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>
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/cgroup.h>
62 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
64 /* let's not notify more than 100 times per second */
65 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
68 * cgroup_mutex is the master lock. Any modification to cgroup or its
69 * hierarchy must be performed while holding it.
71 * css_set_lock protects task->cgroups pointer, the list of css_set
72 * objects, and the chain of tasks off each css_set.
74 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
75 * cgroup.h can use them for lockdep annotations.
77 DEFINE_MUTEX(cgroup_mutex
);
78 DEFINE_SPINLOCK(css_set_lock
);
80 #ifdef CONFIG_PROVE_RCU
81 EXPORT_SYMBOL_GPL(cgroup_mutex
);
82 EXPORT_SYMBOL_GPL(css_set_lock
);
86 * Protects cgroup_idr and css_idr so that IDs can be released without
87 * grabbing cgroup_mutex.
89 static DEFINE_SPINLOCK(cgroup_idr_lock
);
92 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
93 * against file removal/re-creation across css hiding.
95 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
97 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
99 #define cgroup_assert_mutex_or_rcu_locked() \
100 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
101 !lockdep_is_held(&cgroup_mutex), \
102 "cgroup_mutex or RCU read lock required");
105 * cgroup destruction makes heavy use of work items and there can be a lot
106 * of concurrent destructions. Use a separate workqueue so that cgroup
107 * destruction work items don't end up filling up max_active of system_wq
108 * which may lead to deadlock.
110 static struct workqueue_struct
*cgroup_destroy_wq
;
112 /* generate an array of cgroup subsystem pointers */
113 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
114 struct cgroup_subsys
*cgroup_subsys
[] = {
115 #include <linux/cgroup_subsys.h>
119 /* array of cgroup subsystem names */
120 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
121 static const char *cgroup_subsys_name
[] = {
122 #include <linux/cgroup_subsys.h>
126 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
128 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
129 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
130 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
131 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
132 #include <linux/cgroup_subsys.h>
135 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
136 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
137 #include <linux/cgroup_subsys.h>
141 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
142 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
143 #include <linux/cgroup_subsys.h>
147 static DEFINE_PER_CPU(struct cgroup_cpu_stat
, cgrp_dfl_root_cpu_stat
);
150 * The default hierarchy, reserved for the subsystems that are otherwise
151 * unattached - it never has more than a single cgroup, and all tasks are
152 * part of that cgroup.
154 struct cgroup_root cgrp_dfl_root
= { .cgrp
.cpu_stat
= &cgrp_dfl_root_cpu_stat
};
155 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
158 * The default hierarchy always exists but is hidden until mounted for the
159 * first time. This is for backward compatibility.
161 static bool cgrp_dfl_visible
;
163 /* some controllers are not supported in the default hierarchy */
164 static u16 cgrp_dfl_inhibit_ss_mask
;
166 /* some controllers are implicitly enabled on the default hierarchy */
167 static u16 cgrp_dfl_implicit_ss_mask
;
169 /* some controllers can be threaded on the default hierarchy */
170 static u16 cgrp_dfl_threaded_ss_mask
;
172 /* The list of hierarchy roots */
173 LIST_HEAD(cgroup_roots
);
174 static int cgroup_root_count
;
176 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
177 static DEFINE_IDR(cgroup_hierarchy_idr
);
180 * Assign a monotonically increasing serial number to csses. It guarantees
181 * cgroups with bigger numbers are newer than those with smaller numbers.
182 * Also, as csses are always appended to the parent's ->children list, it
183 * guarantees that sibling csses are always sorted in the ascending serial
184 * number order on the list. Protected by cgroup_mutex.
186 static u64 css_serial_nr_next
= 1;
189 * These bitmasks identify subsystems with specific features to avoid
190 * having to do iterative checks repeatedly.
192 static u16 have_fork_callback __read_mostly
;
193 static u16 have_exit_callback __read_mostly
;
194 static u16 have_free_callback __read_mostly
;
195 static u16 have_canfork_callback __read_mostly
;
197 /* cgroup namespace for init task */
198 struct cgroup_namespace init_cgroup_ns
= {
199 .count
= REFCOUNT_INIT(2),
200 .user_ns
= &init_user_ns
,
201 .ns
.ops
= &cgroupns_operations
,
202 .ns
.inum
= PROC_CGROUP_INIT_INO
,
203 .root_cset
= &init_css_set
,
206 static struct file_system_type cgroup2_fs_type
;
207 static struct cftype cgroup_base_files
[];
209 static int cgroup_apply_control(struct cgroup
*cgrp
);
210 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
211 static void css_task_iter_advance(struct css_task_iter
*it
);
212 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
213 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
214 struct cgroup_subsys
*ss
);
215 static void css_release(struct percpu_ref
*ref
);
216 static void kill_css(struct cgroup_subsys_state
*css
);
217 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
218 struct cgroup
*cgrp
, struct cftype cfts
[],
222 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
223 * @ssid: subsys ID of interest
225 * cgroup_subsys_enabled() can only be used with literal subsys names which
226 * is fine for individual subsystems but unsuitable for cgroup core. This
227 * is slower static_key_enabled() based test indexed by @ssid.
229 bool cgroup_ssid_enabled(int ssid
)
231 if (CGROUP_SUBSYS_COUNT
== 0)
234 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
238 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
239 * @cgrp: the cgroup of interest
241 * The default hierarchy is the v2 interface of cgroup and this function
242 * can be used to test whether a cgroup is on the default hierarchy for
243 * cases where a subsystem should behave differnetly depending on the
246 * The set of behaviors which change on the default hierarchy are still
247 * being determined and the mount option is prefixed with __DEVEL__.
249 * List of changed behaviors:
251 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
252 * and "name" are disallowed.
254 * - When mounting an existing superblock, mount options should match.
256 * - Remount is disallowed.
258 * - rename(2) is disallowed.
260 * - "tasks" is removed. Everything should be at process granularity. Use
261 * "cgroup.procs" instead.
263 * - "cgroup.procs" is not sorted. pids will be unique unless they got
264 * recycled inbetween reads.
266 * - "release_agent" and "notify_on_release" are removed. Replacement
267 * notification mechanism will be implemented.
269 * - "cgroup.clone_children" is removed.
271 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
272 * and its descendants contain no task; otherwise, 1. The file also
273 * generates kernfs notification which can be monitored through poll and
274 * [di]notify when the value of the file changes.
276 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
277 * take masks of ancestors with non-empty cpus/mems, instead of being
278 * moved to an ancestor.
280 * - cpuset: a task can be moved into an empty cpuset, and again it takes
281 * masks of ancestors.
283 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
286 * - blkcg: blk-throttle becomes properly hierarchical.
288 * - debug: disallowed on the default hierarchy.
290 bool cgroup_on_dfl(const struct cgroup
*cgrp
)
292 return cgrp
->root
== &cgrp_dfl_root
;
295 /* IDR wrappers which synchronize using cgroup_idr_lock */
296 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
301 idr_preload(gfp_mask
);
302 spin_lock_bh(&cgroup_idr_lock
);
303 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
304 spin_unlock_bh(&cgroup_idr_lock
);
309 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
313 spin_lock_bh(&cgroup_idr_lock
);
314 ret
= idr_replace(idr
, ptr
, id
);
315 spin_unlock_bh(&cgroup_idr_lock
);
319 static void cgroup_idr_remove(struct idr
*idr
, int id
)
321 spin_lock_bh(&cgroup_idr_lock
);
323 spin_unlock_bh(&cgroup_idr_lock
);
326 static bool cgroup_has_tasks(struct cgroup
*cgrp
)
328 return cgrp
->nr_populated_csets
;
331 bool cgroup_is_threaded(struct cgroup
*cgrp
)
333 return cgrp
->dom_cgrp
!= cgrp
;
336 /* can @cgrp host both domain and threaded children? */
337 static bool cgroup_is_mixable(struct cgroup
*cgrp
)
340 * Root isn't under domain level resource control exempting it from
341 * the no-internal-process constraint, so it can serve as a thread
342 * root and a parent of resource domains at the same time.
344 return !cgroup_parent(cgrp
);
347 /* can @cgrp become a thread root? should always be true for a thread root */
348 static bool cgroup_can_be_thread_root(struct cgroup
*cgrp
)
350 /* mixables don't care */
351 if (cgroup_is_mixable(cgrp
))
354 /* domain roots can't be nested under threaded */
355 if (cgroup_is_threaded(cgrp
))
358 /* can only have either domain or threaded children */
359 if (cgrp
->nr_populated_domain_children
)
362 /* and no domain controllers can be enabled */
363 if (cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
369 /* is @cgrp root of a threaded subtree? */
370 bool cgroup_is_thread_root(struct cgroup
*cgrp
)
372 /* thread root should be a domain */
373 if (cgroup_is_threaded(cgrp
))
376 /* a domain w/ threaded children is a thread root */
377 if (cgrp
->nr_threaded_children
)
381 * A domain which has tasks and explicit threaded controllers
382 * enabled is a thread root.
384 if (cgroup_has_tasks(cgrp
) &&
385 (cgrp
->subtree_control
& cgrp_dfl_threaded_ss_mask
))
391 /* a domain which isn't connected to the root w/o brekage can't be used */
392 static bool cgroup_is_valid_domain(struct cgroup
*cgrp
)
394 /* the cgroup itself can be a thread root */
395 if (cgroup_is_threaded(cgrp
))
398 /* but the ancestors can't be unless mixable */
399 while ((cgrp
= cgroup_parent(cgrp
))) {
400 if (!cgroup_is_mixable(cgrp
) && cgroup_is_thread_root(cgrp
))
402 if (cgroup_is_threaded(cgrp
))
409 /* subsystems visibly enabled on a cgroup */
410 static u16
cgroup_control(struct cgroup
*cgrp
)
412 struct cgroup
*parent
= cgroup_parent(cgrp
);
413 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
416 u16 ss_mask
= parent
->subtree_control
;
418 /* threaded cgroups can only have threaded controllers */
419 if (cgroup_is_threaded(cgrp
))
420 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
424 if (cgroup_on_dfl(cgrp
))
425 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
426 cgrp_dfl_implicit_ss_mask
);
430 /* subsystems enabled on a cgroup */
431 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
433 struct cgroup
*parent
= cgroup_parent(cgrp
);
436 u16 ss_mask
= parent
->subtree_ss_mask
;
438 /* threaded cgroups can only have threaded controllers */
439 if (cgroup_is_threaded(cgrp
))
440 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
444 return cgrp
->root
->subsys_mask
;
448 * cgroup_css - obtain a cgroup's css for the specified subsystem
449 * @cgrp: the cgroup of interest
450 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
452 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
453 * function must be called either under cgroup_mutex or rcu_read_lock() and
454 * the caller is responsible for pinning the returned css if it wants to
455 * keep accessing it outside the said locks. This function may return
456 * %NULL if @cgrp doesn't have @subsys_id enabled.
458 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
459 struct cgroup_subsys
*ss
)
462 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
463 lockdep_is_held(&cgroup_mutex
));
469 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
470 * @cgrp: the cgroup of interest
471 * @ss: the subsystem of interest
473 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
474 * or is offline, %NULL is returned.
476 static struct cgroup_subsys_state
*cgroup_tryget_css(struct cgroup
*cgrp
,
477 struct cgroup_subsys
*ss
)
479 struct cgroup_subsys_state
*css
;
482 css
= cgroup_css(cgrp
, ss
);
483 if (!css
|| !css_tryget_online(css
))
491 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
492 * @cgrp: the cgroup of interest
493 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
495 * Similar to cgroup_css() but returns the effective css, which is defined
496 * as the matching css of the nearest ancestor including self which has @ss
497 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
498 * function is guaranteed to return non-NULL css.
500 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
501 struct cgroup_subsys
*ss
)
503 lockdep_assert_held(&cgroup_mutex
);
509 * This function is used while updating css associations and thus
510 * can't test the csses directly. Test ss_mask.
512 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
513 cgrp
= cgroup_parent(cgrp
);
518 return cgroup_css(cgrp
, ss
);
522 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
523 * @cgrp: the cgroup of interest
524 * @ss: the subsystem of interest
526 * Find and get the effective css of @cgrp for @ss. The effective css is
527 * defined as the matching css of the nearest ancestor including self which
528 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
529 * the root css is returned, so this function always returns a valid css.
530 * The returned css must be put using css_put().
532 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
533 struct cgroup_subsys
*ss
)
535 struct cgroup_subsys_state
*css
;
540 css
= cgroup_css(cgrp
, ss
);
542 if (css
&& css_tryget_online(css
))
544 cgrp
= cgroup_parent(cgrp
);
547 css
= init_css_set
.subsys
[ss
->id
];
554 static void cgroup_get_live(struct cgroup
*cgrp
)
556 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
557 css_get(&cgrp
->self
);
560 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
562 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
563 struct cftype
*cft
= of_cft(of
);
566 * This is open and unprotected implementation of cgroup_css().
567 * seq_css() is only called from a kernfs file operation which has
568 * an active reference on the file. Because all the subsystem
569 * files are drained before a css is disassociated with a cgroup,
570 * the matching css from the cgroup's subsys table is guaranteed to
571 * be and stay valid until the enclosing operation is complete.
574 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
578 EXPORT_SYMBOL_GPL(of_css
);
581 * for_each_css - iterate all css's of a cgroup
582 * @css: the iteration cursor
583 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
584 * @cgrp: the target cgroup to iterate css's of
586 * Should be called under cgroup_[tree_]mutex.
588 #define for_each_css(css, ssid, cgrp) \
589 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
590 if (!((css) = rcu_dereference_check( \
591 (cgrp)->subsys[(ssid)], \
592 lockdep_is_held(&cgroup_mutex)))) { } \
596 * for_each_e_css - iterate all effective css's of a cgroup
597 * @css: the iteration cursor
598 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
599 * @cgrp: the target cgroup to iterate css's of
601 * Should be called under cgroup_[tree_]mutex.
603 #define for_each_e_css(css, ssid, cgrp) \
604 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
605 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
610 * do_each_subsys_mask - filter for_each_subsys with a bitmask
611 * @ss: the iteration cursor
612 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
613 * @ss_mask: the bitmask
615 * The block will only run for cases where the ssid-th bit (1 << ssid) of
618 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
619 unsigned long __ss_mask = (ss_mask); \
620 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
624 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
625 (ss) = cgroup_subsys[ssid]; \
628 #define while_each_subsys_mask() \
633 /* iterate over child cgrps, lock should be held throughout iteration */
634 #define cgroup_for_each_live_child(child, cgrp) \
635 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
636 if (({ lockdep_assert_held(&cgroup_mutex); \
637 cgroup_is_dead(child); })) \
641 /* walk live descendants in preorder */
642 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
643 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
644 if (({ lockdep_assert_held(&cgroup_mutex); \
645 (dsct) = (d_css)->cgroup; \
646 cgroup_is_dead(dsct); })) \
650 /* walk live descendants in postorder */
651 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
652 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
653 if (({ lockdep_assert_held(&cgroup_mutex); \
654 (dsct) = (d_css)->cgroup; \
655 cgroup_is_dead(dsct); })) \
660 * The default css_set - used by init and its children prior to any
661 * hierarchies being mounted. It contains a pointer to the root state
662 * for each subsystem. Also used to anchor the list of css_sets. Not
663 * reference-counted, to improve performance when child cgroups
664 * haven't been created.
666 struct css_set init_css_set
= {
667 .refcount
= REFCOUNT_INIT(1),
668 .dom_cset
= &init_css_set
,
669 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
670 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
671 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
672 .threaded_csets
= LIST_HEAD_INIT(init_css_set
.threaded_csets
),
673 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
674 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
675 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
678 * The following field is re-initialized when this cset gets linked
679 * in cgroup_init(). However, let's initialize the field
680 * statically too so that the default cgroup can be accessed safely
683 .dfl_cgrp
= &cgrp_dfl_root
.cgrp
,
686 static int css_set_count
= 1; /* 1 for init_css_set */
688 static bool css_set_threaded(struct css_set
*cset
)
690 return cset
->dom_cset
!= cset
;
694 * css_set_populated - does a css_set contain any tasks?
695 * @cset: target css_set
697 * css_set_populated() should be the same as !!cset->nr_tasks at steady
698 * state. However, css_set_populated() can be called while a task is being
699 * added to or removed from the linked list before the nr_tasks is
700 * properly updated. Hence, we can't just look at ->nr_tasks here.
702 static bool css_set_populated(struct css_set
*cset
)
704 lockdep_assert_held(&css_set_lock
);
706 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
710 * cgroup_update_populated - update the populated count of a cgroup
711 * @cgrp: the target cgroup
712 * @populated: inc or dec populated count
714 * One of the css_sets associated with @cgrp is either getting its first
715 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
716 * count is propagated towards root so that a given cgroup's
717 * nr_populated_children is zero iff none of its descendants contain any
720 * @cgrp's interface file "cgroup.populated" is zero if both
721 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
722 * 1 otherwise. When the sum changes from or to zero, userland is notified
723 * that the content of the interface file has changed. This can be used to
724 * detect when @cgrp and its descendants become populated or empty.
