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 <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/percpu-rwsem.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/kthread.h>
59 #include <linux/delay.h>
60 #include <linux/atomic.h>
61 #include <linux/cpuset.h>
62 #include <linux/proc_ns.h>
63 #include <linux/nsproxy.h>
64 #include <linux/proc_ns.h>
65 #include <linux/file.h>
69 * pidlists linger the following amount before being destroyed. The goal
70 * is avoiding frequent destruction in the middle of consecutive read calls
71 * Expiring in the middle is a performance problem not a correctness one.
72 * 1 sec should be enough.
74 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
76 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
80 * cgroup_mutex is the master lock. Any modification to cgroup or its
81 * hierarchy must be performed while holding it.
83 * css_set_lock protects task->cgroups pointer, the list of css_set
84 * objects, and the chain of tasks off each css_set.
86 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
87 * cgroup.h can use them for lockdep annotations.
89 #ifdef CONFIG_PROVE_RCU
90 DEFINE_MUTEX(cgroup_mutex
);
91 DEFINE_SPINLOCK(css_set_lock
);
92 EXPORT_SYMBOL_GPL(cgroup_mutex
);
93 EXPORT_SYMBOL_GPL(css_set_lock
);
95 static DEFINE_MUTEX(cgroup_mutex
);
96 static DEFINE_SPINLOCK(css_set_lock
);
100 * Protects cgroup_idr and css_idr so that IDs can be released without
101 * grabbing cgroup_mutex.
103 static DEFINE_SPINLOCK(cgroup_idr_lock
);
106 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
107 * against file removal/re-creation across css hiding.
109 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
112 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
113 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
115 static DEFINE_SPINLOCK(release_agent_path_lock
);
117 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
119 #define cgroup_assert_mutex_or_rcu_locked() \
120 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
121 !lockdep_is_held(&cgroup_mutex), \
122 "cgroup_mutex or RCU read lock required");
125 * cgroup destruction makes heavy use of work items and there can be a lot
126 * of concurrent destructions. Use a separate workqueue so that cgroup
127 * destruction work items don't end up filling up max_active of system_wq
128 * which may lead to deadlock.
130 static struct workqueue_struct
*cgroup_destroy_wq
;
133 * pidlist destructions need to be flushed on cgroup destruction. Use a
134 * separate workqueue as flush domain.
136 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
138 /* generate an array of cgroup subsystem pointers */
139 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
140 static struct cgroup_subsys
*cgroup_subsys
[] = {
141 #include <linux/cgroup_subsys.h>
145 /* array of cgroup subsystem names */
146 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
147 static const char *cgroup_subsys_name
[] = {
148 #include <linux/cgroup_subsys.h>
152 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
154 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
155 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
156 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
157 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
158 #include <linux/cgroup_subsys.h>
161 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
162 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
163 #include <linux/cgroup_subsys.h>
167 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
168 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
169 #include <linux/cgroup_subsys.h>
174 * The default hierarchy, reserved for the subsystems that are otherwise
175 * unattached - it never has more than a single cgroup, and all tasks are
176 * part of that cgroup.
178 struct cgroup_root cgrp_dfl_root
;
179 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
182 * The default hierarchy always exists but is hidden until mounted for the
183 * first time. This is for backward compatibility.
185 static bool cgrp_dfl_visible
;
187 /* Controllers blocked by the commandline in v1 */
188 static u16 cgroup_no_v1_mask
;
190 /* some controllers are not supported in the default hierarchy */
191 static u16 cgrp_dfl_inhibit_ss_mask
;
193 /* some controllers are implicitly enabled on the default hierarchy */
194 static unsigned long cgrp_dfl_implicit_ss_mask
;
196 /* The list of hierarchy roots */
198 static LIST_HEAD(cgroup_roots
);
199 static int cgroup_root_count
;
201 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
202 static DEFINE_IDR(cgroup_hierarchy_idr
);
205 * Assign a monotonically increasing serial number to csses. It guarantees
206 * cgroups with bigger numbers are newer than those with smaller numbers.
207 * Also, as csses are always appended to the parent's ->children list, it
208 * guarantees that sibling csses are always sorted in the ascending serial
209 * number order on the list. Protected by cgroup_mutex.
211 static u64 css_serial_nr_next
= 1;
214 * These bitmask flags indicate whether tasks in the fork and exit paths have
215 * fork/exit handlers to call. This avoids us having to do extra work in the
216 * fork/exit path to check which subsystems have fork/exit callbacks.
218 static u16 have_fork_callback __read_mostly
;
219 static u16 have_exit_callback __read_mostly
;
220 static u16 have_free_callback __read_mostly
;
222 /* cgroup namespace for init task */
223 struct cgroup_namespace init_cgroup_ns
= {
224 .count
= { .counter
= 2, },
225 .user_ns
= &init_user_ns
,
226 .ns
.ops
= &cgroupns_operations
,
227 .ns
.inum
= PROC_CGROUP_INIT_INO
,
228 .root_cset
= &init_css_set
,
231 /* Ditto for the can_fork callback. */
232 static u16 have_canfork_callback __read_mostly
;
234 static struct file_system_type cgroup2_fs_type
;
235 static struct cftype cgroup_dfl_base_files
[];
236 static struct cftype cgroup_legacy_base_files
[];
238 static int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
);
239 static void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
);
240 static int cgroup_apply_control(struct cgroup
*cgrp
);
241 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
242 static void css_task_iter_advance(struct css_task_iter
*it
);
243 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
244 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
245 struct cgroup_subsys
*ss
);
246 static void css_release(struct percpu_ref
*ref
);
247 static void kill_css(struct cgroup_subsys_state
*css
);
248 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
249 struct cgroup
*cgrp
, struct cftype cfts
[],
253 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
254 * @ssid: subsys ID of interest
256 * cgroup_subsys_enabled() can only be used with literal subsys names which
257 * is fine for individual subsystems but unsuitable for cgroup core. This
258 * is slower static_key_enabled() based test indexed by @ssid.
260 static bool cgroup_ssid_enabled(int ssid
)
262 if (CGROUP_SUBSYS_COUNT
== 0)
265 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
268 static bool cgroup_ssid_no_v1(int ssid
)
270 return cgroup_no_v1_mask
& (1 << ssid
);
274 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
275 * @cgrp: the cgroup of interest
277 * The default hierarchy is the v2 interface of cgroup and this function
278 * can be used to test whether a cgroup is on the default hierarchy for
279 * cases where a subsystem should behave differnetly depending on the
282 * The set of behaviors which change on the default hierarchy are still
283 * being determined and the mount option is prefixed with __DEVEL__.
285 * List of changed behaviors:
287 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
288 * and "name" are disallowed.
290 * - When mounting an existing superblock, mount options should match.
292 * - Remount is disallowed.
294 * - rename(2) is disallowed.
296 * - "tasks" is removed. Everything should be at process granularity. Use
297 * "cgroup.procs" instead.
299 * - "cgroup.procs" is not sorted. pids will be unique unless they got
300 * recycled inbetween reads.
302 * - "release_agent" and "notify_on_release" are removed. Replacement
303 * notification mechanism will be implemented.
305 * - "cgroup.clone_children" is removed.
307 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
308 * and its descendants contain no task; otherwise, 1. The file also
309 * generates kernfs notification which can be monitored through poll and
310 * [di]notify when the value of the file changes.
312 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
313 * take masks of ancestors with non-empty cpus/mems, instead of being
314 * moved to an ancestor.
316 * - cpuset: a task can be moved into an empty cpuset, and again it takes
317 * masks of ancestors.
319 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
322 * - blkcg: blk-throttle becomes properly hierarchical.
324 * - debug: disallowed on the default hierarchy.
326 static bool cgroup_on_dfl(const struct cgroup
*cgrp
)
328 return cgrp
->root
== &cgrp_dfl_root
;
331 /* IDR wrappers which synchronize using cgroup_idr_lock */
332 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
337 idr_preload(gfp_mask
);
338 spin_lock_bh(&cgroup_idr_lock
);
339 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
340 spin_unlock_bh(&cgroup_idr_lock
);
345 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
349 spin_lock_bh(&cgroup_idr_lock
);
350 ret
= idr_replace(idr
, ptr
, id
);
351 spin_unlock_bh(&cgroup_idr_lock
);
355 static void cgroup_idr_remove(struct idr
*idr
, int id
)
357 spin_lock_bh(&cgroup_idr_lock
);
359 spin_unlock_bh(&cgroup_idr_lock
);
362 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
364 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
367 return container_of(parent_css
, struct cgroup
, self
);
371 /* subsystems visibly enabled on a cgroup */
372 static u16
cgroup_control(struct cgroup
*cgrp
)
374 struct cgroup
*parent
= cgroup_parent(cgrp
);
375 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
378 return parent
->subtree_control
;
380 if (cgroup_on_dfl(cgrp
))
381 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
382 cgrp_dfl_implicit_ss_mask
);
386 /* subsystems enabled on a cgroup */
387 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
389 struct cgroup
*parent
= cgroup_parent(cgrp
);
392 return parent
->subtree_ss_mask
;
394 return cgrp
->root
->subsys_mask
;
398 * cgroup_css - obtain a cgroup's css for the specified subsystem
399 * @cgrp: the cgroup of interest
400 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
402 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
403 * function must be called either under cgroup_mutex or rcu_read_lock() and
404 * the caller is responsible for pinning the returned css if it wants to
405 * keep accessing it outside the said locks. This function may return
406 * %NULL if @cgrp doesn't have @subsys_id enabled.
408 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
409 struct cgroup_subsys
*ss
)
412 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
413 lockdep_is_held(&cgroup_mutex
));
419 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
420 * @cgrp: the cgroup of interest
421 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
423 * Similar to cgroup_css() but returns the effective css, which is defined
424 * as the matching css of the nearest ancestor including self which has @ss
425 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
426 * function is guaranteed to return non-NULL css.
428 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
429 struct cgroup_subsys
*ss
)
431 lockdep_assert_held(&cgroup_mutex
);
437 * This function is used while updating css associations and thus
438 * can't test the csses directly. Test ss_mask.
440 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
441 cgrp
= cgroup_parent(cgrp
);
446 return cgroup_css(cgrp
, ss
);
450 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
451 * @cgrp: the cgroup of interest
452 * @ss: the subsystem of interest
454 * Find and get the effective css of @cgrp for @ss. The effective css is
455 * defined as the matching css of the nearest ancestor including self which
456 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
457 * the root css is returned, so this function always returns a valid css.
