2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
57 #include <linux/fs_parser.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/psi.h>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/cgroup.h>
65 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
67 /* let's not notify more than 100 times per second */
68 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
71 * To avoid confusing the compiler (and generating warnings) with code
72 * that attempts to access what would be a 0-element array (i.e. sized
73 * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
74 * constant expression can be added.
76 #define CGROUP_HAS_SUBSYS_CONFIG (CGROUP_SUBSYS_COUNT > 0)
79 * cgroup_mutex is the master lock. Any modification to cgroup or its
80 * hierarchy must be performed while holding it.
82 * css_set_lock protects task->cgroups pointer, the list of css_set
83 * objects, and the chain of tasks off each css_set.
85 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
86 * cgroup.h can use them for lockdep annotations.
88 DEFINE_MUTEX(cgroup_mutex
);
89 DEFINE_SPINLOCK(css_set_lock
);
91 #ifdef CONFIG_PROVE_RCU
92 EXPORT_SYMBOL_GPL(cgroup_mutex
);
93 EXPORT_SYMBOL_GPL(css_set_lock
);
96 DEFINE_SPINLOCK(trace_cgroup_path_lock
);
97 char trace_cgroup_path
[TRACE_CGROUP_PATH_LEN
];
98 bool cgroup_debug __read_mostly
;
101 * Protects cgroup_idr and css_idr so that IDs can be released without
102 * grabbing cgroup_mutex.
104 static DEFINE_SPINLOCK(cgroup_idr_lock
);
107 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
108 * against file removal/re-creation across css hiding.
110 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
112 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem
);
114 #define cgroup_assert_mutex_or_rcu_locked() \
115 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
116 !lockdep_is_held(&cgroup_mutex), \
117 "cgroup_mutex or RCU read lock required");
120 * cgroup destruction makes heavy use of work items and there can be a lot
121 * of concurrent destructions. Use a separate workqueue so that cgroup
122 * destruction work items don't end up filling up max_active of system_wq
123 * which may lead to deadlock.
125 static struct workqueue_struct
*cgroup_destroy_wq
;
127 /* generate an array of cgroup subsystem pointers */
128 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
129 struct cgroup_subsys
*cgroup_subsys
[] = {
130 #include <linux/cgroup_subsys.h>
134 /* array of cgroup subsystem names */
135 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
136 static const char *cgroup_subsys_name
[] = {
137 #include <linux/cgroup_subsys.h>
141 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
143 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
144 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
145 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
146 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
147 #include <linux/cgroup_subsys.h>
150 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
151 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
152 #include <linux/cgroup_subsys.h>
156 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
157 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
158 #include <linux/cgroup_subsys.h>
162 static DEFINE_PER_CPU(struct cgroup_rstat_cpu
, cgrp_dfl_root_rstat_cpu
);
164 /* the default hierarchy */
165 struct cgroup_root cgrp_dfl_root
= { .cgrp
.rstat_cpu
= &cgrp_dfl_root_rstat_cpu
};
166 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
169 * The default hierarchy always exists but is hidden until mounted for the
170 * first time. This is for backward compatibility.
172 static bool cgrp_dfl_visible
;
174 /* some controllers are not supported in the default hierarchy */
175 static u16 cgrp_dfl_inhibit_ss_mask
;
177 /* some controllers are implicitly enabled on the default hierarchy */
178 static u16 cgrp_dfl_implicit_ss_mask
;
180 /* some controllers can be threaded on the default hierarchy */
181 static u16 cgrp_dfl_threaded_ss_mask
;
183 /* The list of hierarchy roots */
184 LIST_HEAD(cgroup_roots
);
185 static int cgroup_root_count
;
187 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
188 static DEFINE_IDR(cgroup_hierarchy_idr
);
191 * Assign a monotonically increasing serial number to csses. It guarantees
192 * cgroups with bigger numbers are newer than those with smaller numbers.
193 * Also, as csses are always appended to the parent's ->children list, it
194 * guarantees that sibling csses are always sorted in the ascending serial
195 * number order on the list. Protected by cgroup_mutex.
197 static u64 css_serial_nr_next
= 1;
200 * These bitmasks identify subsystems with specific features to avoid
201 * having to do iterative checks repeatedly.
203 static u16 have_fork_callback __read_mostly
;
204 static u16 have_exit_callback __read_mostly
;
205 static u16 have_release_callback __read_mostly
;
206 static u16 have_canfork_callback __read_mostly
;
208 /* cgroup namespace for init task */
209 struct cgroup_namespace init_cgroup_ns
= {
210 .ns
.count
= REFCOUNT_INIT(2),
211 .user_ns
= &init_user_ns
,
212 .ns
.ops
= &cgroupns_operations
,
213 .ns
.inum
= PROC_CGROUP_INIT_INO
,
214 .root_cset
= &init_css_set
,
217 static struct file_system_type cgroup2_fs_type
;
218 static struct cftype cgroup_base_files
[];
220 /* cgroup optional features */
221 enum cgroup_opt_features
{
223 OPT_FEATURE_PRESSURE
,
228 static const char *cgroup_opt_feature_names
[OPT_FEATURE_COUNT
] = {
234 static u16 cgroup_feature_disable_mask __read_mostly
;
236 static int cgroup_apply_control(struct cgroup
*cgrp
);
237 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
238 static void css_task_iter_skip(struct css_task_iter
*it
,
239 struct task_struct
*task
);
240 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
241 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
242 struct cgroup_subsys
*ss
);
243 static void css_release(struct percpu_ref
*ref
);
244 static void kill_css(struct cgroup_subsys_state
*css
);
245 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
246 struct cgroup
*cgrp
, struct cftype cfts
[],
250 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
251 * @ssid: subsys ID of interest
253 * cgroup_subsys_enabled() can only be used with literal subsys names which
254 * is fine for individual subsystems but unsuitable for cgroup core. This
255 * is slower static_key_enabled() based test indexed by @ssid.
257 bool cgroup_ssid_enabled(int ssid
)
259 if (!CGROUP_HAS_SUBSYS_CONFIG
)
262 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
266 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
267 * @cgrp: the cgroup of interest
269 * The default hierarchy is the v2 interface of cgroup and this function
270 * can be used to test whether a cgroup is on the default hierarchy for
271 * cases where a subsystem should behave differently depending on the
274 * List of changed behaviors:
276 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
277 * and "name" are disallowed.
279 * - When mounting an existing superblock, mount options should match.
281 * - Remount is disallowed.
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_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
768 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
771 * The following field is re-initialized when this cset gets linked
772 * in cgroup_init(). However, let's initialize the field
773 * statically too so that the default cgroup can be accessed safely
776 .dfl_cgrp
= &cgrp_dfl_root
.cgrp
,
779 static int css_set_count
= 1; /* 1 for init_css_set */
781 static bool css_set_threaded(struct css_set
*cset
)
783 return cset
->dom_cset
!= cset
;
787 * css_set_populated - does a css_set contain any tasks?
788 * @cset: target css_set
790 * css_set_populated() should be the same as !!cset->nr_tasks at steady
791 * state. However, css_set_populated() can be called while a task is being
792 * added to or removed from the linked list before the nr_tasks is
793 * properly updated. Hence, we can't just look at ->nr_tasks here.
795 static bool css_set_populated(struct css_set
*cset
)
797 lockdep_assert_held(&css_set_lock
);
799 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
803 * cgroup_update_populated - update the populated count of a cgroup
804 * @cgrp: the target cgroup
805 * @populated: inc or dec populated count
807 * One of the css_sets associated with @cgrp is either getting its first
808 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
809 * count is propagated towards root so that a given cgroup's
810 * nr_populated_children is zero iff none of its descendants contain any
813 * @cgrp's interface file "cgroup.populated" is zero if both
814 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
815 * 1 otherwise. When the sum changes from or to zero, userland is notified
816 * that the content of the interface file has changed. This can be used to
817 * detect when @cgrp and its descendants become populated or empty.
819 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
821 struct cgroup
*child
= NULL
;
822 int adj
= populated
? 1 : -1;
824 lockdep_assert_held(&css_set_lock
);
827 bool was_populated
= cgroup_is_populated(cgrp
);
830 cgrp
->nr_populated_csets
+= adj
;
832 if (cgroup_is_threaded(child
))
833 cgrp
->nr_populated_threaded_children
+= adj
;
835 cgrp
->nr_populated_domain_children
+= adj
;
838 if (was_populated
== cgroup_is_populated(cgrp
))
841 cgroup1_check_for_release(cgrp
);
842 TRACE_CGROUP_PATH(notify_populated
, cgrp
,
843 cgroup_is_populated(cgrp
));
844 cgroup_file_notify(&cgrp
->events_file
);
847 cgrp
= cgroup_parent(cgrp
);
852 * css_set_update_populated - update populated state of a css_set
853 * @cset: target css_set
854 * @populated: whether @cset is populated or depopulated
856 * @cset is either getting the first task or losing the last. Update the
857 * populated counters of all associated cgroups accordingly.
859 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
861 struct cgrp_cset_link
*link
;
863 lockdep_assert_held(&css_set_lock
);
865 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
866 cgroup_update_populated(link
->cgrp
, populated
);
870 * @task is leaving, advance task iterators which are pointing to it so
871 * that they can resume at the next position. Advancing an iterator might
872 * remove it from the list, use safe walk. See css_task_iter_skip() for
875 static void css_set_skip_task_iters(struct css_set
*cset
,
876 struct task_struct
*task
)
878 struct css_task_iter
*it
, *pos
;
880 list_for_each_entry_safe(it
, pos
, &cset
->task_iters
, iters_node
)
881 css_task_iter_skip(it
, task
);
885 * css_set_move_task - move a task from one css_set to another
886 * @task: task being moved
887 * @from_cset: css_set @task currently belongs to (may be NULL)
888 * @to_cset: new css_set @task is being moved to (may be NULL)
889 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
891 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
892 * css_set, @from_cset can be NULL. If @task is being disassociated
893 * instead of moved, @to_cset can be NULL.
895 * This function automatically handles populated counter updates and
896 * css_task_iter adjustments but the caller is responsible for managing
897 * @from_cset and @to_cset's reference counts.
899 static void css_set_move_task(struct task_struct
*task
,
900 struct css_set
*from_cset
, struct css_set
*to_cset
,
903 lockdep_assert_held(&css_set_lock
);
905 if (to_cset
&& !css_set_populated(to_cset
))
906 css_set_update_populated(to_cset
, true);
909 WARN_ON_ONCE(list_empty(&task
->cg_list
));
911 css_set_skip_task_iters(from_cset
, task
);
912 list_del_init(&task
->cg_list
);
913 if (!css_set_populated(from_cset
))
914 css_set_update_populated(from_cset
, false);
916 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
921 * We are synchronized through cgroup_threadgroup_rwsem
922 * against PF_EXITING setting such that we can't race
923 * against cgroup_exit()/cgroup_free() dropping the css_set.
925 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
927 cgroup_move_task(task
, to_cset
);
928 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
934 * hash table for cgroup groups. This improves the performance to find
935 * an existing css_set. This hash doesn't (currently) take into
936 * account cgroups in empty hierarchies.
938 #define CSS_SET_HASH_BITS 7
939 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
941 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
943 unsigned long key
= 0UL;
944 struct cgroup_subsys
*ss
;
947 for_each_subsys(ss
, i
)
948 key
+= (unsigned long)css
[i
];
949 key
= (key
>> 16) ^ key
;
954 void put_css_set_locked(struct css_set
*cset
)
956 struct cgrp_cset_link
*link
, *tmp_link
;
957 struct cgroup_subsys
*ss
;
960 lockdep_assert_held(&css_set_lock
);
962 if (!refcount_dec_and_test(&cset
->refcount
))
965 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
967 /* This css_set is dead. Unlink it and release cgroup and css refs */
968 for_each_subsys(ss
, ssid
) {
969 list_del(&cset
->e_cset_node
[ssid
]);
970 css_put(cset
->subsys
[ssid
]);
972 hash_del(&cset
->hlist
);
975 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
976 list_del(&link
->cset_link
);
977 list_del(&link
->cgrp_link
);
978 if (cgroup_parent(link
->cgrp
))
979 cgroup_put(link
->cgrp
);
983 if (css_set_threaded(cset
)) {
984 list_del(&cset
->threaded_csets_node
);
985 put_css_set_locked(cset
->dom_cset
);
988 kfree_rcu(cset
, rcu_head
);
992 * compare_css_sets - helper function for find_existing_css_set().
993 * @cset: candidate css_set being tested
994 * @old_cset: existing css_set for a task
995 * @new_cgrp: cgroup that's being entered by the task
996 * @template: desired set of css pointers in css_set (pre-calculated)
998 * Returns true if "cset" matches "old_cset" except for the hierarchy
999 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
1001 static bool compare_css_sets(struct css_set
*cset
,
1002 struct css_set
*old_cset
,
1003 struct cgroup
*new_cgrp
,
1004 struct cgroup_subsys_state
*template[])
1006 struct cgroup
*new_dfl_cgrp
;
1007 struct list_head
*l1
, *l2
;
1010 * On the default hierarchy, there can be csets which are
1011 * associated with the same set of cgroups but different csses.
1012 * Let's first ensure that csses match.
1014 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
1018 /* @cset's domain should match the default cgroup's */
1019 if (cgroup_on_dfl(new_cgrp
))
1020 new_dfl_cgrp
= new_cgrp
;
1022 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
1024 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
1028 * Compare cgroup pointers in order to distinguish between
1029 * different cgroups in hierarchies. As different cgroups may
1030 * share the same effective css, this comparison is always
1033 l1
= &cset
->cgrp_links
;
1034 l2
= &old_cset
->cgrp_links
;
1036 struct cgrp_cset_link
*link1
, *link2
;
1037 struct cgroup
*cgrp1
, *cgrp2
;
1041 /* See if we reached the end - both lists are equal length. */
1042 if (l1
== &cset
->cgrp_links
) {
1043 BUG_ON(l2
!= &old_cset
->cgrp_links
);
1046 BUG_ON(l2
== &old_cset
->cgrp_links
);
1048 /* Locate the cgroups associated with these links. */
1049 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
1050 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
1051 cgrp1
= link1
->cgrp
;
1052 cgrp2
= link2
->cgrp
;
1053 /* Hierarchies should be linked in the same order. */
1054 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
1057 * If this hierarchy is the hierarchy of the cgroup
1058 * that's changing, then we need to check that this
1059 * css_set points to the new cgroup; if it's any other
1060 * hierarchy, then this css_set should point to the
1061 * same cgroup as the old css_set.
1063 if (cgrp1
->root
== new_cgrp
->root
) {
1064 if (cgrp1
!= new_cgrp
)
1075 * find_existing_css_set - init css array and find the matching css_set
1076 * @old_cset: the css_set that we're using before the cgroup transition
1077 * @cgrp: the cgroup that we're moving into
1078 * @template: out param for the new set of csses, should be clear on entry
1080 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
1081 struct cgroup
*cgrp
,
1082 struct cgroup_subsys_state
*template[])
1084 struct cgroup_root
*root
= cgrp
->root
;
1085 struct cgroup_subsys
*ss
;
1086 struct css_set
*cset
;
1091 * Build the set of subsystem state objects that we want to see in the
1092 * new css_set. While subsystems can change globally, the entries here
1093 * won't change, so no need for locking.
1095 for_each_subsys(ss
, i
) {
1096 if (root
->subsys_mask
& (1UL << i
)) {
1098 * @ss is in this hierarchy, so we want the
1099 * effective css from @cgrp.
1101 template[i
] = cgroup_e_css_by_mask(cgrp
, ss
);
1104 * @ss is not in this hierarchy, so we don't want
1105 * to change the css.
1107 template[i
] = old_cset
->subsys
[i
];
1111 key
= css_set_hash(template);
1112 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
1113 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
1116 /* This css_set matches what we need */
1120 /* No existing cgroup group matched */
1124 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
1126 struct cgrp_cset_link
*link
, *tmp_link
;
1128 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
1129 list_del(&link
->cset_link
);
1135 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1136 * @count: the number of links to allocate
1137 * @tmp_links: list_head the allocated links are put on
1139 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1140 * through ->cset_link. Returns 0 on success or -errno.
1142 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1144 struct cgrp_cset_link
*link
;
1147 INIT_LIST_HEAD(tmp_links
);
1149 for (i
= 0; i
< count
; i
++) {
1150 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1152 free_cgrp_cset_links(tmp_links
);
1155 list_add(&link
->cset_link
, tmp_links
);
1161 * link_css_set - a helper function to link a css_set to a cgroup
1162 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1163 * @cset: the css_set to be linked
1164 * @cgrp: the destination cgroup
1166 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1167 struct cgroup
*cgrp
)
1169 struct cgrp_cset_link
*link
;
1171 BUG_ON(list_empty(tmp_links
));
1173 if (cgroup_on_dfl(cgrp
))
1174 cset
->dfl_cgrp
= cgrp
;
1176 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1181 * Always add links to the tail of the lists so that the lists are
1182 * in chronological order.
1184 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1185 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1187 if (cgroup_parent(cgrp
))
1188 cgroup_get_live(cgrp
);
1192 * find_css_set - return a new css_set with one cgroup updated
1193 * @old_cset: the baseline css_set
1194 * @cgrp: the cgroup to be updated
1196 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1197 * substituted into the appropriate hierarchy.
1199 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1200 struct cgroup
*cgrp
)
1202 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1203 struct css_set
*cset
;
1204 struct list_head tmp_links
;
1205 struct cgrp_cset_link
*link
;
1206 struct cgroup_subsys
*ss
;
1210 lockdep_assert_held(&cgroup_mutex
);
1212 /* First see if we already have a cgroup group that matches
1213 * the desired set */
1214 spin_lock_irq(&css_set_lock
);
1215 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1218 spin_unlock_irq(&css_set_lock
);
1223 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1227 /* Allocate all the cgrp_cset_link objects that we'll need */
1228 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1233 refcount_set(&cset
->refcount
, 1);
1234 cset
->dom_cset
= cset
;
1235 INIT_LIST_HEAD(&cset
->tasks
);
1236 INIT_LIST_HEAD(&cset
->mg_tasks
);
1237 INIT_LIST_HEAD(&cset
->dying_tasks
);
1238 INIT_LIST_HEAD(&cset
->task_iters
);
1239 INIT_LIST_HEAD(&cset
->threaded_csets
);
1240 INIT_HLIST_NODE(&cset
->hlist
);
1241 INIT_LIST_HEAD(&cset
->cgrp_links
);
1242 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1243 INIT_LIST_HEAD(&cset
->mg_node
);
1245 /* Copy the set of subsystem state objects generated in
1246 * find_existing_css_set() */
1247 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1249 spin_lock_irq(&css_set_lock
);
1250 /* Add reference counts and links from the new css_set. */
1251 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1252 struct cgroup
*c
= link
->cgrp
;
1254 if (c
->root
== cgrp
->root
)
1256 link_css_set(&tmp_links
, cset
, c
);
1259 BUG_ON(!list_empty(&tmp_links
));
1263 /* Add @cset to the hash table */
1264 key
= css_set_hash(cset
->subsys
);
1265 hash_add(css_set_table
, &cset
->hlist
, key
);
1267 for_each_subsys(ss
, ssid
) {
1268 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1270 list_add_tail(&cset
->e_cset_node
[ssid
],
1271 &css
->cgroup
->e_csets
[ssid
]);
1275 spin_unlock_irq(&css_set_lock
);
1278 * If @cset should be threaded, look up the matching dom_cset and
1279 * link them up. We first fully initialize @cset then look for the
1280 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1281 * to stay empty until we return.