726 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
728 struct cgroup
*child
= NULL
;
729 int adj
= populated
? 1 : -1;
731 lockdep_assert_held(&css_set_lock
);
734 bool was_populated
= cgroup_is_populated(cgrp
);
737 cgrp
->nr_populated_csets
+= adj
;
739 if (cgroup_is_threaded(child
))
740 cgrp
->nr_populated_threaded_children
+= adj
;
742 cgrp
->nr_populated_domain_children
+= adj
;
745 if (was_populated
== cgroup_is_populated(cgrp
))
748 cgroup1_check_for_release(cgrp
);
749 cgroup_file_notify(&cgrp
->events_file
);
752 cgrp
= cgroup_parent(cgrp
);
757 * css_set_update_populated - update populated state of a css_set
758 * @cset: target css_set
759 * @populated: whether @cset is populated or depopulated
761 * @cset is either getting the first task or losing the last. Update the
762 * populated counters of all associated cgroups accordingly.
764 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
766 struct cgrp_cset_link
*link
;
768 lockdep_assert_held(&css_set_lock
);
770 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
771 cgroup_update_populated(link
->cgrp
, populated
);
775 * css_set_move_task - move a task from one css_set to another
776 * @task: task being moved
777 * @from_cset: css_set @task currently belongs to (may be NULL)
778 * @to_cset: new css_set @task is being moved to (may be NULL)
779 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
781 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
782 * css_set, @from_cset can be NULL. If @task is being disassociated
783 * instead of moved, @to_cset can be NULL.
785 * This function automatically handles populated counter updates and
786 * css_task_iter adjustments but the caller is responsible for managing
787 * @from_cset and @to_cset's reference counts.
789 static void css_set_move_task(struct task_struct
*task
,
790 struct css_set
*from_cset
, struct css_set
*to_cset
,
793 lockdep_assert_held(&css_set_lock
);
795 if (to_cset
&& !css_set_populated(to_cset
))
796 css_set_update_populated(to_cset
, true);
799 struct css_task_iter
*it
, *pos
;
801 WARN_ON_ONCE(list_empty(&task
->cg_list
));
804 * @task is leaving, advance task iterators which are
805 * pointing to it so that they can resume at the next
806 * position. Advancing an iterator might remove it from
807 * the list, use safe walk. See css_task_iter_advance*()
810 list_for_each_entry_safe(it
, pos
, &from_cset
->task_iters
,
812 if (it
->task_pos
== &task
->cg_list
)
813 css_task_iter_advance(it
);
815 list_del_init(&task
->cg_list
);
816 if (!css_set_populated(from_cset
))
817 css_set_update_populated(from_cset
, false);
819 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
824 * We are synchronized through cgroup_threadgroup_rwsem
825 * against PF_EXITING setting such that we can't race
826 * against cgroup_exit() changing the css_set to
827 * init_css_set and dropping the old one.
829 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
831 rcu_assign_pointer(task
->cgroups
, to_cset
);
832 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
838 * hash table for cgroup groups. This improves the performance to find
839 * an existing css_set. This hash doesn't (currently) take into
840 * account cgroups in empty hierarchies.
842 #define CSS_SET_HASH_BITS 7
843 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
845 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
847 unsigned long key
= 0UL;
848 struct cgroup_subsys
*ss
;
851 for_each_subsys(ss
, i
)
852 key
+= (unsigned long)css
[i
];
853 key
= (key
>> 16) ^ key
;
858 void put_css_set_locked(struct css_set
*cset
)
860 struct cgrp_cset_link
*link
, *tmp_link
;
861 struct cgroup_subsys
*ss
;
864 lockdep_assert_held(&css_set_lock
);
866 if (!refcount_dec_and_test(&cset
->refcount
))
869 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
871 /* This css_set is dead. unlink it and release cgroup and css refs */
872 for_each_subsys(ss
, ssid
) {
873 list_del(&cset
->e_cset_node
[ssid
]);
874 css_put(cset
->subsys
[ssid
]);
876 hash_del(&cset
->hlist
);
879 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
880 list_del(&link
->cset_link
);
881 list_del(&link
->cgrp_link
);
882 if (cgroup_parent(link
->cgrp
))
883 cgroup_put(link
->cgrp
);
887 if (css_set_threaded(cset
)) {
888 list_del(&cset
->threaded_csets_node
);
889 put_css_set_locked(cset
->dom_cset
);
892 kfree_rcu(cset
, rcu_head
);
896 * compare_css_sets - helper function for find_existing_css_set().
897 * @cset: candidate css_set being tested
898 * @old_cset: existing css_set for a task
899 * @new_cgrp: cgroup that's being entered by the task
900 * @template: desired set of css pointers in css_set (pre-calculated)
902 * Returns true if "cset" matches "old_cset" except for the hierarchy
903 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
905 static bool compare_css_sets(struct css_set
*cset
,
906 struct css_set
*old_cset
,
907 struct cgroup
*new_cgrp
,
908 struct cgroup_subsys_state
*template[])
910 struct cgroup
*new_dfl_cgrp
;
911 struct list_head
*l1
, *l2
;
914 * On the default hierarchy, there can be csets which are
915 * associated with the same set of cgroups but different csses.
916 * Let's first ensure that csses match.
918 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
922 /* @cset's domain should match the default cgroup's */
923 if (cgroup_on_dfl(new_cgrp
))
924 new_dfl_cgrp
= new_cgrp
;
926 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
928 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
932 * Compare cgroup pointers in order to distinguish between
933 * different cgroups in hierarchies. As different cgroups may
934 * share the same effective css, this comparison is always
937 l1
= &cset
->cgrp_links
;
938 l2
= &old_cset
->cgrp_links
;
940 struct cgrp_cset_link
*link1
, *link2
;
941 struct cgroup
*cgrp1
, *cgrp2
;
945 /* See if we reached the end - both lists are equal length. */
946 if (l1
== &cset
->cgrp_links
) {
947 BUG_ON(l2
!= &old_cset
->cgrp_links
);
950 BUG_ON(l2
== &old_cset
->cgrp_links
);
952 /* Locate the cgroups associated with these links. */
953 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
954 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
957 /* Hierarchies should be linked in the same order. */
958 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
961 * If this hierarchy is the hierarchy of the cgroup
962 * that's changing, then we need to check that this
963 * css_set points to the new cgroup; if it's any other
964 * hierarchy, then this css_set should point to the
965 * same cgroup as the old css_set.
967 if (cgrp1
->root
== new_cgrp
->root
) {
968 if (cgrp1
!= new_cgrp
)
979 * find_existing_css_set - init css array and find the matching css_set
980 * @old_cset: the css_set that we're using before the cgroup transition
981 * @cgrp: the cgroup that we're moving into
982 * @template: out param for the new set of csses, should be clear on entry
984 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
986 struct cgroup_subsys_state
*template[])
988 struct cgroup_root
*root
= cgrp
->root
;
989 struct cgroup_subsys
*ss
;
990 struct css_set
*cset
;
995 * Build the set of subsystem state objects that we want to see in the
996 * new css_set. while subsystems can change globally, the entries here
997 * won't change, so no need for locking.
999 for_each_subsys(ss
, i
) {
1000 if (root
->subsys_mask
& (1UL << i
)) {
1002 * @ss is in this hierarchy, so we want the
1003 * effective css from @cgrp.
1005 template[i
] = cgroup_e_css(cgrp
, ss
);
1008 * @ss is not in this hierarchy, so we don't want
1009 * to change the css.
1011 template[i
] = old_cset
->subsys
[i
];
1015 key
= css_set_hash(template);
1016 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
1017 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
1020 /* This css_set matches what we need */
1024 /* No existing cgroup group matched */
1028 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
1030 struct cgrp_cset_link
*link
, *tmp_link
;
1032 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
1033 list_del(&link
->cset_link
);
1039 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1040 * @count: the number of links to allocate
1041 * @tmp_links: list_head the allocated links are put on
1043 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1044 * through ->cset_link. Returns 0 on success or -errno.
1046 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1048 struct cgrp_cset_link
*link
;
1051 INIT_LIST_HEAD(tmp_links
);
1053 for (i
= 0; i
< count
; i
++) {
1054 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1056 free_cgrp_cset_links(tmp_links
);
1059 list_add(&link
->cset_link
, tmp_links
);
1065 * link_css_set - a helper function to link a css_set to a cgroup
1066 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1067 * @cset: the css_set to be linked
1068 * @cgrp: the destination cgroup
1070 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1071 struct cgroup
*cgrp
)
1073 struct cgrp_cset_link
*link
;
1075 BUG_ON(list_empty(tmp_links
));
1077 if (cgroup_on_dfl(cgrp
))
1078 cset
->dfl_cgrp
= cgrp
;
1080 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1085 * Always add links to the tail of the lists so that the lists are
1086 * in choronological order.
1088 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1089 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1091 if (cgroup_parent(cgrp
))
1092 cgroup_get_live(cgrp
);
1096 * find_css_set - return a new css_set with one cgroup updated
1097 * @old_cset: the baseline css_set
1098 * @cgrp: the cgroup to be updated
1100 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1101 * substituted into the appropriate hierarchy.
1103 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1104 struct cgroup
*cgrp
)
1106 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1107 struct css_set
*cset
;
1108 struct list_head tmp_links
;
1109 struct cgrp_cset_link
*link
;
1110 struct cgroup_subsys
*ss
;
1114 lockdep_assert_held(&cgroup_mutex
);
1116 /* First see if we already have a cgroup group that matches
1117 * the desired set */
1118 spin_lock_irq(&css_set_lock
);
1119 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1122 spin_unlock_irq(&css_set_lock
);
1127 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1131 /* Allocate all the cgrp_cset_link objects that we'll need */
1132 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1137 refcount_set(&cset
->refcount
, 1);
1138 cset
->dom_cset
= cset
;
1139 INIT_LIST_HEAD(&cset
->tasks
);
1140 INIT_LIST_HEAD(&cset
->mg_tasks
);
1141 INIT_LIST_HEAD(&cset
->task_iters
);
1142 INIT_LIST_HEAD(&cset
->threaded_csets
);
1143 INIT_HLIST_NODE(&cset
->hlist
);
1144 INIT_LIST_HEAD(&cset
->cgrp_links
);
1145 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1146 INIT_LIST_HEAD(&cset
->mg_node
);
1148 /* Copy the set of subsystem state objects generated in
1149 * find_existing_css_set() */
1150 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1152 spin_lock_irq(&css_set_lock
);
1153 /* Add reference counts and links from the new css_set. */
1154 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1155 struct cgroup
*c
= link
->cgrp
;
1157 if (c
->root
== cgrp
->root
)
1159 link_css_set(&tmp_links
, cset
, c
);
1162 BUG_ON(!list_empty(&tmp_links
));
1166 /* Add @cset to the hash table */
1167 key
= css_set_hash(cset
->subsys
);
1168 hash_add(css_set_table
, &cset
->hlist
, key
);
1170 for_each_subsys(ss
, ssid
) {
1171 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1173 list_add_tail(&cset
->e_cset_node
[ssid
],
1174 &css
->cgroup
->e_csets
[ssid
]);
1178 spin_unlock_irq(&css_set_lock
);
1181 * If @cset should be threaded, look up the matching dom_cset and
1182 * link them up. We first fully initialize @cset then look for the
1183 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1184 * to stay empty until we return.
1186 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1187 struct css_set
*dcset
;
1189 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1195 spin_lock_irq(&css_set_lock
);
1196 cset
->dom_cset
= dcset
;
1197 list_add_tail(&cset
->threaded_csets_node
,
1198 &dcset
->threaded_csets
);
1199 spin_unlock_irq(&css_set_lock
);
1205 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1207 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1209 return root_cgrp
->root
;
1212 static int cgroup_init_root_id(struct cgroup_root
*root
)
1216 lockdep_assert_held(&cgroup_mutex
);
1218 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1222 root
->hierarchy_id
= id
;
1226 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1228 lockdep_assert_held(&cgroup_mutex
);
1230 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1233 void cgroup_free_root(struct cgroup_root
*root
)
1236 idr_destroy(&root
->cgroup_idr
);
1241 static void cgroup_destroy_root(struct cgroup_root
*root
)
1243 struct cgroup
*cgrp
= &root
->cgrp
;
1244 struct cgrp_cset_link
*link
, *tmp_link
;
1246 trace_cgroup_destroy_root(root
);
1248 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1250 BUG_ON(atomic_read(&root
->nr_cgrps
));
1251 BUG_ON(!list_empty(&cgrp
->self
.children
));
1253 /* Rebind all subsystems back to the default hierarchy */
1254 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1257 * Release all the links from cset_links to this hierarchy's
1260 spin_lock_irq(&css_set_lock
);
1262 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1263 list_del(&link
->cset_link
);
1264 list_del(&link
->cgrp_link
);
1268 spin_unlock_irq(&css_set_lock
);
1270 if (!list_empty(&root
->root_list
)) {
1271 list_del(&root
->root_list
);
1272 cgroup_root_count
--;
1275 cgroup_exit_root_id(root
);
1277 mutex_unlock(&cgroup_mutex
);
1279 kernfs_destroy_root(root
->kf_root
);
1280 cgroup_free_root(root
);
1284 * look up cgroup associated with current task's cgroup namespace on the
1285 * specified hierarchy
1287 static struct cgroup
*
1288 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1290 struct cgroup
*res
= NULL
;
1291 struct css_set
*cset
;
1293 lockdep_assert_held(&css_set_lock
);
1297 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1298 if (cset
== &init_css_set
) {
1301 struct cgrp_cset_link
*link
;
1303 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1304 struct cgroup
*c
= link
->cgrp
;
1306 if (c
->root
== root
) {
1318 /* look up cgroup associated with given css_set on the specified hierarchy */
1319 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1320 struct cgroup_root
*root
)
1322 struct cgroup
*res
= NULL
;
1324 lockdep_assert_held(&cgroup_mutex
);
1325 lockdep_assert_held(&css_set_lock
);
1327 if (cset
== &init_css_set
) {
1329 } else if (root
== &cgrp_dfl_root
) {
1330 res
= cset
->dfl_cgrp
;
1332 struct cgrp_cset_link
*link
;
1334 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1335 struct cgroup
*c
= link
->cgrp
;
1337 if (c
->root
== root
) {
1349 * Return the cgroup for "task" from the given hierarchy. Must be
1350 * called with cgroup_mutex and css_set_lock held.
1352 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1353 struct cgroup_root
*root
)
1356 * No need to lock the task - since we hold cgroup_mutex the
1357 * task can't change groups, so the only thing that can happen
1358 * is that it exits and its css is set back to init_css_set.
1360 return cset_cgroup_from_root(task_css_set(task
), root
);
1364 * A task must hold cgroup_mutex to modify cgroups.
1366 * Any task can increment and decrement the count field without lock.
1367 * So in general, code holding cgroup_mutex can't rely on the count
1368 * field not changing. However, if the count goes to zero, then only
1369 * cgroup_attach_task() can increment it again. Because a count of zero
1370 * means that no tasks are currently attached, therefore there is no
1371 * way a task attached to that cgroup can fork (the other way to
1372 * increment the count). So code holding cgroup_mutex can safely
1373 * assume that if the count is zero, it will stay zero. Similarly, if
1374 * a task holds cgroup_mutex on a cgroup with zero count, it
1375 * knows that the cgroup won't be removed, as cgroup_rmdir()
1378 * A cgroup can only be deleted if both its 'count' of using tasks
1379 * is zero, and its list of 'children' cgroups is empty. Since all
1380 * tasks in the system use _some_ cgroup, and since there is always at
1381 * least one task in the system (init, pid == 1), therefore, root cgroup
1382 * always has either children cgroups and/or using tasks. So we don't
1383 * need a special hack to ensure that root cgroup cannot be deleted.
1385 * P.S. One more locking exception. RCU is used to guard the
1386 * update of a tasks cgroup pointer by cgroup_attach_task()
1389 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1391 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1394 struct cgroup_subsys
*ss
= cft
->ss
;
1396 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1397 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1398 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1399 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1402 strscpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1407 * cgroup_file_mode - deduce file mode of a control file
1408 * @cft: the control file in question
1410 * S_IRUGO for read, S_IWUSR for write.
1412 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1416 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1419 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1420 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1430 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1431 * @subtree_control: the new subtree_control mask to consider
1432 * @this_ss_mask: available subsystems
1434 * On the default hierarchy, a subsystem may request other subsystems to be
1435 * enabled together through its ->depends_on mask. In such cases, more
1436 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1438 * This function calculates which subsystems need to be enabled if
1439 * @subtree_control is to be applied while restricted to @this_ss_mask.
1441 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1443 u16 cur_ss_mask
= subtree_control
;
1444 struct cgroup_subsys
*ss
;
1447 lockdep_assert_held(&cgroup_mutex
);
1449 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1452 u16 new_ss_mask
= cur_ss_mask
;
1454 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1455 new_ss_mask
|= ss
->depends_on
;
1456 } while_each_subsys_mask();
1459 * Mask out subsystems which aren't available. This can
1460 * happen only if some depended-upon subsystems were bound
1461 * to non-default hierarchies.
1463 new_ss_mask
&= this_ss_mask
;
1465 if (new_ss_mask
== cur_ss_mask
)
1467 cur_ss_mask
= new_ss_mask
;
1474 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1475 * @kn: the kernfs_node being serviced
1477 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1478 * the method finishes if locking succeeded. Note that once this function
1479 * returns the cgroup returned by cgroup_kn_lock_live() may become
1480 * inaccessible any time. If the caller intends to continue to access the
1481 * cgroup, it should pin it before invoking this function.