458 * The returned css must be put using css_put().
460 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
461 struct cgroup_subsys
*ss
)
463 struct cgroup_subsys_state
*css
;
468 css
= cgroup_css(cgrp
, ss
);
470 if (css
&& css_tryget_online(css
))
472 cgrp
= cgroup_parent(cgrp
);
475 css
= init_css_set
.subsys
[ss
->id
];
482 /* convenient tests for these bits */
483 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
485 return !(cgrp
->self
.flags
& CSS_ONLINE
);
488 static void cgroup_get(struct cgroup
*cgrp
)
490 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
491 css_get(&cgrp
->self
);
494 static bool cgroup_tryget(struct cgroup
*cgrp
)
496 return css_tryget(&cgrp
->self
);
499 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
501 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
502 struct cftype
*cft
= of_cft(of
);
505 * This is open and unprotected implementation of cgroup_css().
506 * seq_css() is only called from a kernfs file operation which has
507 * an active reference on the file. Because all the subsystem
508 * files are drained before a css is disassociated with a cgroup,
509 * the matching css from the cgroup's subsys table is guaranteed to
510 * be and stay valid until the enclosing operation is complete.
513 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
517 EXPORT_SYMBOL_GPL(of_css
);
519 static int notify_on_release(const struct cgroup
*cgrp
)
521 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
525 * for_each_css - iterate all css's of a cgroup
526 * @css: the iteration cursor
527 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
528 * @cgrp: the target cgroup to iterate css's of
530 * Should be called under cgroup_[tree_]mutex.
532 #define for_each_css(css, ssid, cgrp) \
533 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
534 if (!((css) = rcu_dereference_check( \
535 (cgrp)->subsys[(ssid)], \
536 lockdep_is_held(&cgroup_mutex)))) { } \
540 * for_each_e_css - iterate all effective css's of a cgroup
541 * @css: the iteration cursor
542 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
543 * @cgrp: the target cgroup to iterate css's of
545 * Should be called under cgroup_[tree_]mutex.
547 #define for_each_e_css(css, ssid, cgrp) \
548 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
549 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
554 * for_each_subsys - iterate all enabled cgroup subsystems
555 * @ss: the iteration cursor
556 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
558 #define for_each_subsys(ss, ssid) \
559 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
560 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
563 * do_each_subsys_mask - filter for_each_subsys with a bitmask
564 * @ss: the iteration cursor
565 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
566 * @ss_mask: the bitmask
568 * The block will only run for cases where the ssid-th bit (1 << ssid) of
571 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
572 unsigned long __ss_mask = (ss_mask); \
573 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
577 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
578 (ss) = cgroup_subsys[ssid]; \
581 #define while_each_subsys_mask() \
586 /* iterate across the hierarchies */
587 #define for_each_root(root) \
588 list_for_each_entry((root), &cgroup_roots, root_list)
590 /* iterate over child cgrps, lock should be held throughout iteration */
591 #define cgroup_for_each_live_child(child, cgrp) \
592 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
593 if (({ lockdep_assert_held(&cgroup_mutex); \
594 cgroup_is_dead(child); })) \
598 /* walk live descendants in preorder */
599 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
600 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
601 if (({ lockdep_assert_held(&cgroup_mutex); \
602 (dsct) = (d_css)->cgroup; \
603 cgroup_is_dead(dsct); })) \
607 /* walk live descendants in postorder */
608 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
609 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
610 if (({ lockdep_assert_held(&cgroup_mutex); \
611 (dsct) = (d_css)->cgroup; \
612 cgroup_is_dead(dsct); })) \
616 static void cgroup_release_agent(struct work_struct
*work
);
617 static void check_for_release(struct cgroup
*cgrp
);
620 * A cgroup can be associated with multiple css_sets as different tasks may
621 * belong to different cgroups on different hierarchies. In the other
622 * direction, a css_set is naturally associated with multiple cgroups.
623 * This M:N relationship is represented by the following link structure
624 * which exists for each association and allows traversing the associations
627 struct cgrp_cset_link
{
628 /* the cgroup and css_set this link associates */
630 struct css_set
*cset
;
632 /* list of cgrp_cset_links anchored at cgrp->cset_links */
633 struct list_head cset_link
;
635 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
636 struct list_head cgrp_link
;
640 * The default css_set - used by init and its children prior to any
641 * hierarchies being mounted. It contains a pointer to the root state
642 * for each subsystem. Also used to anchor the list of css_sets. Not
643 * reference-counted, to improve performance when child cgroups
644 * haven't been created.
646 struct css_set init_css_set
= {
647 .refcount
= ATOMIC_INIT(1),
648 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
649 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
650 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
651 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
652 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
653 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
656 static int css_set_count
= 1; /* 1 for init_css_set */
659 * css_set_populated - does a css_set contain any tasks?
660 * @cset: target css_set
662 static bool css_set_populated(struct css_set
*cset
)
664 lockdep_assert_held(&css_set_lock
);
666 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
670 * cgroup_update_populated - updated populated count of a cgroup
671 * @cgrp: the target cgroup
672 * @populated: inc or dec populated count
674 * One of the css_sets associated with @cgrp is either getting its first
675 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
676 * count is propagated towards root so that a given cgroup's populated_cnt
677 * is zero iff the cgroup and all its descendants don't contain any tasks.
679 * @cgrp's interface file "cgroup.populated" is zero if
680 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
681 * changes from or to zero, userland is notified that the content of the
682 * interface file has changed. This can be used to detect when @cgrp and
683 * its descendants become populated or empty.
685 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
687 lockdep_assert_held(&css_set_lock
);
693 trigger
= !cgrp
->populated_cnt
++;
695 trigger
= !--cgrp
->populated_cnt
;
700 check_for_release(cgrp
);
701 cgroup_file_notify(&cgrp
->events_file
);
703 cgrp
= cgroup_parent(cgrp
);
708 * css_set_update_populated - update populated state of a css_set
709 * @cset: target css_set
710 * @populated: whether @cset is populated or depopulated
712 * @cset is either getting the first task or losing the last. Update the
713 * ->populated_cnt of all associated cgroups accordingly.
715 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
717 struct cgrp_cset_link
*link
;
719 lockdep_assert_held(&css_set_lock
);
721 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
722 cgroup_update_populated(link
->cgrp
, populated
);
726 * css_set_move_task - move a task from one css_set to another
727 * @task: task being moved
728 * @from_cset: css_set @task currently belongs to (may be NULL)
729 * @to_cset: new css_set @task is being moved to (may be NULL)
730 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
732 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
733 * css_set, @from_cset can be NULL. If @task is being disassociated
734 * instead of moved, @to_cset can be NULL.
736 * This function automatically handles populated_cnt updates and
737 * css_task_iter adjustments but the caller is responsible for managing
738 * @from_cset and @to_cset's reference counts.
740 static void css_set_move_task(struct task_struct
*task
,
741 struct css_set
*from_cset
, struct css_set
*to_cset
,
744 lockdep_assert_held(&css_set_lock
);
746 if (to_cset
&& !css_set_populated(to_cset
))
747 css_set_update_populated(to_cset
, true);
750 struct css_task_iter
*it
, *pos
;
752 WARN_ON_ONCE(list_empty(&task
->cg_list
));
755 * @task is leaving, advance task iterators which are
756 * pointing to it so that they can resume at the next
757 * position. Advancing an iterator might remove it from
758 * the list, use safe walk. See css_task_iter_advance*()
761 list_for_each_entry_safe(it
, pos
, &from_cset
->task_iters
,
763 if (it
->task_pos
== &task
->cg_list
)
764 css_task_iter_advance(it
);
766 list_del_init(&task
->cg_list
);
767 if (!css_set_populated(from_cset
))
768 css_set_update_populated(from_cset
, false);
770 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
775 * We are synchronized through cgroup_threadgroup_rwsem
776 * against PF_EXITING setting such that we can't race
777 * against cgroup_exit() changing the css_set to
778 * init_css_set and dropping the old one.
780 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
782 rcu_assign_pointer(task
->cgroups
, to_cset
);
783 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
789 * hash table for cgroup groups. This improves the performance to find
790 * an existing css_set. This hash doesn't (currently) take into
791 * account cgroups in empty hierarchies.
793 #define CSS_SET_HASH_BITS 7
794 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
796 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
798 unsigned long key
= 0UL;
799 struct cgroup_subsys
*ss
;
802 for_each_subsys(ss
, i
)
803 key
+= (unsigned long)css
[i
];
804 key
= (key
>> 16) ^ key
;
809 static void put_css_set_locked(struct css_set
*cset
)
811 struct cgrp_cset_link
*link
, *tmp_link
;
812 struct cgroup_subsys
*ss
;
815 lockdep_assert_held(&css_set_lock
);
817 if (!atomic_dec_and_test(&cset
->refcount
))
820 /* This css_set is dead. unlink it and release cgroup and css refs */
821 for_each_subsys(ss
, ssid
) {
822 list_del(&cset
->e_cset_node
[ssid
]);
823 css_put(cset
->subsys
[ssid
]);
825 hash_del(&cset
->hlist
);
828 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
829 list_del(&link
->cset_link
);
830 list_del(&link
->cgrp_link
);
831 if (cgroup_parent(link
->cgrp
))
832 cgroup_put(link
->cgrp
);
836 kfree_rcu(cset
, rcu_head
);
839 static void put_css_set(struct css_set
*cset
)
844 * Ensure that the refcount doesn't hit zero while any readers
845 * can see it. Similar to atomic_dec_and_lock(), but for an
848 if (atomic_add_unless(&cset
->refcount
, -1, 1))
851 spin_lock_irqsave(&css_set_lock
, flags
);
852 put_css_set_locked(cset
);
853 spin_unlock_irqrestore(&css_set_lock
, flags
);
857 * refcounted get/put for css_set objects
859 static inline void get_css_set(struct css_set
*cset
)
861 atomic_inc(&cset
->refcount
);
865 * compare_css_sets - helper function for find_existing_css_set().