1283 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1284 struct css_set
*dcset
;
1286 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1292 spin_lock_irq(&css_set_lock
);
1293 cset
->dom_cset
= dcset
;
1294 list_add_tail(&cset
->threaded_csets_node
,
1295 &dcset
->threaded_csets
);
1296 spin_unlock_irq(&css_set_lock
);
1302 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1304 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1306 return root_cgrp
->root
;
1309 static int cgroup_init_root_id(struct cgroup_root
*root
)
1313 lockdep_assert_held(&cgroup_mutex
);
1315 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1319 root
->hierarchy_id
= id
;
1323 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1325 lockdep_assert_held(&cgroup_mutex
);
1327 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1330 void cgroup_free_root(struct cgroup_root
*root
)
1335 static void cgroup_destroy_root(struct cgroup_root
*root
)
1337 struct cgroup
*cgrp
= &root
->cgrp
;
1338 struct cgrp_cset_link
*link
, *tmp_link
;
1340 trace_cgroup_destroy_root(root
);
1342 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1344 BUG_ON(atomic_read(&root
->nr_cgrps
));
1345 BUG_ON(!list_empty(&cgrp
->self
.children
));
1347 /* Rebind all subsystems back to the default hierarchy */
1348 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1351 * Release all the links from cset_links to this hierarchy's
1354 spin_lock_irq(&css_set_lock
);
1356 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1357 list_del(&link
->cset_link
);
1358 list_del(&link
->cgrp_link
);
1362 spin_unlock_irq(&css_set_lock
);
1364 if (!list_empty(&root
->root_list
)) {
1365 list_del(&root
->root_list
);
1366 cgroup_root_count
--;
1369 cgroup_exit_root_id(root
);
1371 mutex_unlock(&cgroup_mutex
);
1373 cgroup_rstat_exit(cgrp
);
1374 kernfs_destroy_root(root
->kf_root
);
1375 cgroup_free_root(root
);
1379 * look up cgroup associated with current task's cgroup namespace on the
1380 * specified hierarchy
1382 static struct cgroup
*
1383 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1385 struct cgroup
*res
= NULL
;
1386 struct css_set
*cset
;
1388 lockdep_assert_held(&css_set_lock
);
1392 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1393 if (cset
== &init_css_set
) {
1395 } else if (root
== &cgrp_dfl_root
) {
1396 res
= cset
->dfl_cgrp
;
1398 struct cgrp_cset_link
*link
;
1400 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1401 struct cgroup
*c
= link
->cgrp
;
1403 if (c
->root
== root
) {
1415 /* look up cgroup associated with given css_set on the specified hierarchy */
1416 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1417 struct cgroup_root
*root
)
1419 struct cgroup
*res
= NULL
;
1421 lockdep_assert_held(&cgroup_mutex
);
1422 lockdep_assert_held(&css_set_lock
);
1424 if (cset
== &init_css_set
) {
1426 } else if (root
== &cgrp_dfl_root
) {
1427 res
= cset
->dfl_cgrp
;
1429 struct cgrp_cset_link
*link
;
1431 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1432 struct cgroup
*c
= link
->cgrp
;
1434 if (c
->root
== root
) {
1446 * Return the cgroup for "task" from the given hierarchy. Must be
1447 * called with cgroup_mutex and css_set_lock held.
1449 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1450 struct cgroup_root
*root
)
1453 * No need to lock the task - since we hold css_set_lock the
1454 * task can't change groups.
1456 return cset_cgroup_from_root(task_css_set(task
), root
);
1460 * A task must hold cgroup_mutex to modify cgroups.
1462 * Any task can increment and decrement the count field without lock.
1463 * So in general, code holding cgroup_mutex can't rely on the count
1464 * field not changing. However, if the count goes to zero, then only
1465 * cgroup_attach_task() can increment it again. Because a count of zero
1466 * means that no tasks are currently attached, therefore there is no
1467 * way a task attached to that cgroup can fork (the other way to
1468 * increment the count). So code holding cgroup_mutex can safely
1469 * assume that if the count is zero, it will stay zero. Similarly, if
1470 * a task holds cgroup_mutex on a cgroup with zero count, it
1471 * knows that the cgroup won't be removed, as cgroup_rmdir()
1474 * A cgroup can only be deleted if both its 'count' of using tasks
1475 * is zero, and its list of 'children' cgroups is empty. Since all
1476 * tasks in the system use _some_ cgroup, and since there is always at
1477 * least one task in the system (init, pid == 1), therefore, root cgroup
1478 * always has either children cgroups and/or using tasks. So we don't
1479 * need a special hack to ensure that root cgroup cannot be deleted.
1481 * P.S. One more locking exception. RCU is used to guard the
1482 * update of a tasks cgroup pointer by cgroup_attach_task()
1485 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1487 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1490 struct cgroup_subsys
*ss
= cft
->ss
;
1492 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1493 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
1494 const char *dbg
= (cft
->flags
& CFTYPE_DEBUG
) ? ".__DEBUG__." : "";
1496 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s%s.%s",
1497 dbg
, cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1500 strscpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1506 * cgroup_file_mode - deduce file mode of a control file
1507 * @cft: the control file in question
1509 * S_IRUGO for read, S_IWUSR for write.
1511 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1515 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1518 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1519 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1529 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1530 * @subtree_control: the new subtree_control mask to consider
1531 * @this_ss_mask: available subsystems
1533 * On the default hierarchy, a subsystem may request other subsystems to be
1534 * enabled together through its ->depends_on mask. In such cases, more
1535 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1537 * This function calculates which subsystems need to be enabled if
1538 * @subtree_control is to be applied while restricted to @this_ss_mask.
1540 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1542 u16 cur_ss_mask
= subtree_control
;
1543 struct cgroup_subsys
*ss
;
1546 lockdep_assert_held(&cgroup_mutex
);
1548 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1551 u16 new_ss_mask
= cur_ss_mask
;
1553 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1554 new_ss_mask
|= ss
->depends_on
;
1555 } while_each_subsys_mask();
1558 * Mask out subsystems which aren't available. This can
1559 * happen only if some depended-upon subsystems were bound
1560 * to non-default hierarchies.
1562 new_ss_mask
&= this_ss_mask
;
1564 if (new_ss_mask
== cur_ss_mask
)
1566 cur_ss_mask
= new_ss_mask
;
1573 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1574 * @kn: the kernfs_node being serviced
1576 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1577 * the method finishes if locking succeeded. Note that once this function
1578 * returns the cgroup returned by cgroup_kn_lock_live() may become
1579 * inaccessible any time. If the caller intends to continue to access the
1580 * cgroup, it should pin it before invoking this function.
1582 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1584 struct cgroup
*cgrp
;
1586 if (kernfs_type(kn
) == KERNFS_DIR
)
1589 cgrp
= kn
->parent
->priv
;
1591 mutex_unlock(&cgroup_mutex
);
1593 kernfs_unbreak_active_protection(kn
);
1598 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1599 * @kn: the kernfs_node being serviced
1600 * @drain_offline: perform offline draining on the cgroup
1602 * This helper is to be used by a cgroup kernfs method currently servicing
1603 * @kn. It breaks the active protection, performs cgroup locking and
1604 * verifies that the associated cgroup is alive. Returns the cgroup if
1605 * alive; otherwise, %NULL. A successful return should be undone by a
1606 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1607 * cgroup is drained of offlining csses before return.
1609 * Any cgroup kernfs method implementation which requires locking the
1610 * associated cgroup should use this helper. It avoids nesting cgroup
1611 * locking under kernfs active protection and allows all kernfs operations
1612 * including self-removal.
1614 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1616 struct cgroup
*cgrp
;
1618 if (kernfs_type(kn
) == KERNFS_DIR
)
1621 cgrp
= kn
->parent
->priv
;
1624 * We're gonna grab cgroup_mutex which nests outside kernfs
1625 * active_ref. cgroup liveliness check alone provides enough
1626 * protection against removal. Ensure @cgrp stays accessible and
1627 * break the active_ref protection.
1629 if (!cgroup_tryget(cgrp
))
1631 kernfs_break_active_protection(kn
);
1634 cgroup_lock_and_drain_offline(cgrp
);
1636 mutex_lock(&cgroup_mutex
);
1638 if (!cgroup_is_dead(cgrp
))
1641 cgroup_kn_unlock(kn
);
1645 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1647 char name
[CGROUP_FILE_NAME_MAX
];
1649 lockdep_assert_held(&cgroup_mutex
);
1651 if (cft
->file_offset
) {
1652 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1653 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1655 spin_lock_irq(&cgroup_file_kn_lock
);
1657 spin_unlock_irq(&cgroup_file_kn_lock
);
1659 del_timer_sync(&cfile
->notify_timer
);
1662 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1666 * css_clear_dir - remove subsys files in a cgroup directory
1669 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1671 struct cgroup
*cgrp
= css
->cgroup
;
1672 struct cftype
*cfts
;
1674 if (!(css
->flags
& CSS_VISIBLE
))
1677 css
->flags
&= ~CSS_VISIBLE
;
1680 if (cgroup_on_dfl(cgrp
))
1681 cfts
= cgroup_base_files
;
1683 cfts
= cgroup1_base_files
;
1685 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1687 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1688 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1693 * css_populate_dir - create subsys files in a cgroup directory
1696 * On failure, no file is added.
1698 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1700 struct cgroup
*cgrp
= css
->cgroup
;
1701 struct cftype
*cfts
, *failed_cfts
;
1704 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1708 if (cgroup_on_dfl(cgrp
))
1709 cfts
= cgroup_base_files
;
1711 cfts
= cgroup1_base_files
;
1713 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1717 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1718 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1726 css
->flags
|= CSS_VISIBLE
;
1730 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1731 if (cfts
== failed_cfts
)
1733 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1738 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1740 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1741 struct cgroup_subsys
*ss
;
1744 lockdep_assert_held(&cgroup_mutex
);
1746 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1748 * If @ss has non-root csses attached to it, can't move.
1749 * If @ss is an implicit controller, it is exempt from this
1750 * rule and can be stolen.
1752 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1753 !ss
->implicit_on_dfl
)
1756 /* can't move between two non-dummy roots either */
1757 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1759 } while_each_subsys_mask();
1761 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1762 struct cgroup_root
*src_root
= ss
->root
;
1763 struct cgroup
*scgrp
= &src_root
->cgrp
;
1764 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1765 struct css_set
*cset
;
1767 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1769 /* disable from the source */
1770 src_root
->subsys_mask
&= ~(1 << ssid
);
1771 WARN_ON(cgroup_apply_control(scgrp
));
1772 cgroup_finalize_control(scgrp
, 0);
1775 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1776 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1777 ss
->root
= dst_root
;
1778 css
->cgroup
= dcgrp
;
1780 spin_lock_irq(&css_set_lock
);
1781 hash_for_each(css_set_table
, i
, cset
, hlist
)
1782 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1783 &dcgrp
->e_csets
[ss
->id
]);
1784 spin_unlock_irq(&css_set_lock
);
1786 if (ss
->css_rstat_flush
) {
1787 list_del_rcu(&css
->rstat_css_node
);
1788 list_add_rcu(&css
->rstat_css_node
,
1789 &dcgrp
->rstat_css_list
);
1792 /* default hierarchy doesn't enable controllers by default */
1793 dst_root
->subsys_mask
|= 1 << ssid
;
1794 if (dst_root
== &cgrp_dfl_root
) {
1795 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1797 dcgrp
->subtree_control
|= 1 << ssid
;
1798 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1801 ret
= cgroup_apply_control(dcgrp
);
1803 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1808 } while_each_subsys_mask();
1810 kernfs_activate(dcgrp
->kn
);
1814 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1815 struct kernfs_root
*kf_root
)
1819 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1820 struct cgroup
*ns_cgroup
;
1822 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1826 spin_lock_irq(&css_set_lock
);
1827 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1828 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1829 spin_unlock_irq(&css_set_lock
);
1831 if (len
>= PATH_MAX
)
1834 seq_escape(sf
, buf
, " \t\n\\");
1841 enum cgroup2_param
{
1843 Opt_memory_localevents
,
1844 Opt_memory_recursiveprot
,
1848 static const struct fs_parameter_spec cgroup2_fs_parameters
[] = {
1849 fsparam_flag("nsdelegate", Opt_nsdelegate
),
1850 fsparam_flag("memory_localevents", Opt_memory_localevents
),
1851 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot
),
1855 static int cgroup2_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
1857 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1858 struct fs_parse_result result
;
1861 opt
= fs_parse(fc
, cgroup2_fs_parameters
, param
, &result
);
1866 case Opt_nsdelegate
:
1867 ctx
->flags
|= CGRP_ROOT_NS_DELEGATE
;
1869 case Opt_memory_localevents
:
1870 ctx
->flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1872 case Opt_memory_recursiveprot
:
1873 ctx
->flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1879 static void apply_cgroup_root_flags(unsigned int root_flags
)
1881 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1882 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1883 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1885 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1887 if (root_flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1888 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1890 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS
;
1892 if (root_flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1893 cgrp_dfl_root
.flags
|= CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1895 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT
;
1899 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1901 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1902 seq_puts(seq
, ",nsdelegate");
1903 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_LOCAL_EVENTS
)
1904 seq_puts(seq
, ",memory_localevents");
1905 if (cgrp_dfl_root
.flags
& CGRP_ROOT_MEMORY_RECURSIVE_PROT
)
1906 seq_puts(seq
, ",memory_recursiveprot");
1910 static int cgroup_reconfigure(struct fs_context
*fc
)
1912 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1914 apply_cgroup_root_flags(ctx
->flags
);
1918 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1920 struct cgroup_subsys
*ss
;
1923 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1924 INIT_LIST_HEAD(&cgrp
->self
.children
);
1925 INIT_LIST_HEAD(&cgrp
->cset_links
);
1926 INIT_LIST_HEAD(&cgrp
->pidlists
);
1927 mutex_init(&cgrp
->pidlist_mutex
);
1928 cgrp
->self
.cgroup
= cgrp
;
1929 cgrp
->self
.flags
|= CSS_ONLINE
;
1930 cgrp
->dom_cgrp
= cgrp
;
1931 cgrp
->max_descendants
= INT_MAX
;
1932 cgrp
->max_depth
= INT_MAX
;
1933 INIT_LIST_HEAD(&cgrp
->rstat_css_list
);
1934 prev_cputime_init(&cgrp
->prev_cputime
);
1936 for_each_subsys(ss
, ssid
)
1937 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1939 init_waitqueue_head(&cgrp
->offline_waitq
);
1940 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1943 void init_cgroup_root(struct cgroup_fs_context
*ctx
)
1945 struct cgroup_root
*root
= ctx
->root
;
1946 struct cgroup
*cgrp
= &root
->cgrp
;
1948 INIT_LIST_HEAD(&root
->root_list
);
1949 atomic_set(&root
->nr_cgrps
, 1);
1951 init_cgroup_housekeeping(cgrp
);
1953 root
->flags
= ctx
->flags
;
1954 if (ctx
->release_agent
)
1955 strscpy(root
->release_agent_path
, ctx
->release_agent
, PATH_MAX
);
1957 strscpy(root
->name
, ctx
->name
, MAX_CGROUP_ROOT_NAMELEN
);
1958 if (ctx
->cpuset_clone_children
)
1959 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1962 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
1964 LIST_HEAD(tmp_links
);
1965 struct cgroup
*root_cgrp
= &root
->cgrp
;
1966 struct kernfs_syscall_ops
*kf_sops
;
1967 struct css_set
*cset
;
1970 lockdep_assert_held(&cgroup_mutex
);
1972 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
1978 * We're accessing css_set_count without locking css_set_lock here,
1979 * but that's OK - it can only be increased by someone holding
1980 * cgroup_lock, and that's us. Later rebinding may disable
1981 * controllers on the default hierarchy and thus create new csets,
1982 * which can't be more than the existing ones. Allocate 2x.
1984 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
1988 ret
= cgroup_init_root_id(root
);
1992 kf_sops
= root
== &cgrp_dfl_root
?
1993 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
1995 root
->kf_root
= kernfs_create_root(kf_sops
,
1996 KERNFS_ROOT_CREATE_DEACTIVATED
|
1997 KERNFS_ROOT_SUPPORT_EXPORTOP
|
1998 KERNFS_ROOT_SUPPORT_USER_XATTR
,
2000 if (IS_ERR(root
->kf_root
)) {
2001 ret
= PTR_ERR(root
->kf_root
);
2004 root_cgrp
->kn
= root
->kf_root
->kn
;
2005 WARN_ON_ONCE(cgroup_ino(root_cgrp
) != 1);
2006 root_cgrp
->ancestor_ids
[0] = cgroup_id(root_cgrp
);
2008 ret
= css_populate_dir(&root_cgrp
->self
);
2012 ret
= cgroup_rstat_init(root_cgrp
);
2016 ret
= rebind_subsystems(root
, ss_mask
);
2020 ret
= cgroup_bpf_inherit(root_cgrp
);
2023 trace_cgroup_setup_root(root
);
2026 * There must be no failure case after here, since rebinding takes
2027 * care of subsystems' refcounts, which are explicitly dropped in
2028 * the failure exit path.
2030 list_add(&root
->root_list
, &cgroup_roots
);
2031 cgroup_root_count
++;
2034 * Link the root cgroup in this hierarchy into all the css_set
2037 spin_lock_irq(&css_set_lock
);
2038 hash_for_each(css_set_table
, i
, cset
, hlist
) {
2039 link_css_set(&tmp_links
, cset
, root_cgrp
);
2040 if (css_set_populated(cset
))
2041 cgroup_update_populated(root_cgrp
, true);
2043 spin_unlock_irq(&css_set_lock
);
2045 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2046 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2052 cgroup_rstat_exit(root_cgrp
);
2054 kernfs_destroy_root(root
->kf_root
);
2055 root
->kf_root
= NULL
;
2057 cgroup_exit_root_id(root
);
2059 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2061 free_cgrp_cset_links(&tmp_links
);
2065 int cgroup_do_get_tree(struct fs_context
*fc
)
2067 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2070 ctx
->kfc
.root
= ctx
->root
->kf_root
;
2071 if (fc
->fs_type
== &cgroup2_fs_type
)
2072 ctx
->kfc
.magic
= CGROUP2_SUPER_MAGIC
;
2074 ctx
->kfc
.magic
= CGROUP_SUPER_MAGIC
;
2075 ret
= kernfs_get_tree(fc
);
2078 * In non-init cgroup namespace, instead of root cgroup's dentry,
2079 * we return the dentry corresponding to the cgroupns->root_cgrp.
