
4289 mutex_unlock(&cgroup_mutex
);
4294 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4295 * @ss: target cgroup subsystem
4296 * @cfts: zero-length name terminated array of cftypes
4298 * Register @cfts to @ss. Files described by @cfts are created for all
4299 * existing cgroups to which @ss is attached and all future cgroups will
4300 * have them too. This function can be called anytime whether @ss is
4303 * Returns 0 on successful registration, -errno on failure. Note that this
4304 * function currently returns 0 as long as @cfts registration is successful
4305 * even if some file creation attempts on existing cgroups fail.
4307 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4311 if (!cgroup_ssid_enabled(ss
->id
))
4314 if (!cfts
|| cfts
[0].name
[0] == '\0')
4317 ret
= cgroup_init_cftypes(ss
, cfts
);
4321 mutex_lock(&cgroup_mutex
);
4323 list_add_tail(&cfts
->node
, &ss
->cfts
);
4324 ret
= cgroup_apply_cftypes(cfts
, true);
4326 cgroup_rm_cftypes_locked(cfts
);
4328 mutex_unlock(&cgroup_mutex
);
4333 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4334 * @ss: target cgroup subsystem
4335 * @cfts: zero-length name terminated array of cftypes
4337 * Similar to cgroup_add_cftypes() but the added files are only used for
4338 * the default hierarchy.
4340 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4344 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4345 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
4346 return cgroup_add_cftypes(ss
, cfts
);
4350 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4351 * @ss: target cgroup subsystem
4352 * @cfts: zero-length name terminated array of cftypes
4354 * Similar to cgroup_add_cftypes() but the added files are only used for
4355 * the legacy hierarchies.
4357 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4361 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4362 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
4363 return cgroup_add_cftypes(ss
, cfts
);
4367 * cgroup_file_notify - generate a file modified event for a cgroup_file
4368 * @cfile: target cgroup_file
4370 * @cfile must have been obtained by setting cftype->file_offset.
4372 void cgroup_file_notify(struct cgroup_file
*cfile
)
4374 unsigned long flags
;
4376 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
4378 unsigned long last
= cfile
->notified_at
;
4379 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
4381 if (time_in_range(jiffies
, last
, next
)) {
4382 timer_reduce(&cfile
->notify_timer
, next
);
4384 kernfs_notify(cfile
->kn
);
4385 cfile
->notified_at
= jiffies
;
4388 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
4392 * cgroup_file_show - show or hide a hidden cgroup file
4393 * @cfile: target cgroup_file obtained by setting cftype->file_offset
4394 * @show: whether to show or hide
4396 void cgroup_file_show(struct cgroup_file
*cfile
, bool show
)
4398 struct kernfs_node
*kn
;
4400 spin_lock_irq(&cgroup_file_kn_lock
);
4403 spin_unlock_irq(&cgroup_file_kn_lock
);
4406 kernfs_show(kn
, show
);
4412 * css_next_child - find the next child of a given css
4413 * @pos: the current position (%NULL to initiate traversal)
4414 * @parent: css whose children to walk
4416 * This function returns the next child of @parent and should be called
4417 * under either cgroup_mutex or RCU read lock. The only requirement is
4418 * that @parent and @pos are accessible. The next sibling is guaranteed to
4419 * be returned regardless of their states.
4421 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4422 * css which finished ->css_online() is guaranteed to be visible in the
4423 * future iterations and will stay visible until the last reference is put.
4424 * A css which hasn't finished ->css_online() or already finished
4425 * ->css_offline() may show up during traversal. It's each subsystem's
4426 * responsibility to synchronize against on/offlining.
4428 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
4429 struct cgroup_subsys_state
*parent
)
4431 struct cgroup_subsys_state
*next
;
4433 cgroup_assert_mutex_or_rcu_locked();
4436 * @pos could already have been unlinked from the sibling list.
4437 * Once a cgroup is removed, its ->sibling.next is no longer
4438 * updated when its next sibling changes. CSS_RELEASED is set when
4439 * @pos is taken off list, at which time its next pointer is valid,
4440 * and, as releases are serialized, the one pointed to by the next
4441 * pointer is guaranteed to not have started release yet. This
4442 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4443 * critical section, the one pointed to by its next pointer is
4444 * guaranteed to not have finished its RCU grace period even if we
4445 * have dropped rcu_read_lock() in-between iterations.
4447 * If @pos has CSS_RELEASED set, its next pointer can't be
4448 * dereferenced; however, as each css is given a monotonically
4449 * increasing unique serial number and always appended to the
4450 * sibling list, the next one can be found by walking the parent's
4451 * children until the first css with higher serial number than
4452 * @pos's. While this path can be slower, it happens iff iteration
4453 * races against release and the race window is very small.
4456 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
4457 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
4458 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
4460 list_for_each_entry_rcu(next
, &parent
->children
, sibling
,
4461 lockdep_is_held(&cgroup_mutex
))
4462 if (next
->serial_nr
> pos
->serial_nr
)
4467 * @next, if not pointing to the head, can be dereferenced and is
4470 if (&next
->sibling
!= &parent
->children
)
4476 * css_next_descendant_pre - find the next descendant for pre-order walk
4477 * @pos: the current position (%NULL to initiate traversal)
4478 * @root: css whose descendants to walk
4480 * To be used by css_for_each_descendant_pre(). Find the next descendant
4481 * to visit for pre-order traversal of @root's descendants. @root is
4482 * included in the iteration and the first node to be visited.
4484 * While this function requires cgroup_mutex or RCU read locking, it
4485 * doesn't require the whole traversal to be contained in a single critical
4486 * section. This function will return the correct next descendant as long
4487 * as both @pos and @root are accessible and @pos is a descendant of @root.
4489 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4490 * css which finished ->css_online() is guaranteed to be visible in the
4491 * future iterations and will stay visible until the last reference is put.
4492 * A css which hasn't finished ->css_online() or already finished
4493 * ->css_offline() may show up during traversal. It's each subsystem's
4494 * responsibility to synchronize against on/offlining.
4496 struct cgroup_subsys_state
*
4497 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4498 struct cgroup_subsys_state
*root
)
4500 struct cgroup_subsys_state
*next
;
4502 cgroup_assert_mutex_or_rcu_locked();
4504 /* if first iteration, visit @root */
4508 /* visit the first child if exists */
4509 next
= css_next_child(NULL
, pos
);
4513 /* no child, visit my or the closest ancestor's next sibling */
4514 while (pos
!= root
) {
4515 next
= css_next_child(pos
, pos
->parent
);
4523 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
4526 * css_rightmost_descendant - return the rightmost descendant of a css
4527 * @pos: css of interest
4529 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4530 * is returned. This can be used during pre-order traversal to skip
4533 * While this function requires cgroup_mutex or RCU read locking, it
4534 * doesn't require the whole traversal to be contained in a single critical
4535 * section. This function will return the correct rightmost descendant as
4536 * long as @pos is accessible.
4538 struct cgroup_subsys_state
*
4539 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4541 struct cgroup_subsys_state
*last
, *tmp
;
4543 cgroup_assert_mutex_or_rcu_locked();
4547 /* ->prev isn't RCU safe, walk ->next till the end */
4549 css_for_each_child(tmp
, last
)
4556 static struct cgroup_subsys_state
*
4557 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4559 struct cgroup_subsys_state
*last
;
4563 pos
= css_next_child(NULL
, pos
);
4570 * css_next_descendant_post - find the next descendant for post-order walk
4571 * @pos: the current position (%NULL to initiate traversal)
4572 * @root: css whose descendants to walk
4574 * To be used by css_for_each_descendant_post(). Find the next descendant
4575 * to visit for post-order traversal of @root's descendants. @root is
4576 * included in the iteration and the last node to be visited.
4578 * While this function requires cgroup_mutex or RCU read locking, it
4579 * doesn't require the whole traversal to be contained in a single critical
4580 * section. This function will return the correct next descendant as long
4581 * as both @pos and @cgroup are accessible and @pos is a descendant of
4584 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4585 * css which finished ->css_online() is guaranteed to be visible in the
4586 * future iterations and will stay visible until the last reference is put.
4587 * A css which hasn't finished ->css_online() or already finished
4588 * ->css_offline() may show up during traversal. It's each subsystem's
4589 * responsibility to synchronize against on/offlining.