866 * @cset: candidate css_set being tested
867 * @old_cset: existing css_set for a task
868 * @new_cgrp: cgroup that's being entered by the task
869 * @template: desired set of css pointers in css_set (pre-calculated)
871 * Returns true if "cset" matches "old_cset" except for the hierarchy
872 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
874 static bool compare_css_sets(struct css_set
*cset
,
875 struct css_set
*old_cset
,
876 struct cgroup
*new_cgrp
,
877 struct cgroup_subsys_state
*template[])
879 struct list_head
*l1
, *l2
;
882 * On the default hierarchy, there can be csets which are
883 * associated with the same set of cgroups but different csses.
884 * Let's first ensure that csses match.
886 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
890 * Compare cgroup pointers in order to distinguish between
891 * different cgroups in hierarchies. As different cgroups may
892 * share the same effective css, this comparison is always
895 l1
= &cset
->cgrp_links
;
896 l2
= &old_cset
->cgrp_links
;
898 struct cgrp_cset_link
*link1
, *link2
;
899 struct cgroup
*cgrp1
, *cgrp2
;
903 /* See if we reached the end - both lists are equal length. */
904 if (l1
== &cset
->cgrp_links
) {
905 BUG_ON(l2
!= &old_cset
->cgrp_links
);
908 BUG_ON(l2
== &old_cset
->cgrp_links
);
910 /* Locate the cgroups associated with these links. */
911 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
912 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
915 /* Hierarchies should be linked in the same order. */
916 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
919 * If this hierarchy is the hierarchy of the cgroup
920 * that's changing, then we need to check that this
921 * css_set points to the new cgroup; if it's any other
922 * hierarchy, then this css_set should point to the
923 * same cgroup as the old css_set.
925 if (cgrp1
->root
== new_cgrp
->root
) {
926 if (cgrp1
!= new_cgrp
)
937 * find_existing_css_set - init css array and find the matching css_set
938 * @old_cset: the css_set that we're using before the cgroup transition
939 * @cgrp: the cgroup that we're moving into
940 * @template: out param for the new set of csses, should be clear on entry
942 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
944 struct cgroup_subsys_state
*template[])
946 struct cgroup_root
*root
= cgrp
->root
;
947 struct cgroup_subsys
*ss
;
948 struct css_set
*cset
;
953 * Build the set of subsystem state objects that we want to see in the
954 * new css_set. while subsystems can change globally, the entries here
955 * won't change, so no need for locking.
957 for_each_subsys(ss
, i
) {
958 if (root
->subsys_mask
& (1UL << i
)) {
960 * @ss is in this hierarchy, so we want the
961 * effective css from @cgrp.
963 template[i
] = cgroup_e_css(cgrp
, ss
);
966 * @ss is not in this hierarchy, so we don't want
969 template[i
] = old_cset
->subsys
[i
];
973 key
= css_set_hash(template);
974 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
975 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
978 /* This css_set matches what we need */
982 /* No existing cgroup group matched */
986 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
988 struct cgrp_cset_link
*link
, *tmp_link
;
990 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
991 list_del(&link
->cset_link
);
997 * allocate_cgrp_cset_links - allocate cgrp_cset_links
998 * @count: the number of links to allocate
999 * @tmp_links: list_head the allocated links are put on
1001 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1002 * through ->cset_link. Returns 0 on success or -errno.
1004 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1006 struct cgrp_cset_link
*link
;
1009 INIT_LIST_HEAD(tmp_links
);
1011 for (i
= 0; i
< count
; i
++) {
1012 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1014 free_cgrp_cset_links(tmp_links
);
1017 list_add(&link
->cset_link
, tmp_links
);
1023 * link_css_set - a helper function to link a css_set to a cgroup
1024 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1025 * @cset: the css_set to be linked
1026 * @cgrp: the destination cgroup
1028 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1029 struct cgroup
*cgrp
)
1031 struct cgrp_cset_link
*link
;
1033 BUG_ON(list_empty(tmp_links
));
1035 if (cgroup_on_dfl(cgrp
))
1036 cset
->dfl_cgrp
= cgrp
;
1038 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1043 * Always add links to the tail of the lists so that the lists are
1044 * in choronological order.
1046 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1047 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1049 if (cgroup_parent(cgrp
))
1054 * find_css_set - return a new css_set with one cgroup updated
1055 * @old_cset: the baseline css_set
1056 * @cgrp: the cgroup to be updated
1058 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1059 * substituted into the appropriate hierarchy.
1061 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1062 struct cgroup
*cgrp
)
1064 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1065 struct css_set
*cset
;
1066 struct list_head tmp_links
;
1067 struct cgrp_cset_link
*link
;
1068 struct cgroup_subsys
*ss
;
1072 lockdep_assert_held(&cgroup_mutex
);
1074 /* First see if we already have a cgroup group that matches
1075 * the desired set */
1076 spin_lock_irq(&css_set_lock
);
1077 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1080 spin_unlock_irq(&css_set_lock
);
1085 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1089 /* Allocate all the cgrp_cset_link objects that we'll need */
1090 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1095 atomic_set(&cset
->refcount
, 1);
1096 INIT_LIST_HEAD(&cset
->cgrp_links
);
1097 INIT_LIST_HEAD(&cset
->tasks
);
1098 INIT_LIST_HEAD(&cset
->mg_tasks
);
1099 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1100 INIT_LIST_HEAD(&cset
->mg_node
);
1101 INIT_LIST_HEAD(&cset
->task_iters
);
1102 INIT_HLIST_NODE(&cset
->hlist
);
1104 /* Copy the set of subsystem state objects generated in
1105 * find_existing_css_set() */
1106 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1108 spin_lock_irq(&css_set_lock
);
1109 /* Add reference counts and links from the new css_set. */
1110 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1111 struct cgroup
*c
= link
->cgrp
;
1113 if (c
->root
== cgrp
->root
)
1115 link_css_set(&tmp_links
, cset
, c
);
1118 BUG_ON(!list_empty(&tmp_links
));
1122 /* Add @cset to the hash table */
1123 key
= css_set_hash(cset
->subsys
);
1124 hash_add(css_set_table
, &cset
->hlist
, key
);
1126 for_each_subsys(ss
, ssid
) {
1127 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1129 list_add_tail(&cset
->e_cset_node
[ssid
],
1130 &css
->cgroup
->e_csets
[ssid
]);
1134 spin_unlock_irq(&css_set_lock
);
1139 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1141 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1143 return root_cgrp
->root
;
1146 static int cgroup_init_root_id(struct cgroup_root
*root
)
1150 lockdep_assert_held(&cgroup_mutex
);
1152 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1156 root
->hierarchy_id
= id
;
1160 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1162 lockdep_assert_held(&cgroup_mutex
);
1164 if (root
->hierarchy_id
) {
1165 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1166 root
->hierarchy_id
= 0;
1170 static void cgroup_free_root(struct cgroup_root
*root
)
1173 /* hierarchy ID should already have been released */
1174 WARN_ON_ONCE(root
->hierarchy_id
);
1176 idr_destroy(&root
->cgroup_idr
);
1181 static void cgroup_destroy_root(struct cgroup_root
*root
)
1183 struct cgroup
*cgrp
= &root
->cgrp
;
1184 struct cgrp_cset_link
*link
, *tmp_link
;
1186 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1188 BUG_ON(atomic_read(&root
->nr_cgrps
));
1189 BUG_ON(!list_empty(&cgrp
->self
.children
));
1191 /* Rebind all subsystems back to the default hierarchy */
1192 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1195 * Release all the links from cset_links to this hierarchy's
1198 spin_lock_irq(&css_set_lock
);
1200 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1201 list_del(&link
->cset_link
);
1202 list_del(&link
->cgrp_link
);
1206 spin_unlock_irq(&css_set_lock
);
1208 if (!list_empty(&root
->root_list
)) {
1209 list_del(&root
->root_list
);
1210 cgroup_root_count
--;
1213 cgroup_exit_root_id(root
);
1215 mutex_unlock(&cgroup_mutex
);
1217 kernfs_destroy_root(root
->kf_root
);
1218 cgroup_free_root(root
);
1222 * look up cgroup associated with current task's cgroup namespace on the
1223 * specified hierarchy
1225 static struct cgroup
*
1226 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1228 struct cgroup
*res
= NULL
;
1229 struct css_set
*cset
;
1231 lockdep_assert_held(&css_set_lock
);
1235 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1236 if (cset
== &init_css_set
) {
1239 struct cgrp_cset_link
*link
;
1241 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1242 struct cgroup
*c
= link
->cgrp
;
1244 if (c
->root
== root
) {
1256 /* look up cgroup associated with given css_set on the specified hierarchy */
1257 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1258 struct cgroup_root
*root
)
1260 struct cgroup
*res
= NULL
;
1262 lockdep_assert_held(&cgroup_mutex
);
1263 lockdep_assert_held(&css_set_lock
);
1265 if (cset
== &init_css_set
) {
1268 struct cgrp_cset_link
*link
;
1270 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1271 struct cgroup
*c
= link
->cgrp
;
1273 if (c
->root
== root
) {
1285 * Return the cgroup for "task" from the given hierarchy. Must be
1286 * called with cgroup_mutex and css_set_lock held.
1288 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1289 struct cgroup_root
*root
)
1292 * No need to lock the task - since we hold cgroup_mutex the
1293 * task can't change groups, so the only thing that can happen
1294 * is that it exits and its css is set back to init_css_set.
1296 return cset_cgroup_from_root(task_css_set(task
), root
);
1300 * A task must hold cgroup_mutex to modify cgroups.
1302 * Any task can increment and decrement the count field without lock.
1303 * So in general, code holding cgroup_mutex can't rely on the count
1304 * field not changing. However, if the count goes to zero, then only
1305 * cgroup_attach_task() can increment it again. Because a count of zero
1306 * means that no tasks are currently attached, therefore there is no
1307 * way a task attached to that cgroup can fork (the other way to
1308 * increment the count). So code holding cgroup_mutex can safely
1309 * assume that if the count is zero, it will stay zero. Similarly, if
1310 * a task holds cgroup_mutex on a cgroup with zero count, it
1311 * knows that the cgroup won't be removed, as cgroup_rmdir()
1314 * A cgroup can only be deleted if both its 'count' of using tasks
1315 * is zero, and its list of 'children' cgroups is empty. Since all
1316 * tasks in the system use _some_ cgroup, and since there is always at
1317 * least one task in the system (init, pid == 1), therefore, root cgroup
1318 * always has either children cgroups and/or using tasks. So we don't
1319 * need a special hack to ensure that root cgroup cannot be deleted.