2081 if (!ret
&& ctx
->ns
!= &init_cgroup_ns
) {
2082 struct dentry
*nsdentry
;
2083 struct super_block
*sb
= fc
->root
->d_sb
;
2084 struct cgroup
*cgrp
;
2086 mutex_lock(&cgroup_mutex
);
2087 spin_lock_irq(&css_set_lock
);
2089 cgrp
= cset_cgroup_from_root(ctx
->ns
->root_cset
, ctx
->root
);
2091 spin_unlock_irq(&css_set_lock
);
2092 mutex_unlock(&cgroup_mutex
);
2094 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
2096 if (IS_ERR(nsdentry
)) {
2097 deactivate_locked_super(sb
);
2098 ret
= PTR_ERR(nsdentry
);
2101 fc
->root
= nsdentry
;
2104 if (!ctx
->kfc
.new_sb_created
)
2105 cgroup_put(&ctx
->root
->cgrp
);
2111 * Destroy a cgroup filesystem context.
2113 static void cgroup_fs_context_free(struct fs_context
*fc
)
2115 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2118 kfree(ctx
->release_agent
);
2119 put_cgroup_ns(ctx
->ns
);
2120 kernfs_free_fs_context(fc
);
2124 static int cgroup_get_tree(struct fs_context
*fc
)
2126 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
2129 cgrp_dfl_visible
= true;
2130 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2131 ctx
->root
= &cgrp_dfl_root
;
2133 ret
= cgroup_do_get_tree(fc
);
2135 apply_cgroup_root_flags(ctx
->flags
);
2139 static const struct fs_context_operations cgroup_fs_context_ops
= {
2140 .free
= cgroup_fs_context_free
,
2141 .parse_param
= cgroup2_parse_param
,
2142 .get_tree
= cgroup_get_tree
,
2143 .reconfigure
= cgroup_reconfigure
,
2146 static const struct fs_context_operations cgroup1_fs_context_ops
= {
2147 .free
= cgroup_fs_context_free
,
2148 .parse_param
= cgroup1_parse_param
,
2149 .get_tree
= cgroup1_get_tree
,
2150 .reconfigure
= cgroup1_reconfigure
,
2154 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2155 * we select the namespace we're going to use.
2157 static int cgroup_init_fs_context(struct fs_context
*fc
)
2159 struct cgroup_fs_context
*ctx
;
2161 ctx
= kzalloc(sizeof(struct cgroup_fs_context
), GFP_KERNEL
);
2165 ctx
->ns
= current
->nsproxy
->cgroup_ns
;
2166 get_cgroup_ns(ctx
->ns
);
2167 fc
->fs_private
= &ctx
->kfc
;
2168 if (fc
->fs_type
== &cgroup2_fs_type
)
2169 fc
->ops
= &cgroup_fs_context_ops
;
2171 fc
->ops
= &cgroup1_fs_context_ops
;
2172 put_user_ns(fc
->user_ns
);
2173 fc
->user_ns
= get_user_ns(ctx
->ns
->user_ns
);
2178 static void cgroup_kill_sb(struct super_block
*sb
)
2180 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2181 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2184 * If @root doesn't have any children, start killing it.
2185 * This prevents new mounts by disabling percpu_ref_tryget_live().
2187 * And don't kill the default root.
2189 if (list_empty(&root
->cgrp
.self
.children
) && root
!= &cgrp_dfl_root
&&
2190 !percpu_ref_is_dying(&root
->cgrp
.self
.refcnt
))
2191 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2192 cgroup_put(&root
->cgrp
);
2196 struct file_system_type cgroup_fs_type
= {
2198 .init_fs_context
= cgroup_init_fs_context
,
2199 .parameters
= cgroup1_fs_parameters
,
2200 .kill_sb
= cgroup_kill_sb
,
2201 .fs_flags
= FS_USERNS_MOUNT
,
2204 static struct file_system_type cgroup2_fs_type
= {
2206 .init_fs_context
= cgroup_init_fs_context
,
2207 .parameters
= cgroup2_fs_parameters
,
2208 .kill_sb
= cgroup_kill_sb
,
2209 .fs_flags
= FS_USERNS_MOUNT
,
2212 #ifdef CONFIG_CPUSETS
2213 static const struct fs_context_operations cpuset_fs_context_ops
= {
2214 .get_tree
= cgroup1_get_tree
,
2215 .free
= cgroup_fs_context_free
,
2219 * This is ugly, but preserves the userspace API for existing cpuset
2220 * users. If someone tries to mount the "cpuset" filesystem, we
2221 * silently switch it to mount "cgroup" instead
2223 static int cpuset_init_fs_context(struct fs_context
*fc
)
2225 char *agent
= kstrdup("/sbin/cpuset_release_agent", GFP_USER
);
2226 struct cgroup_fs_context
*ctx
;
2229 err
= cgroup_init_fs_context(fc
);
2235 fc
->ops
= &cpuset_fs_context_ops
;
2237 ctx
= cgroup_fc2context(fc
);
2238 ctx
->subsys_mask
= 1 << cpuset_cgrp_id
;
2239 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
2240 ctx
->release_agent
= agent
;
2242 get_filesystem(&cgroup_fs_type
);
2243 put_filesystem(fc
->fs_type
);
2244 fc
->fs_type
= &cgroup_fs_type
;
2249 static struct file_system_type cpuset_fs_type
= {
2251 .init_fs_context
= cpuset_init_fs_context
,
2252 .fs_flags
= FS_USERNS_MOUNT
,
2256 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2257 struct cgroup_namespace
*ns
)
2259 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2261 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2264 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2265 struct cgroup_namespace
*ns
)
2269 mutex_lock(&cgroup_mutex
);
2270 spin_lock_irq(&css_set_lock
);
2272 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2274 spin_unlock_irq(&css_set_lock
);
2275 mutex_unlock(&cgroup_mutex
);
2279 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2282 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2283 * @task: target task
2284 * @buf: the buffer to write the path into
2285 * @buflen: the length of the buffer
2287 * Determine @task's cgroup on the first (the one with the lowest non-zero
2288 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2289 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2290 * cgroup controller callbacks.
2292 * Return value is the same as kernfs_path().
2294 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2296 struct cgroup_root
*root
;
2297 struct cgroup
*cgrp
;
2298 int hierarchy_id
= 1;
2301 mutex_lock(&cgroup_mutex
);
2302 spin_lock_irq(&css_set_lock
);
2304 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2307 cgrp
= task_cgroup_from_root(task
, root
);
2308 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2310 /* if no hierarchy exists, everyone is in "/" */
2311 ret
= strlcpy(buf
, "/", buflen
);
2314 spin_unlock_irq(&css_set_lock
);
2315 mutex_unlock(&cgroup_mutex
);
2318 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2321 * cgroup_migrate_add_task - add a migration target task to a migration context
2322 * @task: target task
2323 * @mgctx: target migration context
2325 * Add @task, which is a migration target, to @mgctx->tset. This function
2326 * becomes noop if @task doesn't need to be migrated. @task's css_set
2327 * should have been added as a migration source and @task->cg_list will be
2328 * moved from the css_set's tasks list to mg_tasks one.
2330 static void cgroup_migrate_add_task(struct task_struct
*task
,
2331 struct cgroup_mgctx
*mgctx
)
2333 struct css_set
*cset
;
2335 lockdep_assert_held(&css_set_lock
);
2337 /* @task either already exited or can't exit until the end */
2338 if (task
->flags
& PF_EXITING
)
2341 /* cgroup_threadgroup_rwsem protects racing against forks */
2342 WARN_ON_ONCE(list_empty(&task
->cg_list
));
2344 cset
= task_css_set(task
);
2345 if (!cset
->mg_src_cgrp
)
2348 mgctx
->tset
.nr_tasks
++;
2350 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2351 if (list_empty(&cset
->mg_node
))
2352 list_add_tail(&cset
->mg_node
,
2353 &mgctx
->tset
.src_csets
);
2354 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2355 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2356 &mgctx
->tset
.dst_csets
);
2360 * cgroup_taskset_first - reset taskset and return the first task
2361 * @tset: taskset of interest
2362 * @dst_cssp: output variable for the destination css
2364 * @tset iteration is initialized and the first task is returned.
2366 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2367 struct cgroup_subsys_state
**dst_cssp
)
2369 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2370 tset
->cur_task
= NULL
;
2372 return cgroup_taskset_next(tset
, dst_cssp
);
2376 * cgroup_taskset_next - iterate to the next task in taskset
2377 * @tset: taskset of interest
2378 * @dst_cssp: output variable for the destination css
2380 * Return the next task in @tset. Iteration must have been initialized
2381 * with cgroup_taskset_first().
2383 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2384 struct cgroup_subsys_state
**dst_cssp
)
2386 struct css_set
*cset
= tset
->cur_cset
;
2387 struct task_struct
*task
= tset
->cur_task
;
2389 while (CGROUP_HAS_SUBSYS_CONFIG
&& &cset
->mg_node
!= tset
->csets
) {
2391 task
= list_first_entry(&cset
->mg_tasks
,
2392 struct task_struct
, cg_list
);
2394 task
= list_next_entry(task
, cg_list
);
2396 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2397 tset
->cur_cset
= cset
;
2398 tset
->cur_task
= task
;
2401 * This function may be called both before and
2402 * after cgroup_taskset_migrate(). The two cases
2403 * can be distinguished by looking at whether @cset
2404 * has its ->mg_dst_cset set.
2406 if (cset
->mg_dst_cset
)
2407 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2409 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2414 cset
= list_next_entry(cset
, mg_node
);
2422 * cgroup_migrate_execute - migrate a taskset
2423 * @mgctx: migration context
2425 * Migrate tasks in @mgctx as setup by migration preparation functions.
2426 * This function fails iff one of the ->can_attach callbacks fails and
2427 * guarantees that either all or none of the tasks in @mgctx are migrated.
2428 * @mgctx is consumed regardless of success.
2430 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2432 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2433 struct cgroup_subsys
*ss
;
2434 struct task_struct
*task
, *tmp_task
;
2435 struct css_set
*cset
, *tmp_cset
;
2436 int ssid
, failed_ssid
, ret
;
2438 /* check that we can legitimately attach to the cgroup */
2439 if (tset
->nr_tasks
) {
2440 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2441 if (ss
->can_attach
) {
2443 ret
= ss
->can_attach(tset
);
2446 goto out_cancel_attach
;
2449 } while_each_subsys_mask();
2453 * Now that we're guaranteed success, proceed to move all tasks to
2454 * the new cgroup. There are no failure cases after here, so this
2455 * is the commit point.
2457 spin_lock_irq(&css_set_lock
);
2458 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2459 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2460 struct css_set
*from_cset
= task_css_set(task
);
2461 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2463 get_css_set(to_cset
);
2464 to_cset
->nr_tasks
++;
2465 css_set_move_task(task
, from_cset
, to_cset
, true);
2466 from_cset
->nr_tasks
--;
2468 * If the source or destination cgroup is frozen,
2469 * the task might require to change its state.
2471 cgroup_freezer_migrate_task(task
, from_cset
->dfl_cgrp
,
2473 put_css_set_locked(from_cset
);
2477 spin_unlock_irq(&css_set_lock
);
2480 * Migration is committed, all target tasks are now on dst_csets.
2481 * Nothing is sensitive to fork() after this point. Notify
2482 * controllers that migration is complete.
2484 tset
->csets
= &tset
->dst_csets
;
2486 if (tset
->nr_tasks
) {
2487 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2492 } while_each_subsys_mask();
2496 goto out_release_tset
;
2499 if (tset
->nr_tasks
) {
2500 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2501 if (ssid
== failed_ssid
)
2503 if (ss
->cancel_attach
) {
2505 ss
->cancel_attach(tset
);
2507 } while_each_subsys_mask();
2510 spin_lock_irq(&css_set_lock
);
2511 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2512 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2513 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2514 list_del_init(&cset
->mg_node
);
2516 spin_unlock_irq(&css_set_lock
);
2519 * Re-initialize the cgroup_taskset structure in case it is reused
2520 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2524 tset
->csets
= &tset
->src_csets
;
2529 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2530 * @dst_cgrp: destination cgroup to test
2532 * On the default hierarchy, except for the mixable, (possible) thread root
2533 * and threaded cgroups, subtree_control must be zero for migration
2534 * destination cgroups with tasks so that child cgroups don't compete
2537 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2539 /* v1 doesn't have any restriction */
2540 if (!cgroup_on_dfl(dst_cgrp
))
2543 /* verify @dst_cgrp can host resources */
2544 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2547 /* mixables don't care */
2548 if (cgroup_is_mixable(dst_cgrp
))
2552 * If @dst_cgrp is already or can become a thread root or is
2553 * threaded, it doesn't matter.
2555 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2558 /* apply no-internal-process constraint */
2559 if (dst_cgrp
->subtree_control
)
2566 * cgroup_migrate_finish - cleanup after attach
2567 * @mgctx: migration context
2569 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2570 * those functions for details.
2572 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2574 LIST_HEAD(preloaded
);
2575 struct css_set
*cset
, *tmp_cset
;
2577 lockdep_assert_held(&cgroup_mutex
);
2579 spin_lock_irq(&css_set_lock
);
2581 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2582 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2584 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2585 cset
->mg_src_cgrp
= NULL
;
2586 cset
->mg_dst_cgrp
= NULL
;
2587 cset
->mg_dst_cset
= NULL
;
2588 list_del_init(&cset
->mg_preload_node
);
2589 put_css_set_locked(cset
);
2592 spin_unlock_irq(&css_set_lock
);
2596 * cgroup_migrate_add_src - add a migration source css_set
2597 * @src_cset: the source css_set to add
2598 * @dst_cgrp: the destination cgroup
2599 * @mgctx: migration context
2601 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2602 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2603 * up by cgroup_migrate_finish().
2605 * This function may be called without holding cgroup_threadgroup_rwsem
2606 * even if the target is a process. Threads may be created and destroyed
2607 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2608 * into play and the preloaded css_sets are guaranteed to cover all
2611 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2612 struct cgroup
*dst_cgrp
,
2613 struct cgroup_mgctx
*mgctx
)
2615 struct cgroup
*src_cgrp
;
2617 lockdep_assert_held(&cgroup_mutex
);
2618 lockdep_assert_held(&css_set_lock
);
2621 * If ->dead, @src_set is associated with one or more dead cgroups
2622 * and doesn't contain any migratable tasks. Ignore it early so
2623 * that the rest of migration path doesn't get confused by it.
2628 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2630 if (!list_empty(&src_cset
->mg_preload_node
))
2633 WARN_ON(src_cset
->mg_src_cgrp
);
2634 WARN_ON(src_cset
->mg_dst_cgrp
);
2635 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2636 WARN_ON(!list_empty(&src_cset
->mg_node
));
2638 src_cset
->mg_src_cgrp
= src_cgrp
;
2639 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2640 get_css_set(src_cset
);
2641 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2645 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2646 * @mgctx: migration context
2648 * Tasks are about to be moved and all the source css_sets have been
2649 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2650 * pins all destination css_sets, links each to its source, and append them
2651 * to @mgctx->preloaded_dst_csets.
2653 * This function must be called after cgroup_migrate_add_src() has been
2654 * called on each migration source css_set. After migration is performed
2655 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2658 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2660 struct css_set
*src_cset
, *tmp_cset
;
2662 lockdep_assert_held(&cgroup_mutex
);
2664 /* look up the dst cset for each src cset and link it to src */
2665 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2667 struct css_set
*dst_cset
;
2668 struct cgroup_subsys
*ss
;
2671 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2675 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2678 * If src cset equals dst, it's noop. Drop the src.
2679 * cgroup_migrate() will skip the cset too. Note that we
2680 * can't handle src == dst as some nodes are used by both.
2682 if (src_cset
== dst_cset
) {
2683 src_cset
->mg_src_cgrp
= NULL
;
2684 src_cset
->mg_dst_cgrp
= NULL
;
2685 list_del_init(&src_cset
->mg_preload_node
);
2686 put_css_set(src_cset
);
2687 put_css_set(dst_cset
);
2691 src_cset
->mg_dst_cset
= dst_cset
;
2693 if (list_empty(&dst_cset
->mg_preload_node
))
2694 list_add_tail(&dst_cset
->mg_preload_node
,
2695 &mgctx
->preloaded_dst_csets
);
2697 put_css_set(dst_cset
);
2699 for_each_subsys(ss
, ssid
)
2700 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2701 mgctx
->ss_mask
|= 1 << ssid
;
2708 * cgroup_migrate - migrate a process or task to a cgroup
2709 * @leader: the leader of the process or the task to migrate
2710 * @threadgroup: whether @leader points to the whole process or a single task
2711 * @mgctx: migration context
2713 * Migrate a process or task denoted by @leader. If migrating a process,
2714 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2715 * responsible for invoking cgroup_migrate_add_src() and
2716 * cgroup_migrate_prepare_dst() on the targets before invoking this
2717 * function and following up with cgroup_migrate_finish().
2719 * As long as a controller's ->can_attach() doesn't fail, this function is
2720 * guaranteed to succeed. This means that, excluding ->can_attach()
2721 * failure, when migrating multiple targets, the success or failure can be
2722 * decided for all targets by invoking group_migrate_prepare_dst() before
2723 * actually starting migrating.
2725 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2726 struct cgroup_mgctx
*mgctx
)
2728 struct task_struct
*task
;
2731 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2732 * already PF_EXITING could be freed from underneath us unless we
2733 * take an rcu_read_lock.
2735 spin_lock_irq(&css_set_lock
);
2739 cgroup_migrate_add_task(task
, mgctx
);
2742 } while_each_thread(leader
, task
);
2744 spin_unlock_irq(&css_set_lock
);
2746 return cgroup_migrate_execute(mgctx
);
2750 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2751 * @dst_cgrp: the cgroup to attach to
2752 * @leader: the task or the leader of the threadgroup to be attached
2753 * @threadgroup: attach the whole threadgroup?