4591 struct cgroup_subsys_state
*
4592 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4593 struct cgroup_subsys_state
*root
)
4595 struct cgroup_subsys_state
*next
;
4597 cgroup_assert_mutex_or_rcu_locked();
4599 /* if first iteration, visit leftmost descendant which may be @root */
4601 return css_leftmost_descendant(root
);
4603 /* if we visited @root, we're done */
4607 /* if there's an unvisited sibling, visit its leftmost descendant */
4608 next
= css_next_child(pos
, pos
->parent
);
4610 return css_leftmost_descendant(next
);
4612 /* no sibling left, visit parent */
4617 * css_has_online_children - does a css have online children
4618 * @css: the target css
4620 * Returns %true if @css has any online children; otherwise, %false. This
4621 * function can be called from any context but the caller is responsible
4622 * for synchronizing against on/offlining as necessary.
4624 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4626 struct cgroup_subsys_state
*child
;
4630 css_for_each_child(child
, css
) {
4631 if (child
->flags
& CSS_ONLINE
) {
4640 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4642 struct list_head
*l
;
4643 struct cgrp_cset_link
*link
;
4644 struct css_set
*cset
;
4646 lockdep_assert_held(&css_set_lock
);
4648 /* find the next threaded cset */
4649 if (it
->tcset_pos
) {
4650 l
= it
->tcset_pos
->next
;
4652 if (l
!= it
->tcset_head
) {
4654 return container_of(l
, struct css_set
,
4655 threaded_csets_node
);
4658 it
->tcset_pos
= NULL
;
4661 /* find the next cset */
4664 if (l
== it
->cset_head
) {
4665 it
->cset_pos
= NULL
;
4670 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4672 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4678 /* initialize threaded css_set walking */
4679 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4681 put_css_set_locked(it
->cur_dcset
);
4682 it
->cur_dcset
= cset
;
4685 it
->tcset_head
= &cset
->threaded_csets
;
4686 it
->tcset_pos
= &cset
->threaded_csets
;
4693 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4694 * @it: the iterator to advance
4696 * Advance @it to the next css_set to walk.
4698 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4700 struct css_set
*cset
;
4702 lockdep_assert_held(&css_set_lock
);
4704 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4705 while ((cset
= css_task_iter_next_css_set(it
))) {
4706 if (!list_empty(&cset
->tasks
)) {
4707 it
->cur_tasks_head
= &cset
->tasks
;
4709 } else if (!list_empty(&cset
->mg_tasks
)) {
4710 it
->cur_tasks_head
= &cset
->mg_tasks
;
4712 } else if (!list_empty(&cset
->dying_tasks
)) {
4713 it
->cur_tasks_head
= &cset
->dying_tasks
;
4718 it
->task_pos
= NULL
;
4721 it
->task_pos
= it
->cur_tasks_head
->next
;
4724 * We don't keep css_sets locked across iteration steps and thus
4725 * need to take steps to ensure that iteration can be resumed after
4726 * the lock is re-acquired. Iteration is performed at two levels -
4727 * css_sets and tasks in them.
4729 * Once created, a css_set never leaves its cgroup lists, so a
4730 * pinned css_set is guaranteed to stay put and we can resume
4731 * iteration afterwards.
4733 * Tasks may leave @cset across iteration steps. This is resolved
4734 * by registering each iterator with the css_set currently being
4735 * walked and making css_set_move_task() advance iterators whose
4736 * next task is leaving.
4739 list_del(&it
->iters_node
);
4740 put_css_set_locked(it
->cur_cset
);
4743 it
->cur_cset
= cset
;
4744 list_add(&it
->iters_node
, &cset
->task_iters
);
4747 static void css_task_iter_skip(struct css_task_iter
*it
,
4748 struct task_struct
*task
)
4750 lockdep_assert_held(&css_set_lock
);
4752 if (it
->task_pos
== &task
->cg_list
) {
4753 it
->task_pos
= it
->task_pos
->next
;
4754 it
->flags
|= CSS_TASK_ITER_SKIPPED
;
4758 static void css_task_iter_advance(struct css_task_iter
*it
)
4760 struct task_struct
*task
;
4762 lockdep_assert_held(&css_set_lock
);
4766 * Advance iterator to find next entry. We go through cset
4767 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4770 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4771 it
->flags
&= ~CSS_TASK_ITER_SKIPPED
;
4773 it
->task_pos
= it
->task_pos
->next
;
4775 if (it
->task_pos
== &it
->cur_cset
->tasks
) {
4776 it
->cur_tasks_head
= &it
->cur_cset
->mg_tasks
;
4777 it
->task_pos
= it
->cur_tasks_head
->next
;
4779 if (it
->task_pos
== &it
->cur_cset
->mg_tasks
) {
4780 it
->cur_tasks_head
= &it
->cur_cset
->dying_tasks
;
4781 it
->task_pos
= it
->cur_tasks_head
->next
;
4783 if (it
->task_pos
== &it
->cur_cset
->dying_tasks
)
4784 css_task_iter_advance_css_set(it
);
4786 /* called from start, proceed to the first cset */
4787 css_task_iter_advance_css_set(it
);
4793 task
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
4795 if (it
->flags
& CSS_TASK_ITER_PROCS
) {
4796 /* if PROCS, skip over tasks which aren't group leaders */
4797 if (!thread_group_leader(task
))
4800 /* and dying leaders w/o live member threads */
4801 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
&&
4802 !atomic_read(&task
->signal
->live
))
4805 /* skip all dying ones */
4806 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
)
4812 * css_task_iter_start - initiate task iteration
4813 * @css: the css to walk tasks of
4814 * @flags: CSS_TASK_ITER_* flags
4815 * @it: the task iterator to use
4817 * Initiate iteration through the tasks of @css. The caller can call
4818 * css_task_iter_next() to walk through the tasks until the function
4819 * returns NULL. On completion of iteration, css_task_iter_end() must be
4822 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4823 struct css_task_iter
*it
)
4825 memset(it
, 0, sizeof(*it
));
4827 spin_lock_irq(&css_set_lock
);
4832 if (CGROUP_HAS_SUBSYS_CONFIG
&& it
->ss
)
4833 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4835 it
->cset_pos
= &css
->cgroup
->cset_links
;
4837 it
->cset_head
= it
->cset_pos
;
4839 css_task_iter_advance(it
);
4841 spin_unlock_irq(&css_set_lock
);
4845 * css_task_iter_next - return the next task for the iterator
4846 * @it: the task iterator being iterated
4848 * The "next" function for task iteration. @it should have been
4849 * initialized via css_task_iter_start(). Returns NULL when the iteration
4852 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4855 put_task_struct(it
->cur_task
);
4856 it
->cur_task
= NULL
;
4859 spin_lock_irq(&css_set_lock
);
4861 /* @it may be half-advanced by skips, finish advancing */
4862 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4863 css_task_iter_advance(it
);
4866 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4868 get_task_struct(it
->cur_task
);
4869 css_task_iter_advance(it
);
4872 spin_unlock_irq(&css_set_lock
);
4874 return it
->cur_task
;
4878 * css_task_iter_end - finish task iteration
4879 * @it: the task iterator to finish
4881 * Finish task iteration started by css_task_iter_start().
4883 void css_task_iter_end(struct css_task_iter
*it
)
4886 spin_lock_irq(&css_set_lock
);
4887 list_del(&it
->iters_node
);
4888 put_css_set_locked(it
->cur_cset
);
4889 spin_unlock_irq(&css_set_lock
);
4893 put_css_set(it
->cur_dcset
);
4896 put_task_struct(it
->cur_task
);
4899 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4901 struct cgroup_file_ctx
*ctx
= of
->priv
;
4903 if (ctx
->procs
.started
)
4904 css_task_iter_end(&ctx
->procs
.iter
);
4907 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4909 struct kernfs_open_file
*of
= s
->private;
4910 struct cgroup_file_ctx
*ctx
= of
->priv
;
4915 return css_task_iter_next(&ctx
->procs
.iter
);
4918 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4919 unsigned int iter_flags
)
4921 struct kernfs_open_file
*of
= s
->private;
4922 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4923 struct cgroup_file_ctx
*ctx
= of
->priv
;
4924 struct css_task_iter
*it
= &ctx
->procs
.iter
;
4927 * When a seq_file is seeked, it's always traversed sequentially
4928 * from position 0, so we can simply keep iterating on !0 *pos.