1483 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1485 struct cgroup
*cgrp
;
1487 if (kernfs_type(kn
) == KERNFS_DIR
)
1490 cgrp
= kn
->parent
->priv
;
1492 mutex_unlock(&cgroup_mutex
);
1494 kernfs_unbreak_active_protection(kn
);
1499 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1500 * @kn: the kernfs_node being serviced
1501 * @drain_offline: perform offline draining on the cgroup
1503 * This helper is to be used by a cgroup kernfs method currently servicing
1504 * @kn. It breaks the active protection, performs cgroup locking and
1505 * verifies that the associated cgroup is alive. Returns the cgroup if
1506 * alive; otherwise, %NULL. A successful return should be undone by a
1507 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1508 * cgroup is drained of offlining csses before return.
1510 * Any cgroup kernfs method implementation which requires locking the
1511 * associated cgroup should use this helper. It avoids nesting cgroup
1512 * locking under kernfs active protection and allows all kernfs operations
1513 * including self-removal.
1515 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1517 struct cgroup
*cgrp
;
1519 if (kernfs_type(kn
) == KERNFS_DIR
)
1522 cgrp
= kn
->parent
->priv
;
1525 * We're gonna grab cgroup_mutex which nests outside kernfs
1526 * active_ref. cgroup liveliness check alone provides enough
1527 * protection against removal. Ensure @cgrp stays accessible and
1528 * break the active_ref protection.
1530 if (!cgroup_tryget(cgrp
))
1532 kernfs_break_active_protection(kn
);
1535 cgroup_lock_and_drain_offline(cgrp
);
1537 mutex_lock(&cgroup_mutex
);
1539 if (!cgroup_is_dead(cgrp
))
1542 cgroup_kn_unlock(kn
);
1546 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1548 char name
[CGROUP_FILE_NAME_MAX
];
1550 lockdep_assert_held(&cgroup_mutex
);
1552 if (cft
->file_offset
) {
1553 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1554 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1556 spin_lock_irq(&cgroup_file_kn_lock
);
1558 spin_unlock_irq(&cgroup_file_kn_lock
);
1560 del_timer_sync(&cfile
->notify_timer
);
1563 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1567 * css_clear_dir - remove subsys files in a cgroup directory
1570 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1572 struct cgroup
*cgrp
= css
->cgroup
;
1573 struct cftype
*cfts
;
1575 if (!(css
->flags
& CSS_VISIBLE
))
1578 css
->flags
&= ~CSS_VISIBLE
;
1581 if (cgroup_on_dfl(cgrp
))
1582 cfts
= cgroup_base_files
;
1584 cfts
= cgroup1_base_files
;
1586 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1588 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1589 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1594 * css_populate_dir - create subsys files in a cgroup directory
1597 * On failure, no file is added.
1599 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1601 struct cgroup
*cgrp
= css
->cgroup
;
1602 struct cftype
*cfts
, *failed_cfts
;
1605 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1609 if (cgroup_on_dfl(cgrp
))
1610 cfts
= cgroup_base_files
;
1612 cfts
= cgroup1_base_files
;
1614 ret
= cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1618 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1619 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1627 css
->flags
|= CSS_VISIBLE
;
1631 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1632 if (cfts
== failed_cfts
)
1634 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1639 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1641 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1642 struct cgroup_subsys
*ss
;
1645 lockdep_assert_held(&cgroup_mutex
);
1647 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1649 * If @ss has non-root csses attached to it, can't move.
1650 * If @ss is an implicit controller, it is exempt from this
1651 * rule and can be stolen.
1653 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1654 !ss
->implicit_on_dfl
)
1657 /* can't move between two non-dummy roots either */
1658 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1660 } while_each_subsys_mask();
1662 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1663 struct cgroup_root
*src_root
= ss
->root
;
1664 struct cgroup
*scgrp
= &src_root
->cgrp
;
1665 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1666 struct css_set
*cset
;
1668 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1670 /* disable from the source */
1671 src_root
->subsys_mask
&= ~(1 << ssid
);
1672 WARN_ON(cgroup_apply_control(scgrp
));
1673 cgroup_finalize_control(scgrp
, 0);
1676 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1677 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1678 ss
->root
= dst_root
;
1679 css
->cgroup
= dcgrp
;
1681 spin_lock_irq(&css_set_lock
);
1682 hash_for_each(css_set_table
, i
, cset
, hlist
)
1683 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1684 &dcgrp
->e_csets
[ss
->id
]);
1685 spin_unlock_irq(&css_set_lock
);
1687 /* default hierarchy doesn't enable controllers by default */
1688 dst_root
->subsys_mask
|= 1 << ssid
;
1689 if (dst_root
== &cgrp_dfl_root
) {
1690 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1692 dcgrp
->subtree_control
|= 1 << ssid
;
1693 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1696 ret
= cgroup_apply_control(dcgrp
);
1698 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1703 } while_each_subsys_mask();
1705 kernfs_activate(dcgrp
->kn
);
1709 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1710 struct kernfs_root
*kf_root
)
1714 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1715 struct cgroup
*ns_cgroup
;
1717 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1721 spin_lock_irq(&css_set_lock
);
1722 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1723 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1724 spin_unlock_irq(&css_set_lock
);
1726 if (len
>= PATH_MAX
)
1729 seq_escape(sf
, buf
, " \t\n\\");
1736 static int parse_cgroup_root_flags(char *data
, unsigned int *root_flags
)
1745 while ((token
= strsep(&data
, ",")) != NULL
) {
1746 if (!strcmp(token
, "nsdelegate")) {
1747 *root_flags
|= CGRP_ROOT_NS_DELEGATE
;
1751 pr_err("cgroup2: unknown option \"%s\"\n", token
);
1758 static void apply_cgroup_root_flags(unsigned int root_flags
)
1760 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1761 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1762 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1764 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1768 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1770 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1771 seq_puts(seq
, ",nsdelegate");
1775 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1777 unsigned int root_flags
;
1780 ret
= parse_cgroup_root_flags(data
, &root_flags
);
1784 apply_cgroup_root_flags(root_flags
);
1789 * To reduce the fork() overhead for systems that are not actually using
1790 * their cgroups capability, we don't maintain the lists running through
1791 * each css_set to its tasks until we see the list actually used - in other
1792 * words after the first mount.
1794 static bool use_task_css_set_links __read_mostly
;
1796 static void cgroup_enable_task_cg_lists(void)
1798 struct task_struct
*p
, *g
;
1800 spin_lock_irq(&css_set_lock
);
1802 if (use_task_css_set_links
)
1805 use_task_css_set_links
= true;
1808 * We need tasklist_lock because RCU is not safe against
1809 * while_each_thread(). Besides, a forking task that has passed
1810 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1811 * is not guaranteed to have its child immediately visible in the
1812 * tasklist if we walk through it with RCU.
1814 read_lock(&tasklist_lock
);
1815 do_each_thread(g
, p
) {
1816 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1817 task_css_set(p
) != &init_css_set
);
1820 * We should check if the process is exiting, otherwise
1821 * it will race with cgroup_exit() in that the list
1822 * entry won't be deleted though the process has exited.
1823 * Do it while holding siglock so that we don't end up
1824 * racing against cgroup_exit().
1826 * Interrupts were already disabled while acquiring
1827 * the css_set_lock, so we do not need to disable it
1828 * again when acquiring the sighand->siglock here.
1830 spin_lock(&p
->sighand
->siglock
);
1831 if (!(p
->flags
& PF_EXITING
)) {
1832 struct css_set
*cset
= task_css_set(p
);
1834 if (!css_set_populated(cset
))
1835 css_set_update_populated(cset
, true);
1836 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1840 spin_unlock(&p
->sighand
->siglock
);
1841 } while_each_thread(g
, p
);
1842 read_unlock(&tasklist_lock
);
1844 spin_unlock_irq(&css_set_lock
);
1847 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1849 struct cgroup_subsys
*ss
;
1852 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1853 INIT_LIST_HEAD(&cgrp
->self
.children
);
1854 INIT_LIST_HEAD(&cgrp
->cset_links
);
1855 INIT_LIST_HEAD(&cgrp
->pidlists
);
1856 mutex_init(&cgrp
->pidlist_mutex
);
1857 cgrp
->self
.cgroup
= cgrp
;
1858 cgrp
->self
.flags
|= CSS_ONLINE
;
1859 cgrp
->dom_cgrp
= cgrp
;
1860 cgrp
->max_descendants
= INT_MAX
;
1861 cgrp
->max_depth
= INT_MAX
;
1863 for_each_subsys(ss
, ssid
)
1864 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1866 init_waitqueue_head(&cgrp
->offline_waitq
);
1867 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1870 void init_cgroup_root(struct cgroup_root
*root
, struct cgroup_sb_opts
*opts
)
1872 struct cgroup
*cgrp
= &root
->cgrp
;
1874 INIT_LIST_HEAD(&root
->root_list
);
1875 atomic_set(&root
->nr_cgrps
, 1);
1877 init_cgroup_housekeeping(cgrp
);
1878 idr_init(&root
->cgroup_idr
);
1880 root
->flags
= opts
->flags
;
1881 if (opts
->release_agent
)
1882 strscpy(root
->release_agent_path
, opts
->release_agent
, PATH_MAX
);
1884 strscpy(root
->name
, opts
->name
, MAX_CGROUP_ROOT_NAMELEN
);
1885 if (opts
->cpuset_clone_children
)
1886 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1889 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
, int ref_flags
)
1891 LIST_HEAD(tmp_links
);
1892 struct cgroup
*root_cgrp
= &root
->cgrp
;
1893 struct kernfs_syscall_ops
*kf_sops
;
1894 struct css_set
*cset
;
1897 lockdep_assert_held(&cgroup_mutex
);
1899 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1902 root_cgrp
->id
= ret
;
1903 root_cgrp
->ancestor_ids
[0] = ret
;
1905 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
1906 ref_flags
, GFP_KERNEL
);
1911 * We're accessing css_set_count without locking css_set_lock here,
1912 * but that's OK - it can only be increased by someone holding
1913 * cgroup_lock, and that's us. Later rebinding may disable
1914 * controllers on the default hierarchy and thus create new csets,
1915 * which can't be more than the existing ones. Allocate 2x.
1917 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
1921 ret
= cgroup_init_root_id(root
);
1925 kf_sops
= root
== &cgrp_dfl_root
?
1926 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
1928 root
->kf_root
= kernfs_create_root(kf_sops
,
1929 KERNFS_ROOT_CREATE_DEACTIVATED
|
1930 KERNFS_ROOT_SUPPORT_EXPORTOP
,
1932 if (IS_ERR(root
->kf_root
)) {
1933 ret
= PTR_ERR(root
->kf_root
);
1936 root_cgrp
->kn
= root
->kf_root
->kn
;
1938 ret
= css_populate_dir(&root_cgrp
->self
);
1942 ret
= rebind_subsystems(root
, ss_mask
);
1946 ret
= cgroup_bpf_inherit(root_cgrp
);
1949 trace_cgroup_setup_root(root
);
1952 * There must be no failure case after here, since rebinding takes
1953 * care of subsystems' refcounts, which are explicitly dropped in
1954 * the failure exit path.
1956 list_add(&root
->root_list
, &cgroup_roots
);
1957 cgroup_root_count
++;
1960 * Link the root cgroup in this hierarchy into all the css_set
1963 spin_lock_irq(&css_set_lock
);
1964 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1965 link_css_set(&tmp_links
, cset
, root_cgrp
);
1966 if (css_set_populated(cset
))
1967 cgroup_update_populated(root_cgrp
, true);
1969 spin_unlock_irq(&css_set_lock
);
1971 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1972 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1974 kernfs_activate(root_cgrp
->kn
);
1979 kernfs_destroy_root(root
->kf_root
);
1980 root
->kf_root
= NULL
;
1982 cgroup_exit_root_id(root
);
1984 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1986 free_cgrp_cset_links(&tmp_links
);
1990 struct dentry
*cgroup_do_mount(struct file_system_type
*fs_type
, int flags
,
1991 struct cgroup_root
*root
, unsigned long magic
,
1992 struct cgroup_namespace
*ns
)
1994 struct dentry
*dentry
;
1997 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
, magic
, &new_sb
);
2000 * In non-init cgroup namespace, instead of root cgroup's dentry,
2001 * we return the dentry corresponding to the cgroupns->root_cgrp.
2003 if (!IS_ERR(dentry
) && ns
!= &init_cgroup_ns
) {
2004 struct dentry
*nsdentry
;
2005 struct cgroup
*cgrp
;
2007 mutex_lock(&cgroup_mutex
);
2008 spin_lock_irq(&css_set_lock
);
2010 cgrp
= cset_cgroup_from_root(ns
->root_cset
, root
);
2012 spin_unlock_irq(&css_set_lock
);
2013 mutex_unlock(&cgroup_mutex
);
2015 nsdentry
= kernfs_node_dentry(cgrp
->kn
, dentry
->d_sb
);
2020 if (IS_ERR(dentry
) || !new_sb
)
2021 cgroup_put(&root
->cgrp
);
2026 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
2027 int flags
, const char *unused_dev_name
,
2030 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
2031 struct dentry
*dentry
;
2036 /* Check if the caller has permission to mount. */
2037 if (!ns_capable(ns
->user_ns
, CAP_SYS_ADMIN
)) {
2039 return ERR_PTR(-EPERM
);
2043 * The first time anyone tries to mount a cgroup, enable the list
2044 * linking each css_set to its tasks and fix up all existing tasks.
2046 if (!use_task_css_set_links
)
2047 cgroup_enable_task_cg_lists();
2049 if (fs_type
== &cgroup2_fs_type
) {
2050 unsigned int root_flags
;
2052 ret
= parse_cgroup_root_flags(data
, &root_flags
);
2055 return ERR_PTR(ret
);
2058 cgrp_dfl_visible
= true;
2059 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2061 dentry
= cgroup_do_mount(&cgroup2_fs_type
, flags
, &cgrp_dfl_root
,
2062 CGROUP2_SUPER_MAGIC
, ns
);
2063 if (!IS_ERR(dentry
))
2064 apply_cgroup_root_flags(root_flags
);
2066 dentry
= cgroup1_mount(&cgroup_fs_type
, flags
, data
,
2067 CGROUP_SUPER_MAGIC
, ns
);
2074 static void cgroup_kill_sb(struct super_block
*sb
)
2076 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2077 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2080 * If @root doesn't have any mounts or children, start killing it.
2081 * This prevents new mounts by disabling percpu_ref_tryget_live().
2082 * cgroup_mount() may wait for @root's release.
2084 * And don't kill the default root.
2086 if (!list_empty(&root
->cgrp
.self
.children
) ||
2087 root
== &cgrp_dfl_root
)
2088 cgroup_put(&root
->cgrp
);
2090 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2095 struct file_system_type cgroup_fs_type
= {
2097 .mount
= cgroup_mount
,
2098 .kill_sb
= cgroup_kill_sb
,
2099 .fs_flags
= FS_USERNS_MOUNT
,
2102 static struct file_system_type cgroup2_fs_type
= {
2104 .mount
= cgroup_mount
,
2105 .kill_sb
= cgroup_kill_sb
,
2106 .fs_flags
= FS_USERNS_MOUNT
,
2109 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2110 struct cgroup_namespace
*ns
)
2112 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2114 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2117 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2118 struct cgroup_namespace
*ns
)
2122 mutex_lock(&cgroup_mutex
);
2123 spin_lock_irq(&css_set_lock
);
2125 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2127 spin_unlock_irq(&css_set_lock
);
2128 mutex_unlock(&cgroup_mutex
);
2132 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2135 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2136 * @task: target task
2137 * @buf: the buffer to write the path into
2138 * @buflen: the length of the buffer
2140 * Determine @task's cgroup on the first (the one with the lowest non-zero
2141 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2142 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2143 * cgroup controller callbacks.
2145 * Return value is the same as kernfs_path().
2147 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2149 struct cgroup_root
*root
;
2150 struct cgroup
*cgrp
;
2151 int hierarchy_id
= 1;
2154 mutex_lock(&cgroup_mutex
);
2155 spin_lock_irq(&css_set_lock
);
2157 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2160 cgrp
= task_cgroup_from_root(task
, root
);
2161 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2163 /* if no hierarchy exists, everyone is in "/" */
2164 ret
= strlcpy(buf
, "/", buflen
);
2167 spin_unlock_irq(&css_set_lock
);
2168 mutex_unlock(&cgroup_mutex
);
2171 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2174 * cgroup_migrate_add_task - add a migration target task to a migration context
2175 * @task: target task
2176 * @mgctx: target migration context
2178 * Add @task, which is a migration target, to @mgctx->tset. This function
2179 * becomes noop if @task doesn't need to be migrated. @task's css_set
2180 * should have been added as a migration source and @task->cg_list will be
2181 * moved from the css_set's tasks list to mg_tasks one.
2183 static void cgroup_migrate_add_task(struct task_struct
*task
,
2184 struct cgroup_mgctx
*mgctx
)
2186 struct css_set
*cset
;
2188 lockdep_assert_held(&css_set_lock
);
2190 /* @task either already exited or can't exit until the end */
2191 if (task
->flags
& PF_EXITING
)
2194 /* leave @task alone if post_fork() hasn't linked it yet */
2195 if (list_empty(&task
->cg_list
))
2198 cset
= task_css_set(task
);
2199 if (!cset
->mg_src_cgrp
)
2202 mgctx
->tset
.nr_tasks
++;
2204 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2205 if (list_empty(&cset
->mg_node
))
2206 list_add_tail(&cset
->mg_node
,
2207 &mgctx
->tset
.src_csets
);
2208 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2209 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2210 &mgctx
->tset
.dst_csets
);
2214 * cgroup_taskset_first - reset taskset and return the first task
2215 * @tset: taskset of interest
2216 * @dst_cssp: output variable for the destination css
2218 * @tset iteration is initialized and the first task is returned.