2755 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2757 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2760 DEFINE_CGROUP_MGCTX(mgctx
);
2761 struct task_struct
*task
;
2764 /* look up all src csets */
2765 spin_lock_irq(&css_set_lock
);
2769 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2772 } while_each_thread(leader
, task
);
2774 spin_unlock_irq(&css_set_lock
);
2776 /* prepare dst csets and commit */
2777 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2779 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2781 cgroup_migrate_finish(&mgctx
);
2784 TRACE_CGROUP_PATH(attach_task
, dst_cgrp
, leader
, threadgroup
);
2789 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
,
2791 __acquires(&cgroup_threadgroup_rwsem
)
2793 struct task_struct
*tsk
;
2796 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2797 return ERR_PTR(-EINVAL
);
2800 * If we migrate a single thread, we don't care about threadgroup
2801 * stability. If the thread is `current`, it won't exit(2) under our
2802 * hands or change PID through exec(2). We exclude
2803 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2804 * callers by cgroup_mutex.
2805 * Therefore, we can skip the global lock.
2807 lockdep_assert_held(&cgroup_mutex
);
2808 if (pid
|| threadgroup
) {
2809 percpu_down_write(&cgroup_threadgroup_rwsem
);
2817 tsk
= find_task_by_vpid(pid
);
2819 tsk
= ERR_PTR(-ESRCH
);
2820 goto out_unlock_threadgroup
;
2827 tsk
= tsk
->group_leader
;
2830 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2831 * If userland migrates such a kthread to a non-root cgroup, it can
2832 * become trapped in a cpuset, or RT kthread may be born in a
2833 * cgroup with no rt_runtime allocated. Just say no.
2835 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2836 tsk
= ERR_PTR(-EINVAL
);
2837 goto out_unlock_threadgroup
;
2840 get_task_struct(tsk
);
2841 goto out_unlock_rcu
;
2843 out_unlock_threadgroup
:
2845 percpu_up_write(&cgroup_threadgroup_rwsem
);
2853 void cgroup_procs_write_finish(struct task_struct
*task
, bool locked
)
2854 __releases(&cgroup_threadgroup_rwsem
)
2856 struct cgroup_subsys
*ss
;
2859 /* release reference from cgroup_procs_write_start() */
2860 put_task_struct(task
);
2863 percpu_up_write(&cgroup_threadgroup_rwsem
);
2864 for_each_subsys(ss
, ssid
)
2865 if (ss
->post_attach
)
2869 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2871 struct cgroup_subsys
*ss
;
2872 bool printed
= false;
2875 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2878 seq_puts(seq
, ss
->name
);
2880 } while_each_subsys_mask();
2882 seq_putc(seq
, '\n');
2885 /* show controllers which are enabled from the parent */
2886 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2888 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2890 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2894 /* show controllers which are enabled for a given cgroup's children */
2895 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2897 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2899 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2904 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2905 * @cgrp: root of the subtree to update csses for
2907 * @cgrp's control masks have changed and its subtree's css associations
2908 * need to be updated accordingly. This function looks up all css_sets
2909 * which are attached to the subtree, creates the matching updated css_sets
2910 * and migrates the tasks to the new ones.
2912 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2914 DEFINE_CGROUP_MGCTX(mgctx
);
2915 struct cgroup_subsys_state
*d_css
;
2916 struct cgroup
*dsct
;
2917 struct css_set
*src_cset
;
2920 lockdep_assert_held(&cgroup_mutex
);
2922 percpu_down_write(&cgroup_threadgroup_rwsem
);
2924 /* look up all csses currently attached to @cgrp's subtree */
2925 spin_lock_irq(&css_set_lock
);
2926 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2927 struct cgrp_cset_link
*link
;
2929 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2930 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2932 spin_unlock_irq(&css_set_lock
);
2934 /* NULL dst indicates self on default hierarchy */
2935 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2939 spin_lock_irq(&css_set_lock
);
2940 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2941 struct task_struct
*task
, *ntask
;
2943 /* all tasks in src_csets need to be migrated */
2944 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2945 cgroup_migrate_add_task(task
, &mgctx
);
2947 spin_unlock_irq(&css_set_lock
);
2949 ret
= cgroup_migrate_execute(&mgctx
);
2951 cgroup_migrate_finish(&mgctx
);
2952 percpu_up_write(&cgroup_threadgroup_rwsem
);
2957 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2958 * @cgrp: root of the target subtree
2960 * Because css offlining is asynchronous, userland may try to re-enable a
2961 * controller while the previous css is still around. This function grabs
2962 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2964 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
2965 __acquires(&cgroup_mutex
)
2967 struct cgroup
*dsct
;
2968 struct cgroup_subsys_state
*d_css
;
2969 struct cgroup_subsys
*ss
;
2973 mutex_lock(&cgroup_mutex
);
2975 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2976 for_each_subsys(ss
, ssid
) {
2977 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2980 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
2983 cgroup_get_live(dsct
);
2984 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
2985 TASK_UNINTERRUPTIBLE
);
2987 mutex_unlock(&cgroup_mutex
);
2989 finish_wait(&dsct
->offline_waitq
, &wait
);
2998 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2999 * @cgrp: root of the target subtree
3001 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3002 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3005 static void cgroup_save_control(struct cgroup
*cgrp
)
3007 struct cgroup
*dsct
;
3008 struct cgroup_subsys_state
*d_css
;
3010 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3011 dsct
->old_subtree_control
= dsct
->subtree_control
;
3012 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
3013 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
3018 * cgroup_propagate_control - refresh control masks of a subtree
3019 * @cgrp: root of the target subtree
3021 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3022 * ->subtree_control and propagate controller availability through the
3023 * subtree so that descendants don't have unavailable controllers enabled.
3025 static void cgroup_propagate_control(struct cgroup
*cgrp
)
3027 struct cgroup
*dsct
;
3028 struct cgroup_subsys_state
*d_css
;
3030 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3031 dsct
->subtree_control
&= cgroup_control(dsct
);
3032 dsct
->subtree_ss_mask
=
3033 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
3034 cgroup_ss_mask(dsct
));
3039 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3040 * @cgrp: root of the target subtree
3042 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3043 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3046 static void cgroup_restore_control(struct cgroup
*cgrp
)
3048 struct cgroup
*dsct
;
3049 struct cgroup_subsys_state
*d_css
;
3051 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3052 dsct
->subtree_control
= dsct
->old_subtree_control
;
3053 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3054 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
3058 static bool css_visible(struct cgroup_subsys_state
*css
)
3060 struct cgroup_subsys
*ss
= css
->ss
;
3061 struct cgroup
*cgrp
= css
->cgroup
;
3063 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3065 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3067 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3071 * cgroup_apply_control_enable - enable or show csses according to control
3072 * @cgrp: root of the target subtree
3074 * Walk @cgrp's subtree and create new csses or make the existing ones
3075 * visible. A css is created invisible if it's being implicitly enabled
3076 * through dependency. An invisible css is made visible when the userland
3077 * explicitly enables it.
3079 * Returns 0 on success, -errno on failure. On failure, csses which have
3080 * been processed already aren't cleaned up. The caller is responsible for
3081 * cleaning up with cgroup_apply_control_disable().
3083 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3085 struct cgroup
*dsct
;
3086 struct cgroup_subsys_state
*d_css
;
3087 struct cgroup_subsys
*ss
;
3090 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3091 for_each_subsys(ss
, ssid
) {
3092 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3094 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3098 css
= css_create(dsct
, ss
);
3100 return PTR_ERR(css
);
3103 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3105 if (css_visible(css
)) {
3106 ret
= css_populate_dir(css
);
3117 * cgroup_apply_control_disable - kill or hide csses according to control
3118 * @cgrp: root of the target subtree
3120 * Walk @cgrp's subtree and kill and hide csses so that they match
3121 * cgroup_ss_mask() and cgroup_visible_mask().
3123 * A css is hidden when the userland requests it to be disabled while other
3124 * subsystems are still depending on it. The css must not actively control
3125 * resources and be in the vanilla state if it's made visible again later.
3126 * Controllers which may be depended upon should provide ->css_reset() for
3129 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3131 struct cgroup
*dsct
;
3132 struct cgroup_subsys_state
*d_css
;
3133 struct cgroup_subsys
*ss
;
3136 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3137 for_each_subsys(ss
, ssid
) {
3138 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3143 WARN_ON_ONCE(percpu_ref_is_dying(&css
->refcnt
));
3146 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3148 } else if (!css_visible(css
)) {
3158 * cgroup_apply_control - apply control mask updates to the subtree
3159 * @cgrp: root of the target subtree
3161 * subsystems can be enabled and disabled in a subtree using the following
3164 * 1. Call cgroup_save_control() to stash the current state.
3165 * 2. Update ->subtree_control masks in the subtree as desired.
3166 * 3. Call cgroup_apply_control() to apply the changes.
3167 * 4. Optionally perform other related operations.
3168 * 5. Call cgroup_finalize_control() to finish up.
3170 * This function implements step 3 and propagates the mask changes
3171 * throughout @cgrp's subtree, updates csses accordingly and perform
3172 * process migrations.
3174 static int cgroup_apply_control(struct cgroup
*cgrp
)
3178 cgroup_propagate_control(cgrp
);
3180 ret
= cgroup_apply_control_enable(cgrp
);
3185 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3186 * making the following cgroup_update_dfl_csses() properly update
3187 * css associations of all tasks in the subtree.
3189 ret
= cgroup_update_dfl_csses(cgrp
);
3197 * cgroup_finalize_control - finalize control mask update
3198 * @cgrp: root of the target subtree
3199 * @ret: the result of the update
3201 * Finalize control mask update. See cgroup_apply_control() for more info.
3203 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3206 cgroup_restore_control(cgrp
);
3207 cgroup_propagate_control(cgrp
);
3210 cgroup_apply_control_disable(cgrp
);
3213 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3215 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3217 /* if nothing is getting enabled, nothing to worry about */
3221 /* can @cgrp host any resources? */
3222 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3225 /* mixables don't care */
3226 if (cgroup_is_mixable(cgrp
))
3229 if (domain_enable
) {
3230 /* can't enable domain controllers inside a thread subtree */
3231 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3235 * Threaded controllers can handle internal competitions
3236 * and are always allowed inside a (prospective) thread
3239 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3244 * Controllers can't be enabled for a cgroup with tasks to avoid
3245 * child cgroups competing against tasks.
3247 if (cgroup_has_tasks(cgrp
))
3253 /* change the enabled child controllers for a cgroup in the default hierarchy */
3254 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3255 char *buf
, size_t nbytes
,
3258 u16 enable
= 0, disable
= 0;
3259 struct cgroup
*cgrp
, *child
;
3260 struct cgroup_subsys
*ss
;
3265 * Parse input - space separated list of subsystem names prefixed
3266 * with either + or -.
3268 buf
= strstrip(buf
);
3269 while ((tok
= strsep(&buf
, " "))) {
3272 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3273 if (!cgroup_ssid_enabled(ssid
) ||
3274 strcmp(tok
+ 1, ss
->name
))
3278 enable
|= 1 << ssid
;
3279 disable
&= ~(1 << ssid
);
3280 } else if (*tok
== '-') {
3281 disable
|= 1 << ssid
;
3282 enable
&= ~(1 << ssid
);
3287 } while_each_subsys_mask();
3288 if (ssid
== CGROUP_SUBSYS_COUNT
)
3292 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3296 for_each_subsys(ss
, ssid
) {
3297 if (enable
& (1 << ssid
)) {
3298 if (cgrp
->subtree_control
& (1 << ssid
)) {
3299 enable
&= ~(1 << ssid
);
3303 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3307 } else if (disable
& (1 << ssid
)) {
3308 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3309 disable
&= ~(1 << ssid
);
3313 /* a child has it enabled? */
3314 cgroup_for_each_live_child(child
, cgrp
) {
3315 if (child
->subtree_control
& (1 << ssid
)) {
3323 if (!enable
&& !disable
) {
3328 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3332 /* save and update control masks and prepare csses */
3333 cgroup_save_control(cgrp
);
3335 cgrp
->subtree_control
|= enable
;
3336 cgrp
->subtree_control
&= ~disable
;
3338 ret
= cgroup_apply_control(cgrp
);
3339 cgroup_finalize_control(cgrp
, ret
);
3343 kernfs_activate(cgrp
->kn
);
3345 cgroup_kn_unlock(of
->kn
);
3346 return ret
?: nbytes
;
3350 * cgroup_enable_threaded - make @cgrp threaded
3351 * @cgrp: the target cgroup
3353 * Called when "threaded" is written to the cgroup.type interface file and
3354 * tries to make @cgrp threaded and join the parent's resource domain.
3355 * This function is never called on the root cgroup as cgroup.type doesn't
3358 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3360 struct cgroup
*parent
= cgroup_parent(cgrp
);
3361 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3362 struct cgroup
*dsct
;
3363 struct cgroup_subsys_state
*d_css
;
3366 lockdep_assert_held(&cgroup_mutex
);
3368 /* noop if already threaded */
3369 if (cgroup_is_threaded(cgrp
))
3373 * If @cgroup is populated or has domain controllers enabled, it
3374 * can't be switched. While the below cgroup_can_be_thread_root()
3375 * test can catch the same conditions, that's only when @parent is
3376 * not mixable, so let's check it explicitly.
3378 if (cgroup_is_populated(cgrp
) ||
3379 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3382 /* we're joining the parent's domain, ensure its validity */
3383 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3384 !cgroup_can_be_thread_root(dom_cgrp
))
3388 * The following shouldn't cause actual migrations and should
3391 cgroup_save_control(cgrp
);
3393 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3394 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3395 dsct
->dom_cgrp
= dom_cgrp
;
3397 ret
= cgroup_apply_control(cgrp
);
3399 parent
->nr_threaded_children
++;
3401 cgroup_finalize_control(cgrp
, ret
);
3405 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3407 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3409 if (cgroup_is_threaded(cgrp
))
3410 seq_puts(seq
, "threaded\n");
3411 else if (!cgroup_is_valid_domain(cgrp
))
3412 seq_puts(seq
, "domain invalid\n");
3413 else if (cgroup_is_thread_root(cgrp
))
3414 seq_puts(seq
, "domain threaded\n");
3416 seq_puts(seq
, "domain\n");
3421 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3422 size_t nbytes
, loff_t off
)
3424 struct cgroup
*cgrp
;
3427 /* only switching to threaded mode is supported */
3428 if (strcmp(strstrip(buf
), "threaded"))
3431 /* drain dying csses before we re-apply (threaded) subtree control */
3432 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3436 /* threaded can only be enabled */
3437 ret
= cgroup_enable_threaded(cgrp
);
3439 cgroup_kn_unlock(of
->kn
);
3440 return ret
?: nbytes
;
3443 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3445 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3446 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3448 if (descendants
== INT_MAX
)
3449 seq_puts(seq
, "max\n");
3451 seq_printf(seq
, "%d\n", descendants
);
3456 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3457 char *buf
, size_t nbytes
, loff_t off
)
3459 struct cgroup
*cgrp
;
3463 buf
= strstrip(buf
);
3464 if (!strcmp(buf
, "max")) {
3465 descendants
= INT_MAX
;
3467 ret
= kstrtoint(buf
, 0, &descendants
);
3472 if (descendants
< 0)
3475 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3479 cgrp
->max_descendants
= descendants
;
3481 cgroup_kn_unlock(of
->kn
);
3486 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3488 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3489 int depth
= READ_ONCE(cgrp
->max_depth
);
3491 if (depth
== INT_MAX
)
3492 seq_puts(seq
, "max\n");
3494 seq_printf(seq
, "%d\n", depth
);
3499 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3500 char *buf
, size_t nbytes
, loff_t off
)
3502 struct cgroup
*cgrp
;
3506 buf
= strstrip(buf
);
3507 if (!strcmp(buf
, "max")) {
3510 ret
= kstrtoint(buf
, 0, &depth
);
3518 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3522 cgrp
->max_depth
= depth
;
3524 cgroup_kn_unlock(of
->kn
);
3529 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3531 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3533 seq_printf(seq
, "populated %d\n", cgroup_is_populated(cgrp
));
3534 seq_printf(seq
, "frozen %d\n", test_bit(CGRP_FROZEN
, &cgrp
->flags
));
3539 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3541 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3543 seq_printf(seq
, "nr_descendants %d\n",
3544 cgroup
->nr_descendants
);
3545 seq_printf(seq
, "nr_dying_descendants %d\n",
3546 cgroup
->nr_dying_descendants
);
3551 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3552 struct cgroup
*cgrp
, int ssid
)
3554 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3555 struct cgroup_subsys_state
*css
;
3558 if (!ss
->css_extra_stat_show
)
3561 css
= cgroup_tryget_css(cgrp
, ss
);
3565 ret
= ss
->css_extra_stat_show(seq
, css
);
3570 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3572 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3575 cgroup_base_stat_cputime_show(seq
);
3576 #ifdef CONFIG_CGROUP_SCHED
3577 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3583 static int cgroup_io_pressure_show(struct seq_file
*seq
, void *v
)
3585 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3586 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3588 return psi_show(seq
, psi
, PSI_IO
);
3590 static int cgroup_memory_pressure_show(struct seq_file
*seq
, void *v
)
3592 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3593 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3595 return psi_show(seq
, psi
, PSI_MEM
);
3597 static int cgroup_cpu_pressure_show(struct seq_file
*seq
, void *v
)
3599 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3600 struct psi_group
*psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3602 return psi_show(seq
, psi
, PSI_CPU
);
3605 static ssize_t
cgroup_pressure_write(struct kernfs_open_file
*of
, char *buf
,
3606 size_t nbytes
, enum psi_res res
)
3608 struct psi_trigger
*new;
3609 struct cgroup
*cgrp
;
3610 struct psi_group
*psi
;
3612 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3617 cgroup_kn_unlock(of
->kn
);
3619 psi
= cgroup_ino(cgrp
) == 1 ? &psi_system
: &cgrp
->psi
;
3620 new = psi_trigger_create(psi
, buf
, nbytes
, res
);
3623 return PTR_ERR(new);
3626 psi_trigger_replace(&of
->priv
, new);
3633 static ssize_t
cgroup_io_pressure_write(struct kernfs_open_file
*of
,
3634 char *buf
, size_t nbytes
,
3637 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_IO
);
3640 static ssize_t
cgroup_memory_pressure_write(struct kernfs_open_file
*of
,
3641 char *buf
, size_t nbytes
,
3644 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_MEM
);
3647 static ssize_t
cgroup_cpu_pressure_write(struct kernfs_open_file
*of
,
3648 char *buf
, size_t nbytes
,
3651 return cgroup_pressure_write(of
, buf
, nbytes
, PSI_CPU
);
3654 static __poll_t
cgroup_pressure_poll(struct kernfs_open_file
*of
,
3657 return psi_trigger_poll(&of
->priv
, of
->file
, pt
);
3660 static void cgroup_pressure_release(struct kernfs_open_file
*of
)
3662 psi_trigger_replace(&of
->priv
, NULL
);
3665 bool cgroup_psi_enabled(void)
3667 return (cgroup_feature_disable_mask
& (1 << OPT_FEATURE_PRESSURE
)) == 0;
3670 #else /* CONFIG_PSI */
3671 bool cgroup_psi_enabled(void)
3676 #endif /* CONFIG_PSI */
3678 static int cgroup_freeze_show(struct seq_file
*seq
, void *v
)
3680 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3682 seq_printf(seq
, "%d\n", cgrp
->freezer
.freeze
);
3687 static ssize_t
cgroup_freeze_write(struct kernfs_open_file
*of
,
3688 char *buf
, size_t nbytes
, loff_t off
)
3690 struct cgroup
*cgrp
;
3694 ret
= kstrtoint(strstrip(buf
), 0, &freeze
);
3698 if (freeze
< 0 || freeze
> 1)
3701 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3705 cgroup_freeze(cgrp
, freeze
);
3707 cgroup_kn_unlock(of
->kn
);
3712 static void __cgroup_kill(struct cgroup
*cgrp
)
3714 struct css_task_iter it
;
3715 struct task_struct
*task
;
3717 lockdep_assert_held(&cgroup_mutex
);
3719 spin_lock_irq(&css_set_lock
);
3720 set_bit(CGRP_KILL
, &cgrp
->flags
);
3721 spin_unlock_irq(&css_set_lock
);
3723 css_task_iter_start(&cgrp
->self
, CSS_TASK_ITER_PROCS
| CSS_TASK_ITER_THREADED
, &it
);
3724 while ((task
= css_task_iter_next(&it
))) {
3725 /* Ignore kernel threads here. */
3726 if (task
->flags
& PF_KTHREAD
)
3729 /* Skip tasks that are already dying. */
3730 if (__fatal_signal_pending(task
))
3733 send_sig(SIGKILL
, task
, 0);
3735 css_task_iter_end(&it
);
3737 spin_lock_irq(&css_set_lock
);
3738 clear_bit(CGRP_KILL
, &cgrp
->flags
);
3739 spin_unlock_irq(&css_set_lock
);
3742 static void cgroup_kill(struct cgroup
*cgrp
)
3744 struct cgroup_subsys_state
*css
;
3745 struct cgroup
*dsct
;
3747 lockdep_assert_held(&cgroup_mutex
);
3749 cgroup_for_each_live_descendant_pre(dsct
, css
, cgrp
)
3750 __cgroup_kill(dsct
);
3753 static ssize_t
cgroup_kill_write(struct kernfs_open_file
*of
, char *buf
,
3754 size_t nbytes
, loff_t off
)
3758 struct cgroup
*cgrp
;
3760 ret
= kstrtoint(strstrip(buf
), 0, &kill
);
3767 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3772 * Killing is a process directed operation, i.e. the whole thread-group
3773 * is taken down so act like we do for cgroup.procs and only make this
3774 * writable in non-threaded cgroups.