4930 if (!ctx
->procs
.started
) {
4931 if (WARN_ON_ONCE((*pos
)))
4932 return ERR_PTR(-EINVAL
);
4933 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4934 ctx
->procs
.started
= true;
4935 } else if (!(*pos
)) {
4936 css_task_iter_end(it
);
4937 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4939 return it
->cur_task
;
4941 return cgroup_procs_next(s
, NULL
, NULL
);
4944 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4946 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4949 * All processes of a threaded subtree belong to the domain cgroup
4950 * of the subtree. Only threads can be distributed across the
4951 * subtree. Reject reads on cgroup.procs in the subtree proper.
4952 * They're always empty anyway.
4954 if (cgroup_is_threaded(cgrp
))
4955 return ERR_PTR(-EOPNOTSUPP
);
4957 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4958 CSS_TASK_ITER_THREADED
);
4961 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4963 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4967 static int cgroup_may_write(const struct cgroup
*cgrp
, struct super_block
*sb
)
4970 struct inode
*inode
;
4972 lockdep_assert_held(&cgroup_mutex
);
4974 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
4978 ret
= inode_permission(&init_user_ns
, inode
, MAY_WRITE
);
4983 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4984 struct cgroup
*dst_cgrp
,
4985 struct super_block
*sb
,
4986 struct cgroup_namespace
*ns
)
4988 struct cgroup
*com_cgrp
= src_cgrp
;
4991 lockdep_assert_held(&cgroup_mutex
);
4993 /* find the common ancestor */
4994 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4995 com_cgrp
= cgroup_parent(com_cgrp
);
4997 /* %current should be authorized to migrate to the common ancestor */
4998 ret
= cgroup_may_write(com_cgrp
, sb
);
5003 * If namespaces are delegation boundaries, %current must be able
5004 * to see both source and destination cgroups from its namespace.
5006 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
5007 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
5008 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
5014 static int cgroup_attach_permissions(struct cgroup
*src_cgrp
,
5015 struct cgroup
*dst_cgrp
,
5016 struct super_block
*sb
, bool threadgroup
,
5017 struct cgroup_namespace
*ns
)
5021 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
, sb
, ns
);
5025 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
5029 if (!threadgroup
&& (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
))
5035 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
5038 struct cgroup_file_ctx
*ctx
= of
->priv
;
5039 struct cgroup
*src_cgrp
, *dst_cgrp
;
5040 struct task_struct
*task
;
5041 const struct cred
*saved_cred
;
5043 bool threadgroup_locked
;
5045 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
5049 task
= cgroup_procs_write_start(buf
, threadgroup
, &threadgroup_locked
);
5050 ret
= PTR_ERR_OR_ZERO(task
);
5054 /* find the source cgroup */
5055 spin_lock_irq(&css_set_lock
);
5056 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
5057 spin_unlock_irq(&css_set_lock
);
5060 * Process and thread migrations follow same delegation rule. Check
5061 * permissions using the credentials from file open to protect against
5062 * inherited fd attacks.
5064 saved_cred
= override_creds(of
->file
->f_cred
);
5065 ret
= cgroup_attach_permissions(src_cgrp
, dst_cgrp
,
5066 of
->file
->f_path
.dentry
->d_sb
,
5067 threadgroup
, ctx
->ns
);
5068 revert_creds(saved_cred
);
5072 ret
= cgroup_attach_task(dst_cgrp
, task
, threadgroup
);
5075 cgroup_procs_write_finish(task
, threadgroup_locked
);
5077 cgroup_kn_unlock(of
->kn
);
5082 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
5083 char *buf
, size_t nbytes
, loff_t off
)
5085 return __cgroup_procs_write(of
, buf
, true) ?: nbytes
;
5088 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
5090 return __cgroup_procs_start(s
, pos
, 0);
5093 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
5094 char *buf
, size_t nbytes
, loff_t off
)
5096 return __cgroup_procs_write(of
, buf
, false) ?: nbytes
;
5099 /* cgroup core interface files for the default hierarchy */
5100 static struct cftype cgroup_base_files
[] = {
5102 .name
= "cgroup.type",
5103 .flags
= CFTYPE_NOT_ON_ROOT
,
5104 .seq_show
= cgroup_type_show
,
5105 .write
= cgroup_type_write
,
5108 .name
= "cgroup.procs",
5109 .flags
= CFTYPE_NS_DELEGATABLE
,
5110 .file_offset
= offsetof(struct cgroup
, procs_file
),
5111 .release
= cgroup_procs_release
,
5112 .seq_start
= cgroup_procs_start
,
5113 .seq_next
= cgroup_procs_next
,
5114 .seq_show
= cgroup_procs_show
,
5115 .write
= cgroup_procs_write
,
5118 .name
= "cgroup.threads",
5119 .flags
= CFTYPE_NS_DELEGATABLE
,
5120 .release
= cgroup_procs_release
,
5121 .seq_start
= cgroup_threads_start
,
5122 .seq_next
= cgroup_procs_next
,
5123 .seq_show
= cgroup_procs_show
,
5124 .write
= cgroup_threads_write
,
5127 .name
= "cgroup.controllers",
5128 .seq_show
= cgroup_controllers_show
,
5131 .name
= "cgroup.subtree_control",
5132 .flags
= CFTYPE_NS_DELEGATABLE
,
5133 .seq_show
= cgroup_subtree_control_show
,
5134 .write
= cgroup_subtree_control_write
,
5137 .name
= "cgroup.events",
5138 .flags
= CFTYPE_NOT_ON_ROOT
,
5139 .file_offset
= offsetof(struct cgroup
, events_file
),
5140 .seq_show
= cgroup_events_show
,
5143 .name
= "cgroup.max.descendants",
5144 .seq_show
= cgroup_max_descendants_show
,
5145 .write
= cgroup_max_descendants_write
,
5148 .name
= "cgroup.max.depth",
5149 .seq_show
= cgroup_max_depth_show
,
5150 .write
= cgroup_max_depth_write
,
5153 .name
= "cgroup.stat",
5154 .seq_show
= cgroup_stat_show
,
5157 .name
= "cgroup.freeze",
5158 .flags
= CFTYPE_NOT_ON_ROOT
,
5159 .seq_show
= cgroup_freeze_show
,
5160 .write
= cgroup_freeze_write
,
5163 .name
= "cgroup.kill",
5164 .flags
= CFTYPE_NOT_ON_ROOT
,
5165 .write
= cgroup_kill_write
,
5169 .seq_show
= cpu_stat_show
,
5174 static struct cftype cgroup_psi_files
[] = {
5177 .name
= "io.pressure",
5178 .seq_show
= cgroup_io_pressure_show
,
5179 .write
= cgroup_io_pressure_write
,
5180 .poll
= cgroup_pressure_poll
,
5181 .release
= cgroup_pressure_release
,
5184 .name
= "memory.pressure",
5185 .seq_show
= cgroup_memory_pressure_show
,
5186 .write
= cgroup_memory_pressure_write
,
5187 .poll
= cgroup_pressure_poll
,
5188 .release
= cgroup_pressure_release
,
5191 .name
= "cpu.pressure",
5192 .seq_show
= cgroup_cpu_pressure_show
,
5193 .write
= cgroup_cpu_pressure_write
,
5194 .poll
= cgroup_pressure_poll
,
5195 .release
= cgroup_pressure_release
,
5197 #endif /* CONFIG_PSI */
5202 * css destruction is four-stage process.
5204 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5205 * Implemented in kill_css().
5207 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5208 * and thus css_tryget_online() is guaranteed to fail, the css can be
5209 * offlined by invoking offline_css(). After offlining, the base ref is
5210 * put. Implemented in css_killed_work_fn().
5212 * 3. When the percpu_ref reaches zero, the only possible remaining
5213 * accessors are inside RCU read sections. css_release() schedules the
5216 * 4. After the grace period, the css can be freed. Implemented in
5217 * css_free_work_fn().
5219 * It is actually hairier because both step 2 and 4 require process context
5220 * and thus involve punting to css->destroy_work adding two additional
5221 * steps to the already complex sequence.