1321 * P.S. One more locking exception. RCU is used to guard the
1322 * update of a tasks cgroup pointer by cgroup_attach_task()
1325 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1326 static const struct file_operations proc_cgroupstats_operations
;
1328 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1331 struct cgroup_subsys
*ss
= cft
->ss
;
1333 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1334 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1335 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1336 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1339 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1344 * cgroup_file_mode - deduce file mode of a control file
1345 * @cft: the control file in question
1347 * S_IRUGO for read, S_IWUSR for write.
1349 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1353 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1356 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1357 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1367 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1368 * @subtree_control: the new subtree_control mask to consider
1369 * @this_ss_mask: available subsystems
1371 * On the default hierarchy, a subsystem may request other subsystems to be
1372 * enabled together through its ->depends_on mask. In such cases, more
1373 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1375 * This function calculates which subsystems need to be enabled if
1376 * @subtree_control is to be applied while restricted to @this_ss_mask.
1378 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1380 u16 cur_ss_mask
= subtree_control
;
1381 struct cgroup_subsys
*ss
;
1384 lockdep_assert_held(&cgroup_mutex
);
1386 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1389 u16 new_ss_mask
= cur_ss_mask
;
1391 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1392 new_ss_mask
|= ss
->depends_on
;
1393 } while_each_subsys_mask();
1396 * Mask out subsystems which aren't available. This can
1397 * happen only if some depended-upon subsystems were bound
1398 * to non-default hierarchies.
1400 new_ss_mask
&= this_ss_mask
;
1402 if (new_ss_mask
== cur_ss_mask
)
1404 cur_ss_mask
= new_ss_mask
;
1411 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1412 * @kn: the kernfs_node being serviced
1414 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1415 * the method finishes if locking succeeded. Note that once this function
1416 * returns the cgroup returned by cgroup_kn_lock_live() may become
1417 * inaccessible any time. If the caller intends to continue to access the
1418 * cgroup, it should pin it before invoking this function.
1420 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1422 struct cgroup
*cgrp
;
1424 if (kernfs_type(kn
) == KERNFS_DIR
)
1427 cgrp
= kn
->parent
->priv
;
1429 mutex_unlock(&cgroup_mutex
);
1431 kernfs_unbreak_active_protection(kn
);
1436 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1437 * @kn: the kernfs_node being serviced
1438 * @drain_offline: perform offline draining on the cgroup
1440 * This helper is to be used by a cgroup kernfs method currently servicing
1441 * @kn. It breaks the active protection, performs cgroup locking and
1442 * verifies that the associated cgroup is alive. Returns the cgroup if
1443 * alive; otherwise, %NULL. A successful return should be undone by a
1444 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1445 * cgroup is drained of offlining csses before return.
1447 * Any cgroup kernfs method implementation which requires locking the
1448 * associated cgroup should use this helper. It avoids nesting cgroup
1449 * locking under kernfs active protection and allows all kernfs operations
1450 * including self-removal.
1452 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
,
1455 struct cgroup
*cgrp
;
1457 if (kernfs_type(kn
) == KERNFS_DIR
)
1460 cgrp
= kn
->parent
->priv
;
1463 * We're gonna grab cgroup_mutex which nests outside kernfs
1464 * active_ref. cgroup liveliness check alone provides enough
1465 * protection against removal. Ensure @cgrp stays accessible and
1466 * break the active_ref protection.
1468 if (!cgroup_tryget(cgrp
))
1470 kernfs_break_active_protection(kn
);
1473 cgroup_lock_and_drain_offline(cgrp
);
1475 mutex_lock(&cgroup_mutex
);
1477 if (!cgroup_is_dead(cgrp
))
1480 cgroup_kn_unlock(kn
);
1484 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1486 char name
[CGROUP_FILE_NAME_MAX
];
1488 lockdep_assert_held(&cgroup_mutex
);
1490 if (cft
->file_offset
) {
1491 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1492 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1494 spin_lock_irq(&cgroup_file_kn_lock
);
1496 spin_unlock_irq(&cgroup_file_kn_lock
);
1499 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1503 * css_clear_dir - remove subsys files in a cgroup directory
1506 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1508 struct cgroup
*cgrp
= css
->cgroup
;
1509 struct cftype
*cfts
;
1511 if (!(css
->flags
& CSS_VISIBLE
))
1514 css
->flags
&= ~CSS_VISIBLE
;
1516 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1517 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1521 * css_populate_dir - create subsys files in a cgroup directory
1524 * On failure, no file is added.
1526 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1528 struct cgroup
*cgrp
= css
->cgroup
;
1529 struct cftype
*cfts
, *failed_cfts
;
1532 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1536 if (cgroup_on_dfl(cgrp
))
1537 cfts
= cgroup_dfl_base_files
;
1539 cfts
= cgroup_legacy_base_files
;
1541 return cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1544 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1545 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1552 css
->flags
|= CSS_VISIBLE
;
1556 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1557 if (cfts
== failed_cfts
)
1559 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1564 static int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1566 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1567 struct cgroup_subsys
*ss
;
1570 lockdep_assert_held(&cgroup_mutex
);
1572 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1574 * If @ss has non-root csses attached to it, can't move.
1575 * If @ss is an implicit controller, it is exempt from this
1576 * rule and can be stolen.
1578 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1579 !ss
->implicit_on_dfl
)
1582 /* can't move between two non-dummy roots either */
1583 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1585 } while_each_subsys_mask();
1587 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1588 struct cgroup_root
*src_root
= ss
->root
;
1589 struct cgroup
*scgrp
= &src_root
->cgrp
;
1590 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1591 struct css_set
*cset
;
1593 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1595 /* disable from the source */
1596 src_root
->subsys_mask
&= ~(1 << ssid
);
1597 WARN_ON(cgroup_apply_control(scgrp
));
1598 cgroup_finalize_control(scgrp
, 0);
1601 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1602 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1603 ss
->root
= dst_root
;
1604 css
->cgroup
= dcgrp
;
1606 spin_lock_irq(&css_set_lock
);
1607 hash_for_each(css_set_table
, i
, cset
, hlist
)
1608 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1609 &dcgrp
->e_csets
[ss
->id
]);
1610 spin_unlock_irq(&css_set_lock
);
1612 /* default hierarchy doesn't enable controllers by default */
1613 dst_root
->subsys_mask
|= 1 << ssid
;
1614 if (dst_root
== &cgrp_dfl_root
) {
1615 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1617 dcgrp
->subtree_control
|= 1 << ssid
;
1618 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1621 ret
= cgroup_apply_control(dcgrp
);
1623 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1628 } while_each_subsys_mask();
1630 kernfs_activate(dcgrp
->kn
);
1634 static int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1635 struct kernfs_root
*kf_root
)
1639 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1640 struct cgroup
*ns_cgroup
;
1642 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1646 spin_lock_irq(&css_set_lock
);
1647 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1648 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1649 spin_unlock_irq(&css_set_lock
);
1651 if (len
>= PATH_MAX
)
1654 seq_escape(sf
, buf
, " \t\n\\");
1661 static int cgroup_show_options(struct seq_file
*seq
,
1662 struct kernfs_root
*kf_root
)
1664 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1665 struct cgroup_subsys
*ss
;
1668 if (root
!= &cgrp_dfl_root
)
1669 for_each_subsys(ss
, ssid
)
1670 if (root
->subsys_mask
& (1 << ssid
))
1671 seq_show_option(seq
, ss
->legacy_name
, NULL
);
1672 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1673 seq_puts(seq
, ",noprefix");
1674 if (root
->flags
& CGRP_ROOT_XATTR
)
1675 seq_puts(seq
, ",xattr");
1677 spin_lock(&release_agent_path_lock
);
1678 if (strlen(root
->release_agent_path
))
1679 seq_show_option(seq
, "release_agent",
1680 root
->release_agent_path
);
1681 spin_unlock(&release_agent_path_lock
);
1683 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1684 seq_puts(seq
, ",clone_children");
1685 if (strlen(root
->name
))
1686 seq_show_option(seq
, "name", root
->name
);
1690 struct cgroup_sb_opts
{
1693 char *release_agent
;
1694 bool cpuset_clone_children
;
1696 /* User explicitly requested empty subsystem */
1700 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1702 char *token
, *o
= data
;
1703 bool all_ss
= false, one_ss
= false;
1705 struct cgroup_subsys
*ss
;
1709 #ifdef CONFIG_CPUSETS
1710 mask
= ~((u16
)1 << cpuset_cgrp_id
);
1713 memset(opts
, 0, sizeof(*opts
));
1715 while ((token
= strsep(&o
, ",")) != NULL
) {
1720 if (!strcmp(token
, "none")) {
1721 /* Explicitly have no subsystems */
1725 if (!strcmp(token
, "all")) {
1726 /* Mutually exclusive option 'all' + subsystem name */
1732 if (!strcmp(token
, "noprefix")) {
1733 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1736 if (!strcmp(token
, "clone_children")) {
1737 opts
->cpuset_clone_children
= true;
1740 if (!strcmp(token
, "xattr")) {
1741 opts
->flags
|= CGRP_ROOT_XATTR
;
1744 if (!strncmp(token
, "release_agent=", 14)) {
1745 /* Specifying two release agents is forbidden */
1746 if (opts
->release_agent
)
1748 opts
->release_agent
=
1749 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1750 if (!opts
->release_agent
)
1754 if (!strncmp(token
, "name=", 5)) {
1755 const char *name
= token
+ 5;
1756 /* Can't specify an empty name */
1759 /* Must match [\w.-]+ */
1760 for (i
= 0; i
< strlen(name
); i
++) {
1764 if ((c
== '.') || (c
== '-') || (c
== '_'))
1768 /* Specifying two names is forbidden */
1771 opts
->name
= kstrndup(name
,
1772 MAX_CGROUP_ROOT_NAMELEN
- 1,
1780 for_each_subsys(ss
, i
) {
1781 if (strcmp(token
, ss
->legacy_name
))
1783 if (!cgroup_ssid_enabled(i
))
1785 if (cgroup_ssid_no_v1(i
))
1788 /* Mutually exclusive option 'all' + subsystem name */
1791 opts
->subsys_mask
|= (1 << i
);
1796 if (i
== CGROUP_SUBSYS_COUNT
)
1801 * If the 'all' option was specified select all the subsystems,
1802 * otherwise if 'none', 'name=' and a subsystem name options were
1803 * not specified, let's default to 'all'
1805 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1806 for_each_subsys(ss
, i
)
1807 if (cgroup_ssid_enabled(i
) && !cgroup_ssid_no_v1(i
))
1808 opts
->subsys_mask
|= (1 << i
);
1811 * We either have to specify by name or by subsystems. (So all
1812 * empty hierarchies must have a name).