3776 if (cgroup_is_threaded(cgrp
))
3781 cgroup_kn_unlock(of
->kn
);
3783 return ret
?: nbytes
;
3786 static int cgroup_file_open(struct kernfs_open_file
*of
)
3788 struct cftype
*cft
= of_cft(of
);
3791 return cft
->open(of
);
3795 static void cgroup_file_release(struct kernfs_open_file
*of
)
3797 struct cftype
*cft
= of_cft(of
);
3803 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3804 size_t nbytes
, loff_t off
)
3806 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3807 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3808 struct cftype
*cft
= of_cft(of
);
3809 struct cgroup_subsys_state
*css
;
3816 * If namespaces are delegation boundaries, disallow writes to
3817 * files in an non-init namespace root from inside the namespace
3818 * except for the files explicitly marked delegatable -
3819 * cgroup.procs and cgroup.subtree_control.
3821 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3822 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3823 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3827 return cft
->write(of
, buf
, nbytes
, off
);
3830 * kernfs guarantees that a file isn't deleted with operations in
3831 * flight, which means that the matching css is and stays alive and
3832 * doesn't need to be pinned. The RCU locking is not necessary
3833 * either. It's just for the convenience of using cgroup_css().
3836 css
= cgroup_css(cgrp
, cft
->ss
);
3839 if (cft
->write_u64
) {
3840 unsigned long long v
;
3841 ret
= kstrtoull(buf
, 0, &v
);
3843 ret
= cft
->write_u64(css
, cft
, v
);
3844 } else if (cft
->write_s64
) {
3846 ret
= kstrtoll(buf
, 0, &v
);
3848 ret
= cft
->write_s64(css
, cft
, v
);
3853 return ret
?: nbytes
;
3856 static __poll_t
cgroup_file_poll(struct kernfs_open_file
*of
, poll_table
*pt
)
3858 struct cftype
*cft
= of_cft(of
);
3861 return cft
->poll(of
, pt
);
3863 return kernfs_generic_poll(of
, pt
);
3866 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3868 return seq_cft(seq
)->seq_start(seq
, ppos
);
3871 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3873 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3876 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3878 if (seq_cft(seq
)->seq_stop
)
3879 seq_cft(seq
)->seq_stop(seq
, v
);
3882 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3884 struct cftype
*cft
= seq_cft(m
);
3885 struct cgroup_subsys_state
*css
= seq_css(m
);
3888 return cft
->seq_show(m
, arg
);
3891 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3892 else if (cft
->read_s64
)
3893 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3899 static struct kernfs_ops cgroup_kf_single_ops
= {
3900 .atomic_write_len
= PAGE_SIZE
,
3901 .open
= cgroup_file_open
,
3902 .release
= cgroup_file_release
,
3903 .write
= cgroup_file_write
,
3904 .poll
= cgroup_file_poll
,
3905 .seq_show
= cgroup_seqfile_show
,
3908 static struct kernfs_ops cgroup_kf_ops
= {
3909 .atomic_write_len
= PAGE_SIZE
,
3910 .open
= cgroup_file_open
,
3911 .release
= cgroup_file_release
,
3912 .write
= cgroup_file_write
,
3913 .poll
= cgroup_file_poll
,
3914 .seq_start
= cgroup_seqfile_start
,
3915 .seq_next
= cgroup_seqfile_next
,
3916 .seq_stop
= cgroup_seqfile_stop
,
3917 .seq_show
= cgroup_seqfile_show
,
3920 /* set uid and gid of cgroup dirs and files to that of the creator */
3921 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3923 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3924 .ia_uid
= current_fsuid(),
3925 .ia_gid
= current_fsgid(), };
3927 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3928 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3931 return kernfs_setattr(kn
, &iattr
);
3934 static void cgroup_file_notify_timer(struct timer_list
*timer
)
3936 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
3940 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3943 char name
[CGROUP_FILE_NAME_MAX
];
3944 struct kernfs_node
*kn
;
3945 struct lock_class_key
*key
= NULL
;
3948 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3949 key
= &cft
->lockdep_key
;
3951 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3952 cgroup_file_mode(cft
),
3953 GLOBAL_ROOT_UID
, GLOBAL_ROOT_GID
,
3954 0, cft
->kf_ops
, cft
,
3959 ret
= cgroup_kn_set_ugid(kn
);
3965 if (cft
->file_offset
) {
3966 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3968 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
3970 spin_lock_irq(&cgroup_file_kn_lock
);
3972 spin_unlock_irq(&cgroup_file_kn_lock
);
3979 * cgroup_addrm_files - add or remove files to a cgroup directory
3980 * @css: the target css
3981 * @cgrp: the target cgroup (usually css->cgroup)
3982 * @cfts: array of cftypes to be added
3983 * @is_add: whether to add or remove
3985 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3986 * For removals, this function never fails.
3988 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3989 struct cgroup
*cgrp
, struct cftype cfts
[],
3992 struct cftype
*cft
, *cft_end
= NULL
;
3995 lockdep_assert_held(&cgroup_mutex
);
3998 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3999 /* does cft->flags tell us to skip this file on @cgrp? */
4000 if ((cft
->flags
& CFTYPE_PRESSURE
) && !cgroup_psi_enabled())
4002 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
4004 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
4006 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
4008 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
4010 if ((cft
->flags
& CFTYPE_DEBUG
) && !cgroup_debug
)
4013 ret
= cgroup_add_file(css
, cgrp
, cft
);
4015 pr_warn("%s: failed to add %s, err=%d\n",
4016 __func__
, cft
->name
, ret
);
4022 cgroup_rm_file(cgrp
, cft
);
4028 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
4030 struct cgroup_subsys
*ss
= cfts
[0].ss
;
4031 struct cgroup
*root
= &ss
->root
->cgrp
;
4032 struct cgroup_subsys_state
*css
;
4035 lockdep_assert_held(&cgroup_mutex
);
4037 /* add/rm files for all cgroups created before */
4038 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
4039 struct cgroup
*cgrp
= css
->cgroup
;
4041 if (!(css
->flags
& CSS_VISIBLE
))
4044 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
4050 kernfs_activate(root
->kn
);
4054 static void cgroup_exit_cftypes(struct cftype
*cfts
)
4058 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4059 /* free copy for custom atomic_write_len, see init_cftypes() */
4060 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
4065 /* revert flags set by cgroup core while adding @cfts */
4066 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
4070 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4074 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
4075 struct kernfs_ops
*kf_ops
;
4077 WARN_ON(cft
->ss
|| cft
->kf_ops
);
4079 if ((cft
->flags
& CFTYPE_PRESSURE
) && !cgroup_psi_enabled())
4083 kf_ops
= &cgroup_kf_ops
;
4085 kf_ops
= &cgroup_kf_single_ops
;
4088 * Ugh... if @cft wants a custom max_write_len, we need to
4089 * make a copy of kf_ops to set its atomic_write_len.
4091 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
4092 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
4094 cgroup_exit_cftypes(cfts
);
4097 kf_ops
->atomic_write_len
= cft
->max_write_len
;
4100 cft
->kf_ops
= kf_ops
;
4107 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
4109 lockdep_assert_held(&cgroup_mutex
);
4111 if (!cfts
|| !cfts
[0].ss
)
4114 list_del(&cfts
->node
);
4115 cgroup_apply_cftypes(cfts
, false);
4116 cgroup_exit_cftypes(cfts
);
4121 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4122 * @cfts: zero-length name terminated array of cftypes
4124 * Unregister @cfts. Files described by @cfts are removed from all
4125 * existing cgroups and all future cgroups won't have them either. This
4126 * function can be called anytime whether @cfts' subsys is attached or not.
4128 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4131 int cgroup_rm_cftypes(struct cftype
*cfts
)
4135 mutex_lock(&cgroup_mutex
);
4136 ret
= cgroup_rm_cftypes_locked(cfts
);
4137 mutex_unlock(&cgroup_mutex
);
4142 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4143 * @ss: target cgroup subsystem
4144 * @cfts: zero-length name terminated array of cftypes
4146 * Register @cfts to @ss. Files described by @cfts are created for all
4147 * existing cgroups to which @ss is attached and all future cgroups will
4148 * have them too. This function can be called anytime whether @ss is
4151 * Returns 0 on successful registration, -errno on failure. Note that this
4152 * function currently returns 0 as long as @cfts registration is successful
4153 * even if some file creation attempts on existing cgroups fail.
4155 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4159 if (!cgroup_ssid_enabled(ss
->id
))
4162 if (!cfts
|| cfts
[0].name
[0] == '\0')
4165 ret
= cgroup_init_cftypes(ss
, cfts
);
4169 mutex_lock(&cgroup_mutex
);
4171 list_add_tail(&cfts
->node
, &ss
->cfts
);
4172 ret
= cgroup_apply_cftypes(cfts
, true);
4174 cgroup_rm_cftypes_locked(cfts
);
4176 mutex_unlock(&cgroup_mutex
);
4181 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4182 * @ss: target cgroup subsystem
4183 * @cfts: zero-length name terminated array of cftypes
4185 * Similar to cgroup_add_cftypes() but the added files are only used for
4186 * the default hierarchy.
4188 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4192 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4193 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
4194 return cgroup_add_cftypes(ss
, cfts
);
4198 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4199 * @ss: target cgroup subsystem
4200 * @cfts: zero-length name terminated array of cftypes
4202 * Similar to cgroup_add_cftypes() but the added files are only used for
4203 * the legacy hierarchies.
4205 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
4209 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
4210 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
4211 return cgroup_add_cftypes(ss
, cfts
);
4215 * cgroup_file_notify - generate a file modified event for a cgroup_file
4216 * @cfile: target cgroup_file
4218 * @cfile must have been obtained by setting cftype->file_offset.
4220 void cgroup_file_notify(struct cgroup_file
*cfile
)
4222 unsigned long flags
;
4224 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
4226 unsigned long last
= cfile
->notified_at
;
4227 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
4229 if (time_in_range(jiffies
, last
, next
)) {
4230 timer_reduce(&cfile
->notify_timer
, next
);
4232 kernfs_notify(cfile
->kn
);
4233 cfile
->notified_at
= jiffies
;
4236 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
4240 * css_next_child - find the next child of a given css
4241 * @pos: the current position (%NULL to initiate traversal)
4242 * @parent: css whose children to walk
4244 * This function returns the next child of @parent and should be called
4245 * under either cgroup_mutex or RCU read lock. The only requirement is
4246 * that @parent and @pos are accessible. The next sibling is guaranteed to
4247 * be returned regardless of their states.
4249 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4250 * css which finished ->css_online() is guaranteed to be visible in the
4251 * future iterations and will stay visible until the last reference is put.
4252 * A css which hasn't finished ->css_online() or already finished
4253 * ->css_offline() may show up during traversal. It's each subsystem's
4254 * responsibility to synchronize against on/offlining.
4256 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
4257 struct cgroup_subsys_state
*parent
)
4259 struct cgroup_subsys_state
*next
;
4261 cgroup_assert_mutex_or_rcu_locked();
4264 * @pos could already have been unlinked from the sibling list.
4265 * Once a cgroup is removed, its ->sibling.next is no longer
4266 * updated when its next sibling changes. CSS_RELEASED is set when
4267 * @pos is taken off list, at which time its next pointer is valid,
4268 * and, as releases are serialized, the one pointed to by the next
4269 * pointer is guaranteed to not have started release yet. This
4270 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4271 * critical section, the one pointed to by its next pointer is
4272 * guaranteed to not have finished its RCU grace period even if we
4273 * have dropped rcu_read_lock() in-between iterations.
4275 * If @pos has CSS_RELEASED set, its next pointer can't be
4276 * dereferenced; however, as each css is given a monotonically
4277 * increasing unique serial number and always appended to the
4278 * sibling list, the next one can be found by walking the parent's
4279 * children until the first css with higher serial number than
4280 * @pos's. While this path can be slower, it happens iff iteration
4281 * races against release and the race window is very small.
4284 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
4285 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
4286 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
4288 list_for_each_entry_rcu(next
, &parent
->children
, sibling
,
4289 lockdep_is_held(&cgroup_mutex
))
4290 if (next
->serial_nr
> pos
->serial_nr
)
4295 * @next, if not pointing to the head, can be dereferenced and is
4298 if (&next
->sibling
!= &parent
->children
)
4304 * css_next_descendant_pre - find the next descendant for pre-order walk
4305 * @pos: the current position (%NULL to initiate traversal)
4306 * @root: css whose descendants to walk
4308 * To be used by css_for_each_descendant_pre(). Find the next descendant
4309 * to visit for pre-order traversal of @root's descendants. @root is
4310 * included in the iteration and the first node to be visited.
4312 * While this function requires cgroup_mutex or RCU read locking, it
4313 * doesn't require the whole traversal to be contained in a single critical
4314 * section. This function will return the correct next descendant as long
4315 * as both @pos and @root are accessible and @pos is a descendant of @root.
4317 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4318 * css which finished ->css_online() is guaranteed to be visible in the
4319 * future iterations and will stay visible until the last reference is put.
4320 * A css which hasn't finished ->css_online() or already finished
4321 * ->css_offline() may show up during traversal. It's each subsystem's
4322 * responsibility to synchronize against on/offlining.
4324 struct cgroup_subsys_state
*
4325 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4326 struct cgroup_subsys_state
*root
)
4328 struct cgroup_subsys_state
*next
;
4330 cgroup_assert_mutex_or_rcu_locked();
4332 /* if first iteration, visit @root */
4336 /* visit the first child if exists */
4337 next
= css_next_child(NULL
, pos
);
4341 /* no child, visit my or the closest ancestor's next sibling */
4342 while (pos
!= root
) {
4343 next
= css_next_child(pos
, pos
->parent
);
4351 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
4354 * css_rightmost_descendant - return the rightmost descendant of a css
4355 * @pos: css of interest
4357 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4358 * is returned. This can be used during pre-order traversal to skip
4361 * While this function requires cgroup_mutex or RCU read locking, it
4362 * doesn't require the whole traversal to be contained in a single critical
4363 * section. This function will return the correct rightmost descendant as
4364 * long as @pos is accessible.
4366 struct cgroup_subsys_state
*
4367 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4369 struct cgroup_subsys_state
*last
, *tmp
;
4371 cgroup_assert_mutex_or_rcu_locked();
4375 /* ->prev isn't RCU safe, walk ->next till the end */
4377 css_for_each_child(tmp
, last
)
4384 static struct cgroup_subsys_state
*
4385 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4387 struct cgroup_subsys_state
*last
;
4391 pos
= css_next_child(NULL
, pos
);
4398 * css_next_descendant_post - find the next descendant for post-order walk
4399 * @pos: the current position (%NULL to initiate traversal)
4400 * @root: css whose descendants to walk
4402 * To be used by css_for_each_descendant_post(). Find the next descendant
4403 * to visit for post-order traversal of @root's descendants. @root is
4404 * included in the iteration and the last node to be visited.
4406 * While this function requires cgroup_mutex or RCU read locking, it
4407 * doesn't require the whole traversal to be contained in a single critical
4408 * section. This function will return the correct next descendant as long
4409 * as both @pos and @cgroup are accessible and @pos is a descendant of
4412 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4413 * css which finished ->css_online() is guaranteed to be visible in the
4414 * future iterations and will stay visible until the last reference is put.
4415 * A css which hasn't finished ->css_online() or already finished
4416 * ->css_offline() may show up during traversal. It's each subsystem's
4417 * responsibility to synchronize against on/offlining.
4419 struct cgroup_subsys_state
*
4420 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4421 struct cgroup_subsys_state
*root
)
4423 struct cgroup_subsys_state
*next
;
4425 cgroup_assert_mutex_or_rcu_locked();
4427 /* if first iteration, visit leftmost descendant which may be @root */
4429 return css_leftmost_descendant(root
);
4431 /* if we visited @root, we're done */
4435 /* if there's an unvisited sibling, visit its leftmost descendant */
4436 next
= css_next_child(pos
, pos
->parent
);
4438 return css_leftmost_descendant(next
);
4440 /* no sibling left, visit parent */
4445 * css_has_online_children - does a css have online children
4446 * @css: the target css
4448 * Returns %true if @css has any online children; otherwise, %false. This
4449 * function can be called from any context but the caller is responsible
4450 * for synchronizing against on/offlining as necessary.