5223 static void css_free_rwork_fn(struct work_struct
*work
)
5225 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
5226 struct cgroup_subsys_state
, destroy_rwork
);
5227 struct cgroup_subsys
*ss
= css
->ss
;
5228 struct cgroup
*cgrp
= css
->cgroup
;
5230 percpu_ref_exit(&css
->refcnt
);
5234 struct cgroup_subsys_state
*parent
= css
->parent
;
5238 cgroup_idr_remove(&ss
->css_idr
, id
);
5244 /* cgroup free path */
5245 atomic_dec(&cgrp
->root
->nr_cgrps
);
5246 cgroup1_pidlist_destroy_all(cgrp
);
5247 cancel_work_sync(&cgrp
->release_agent_work
);
5249 if (cgroup_parent(cgrp
)) {
5251 * We get a ref to the parent, and put the ref when
5252 * this cgroup is being freed, so it's guaranteed
5253 * that the parent won't be destroyed before its
5256 cgroup_put(cgroup_parent(cgrp
));
5257 kernfs_put(cgrp
->kn
);
5258 psi_cgroup_free(cgrp
);
5259 cgroup_rstat_exit(cgrp
);
5263 * This is root cgroup's refcnt reaching zero,
5264 * which indicates that the root should be
5267 cgroup_destroy_root(cgrp
->root
);
5272 static void css_release_work_fn(struct work_struct
*work
)
5274 struct cgroup_subsys_state
*css
=
5275 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5276 struct cgroup_subsys
*ss
= css
->ss
;
5277 struct cgroup
*cgrp
= css
->cgroup
;
5279 mutex_lock(&cgroup_mutex
);
5281 css
->flags
|= CSS_RELEASED
;
5282 list_del_rcu(&css
->sibling
);
5285 /* css release path */
5286 if (!list_empty(&css
->rstat_css_node
)) {
5287 cgroup_rstat_flush(cgrp
);
5288 list_del_rcu(&css
->rstat_css_node
);
5291 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
5292 if (ss
->css_released
)
5293 ss
->css_released(css
);
5295 struct cgroup
*tcgrp
;
5297 /* cgroup release path */
5298 TRACE_CGROUP_PATH(release
, cgrp
);
5300 cgroup_rstat_flush(cgrp
);
5302 spin_lock_irq(&css_set_lock
);
5303 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
5304 tcgrp
= cgroup_parent(tcgrp
))
5305 tcgrp
->nr_dying_descendants
--;
5306 spin_unlock_irq(&css_set_lock
);
5309 * There are two control paths which try to determine
5310 * cgroup from dentry without going through kernfs -
5311 * cgroupstats_build() and css_tryget_online_from_dir().
5312 * Those are supported by RCU protecting clearing of
5313 * cgrp->kn->priv backpointer.
5316 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5320 mutex_unlock(&cgroup_mutex
);
5322 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5323 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5326 static void css_release(struct percpu_ref
*ref
)
5328 struct cgroup_subsys_state
*css
=
5329 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5331 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5332 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5335 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5336 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5338 lockdep_assert_held(&cgroup_mutex
);
5340 cgroup_get_live(cgrp
);
5342 memset(css
, 0, sizeof(*css
));
5346 INIT_LIST_HEAD(&css
->sibling
);
5347 INIT_LIST_HEAD(&css
->children
);
5348 INIT_LIST_HEAD(&css
->rstat_css_node
);
5349 css
->serial_nr
= css_serial_nr_next
++;
5350 atomic_set(&css
->online_cnt
, 0);
5352 if (cgroup_parent(cgrp
)) {
5353 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5354 css_get(css
->parent
);
5357 if (ss
->css_rstat_flush
)
5358 list_add_rcu(&css
->rstat_css_node
, &cgrp
->rstat_css_list
);
5360 BUG_ON(cgroup_css(cgrp
, ss
));
5363 /* invoke ->css_online() on a new CSS and mark it online if successful */
5364 static int online_css(struct cgroup_subsys_state
*css
)
5366 struct cgroup_subsys
*ss
= css
->ss
;
5369 lockdep_assert_held(&cgroup_mutex
);
5372 ret
= ss
->css_online(css
);
5374 css
->flags
|= CSS_ONLINE
;
5375 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5377 atomic_inc(&css
->online_cnt
);
5379 atomic_inc(&css
->parent
->online_cnt
);
5384 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5385 static void offline_css(struct cgroup_subsys_state
*css
)
5387 struct cgroup_subsys
*ss
= css
->ss
;
5389 lockdep_assert_held(&cgroup_mutex
);
5391 if (!(css
->flags
& CSS_ONLINE
))
5394 if (ss
->css_offline
)
5395 ss
->css_offline(css
);
5397 css
->flags
&= ~CSS_ONLINE
;
5398 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5400 wake_up_all(&css
->cgroup
->offline_waitq
);
5404 * css_create - create a cgroup_subsys_state
5405 * @cgrp: the cgroup new css will be associated with
5406 * @ss: the subsys of new css
5408 * Create a new css associated with @cgrp - @ss pair. On success, the new
5409 * css is online and installed in @cgrp. This function doesn't create the
5410 * interface files. Returns 0 on success, -errno on failure.
5412 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5413 struct cgroup_subsys
*ss
)
5415 struct cgroup
*parent
= cgroup_parent(cgrp
);
5416 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5417 struct cgroup_subsys_state
*css
;
5420 lockdep_assert_held(&cgroup_mutex
);
5422 css
= ss
->css_alloc(parent_css
);
5424 css
= ERR_PTR(-ENOMEM
);
5428 init_and_link_css(css
, ss
, cgrp
);
5430 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5434 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5439 /* @css is ready to be brought online now, make it visible */
5440 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5441 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5443 err
= online_css(css
);
5450 list_del_rcu(&css
->sibling
);
5452 list_del_rcu(&css
->rstat_css_node
);
5453 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5454 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5455 return ERR_PTR(err
);
5459 * The returned cgroup is fully initialized including its control mask, but
5460 * it isn't associated with its kernfs_node and doesn't have the control
5463 static struct cgroup
*cgroup_create(struct cgroup
*parent
, const char *name
,
5466 struct cgroup_root
*root
= parent
->root
;
5467 struct cgroup
*cgrp
, *tcgrp
;
5468 struct kernfs_node
*kn
;
5469 int level
= parent
->level
+ 1;
5472 /* allocate the cgroup and its ID, 0 is reserved for the root */
5473 cgrp
= kzalloc(struct_size(cgrp
, ancestors
, (level
+ 1)), GFP_KERNEL
);
5475 return ERR_PTR(-ENOMEM
);
5477 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5481 ret
= cgroup_rstat_init(cgrp
);
5483 goto out_cancel_ref
;
5485 /* create the directory */
5486 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5493 init_cgroup_housekeeping(cgrp
);
5495 cgrp
->self
.parent
= &parent
->self
;
5497 cgrp
->level
= level
;
5499 ret
= psi_cgroup_alloc(cgrp
);
5501 goto out_kernfs_remove
;
5503 ret
= cgroup_bpf_inherit(cgrp
);
5508 * New cgroup inherits effective freeze counter, and
5509 * if the parent has to be frozen, the child has too.
5511 cgrp
->freezer
.e_freeze
= parent
->freezer
.e_freeze
;
5512 if (cgrp
->freezer
.e_freeze
) {
5514 * Set the CGRP_FREEZE flag, so when a process will be
5515 * attached to the child cgroup, it will become frozen.
5516 * At this point the new cgroup is unpopulated, so we can
5517 * consider it frozen immediately.
5519 set_bit(CGRP_FREEZE
, &cgrp
->flags
);
5520 set_bit(CGRP_FROZEN
, &cgrp
->flags
);
5523 spin_lock_irq(&css_set_lock
);
5524 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5525 cgrp
->ancestors
[tcgrp
->level
] = tcgrp
;
5527 if (tcgrp
!= cgrp
) {
5528 tcgrp
->nr_descendants
++;
5531 * If the new cgroup is frozen, all ancestor cgroups
5532 * get a new frozen descendant, but their state can't
5533 * change because of this.
5535 if (cgrp
->freezer
.e_freeze
)
5536 tcgrp
->freezer
.nr_frozen_descendants
++;
5539 spin_unlock_irq(&css_set_lock
);
5541 if (notify_on_release(parent
))
5542 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5544 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5545 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5547 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5549 /* allocation complete, commit to creation */
5550 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5551 atomic_inc(&root
->nr_cgrps
);
5552 cgroup_get_live(parent
);
5555 * On the default hierarchy, a child doesn't automatically inherit
5556 * subtree_control from the parent. Each is configured manually.