1814 if (!opts
->subsys_mask
&& !opts
->name
)
1818 * Option noprefix was introduced just for backward compatibility
1819 * with the old cpuset, so we allow noprefix only if mounting just
1820 * the cpuset subsystem.
1822 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1825 /* Can't specify "none" and some subsystems */
1826 if (opts
->subsys_mask
&& opts
->none
)
1832 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1835 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1836 struct cgroup_sb_opts opts
;
1837 u16 added_mask
, removed_mask
;
1839 if (root
== &cgrp_dfl_root
) {
1840 pr_err("remount is not allowed\n");
1844 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1846 /* See what subsystems are wanted */
1847 ret
= parse_cgroupfs_options(data
, &opts
);
1851 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1852 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1853 task_tgid_nr(current
), current
->comm
);
1855 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1856 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1858 /* Don't allow flags or name to change at remount */
1859 if ((opts
.flags
^ root
->flags
) ||
1860 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1861 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1862 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1867 /* remounting is not allowed for populated hierarchies */
1868 if (!list_empty(&root
->cgrp
.self
.children
)) {
1873 ret
= rebind_subsystems(root
, added_mask
);
1877 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, removed_mask
));
1879 if (opts
.release_agent
) {
1880 spin_lock(&release_agent_path_lock
);
1881 strcpy(root
->release_agent_path
, opts
.release_agent
);
1882 spin_unlock(&release_agent_path_lock
);
1885 kfree(opts
.release_agent
);
1887 mutex_unlock(&cgroup_mutex
);
1892 * To reduce the fork() overhead for systems that are not actually using
1893 * their cgroups capability, we don't maintain the lists running through
1894 * each css_set to its tasks until we see the list actually used - in other
1895 * words after the first mount.
1897 static bool use_task_css_set_links __read_mostly
;
1899 static void cgroup_enable_task_cg_lists(void)
1901 struct task_struct
*p
, *g
;
1903 spin_lock_irq(&css_set_lock
);
1905 if (use_task_css_set_links
)
1908 use_task_css_set_links
= true;
1911 * We need tasklist_lock because RCU is not safe against
1912 * while_each_thread(). Besides, a forking task that has passed
1913 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1914 * is not guaranteed to have its child immediately visible in the
1915 * tasklist if we walk through it with RCU.
1917 read_lock(&tasklist_lock
);
1918 do_each_thread(g
, p
) {
1919 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1920 task_css_set(p
) != &init_css_set
);
1923 * We should check if the process is exiting, otherwise
1924 * it will race with cgroup_exit() in that the list
1925 * entry won't be deleted though the process has exited.
1926 * Do it while holding siglock so that we don't end up
1927 * racing against cgroup_exit().
1929 * Interrupts were already disabled while acquiring
1930 * the css_set_lock, so we do not need to disable it
1931 * again when acquiring the sighand->siglock here.
1933 spin_lock(&p
->sighand
->siglock
);
1934 if (!(p
->flags
& PF_EXITING
)) {
1935 struct css_set
*cset
= task_css_set(p
);
1937 if (!css_set_populated(cset
))
1938 css_set_update_populated(cset
, true);
1939 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1942 spin_unlock(&p
->sighand
->siglock
);
1943 } while_each_thread(g
, p
);
1944 read_unlock(&tasklist_lock
);
1946 spin_unlock_irq(&css_set_lock
);
1949 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1951 struct cgroup_subsys
*ss
;
1954 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1955 INIT_LIST_HEAD(&cgrp
->self
.children
);
1956 INIT_LIST_HEAD(&cgrp
->cset_links
);
1957 INIT_LIST_HEAD(&cgrp
->pidlists
);
1958 mutex_init(&cgrp
->pidlist_mutex
);
1959 cgrp
->self
.cgroup
= cgrp
;
1960 cgrp
->self
.flags
|= CSS_ONLINE
;
1962 for_each_subsys(ss
, ssid
)
1963 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1965 init_waitqueue_head(&cgrp
->offline_waitq
);
1966 INIT_WORK(&cgrp
->release_agent_work
, cgroup_release_agent
);
1969 static void init_cgroup_root(struct cgroup_root
*root
,
1970 struct cgroup_sb_opts
*opts
)
1972 struct cgroup
*cgrp
= &root
->cgrp
;
1974 INIT_LIST_HEAD(&root
->root_list
);
1975 atomic_set(&root
->nr_cgrps
, 1);
1977 init_cgroup_housekeeping(cgrp
);
1978 idr_init(&root
->cgroup_idr
);
1980 root
->flags
= opts
->flags
;
1981 if (opts
->release_agent
)
1982 strcpy(root
->release_agent_path
, opts
->release_agent
);
1984 strcpy(root
->name
, opts
->name
);
1985 if (opts
->cpuset_clone_children
)
1986 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1989 static int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
)
1991 LIST_HEAD(tmp_links
);
1992 struct cgroup
*root_cgrp
= &root
->cgrp
;
1993 struct css_set
*cset
;
1996 lockdep_assert_held(&cgroup_mutex
);
1998 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
2001 root_cgrp
->id
= ret
;
2002 root_cgrp
->ancestor_ids
[0] = ret
;
2004 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
, 0,
2010 * We're accessing css_set_count without locking css_set_lock here,
2011 * but that's OK - it can only be increased by someone holding
2012 * cgroup_lock, and that's us. Later rebinding may disable
2013 * controllers on the default hierarchy and thus create new csets,
2014 * which can't be more than the existing ones. Allocate 2x.
2016 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
2020 ret
= cgroup_init_root_id(root
);
2024 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
2025 KERNFS_ROOT_CREATE_DEACTIVATED
,
2027 if (IS_ERR(root
->kf_root
)) {
2028 ret
= PTR_ERR(root
->kf_root
);
2031 root_cgrp
->kn
= root
->kf_root
->kn
;
2033 ret
= css_populate_dir(&root_cgrp
->self
);
2037 ret
= rebind_subsystems(root
, ss_mask
);
2042 * There must be no failure case after here, since rebinding takes
2043 * care of subsystems' refcounts, which are explicitly dropped in
2044 * the failure exit path.
2046 list_add(&root
->root_list
, &cgroup_roots
);
2047 cgroup_root_count
++;
2050 * Link the root cgroup in this hierarchy into all the css_set
2053 spin_lock_irq(&css_set_lock
);
2054 hash_for_each(css_set_table
, i
, cset
, hlist
) {
2055 link_css_set(&tmp_links
, cset
, root_cgrp
);
2056 if (css_set_populated(cset
))
2057 cgroup_update_populated(root_cgrp
, true);
2059 spin_unlock_irq(&css_set_lock
);
2061 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
2062 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
2064 kernfs_activate(root_cgrp
->kn
);
2069 kernfs_destroy_root(root
->kf_root
);
2070 root
->kf_root
= NULL
;
2072 cgroup_exit_root_id(root
);
2074 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
2076 free_cgrp_cset_links(&tmp_links
);
2080 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
2081 int flags
, const char *unused_dev_name
,
2084 bool is_v2
= fs_type
== &cgroup2_fs_type
;
2085 struct super_block
*pinned_sb
= NULL
;
2086 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
2087 struct cgroup_subsys
*ss
;
2088 struct cgroup_root
*root
;
2089 struct cgroup_sb_opts opts
;
2090 struct dentry
*dentry
;
2097 /* Check if the caller has permission to mount. */
2098 if (!ns_capable(ns
->user_ns
, CAP_SYS_ADMIN
)) {
2100 return ERR_PTR(-EPERM
);
2104 * The first time anyone tries to mount a cgroup, enable the list
2105 * linking each css_set to its tasks and fix up all existing tasks.
2107 if (!use_task_css_set_links
)
2108 cgroup_enable_task_cg_lists();
2112 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data
);
2114 return ERR_PTR(-EINVAL
);
2116 cgrp_dfl_visible
= true;
2117 root
= &cgrp_dfl_root
;
2118 cgroup_get(&root
->cgrp
);
2122 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
2124 /* First find the desired set of subsystems */
2125 ret
= parse_cgroupfs_options(data
, &opts
);
2130 * Destruction of cgroup root is asynchronous, so subsystems may
2131 * still be dying after the previous unmount. Let's drain the
2132 * dying subsystems. We just need to ensure that the ones
2133 * unmounted previously finish dying and don't care about new ones
2134 * starting. Testing ref liveliness is good enough.
2136 for_each_subsys(ss
, i
) {
2137 if (!(opts
.subsys_mask
& (1 << i
)) ||
2138 ss
->root
== &cgrp_dfl_root
)
2141 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
2142 mutex_unlock(&cgroup_mutex
);
2144 ret
= restart_syscall();
2147 cgroup_put(&ss
->root
->cgrp
);
2150 for_each_root(root
) {
2151 bool name_match
= false;
2153 if (root
== &cgrp_dfl_root
)
2157 * If we asked for a name then it must match. Also, if
2158 * name matches but sybsys_mask doesn't, we should fail.
2159 * Remember whether name matched.
2162 if (strcmp(opts
.name
, root
->name
))
2168 * If we asked for subsystems (or explicitly for no
2169 * subsystems) then they must match.
2171 if ((opts
.subsys_mask
|| opts
.none
) &&
2172 (opts
.subsys_mask
!= root
->subsys_mask
)) {
2179 if (root
->flags
^ opts
.flags
)
2180 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2183 * We want to reuse @root whose lifetime is governed by its
2184 * ->cgrp. Let's check whether @root is alive and keep it
2185 * that way. As cgroup_kill_sb() can happen anytime, we
2186 * want to block it by pinning the sb so that @root doesn't
2187 * get killed before mount is complete.
2189 * With the sb pinned, tryget_live can reliably indicate
2190 * whether @root can be reused. If it's being killed,
2191 * drain it. We can use wait_queue for the wait but this
2192 * path is super cold. Let's just sleep a bit and retry.