2220 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2221 struct cgroup_subsys_state
**dst_cssp
)
2223 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2224 tset
->cur_task
= NULL
;
2226 return cgroup_taskset_next(tset
, dst_cssp
);
2230 * cgroup_taskset_next - iterate to the next task in taskset
2231 * @tset: taskset of interest
2232 * @dst_cssp: output variable for the destination css
2234 * Return the next task in @tset. Iteration must have been initialized
2235 * with cgroup_taskset_first().
2237 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2238 struct cgroup_subsys_state
**dst_cssp
)
2240 struct css_set
*cset
= tset
->cur_cset
;
2241 struct task_struct
*task
= tset
->cur_task
;
2243 while (&cset
->mg_node
!= tset
->csets
) {
2245 task
= list_first_entry(&cset
->mg_tasks
,
2246 struct task_struct
, cg_list
);
2248 task
= list_next_entry(task
, cg_list
);
2250 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2251 tset
->cur_cset
= cset
;
2252 tset
->cur_task
= task
;
2255 * This function may be called both before and
2256 * after cgroup_taskset_migrate(). The two cases
2257 * can be distinguished by looking at whether @cset
2258 * has its ->mg_dst_cset set.
2260 if (cset
->mg_dst_cset
)
2261 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2263 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2268 cset
= list_next_entry(cset
, mg_node
);
2276 * cgroup_taskset_migrate - migrate a taskset
2277 * @mgctx: migration context
2279 * Migrate tasks in @mgctx as setup by migration preparation functions.
2280 * This function fails iff one of the ->can_attach callbacks fails and
2281 * guarantees that either all or none of the tasks in @mgctx are migrated.
2282 * @mgctx is consumed regardless of success.
2284 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2286 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2287 struct cgroup_subsys
*ss
;
2288 struct task_struct
*task
, *tmp_task
;
2289 struct css_set
*cset
, *tmp_cset
;
2290 int ssid
, failed_ssid
, ret
;
2292 /* check that we can legitimately attach to the cgroup */
2293 if (tset
->nr_tasks
) {
2294 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2295 if (ss
->can_attach
) {
2297 ret
= ss
->can_attach(tset
);
2300 goto out_cancel_attach
;
2303 } while_each_subsys_mask();
2307 * Now that we're guaranteed success, proceed to move all tasks to
2308 * the new cgroup. There are no failure cases after here, so this
2309 * is the commit point.
2311 spin_lock_irq(&css_set_lock
);
2312 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2313 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2314 struct css_set
*from_cset
= task_css_set(task
);
2315 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2317 get_css_set(to_cset
);
2318 to_cset
->nr_tasks
++;
2319 css_set_move_task(task
, from_cset
, to_cset
, true);
2320 put_css_set_locked(from_cset
);
2321 from_cset
->nr_tasks
--;
2324 spin_unlock_irq(&css_set_lock
);
2327 * Migration is committed, all target tasks are now on dst_csets.
2328 * Nothing is sensitive to fork() after this point. Notify
2329 * controllers that migration is complete.
2331 tset
->csets
= &tset
->dst_csets
;
2333 if (tset
->nr_tasks
) {
2334 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2339 } while_each_subsys_mask();
2343 goto out_release_tset
;
2346 if (tset
->nr_tasks
) {
2347 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2348 if (ssid
== failed_ssid
)
2350 if (ss
->cancel_attach
) {
2352 ss
->cancel_attach(tset
);
2354 } while_each_subsys_mask();
2357 spin_lock_irq(&css_set_lock
);
2358 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2359 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2360 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2361 list_del_init(&cset
->mg_node
);
2363 spin_unlock_irq(&css_set_lock
);
2366 * Re-initialize the cgroup_taskset structure in case it is reused
2367 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2371 tset
->csets
= &tset
->src_csets
;
2376 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2377 * @dst_cgrp: destination cgroup to test
2379 * On the default hierarchy, except for the mixable, (possible) thread root
2380 * and threaded cgroups, subtree_control must be zero for migration
2381 * destination cgroups with tasks so that child cgroups don't compete
2384 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2386 /* v1 doesn't have any restriction */
2387 if (!cgroup_on_dfl(dst_cgrp
))
2390 /* verify @dst_cgrp can host resources */
2391 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2394 /* mixables don't care */
2395 if (cgroup_is_mixable(dst_cgrp
))
2399 * If @dst_cgrp is already or can become a thread root or is
2400 * threaded, it doesn't matter.
2402 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2405 /* apply no-internal-process constraint */
2406 if (dst_cgrp
->subtree_control
)
2413 * cgroup_migrate_finish - cleanup after attach
2414 * @mgctx: migration context
2416 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2417 * those functions for details.
2419 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2421 LIST_HEAD(preloaded
);
2422 struct css_set
*cset
, *tmp_cset
;
2424 lockdep_assert_held(&cgroup_mutex
);
2426 spin_lock_irq(&css_set_lock
);
2428 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2429 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2431 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2432 cset
->mg_src_cgrp
= NULL
;
2433 cset
->mg_dst_cgrp
= NULL
;
2434 cset
->mg_dst_cset
= NULL
;
2435 list_del_init(&cset
->mg_preload_node
);
2436 put_css_set_locked(cset
);
2439 spin_unlock_irq(&css_set_lock
);
2443 * cgroup_migrate_add_src - add a migration source css_set
2444 * @src_cset: the source css_set to add
2445 * @dst_cgrp: the destination cgroup
2446 * @mgctx: migration context
2448 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2449 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2450 * up by cgroup_migrate_finish().
2452 * This function may be called without holding cgroup_threadgroup_rwsem
2453 * even if the target is a process. Threads may be created and destroyed
2454 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2455 * into play and the preloaded css_sets are guaranteed to cover all
2458 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2459 struct cgroup
*dst_cgrp
,
2460 struct cgroup_mgctx
*mgctx
)
2462 struct cgroup
*src_cgrp
;
2464 lockdep_assert_held(&cgroup_mutex
);
2465 lockdep_assert_held(&css_set_lock
);
2468 * If ->dead, @src_set is associated with one or more dead cgroups
2469 * and doesn't contain any migratable tasks. Ignore it early so
2470 * that the rest of migration path doesn't get confused by it.
2475 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2477 if (!list_empty(&src_cset
->mg_preload_node
))
2480 WARN_ON(src_cset
->mg_src_cgrp
);
2481 WARN_ON(src_cset
->mg_dst_cgrp
);
2482 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2483 WARN_ON(!list_empty(&src_cset
->mg_node
));
2485 src_cset
->mg_src_cgrp
= src_cgrp
;
2486 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2487 get_css_set(src_cset
);
2488 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2492 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2493 * @mgctx: migration context
2495 * Tasks are about to be moved and all the source css_sets have been
2496 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2497 * pins all destination css_sets, links each to its source, and append them
2498 * to @mgctx->preloaded_dst_csets.
2500 * This function must be called after cgroup_migrate_add_src() has been
2501 * called on each migration source css_set. After migration is performed
2502 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2505 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2507 struct css_set
*src_cset
, *tmp_cset
;
2509 lockdep_assert_held(&cgroup_mutex
);
2511 /* look up the dst cset for each src cset and link it to src */
2512 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2514 struct css_set
*dst_cset
;
2515 struct cgroup_subsys
*ss
;
2518 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2522 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2525 * If src cset equals dst, it's noop. Drop the src.
2526 * cgroup_migrate() will skip the cset too. Note that we
2527 * can't handle src == dst as some nodes are used by both.
2529 if (src_cset
== dst_cset
) {
2530 src_cset
->mg_src_cgrp
= NULL
;
2531 src_cset
->mg_dst_cgrp
= NULL
;
2532 list_del_init(&src_cset
->mg_preload_node
);
2533 put_css_set(src_cset
);
2534 put_css_set(dst_cset
);
2538 src_cset
->mg_dst_cset
= dst_cset
;
2540 if (list_empty(&dst_cset
->mg_preload_node
))
2541 list_add_tail(&dst_cset
->mg_preload_node
,
2542 &mgctx
->preloaded_dst_csets
);
2544 put_css_set(dst_cset
);
2546 for_each_subsys(ss
, ssid
)
2547 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2548 mgctx
->ss_mask
|= 1 << ssid
;
2553 cgroup_migrate_finish(mgctx
);
2558 * cgroup_migrate - migrate a process or task to a cgroup
2559 * @leader: the leader of the process or the task to migrate
2560 * @threadgroup: whether @leader points to the whole process or a single task
2561 * @mgctx: migration context
2563 * Migrate a process or task denoted by @leader. If migrating a process,
2564 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2565 * responsible for invoking cgroup_migrate_add_src() and
2566 * cgroup_migrate_prepare_dst() on the targets before invoking this
2567 * function and following up with cgroup_migrate_finish().
2569 * As long as a controller's ->can_attach() doesn't fail, this function is
2570 * guaranteed to succeed. This means that, excluding ->can_attach()
2571 * failure, when migrating multiple targets, the success or failure can be
2572 * decided for all targets by invoking group_migrate_prepare_dst() before
2573 * actually starting migrating.
2575 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2576 struct cgroup_mgctx
*mgctx
)
2578 struct task_struct
*task
;
2581 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2582 * already PF_EXITING could be freed from underneath us unless we
2583 * take an rcu_read_lock.
2585 spin_lock_irq(&css_set_lock
);
2589 cgroup_migrate_add_task(task
, mgctx
);
2592 } while_each_thread(leader
, task
);
2594 spin_unlock_irq(&css_set_lock
);
2596 return cgroup_migrate_execute(mgctx
);
2600 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2601 * @dst_cgrp: the cgroup to attach to
2602 * @leader: the task or the leader of the threadgroup to be attached
2603 * @threadgroup: attach the whole threadgroup?
2605 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2607 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2610 DEFINE_CGROUP_MGCTX(mgctx
);
2611 struct task_struct
*task
;
2614 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
2618 /* look up all src csets */
2619 spin_lock_irq(&css_set_lock
);
2623 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2626 } while_each_thread(leader
, task
);
2628 spin_unlock_irq(&css_set_lock
);
2630 /* prepare dst csets and commit */
2631 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2633 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2635 cgroup_migrate_finish(&mgctx
);
2638 trace_cgroup_attach_task(dst_cgrp
, leader
, threadgroup
);
2643 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
)
2644 __acquires(&cgroup_threadgroup_rwsem
)
2646 struct task_struct
*tsk
;
2649 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2650 return ERR_PTR(-EINVAL
);
2652 percpu_down_write(&cgroup_threadgroup_rwsem
);
2656 tsk
= find_task_by_vpid(pid
);
2658 tsk
= ERR_PTR(-ESRCH
);
2659 goto out_unlock_threadgroup
;
2666 tsk
= tsk
->group_leader
;
2669 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2670 * If userland migrates such a kthread to a non-root cgroup, it can
2671 * become trapped in a cpuset, or RT kthread may be born in a
2672 * cgroup with no rt_runtime allocated. Just say no.
2674 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2675 tsk
= ERR_PTR(-EINVAL
);
2676 goto out_unlock_threadgroup
;
2679 get_task_struct(tsk
);
2680 goto out_unlock_rcu
;
2682 out_unlock_threadgroup
:
2683 percpu_up_write(&cgroup_threadgroup_rwsem
);
2689 void cgroup_procs_write_finish(struct task_struct
*task
)
2690 __releases(&cgroup_threadgroup_rwsem
)
2692 struct cgroup_subsys
*ss
;
2695 /* release reference from cgroup_procs_write_start() */
2696 put_task_struct(task
);
2698 percpu_up_write(&cgroup_threadgroup_rwsem
);
2699 for_each_subsys(ss
, ssid
)
2700 if (ss
->post_attach
)
2704 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2706 struct cgroup_subsys
*ss
;
2707 bool printed
= false;
2710 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2713 seq_printf(seq
, "%s", ss
->name
);
2715 } while_each_subsys_mask();
2717 seq_putc(seq
, '\n');
2720 /* show controllers which are enabled from the parent */
2721 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2723 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2725 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2729 /* show controllers which are enabled for a given cgroup's children */
2730 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2732 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2734 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2739 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2740 * @cgrp: root of the subtree to update csses for
2742 * @cgrp's control masks have changed and its subtree's css associations
2743 * need to be updated accordingly. This function looks up all css_sets
2744 * which are attached to the subtree, creates the matching updated css_sets
2745 * and migrates the tasks to the new ones.
2747 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2749 DEFINE_CGROUP_MGCTX(mgctx
);
2750 struct cgroup_subsys_state
*d_css
;
2751 struct cgroup
*dsct
;
2752 struct css_set
*src_cset
;
2755 lockdep_assert_held(&cgroup_mutex
);
2757 percpu_down_write(&cgroup_threadgroup_rwsem
);
2759 /* look up all csses currently attached to @cgrp's subtree */
2760 spin_lock_irq(&css_set_lock
);
2761 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2762 struct cgrp_cset_link
*link
;
2764 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2765 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2767 spin_unlock_irq(&css_set_lock
);
2769 /* NULL dst indicates self on default hierarchy */
2770 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2774 spin_lock_irq(&css_set_lock
);
2775 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2776 struct task_struct
*task
, *ntask
;
2778 /* all tasks in src_csets need to be migrated */
2779 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2780 cgroup_migrate_add_task(task
, &mgctx
);
2782 spin_unlock_irq(&css_set_lock
);
2784 ret
= cgroup_migrate_execute(&mgctx
);
2786 cgroup_migrate_finish(&mgctx
);
2787 percpu_up_write(&cgroup_threadgroup_rwsem
);
2792 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2793 * @cgrp: root of the target subtree
2795 * Because css offlining is asynchronous, userland may try to re-enable a
2796 * controller while the previous css is still around. This function grabs
2797 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2799 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
2800 __acquires(&cgroup_mutex
)
2802 struct cgroup
*dsct
;
2803 struct cgroup_subsys_state
*d_css
;
2804 struct cgroup_subsys
*ss
;
2808 mutex_lock(&cgroup_mutex
);
2810 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2811 for_each_subsys(ss
, ssid
) {
2812 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2815 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
2818 cgroup_get_live(dsct
);
2819 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
2820 TASK_UNINTERRUPTIBLE
);
2822 mutex_unlock(&cgroup_mutex
);
2824 finish_wait(&dsct
->offline_waitq
, &wait
);
2833 * cgroup_save_control - save control masks of a subtree
2834 * @cgrp: root of the target subtree
2836 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
2837 * prefixed fields for @cgrp's subtree including @cgrp itself.
2839 static void cgroup_save_control(struct cgroup
*cgrp
)
2841 struct cgroup
*dsct
;
2842 struct cgroup_subsys_state
*d_css
;
2844 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2845 dsct
->old_subtree_control
= dsct
->subtree_control
;
2846 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
2851 * cgroup_propagate_control - refresh control masks of a subtree
2852 * @cgrp: root of the target subtree
2854 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2855 * ->subtree_control and propagate controller availability through the
2856 * subtree so that descendants don't have unavailable controllers enabled.
2858 static void cgroup_propagate_control(struct cgroup
*cgrp
)
2860 struct cgroup
*dsct
;
2861 struct cgroup_subsys_state
*d_css
;
2863 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2864 dsct
->subtree_control
&= cgroup_control(dsct
);
2865 dsct
->subtree_ss_mask
=
2866 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
2867 cgroup_ss_mask(dsct
));
2872 * cgroup_restore_control - restore control masks of a subtree
2873 * @cgrp: root of the target subtree
2875 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
2876 * prefixed fields for @cgrp's subtree including @cgrp itself.
2878 static void cgroup_restore_control(struct cgroup
*cgrp
)
2880 struct cgroup
*dsct
;
2881 struct cgroup_subsys_state
*d_css
;
2883 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2884 dsct
->subtree_control
= dsct
->old_subtree_control
;
2885 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
2889 static bool css_visible(struct cgroup_subsys_state
*css
)
2891 struct cgroup_subsys
*ss
= css
->ss
;
2892 struct cgroup
*cgrp
= css
->cgroup
;
2894 if (cgroup_control(cgrp
) & (1 << ss
->id
))
2896 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
2898 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
2902 * cgroup_apply_control_enable - enable or show csses according to control
2903 * @cgrp: root of the target subtree
2905 * Walk @cgrp's subtree and create new csses or make the existing ones
2906 * visible. A css is created invisible if it's being implicitly enabled
2907 * through dependency. An invisible css is made visible when the userland
2908 * explicitly enables it.
2910 * Returns 0 on success, -errno on failure. On failure, csses which have
2911 * been processed already aren't cleaned up. The caller is responsible for
2912 * cleaning up with cgroup_apply_control_disable().
2914 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
2916 struct cgroup
*dsct
;
2917 struct cgroup_subsys_state
*d_css
;
2918 struct cgroup_subsys
*ss
;
2921 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2922 for_each_subsys(ss
, ssid
) {
2923 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2925 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
2927 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
2931 css
= css_create(dsct
, ss
);
2933 return PTR_ERR(css
);
2936 if (css_visible(css
)) {
2937 ret
= css_populate_dir(css
);
2948 * cgroup_apply_control_disable - kill or hide csses according to control
2949 * @cgrp: root of the target subtree
2951 * Walk @cgrp's subtree and kill and hide csses so that they match
2952 * cgroup_ss_mask() and cgroup_visible_mask().