5558 if (!cgroup_on_dfl(cgrp
))
5559 cgrp
->subtree_control
= cgroup_control(cgrp
);
5561 cgroup_propagate_control(cgrp
);
5566 psi_cgroup_free(cgrp
);
5568 kernfs_remove(cgrp
->kn
);
5570 cgroup_rstat_exit(cgrp
);
5572 percpu_ref_exit(&cgrp
->self
.refcnt
);
5575 return ERR_PTR(ret
);
5578 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
5580 struct cgroup
*cgroup
;
5584 lockdep_assert_held(&cgroup_mutex
);
5586 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
5587 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
5590 if (level
> cgroup
->max_depth
)
5601 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
5603 struct cgroup
*parent
, *cgrp
;
5606 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5607 if (strchr(name
, '\n'))
5610 parent
= cgroup_kn_lock_live(parent_kn
, false);
5614 if (!cgroup_check_hierarchy_limits(parent
)) {
5619 cgrp
= cgroup_create(parent
, name
, mode
);
5621 ret
= PTR_ERR(cgrp
);
5626 * This extra ref will be put in cgroup_free_fn() and guarantees
5627 * that @cgrp->kn is always accessible.
5629 kernfs_get(cgrp
->kn
);
5631 ret
= cgroup_kn_set_ugid(cgrp
->kn
);
5635 ret
= css_populate_dir(&cgrp
->self
);
5639 ret
= cgroup_apply_control_enable(cgrp
);
5643 TRACE_CGROUP_PATH(mkdir
, cgrp
);
5645 /* let's create and online css's */
5646 kernfs_activate(cgrp
->kn
);
5652 cgroup_destroy_locked(cgrp
);
5654 cgroup_kn_unlock(parent_kn
);
5659 * This is called when the refcnt of a css is confirmed to be killed.
5660 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5661 * initiate destruction and put the css ref from kill_css().
5663 static void css_killed_work_fn(struct work_struct
*work
)
5665 struct cgroup_subsys_state
*css
=
5666 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5668 mutex_lock(&cgroup_mutex
);
5673 /* @css can't go away while we're holding cgroup_mutex */
5675 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5677 mutex_unlock(&cgroup_mutex
);
5680 /* css kill confirmation processing requires process context, bounce */
5681 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5683 struct cgroup_subsys_state
*css
=
5684 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5686 if (atomic_dec_and_test(&css
->online_cnt
)) {
5687 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5688 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5693 * kill_css - destroy a css
5694 * @css: css to destroy
5696 * This function initiates destruction of @css by removing cgroup interface
5697 * files and putting its base reference. ->css_offline() will be invoked
5698 * asynchronously once css_tryget_online() is guaranteed to fail and when
5699 * the reference count reaches zero, @css will be released.
5701 static void kill_css(struct cgroup_subsys_state
*css
)
5703 lockdep_assert_held(&cgroup_mutex
);
5705 if (css
->flags
& CSS_DYING
)
5708 css
->flags
|= CSS_DYING
;
5711 * This must happen before css is disassociated with its cgroup.
5712 * See seq_css() for details.
5717 * Killing would put the base ref, but we need to keep it alive
5718 * until after ->css_offline().
5723 * cgroup core guarantees that, by the time ->css_offline() is
5724 * invoked, no new css reference will be given out via
5725 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5726 * proceed to offlining css's because percpu_ref_kill() doesn't
5727 * guarantee that the ref is seen as killed on all CPUs on return.
5729 * Use percpu_ref_kill_and_confirm() to get notifications as each
5730 * css is confirmed to be seen as killed on all CPUs.
5732 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5736 * cgroup_destroy_locked - the first stage of cgroup destruction
5737 * @cgrp: cgroup to be destroyed
5739 * css's make use of percpu refcnts whose killing latency shouldn't be
5740 * exposed to userland and are RCU protected. Also, cgroup core needs to
5741 * guarantee that css_tryget_online() won't succeed by the time
5742 * ->css_offline() is invoked. To satisfy all the requirements,
5743 * destruction is implemented in the following two steps.
5745 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5746 * userland visible parts and start killing the percpu refcnts of
5747 * css's. Set up so that the next stage will be kicked off once all
5748 * the percpu refcnts are confirmed to be killed.
5750 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5751 * rest of destruction. Once all cgroup references are gone, the
5752 * cgroup is RCU-freed.
5754 * This function implements s1. After this step, @cgrp is gone as far as
5755 * the userland is concerned and a new cgroup with the same name may be
5756 * created. As cgroup doesn't care about the names internally, this
5757 * doesn't cause any problem.
5759 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5760 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5762 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5763 struct cgroup_subsys_state
*css
;
5764 struct cgrp_cset_link
*link
;
5767 lockdep_assert_held(&cgroup_mutex
);
5770 * Only migration can raise populated from zero and we're already
5771 * holding cgroup_mutex.
5773 if (cgroup_is_populated(cgrp
))
5777 * Make sure there's no live children. We can't test emptiness of
5778 * ->self.children as dead children linger on it while being
5779 * drained; otherwise, "rmdir parent/child parent" may fail.
5781 if (css_has_online_children(&cgrp
->self
))
5785 * Mark @cgrp and the associated csets dead. The former prevents
5786 * further task migration and child creation by disabling
5787 * cgroup_lock_live_group(). The latter makes the csets ignored by
5788 * the migration path.
5790 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5792 spin_lock_irq(&css_set_lock
);
5793 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5794 link
->cset
->dead
= true;
5795 spin_unlock_irq(&css_set_lock
);
5797 /* initiate massacre of all css's */
5798 for_each_css(css
, ssid
, cgrp
)
5801 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5802 css_clear_dir(&cgrp
->self
);
5803 kernfs_remove(cgrp
->kn
);
5805 if (cgroup_is_threaded(cgrp
))
5806 parent
->nr_threaded_children
--;
5808 spin_lock_irq(&css_set_lock
);
5809 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5810 tcgrp
->nr_descendants
--;
5811 tcgrp
->nr_dying_descendants
++;
5813 * If the dying cgroup is frozen, decrease frozen descendants
5814 * counters of ancestor cgroups.
5816 if (test_bit(CGRP_FROZEN
, &cgrp
->flags
))
5817 tcgrp
->freezer
.nr_frozen_descendants
--;
5819 spin_unlock_irq(&css_set_lock
);
5821 cgroup1_check_for_release(parent
);
5823 cgroup_bpf_offline(cgrp
);
5825 /* put the base reference */
5826 percpu_ref_kill(&cgrp
->self
.refcnt
);
5831 int cgroup_rmdir(struct kernfs_node
*kn
)
5833 struct cgroup
*cgrp
;
5836 cgrp
= cgroup_kn_lock_live(kn
, false);
5840 ret
= cgroup_destroy_locked(cgrp
);
5842 TRACE_CGROUP_PATH(rmdir
, cgrp
);
5844 cgroup_kn_unlock(kn
);
5848 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5849 .show_options
= cgroup_show_options
,
5850 .mkdir
= cgroup_mkdir
,
5851 .rmdir
= cgroup_rmdir
,
5852 .show_path
= cgroup_show_path
,
5855 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5857 struct cgroup_subsys_state
*css
;
5859 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5861 mutex_lock(&cgroup_mutex
);
5863 idr_init(&ss
->css_idr
);
5864 INIT_LIST_HEAD(&ss
->cfts
);
5866 /* Create the root cgroup state for this subsystem */
5867 ss
->root
= &cgrp_dfl_root
;
5868 css
= ss
->css_alloc(NULL
);
5869 /* We don't handle early failures gracefully */
5870 BUG_ON(IS_ERR(css
));
5871 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5874 * Root csses are never destroyed and we can't initialize
5875 * percpu_ref during early init. Disable refcnting.
5877 css
->flags
|= CSS_NO_REF
;
5880 /* allocation can't be done safely during early init */
5883 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5884 BUG_ON(css
->id
< 0);
5887 /* Update the init_css_set to contain a subsys
5888 * pointer to this state - since the subsystem is
5889 * newly registered, all tasks and hence the
5890 * init_css_set is in the subsystem's root cgroup. */
5891 init_css_set
.subsys
[ss
->id
] = css
;
5893 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5894 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5895 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5896 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5898 /* At system boot, before all subsystems have been
5899 * registered, no tasks have been forked, so we don't
5900 * need to invoke fork callbacks here. */
5901 BUG_ON(!list_empty(&init_task
.tasks
));
5903 BUG_ON(online_css(css
));
5905 mutex_unlock(&cgroup_mutex
);
5909 * cgroup_init_early - cgroup initialization at system boot
5911 * Initialize cgroups at system boot, and initialize any
5912 * subsystems that request early init.