2194 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
2195 if (IS_ERR(pinned_sb
) ||
2196 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
2197 mutex_unlock(&cgroup_mutex
);
2198 if (!IS_ERR_OR_NULL(pinned_sb
))
2199 deactivate_super(pinned_sb
);
2201 ret
= restart_syscall();
2210 * No such thing, create a new one. name= matching without subsys
2211 * specification is allowed for already existing hierarchies but we
2212 * can't create new one without subsys specification.
2214 if (!opts
.subsys_mask
&& !opts
.none
) {
2220 * We know this subsystem has not yet been bound. Users in a non-init
2221 * user namespace may only mount hierarchies with no bound subsystems,
2222 * i.e. 'none,name=user1'
2224 if (!opts
.none
&& !capable(CAP_SYS_ADMIN
)) {
2229 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
2235 init_cgroup_root(root
, &opts
);
2237 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
2239 cgroup_free_root(root
);
2242 mutex_unlock(&cgroup_mutex
);
2244 kfree(opts
.release_agent
);
2249 return ERR_PTR(ret
);
2252 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
2253 is_v2
? CGROUP2_SUPER_MAGIC
: CGROUP_SUPER_MAGIC
,
2257 * In non-init cgroup namespace, instead of root cgroup's
2258 * dentry, we return the dentry corresponding to the
2259 * cgroupns->root_cgrp.
2261 if (!IS_ERR(dentry
) && ns
!= &init_cgroup_ns
) {
2262 struct dentry
*nsdentry
;
2263 struct cgroup
*cgrp
;
2265 mutex_lock(&cgroup_mutex
);
2266 spin_lock_irq(&css_set_lock
);
2268 cgrp
= cset_cgroup_from_root(ns
->root_cset
, root
);
2270 spin_unlock_irq(&css_set_lock
);
2271 mutex_unlock(&cgroup_mutex
);
2273 nsdentry
= kernfs_node_dentry(cgrp
->kn
, dentry
->d_sb
);
2278 if (IS_ERR(dentry
) || !new_sb
)
2279 cgroup_put(&root
->cgrp
);
2282 * If @pinned_sb, we're reusing an existing root and holding an
2283 * extra ref on its sb. Mount is complete. Put the extra ref.
2287 deactivate_super(pinned_sb
);
2294 static void cgroup_kill_sb(struct super_block
*sb
)
2296 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2297 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2300 * If @root doesn't have any mounts or children, start killing it.
2301 * This prevents new mounts by disabling percpu_ref_tryget_live().
2302 * cgroup_mount() may wait for @root's release.
2304 * And don't kill the default root.
2306 if (!list_empty(&root
->cgrp
.self
.children
) ||
2307 root
== &cgrp_dfl_root
)
2308 cgroup_put(&root
->cgrp
);
2310 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2315 static struct file_system_type cgroup_fs_type
= {
2317 .mount
= cgroup_mount
,
2318 .kill_sb
= cgroup_kill_sb
,
2319 .fs_flags
= FS_USERNS_MOUNT
,
2322 static struct file_system_type cgroup2_fs_type
= {
2324 .mount
= cgroup_mount
,
2325 .kill_sb
= cgroup_kill_sb
,
2326 .fs_flags
= FS_USERNS_MOUNT
,
2329 static char *cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2330 struct cgroup_namespace
*ns
)
2332 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2335 ret
= kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2336 if (ret
< 0 || ret
>= buflen
)
2341 char *cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2342 struct cgroup_namespace
*ns
)
2346 mutex_lock(&cgroup_mutex
);
2347 spin_lock_irq(&css_set_lock
);
2349 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2351 spin_unlock_irq(&css_set_lock
);
2352 mutex_unlock(&cgroup_mutex
);
2356 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2359 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2360 * @task: target task
2361 * @buf: the buffer to write the path into
2362 * @buflen: the length of the buffer
2364 * Determine @task's cgroup on the first (the one with the lowest non-zero
2365 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2366 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2367 * cgroup controller callbacks.
2369 * Return value is the same as kernfs_path().
2371 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2373 struct cgroup_root
*root
;
2374 struct cgroup
*cgrp
;
2375 int hierarchy_id
= 1;
2378 mutex_lock(&cgroup_mutex
);
2379 spin_lock_irq(&css_set_lock
);
2381 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2384 cgrp
= task_cgroup_from_root(task
, root
);
2385 path
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2387 /* if no hierarchy exists, everyone is in "/" */
2388 if (strlcpy(buf
, "/", buflen
) < buflen
)
2392 spin_unlock_irq(&css_set_lock
);
2393 mutex_unlock(&cgroup_mutex
);
2396 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2398 /* used to track tasks and other necessary states during migration */
2399 struct cgroup_taskset
{
2400 /* the src and dst cset list running through cset->mg_node */
2401 struct list_head src_csets
;
2402 struct list_head dst_csets
;
2404 /* the subsys currently being processed */
2408 * Fields for cgroup_taskset_*() iteration.
2410 * Before migration is committed, the target migration tasks are on
2411 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2412 * the csets on ->dst_csets. ->csets point to either ->src_csets
2413 * or ->dst_csets depending on whether migration is committed.
2415 * ->cur_csets and ->cur_task point to the current task position
2418 struct list_head
*csets
;
2419 struct css_set
*cur_cset
;
2420 struct task_struct
*cur_task
;
2423 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2424 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2425 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2426 .csets = &tset.src_csets, \
2430 * cgroup_taskset_add - try to add a migration target task to a taskset
2431 * @task: target task
2432 * @tset: target taskset
2434 * Add @task, which is a migration target, to @tset. This function becomes
2435 * noop if @task doesn't need to be migrated. @task's css_set should have
2436 * been added as a migration source and @task->cg_list will be moved from
2437 * the css_set's tasks list to mg_tasks one.
2439 static void cgroup_taskset_add(struct task_struct
*task
,
2440 struct cgroup_taskset
*tset
)
2442 struct css_set
*cset
;
2444 lockdep_assert_held(&css_set_lock
);
2446 /* @task either already exited or can't exit until the end */
2447 if (task
->flags
& PF_EXITING
)
2450 /* leave @task alone if post_fork() hasn't linked it yet */
2451 if (list_empty(&task
->cg_list
))
2454 cset
= task_css_set(task
);
2455 if (!cset
->mg_src_cgrp
)
2458 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2459 if (list_empty(&cset
->mg_node
))
2460 list_add_tail(&cset
->mg_node
, &tset
->src_csets
);
2461 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2462 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2467 * cgroup_taskset_first - reset taskset and return the first task
2468 * @tset: taskset of interest
2469 * @dst_cssp: output variable for the destination css
2471 * @tset iteration is initialized and the first task is returned.
2473 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2474 struct cgroup_subsys_state
**dst_cssp
)
2476 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2477 tset
->cur_task
= NULL
;
2479 return cgroup_taskset_next(tset
, dst_cssp
);
2483 * cgroup_taskset_next - iterate to the next task in taskset
2484 * @tset: taskset of interest
2485 * @dst_cssp: output variable for the destination css
2487 * Return the next task in @tset. Iteration must have been initialized
2488 * with cgroup_taskset_first().
2490 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2491 struct cgroup_subsys_state
**dst_cssp
)
2493 struct css_set
*cset
= tset
->cur_cset
;
2494 struct task_struct
*task
= tset
->cur_task
;
2496 while (&cset
->mg_node
!= tset
->csets
) {
2498 task
= list_first_entry(&cset
->mg_tasks
,
2499 struct task_struct
, cg_list
);
2501 task
= list_next_entry(task
, cg_list
);
2503 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2504 tset
->cur_cset
= cset
;
2505 tset
->cur_task
= task
;
2508 * This function may be called both before and
2509 * after cgroup_taskset_migrate(). The two cases
2510 * can be distinguished by looking at whether @cset
2511 * has its ->mg_dst_cset set.
2513 if (cset
->mg_dst_cset
)
2514 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2516 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2521 cset
= list_next_entry(cset
, mg_node
);
2529 * cgroup_taskset_migrate - migrate a taskset
2530 * @tset: taget taskset
2531 * @root: cgroup root the migration is taking place on
2533 * Migrate tasks in @tset as setup by migration preparation functions.
2534 * This function fails iff one of the ->can_attach callbacks fails and
2535 * guarantees that either all or none of the tasks in @tset are migrated.
2536 * @tset is consumed regardless of success.
2538 static int cgroup_taskset_migrate(struct cgroup_taskset
*tset
,
2539 struct cgroup_root
*root
)
2541 struct cgroup_subsys
*ss
;
2542 struct task_struct
*task
, *tmp_task
;
2543 struct css_set
*cset
, *tmp_cset
;
2544 int ssid
, failed_ssid
, ret
;
2546 /* methods shouldn't be called if no task is actually migrating */
2547 if (list_empty(&tset
->src_csets
))
2550 /* check that we can legitimately attach to the cgroup */
2551 do_each_subsys_mask(ss
, ssid
, root
->subsys_mask
) {
2552 if (ss
->can_attach
) {
2554 ret
= ss
->can_attach(tset
);
2557 goto out_cancel_attach
;
2560 } while_each_subsys_mask();
2563 * Now that we're guaranteed success, proceed to move all tasks to
2564 * the new cgroup. There are no failure cases after here, so this
2565 * is the commit point.
2567 spin_lock_irq(&css_set_lock
);
2568 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2569 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2570 struct css_set
*from_cset
= task_css_set(task
);
2571 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2573 get_css_set(to_cset
);
2574 css_set_move_task(task
, from_cset
, to_cset
, true);
2575 put_css_set_locked(from_cset
);
2578 spin_unlock_irq(&css_set_lock
);
2581 * Migration is committed, all target tasks are now on dst_csets.
2582 * Nothing is sensitive to fork() after this point. Notify
2583 * controllers that migration is complete.