2954 * A css is hidden when the userland requests it to be disabled while other
2955 * subsystems are still depending on it. The css must not actively control
2956 * resources and be in the vanilla state if it's made visible again later.
2957 * Controllers which may be depended upon should provide ->css_reset() for
2960 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
2962 struct cgroup
*dsct
;
2963 struct cgroup_subsys_state
*d_css
;
2964 struct cgroup_subsys
*ss
;
2967 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2968 for_each_subsys(ss
, ssid
) {
2969 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2971 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
2977 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
2979 } else if (!css_visible(css
)) {
2989 * cgroup_apply_control - apply control mask updates to the subtree
2990 * @cgrp: root of the target subtree
2992 * subsystems can be enabled and disabled in a subtree using the following
2995 * 1. Call cgroup_save_control() to stash the current state.
2996 * 2. Update ->subtree_control masks in the subtree as desired.
2997 * 3. Call cgroup_apply_control() to apply the changes.
2998 * 4. Optionally perform other related operations.
2999 * 5. Call cgroup_finalize_control() to finish up.
3001 * This function implements step 3 and propagates the mask changes
3002 * throughout @cgrp's subtree, updates csses accordingly and perform
3003 * process migrations.
3005 static int cgroup_apply_control(struct cgroup
*cgrp
)
3009 cgroup_propagate_control(cgrp
);
3011 ret
= cgroup_apply_control_enable(cgrp
);
3016 * At this point, cgroup_e_css() results reflect the new csses
3017 * making the following cgroup_update_dfl_csses() properly update
3018 * css associations of all tasks in the subtree.
3020 ret
= cgroup_update_dfl_csses(cgrp
);
3028 * cgroup_finalize_control - finalize control mask update
3029 * @cgrp: root of the target subtree
3030 * @ret: the result of the update
3032 * Finalize control mask update. See cgroup_apply_control() for more info.
3034 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3037 cgroup_restore_control(cgrp
);
3038 cgroup_propagate_control(cgrp
);
3041 cgroup_apply_control_disable(cgrp
);
3044 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3046 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3048 /* if nothing is getting enabled, nothing to worry about */
3052 /* can @cgrp host any resources? */
3053 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3056 /* mixables don't care */
3057 if (cgroup_is_mixable(cgrp
))
3060 if (domain_enable
) {
3061 /* can't enable domain controllers inside a thread subtree */
3062 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3066 * Threaded controllers can handle internal competitions
3067 * and are always allowed inside a (prospective) thread
3070 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3075 * Controllers can't be enabled for a cgroup with tasks to avoid
3076 * child cgroups competing against tasks.
3078 if (cgroup_has_tasks(cgrp
))
3084 /* change the enabled child controllers for a cgroup in the default hierarchy */
3085 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3086 char *buf
, size_t nbytes
,
3089 u16 enable
= 0, disable
= 0;
3090 struct cgroup
*cgrp
, *child
;
3091 struct cgroup_subsys
*ss
;
3096 * Parse input - space separated list of subsystem names prefixed
3097 * with either + or -.
3099 buf
= strstrip(buf
);
3100 while ((tok
= strsep(&buf
, " "))) {
3103 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3104 if (!cgroup_ssid_enabled(ssid
) ||
3105 strcmp(tok
+ 1, ss
->name
))
3109 enable
|= 1 << ssid
;
3110 disable
&= ~(1 << ssid
);
3111 } else if (*tok
== '-') {
3112 disable
|= 1 << ssid
;
3113 enable
&= ~(1 << ssid
);
3118 } while_each_subsys_mask();
3119 if (ssid
== CGROUP_SUBSYS_COUNT
)
3123 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3127 for_each_subsys(ss
, ssid
) {
3128 if (enable
& (1 << ssid
)) {
3129 if (cgrp
->subtree_control
& (1 << ssid
)) {
3130 enable
&= ~(1 << ssid
);
3134 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3138 } else if (disable
& (1 << ssid
)) {
3139 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3140 disable
&= ~(1 << ssid
);
3144 /* a child has it enabled? */
3145 cgroup_for_each_live_child(child
, cgrp
) {
3146 if (child
->subtree_control
& (1 << ssid
)) {
3154 if (!enable
&& !disable
) {
3159 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3163 /* save and update control masks and prepare csses */
3164 cgroup_save_control(cgrp
);
3166 cgrp
->subtree_control
|= enable
;
3167 cgrp
->subtree_control
&= ~disable
;
3169 ret
= cgroup_apply_control(cgrp
);
3170 cgroup_finalize_control(cgrp
, ret
);
3174 kernfs_activate(cgrp
->kn
);
3176 cgroup_kn_unlock(of
->kn
);
3177 return ret
?: nbytes
;
3181 * cgroup_enable_threaded - make @cgrp threaded
3182 * @cgrp: the target cgroup
3184 * Called when "threaded" is written to the cgroup.type interface file and
3185 * tries to make @cgrp threaded and join the parent's resource domain.
3186 * This function is never called on the root cgroup as cgroup.type doesn't
3189 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3191 struct cgroup
*parent
= cgroup_parent(cgrp
);
3192 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3195 lockdep_assert_held(&cgroup_mutex
);
3197 /* noop if already threaded */
3198 if (cgroup_is_threaded(cgrp
))
3202 * If @cgroup is populated or has domain controllers enabled, it
3203 * can't be switched. While the below cgroup_can_be_thread_root()
3204 * test can catch the same conditions, that's only when @parent is
3205 * not mixable, so let's check it explicitly.
3207 if (cgroup_is_populated(cgrp
) ||
3208 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3211 /* we're joining the parent's domain, ensure its validity */
3212 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3213 !cgroup_can_be_thread_root(dom_cgrp
))
3217 * The following shouldn't cause actual migrations and should
3220 cgroup_save_control(cgrp
);
3222 cgrp
->dom_cgrp
= dom_cgrp
;
3223 ret
= cgroup_apply_control(cgrp
);
3225 parent
->nr_threaded_children
++;
3227 cgrp
->dom_cgrp
= cgrp
;
3229 cgroup_finalize_control(cgrp
, ret
);
3233 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3235 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3237 if (cgroup_is_threaded(cgrp
))
3238 seq_puts(seq
, "threaded\n");
3239 else if (!cgroup_is_valid_domain(cgrp
))
3240 seq_puts(seq
, "domain invalid\n");
3241 else if (cgroup_is_thread_root(cgrp
))
3242 seq_puts(seq
, "domain threaded\n");
3244 seq_puts(seq
, "domain\n");
3249 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3250 size_t nbytes
, loff_t off
)
3252 struct cgroup
*cgrp
;
3255 /* only switching to threaded mode is supported */
3256 if (strcmp(strstrip(buf
), "threaded"))
3259 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3263 /* threaded can only be enabled */
3264 ret
= cgroup_enable_threaded(cgrp
);
3266 cgroup_kn_unlock(of
->kn
);
3267 return ret
?: nbytes
;
3270 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3272 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3273 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3275 if (descendants
== INT_MAX
)
3276 seq_puts(seq
, "max\n");
3278 seq_printf(seq
, "%d\n", descendants
);
3283 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3284 char *buf
, size_t nbytes
, loff_t off
)
3286 struct cgroup
*cgrp
;
3290 buf
= strstrip(buf
);
3291 if (!strcmp(buf
, "max")) {
3292 descendants
= INT_MAX
;
3294 ret
= kstrtoint(buf
, 0, &descendants
);
3299 if (descendants
< 0)
3302 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3306 cgrp
->max_descendants
= descendants
;
3308 cgroup_kn_unlock(of
->kn
);
3313 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3315 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3316 int depth
= READ_ONCE(cgrp
->max_depth
);
3318 if (depth
== INT_MAX
)
3319 seq_puts(seq
, "max\n");
3321 seq_printf(seq
, "%d\n", depth
);
3326 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3327 char *buf
, size_t nbytes
, loff_t off
)
3329 struct cgroup
*cgrp
;
3333 buf
= strstrip(buf
);
3334 if (!strcmp(buf
, "max")) {
3337 ret
= kstrtoint(buf
, 0, &depth
);
3345 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3349 cgrp
->max_depth
= depth
;
3351 cgroup_kn_unlock(of
->kn
);
3356 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3358 seq_printf(seq
, "populated %d\n",
3359 cgroup_is_populated(seq_css(seq
)->cgroup
));
3363 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3365 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3367 seq_printf(seq
, "nr_descendants %d\n",
3368 cgroup
->nr_descendants
);
3369 seq_printf(seq
, "nr_dying_descendants %d\n",
3370 cgroup
->nr_dying_descendants
);
3375 static int __maybe_unused
cgroup_extra_stat_show(struct seq_file
*seq
,
3376 struct cgroup
*cgrp
, int ssid
)
3378 struct cgroup_subsys
*ss
= cgroup_subsys
[ssid
];
3379 struct cgroup_subsys_state
*css
;
3382 if (!ss
->css_extra_stat_show
)
3385 css
= cgroup_tryget_css(cgrp
, ss
);
3389 ret
= ss
->css_extra_stat_show(seq
, css
);
3394 static int cpu_stat_show(struct seq_file
*seq
, void *v
)
3396 struct cgroup __maybe_unused
*cgrp
= seq_css(seq
)->cgroup
;
3399 cgroup_stat_show_cputime(seq
);
3400 #ifdef CONFIG_CGROUP_SCHED
3401 ret
= cgroup_extra_stat_show(seq
, cgrp
, cpu_cgrp_id
);
3406 static int cgroup_file_open(struct kernfs_open_file
*of
)
3408 struct cftype
*cft
= of
->kn
->priv
;
3411 return cft
->open(of
);
3415 static void cgroup_file_release(struct kernfs_open_file
*of
)
3417 struct cftype
*cft
= of
->kn
->priv
;
3423 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3424 size_t nbytes
, loff_t off
)
3426 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3427 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3428 struct cftype
*cft
= of
->kn
->priv
;
3429 struct cgroup_subsys_state
*css
;
3433 * If namespaces are delegation boundaries, disallow writes to
3434 * files in an non-init namespace root from inside the namespace
3435 * except for the files explicitly marked delegatable -
3436 * cgroup.procs and cgroup.subtree_control.
3438 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3439 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3440 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3444 return cft
->write(of
, buf
, nbytes
, off
);
3447 * kernfs guarantees that a file isn't deleted with operations in
3448 * flight, which means that the matching css is and stays alive and
3449 * doesn't need to be pinned. The RCU locking is not necessary
3450 * either. It's just for the convenience of using cgroup_css().
3453 css
= cgroup_css(cgrp
, cft
->ss
);
3456 if (cft
->write_u64
) {
3457 unsigned long long v
;
3458 ret
= kstrtoull(buf
, 0, &v
);
3460 ret
= cft
->write_u64(css
, cft
, v
);
3461 } else if (cft
->write_s64
) {
3463 ret
= kstrtoll(buf
, 0, &v
);
3465 ret
= cft
->write_s64(css
, cft
, v
);
3470 return ret
?: nbytes
;
3473 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3475 return seq_cft(seq
)->seq_start(seq
, ppos
);
3478 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3480 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3483 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3485 if (seq_cft(seq
)->seq_stop
)
3486 seq_cft(seq
)->seq_stop(seq
, v
);
3489 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3491 struct cftype
*cft
= seq_cft(m
);
3492 struct cgroup_subsys_state
*css
= seq_css(m
);
3495 return cft
->seq_show(m
, arg
);
3498 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3499 else if (cft
->read_s64
)
3500 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3506 static struct kernfs_ops cgroup_kf_single_ops
= {
3507 .atomic_write_len
= PAGE_SIZE
,
3508 .open
= cgroup_file_open
,
3509 .release
= cgroup_file_release
,
3510 .write
= cgroup_file_write
,
3511 .seq_show
= cgroup_seqfile_show
,
3514 static struct kernfs_ops cgroup_kf_ops
= {
3515 .atomic_write_len
= PAGE_SIZE
,
3516 .open
= cgroup_file_open
,
3517 .release
= cgroup_file_release
,
3518 .write
= cgroup_file_write
,
3519 .seq_start
= cgroup_seqfile_start
,
3520 .seq_next
= cgroup_seqfile_next
,
3521 .seq_stop
= cgroup_seqfile_stop
,
3522 .seq_show
= cgroup_seqfile_show
,
3525 /* set uid and gid of cgroup dirs and files to that of the creator */
3526 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3528 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3529 .ia_uid
= current_fsuid(),
3530 .ia_gid
= current_fsgid(), };
3532 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3533 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3536 return kernfs_setattr(kn
, &iattr
);
3539 static void cgroup_file_notify_timer(struct timer_list
*timer
)
3541 cgroup_file_notify(container_of(timer
, struct cgroup_file
,
3545 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3548 char name
[CGROUP_FILE_NAME_MAX
];
3549 struct kernfs_node
*kn
;
3550 struct lock_class_key
*key
= NULL
;
3553 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3554 key
= &cft
->lockdep_key
;
3556 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3557 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3562 ret
= cgroup_kn_set_ugid(kn
);
3568 if (cft
->file_offset
) {
3569 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3571 timer_setup(&cfile
->notify_timer
, cgroup_file_notify_timer
, 0);
3573 spin_lock_irq(&cgroup_file_kn_lock
);
3575 spin_unlock_irq(&cgroup_file_kn_lock
);
3582 * cgroup_addrm_files - add or remove files to a cgroup directory
3583 * @css: the target css
3584 * @cgrp: the target cgroup (usually css->cgroup)
3585 * @cfts: array of cftypes to be added
3586 * @is_add: whether to add or remove
3588 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3589 * For removals, this function never fails.
3591 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3592 struct cgroup
*cgrp
, struct cftype cfts
[],
3595 struct cftype
*cft
, *cft_end
= NULL
;
3598 lockdep_assert_held(&cgroup_mutex
);
3601 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3602 /* does cft->flags tell us to skip this file on @cgrp? */
3603 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3605 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3607 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3609 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3613 ret
= cgroup_add_file(css
, cgrp
, cft
);
3615 pr_warn("%s: failed to add %s, err=%d\n",
3616 __func__
, cft
->name
, ret
);
3622 cgroup_rm_file(cgrp
, cft
);
3628 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3630 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3631 struct cgroup
*root
= &ss
->root
->cgrp
;
3632 struct cgroup_subsys_state
*css
;
3635 lockdep_assert_held(&cgroup_mutex
);
3637 /* add/rm files for all cgroups created before */
3638 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3639 struct cgroup
*cgrp
= css
->cgroup
;
3641 if (!(css
->flags
& CSS_VISIBLE
))
3644 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3650 kernfs_activate(root
->kn
);
3654 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3658 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3659 /* free copy for custom atomic_write_len, see init_cftypes() */
3660 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3665 /* revert flags set by cgroup core while adding @cfts */
3666 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3670 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3674 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3675 struct kernfs_ops
*kf_ops
;
3677 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3680 kf_ops
= &cgroup_kf_ops
;
3682 kf_ops
= &cgroup_kf_single_ops
;
3685 * Ugh... if @cft wants a custom max_write_len, we need to
3686 * make a copy of kf_ops to set its atomic_write_len.
3688 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3689 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3691 cgroup_exit_cftypes(cfts
);
3694 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3697 cft
->kf_ops
= kf_ops
;
3704 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3706 lockdep_assert_held(&cgroup_mutex
);
3708 if (!cfts
|| !cfts
[0].ss
)
3711 list_del(&cfts
->node
);
3712 cgroup_apply_cftypes(cfts
, false);
3713 cgroup_exit_cftypes(cfts
);
3718 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3719 * @cfts: zero-length name terminated array of cftypes
3721 * Unregister @cfts. Files described by @cfts are removed from all
3722 * existing cgroups and all future cgroups won't have them either. This
3723 * function can be called anytime whether @cfts' subsys is attached or not.
3725 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3728 int cgroup_rm_cftypes(struct cftype
*cfts
)
3732 mutex_lock(&cgroup_mutex
);
3733 ret
= cgroup_rm_cftypes_locked(cfts
);
3734 mutex_unlock(&cgroup_mutex
);
3739 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3740 * @ss: target cgroup subsystem
3741 * @cfts: zero-length name terminated array of cftypes
3743 * Register @cfts to @ss. Files described by @cfts are created for all
3744 * existing cgroups to which @ss is attached and all future cgroups will
3745 * have them too. This function can be called anytime whether @ss is
3748 * Returns 0 on successful registration, -errno on failure. Note that this
3749 * function currently returns 0 as long as @cfts registration is successful
3750 * even if some file creation attempts on existing cgroups fail.
3752 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3756 if (!cgroup_ssid_enabled(ss
->id
))
3759 if (!cfts
|| cfts
[0].name
[0] == '\0')
3762 ret
= cgroup_init_cftypes(ss
, cfts
);
3766 mutex_lock(&cgroup_mutex
);
3768 list_add_tail(&cfts
->node
, &ss
->cfts
);
3769 ret
= cgroup_apply_cftypes(cfts
, true);
3771 cgroup_rm_cftypes_locked(cfts
);
3773 mutex_unlock(&cgroup_mutex
);
3778 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3779 * @ss: target cgroup subsystem
3780 * @cfts: zero-length name terminated array of cftypes
3782 * Similar to cgroup_add_cftypes() but the added files are only used for
3783 * the default hierarchy.
3785 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3789 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3790 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3791 return cgroup_add_cftypes(ss
, cfts
);
3795 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3796 * @ss: target cgroup subsystem
3797 * @cfts: zero-length name terminated array of cftypes
3799 * Similar to cgroup_add_cftypes() but the added files are only used for
3800 * the legacy hierarchies.