5914 int __init
cgroup_init_early(void)
5916 static struct cgroup_fs_context __initdata ctx
;
5917 struct cgroup_subsys
*ss
;
5920 ctx
.root
= &cgrp_dfl_root
;
5921 init_cgroup_root(&ctx
);
5922 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5924 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5926 for_each_subsys(ss
, i
) {
5927 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5928 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5929 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5931 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5932 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5935 ss
->name
= cgroup_subsys_name
[i
];
5936 if (!ss
->legacy_name
)
5937 ss
->legacy_name
= cgroup_subsys_name
[i
];
5940 cgroup_init_subsys(ss
, true);
5946 * cgroup_init - cgroup initialization
5948 * Register cgroup filesystem and /proc file, and initialize
5949 * any subsystems that didn't request early init.
5951 int __init
cgroup_init(void)
5953 struct cgroup_subsys
*ss
;
5956 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5957 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5958 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_psi_files
));
5959 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5961 cgroup_rstat_boot();
5963 get_user_ns(init_cgroup_ns
.user_ns
);
5965 mutex_lock(&cgroup_mutex
);
5968 * Add init_css_set to the hash table so that dfl_root can link to
5971 hash_add(css_set_table
, &init_css_set
.hlist
,
5972 css_set_hash(init_css_set
.subsys
));
5974 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5976 mutex_unlock(&cgroup_mutex
);
5978 for_each_subsys(ss
, ssid
) {
5979 if (ss
->early_init
) {
5980 struct cgroup_subsys_state
*css
=
5981 init_css_set
.subsys
[ss
->id
];
5983 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5985 BUG_ON(css
->id
< 0);
5987 cgroup_init_subsys(ss
, false);
5990 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5991 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5994 * Setting dfl_root subsys_mask needs to consider the
5995 * disabled flag and cftype registration needs kmalloc,
5996 * both of which aren't available during early_init.
5998 if (!cgroup_ssid_enabled(ssid
))
6001 if (cgroup1_ssid_disabled(ssid
))
6002 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
6005 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
6007 /* implicit controllers must be threaded too */
6008 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
6010 if (ss
->implicit_on_dfl
)
6011 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
6012 else if (!ss
->dfl_cftypes
)
6013 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
6016 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
6018 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
6019 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
6021 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
6022 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
6026 ss
->bind(init_css_set
.subsys
[ssid
]);
6028 mutex_lock(&cgroup_mutex
);
6029 css_populate_dir(init_css_set
.subsys
[ssid
]);
6030 mutex_unlock(&cgroup_mutex
);
6033 /* init_css_set.subsys[] has been updated, re-hash */
6034 hash_del(&init_css_set
.hlist
);
6035 hash_add(css_set_table
, &init_css_set
.hlist
,
6036 css_set_hash(init_css_set
.subsys
));
6038 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
6039 WARN_ON(register_filesystem(&cgroup_fs_type
));
6040 WARN_ON(register_filesystem(&cgroup2_fs_type
));
6041 WARN_ON(!proc_create_single("cgroups", 0, NULL
, proc_cgroupstats_show
));
6042 #ifdef CONFIG_CPUSETS
6043 WARN_ON(register_filesystem(&cpuset_fs_type
));
6049 static int __init
cgroup_wq_init(void)
6052 * There isn't much point in executing destruction path in
6053 * parallel. Good chunk is serialized with cgroup_mutex anyway.
6054 * Use 1 for @max_active.
6056 * We would prefer to do this in cgroup_init() above, but that
6057 * is called before init_workqueues(): so leave this until after.
6059 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
6060 BUG_ON(!cgroup_destroy_wq
);
6063 core_initcall(cgroup_wq_init
);
6065 void cgroup_path_from_kernfs_id(u64 id
, char *buf
, size_t buflen
)
6067 struct kernfs_node
*kn
;
6069 kn
= kernfs_find_and_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
6072 kernfs_path(kn
, buf
, buflen
);
6077 * cgroup_get_from_id : get the cgroup associated with cgroup id
6079 * On success return the cgrp or ERR_PTR on failure
6080 * Only cgroups within current task's cgroup NS are valid.
6082 struct cgroup
*cgroup_get_from_id(u64 id
)
6084 struct kernfs_node
*kn
;
6085 struct cgroup
*cgrp
, *root_cgrp
;
6087 kn
= kernfs_find_and_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
6089 return ERR_PTR(-ENOENT
);
6091 if (kernfs_type(kn
) != KERNFS_DIR
) {
6093 return ERR_PTR(-ENOENT
);
6098 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6099 if (cgrp
&& !cgroup_tryget(cgrp
))
6106 return ERR_PTR(-ENOENT
);
6108 spin_lock_irq(&css_set_lock
);
6109 root_cgrp
= current_cgns_cgroup_from_root(&cgrp_dfl_root
);
6110 spin_unlock_irq(&css_set_lock
);
6111 if (!cgroup_is_descendant(cgrp
, root_cgrp
)) {
6113 return ERR_PTR(-ENOENT
);
6118 EXPORT_SYMBOL_GPL(cgroup_get_from_id
);
6121 * proc_cgroup_show()
6122 * - Print task's cgroup paths into seq_file, one line for each hierarchy
6123 * - Used for /proc/<pid>/cgroup.
6125 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
6126 struct pid
*pid
, struct task_struct
*tsk
)
6130 struct cgroup_root
*root
;
6133 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
6137 mutex_lock(&cgroup_mutex
);
6138 spin_lock_irq(&css_set_lock
);
6140 for_each_root(root
) {
6141 struct cgroup_subsys
*ss
;
6142 struct cgroup
*cgrp
;
6143 int ssid
, count
= 0;
6145 if (root
== &cgrp_dfl_root
&& !READ_ONCE(cgrp_dfl_visible
))
6148 seq_printf(m
, "%d:", root
->hierarchy_id
);
6149 if (root
!= &cgrp_dfl_root
)
6150 for_each_subsys(ss
, ssid
)
6151 if (root
->subsys_mask
& (1 << ssid
))
6152 seq_printf(m
, "%s%s", count
++ ? "," : "",
6154 if (strlen(root
->name
))
6155 seq_printf(m
, "%sname=%s", count
? "," : "",
6159 cgrp
= task_cgroup_from_root(tsk
, root
);
6162 * On traditional hierarchies, all zombie tasks show up as
6163 * belonging to the root cgroup. On the default hierarchy,
6164 * while a zombie doesn't show up in "cgroup.procs" and
6165 * thus can't be migrated, its /proc/PID/cgroup keeps
6166 * reporting the cgroup it belonged to before exiting. If
6167 * the cgroup is removed before the zombie is reaped,
6168 * " (deleted)" is appended to the cgroup path.
6170 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
6171 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
6172 current
->nsproxy
->cgroup_ns
);
6173 if (retval
>= PATH_MAX
)
6174 retval
= -ENAMETOOLONG
;
6183 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
6184 seq_puts(m
, " (deleted)\n");
6191 spin_unlock_irq(&css_set_lock
);
6192 mutex_unlock(&cgroup_mutex
);
6199 * cgroup_fork - initialize cgroup related fields during copy_process()
6200 * @child: pointer to task_struct of forking parent process.
6202 * A task is associated with the init_css_set until cgroup_post_fork()
6203 * attaches it to the target css_set.
6205 void cgroup_fork(struct task_struct
*child
)
6207 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
6208 INIT_LIST_HEAD(&child
->cg_list
);
6211 static struct cgroup
*cgroup_get_from_file(struct file
*f
)
6213 struct cgroup_subsys_state
*css
;
6214 struct cgroup
*cgrp
;
6216 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
6218 return ERR_CAST(css
);
6225 * cgroup_css_set_fork - find or create a css_set for a child process
6226 * @kargs: the arguments passed to create the child process
6228 * This functions finds or creates a new css_set which the child
6229 * process will be attached to in cgroup_post_fork(). By default,
6230 * the child process will be given the same css_set as its parent.
6232 * If CLONE_INTO_CGROUP is specified this function will try to find an
6233 * existing css_set which includes the requested cgroup and if not create
6234 * a new css_set that the child will be attached to later. If this function
6235 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6236 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6237 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6238 * to the target cgroup.