4452 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4454 struct cgroup_subsys_state
*child
;
4458 css_for_each_child(child
, css
) {
4459 if (child
->flags
& CSS_ONLINE
) {
4468 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4470 struct list_head
*l
;
4471 struct cgrp_cset_link
*link
;
4472 struct css_set
*cset
;
4474 lockdep_assert_held(&css_set_lock
);
4476 /* find the next threaded cset */
4477 if (it
->tcset_pos
) {
4478 l
= it
->tcset_pos
->next
;
4480 if (l
!= it
->tcset_head
) {
4482 return container_of(l
, struct css_set
,
4483 threaded_csets_node
);
4486 it
->tcset_pos
= NULL
;
4489 /* find the next cset */
4492 if (l
== it
->cset_head
) {
4493 it
->cset_pos
= NULL
;
4498 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4500 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4506 /* initialize threaded css_set walking */
4507 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4509 put_css_set_locked(it
->cur_dcset
);
4510 it
->cur_dcset
= cset
;
4513 it
->tcset_head
= &cset
->threaded_csets
;
4514 it
->tcset_pos
= &cset
->threaded_csets
;
4521 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4522 * @it: the iterator to advance
4524 * Advance @it to the next css_set to walk.
4526 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4528 struct css_set
*cset
;
4530 lockdep_assert_held(&css_set_lock
);
4532 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4533 while ((cset
= css_task_iter_next_css_set(it
))) {
4534 if (!list_empty(&cset
->tasks
)) {
4535 it
->cur_tasks_head
= &cset
->tasks
;
4537 } else if (!list_empty(&cset
->mg_tasks
)) {
4538 it
->cur_tasks_head
= &cset
->mg_tasks
;
4540 } else if (!list_empty(&cset
->dying_tasks
)) {
4541 it
->cur_tasks_head
= &cset
->dying_tasks
;
4546 it
->task_pos
= NULL
;
4549 it
->task_pos
= it
->cur_tasks_head
->next
;
4552 * We don't keep css_sets locked across iteration steps and thus
4553 * need to take steps to ensure that iteration can be resumed after
4554 * the lock is re-acquired. Iteration is performed at two levels -
4555 * css_sets and tasks in them.
4557 * Once created, a css_set never leaves its cgroup lists, so a
4558 * pinned css_set is guaranteed to stay put and we can resume
4559 * iteration afterwards.
4561 * Tasks may leave @cset across iteration steps. This is resolved
4562 * by registering each iterator with the css_set currently being
4563 * walked and making css_set_move_task() advance iterators whose
4564 * next task is leaving.
4567 list_del(&it
->iters_node
);
4568 put_css_set_locked(it
->cur_cset
);
4571 it
->cur_cset
= cset
;
4572 list_add(&it
->iters_node
, &cset
->task_iters
);
4575 static void css_task_iter_skip(struct css_task_iter
*it
,
4576 struct task_struct
*task
)
4578 lockdep_assert_held(&css_set_lock
);
4580 if (it
->task_pos
== &task
->cg_list
) {
4581 it
->task_pos
= it
->task_pos
->next
;
4582 it
->flags
|= CSS_TASK_ITER_SKIPPED
;
4586 static void css_task_iter_advance(struct css_task_iter
*it
)
4588 struct task_struct
*task
;
4590 lockdep_assert_held(&css_set_lock
);
4594 * Advance iterator to find next entry. We go through cset
4595 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4598 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4599 it
->flags
&= ~CSS_TASK_ITER_SKIPPED
;
4601 it
->task_pos
= it
->task_pos
->next
;
4603 if (it
->task_pos
== &it
->cur_cset
->tasks
) {
4604 it
->cur_tasks_head
= &it
->cur_cset
->mg_tasks
;
4605 it
->task_pos
= it
->cur_tasks_head
->next
;
4607 if (it
->task_pos
== &it
->cur_cset
->mg_tasks
) {
4608 it
->cur_tasks_head
= &it
->cur_cset
->dying_tasks
;
4609 it
->task_pos
= it
->cur_tasks_head
->next
;
4611 if (it
->task_pos
== &it
->cur_cset
->dying_tasks
)
4612 css_task_iter_advance_css_set(it
);
4614 /* called from start, proceed to the first cset */
4615 css_task_iter_advance_css_set(it
);
4621 task
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
4623 if (it
->flags
& CSS_TASK_ITER_PROCS
) {
4624 /* if PROCS, skip over tasks which aren't group leaders */
4625 if (!thread_group_leader(task
))
4628 /* and dying leaders w/o live member threads */
4629 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
&&
4630 !atomic_read(&task
->signal
->live
))
4633 /* skip all dying ones */
4634 if (it
->cur_tasks_head
== &it
->cur_cset
->dying_tasks
)
4640 * css_task_iter_start - initiate task iteration
4641 * @css: the css to walk tasks of
4642 * @flags: CSS_TASK_ITER_* flags
4643 * @it: the task iterator to use
4645 * Initiate iteration through the tasks of @css. The caller can call
4646 * css_task_iter_next() to walk through the tasks until the function
4647 * returns NULL. On completion of iteration, css_task_iter_end() must be
4650 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4651 struct css_task_iter
*it
)
4653 memset(it
, 0, sizeof(*it
));
4655 spin_lock_irq(&css_set_lock
);
4660 if (CGROUP_HAS_SUBSYS_CONFIG
&& it
->ss
)
4661 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4663 it
->cset_pos
= &css
->cgroup
->cset_links
;
4665 it
->cset_head
= it
->cset_pos
;
4667 css_task_iter_advance(it
);
4669 spin_unlock_irq(&css_set_lock
);
4673 * css_task_iter_next - return the next task for the iterator
4674 * @it: the task iterator being iterated
4676 * The "next" function for task iteration. @it should have been
4677 * initialized via css_task_iter_start(). Returns NULL when the iteration
4680 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4683 put_task_struct(it
->cur_task
);
4684 it
->cur_task
= NULL
;
4687 spin_lock_irq(&css_set_lock
);
4689 /* @it may be half-advanced by skips, finish advancing */
4690 if (it
->flags
& CSS_TASK_ITER_SKIPPED
)
4691 css_task_iter_advance(it
);
4694 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4696 get_task_struct(it
->cur_task
);
4697 css_task_iter_advance(it
);
4700 spin_unlock_irq(&css_set_lock
);
4702 return it
->cur_task
;
4706 * css_task_iter_end - finish task iteration
4707 * @it: the task iterator to finish
4709 * Finish task iteration started by css_task_iter_start().
4711 void css_task_iter_end(struct css_task_iter
*it
)
4714 spin_lock_irq(&css_set_lock
);
4715 list_del(&it
->iters_node
);
4716 put_css_set_locked(it
->cur_cset
);
4717 spin_unlock_irq(&css_set_lock
);
4721 put_css_set(it
->cur_dcset
);
4724 put_task_struct(it
->cur_task
);
4727 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4730 css_task_iter_end(of
->priv
);
4735 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4737 struct kernfs_open_file
*of
= s
->private;
4738 struct css_task_iter
*it
= of
->priv
;
4743 return css_task_iter_next(it
);
4746 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4747 unsigned int iter_flags
)
4749 struct kernfs_open_file
*of
= s
->private;
4750 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4751 struct css_task_iter
*it
= of
->priv
;
4754 * When a seq_file is seeked, it's always traversed sequentially
4755 * from position 0, so we can simply keep iterating on !0 *pos.
4758 if (WARN_ON_ONCE((*pos
)))
4759 return ERR_PTR(-EINVAL
);
4761 it
= kzalloc(sizeof(*it
), GFP_KERNEL
);
4763 return ERR_PTR(-ENOMEM
);
4765 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4766 } else if (!(*pos
)) {
4767 css_task_iter_end(it
);
4768 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4770 return it
->cur_task
;
4772 return cgroup_procs_next(s
, NULL
, NULL
);
4775 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4777 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4780 * All processes of a threaded subtree belong to the domain cgroup
4781 * of the subtree. Only threads can be distributed across the
4782 * subtree. Reject reads on cgroup.procs in the subtree proper.
4783 * They're always empty anyway.
4785 if (cgroup_is_threaded(cgrp
))
4786 return ERR_PTR(-EOPNOTSUPP
);
4788 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4789 CSS_TASK_ITER_THREADED
);
4792 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4794 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4798 static int cgroup_may_write(const struct cgroup
*cgrp
, struct super_block
*sb
)
4801 struct inode
*inode
;
4803 lockdep_assert_held(&cgroup_mutex
);
4805 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
4809 ret
= inode_permission(&init_user_ns
, inode
, MAY_WRITE
);
4814 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4815 struct cgroup
*dst_cgrp
,
4816 struct super_block
*sb
)
4818 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4819 struct cgroup
*com_cgrp
= src_cgrp
;
4822 lockdep_assert_held(&cgroup_mutex
);
4824 /* find the common ancestor */
4825 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4826 com_cgrp
= cgroup_parent(com_cgrp
);
4828 /* %current should be authorized to migrate to the common ancestor */
4829 ret
= cgroup_may_write(com_cgrp
, sb
);
4834 * If namespaces are delegation boundaries, %current must be able
4835 * to see both source and destination cgroups from its namespace.
4837 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4838 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4839 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4845 static int cgroup_attach_permissions(struct cgroup
*src_cgrp
,
4846 struct cgroup
*dst_cgrp
,
4847 struct super_block
*sb
, bool threadgroup
)
4851 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
, sb
);
4855 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
4859 if (!threadgroup
&& (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
))
4865 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
4868 struct cgroup
*src_cgrp
, *dst_cgrp
;
4869 struct task_struct
*task
;
4873 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4877 task
= cgroup_procs_write_start(buf
, threadgroup
, &locked
);
4878 ret
= PTR_ERR_OR_ZERO(task
);
4882 /* find the source cgroup */
4883 spin_lock_irq(&css_set_lock
);
4884 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4885 spin_unlock_irq(&css_set_lock
);
4887 /* process and thread migrations follow same delegation rule */
4888 ret
= cgroup_attach_permissions(src_cgrp
, dst_cgrp
,
4889 of
->file
->f_path
.dentry
->d_sb
, threadgroup
);
4893 ret
= cgroup_attach_task(dst_cgrp
, task
, threadgroup
);
4896 cgroup_procs_write_finish(task
, locked
);
4898 cgroup_kn_unlock(of
->kn
);
4903 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4904 char *buf
, size_t nbytes
, loff_t off
)
4906 return __cgroup_procs_write(of
, buf
, true) ?: nbytes
;
4909 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4911 return __cgroup_procs_start(s
, pos
, 0);
4914 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4915 char *buf
, size_t nbytes
, loff_t off
)
4917 return __cgroup_procs_write(of
, buf
, false) ?: nbytes
;
4920 /* cgroup core interface files for the default hierarchy */
4921 static struct cftype cgroup_base_files
[] = {
4923 .name
= "cgroup.type",
4924 .flags
= CFTYPE_NOT_ON_ROOT
,
4925 .seq_show
= cgroup_type_show
,
4926 .write
= cgroup_type_write
,
4929 .name
= "cgroup.procs",
4930 .flags
= CFTYPE_NS_DELEGATABLE
,
4931 .file_offset
= offsetof(struct cgroup
, procs_file
),
4932 .release
= cgroup_procs_release
,
4933 .seq_start
= cgroup_procs_start
,
4934 .seq_next
= cgroup_procs_next
,
4935 .seq_show
= cgroup_procs_show
,
4936 .write
= cgroup_procs_write
,
4939 .name
= "cgroup.threads",
4940 .flags
= CFTYPE_NS_DELEGATABLE
,
4941 .release
= cgroup_procs_release
,
4942 .seq_start
= cgroup_threads_start
,
4943 .seq_next
= cgroup_procs_next
,
4944 .seq_show
= cgroup_procs_show
,
4945 .write
= cgroup_threads_write
,
4948 .name
= "cgroup.controllers",
4949 .seq_show
= cgroup_controllers_show
,
4952 .name
= "cgroup.subtree_control",
4953 .flags
= CFTYPE_NS_DELEGATABLE
,
4954 .seq_show
= cgroup_subtree_control_show
,
4955 .write
= cgroup_subtree_control_write
,
4958 .name
= "cgroup.events",
4959 .flags
= CFTYPE_NOT_ON_ROOT
,
4960 .file_offset
= offsetof(struct cgroup
, events_file
),
4961 .seq_show
= cgroup_events_show
,
4964 .name
= "cgroup.max.descendants",
4965 .seq_show
= cgroup_max_descendants_show
,
4966 .write
= cgroup_max_descendants_write
,
4969 .name
= "cgroup.max.depth",
4970 .seq_show
= cgroup_max_depth_show
,
4971 .write
= cgroup_max_depth_write
,
4974 .name
= "cgroup.stat",
4975 .seq_show
= cgroup_stat_show
,
4978 .name
= "cgroup.freeze",
4979 .flags
= CFTYPE_NOT_ON_ROOT
,
4980 .seq_show
= cgroup_freeze_show
,
4981 .write
= cgroup_freeze_write
,
4984 .name
= "cgroup.kill",
4985 .flags
= CFTYPE_NOT_ON_ROOT
,
4986 .write
= cgroup_kill_write
,
4990 .seq_show
= cpu_stat_show
,
4994 .name
= "io.pressure",
4995 .flags
= CFTYPE_PRESSURE
,
4996 .seq_show
= cgroup_io_pressure_show
,
4997 .write
= cgroup_io_pressure_write
,
4998 .poll
= cgroup_pressure_poll
,
4999 .release
= cgroup_pressure_release
,
5002 .name
= "memory.pressure",
5003 .flags
= CFTYPE_PRESSURE
,
5004 .seq_show
= cgroup_memory_pressure_show
,
5005 .write
= cgroup_memory_pressure_write
,
5006 .poll
= cgroup_pressure_poll
,
5007 .release
= cgroup_pressure_release
,
5010 .name
= "cpu.pressure",
5011 .flags
= CFTYPE_PRESSURE
,
5012 .seq_show
= cgroup_cpu_pressure_show
,
5013 .write
= cgroup_cpu_pressure_write
,
5014 .poll
= cgroup_pressure_poll
,
5015 .release
= cgroup_pressure_release
,
5017 #endif /* CONFIG_PSI */
5022 * css destruction is four-stage process.
5024 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5025 * Implemented in kill_css().
5027 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5028 * and thus css_tryget_online() is guaranteed to fail, the css can be
5029 * offlined by invoking offline_css(). After offlining, the base ref is
5030 * put. Implemented in css_killed_work_fn().
5032 * 3. When the percpu_ref reaches zero, the only possible remaining
5033 * accessors are inside RCU read sections. css_release() schedules the
5036 * 4. After the grace period, the css can be freed. Implemented in
5037 * css_free_work_fn().
5039 * It is actually hairier because both step 2 and 4 require process context
5040 * and thus involve punting to css->destroy_work adding two additional
5041 * steps to the already complex sequence.
5043 static void css_free_rwork_fn(struct work_struct
*work
)
5045 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
5046 struct cgroup_subsys_state
, destroy_rwork
);
5047 struct cgroup_subsys
*ss
= css
->ss
;
5048 struct cgroup
*cgrp
= css
->cgroup
;
5050 percpu_ref_exit(&css
->refcnt
);
5054 struct cgroup_subsys_state
*parent
= css
->parent
;
5058 cgroup_idr_remove(&ss
->css_idr
, id
);
5064 /* cgroup free path */
5065 atomic_dec(&cgrp
->root
->nr_cgrps
);
5066 cgroup1_pidlist_destroy_all(cgrp
);
5067 cancel_work_sync(&cgrp
->release_agent_work
);
5069 if (cgroup_parent(cgrp
)) {
5071 * We get a ref to the parent, and put the ref when
5072 * this cgroup is being freed, so it's guaranteed
5073 * that the parent won't be destroyed before its
5076 cgroup_put(cgroup_parent(cgrp
));
5077 kernfs_put(cgrp
->kn
);
5078 psi_cgroup_free(cgrp
);
5079 cgroup_rstat_exit(cgrp
);
5083 * This is root cgroup's refcnt reaching zero,
5084 * which indicates that the root should be
5087 cgroup_destroy_root(cgrp
->root
);
5092 static void css_release_work_fn(struct work_struct
*work
)
5094 struct cgroup_subsys_state
*css
=
5095 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5096 struct cgroup_subsys
*ss
= css
->ss
;
5097 struct cgroup
*cgrp
= css
->cgroup
;
5099 mutex_lock(&cgroup_mutex
);
5101 css
->flags
|= CSS_RELEASED
;
5102 list_del_rcu(&css
->sibling
);
5105 /* css release path */
5106 if (!list_empty(&css
->rstat_css_node
)) {
5107 cgroup_rstat_flush(cgrp
);
5108 list_del_rcu(&css
->rstat_css_node
);
5111 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
5112 if (ss
->css_released
)
5113 ss
->css_released(css
);
5115 struct cgroup
*tcgrp
;
5117 /* cgroup release path */
5118 TRACE_CGROUP_PATH(release
, cgrp
);
5120 cgroup_rstat_flush(cgrp
);
5122 spin_lock_irq(&css_set_lock
);
5123 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
5124 tcgrp
= cgroup_parent(tcgrp
))
5125 tcgrp
->nr_dying_descendants
--;
5126 spin_unlock_irq(&css_set_lock
);
5129 * There are two control paths which try to determine
5130 * cgroup from dentry without going through kernfs -
5131 * cgroupstats_build() and css_tryget_online_from_dir().
5132 * Those are supported by RCU protecting clearing of
5133 * cgrp->kn->priv backpointer.