3802 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3806 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3807 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3808 return cgroup_add_cftypes(ss
, cfts
);
3812 * cgroup_file_notify - generate a file modified event for a cgroup_file
3813 * @cfile: target cgroup_file
3815 * @cfile must have been obtained by setting cftype->file_offset.
3817 void cgroup_file_notify(struct cgroup_file
*cfile
)
3819 unsigned long flags
;
3821 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
3823 unsigned long last
= cfile
->notified_at
;
3824 unsigned long next
= last
+ CGROUP_FILE_NOTIFY_MIN_INTV
;
3826 if (time_in_range(jiffies
, last
, next
)) {
3827 timer_reduce(&cfile
->notify_timer
, next
);
3829 kernfs_notify(cfile
->kn
);
3830 cfile
->notified_at
= jiffies
;
3833 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
3837 * css_next_child - find the next child of a given css
3838 * @pos: the current position (%NULL to initiate traversal)
3839 * @parent: css whose children to walk
3841 * This function returns the next child of @parent and should be called
3842 * under either cgroup_mutex or RCU read lock. The only requirement is
3843 * that @parent and @pos are accessible. The next sibling is guaranteed to
3844 * be returned regardless of their states.
3846 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3847 * css which finished ->css_online() is guaranteed to be visible in the
3848 * future iterations and will stay visible until the last reference is put.
3849 * A css which hasn't finished ->css_online() or already finished
3850 * ->css_offline() may show up during traversal. It's each subsystem's
3851 * responsibility to synchronize against on/offlining.
3853 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3854 struct cgroup_subsys_state
*parent
)
3856 struct cgroup_subsys_state
*next
;
3858 cgroup_assert_mutex_or_rcu_locked();
3861 * @pos could already have been unlinked from the sibling list.
3862 * Once a cgroup is removed, its ->sibling.next is no longer
3863 * updated when its next sibling changes. CSS_RELEASED is set when
3864 * @pos is taken off list, at which time its next pointer is valid,
3865 * and, as releases are serialized, the one pointed to by the next
3866 * pointer is guaranteed to not have started release yet. This
3867 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3868 * critical section, the one pointed to by its next pointer is
3869 * guaranteed to not have finished its RCU grace period even if we
3870 * have dropped rcu_read_lock() inbetween iterations.
3872 * If @pos has CSS_RELEASED set, its next pointer can't be
3873 * dereferenced; however, as each css is given a monotonically
3874 * increasing unique serial number and always appended to the
3875 * sibling list, the next one can be found by walking the parent's
3876 * children until the first css with higher serial number than
3877 * @pos's. While this path can be slower, it happens iff iteration
3878 * races against release and the race window is very small.
3881 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3882 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3883 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3885 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3886 if (next
->serial_nr
> pos
->serial_nr
)
3891 * @next, if not pointing to the head, can be dereferenced and is
3894 if (&next
->sibling
!= &parent
->children
)
3900 * css_next_descendant_pre - find the next descendant for pre-order walk
3901 * @pos: the current position (%NULL to initiate traversal)
3902 * @root: css whose descendants to walk
3904 * To be used by css_for_each_descendant_pre(). Find the next descendant
3905 * to visit for pre-order traversal of @root's descendants. @root is
3906 * included in the iteration and the first node to be visited.
3908 * While this function requires cgroup_mutex or RCU read locking, it
3909 * doesn't require the whole traversal to be contained in a single critical
3910 * section. This function will return the correct next descendant as long
3911 * as both @pos and @root are accessible and @pos is a descendant of @root.
3913 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3914 * css which finished ->css_online() is guaranteed to be visible in the
3915 * future iterations and will stay visible until the last reference is put.
3916 * A css which hasn't finished ->css_online() or already finished
3917 * ->css_offline() may show up during traversal. It's each subsystem's
3918 * responsibility to synchronize against on/offlining.
3920 struct cgroup_subsys_state
*
3921 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3922 struct cgroup_subsys_state
*root
)
3924 struct cgroup_subsys_state
*next
;
3926 cgroup_assert_mutex_or_rcu_locked();
3928 /* if first iteration, visit @root */
3932 /* visit the first child if exists */
3933 next
= css_next_child(NULL
, pos
);
3937 /* no child, visit my or the closest ancestor's next sibling */
3938 while (pos
!= root
) {
3939 next
= css_next_child(pos
, pos
->parent
);
3949 * css_rightmost_descendant - return the rightmost descendant of a css
3950 * @pos: css of interest
3952 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3953 * is returned. This can be used during pre-order traversal to skip
3956 * While this function requires cgroup_mutex or RCU read locking, it
3957 * doesn't require the whole traversal to be contained in a single critical
3958 * section. This function will return the correct rightmost descendant as
3959 * long as @pos is accessible.
3961 struct cgroup_subsys_state
*
3962 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3964 struct cgroup_subsys_state
*last
, *tmp
;
3966 cgroup_assert_mutex_or_rcu_locked();
3970 /* ->prev isn't RCU safe, walk ->next till the end */
3972 css_for_each_child(tmp
, last
)
3979 static struct cgroup_subsys_state
*
3980 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3982 struct cgroup_subsys_state
*last
;
3986 pos
= css_next_child(NULL
, pos
);
3993 * css_next_descendant_post - find the next descendant for post-order walk
3994 * @pos: the current position (%NULL to initiate traversal)
3995 * @root: css whose descendants to walk
3997 * To be used by css_for_each_descendant_post(). Find the next descendant
3998 * to visit for post-order traversal of @root's descendants. @root is
3999 * included in the iteration and the last node to be visited.
4001 * While this function requires cgroup_mutex or RCU read locking, it
4002 * doesn't require the whole traversal to be contained in a single critical
4003 * section. This function will return the correct next descendant as long
4004 * as both @pos and @cgroup are accessible and @pos is a descendant of
4007 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4008 * css which finished ->css_online() is guaranteed to be visible in the
4009 * future iterations and will stay visible until the last reference is put.
4010 * A css which hasn't finished ->css_online() or already finished
4011 * ->css_offline() may show up during traversal. It's each subsystem's
4012 * responsibility to synchronize against on/offlining.
4014 struct cgroup_subsys_state
*
4015 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4016 struct cgroup_subsys_state
*root
)
4018 struct cgroup_subsys_state
*next
;
4020 cgroup_assert_mutex_or_rcu_locked();
4022 /* if first iteration, visit leftmost descendant which may be @root */
4024 return css_leftmost_descendant(root
);
4026 /* if we visited @root, we're done */
4030 /* if there's an unvisited sibling, visit its leftmost descendant */
4031 next
= css_next_child(pos
, pos
->parent
);
4033 return css_leftmost_descendant(next
);
4035 /* no sibling left, visit parent */
4040 * css_has_online_children - does a css have online children
4041 * @css: the target css
4043 * Returns %true if @css has any online children; otherwise, %false. This
4044 * function can be called from any context but the caller is responsible
4045 * for synchronizing against on/offlining as necessary.
4047 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4049 struct cgroup_subsys_state
*child
;
4053 css_for_each_child(child
, css
) {
4054 if (child
->flags
& CSS_ONLINE
) {
4063 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
4065 struct list_head
*l
;
4066 struct cgrp_cset_link
*link
;
4067 struct css_set
*cset
;
4069 lockdep_assert_held(&css_set_lock
);
4071 /* find the next threaded cset */
4072 if (it
->tcset_pos
) {
4073 l
= it
->tcset_pos
->next
;
4075 if (l
!= it
->tcset_head
) {
4077 return container_of(l
, struct css_set
,
4078 threaded_csets_node
);
4081 it
->tcset_pos
= NULL
;
4084 /* find the next cset */
4087 if (l
== it
->cset_head
) {
4088 it
->cset_pos
= NULL
;
4093 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4095 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4101 /* initialize threaded css_set walking */
4102 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4104 put_css_set_locked(it
->cur_dcset
);
4105 it
->cur_dcset
= cset
;
4108 it
->tcset_head
= &cset
->threaded_csets
;
4109 it
->tcset_pos
= &cset
->threaded_csets
;
4116 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4117 * @it: the iterator to advance
4119 * Advance @it to the next css_set to walk.
4121 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4123 struct css_set
*cset
;
4125 lockdep_assert_held(&css_set_lock
);
4127 /* Advance to the next non-empty css_set */
4129 cset
= css_task_iter_next_css_set(it
);
4131 it
->task_pos
= NULL
;
4134 } while (!css_set_populated(cset
));
4136 if (!list_empty(&cset
->tasks
))
4137 it
->task_pos
= cset
->tasks
.next
;
4139 it
->task_pos
= cset
->mg_tasks
.next
;
4141 it
->tasks_head
= &cset
->tasks
;
4142 it
->mg_tasks_head
= &cset
->mg_tasks
;
4145 * We don't keep css_sets locked across iteration steps and thus
4146 * need to take steps to ensure that iteration can be resumed after
4147 * the lock is re-acquired. Iteration is performed at two levels -
4148 * css_sets and tasks in them.
4150 * Once created, a css_set never leaves its cgroup lists, so a
4151 * pinned css_set is guaranteed to stay put and we can resume
4152 * iteration afterwards.
4154 * Tasks may leave @cset across iteration steps. This is resolved
4155 * by registering each iterator with the css_set currently being
4156 * walked and making css_set_move_task() advance iterators whose
4157 * next task is leaving.
4160 list_del(&it
->iters_node
);
4161 put_css_set_locked(it
->cur_cset
);
4164 it
->cur_cset
= cset
;
4165 list_add(&it
->iters_node
, &cset
->task_iters
);
4168 static void css_task_iter_advance(struct css_task_iter
*it
)
4170 struct list_head
*next
;
4172 lockdep_assert_held(&css_set_lock
);
4175 * Advance iterator to find next entry. cset->tasks is consumed
4176 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4179 next
= it
->task_pos
->next
;
4181 if (next
== it
->tasks_head
)
4182 next
= it
->mg_tasks_head
->next
;
4184 if (next
== it
->mg_tasks_head
)
4185 css_task_iter_advance_css_set(it
);
4187 it
->task_pos
= next
;
4189 /* if PROCS, skip over tasks which aren't group leaders */
4190 if ((it
->flags
& CSS_TASK_ITER_PROCS
) && it
->task_pos
&&
4191 !thread_group_leader(list_entry(it
->task_pos
, struct task_struct
,
4197 * css_task_iter_start - initiate task iteration
4198 * @css: the css to walk tasks of
4199 * @flags: CSS_TASK_ITER_* flags
4200 * @it: the task iterator to use
4202 * Initiate iteration through the tasks of @css. The caller can call
4203 * css_task_iter_next() to walk through the tasks until the function
4204 * returns NULL. On completion of iteration, css_task_iter_end() must be
4207 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4208 struct css_task_iter
*it
)
4210 /* no one should try to iterate before mounting cgroups */
4211 WARN_ON_ONCE(!use_task_css_set_links
);
4213 memset(it
, 0, sizeof(*it
));
4215 spin_lock_irq(&css_set_lock
);
4221 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4223 it
->cset_pos
= &css
->cgroup
->cset_links
;
4225 it
->cset_head
= it
->cset_pos
;
4227 css_task_iter_advance_css_set(it
);
4229 spin_unlock_irq(&css_set_lock
);
4233 * css_task_iter_next - return the next task for the iterator
4234 * @it: the task iterator being iterated
4236 * The "next" function for task iteration. @it should have been
4237 * initialized via css_task_iter_start(). Returns NULL when the iteration
4240 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4243 put_task_struct(it
->cur_task
);
4244 it
->cur_task
= NULL
;
4247 spin_lock_irq(&css_set_lock
);
4250 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4252 get_task_struct(it
->cur_task
);
4253 css_task_iter_advance(it
);
4256 spin_unlock_irq(&css_set_lock
);
4258 return it
->cur_task
;
4262 * css_task_iter_end - finish task iteration
4263 * @it: the task iterator to finish
4265 * Finish task iteration started by css_task_iter_start().
4267 void css_task_iter_end(struct css_task_iter
*it
)
4270 spin_lock_irq(&css_set_lock
);
4271 list_del(&it
->iters_node
);
4272 put_css_set_locked(it
->cur_cset
);
4273 spin_unlock_irq(&css_set_lock
);
4277 put_css_set(it
->cur_dcset
);
4280 put_task_struct(it
->cur_task
);
4283 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4286 css_task_iter_end(of
->priv
);
4291 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4293 struct kernfs_open_file
*of
= s
->private;
4294 struct css_task_iter
*it
= of
->priv
;
4296 return css_task_iter_next(it
);
4299 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4300 unsigned int iter_flags
)
4302 struct kernfs_open_file
*of
= s
->private;
4303 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4304 struct css_task_iter
*it
= of
->priv
;
4307 * When a seq_file is seeked, it's always traversed sequentially
4308 * from position 0, so we can simply keep iterating on !0 *pos.
4311 if (WARN_ON_ONCE((*pos
)++))
4312 return ERR_PTR(-EINVAL
);
4314 it
= kzalloc(sizeof(*it
), GFP_KERNEL
);
4316 return ERR_PTR(-ENOMEM
);
4318 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4319 } else if (!(*pos
)++) {
4320 css_task_iter_end(it
);
4321 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4324 return cgroup_procs_next(s
, NULL
, NULL
);
4327 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4329 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4332 * All processes of a threaded subtree belong to the domain cgroup
4333 * of the subtree. Only threads can be distributed across the
4334 * subtree. Reject reads on cgroup.procs in the subtree proper.
4335 * They're always empty anyway.
4337 if (cgroup_is_threaded(cgrp
))
4338 return ERR_PTR(-EOPNOTSUPP
);
4340 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4341 CSS_TASK_ITER_THREADED
);
4344 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4346 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4350 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4351 struct cgroup
*dst_cgrp
,
4352 struct super_block
*sb
)
4354 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4355 struct cgroup
*com_cgrp
= src_cgrp
;
4356 struct inode
*inode
;
4359 lockdep_assert_held(&cgroup_mutex
);
4361 /* find the common ancestor */
4362 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4363 com_cgrp
= cgroup_parent(com_cgrp
);
4365 /* %current should be authorized to migrate to the common ancestor */
4366 inode
= kernfs_get_inode(sb
, com_cgrp
->procs_file
.kn
);
4370 ret
= inode_permission(inode
, MAY_WRITE
);
4376 * If namespaces are delegation boundaries, %current must be able
4377 * to see both source and destination cgroups from its namespace.
4379 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4380 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4381 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4387 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4388 char *buf
, size_t nbytes
, loff_t off
)
4390 struct cgroup
*src_cgrp
, *dst_cgrp
;
4391 struct task_struct
*task
;
4394 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4398 task
= cgroup_procs_write_start(buf
, true);
4399 ret
= PTR_ERR_OR_ZERO(task
);
4403 /* find the source cgroup */
4404 spin_lock_irq(&css_set_lock
);
4405 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4406 spin_unlock_irq(&css_set_lock
);
4408 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4409 of
->file
->f_path
.dentry
->d_sb
);
4413 ret
= cgroup_attach_task(dst_cgrp
, task
, true);
4416 cgroup_procs_write_finish(task
);
4418 cgroup_kn_unlock(of
->kn
);
4420 return ret
?: nbytes
;
4423 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4425 return __cgroup_procs_start(s
, pos
, 0);
4428 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4429 char *buf
, size_t nbytes
, loff_t off
)
4431 struct cgroup
*src_cgrp
, *dst_cgrp
;
4432 struct task_struct
*task
;
4435 buf
= strstrip(buf
);
4437 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4441 task
= cgroup_procs_write_start(buf
, false);
4442 ret
= PTR_ERR_OR_ZERO(task
);
4446 /* find the source cgroup */
4447 spin_lock_irq(&css_set_lock
);
4448 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4449 spin_unlock_irq(&css_set_lock
);
4451 /* thread migrations follow the cgroup.procs delegation rule */
4452 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4453 of
->file
->f_path
.dentry
->d_sb
);
4457 /* and must be contained in the same domain */
4459 if (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
)
4462 ret
= cgroup_attach_task(dst_cgrp
, task
, false);
4465 cgroup_procs_write_finish(task
);
4467 cgroup_kn_unlock(of
->kn
);
4469 return ret
?: nbytes
;
4472 /* cgroup core interface files for the default hierarchy */
4473 static struct cftype cgroup_base_files
[] = {
4475 .name
= "cgroup.type",
4476 .flags
= CFTYPE_NOT_ON_ROOT
,
4477 .seq_show
= cgroup_type_show
,
4478 .write
= cgroup_type_write
,
4481 .name
= "cgroup.procs",
4482 .flags
= CFTYPE_NS_DELEGATABLE
,
4483 .file_offset
= offsetof(struct cgroup
, procs_file
),
4484 .release
= cgroup_procs_release
,
4485 .seq_start
= cgroup_procs_start
,
4486 .seq_next
= cgroup_procs_next
,
4487 .seq_show
= cgroup_procs_show
,
4488 .write
= cgroup_procs_write
,
4491 .name
= "cgroup.threads",
4492 .flags
= CFTYPE_NS_DELEGATABLE
,
4493 .release
= cgroup_procs_release
,
4494 .seq_start
= cgroup_threads_start
,
4495 .seq_next
= cgroup_procs_next
,
4496 .seq_show
= cgroup_procs_show
,
4497 .write
= cgroup_threads_write
,
4500 .name
= "cgroup.controllers",
4501 .seq_show
= cgroup_controllers_show
,
4504 .name
= "cgroup.subtree_control",
4505 .flags
= CFTYPE_NS_DELEGATABLE
,
4506 .seq_show
= cgroup_subtree_control_show
,
4507 .write
= cgroup_subtree_control_write
,
4510 .name
= "cgroup.events",
4511 .flags
= CFTYPE_NOT_ON_ROOT
,
4512 .file_offset
= offsetof(struct cgroup
, events_file
),
4513 .seq_show
= cgroup_events_show
,
4516 .name
= "cgroup.max.descendants",
4517 .seq_show
= cgroup_max_descendants_show
,
4518 .write
= cgroup_max_descendants_write
,
4521 .name
= "cgroup.max.depth",
4522 .seq_show
= cgroup_max_depth_show
,
4523 .write
= cgroup_max_depth_write
,
4526 .name
= "cgroup.stat",
4527 .seq_show
= cgroup_stat_show
,
4531 .flags
= CFTYPE_NOT_ON_ROOT
,
4532 .seq_show
= cpu_stat_show
,
4538 * css destruction is four-stage process.