6240 static int cgroup_css_set_fork(struct kernel_clone_args
*kargs
)
6241 __acquires(&cgroup_mutex
) __acquires(&cgroup_threadgroup_rwsem
)
6244 struct cgroup
*dst_cgrp
= NULL
;
6245 struct css_set
*cset
;
6246 struct super_block
*sb
;
6249 if (kargs
->flags
& CLONE_INTO_CGROUP
)
6250 mutex_lock(&cgroup_mutex
);
6252 cgroup_threadgroup_change_begin(current
);
6254 spin_lock_irq(&css_set_lock
);
6255 cset
= task_css_set(current
);
6257 spin_unlock_irq(&css_set_lock
);
6259 if (!(kargs
->flags
& CLONE_INTO_CGROUP
)) {
6264 f
= fget_raw(kargs
->cgroup
);
6269 sb
= f
->f_path
.dentry
->d_sb
;
6271 dst_cgrp
= cgroup_get_from_file(f
);
6272 if (IS_ERR(dst_cgrp
)) {
6273 ret
= PTR_ERR(dst_cgrp
);
6278 if (cgroup_is_dead(dst_cgrp
)) {
6284 * Verify that we the target cgroup is writable for us. This is
6285 * usually done by the vfs layer but since we're not going through
6286 * the vfs layer here we need to do it "manually".
6288 ret
= cgroup_may_write(dst_cgrp
, sb
);
6293 * Spawning a task directly into a cgroup works by passing a file
6294 * descriptor to the target cgroup directory. This can even be an O_PATH
6295 * file descriptor. But it can never be a cgroup.procs file descriptor.
6296 * This was done on purpose so spawning into a cgroup could be
6297 * conceptualized as an atomic
6299 * fd = openat(dfd_cgroup, "cgroup.procs", ...);
6300 * write(fd, <child-pid>, ...);
6302 * sequence, i.e. it's a shorthand for the caller opening and writing
6303 * cgroup.procs of the cgroup indicated by @dfd_cgroup. This allows us
6304 * to always use the caller's credentials.
6306 ret
= cgroup_attach_permissions(cset
->dfl_cgrp
, dst_cgrp
, sb
,
6307 !(kargs
->flags
& CLONE_THREAD
),
6308 current
->nsproxy
->cgroup_ns
);
6312 kargs
->cset
= find_css_set(cset
, dst_cgrp
);
6320 kargs
->cgrp
= dst_cgrp
;
6324 cgroup_threadgroup_change_end(current
);
6325 mutex_unlock(&cgroup_mutex
);
6329 cgroup_put(dst_cgrp
);
6332 put_css_set(kargs
->cset
);
6337 * cgroup_css_set_put_fork - drop references we took during fork
6338 * @kargs: the arguments passed to create the child process
6340 * Drop references to the prepared css_set and target cgroup if
6341 * CLONE_INTO_CGROUP was requested.
6343 static void cgroup_css_set_put_fork(struct kernel_clone_args
*kargs
)
6344 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6346 cgroup_threadgroup_change_end(current
);
6348 if (kargs
->flags
& CLONE_INTO_CGROUP
) {
6349 struct cgroup
*cgrp
= kargs
->cgrp
;
6350 struct css_set
*cset
= kargs
->cset
;
6352 mutex_unlock(&cgroup_mutex
);
6367 * cgroup_can_fork - called on a new task before the process is exposed
6368 * @child: the child process
6369 * @kargs: the arguments passed to create the child process
6371 * This prepares a new css_set for the child process which the child will
6372 * be attached to in cgroup_post_fork().
6373 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6374 * callback returns an error, the fork aborts with that error code. This
6375 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6377 int cgroup_can_fork(struct task_struct
*child
, struct kernel_clone_args
*kargs
)
6379 struct cgroup_subsys
*ss
;
6382 ret
= cgroup_css_set_fork(kargs
);
6386 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
6387 ret
= ss
->can_fork(child
, kargs
->cset
);
6390 } while_each_subsys_mask();
6395 for_each_subsys(ss
, j
) {
6398 if (ss
->cancel_fork
)
6399 ss
->cancel_fork(child
, kargs
->cset
);
6402 cgroup_css_set_put_fork(kargs
);
6408 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6409 * @child: the child process
6410 * @kargs: the arguments passed to create the child process
6412 * This calls the cancel_fork() callbacks if a fork failed *after*
6413 * cgroup_can_fork() succeeded and cleans up references we took to
6414 * prepare a new css_set for the child process in cgroup_can_fork().
6416 void cgroup_cancel_fork(struct task_struct
*child
,
6417 struct kernel_clone_args
*kargs
)
6419 struct cgroup_subsys
*ss
;
6422 for_each_subsys(ss
, i
)
6423 if (ss
->cancel_fork
)
6424 ss
->cancel_fork(child
, kargs
->cset
);
6426 cgroup_css_set_put_fork(kargs
);
6430 * cgroup_post_fork - finalize cgroup setup for the child process
6431 * @child: the child process
6432 * @kargs: the arguments passed to create the child process
6434 * Attach the child process to its css_set calling the subsystem fork()
6437 void cgroup_post_fork(struct task_struct
*child
,
6438 struct kernel_clone_args
*kargs
)
6439 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6441 unsigned long cgrp_flags
= 0;
6443 struct cgroup_subsys
*ss
;
6444 struct css_set
*cset
;
6450 spin_lock_irq(&css_set_lock
);
6452 /* init tasks are special, only link regular threads */
6453 if (likely(child
->pid
)) {
6455 cgrp_flags
= kargs
->cgrp
->flags
;
6457 cgrp_flags
= cset
->dfl_cgrp
->flags
;
6459 WARN_ON_ONCE(!list_empty(&child
->cg_list
));
6461 css_set_move_task(child
, NULL
, cset
, false);
6467 if (!(child
->flags
& PF_KTHREAD
)) {
6468 if (unlikely(test_bit(CGRP_FREEZE
, &cgrp_flags
))) {
6470 * If the cgroup has to be frozen, the new task has
6471 * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6472 * get the task into the frozen state.
6474 spin_lock(&child
->sighand
->siglock
);
6475 WARN_ON_ONCE(child
->frozen
);
6476 child
->jobctl
|= JOBCTL_TRAP_FREEZE
;
6477 spin_unlock(&child
->sighand
->siglock
);
6480 * Calling cgroup_update_frozen() isn't required here,
6481 * because it will be called anyway a bit later from
6482 * do_freezer_trap(). So we avoid cgroup's transient
6483 * switch from the frozen state and back.
6488 * If the cgroup is to be killed notice it now and take the
6489 * child down right after we finished preparing it for
6492 kill
= test_bit(CGRP_KILL
, &cgrp_flags
);
6495 spin_unlock_irq(&css_set_lock
);
6498 * Call ss->fork(). This must happen after @child is linked on
6499 * css_set; otherwise, @child might change state between ->fork()
6500 * and addition to css_set.
6502 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
6504 } while_each_subsys_mask();
6506 /* Make the new cset the root_cset of the new cgroup namespace. */
6507 if (kargs
->flags
& CLONE_NEWCGROUP
) {
6508 struct css_set
*rcset
= child
->nsproxy
->cgroup_ns
->root_cset
;
6511 child
->nsproxy
->cgroup_ns
->root_cset
= cset
;
6515 /* Cgroup has to be killed so take down child immediately. */
6517 do_send_sig_info(SIGKILL
, SEND_SIG_NOINFO
, child
, PIDTYPE_TGID
);
6519 cgroup_css_set_put_fork(kargs
);
6523 * cgroup_exit - detach cgroup from exiting task
6524 * @tsk: pointer to task_struct of exiting process
6526 * Description: Detach cgroup from @tsk.