5136 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5140 mutex_unlock(&cgroup_mutex
);
5142 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5143 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5146 static void css_release(struct percpu_ref
*ref
)
5148 struct cgroup_subsys_state
*css
=
5149 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5151 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5152 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5155 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5156 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5158 lockdep_assert_held(&cgroup_mutex
);
5160 cgroup_get_live(cgrp
);
5162 memset(css
, 0, sizeof(*css
));
5166 INIT_LIST_HEAD(&css
->sibling
);
5167 INIT_LIST_HEAD(&css
->children
);
5168 INIT_LIST_HEAD(&css
->rstat_css_node
);
5169 css
->serial_nr
= css_serial_nr_next
++;
5170 atomic_set(&css
->online_cnt
, 0);
5172 if (cgroup_parent(cgrp
)) {
5173 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5174 css_get(css
->parent
);
5177 if (ss
->css_rstat_flush
)
5178 list_add_rcu(&css
->rstat_css_node
, &cgrp
->rstat_css_list
);
5180 BUG_ON(cgroup_css(cgrp
, ss
));
5183 /* invoke ->css_online() on a new CSS and mark it online if successful */
5184 static int online_css(struct cgroup_subsys_state
*css
)
5186 struct cgroup_subsys
*ss
= css
->ss
;
5189 lockdep_assert_held(&cgroup_mutex
);
5192 ret
= ss
->css_online(css
);
5194 css
->flags
|= CSS_ONLINE
;
5195 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5197 atomic_inc(&css
->online_cnt
);
5199 atomic_inc(&css
->parent
->online_cnt
);
5204 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5205 static void offline_css(struct cgroup_subsys_state
*css
)
5207 struct cgroup_subsys
*ss
= css
->ss
;
5209 lockdep_assert_held(&cgroup_mutex
);
5211 if (!(css
->flags
& CSS_ONLINE
))
5214 if (ss
->css_offline
)
5215 ss
->css_offline(css
);
5217 css
->flags
&= ~CSS_ONLINE
;
5218 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5220 wake_up_all(&css
->cgroup
->offline_waitq
);
5224 * css_create - create a cgroup_subsys_state
5225 * @cgrp: the cgroup new css will be associated with
5226 * @ss: the subsys of new css
5228 * Create a new css associated with @cgrp - @ss pair. On success, the new
5229 * css is online and installed in @cgrp. This function doesn't create the
5230 * interface files. Returns 0 on success, -errno on failure.
5232 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5233 struct cgroup_subsys
*ss
)
5235 struct cgroup
*parent
= cgroup_parent(cgrp
);
5236 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5237 struct cgroup_subsys_state
*css
;
5240 lockdep_assert_held(&cgroup_mutex
);
5242 css
= ss
->css_alloc(parent_css
);
5244 css
= ERR_PTR(-ENOMEM
);
5248 init_and_link_css(css
, ss
, cgrp
);
5250 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5254 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5259 /* @css is ready to be brought online now, make it visible */
5260 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5261 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5263 err
= online_css(css
);
5270 list_del_rcu(&css
->sibling
);
5272 list_del_rcu(&css
->rstat_css_node
);
5273 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
5274 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
5275 return ERR_PTR(err
);
5279 * The returned cgroup is fully initialized including its control mask, but
5280 * it isn't associated with its kernfs_node and doesn't have the control
5283 static struct cgroup
*cgroup_create(struct cgroup
*parent
, const char *name
,
5286 struct cgroup_root
*root
= parent
->root
;
5287 struct cgroup
*cgrp
, *tcgrp
;
5288 struct kernfs_node
*kn
;
5289 int level
= parent
->level
+ 1;
5292 /* allocate the cgroup and its ID, 0 is reserved for the root */
5293 cgrp
= kzalloc(struct_size(cgrp
, ancestor_ids
, (level
+ 1)),
5296 return ERR_PTR(-ENOMEM
);
5298 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5302 ret
= cgroup_rstat_init(cgrp
);
5304 goto out_cancel_ref
;
5306 /* create the directory */
5307 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5314 init_cgroup_housekeeping(cgrp
);
5316 cgrp
->self
.parent
= &parent
->self
;
5318 cgrp
->level
= level
;
5320 ret
= psi_cgroup_alloc(cgrp
);
5322 goto out_kernfs_remove
;
5324 ret
= cgroup_bpf_inherit(cgrp
);
5329 * New cgroup inherits effective freeze counter, and
5330 * if the parent has to be frozen, the child has too.
5332 cgrp
->freezer
.e_freeze
= parent
->freezer
.e_freeze
;
5333 if (cgrp
->freezer
.e_freeze
) {
5335 * Set the CGRP_FREEZE flag, so when a process will be
5336 * attached to the child cgroup, it will become frozen.
5337 * At this point the new cgroup is unpopulated, so we can
5338 * consider it frozen immediately.
5340 set_bit(CGRP_FREEZE
, &cgrp
->flags
);
5341 set_bit(CGRP_FROZEN
, &cgrp
->flags
);
5344 spin_lock_irq(&css_set_lock
);
5345 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5346 cgrp
->ancestor_ids
[tcgrp
->level
] = cgroup_id(tcgrp
);
5348 if (tcgrp
!= cgrp
) {
5349 tcgrp
->nr_descendants
++;
5352 * If the new cgroup is frozen, all ancestor cgroups
5353 * get a new frozen descendant, but their state can't
5354 * change because of this.
5356 if (cgrp
->freezer
.e_freeze
)
5357 tcgrp
->freezer
.nr_frozen_descendants
++;
5360 spin_unlock_irq(&css_set_lock
);
5362 if (notify_on_release(parent
))
5363 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5365 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5366 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5368 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5370 /* allocation complete, commit to creation */
5371 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5372 atomic_inc(&root
->nr_cgrps
);
5373 cgroup_get_live(parent
);
5376 * On the default hierarchy, a child doesn't automatically inherit
5377 * subtree_control from the parent. Each is configured manually.
5379 if (!cgroup_on_dfl(cgrp
))
5380 cgrp
->subtree_control
= cgroup_control(cgrp
);
5382 cgroup_propagate_control(cgrp
);
5387 psi_cgroup_free(cgrp
);
5389 kernfs_remove(cgrp
->kn
);
5391 cgroup_rstat_exit(cgrp
);
5393 percpu_ref_exit(&cgrp
->self
.refcnt
);
5396 return ERR_PTR(ret
);
5399 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
5401 struct cgroup
*cgroup
;
5405 lockdep_assert_held(&cgroup_mutex
);
5407 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
5408 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
5411 if (level
> cgroup
->max_depth
)
5422 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
5424 struct cgroup
*parent
, *cgrp
;
5427 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5428 if (strchr(name
, '\n'))
5431 parent
= cgroup_kn_lock_live(parent_kn
, false);
5435 if (!cgroup_check_hierarchy_limits(parent
)) {
5440 cgrp
= cgroup_create(parent
, name
, mode
);
5442 ret
= PTR_ERR(cgrp
);
5447 * This extra ref will be put in cgroup_free_fn() and guarantees
5448 * that @cgrp->kn is always accessible.
5450 kernfs_get(cgrp
->kn
);
5452 ret
= cgroup_kn_set_ugid(cgrp
->kn
);
5456 ret
= css_populate_dir(&cgrp
->self
);
5460 ret
= cgroup_apply_control_enable(cgrp
);
5464 TRACE_CGROUP_PATH(mkdir
, cgrp
);
5466 /* let's create and online css's */
5467 kernfs_activate(cgrp
->kn
);
5473 cgroup_destroy_locked(cgrp
);
5475 cgroup_kn_unlock(parent_kn
);
5480 * This is called when the refcnt of a css is confirmed to be killed.
5481 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5482 * initiate destruction and put the css ref from kill_css().
5484 static void css_killed_work_fn(struct work_struct
*work
)
5486 struct cgroup_subsys_state
*css
=
5487 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5489 mutex_lock(&cgroup_mutex
);
5494 /* @css can't go away while we're holding cgroup_mutex */
5496 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5498 mutex_unlock(&cgroup_mutex
);
5501 /* css kill confirmation processing requires process context, bounce */
5502 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5504 struct cgroup_subsys_state
*css
=
5505 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5507 if (atomic_dec_and_test(&css
->online_cnt
)) {
5508 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5509 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5514 * kill_css - destroy a css
5515 * @css: css to destroy
5517 * This function initiates destruction of @css by removing cgroup interface
5518 * files and putting its base reference. ->css_offline() will be invoked
5519 * asynchronously once css_tryget_online() is guaranteed to fail and when
5520 * the reference count reaches zero, @css will be released.
5522 static void kill_css(struct cgroup_subsys_state
*css
)
5524 lockdep_assert_held(&cgroup_mutex
);
5526 if (css
->flags
& CSS_DYING
)
5529 css
->flags
|= CSS_DYING
;
5532 * This must happen before css is disassociated with its cgroup.
5533 * See seq_css() for details.
5538 * Killing would put the base ref, but we need to keep it alive
5539 * until after ->css_offline().
5544 * cgroup core guarantees that, by the time ->css_offline() is
5545 * invoked, no new css reference will be given out via
5546 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5547 * proceed to offlining css's because percpu_ref_kill() doesn't
5548 * guarantee that the ref is seen as killed on all CPUs on return.
5550 * Use percpu_ref_kill_and_confirm() to get notifications as each
5551 * css is confirmed to be seen as killed on all CPUs.
5553 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5557 * cgroup_destroy_locked - the first stage of cgroup destruction
5558 * @cgrp: cgroup to be destroyed
5560 * css's make use of percpu refcnts whose killing latency shouldn't be
5561 * exposed to userland and are RCU protected. Also, cgroup core needs to
5562 * guarantee that css_tryget_online() won't succeed by the time
5563 * ->css_offline() is invoked. To satisfy all the requirements,
5564 * destruction is implemented in the following two steps.
5566 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5567 * userland visible parts and start killing the percpu refcnts of
5568 * css's. Set up so that the next stage will be kicked off once all
5569 * the percpu refcnts are confirmed to be killed.
5571 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5572 * rest of destruction. Once all cgroup references are gone, the
5573 * cgroup is RCU-freed.
5575 * This function implements s1. After this step, @cgrp is gone as far as
5576 * the userland is concerned and a new cgroup with the same name may be
5577 * created. As cgroup doesn't care about the names internally, this
5578 * doesn't cause any problem.
5580 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5581 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5583 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5584 struct cgroup_subsys_state
*css
;
5585 struct cgrp_cset_link
*link
;
5588 lockdep_assert_held(&cgroup_mutex
);
5591 * Only migration can raise populated from zero and we're already
5592 * holding cgroup_mutex.
5594 if (cgroup_is_populated(cgrp
))
5598 * Make sure there's no live children. We can't test emptiness of
5599 * ->self.children as dead children linger on it while being
5600 * drained; otherwise, "rmdir parent/child parent" may fail.
5602 if (css_has_online_children(&cgrp
->self
))
5606 * Mark @cgrp and the associated csets dead. The former prevents
5607 * further task migration and child creation by disabling
5608 * cgroup_lock_live_group(). The latter makes the csets ignored by
5609 * the migration path.
5611 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5613 spin_lock_irq(&css_set_lock
);
5614 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5615 link
->cset
->dead
= true;
5616 spin_unlock_irq(&css_set_lock
);
5618 /* initiate massacre of all css's */
5619 for_each_css(css
, ssid
, cgrp
)
5622 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5623 css_clear_dir(&cgrp
->self
);
5624 kernfs_remove(cgrp
->kn
);
5626 if (parent
&& cgroup_is_threaded(cgrp
))
5627 parent
->nr_threaded_children
--;
5629 spin_lock_irq(&css_set_lock
);
5630 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5631 tcgrp
->nr_descendants
--;
5632 tcgrp
->nr_dying_descendants
++;
5634 * If the dying cgroup is frozen, decrease frozen descendants
5635 * counters of ancestor cgroups.
5637 if (test_bit(CGRP_FROZEN
, &cgrp
->flags
))
5638 tcgrp
->freezer
.nr_frozen_descendants
--;
5640 spin_unlock_irq(&css_set_lock
);
5642 cgroup1_check_for_release(parent
);
5644 cgroup_bpf_offline(cgrp
);
5646 /* put the base reference */
5647 percpu_ref_kill(&cgrp
->self
.refcnt
);
5652 int cgroup_rmdir(struct kernfs_node
*kn
)
5654 struct cgroup
*cgrp
;
5657 cgrp
= cgroup_kn_lock_live(kn
, false);
5661 ret
= cgroup_destroy_locked(cgrp
);
5663 TRACE_CGROUP_PATH(rmdir
, cgrp
);
5665 cgroup_kn_unlock(kn
);
5669 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5670 .show_options
= cgroup_show_options
,
5671 .mkdir
= cgroup_mkdir
,
5672 .rmdir
= cgroup_rmdir
,
5673 .show_path
= cgroup_show_path
,
5676 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5678 struct cgroup_subsys_state
*css
;
5680 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5682 mutex_lock(&cgroup_mutex
);
5684 idr_init(&ss
->css_idr
);
5685 INIT_LIST_HEAD(&ss
->cfts
);
5687 /* Create the root cgroup state for this subsystem */
5688 ss
->root
= &cgrp_dfl_root
;
5689 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5690 /* We don't handle early failures gracefully */
5691 BUG_ON(IS_ERR(css
));
5692 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5695 * Root csses are never destroyed and we can't initialize
5696 * percpu_ref during early init. Disable refcnting.
5698 css
->flags
|= CSS_NO_REF
;
5701 /* allocation can't be done safely during early init */
5704 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5705 BUG_ON(css
->id
< 0);
5708 /* Update the init_css_set to contain a subsys
5709 * pointer to this state - since the subsystem is
5710 * newly registered, all tasks and hence the
5711 * init_css_set is in the subsystem's root cgroup. */
5712 init_css_set
.subsys
[ss
->id
] = css
;
5714 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5715 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5716 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5717 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5719 /* At system boot, before all subsystems have been
5720 * registered, no tasks have been forked, so we don't
5721 * need to invoke fork callbacks here. */
5722 BUG_ON(!list_empty(&init_task
.tasks
));
5724 BUG_ON(online_css(css
));
5726 mutex_unlock(&cgroup_mutex
);
5730 * cgroup_init_early - cgroup initialization at system boot
5732 * Initialize cgroups at system boot, and initialize any
5733 * subsystems that request early init.
5735 int __init
cgroup_init_early(void)
5737 static struct cgroup_fs_context __initdata ctx
;
5738 struct cgroup_subsys
*ss
;
5741 ctx
.root
= &cgrp_dfl_root
;
5742 init_cgroup_root(&ctx
);
5743 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5745 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5747 for_each_subsys(ss
, i
) {
5748 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5749 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5750 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5752 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5753 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5756 ss
->name
= cgroup_subsys_name
[i
];
5757 if (!ss
->legacy_name
)
5758 ss
->legacy_name
= cgroup_subsys_name
[i
];
5761 cgroup_init_subsys(ss
, true);
5767 * cgroup_init - cgroup initialization
5769 * Register cgroup filesystem and /proc file, and initialize
5770 * any subsystems that didn't request early init.
5772 int __init
cgroup_init(void)
5774 struct cgroup_subsys
*ss
;
5777 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5778 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5779 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5781 cgroup_rstat_boot();
5784 * The latency of the synchronize_rcu() is too high for cgroups,
5785 * avoid it at the cost of forcing all readers into the slow path.
5787 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5789 get_user_ns(init_cgroup_ns
.user_ns
);
5791 mutex_lock(&cgroup_mutex
);
5794 * Add init_css_set to the hash table so that dfl_root can link to
5797 hash_add(css_set_table
, &init_css_set
.hlist
,
5798 css_set_hash(init_css_set
.subsys
));
5800 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5802 mutex_unlock(&cgroup_mutex
);
5804 for_each_subsys(ss
, ssid
) {
5805 if (ss
->early_init
) {
5806 struct cgroup_subsys_state
*css
=
5807 init_css_set
.subsys
[ss
->id
];
5809 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5811 BUG_ON(css
->id
< 0);
5813 cgroup_init_subsys(ss
, false);
5816 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5817 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5820 * Setting dfl_root subsys_mask needs to consider the
5821 * disabled flag and cftype registration needs kmalloc,
5822 * both of which aren't available during early_init.
5824 if (!cgroup_ssid_enabled(ssid
))
5827 if (cgroup1_ssid_disabled(ssid
))
5828 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5831 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5833 /* implicit controllers must be threaded too */
5834 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5836 if (ss
->implicit_on_dfl
)
5837 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5838 else if (!ss
->dfl_cftypes
)
5839 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5842 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5844 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5845 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5847 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5848 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5852 ss
->bind(init_css_set
.subsys
[ssid
]);
5854 mutex_lock(&cgroup_mutex
);
5855 css_populate_dir(init_css_set
.subsys
[ssid
]);
5856 mutex_unlock(&cgroup_mutex
);
5859 /* init_css_set.subsys[] has been updated, re-hash */
5860 hash_del(&init_css_set
.hlist
);
5861 hash_add(css_set_table
, &init_css_set
.hlist
,
5862 css_set_hash(init_css_set
.subsys
));
5864 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5865 WARN_ON(register_filesystem(&cgroup_fs_type
));
5866 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5867 WARN_ON(!proc_create_single("cgroups", 0, NULL
, proc_cgroupstats_show
));
5868 #ifdef CONFIG_CPUSETS
5869 WARN_ON(register_filesystem(&cpuset_fs_type
));
5875 static int __init
cgroup_wq_init(void)
5878 * There isn't much point in executing destruction path in
5879 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5880 * Use 1 for @max_active.
5882 * We would prefer to do this in cgroup_init() above, but that
5883 * is called before init_workqueues(): so leave this until after.
5885 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5886 BUG_ON(!cgroup_destroy_wq
);
5889 core_initcall(cgroup_wq_init
);
5891 void cgroup_path_from_kernfs_id(u64 id
, char *buf
, size_t buflen
)
5893 struct kernfs_node
*kn
;
5895 kn
= kernfs_find_and_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5898 kernfs_path(kn
, buf
, buflen
);
5903 * cgroup_get_from_id : get the cgroup associated with cgroup id
5905 * On success return the cgrp, on failure return NULL
5907 struct cgroup
*cgroup_get_from_id(u64 id
)
5909 struct kernfs_node
*kn
;
5910 struct cgroup
*cgrp
= NULL
;
5912 mutex_lock(&cgroup_mutex
);
5913 kn
= kernfs_find_and_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5918 if (cgroup_is_dead(cgrp
) || !cgroup_tryget(cgrp
))
5922 mutex_unlock(&cgroup_mutex
);
5925 EXPORT_SYMBOL_GPL(cgroup_get_from_id
);
5928 * proc_cgroup_show()
5929 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5930 * - Used for /proc/<pid>/cgroup.
5932 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5933 struct pid
*pid
, struct task_struct
*tsk
)
5937 struct cgroup_root
*root
;
5940 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5944 mutex_lock(&cgroup_mutex
);
5945 spin_lock_irq(&css_set_lock
);
5947 for_each_root(root
) {
5948 struct cgroup_subsys
*ss
;
5949 struct cgroup
*cgrp
;
5950 int ssid
, count
= 0;
5952 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5955 seq_printf(m
, "%d:", root
->hierarchy_id
);
5956 if (root
!= &cgrp_dfl_root
)
5957 for_each_subsys(ss
, ssid
)
5958 if (root
->subsys_mask
& (1 << ssid
))
5959 seq_printf(m
, "%s%s", count
++ ? "," : "",
5961 if (strlen(root
->name
))
5962 seq_printf(m
, "%sname=%s", count
? "," : "",
5966 cgrp
= task_cgroup_from_root(tsk
, root
);
5969 * On traditional hierarchies, all zombie tasks show up as
5970 * belonging to the root cgroup. On the default hierarchy,
5971 * while a zombie doesn't show up in "cgroup.procs" and
5972 * thus can't be migrated, its /proc/PID/cgroup keeps
5973 * reporting the cgroup it belonged to before exiting. If
5974 * the cgroup is removed before the zombie is reaped,
5975 * " (deleted)" is appended to the cgroup path.