4540 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4541 * Implemented in kill_css().
4543 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4544 * and thus css_tryget_online() is guaranteed to fail, the css can be
4545 * offlined by invoking offline_css(). After offlining, the base ref is
4546 * put. Implemented in css_killed_work_fn().
4548 * 3. When the percpu_ref reaches zero, the only possible remaining
4549 * accessors are inside RCU read sections. css_release() schedules the
4552 * 4. After the grace period, the css can be freed. Implemented in
4553 * css_free_work_fn().
4555 * It is actually hairier because both step 2 and 4 require process context
4556 * and thus involve punting to css->destroy_work adding two additional
4557 * steps to the already complex sequence.
4559 static void css_free_rwork_fn(struct work_struct
*work
)
4561 struct cgroup_subsys_state
*css
= container_of(to_rcu_work(work
),
4562 struct cgroup_subsys_state
, destroy_rwork
);
4563 struct cgroup_subsys
*ss
= css
->ss
;
4564 struct cgroup
*cgrp
= css
->cgroup
;
4566 percpu_ref_exit(&css
->refcnt
);
4570 struct cgroup_subsys_state
*parent
= css
->parent
;
4574 cgroup_idr_remove(&ss
->css_idr
, id
);
4580 /* cgroup free path */
4581 atomic_dec(&cgrp
->root
->nr_cgrps
);
4582 cgroup1_pidlist_destroy_all(cgrp
);
4583 cancel_work_sync(&cgrp
->release_agent_work
);
4585 if (cgroup_parent(cgrp
)) {
4587 * We get a ref to the parent, and put the ref when
4588 * this cgroup is being freed, so it's guaranteed
4589 * that the parent won't be destroyed before its
4592 cgroup_put(cgroup_parent(cgrp
));
4593 kernfs_put(cgrp
->kn
);
4594 if (cgroup_on_dfl(cgrp
))
4595 cgroup_stat_exit(cgrp
);
4599 * This is root cgroup's refcnt reaching zero,
4600 * which indicates that the root should be
4603 cgroup_destroy_root(cgrp
->root
);
4608 static void css_release_work_fn(struct work_struct
*work
)
4610 struct cgroup_subsys_state
*css
=
4611 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4612 struct cgroup_subsys
*ss
= css
->ss
;
4613 struct cgroup
*cgrp
= css
->cgroup
;
4615 mutex_lock(&cgroup_mutex
);
4617 css
->flags
|= CSS_RELEASED
;
4618 list_del_rcu(&css
->sibling
);
4621 /* css release path */
4622 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4623 if (ss
->css_released
)
4624 ss
->css_released(css
);
4626 struct cgroup
*tcgrp
;
4628 /* cgroup release path */
4629 trace_cgroup_release(cgrp
);
4631 if (cgroup_on_dfl(cgrp
))
4632 cgroup_stat_flush(cgrp
);
4634 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
4635 tcgrp
= cgroup_parent(tcgrp
))
4636 tcgrp
->nr_dying_descendants
--;
4638 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4642 * There are two control paths which try to determine
4643 * cgroup from dentry without going through kernfs -
4644 * cgroupstats_build() and css_tryget_online_from_dir().
4645 * Those are supported by RCU protecting clearing of
4646 * cgrp->kn->priv backpointer.
4649 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
4652 cgroup_bpf_put(cgrp
);
4655 mutex_unlock(&cgroup_mutex
);
4657 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
4658 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
4661 static void css_release(struct percpu_ref
*ref
)
4663 struct cgroup_subsys_state
*css
=
4664 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4666 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4667 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4670 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4671 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4673 lockdep_assert_held(&cgroup_mutex
);
4675 cgroup_get_live(cgrp
);
4677 memset(css
, 0, sizeof(*css
));
4681 INIT_LIST_HEAD(&css
->sibling
);
4682 INIT_LIST_HEAD(&css
->children
);
4683 css
->serial_nr
= css_serial_nr_next
++;
4684 atomic_set(&css
->online_cnt
, 0);
4686 if (cgroup_parent(cgrp
)) {
4687 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4688 css_get(css
->parent
);
4691 BUG_ON(cgroup_css(cgrp
, ss
));
4694 /* invoke ->css_online() on a new CSS and mark it online if successful */
4695 static int online_css(struct cgroup_subsys_state
*css
)
4697 struct cgroup_subsys
*ss
= css
->ss
;
4700 lockdep_assert_held(&cgroup_mutex
);
4703 ret
= ss
->css_online(css
);
4705 css
->flags
|= CSS_ONLINE
;
4706 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4708 atomic_inc(&css
->online_cnt
);
4710 atomic_inc(&css
->parent
->online_cnt
);
4715 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4716 static void offline_css(struct cgroup_subsys_state
*css
)
4718 struct cgroup_subsys
*ss
= css
->ss
;
4720 lockdep_assert_held(&cgroup_mutex
);
4722 if (!(css
->flags
& CSS_ONLINE
))
4725 if (ss
->css_offline
)
4726 ss
->css_offline(css
);
4728 css
->flags
&= ~CSS_ONLINE
;
4729 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4731 wake_up_all(&css
->cgroup
->offline_waitq
);
4735 * css_create - create a cgroup_subsys_state
4736 * @cgrp: the cgroup new css will be associated with
4737 * @ss: the subsys of new css
4739 * Create a new css associated with @cgrp - @ss pair. On success, the new
4740 * css is online and installed in @cgrp. This function doesn't create the
4741 * interface files. Returns 0 on success, -errno on failure.
4743 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
4744 struct cgroup_subsys
*ss
)
4746 struct cgroup
*parent
= cgroup_parent(cgrp
);
4747 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4748 struct cgroup_subsys_state
*css
;
4751 lockdep_assert_held(&cgroup_mutex
);
4753 css
= ss
->css_alloc(parent_css
);
4755 css
= ERR_PTR(-ENOMEM
);
4759 init_and_link_css(css
, ss
, cgrp
);
4761 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4765 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4770 /* @css is ready to be brought online now, make it visible */
4771 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4772 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4774 err
= online_css(css
);
4778 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4779 cgroup_parent(parent
)) {
4780 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4781 current
->comm
, current
->pid
, ss
->name
);
4782 if (!strcmp(ss
->name
, "memory"))
4783 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4784 ss
->warned_broken_hierarchy
= true;
4790 list_del_rcu(&css
->sibling
);
4792 INIT_RCU_WORK(&css
->destroy_rwork
, css_free_rwork_fn
);
4793 queue_rcu_work(cgroup_destroy_wq
, &css
->destroy_rwork
);
4794 return ERR_PTR(err
);
4798 * The returned cgroup is fully initialized including its control mask, but
4799 * it isn't associated with its kernfs_node and doesn't have the control
4802 static struct cgroup
*cgroup_create(struct cgroup
*parent
)
4804 struct cgroup_root
*root
= parent
->root
;
4805 struct cgroup
*cgrp
, *tcgrp
;
4806 int level
= parent
->level
+ 1;
4809 /* allocate the cgroup and its ID, 0 is reserved for the root */
4810 cgrp
= kzalloc(sizeof(*cgrp
) +
4811 sizeof(cgrp
->ancestor_ids
[0]) * (level
+ 1), GFP_KERNEL
);
4813 return ERR_PTR(-ENOMEM
);
4815 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
4819 if (cgroup_on_dfl(parent
)) {
4820 ret
= cgroup_stat_init(cgrp
);
4822 goto out_cancel_ref
;
4826 * Temporarily set the pointer to NULL, so idr_find() won't return
4827 * a half-baked cgroup.
4829 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
4835 init_cgroup_housekeeping(cgrp
);
4837 cgrp
->self
.parent
= &parent
->self
;
4839 cgrp
->level
= level
;
4840 ret
= cgroup_bpf_inherit(cgrp
);
4844 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
4845 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
4848 tcgrp
->nr_descendants
++;
4851 if (notify_on_release(parent
))
4852 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4854 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4855 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4857 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4859 /* allocation complete, commit to creation */
4860 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4861 atomic_inc(&root
->nr_cgrps
);
4862 cgroup_get_live(parent
);
4865 * @cgrp is now fully operational. If something fails after this
4866 * point, it'll be released via the normal destruction path.
4868 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4871 * On the default hierarchy, a child doesn't automatically inherit
4872 * subtree_control from the parent. Each is configured manually.
4874 if (!cgroup_on_dfl(cgrp
))
4875 cgrp
->subtree_control
= cgroup_control(cgrp
);
4877 cgroup_propagate_control(cgrp
);
4882 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4884 if (cgroup_on_dfl(parent
))
4885 cgroup_stat_exit(cgrp
);
4887 percpu_ref_exit(&cgrp
->self
.refcnt
);
4890 return ERR_PTR(ret
);
4893 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
4895 struct cgroup
*cgroup
;
4899 lockdep_assert_held(&cgroup_mutex
);
4901 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
4902 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
4905 if (level
> cgroup
->max_depth
)
4916 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
4918 struct cgroup
*parent
, *cgrp
;
4919 struct kernfs_node
*kn
;
4922 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4923 if (strchr(name
, '\n'))
4926 parent
= cgroup_kn_lock_live(parent_kn
, false);
4930 if (!cgroup_check_hierarchy_limits(parent
)) {
4935 cgrp
= cgroup_create(parent
);
4937 ret
= PTR_ERR(cgrp
);
4941 /* create the directory */
4942 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4950 * This extra ref will be put in cgroup_free_fn() and guarantees
4951 * that @cgrp->kn is always accessible.
4955 ret
= cgroup_kn_set_ugid(kn
);
4959 ret
= css_populate_dir(&cgrp
->self
);
4963 ret
= cgroup_apply_control_enable(cgrp
);
4967 trace_cgroup_mkdir(cgrp
);
4969 /* let's create and online css's */
4970 kernfs_activate(kn
);
4976 cgroup_destroy_locked(cgrp
);
4978 cgroup_kn_unlock(parent_kn
);
4983 * This is called when the refcnt of a css is confirmed to be killed.
4984 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4985 * initate destruction and put the css ref from kill_css().
4987 static void css_killed_work_fn(struct work_struct
*work
)
4989 struct cgroup_subsys_state
*css
=
4990 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4992 mutex_lock(&cgroup_mutex
);
4997 /* @css can't go away while we're holding cgroup_mutex */
4999 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5001 mutex_unlock(&cgroup_mutex
);
5004 /* css kill confirmation processing requires process context, bounce */
5005 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5007 struct cgroup_subsys_state
*css
=
5008 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5010 if (atomic_dec_and_test(&css
->online_cnt
)) {
5011 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5012 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5017 * kill_css - destroy a css
5018 * @css: css to destroy
5020 * This function initiates destruction of @css by removing cgroup interface
5021 * files and putting its base reference. ->css_offline() will be invoked
5022 * asynchronously once css_tryget_online() is guaranteed to fail and when
5023 * the reference count reaches zero, @css will be released.
5025 static void kill_css(struct cgroup_subsys_state
*css
)
5027 lockdep_assert_held(&cgroup_mutex
);
5029 if (css
->flags
& CSS_DYING
)
5032 css
->flags
|= CSS_DYING
;
5035 * This must happen before css is disassociated with its cgroup.
5036 * See seq_css() for details.
5041 * Killing would put the base ref, but we need to keep it alive
5042 * until after ->css_offline().
5047 * cgroup core guarantees that, by the time ->css_offline() is
5048 * invoked, no new css reference will be given out via
5049 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5050 * proceed to offlining css's because percpu_ref_kill() doesn't
5051 * guarantee that the ref is seen as killed on all CPUs on return.
5053 * Use percpu_ref_kill_and_confirm() to get notifications as each
5054 * css is confirmed to be seen as killed on all CPUs.
5056 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5060 * cgroup_destroy_locked - the first stage of cgroup destruction
5061 * @cgrp: cgroup to be destroyed
5063 * css's make use of percpu refcnts whose killing latency shouldn't be
5064 * exposed to userland and are RCU protected. Also, cgroup core needs to
5065 * guarantee that css_tryget_online() won't succeed by the time
5066 * ->css_offline() is invoked. To satisfy all the requirements,
5067 * destruction is implemented in the following two steps.
5069 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5070 * userland visible parts and start killing the percpu refcnts of
5071 * css's. Set up so that the next stage will be kicked off once all
5072 * the percpu refcnts are confirmed to be killed.
5074 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5075 * rest of destruction. Once all cgroup references are gone, the
5076 * cgroup is RCU-freed.
5078 * This function implements s1. After this step, @cgrp is gone as far as
5079 * the userland is concerned and a new cgroup with the same name may be
5080 * created. As cgroup doesn't care about the names internally, this
5081 * doesn't cause any problem.
5083 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5084 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5086 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
5087 struct cgroup_subsys_state
*css
;
5088 struct cgrp_cset_link
*link
;
5091 lockdep_assert_held(&cgroup_mutex
);
5094 * Only migration can raise populated from zero and we're already
5095 * holding cgroup_mutex.
5097 if (cgroup_is_populated(cgrp
))
5101 * Make sure there's no live children. We can't test emptiness of
5102 * ->self.children as dead children linger on it while being
5103 * drained; otherwise, "rmdir parent/child parent" may fail.
5105 if (css_has_online_children(&cgrp
->self
))
5109 * Mark @cgrp and the associated csets dead. The former prevents
5110 * further task migration and child creation by disabling
5111 * cgroup_lock_live_group(). The latter makes the csets ignored by
5112 * the migration path.
5114 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5116 spin_lock_irq(&css_set_lock
);
5117 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5118 link
->cset
->dead
= true;
5119 spin_unlock_irq(&css_set_lock
);
5121 /* initiate massacre of all css's */
5122 for_each_css(css
, ssid
, cgrp
)
5125 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5126 css_clear_dir(&cgrp
->self
);
5127 kernfs_remove(cgrp
->kn
);
5129 if (parent
&& cgroup_is_threaded(cgrp
))
5130 parent
->nr_threaded_children
--;
5132 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5133 tcgrp
->nr_descendants
--;
5134 tcgrp
->nr_dying_descendants
++;
5137 cgroup1_check_for_release(parent
);
5139 /* put the base reference */
5140 percpu_ref_kill(&cgrp
->self
.refcnt
);
5145 int cgroup_rmdir(struct kernfs_node
*kn
)
5147 struct cgroup
*cgrp
;
5150 cgrp
= cgroup_kn_lock_live(kn
, false);
5154 ret
= cgroup_destroy_locked(cgrp
);
5157 trace_cgroup_rmdir(cgrp
);
5159 cgroup_kn_unlock(kn
);
5163 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5164 .show_options
= cgroup_show_options
,
5165 .remount_fs
= cgroup_remount
,
5166 .mkdir
= cgroup_mkdir
,
5167 .rmdir
= cgroup_rmdir
,
5168 .show_path
= cgroup_show_path
,
5171 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5173 struct cgroup_subsys_state
*css
;
5175 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5177 mutex_lock(&cgroup_mutex
);
5179 idr_init(&ss
->css_idr
);
5180 INIT_LIST_HEAD(&ss
->cfts
);
5182 /* Create the root cgroup state for this subsystem */
5183 ss
->root
= &cgrp_dfl_root
;
5184 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5185 /* We don't handle early failures gracefully */
5186 BUG_ON(IS_ERR(css
));
5187 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5190 * Root csses are never destroyed and we can't initialize
5191 * percpu_ref during early init. Disable refcnting.
5193 css
->flags
|= CSS_NO_REF
;
5196 /* allocation can't be done safely during early init */
5199 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5200 BUG_ON(css
->id
< 0);
5203 /* Update the init_css_set to contain a subsys
5204 * pointer to this state - since the subsystem is
5205 * newly registered, all tasks and hence the
5206 * init_css_set is in the subsystem's root cgroup. */
5207 init_css_set
.subsys
[ss
->id
] = css
;
5209 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5210 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5211 have_free_callback
|= (bool)ss
->free
<< ss
->id
;
5212 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5214 /* At system boot, before all subsystems have been
5215 * registered, no tasks have been forked, so we don't
5216 * need to invoke fork callbacks here. */
5217 BUG_ON(!list_empty(&init_task
.tasks
));
5219 BUG_ON(online_css(css
));
5221 mutex_unlock(&cgroup_mutex
);
5225 * cgroup_init_early - cgroup initialization at system boot
5227 * Initialize cgroups at system boot, and initialize any
5228 * subsystems that request early init.