6529 void cgroup_exit(struct task_struct
*tsk
)
6531 struct cgroup_subsys
*ss
;
6532 struct css_set
*cset
;
6535 spin_lock_irq(&css_set_lock
);
6537 WARN_ON_ONCE(list_empty(&tsk
->cg_list
));
6538 cset
= task_css_set(tsk
);
6539 css_set_move_task(tsk
, cset
, NULL
, false);
6540 list_add_tail(&tsk
->cg_list
, &cset
->dying_tasks
);
6543 WARN_ON_ONCE(cgroup_task_frozen(tsk
));
6544 if (unlikely(!(tsk
->flags
& PF_KTHREAD
) &&
6545 test_bit(CGRP_FREEZE
, &task_dfl_cgroup(tsk
)->flags
)))
6546 cgroup_update_frozen(task_dfl_cgroup(tsk
));
6548 spin_unlock_irq(&css_set_lock
);
6550 /* see cgroup_post_fork() for details */
6551 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
6553 } while_each_subsys_mask();
6556 void cgroup_release(struct task_struct
*task
)
6558 struct cgroup_subsys
*ss
;
6561 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
6563 } while_each_subsys_mask();
6565 spin_lock_irq(&css_set_lock
);
6566 css_set_skip_task_iters(task_css_set(task
), task
);
6567 list_del_init(&task
->cg_list
);
6568 spin_unlock_irq(&css_set_lock
);
6571 void cgroup_free(struct task_struct
*task
)
6573 struct css_set
*cset
= task_css_set(task
);
6577 static int __init
cgroup_disable(char *str
)
6579 struct cgroup_subsys
*ss
;
6583 while ((token
= strsep(&str
, ",")) != NULL
) {
6587 for_each_subsys(ss
, i
) {
6588 if (strcmp(token
, ss
->name
) &&
6589 strcmp(token
, ss
->legacy_name
))
6592 static_branch_disable(cgroup_subsys_enabled_key
[i
]);
6593 pr_info("Disabling %s control group subsystem\n",
6597 for (i
= 0; i
< OPT_FEATURE_COUNT
; i
++) {
6598 if (strcmp(token
, cgroup_opt_feature_names
[i
]))
6600 cgroup_feature_disable_mask
|= 1 << i
;
6601 pr_info("Disabling %s control group feature\n",
6602 cgroup_opt_feature_names
[i
]);
6608 __setup("cgroup_disable=", cgroup_disable
);
6610 void __init __weak
enable_debug_cgroup(void) { }
6612 static int __init
enable_cgroup_debug(char *str
)
6614 cgroup_debug
= true;
6615 enable_debug_cgroup();
6618 __setup("cgroup_debug", enable_cgroup_debug
);
6621 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6622 * @dentry: directory dentry of interest
6623 * @ss: subsystem of interest
6625 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6626 * to get the corresponding css and return it. If such css doesn't exist
6627 * or can't be pinned, an ERR_PTR value is returned.
6629 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6630 struct cgroup_subsys
*ss
)
6632 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6633 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6634 struct cgroup_subsys_state
*css
= NULL
;
6635 struct cgroup
*cgrp
;
6637 /* is @dentry a cgroup dir? */
6638 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6639 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6640 return ERR_PTR(-EBADF
);
6645 * This path doesn't originate from kernfs and @kn could already
6646 * have been or be removed at any point. @kn->priv is RCU
6647 * protected for this access. See css_release_work_fn() for details.
6649 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6651 css
= cgroup_css(cgrp
, ss
);
6653 if (!css
|| !css_tryget_online(css
))
6654 css
= ERR_PTR(-ENOENT
);
6661 * css_from_id - lookup css by id
6662 * @id: the cgroup id
6663 * @ss: cgroup subsys to be looked into
6665 * Returns the css if there's valid one with @id, otherwise returns NULL.
6666 * Should be called under rcu_read_lock().
6668 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6670 WARN_ON_ONCE(!rcu_read_lock_held());
6671 return idr_find(&ss
->css_idr
, id
);
6675 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6676 * @path: path on the default hierarchy
6678 * Find the cgroup at @path on the default hierarchy, increment its
6679 * reference count and return it. Returns pointer to the found cgroup on
6680 * success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already
6681 * been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory.
6683 struct cgroup
*cgroup_get_from_path(const char *path
)
6685 struct kernfs_node
*kn
;
6686 struct cgroup
*cgrp
= ERR_PTR(-ENOENT
);
6687 struct cgroup
*root_cgrp
;
6689 spin_lock_irq(&css_set_lock
);
6690 root_cgrp
= current_cgns_cgroup_from_root(&cgrp_dfl_root
);
6691 kn
= kernfs_walk_and_get(root_cgrp
->kn
, path
);
6692 spin_unlock_irq(&css_set_lock
);
6696 if (kernfs_type(kn
) != KERNFS_DIR
) {
6697 cgrp
= ERR_PTR(-ENOTDIR
);
6703 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6704 if (!cgrp
|| !cgroup_tryget(cgrp
))
6705 cgrp
= ERR_PTR(-ENOENT
);
6714 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6717 * cgroup_get_from_fd - get a cgroup pointer from a fd
6718 * @fd: fd obtained by open(cgroup2_dir)
6720 * Find the cgroup from a fd which should be obtained
6721 * by opening a cgroup directory. Returns a pointer to the
6722 * cgroup on success. ERR_PTR is returned if the cgroup
6725 struct cgroup
*cgroup_get_from_fd(int fd
)
6727 struct cgroup
*cgrp
;
6732 return ERR_PTR(-EBADF
);
6734 cgrp
= cgroup_get_from_file(f
);
6738 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6740 static u64
power_of_ten(int power
)
6749 * cgroup_parse_float - parse a floating number
6750 * @input: input string
6751 * @dec_shift: number of decimal digits to shift
6754 * Parse a decimal floating point number in @input and store the result in
6755 * @v with decimal point right shifted @dec_shift times. For example, if
6756 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6757 * Returns 0 on success, -errno otherwise.
6759 * There's nothing cgroup specific about this function except that it's
6760 * currently the only user.
6762 int cgroup_parse_float(const char *input
, unsigned dec_shift
, s64
*v
)
6764 s64 whole
, frac
= 0;
6765 int fstart
= 0, fend
= 0, flen
;
6767 if (!sscanf(input
, "%lld.%n%lld%n", &whole
, &fstart
, &frac
, &fend
))
6772 flen
= fend
> fstart
? fend
- fstart
: 0;
6773 if (flen
< dec_shift
)
6774 frac
*= power_of_ten(dec_shift
- flen
);
6776 frac
= DIV_ROUND_CLOSEST_ULL(frac
, power_of_ten(flen
- dec_shift
));
6778 *v
= whole
* power_of_ten(dec_shift
) + frac
;
6783 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6784 * definition in cgroup-defs.h.
6786 #ifdef CONFIG_SOCK_CGROUP_DATA
6788 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6790 struct cgroup
*cgroup
;
6793 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6794 if (in_interrupt()) {
6795 cgroup
= &cgrp_dfl_root
.cgrp
;
6801 struct css_set
*cset
;
6803 cset
= task_css_set(current
);
6804 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6805 cgroup
= cset
->dfl_cgrp
;
6811 skcd
->cgroup
= cgroup
;
6812 cgroup_bpf_get(cgroup
);
6816 void cgroup_sk_clone(struct sock_cgroup_data
*skcd
)
6818 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6821 * We might be cloning a socket which is left in an empty
6822 * cgroup and the cgroup might have already been rmdir'd.
6823 * Don't use cgroup_get_live().
6826 cgroup_bpf_get(cgrp
);
6829 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6831 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6833 cgroup_bpf_put(cgrp
);
6837 #endif /* CONFIG_SOCK_CGROUP_DATA */
6840 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
6841 ssize_t size
, const char *prefix
)
6846 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
6847 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
6851 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
6853 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
6855 if (WARN_ON(ret
>= size
))
6862 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6865 struct cgroup_subsys
*ss
;
6869 ret
= show_delegatable_files(cgroup_base_files
, buf
+ ret
,
6870 PAGE_SIZE
- ret
, NULL
);
6871 if (cgroup_psi_enabled())
6872 ret
+= show_delegatable_files(cgroup_psi_files
, buf
+ ret
,
6873 PAGE_SIZE
- ret
, NULL
);
6875 for_each_subsys(ss
, ssid
)
6876 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
6878 cgroup_subsys_name
[ssid
]);
6882 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
6884 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6887 return snprintf(buf
, PAGE_SIZE
,
6890 "memory_localevents\n"
6891 "memory_recursiveprot\n");
6893 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
6895 static struct attribute
*cgroup_sysfs_attrs
[] = {
6896 &cgroup_delegate_attr
.attr
,
6897 &cgroup_features_attr
.attr
,
6901 static const struct attribute_group cgroup_sysfs_attr_group
= {
6902 .attrs
= cgroup_sysfs_attrs
,
6906 static int __init
cgroup_sysfs_init(void)
6908 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
6910 subsys_initcall(cgroup_sysfs_init
);
6912 #endif /* CONFIG_SYSFS */