2585 tset
->csets
= &tset
->dst_csets
;
2587 do_each_subsys_mask(ss
, ssid
, root
->subsys_mask
) {
2592 } while_each_subsys_mask();
2595 goto out_release_tset
;
2598 do_each_subsys_mask(ss
, ssid
, root
->subsys_mask
) {
2599 if (ssid
== failed_ssid
)
2601 if (ss
->cancel_attach
) {
2603 ss
->cancel_attach(tset
);
2605 } while_each_subsys_mask();
2607 spin_lock_irq(&css_set_lock
);
2608 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2609 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2610 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2611 list_del_init(&cset
->mg_node
);
2613 spin_unlock_irq(&css_set_lock
);
2618 * cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2619 * @dst_cgrp: destination cgroup to test
2621 * On the default hierarchy, except for the root, subtree_control must be
2622 * zero for migration destination cgroups with tasks so that child cgroups
2623 * don't compete against tasks.
2625 static bool cgroup_may_migrate_to(struct cgroup
*dst_cgrp
)
2627 return !cgroup_on_dfl(dst_cgrp
) || !cgroup_parent(dst_cgrp
) ||
2628 !dst_cgrp
->subtree_control
;
2632 * cgroup_migrate_finish - cleanup after attach
2633 * @preloaded_csets: list of preloaded css_sets
2635 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2636 * those functions for details.
2638 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2640 struct css_set
*cset
, *tmp_cset
;
2642 lockdep_assert_held(&cgroup_mutex
);
2644 spin_lock_irq(&css_set_lock
);
2645 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2646 cset
->mg_src_cgrp
= NULL
;
2647 cset
->mg_dst_cgrp
= NULL
;
2648 cset
->mg_dst_cset
= NULL
;
2649 list_del_init(&cset
->mg_preload_node
);
2650 put_css_set_locked(cset
);
2652 spin_unlock_irq(&css_set_lock
);
2656 * cgroup_migrate_add_src - add a migration source css_set
2657 * @src_cset: the source css_set to add
2658 * @dst_cgrp: the destination cgroup
2659 * @preloaded_csets: list of preloaded css_sets
2661 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2662 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2663 * up by cgroup_migrate_finish().
2665 * This function may be called without holding cgroup_threadgroup_rwsem
2666 * even if the target is a process. Threads may be created and destroyed
2667 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2668 * into play and the preloaded css_sets are guaranteed to cover all
2671 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2672 struct cgroup
*dst_cgrp
,
2673 struct list_head
*preloaded_csets
)
2675 struct cgroup
*src_cgrp
;
2677 lockdep_assert_held(&cgroup_mutex
);
2678 lockdep_assert_held(&css_set_lock
);
2681 * If ->dead, @src_set is associated with one or more dead cgroups
2682 * and doesn't contain any migratable tasks. Ignore it early so
2683 * that the rest of migration path doesn't get confused by it.
2688 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2690 if (!list_empty(&src_cset
->mg_preload_node
))
2693 WARN_ON(src_cset
->mg_src_cgrp
);
2694 WARN_ON(src_cset
->mg_dst_cgrp
);
2695 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2696 WARN_ON(!list_empty(&src_cset
->mg_node
));
2698 src_cset
->mg_src_cgrp
= src_cgrp
;
2699 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2700 get_css_set(src_cset
);
2701 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2705 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2706 * @preloaded_csets: list of preloaded source css_sets
2708 * Tasks are about to be moved and all the source css_sets have been
2709 * preloaded to @preloaded_csets. This function looks up and pins all
2710 * destination css_sets, links each to its source, and append them to
2713 * This function must be called after cgroup_migrate_add_src() has been
2714 * called on each migration source css_set. After migration is performed
2715 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2718 static int cgroup_migrate_prepare_dst(struct list_head
*preloaded_csets
)
2721 struct css_set
*src_cset
, *tmp_cset
;
2723 lockdep_assert_held(&cgroup_mutex
);
2725 /* look up the dst cset for each src cset and link it to src */
2726 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2727 struct css_set
*dst_cset
;
2729 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2733 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2736 * If src cset equals dst, it's noop. Drop the src.
2737 * cgroup_migrate() will skip the cset too. Note that we
2738 * can't handle src == dst as some nodes are used by both.
2740 if (src_cset
== dst_cset
) {
2741 src_cset
->mg_src_cgrp
= NULL
;
2742 src_cset
->mg_dst_cgrp
= NULL
;
2743 list_del_init(&src_cset
->mg_preload_node
);
2744 put_css_set(src_cset
);
2745 put_css_set(dst_cset
);
2749 src_cset
->mg_dst_cset
= dst_cset
;
2751 if (list_empty(&dst_cset
->mg_preload_node
))
2752 list_add(&dst_cset
->mg_preload_node
, &csets
);
2754 put_css_set(dst_cset
);
2757 list_splice_tail(&csets
, preloaded_csets
);
2760 cgroup_migrate_finish(&csets
);
2765 * cgroup_migrate - migrate a process or task to a cgroup
2766 * @leader: the leader of the process or the task to migrate
2767 * @threadgroup: whether @leader points to the whole process or a single task
2768 * @root: cgroup root migration is taking place on
2770 * Migrate a process or task denoted by @leader. If migrating a process,
2771 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2772 * responsible for invoking cgroup_migrate_add_src() and
2773 * cgroup_migrate_prepare_dst() on the targets before invoking this
2774 * function and following up with cgroup_migrate_finish().
2776 * As long as a controller's ->can_attach() doesn't fail, this function is
2777 * guaranteed to succeed. This means that, excluding ->can_attach()
2778 * failure, when migrating multiple targets, the success or failure can be
2779 * decided for all targets by invoking group_migrate_prepare_dst() before
2780 * actually starting migrating.
2782 static int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2783 struct cgroup_root
*root
)
2785 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2786 struct task_struct
*task
;
2789 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2790 * already PF_EXITING could be freed from underneath us unless we
2791 * take an rcu_read_lock.
2793 spin_lock_irq(&css_set_lock
);
2797 cgroup_taskset_add(task
, &tset
);
2800 } while_each_thread(leader
, task
);
2802 spin_unlock_irq(&css_set_lock
);
2804 return cgroup_taskset_migrate(&tset
, root
);
2808 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2809 * @dst_cgrp: the cgroup to attach to
2810 * @leader: the task or the leader of the threadgroup to be attached
2811 * @threadgroup: attach the whole threadgroup?
2813 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2815 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2816 struct task_struct
*leader
, bool threadgroup
)
2818 LIST_HEAD(preloaded_csets
);
2819 struct task_struct
*task
;
2822 if (!cgroup_may_migrate_to(dst_cgrp
))
2825 /* look up all src csets */
2826 spin_lock_irq(&css_set_lock
);
2830 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2834 } while_each_thread(leader
, task
);
2836 spin_unlock_irq(&css_set_lock
);
2838 /* prepare dst csets and commit */
2839 ret
= cgroup_migrate_prepare_dst(&preloaded_csets
);
2841 ret
= cgroup_migrate(leader
, threadgroup
, dst_cgrp
->root
);
2843 cgroup_migrate_finish(&preloaded_csets
);
2847 static int cgroup_procs_write_permission(struct task_struct
*task
,
2848 struct cgroup
*dst_cgrp
,
2849 struct kernfs_open_file
*of
)
2851 const struct cred
*cred
= current_cred();
2852 const struct cred
*tcred
= get_task_cred(task
);
2856 * even if we're attaching all tasks in the thread group, we only
2857 * need to check permissions on one of them.
2859 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2860 !uid_eq(cred
->euid
, tcred
->uid
) &&
2861 !uid_eq(cred
->euid
, tcred
->suid
))
2864 if (!ret
&& cgroup_on_dfl(dst_cgrp
)) {
2865 struct super_block
*sb
= of
->file
->f_path
.dentry
->d_sb
;
2866 struct cgroup
*cgrp
;
2867 struct inode
*inode
;
2869 spin_lock_irq(&css_set_lock
);
2870 cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
2871 spin_unlock_irq(&css_set_lock
);
2873 while (!cgroup_is_descendant(dst_cgrp
, cgrp
))
2874 cgrp
= cgroup_parent(cgrp
);
2877 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
2879 ret
= inode_permission(inode
, MAY_WRITE
);
2889 * Find the task_struct of the task to attach by vpid and pass it along to the
2890 * function to attach either it or all tasks in its threadgroup. Will lock
2891 * cgroup_mutex and threadgroup.
2893 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2894 size_t nbytes
, loff_t off
, bool threadgroup
)
2896 struct task_struct
*tsk
;
2897 struct cgroup_subsys
*ss
;
2898 struct cgroup
*cgrp
;
2902 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2905 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
2909 percpu_down_write(&cgroup_threadgroup_rwsem
);
2912 tsk
= find_task_by_vpid(pid
);
2915 goto out_unlock_rcu
;
2922 tsk
= tsk
->group_leader
;
2925 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2926 * trapped in a cpuset, or RT worker may be born in a cgroup
2927 * with no rt_runtime allocated. Just say no.
2929 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2931 goto out_unlock_rcu
;
2934 get_task_struct(tsk
);
2937 ret
= cgroup_procs_write_permission(tsk
, cgrp
, of
);
2939 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2941 put_task_struct(tsk
);
2942 goto out_unlock_threadgroup
;
2946 out_unlock_threadgroup
:
2947 percpu_up_write(&cgroup_threadgroup_rwsem
);
2948 for_each_subsys(ss
, ssid
)
2949 if (ss
->post_attach
)
2951 cgroup_kn_unlock(of
->kn
);
2952 return ret
?: nbytes
;
2956 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2957 * @from: attach to all cgroups of a given task
2958 * @tsk: the task to be attached
2960 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2962 struct cgroup_root
*root
;
2965 mutex_lock(&cgroup_mutex
);
2966 for_each_root(root
) {
2967 struct cgroup
*from_cgrp
;
2969 if (root
== &cgrp_dfl_root
)
2972 spin_lock_irq(&css_set_lock
);
2973 from_cgrp
= task_cgroup_from_root(from
, root
);
2974 spin_unlock_irq(&css_set_lock
);
2976 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2980 mutex_unlock(&cgroup_mutex
);
2984 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2986 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2987 char *buf
, size_t nbytes
, loff_t off
)
2989 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2992 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2993 char *buf
, size_t nbytes
, loff_t off
)
2995 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2998 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2999 char *buf
, size_t nbytes
, loff_t off
)
3001 struct cgroup
*cgrp
;
3003 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
3005 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3008 spin_lock(&release_agent_path_lock
);
3009 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
3010 sizeof(cgrp
->root
->release_agent_path
));
3011 spin_unlock(&release_agent_path_lock
);
3012 cgroup_kn_unlock(of
->kn
);
3016 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
3018 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3020 spin_lock(&release_agent_path_lock
);
3021 seq_puts(seq
, cgrp
->root
->release_agent_path
);
3022 spin_unlock(&release_agent_path_lock
);
3023 seq_putc(seq
, '\n');
3027 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
3029 seq_puts(seq
, "0\n");
3033 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
3035 struct cgroup_subsys
*ss
;
3036 bool printed
= false;
3039 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
3042 seq_printf(seq
, "%s", ss
->name
);
3044 } while_each_subsys_mask();
3046 seq_putc(seq
, '\n');
3049 /* show controllers which are enabled from the parent */
3050 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
3052 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3054 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
3058 /* show controllers which are enabled for a given cgroup's children */
3059 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
3061 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3063 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
3068 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3069 * @cgrp: root of the subtree to update csses for
3071 * @cgrp's control masks have changed and its subtree's css associations
3072 * need to be updated accordingly. This function looks up all css_sets
3073 * which are attached to the subtree, creates the matching updated css_sets
3074 * and migrates the tasks to the new ones.