5230 int __init
cgroup_init_early(void)
5232 static struct cgroup_sb_opts __initdata opts
;
5233 struct cgroup_subsys
*ss
;
5236 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5237 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5239 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5241 for_each_subsys(ss
, i
) {
5242 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5243 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5244 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5246 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5247 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5250 ss
->name
= cgroup_subsys_name
[i
];
5251 if (!ss
->legacy_name
)
5252 ss
->legacy_name
= cgroup_subsys_name
[i
];
5255 cgroup_init_subsys(ss
, true);
5260 static u16 cgroup_disable_mask __initdata
;
5263 * cgroup_init - cgroup initialization
5265 * Register cgroup filesystem and /proc file, and initialize
5266 * any subsystems that didn't request early init.
5268 int __init
cgroup_init(void)
5270 struct cgroup_subsys
*ss
;
5273 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5274 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5275 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5276 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5281 * The latency of the synchronize_sched() is too high for cgroups,
5282 * avoid it at the cost of forcing all readers into the slow path.
5284 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5286 get_user_ns(init_cgroup_ns
.user_ns
);
5288 mutex_lock(&cgroup_mutex
);
5291 * Add init_css_set to the hash table so that dfl_root can link to
5294 hash_add(css_set_table
, &init_css_set
.hlist
,
5295 css_set_hash(init_css_set
.subsys
));
5297 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0, 0));
5299 mutex_unlock(&cgroup_mutex
);
5301 for_each_subsys(ss
, ssid
) {
5302 if (ss
->early_init
) {
5303 struct cgroup_subsys_state
*css
=
5304 init_css_set
.subsys
[ss
->id
];
5306 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5308 BUG_ON(css
->id
< 0);
5310 cgroup_init_subsys(ss
, false);
5313 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5314 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5317 * Setting dfl_root subsys_mask needs to consider the
5318 * disabled flag and cftype registration needs kmalloc,
5319 * both of which aren't available during early_init.
5321 if (cgroup_disable_mask
& (1 << ssid
)) {
5322 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5323 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5328 if (cgroup1_ssid_disabled(ssid
))
5329 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5332 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5334 /* implicit controllers must be threaded too */
5335 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5337 if (ss
->implicit_on_dfl
)
5338 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5339 else if (!ss
->dfl_cftypes
)
5340 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5343 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5345 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5346 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5348 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5349 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5353 ss
->bind(init_css_set
.subsys
[ssid
]);
5355 mutex_lock(&cgroup_mutex
);
5356 css_populate_dir(init_css_set
.subsys
[ssid
]);
5357 mutex_unlock(&cgroup_mutex
);
5360 /* init_css_set.subsys[] has been updated, re-hash */
5361 hash_del(&init_css_set
.hlist
);
5362 hash_add(css_set_table
, &init_css_set
.hlist
,
5363 css_set_hash(init_css_set
.subsys
));
5365 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5366 WARN_ON(register_filesystem(&cgroup_fs_type
));
5367 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5368 WARN_ON(!proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
));
5373 static int __init
cgroup_wq_init(void)
5376 * There isn't much point in executing destruction path in
5377 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5378 * Use 1 for @max_active.
5380 * We would prefer to do this in cgroup_init() above, but that
5381 * is called before init_workqueues(): so leave this until after.
5383 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5384 BUG_ON(!cgroup_destroy_wq
);
5387 core_initcall(cgroup_wq_init
);
5389 void cgroup_path_from_kernfs_id(const union kernfs_node_id
*id
,
5390 char *buf
, size_t buflen
)
5392 struct kernfs_node
*kn
;
5394 kn
= kernfs_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5397 kernfs_path(kn
, buf
, buflen
);
5402 * proc_cgroup_show()
5403 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5404 * - Used for /proc/<pid>/cgroup.
5406 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5407 struct pid
*pid
, struct task_struct
*tsk
)
5411 struct cgroup_root
*root
;
5414 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5418 mutex_lock(&cgroup_mutex
);
5419 spin_lock_irq(&css_set_lock
);
5421 for_each_root(root
) {
5422 struct cgroup_subsys
*ss
;
5423 struct cgroup
*cgrp
;
5424 int ssid
, count
= 0;
5426 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5429 seq_printf(m
, "%d:", root
->hierarchy_id
);
5430 if (root
!= &cgrp_dfl_root
)
5431 for_each_subsys(ss
, ssid
)
5432 if (root
->subsys_mask
& (1 << ssid
))
5433 seq_printf(m
, "%s%s", count
++ ? "," : "",
5435 if (strlen(root
->name
))
5436 seq_printf(m
, "%sname=%s", count
? "," : "",
5440 cgrp
= task_cgroup_from_root(tsk
, root
);
5443 * On traditional hierarchies, all zombie tasks show up as
5444 * belonging to the root cgroup. On the default hierarchy,
5445 * while a zombie doesn't show up in "cgroup.procs" and
5446 * thus can't be migrated, its /proc/PID/cgroup keeps
5447 * reporting the cgroup it belonged to before exiting. If
5448 * the cgroup is removed before the zombie is reaped,
5449 * " (deleted)" is appended to the cgroup path.
5451 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5452 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5453 current
->nsproxy
->cgroup_ns
);
5454 if (retval
>= PATH_MAX
)
5455 retval
= -ENAMETOOLONG
;
5464 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5465 seq_puts(m
, " (deleted)\n");
5472 spin_unlock_irq(&css_set_lock
);
5473 mutex_unlock(&cgroup_mutex
);
5480 * cgroup_fork - initialize cgroup related fields during copy_process()
5481 * @child: pointer to task_struct of forking parent process.
5483 * A task is associated with the init_css_set until cgroup_post_fork()
5484 * attaches it to the parent's css_set. Empty cg_list indicates that
5485 * @child isn't holding reference to its css_set.
5487 void cgroup_fork(struct task_struct
*child
)
5489 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5490 INIT_LIST_HEAD(&child
->cg_list
);
5494 * cgroup_can_fork - called on a new task before the process is exposed
5495 * @child: the task in question.
5497 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5498 * returns an error, the fork aborts with that error code. This allows for
5499 * a cgroup subsystem to conditionally allow or deny new forks.
5501 int cgroup_can_fork(struct task_struct
*child
)
5503 struct cgroup_subsys
*ss
;
5506 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
5507 ret
= ss
->can_fork(child
);
5510 } while_each_subsys_mask();
5515 for_each_subsys(ss
, j
) {
5518 if (ss
->cancel_fork
)
5519 ss
->cancel_fork(child
);
5526 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5527 * @child: the task in question
5529 * This calls the cancel_fork() callbacks if a fork failed *after*
5530 * cgroup_can_fork() succeded.
5532 void cgroup_cancel_fork(struct task_struct
*child
)
5534 struct cgroup_subsys
*ss
;
5537 for_each_subsys(ss
, i
)
5538 if (ss
->cancel_fork
)
5539 ss
->cancel_fork(child
);
5543 * cgroup_post_fork - called on a new task after adding it to the task list
5544 * @child: the task in question
5546 * Adds the task to the list running through its css_set if necessary and
5547 * call the subsystem fork() callbacks. Has to be after the task is
5548 * visible on the task list in case we race with the first call to
5549 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5552 void cgroup_post_fork(struct task_struct
*child
)
5554 struct cgroup_subsys
*ss
;
5558 * This may race against cgroup_enable_task_cg_lists(). As that
5559 * function sets use_task_css_set_links before grabbing
5560 * tasklist_lock and we just went through tasklist_lock to add
5561 * @child, it's guaranteed that either we see the set
5562 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5563 * @child during its iteration.
5565 * If we won the race, @child is associated with %current's
5566 * css_set. Grabbing css_set_lock guarantees both that the
5567 * association is stable, and, on completion of the parent's
5568 * migration, @child is visible in the source of migration or
5569 * already in the destination cgroup. This guarantee is necessary
5570 * when implementing operations which need to migrate all tasks of
5571 * a cgroup to another.
5573 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5574 * will remain in init_css_set. This is safe because all tasks are
5575 * in the init_css_set before cg_links is enabled and there's no
5576 * operation which transfers all tasks out of init_css_set.
5578 if (use_task_css_set_links
) {
5579 struct css_set
*cset
;
5581 spin_lock_irq(&css_set_lock
);
5582 cset
= task_css_set(current
);
5583 if (list_empty(&child
->cg_list
)) {
5586 css_set_move_task(child
, NULL
, cset
, false);
5588 spin_unlock_irq(&css_set_lock
);
5592 * Call ss->fork(). This must happen after @child is linked on
5593 * css_set; otherwise, @child might change state between ->fork()
5594 * and addition to css_set.
5596 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
5598 } while_each_subsys_mask();
5602 * cgroup_exit - detach cgroup from exiting task
5603 * @tsk: pointer to task_struct of exiting process
5605 * Description: Detach cgroup from @tsk and release it.
5607 * Note that cgroups marked notify_on_release force every task in
5608 * them to take the global cgroup_mutex mutex when exiting.
5609 * This could impact scaling on very large systems. Be reluctant to
5610 * use notify_on_release cgroups where very high task exit scaling
5611 * is required on large systems.
5613 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5614 * call cgroup_exit() while the task is still competent to handle
5615 * notify_on_release(), then leave the task attached to the root cgroup in
5616 * each hierarchy for the remainder of its exit. No need to bother with
5617 * init_css_set refcnting. init_css_set never goes away and we can't race
5618 * with migration path - PF_EXITING is visible to migration path.
5620 void cgroup_exit(struct task_struct
*tsk
)
5622 struct cgroup_subsys
*ss
;
5623 struct css_set
*cset
;
5627 * Unlink from @tsk from its css_set. As migration path can't race
5628 * with us, we can check css_set and cg_list without synchronization.
5630 cset
= task_css_set(tsk
);
5632 if (!list_empty(&tsk
->cg_list
)) {
5633 spin_lock_irq(&css_set_lock
);
5634 css_set_move_task(tsk
, cset
, NULL
, false);
5636 spin_unlock_irq(&css_set_lock
);
5641 /* see cgroup_post_fork() for details */
5642 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
5644 } while_each_subsys_mask();
5647 void cgroup_free(struct task_struct
*task
)
5649 struct css_set
*cset
= task_css_set(task
);
5650 struct cgroup_subsys
*ss
;
5653 do_each_subsys_mask(ss
, ssid
, have_free_callback
) {
5655 } while_each_subsys_mask();
5660 static int __init
cgroup_disable(char *str
)
5662 struct cgroup_subsys
*ss
;
5666 while ((token
= strsep(&str
, ",")) != NULL
) {
5670 for_each_subsys(ss
, i
) {
5671 if (strcmp(token
, ss
->name
) &&
5672 strcmp(token
, ss
->legacy_name
))
5674 cgroup_disable_mask
|= 1 << i
;
5679 __setup("cgroup_disable=", cgroup_disable
);
5682 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5683 * @dentry: directory dentry of interest
5684 * @ss: subsystem of interest
5686 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5687 * to get the corresponding css and return it. If such css doesn't exist
5688 * or can't be pinned, an ERR_PTR value is returned.
5690 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5691 struct cgroup_subsys
*ss
)
5693 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5694 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
5695 struct cgroup_subsys_state
*css
= NULL
;
5696 struct cgroup
*cgrp
;
5698 /* is @dentry a cgroup dir? */
5699 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
5700 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
5701 return ERR_PTR(-EBADF
);
5706 * This path doesn't originate from kernfs and @kn could already
5707 * have been or be removed at any point. @kn->priv is RCU
5708 * protected for this access. See css_release_work_fn() for details.
5710 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
5712 css
= cgroup_css(cgrp
, ss
);
5714 if (!css
|| !css_tryget_online(css
))
5715 css
= ERR_PTR(-ENOENT
);
5722 * css_from_id - lookup css by id
5723 * @id: the cgroup id
5724 * @ss: cgroup subsys to be looked into
5726 * Returns the css if there's valid one with @id, otherwise returns NULL.
5727 * Should be called under rcu_read_lock().
5729 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5731 WARN_ON_ONCE(!rcu_read_lock_held());
5732 return idr_find(&ss
->css_idr
, id
);
5736 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5737 * @path: path on the default hierarchy
5739 * Find the cgroup at @path on the default hierarchy, increment its
5740 * reference count and return it. Returns pointer to the found cgroup on
5741 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5742 * if @path points to a non-directory.
5744 struct cgroup
*cgroup_get_from_path(const char *path
)
5746 struct kernfs_node
*kn
;
5747 struct cgroup
*cgrp
;
5749 mutex_lock(&cgroup_mutex
);
5751 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
5753 if (kernfs_type(kn
) == KERNFS_DIR
) {
5755 cgroup_get_live(cgrp
);
5757 cgrp
= ERR_PTR(-ENOTDIR
);
5761 cgrp
= ERR_PTR(-ENOENT
);
5764 mutex_unlock(&cgroup_mutex
);
5767 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
5770 * cgroup_get_from_fd - get a cgroup pointer from a fd
5771 * @fd: fd obtained by open(cgroup2_dir)
5773 * Find the cgroup from a fd which should be obtained
5774 * by opening a cgroup directory. Returns a pointer to the
5775 * cgroup on success. ERR_PTR is returned if the cgroup
5778 struct cgroup
*cgroup_get_from_fd(int fd
)
5780 struct cgroup_subsys_state
*css
;
5781 struct cgroup
*cgrp
;
5786 return ERR_PTR(-EBADF
);
5788 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
5791 return ERR_CAST(css
);
5794 if (!cgroup_on_dfl(cgrp
)) {
5796 return ERR_PTR(-EBADF
);
5801 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
5804 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5805 * definition in cgroup-defs.h.
5807 #ifdef CONFIG_SOCK_CGROUP_DATA
5809 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5811 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
5812 static bool cgroup_sk_alloc_disabled __read_mostly
;
5814 void cgroup_sk_alloc_disable(void)
5816 if (cgroup_sk_alloc_disabled
)
5818 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5819 cgroup_sk_alloc_disabled
= true;
5824 #define cgroup_sk_alloc_disabled false
5828 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
5830 if (cgroup_sk_alloc_disabled
)
5833 /* Socket clone path */
5836 * We might be cloning a socket which is left in an empty
5837 * cgroup and the cgroup might have already been rmdir'd.
5838 * Don't use cgroup_get_live().
5840 cgroup_get(sock_cgroup_ptr(skcd
));
5847 struct css_set
*cset
;
5849 cset
= task_css_set(current
);
5850 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
5851 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
5860 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
5862 cgroup_put(sock_cgroup_ptr(skcd
));
5865 #endif /* CONFIG_SOCK_CGROUP_DATA */
5867 #ifdef CONFIG_CGROUP_BPF
5868 int cgroup_bpf_attach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
5869 enum bpf_attach_type type
, u32 flags
)
5873 mutex_lock(&cgroup_mutex
);
5874 ret
= __cgroup_bpf_attach(cgrp
, prog
, type
, flags
);
5875 mutex_unlock(&cgroup_mutex
);
5878 int cgroup_bpf_detach(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
5879 enum bpf_attach_type type
, u32 flags
)
5883 mutex_lock(&cgroup_mutex
);
5884 ret
= __cgroup_bpf_detach(cgrp
, prog
, type
, flags
);
5885 mutex_unlock(&cgroup_mutex
);
5888 int cgroup_bpf_query(struct cgroup
*cgrp
, const union bpf_attr
*attr
,
5889 union bpf_attr __user
*uattr
)
5893 mutex_lock(&cgroup_mutex
);
5894 ret
= __cgroup_bpf_query(cgrp
, attr
, uattr
);
5895 mutex_unlock(&cgroup_mutex
);
5898 #endif /* CONFIG_CGROUP_BPF */
5901 static ssize_t
show_delegatable_files(struct cftype
*files
, char *buf
,
5902 ssize_t size
, const char *prefix
)
5907 for (cft
= files
; cft
&& cft
->name
[0] != '\0'; cft
++) {
5908 if (!(cft
->flags
& CFTYPE_NS_DELEGATABLE
))
5912 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s.", prefix
);
5914 ret
+= snprintf(buf
+ ret
, size
- ret
, "%s\n", cft
->name
);
5916 if (unlikely(ret
>= size
)) {
5925 static ssize_t
delegate_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
5928 struct cgroup_subsys
*ss
;
5932 ret
= show_delegatable_files(cgroup_base_files
, buf
, PAGE_SIZE
- ret
,
5935 for_each_subsys(ss
, ssid
)
5936 ret
+= show_delegatable_files(ss
->dfl_cftypes
, buf
+ ret
,
5938 cgroup_subsys_name
[ssid
]);
5942 static struct kobj_attribute cgroup_delegate_attr
= __ATTR_RO(delegate
);
5944 static ssize_t
features_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
5947 return snprintf(buf
, PAGE_SIZE
, "nsdelegate\n");
5949 static struct kobj_attribute cgroup_features_attr
= __ATTR_RO(features
);
5951 static struct attribute
*cgroup_sysfs_attrs
[] = {
5952 &cgroup_delegate_attr
.attr
,
5953 &cgroup_features_attr
.attr
,
5957 static const struct attribute_group cgroup_sysfs_attr_group
= {
5958 .attrs
= cgroup_sysfs_attrs
,
5962 static int __init
cgroup_sysfs_init(void)
5964 return sysfs_create_group(kernel_kobj
, &cgroup_sysfs_attr_group
);
5966 subsys_initcall(cgroup_sysfs_init
);
5967 #endif /* CONFIG_SYSFS */