5977 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5978 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5979 current
->nsproxy
->cgroup_ns
);
5980 if (retval
>= PATH_MAX
)
5981 retval
= -ENAMETOOLONG
;
5990 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5991 seq_puts(m
, " (deleted)\n");
5998 spin_unlock_irq(&css_set_lock
);
5999 mutex_unlock(&cgroup_mutex
);
6006 * cgroup_fork - initialize cgroup related fields during copy_process()
6007 * @child: pointer to task_struct of forking parent process.
6009 * A task is associated with the init_css_set until cgroup_post_fork()
6010 * attaches it to the target css_set.
6012 void cgroup_fork(struct task_struct
*child
)
6014 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
6015 INIT_LIST_HEAD(&child
->cg_list
);
6018 static struct cgroup
*cgroup_get_from_file(struct file
*f
)
6020 struct cgroup_subsys_state
*css
;
6021 struct cgroup
*cgrp
;
6023 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
6025 return ERR_CAST(css
);
6028 if (!cgroup_on_dfl(cgrp
)) {
6030 return ERR_PTR(-EBADF
);
6037 * cgroup_css_set_fork - find or create a css_set for a child process
6038 * @kargs: the arguments passed to create the child process
6040 * This functions finds or creates a new css_set which the child
6041 * process will be attached to in cgroup_post_fork(). By default,
6042 * the child process will be given the same css_set as its parent.
6044 * If CLONE_INTO_CGROUP is specified this function will try to find an
6045 * existing css_set which includes the requested cgroup and if not create
6046 * a new css_set that the child will be attached to later. If this function
6047 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6048 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6049 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6050 * to the target cgroup.
6052 static int cgroup_css_set_fork(struct kernel_clone_args
*kargs
)
6053 __acquires(&cgroup_mutex
) __acquires(&cgroup_threadgroup_rwsem
)
6056 struct cgroup
*dst_cgrp
= NULL
;
6057 struct css_set
*cset
;
6058 struct super_block
*sb
;
6061 if (kargs
->flags
& CLONE_INTO_CGROUP
)
6062 mutex_lock(&cgroup_mutex
);
6064 cgroup_threadgroup_change_begin(current
);
6066 spin_lock_irq(&css_set_lock
);
6067 cset
= task_css_set(current
);
6069 spin_unlock_irq(&css_set_lock
);
6071 if (!(kargs
->flags
& CLONE_INTO_CGROUP
)) {
6076 f
= fget_raw(kargs
->cgroup
);
6081 sb
= f
->f_path
.dentry
->d_sb
;
6083 dst_cgrp
= cgroup_get_from_file(f
);
6084 if (IS_ERR(dst_cgrp
)) {
6085 ret
= PTR_ERR(dst_cgrp
);
6090 if (cgroup_is_dead(dst_cgrp
)) {
6096 * Verify that we the target cgroup is writable for us. This is
6097 * usually done by the vfs layer but since we're not going through
6098 * the vfs layer here we need to do it "manually".
6100 ret
= cgroup_may_write(dst_cgrp
, sb
);
6104 ret
= cgroup_attach_permissions(cset
->dfl_cgrp
, dst_cgrp
, sb
,
6105 !(kargs
->flags
& CLONE_THREAD
));
6109 kargs
->cset
= find_css_set(cset
, dst_cgrp
);
6117 kargs
->cgrp
= dst_cgrp
;
6121 cgroup_threadgroup_change_end(current
);
6122 mutex_unlock(&cgroup_mutex
);
6126 cgroup_put(dst_cgrp
);
6129 put_css_set(kargs
->cset
);
6134 * cgroup_css_set_put_fork - drop references we took during fork
6135 * @kargs: the arguments passed to create the child process
6137 * Drop references to the prepared css_set and target cgroup if
6138 * CLONE_INTO_CGROUP was requested.
6140 static void cgroup_css_set_put_fork(struct kernel_clone_args
*kargs
)
6141 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6143 cgroup_threadgroup_change_end(current
);
6145 if (kargs
->flags
& CLONE_INTO_CGROUP
) {
6146 struct cgroup
*cgrp
= kargs
->cgrp
;
6147 struct css_set
*cset
= kargs
->cset
;
6149 mutex_unlock(&cgroup_mutex
);
6164 * cgroup_can_fork - called on a new task before the process is exposed
6165 * @child: the child process
6167 * This prepares a new css_set for the child process which the child will
6168 * be attached to in cgroup_post_fork().
6169 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6170 * callback returns an error, the fork aborts with that error code. This
6171 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6173 int cgroup_can_fork(struct task_struct
*child
, struct kernel_clone_args
*kargs
)
6175 struct cgroup_subsys
*ss
;
6178 ret
= cgroup_css_set_fork(kargs
);
6182 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
6183 ret
= ss
->can_fork(child
, kargs
->cset
);
6186 } while_each_subsys_mask();
6191 for_each_subsys(ss
, j
) {
6194 if (ss
->cancel_fork
)
6195 ss
->cancel_fork(child
, kargs
->cset
);
6198 cgroup_css_set_put_fork(kargs
);
6204 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6205 * @child: the child process
6206 * @kargs: the arguments passed to create the child process
6208 * This calls the cancel_fork() callbacks if a fork failed *after*
6209 * cgroup_can_fork() succeeded and cleans up references we took to
6210 * prepare a new css_set for the child process in cgroup_can_fork().
6212 void cgroup_cancel_fork(struct task_struct
*child
,
6213 struct kernel_clone_args
*kargs
)
6215 struct cgroup_subsys
*ss
;
6218 for_each_subsys(ss
, i
)
6219 if (ss
->cancel_fork
)
6220 ss
->cancel_fork(child
, kargs
->cset
);
6222 cgroup_css_set_put_fork(kargs
);
6226 * cgroup_post_fork - finalize cgroup setup for the child process
6227 * @child: the child process
6229 * Attach the child process to its css_set calling the subsystem fork()
6232 void cgroup_post_fork(struct task_struct
*child
,
6233 struct kernel_clone_args
*kargs
)
6234 __releases(&cgroup_threadgroup_rwsem
) __releases(&cgroup_mutex
)
6236 unsigned long cgrp_flags
= 0;
6238 struct cgroup_subsys
*ss
;
6239 struct css_set
*cset
;
6245 spin_lock_irq(&css_set_lock
);
6247 /* init tasks are special, only link regular threads */
6248 if (likely(child
->pid
)) {
6250 cgrp_flags
= kargs
->cgrp
->flags
;
6252 cgrp_flags
= cset
->dfl_cgrp
->flags
;
6254 WARN_ON_ONCE(!list_empty(&child
->cg_list
));
6256 css_set_move_task(child
, NULL
, cset
, false);
6262 if (!(child
->flags
& PF_KTHREAD
)) {
6263 if (unlikely(test_bit(CGRP_FREEZE
, &cgrp_flags
))) {
6265 * If the cgroup has to be frozen, the new task has
6266 * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6267 * get the task into the frozen state.
6269 spin_lock(&child
->sighand
->siglock
);
6270 WARN_ON_ONCE(child
->frozen
);
6271 child
->jobctl
|= JOBCTL_TRAP_FREEZE
;
6272 spin_unlock(&child
->sighand
->siglock
);
6275 * Calling cgroup_update_frozen() isn't required here,
6276 * because it will be called anyway a bit later from
6277 * do_freezer_trap(). So we avoid cgroup's transient
6278 * switch from the frozen state and back.
6283 * If the cgroup is to be killed notice it now and take the
6284 * child down right after we finished preparing it for
6287 kill
= test_bit(CGRP_KILL
, &cgrp_flags
);
6290 spin_unlock_irq(&css_set_lock
);
6293 * Call ss->fork(). This must happen after @child is linked on
6294 * css_set; otherwise, @child might change state between ->fork()
6295 * and addition to css_set.
6297 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
6299 } while_each_subsys_mask();
6301 /* Make the new cset the root_cset of the new cgroup namespace. */
6302 if (kargs
->flags
& CLONE_NEWCGROUP
) {
6303 struct css_set
*rcset
= child
->nsproxy
->cgroup_ns
->root_cset
;
6306 child
->nsproxy
->cgroup_ns
->root_cset
= cset
;
6310 /* Cgroup has to be killed so take down child immediately. */
6312 do_send_sig_info(SIGKILL
, SEND_SIG_NOINFO
, child
, PIDTYPE_TGID
);
6314 cgroup_css_set_put_fork(kargs
);
6318 * cgroup_exit - detach cgroup from exiting task
6319 * @tsk: pointer to task_struct of exiting process
6321 * Description: Detach cgroup from @tsk.
6324 void cgroup_exit(struct task_struct
*tsk
)
6326 struct cgroup_subsys
*ss
;
6327 struct css_set
*cset
;
6330 spin_lock_irq(&css_set_lock
);
6332 WARN_ON_ONCE(list_empty(&tsk
->cg_list
));
6333 cset
= task_css_set(tsk
);
6334 css_set_move_task(tsk
, cset
, NULL
, false);
6335 list_add_tail(&tsk
->cg_list
, &cset
->dying_tasks
);
6338 WARN_ON_ONCE(cgroup_task_frozen(tsk
));
6339 if (unlikely(!(tsk
->flags
& PF_KTHREAD
) &&
6340 test_bit(CGRP_FREEZE
, &task_dfl_cgroup(tsk
)->flags
)))
6341 cgroup_update_frozen(task_dfl_cgroup(tsk
));
6343 spin_unlock_irq(&css_set_lock
);
6345 /* see cgroup_post_fork() for details */
6346 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
6348 } while_each_subsys_mask();
6351 void cgroup_release(struct task_struct
*task
)
6353 struct cgroup_subsys
*ss
;
6356 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
6358 } while_each_subsys_mask();
6360 spin_lock_irq(&css_set_lock
);
6361 css_set_skip_task_iters(task_css_set(task
), task
);
6362 list_del_init(&task
->cg_list
);
6363 spin_unlock_irq(&css_set_lock
);
6366 void cgroup_free(struct task_struct
*task
)
6368 struct css_set
*cset
= task_css_set(task
);
6372 static int __init
cgroup_disable(char *str
)
6374 struct cgroup_subsys
*ss
;
6378 while ((token
= strsep(&str
, ",")) != NULL
) {
6382 for_each_subsys(ss
, i
) {
6383 if (strcmp(token
, ss
->name
) &&
6384 strcmp(token
, ss
->legacy_name
))
6387 static_branch_disable(cgroup_subsys_enabled_key
[i
]);
6388 pr_info("Disabling %s control group subsystem\n",
6392 for (i
= 0; i
< OPT_FEATURE_COUNT
; i
++) {
6393 if (strcmp(token
, cgroup_opt_feature_names
[i
]))
6395 cgroup_feature_disable_mask
|= 1 << i
;
6396 pr_info("Disabling %s control group feature\n",
6397 cgroup_opt_feature_names
[i
]);
6403 __setup("cgroup_disable=", cgroup_disable
);
6405 void __init __weak
enable_debug_cgroup(void) { }
6407 static int __init
enable_cgroup_debug(char *str
)
6409 cgroup_debug
= true;
6410 enable_debug_cgroup();
6413 __setup("cgroup_debug", enable_cgroup_debug
);
6416 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6417 * @dentry: directory dentry of interest
6418 * @ss: subsystem of interest
6420 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6421 * to get the corresponding css and return it. If such css doesn't exist
6422 * or can't be pinned, an ERR_PTR value is returned.
6424 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6425 struct cgroup_subsys
*ss
)
6427 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6428 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6429 struct cgroup_subsys_state
*css
= NULL
;
6430 struct cgroup
*cgrp
;
6432 /* is @dentry a cgroup dir? */
6433 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6434 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6435 return ERR_PTR(-EBADF
);
6440 * This path doesn't originate from kernfs and @kn could already
6441 * have been or be removed at any point. @kn->priv is RCU
6442 * protected for this access. See css_release_work_fn() for details.
6444 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
6446 css
= cgroup_css(cgrp
, ss
);
6448 if (!css
|| !css_tryget_online(css
))
6449 css
= ERR_PTR(-ENOENT
);
6456 * css_from_id - lookup css by id
6457 * @id: the cgroup id
6458 * @ss: cgroup subsys to be looked into
6460 * Returns the css if there's valid one with @id, otherwise returns NULL.
6461 * Should be called under rcu_read_lock().
6463 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6465 WARN_ON_ONCE(!rcu_read_lock_held());
6466 return idr_find(&ss
->css_idr
, id
);
6470 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6471 * @path: path on the default hierarchy
6473 * Find the cgroup at @path on the default hierarchy, increment its
6474 * reference count and return it. Returns pointer to the found cgroup on
6475 * success, ERR_PTR(-ENOENT) if @path doesn't exist and ERR_PTR(-ENOTDIR)
6476 * if @path points to a non-directory.
6478 struct cgroup
*cgroup_get_from_path(const char *path
)
6480 struct kernfs_node
*kn
;
6481 struct cgroup
*cgrp
;
6483 mutex_lock(&cgroup_mutex
);
6485 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
6487 if (kernfs_type(kn
) == KERNFS_DIR
) {
6489 cgroup_get_live(cgrp
);
6491 cgrp
= ERR_PTR(-ENOTDIR
);
6495 cgrp
= ERR_PTR(-ENOENT
);
6498 mutex_unlock(&cgroup_mutex
);
6501 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6504 * cgroup_get_from_fd - get a cgroup pointer from a fd
6505 * @fd: fd obtained by open(cgroup2_dir)
6507 * Find the cgroup from a fd which should be obtained
6508 * by opening a cgroup directory. Returns a pointer to the
6509 * cgroup on success. ERR_PTR is returned if the cgroup
6512 struct cgroup
*cgroup_get_from_fd(int fd
)
6514 struct cgroup
*cgrp
;
6519 return ERR_PTR(-EBADF
);
6521 cgrp
= cgroup_get_from_file(f
);
6525 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6527 static u64
power_of_ten(int power
)
6536 * cgroup_parse_float - parse a floating number
6537 * @input: input string
6538 * @dec_shift: number of decimal digits to shift
6541 * Parse a decimal floating point number in @input and store the result in
6542 * @v with decimal point right shifted @dec_shift times. For example, if
6543 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6544 * Returns 0 on success, -errno otherwise.
6546 * There's nothing cgroup specific about this function except that it's
6547 * currently the only user.
6549 int cgroup_parse_float(const char *input
, unsigned dec_shift
, s64
*v
)
6551 s64 whole
, frac
= 0;
6552 int fstart
= 0, fend
= 0, flen
;
6554 if (!sscanf(input
, "%lld.%n%lld%n", &whole
, &fstart
, &frac
, &fend
))
6559 flen
= fend
> fstart
? fend
- fstart
: 0;
6560 if (flen
< dec_shift
)
6561 frac
*= power_of_ten(dec_shift
- flen
);
6563 frac
= DIV_ROUND_CLOSEST_ULL(frac
, power_of_ten(flen
- dec_shift
));
6565 *v
= whole
* power_of_ten(dec_shift
) + frac
;
6570 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6571 * definition in cgroup-defs.h.
6573 #ifdef CONFIG_SOCK_CGROUP_DATA
6575 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6577 struct cgroup
*cgroup
;
6580 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6581 if (in_interrupt()) {
6582 cgroup
= &cgrp_dfl_root
.cgrp
;
6588 struct css_set
*cset
;
6590 cset
= task_css_set(current
);
6591 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6592 cgroup
= cset
->dfl_cgrp
;
6598 skcd
->cgroup
= cgroup
;
6599 cgroup_bpf_get(cgroup
);
6603 void cgroup_sk_clone(struct sock_cgroup_data
*skcd
)
6605 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6608 * We might be cloning a socket which is left in an empty
6609 * cgroup and the cgroup might have already been rmdir'd.
6610 * Don't use cgroup_get_live().
6613 cgroup_bpf_get(cgrp
);
6616 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6618 struct cgroup
*cgrp
= sock_cgroup_ptr(skcd
);
6620 cgroup_bpf_put(cgrp
);
6624 #endif /* CONFIG_SOCK_CGROUP_DATA */
6626 #ifdef CONFIG_CGROUP_BPF
6627 int cgroup_bpf_attach(struct cgroup
*cgrp
,
6628 struct bpf_prog
*prog
, struct bpf_prog
*replace_prog
,
6629 struct bpf_cgroup_link
*link
,
6630 enum bpf_attach_type type
,
6635 mutex_lock(&cgroup_mutex
);
6636 ret
= __cgroup_bpf_attach(cgrp
, prog
, replace_prog
, link
, type
, flags
);
6637 mutex_unlock(&cgroup_mutex
);
6641 int cgroup_bpf_detach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
6642 enum bpf_attach_type type
)
6646 mutex_lock(&cgroup_mutex
);
6647 ret
= __cgroup_bpf_detach(cgrp
, prog
, NULL
, type
);
6648 mutex_unlock(&cgroup_mutex
);
6652 int cgroup_bpf_query(struct cgroup
*cgrp
, const union bpf_attr
*attr
,
6653 union bpf_attr __user
*uattr
)
6657 mutex_lock(&cgroup_mutex
);
6658 ret
= __cgroup_bpf_query(cgrp
, attr
, uattr
);
6659 mutex_unlock(&cgroup_mutex
);
6662 #endif /* CONFIG_CGROUP_BPF */
6665 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
6666 ssize_t size
, const char *prefix
)
6671 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
6672 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
6675 if ((cft
->flags
& CFTYPE_PRESSURE
) && !cgroup_psi_enabled())
6679 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
6681 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
6683 if (WARN_ON(ret
>= size
))
6690 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6693 struct cgroup_subsys
*ss
;
6697 ret
= show_delegatable_files(cgroup_base_files
, buf
, PAGE_SIZE
- ret
,
6700 for_each_subsys(ss
, ssid
)
6701 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
6703 cgroup_subsys_name
[ssid
]);
6707 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
6709 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
6712 return snprintf(buf
, PAGE_SIZE
,
6714 "memory_localevents\n"
6715 "memory_recursiveprot\n");
6717 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
6719 static struct attribute
*cgroup_sysfs_attrs
[] = {
6720 &cgroup_delegate_attr
.attr
,
6721 &cgroup_features_attr
.attr
,
6725 static const struct attribute_group cgroup_sysfs_attr_group
= {
6726 .attrs
= cgroup_sysfs_attrs
,
6730 static int __init
cgroup_sysfs_init(void)
6732 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
6734 subsys_initcall(cgroup_sysfs_init
);
6736 #endif /* CONFIG_SYSFS */