3076 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
3078 LIST_HEAD(preloaded_csets
);
3079 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
3080 struct cgroup_subsys_state
*d_css
;
3081 struct cgroup
*dsct
;
3082 struct css_set
*src_cset
;
3085 lockdep_assert_held(&cgroup_mutex
);
3087 percpu_down_write(&cgroup_threadgroup_rwsem
);
3089 /* look up all csses currently attached to @cgrp's subtree */
3090 spin_lock_irq(&css_set_lock
);
3091 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3092 struct cgrp_cset_link
*link
;
3094 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
3095 cgroup_migrate_add_src(link
->cset
, dsct
,
3098 spin_unlock_irq(&css_set_lock
);
3100 /* NULL dst indicates self on default hierarchy */
3101 ret
= cgroup_migrate_prepare_dst(&preloaded_csets
);
3105 spin_lock_irq(&css_set_lock
);
3106 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
3107 struct task_struct
*task
, *ntask
;
3109 /* src_csets precede dst_csets, break on the first dst_cset */
3110 if (!src_cset
->mg_src_cgrp
)
3113 /* all tasks in src_csets need to be migrated */
3114 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
3115 cgroup_taskset_add(task
, &tset
);
3117 spin_unlock_irq(&css_set_lock
);
3119 ret
= cgroup_taskset_migrate(&tset
, cgrp
->root
);
3121 cgroup_migrate_finish(&preloaded_csets
);
3122 percpu_up_write(&cgroup_threadgroup_rwsem
);
3127 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3128 * @cgrp: root of the target subtree
3130 * Because css offlining is asynchronous, userland may try to re-enable a
3131 * controller while the previous css is still around. This function grabs
3132 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3134 static void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
3135 __acquires(&cgroup_mutex
)
3137 struct cgroup
*dsct
;
3138 struct cgroup_subsys_state
*d_css
;
3139 struct cgroup_subsys
*ss
;
3143 mutex_lock(&cgroup_mutex
);
3145 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3146 for_each_subsys(ss
, ssid
) {
3147 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3150 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
3154 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
3155 TASK_UNINTERRUPTIBLE
);
3157 mutex_unlock(&cgroup_mutex
);
3159 finish_wait(&dsct
->offline_waitq
, &wait
);
3168 * cgroup_save_control - save control masks of a subtree
3169 * @cgrp: root of the target subtree
3171 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
3172 * prefixed fields for @cgrp's subtree including @cgrp itself.
3174 static void cgroup_save_control(struct cgroup
*cgrp
)
3176 struct cgroup
*dsct
;
3177 struct cgroup_subsys_state
*d_css
;
3179 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3180 dsct
->old_subtree_control
= dsct
->subtree_control
;
3181 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
3186 * cgroup_propagate_control - refresh control masks of a subtree
3187 * @cgrp: root of the target subtree
3189 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3190 * ->subtree_control and propagate controller availability through the
3191 * subtree so that descendants don't have unavailable controllers enabled.
3193 static void cgroup_propagate_control(struct cgroup
*cgrp
)
3195 struct cgroup
*dsct
;
3196 struct cgroup_subsys_state
*d_css
;
3198 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3199 dsct
->subtree_control
&= cgroup_control(dsct
);
3200 dsct
->subtree_ss_mask
=
3201 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
3202 cgroup_ss_mask(dsct
));
3207 * cgroup_restore_control - restore control masks of a subtree
3208 * @cgrp: root of the target subtree
3210 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
3211 * prefixed fields for @cgrp's subtree including @cgrp itself.
3213 static void cgroup_restore_control(struct cgroup
*cgrp
)
3215 struct cgroup
*dsct
;
3216 struct cgroup_subsys_state
*d_css
;
3218 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3219 dsct
->subtree_control
= dsct
->old_subtree_control
;
3220 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
3224 static bool css_visible(struct cgroup_subsys_state
*css
)
3226 struct cgroup_subsys
*ss
= css
->ss
;
3227 struct cgroup
*cgrp
= css
->cgroup
;
3229 if (cgroup_control(cgrp
) & (1 << ss
->id
))
3231 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
3233 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
3237 * cgroup_apply_control_enable - enable or show csses according to control
3238 * @cgrp: root of the target subtree
3240 * Walk @cgrp's subtree and create new csses or make the existing ones
3241 * visible. A css is created invisible if it's being implicitly enabled
3242 * through dependency. An invisible css is made visible when the userland
3243 * explicitly enables it.
3245 * Returns 0 on success, -errno on failure. On failure, csses which have
3246 * been processed already aren't cleaned up. The caller is responsible for
3247 * cleaning up with cgroup_apply_control_disble().
3249 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
3251 struct cgroup
*dsct
;
3252 struct cgroup_subsys_state
*d_css
;
3253 struct cgroup_subsys
*ss
;
3256 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
3257 for_each_subsys(ss
, ssid
) {
3258 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3260 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
3262 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
3266 css
= css_create(dsct
, ss
);
3268 return PTR_ERR(css
);
3271 if (css_visible(css
)) {
3272 ret
= css_populate_dir(css
);
3283 * cgroup_apply_control_disable - kill or hide csses according to control
3284 * @cgrp: root of the target subtree
3286 * Walk @cgrp's subtree and kill and hide csses so that they match
3287 * cgroup_ss_mask() and cgroup_visible_mask().
3289 * A css is hidden when the userland requests it to be disabled while other
3290 * subsystems are still depending on it. The css must not actively control
3291 * resources and be in the vanilla state if it's made visible again later.
3292 * Controllers which may be depended upon should provide ->css_reset() for
3295 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
3297 struct cgroup
*dsct
;
3298 struct cgroup_subsys_state
*d_css
;
3299 struct cgroup_subsys
*ss
;
3302 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
3303 for_each_subsys(ss
, ssid
) {
3304 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
3306 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
3312 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
3314 } else if (!css_visible(css
)) {
3324 * cgroup_apply_control - apply control mask updates to the subtree
3325 * @cgrp: root of the target subtree
3327 * subsystems can be enabled and disabled in a subtree using the following
3330 * 1. Call cgroup_save_control() to stash the current state.
3331 * 2. Update ->subtree_control masks in the subtree as desired.
3332 * 3. Call cgroup_apply_control() to apply the changes.
3333 * 4. Optionally perform other related operations.
3334 * 5. Call cgroup_finalize_control() to finish up.
3336 * This function implements step 3 and propagates the mask changes
3337 * throughout @cgrp's subtree, updates csses accordingly and perform
3338 * process migrations.
3340 static int cgroup_apply_control(struct cgroup
*cgrp
)
3344 cgroup_propagate_control(cgrp
);
3346 ret
= cgroup_apply_control_enable(cgrp
);
3351 * At this point, cgroup_e_css() results reflect the new csses
3352 * making the following cgroup_update_dfl_csses() properly update
3353 * css associations of all tasks in the subtree.
3355 ret
= cgroup_update_dfl_csses(cgrp
);
3363 * cgroup_finalize_control - finalize control mask update
3364 * @cgrp: root of the target subtree
3365 * @ret: the result of the update
3367 * Finalize control mask update. See cgroup_apply_control() for more info.
3369 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
3372 cgroup_restore_control(cgrp
);
3373 cgroup_propagate_control(cgrp
);
3376 cgroup_apply_control_disable(cgrp
);
3379 /* change the enabled child controllers for a cgroup in the default hierarchy */
3380 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3381 char *buf
, size_t nbytes
,
3384 u16 enable
= 0, disable
= 0;
3385 struct cgroup
*cgrp
, *child
;
3386 struct cgroup_subsys
*ss
;
3391 * Parse input - space separated list of subsystem names prefixed
3392 * with either + or -.
3394 buf
= strstrip(buf
);
3395 while ((tok
= strsep(&buf
, " "))) {
3398 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3399 if (!cgroup_ssid_enabled(ssid
) ||
3400 strcmp(tok
+ 1, ss
->name
))
3404 enable
|= 1 << ssid
;
3405 disable
&= ~(1 << ssid
);
3406 } else if (*tok
== '-') {
3407 disable
|= 1 << ssid
;
3408 enable
&= ~(1 << ssid
);
3413 } while_each_subsys_mask();
3414 if (ssid
== CGROUP_SUBSYS_COUNT
)
3418 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3422 for_each_subsys(ss
, ssid
) {
3423 if (enable
& (1 << ssid
)) {
3424 if (cgrp
->subtree_control
& (1 << ssid
)) {
3425 enable
&= ~(1 << ssid
);
3429 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3433 } else if (disable
& (1 << ssid
)) {
3434 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3435 disable
&= ~(1 << ssid
);
3439 /* a child has it enabled? */
3440 cgroup_for_each_live_child(child
, cgrp
) {
3441 if (child
->subtree_control
& (1 << ssid
)) {
3449 if (!enable
&& !disable
) {
3455 * Except for the root, subtree_control must be zero for a cgroup
3456 * with tasks so that child cgroups don't compete against tasks.
3458 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
3463 /* save and update control masks and prepare csses */
3464 cgroup_save_control(cgrp
);
3466 cgrp
->subtree_control
|= enable
;
3467 cgrp
->subtree_control
&= ~disable
;
3469 ret
= cgroup_apply_control(cgrp
);
3471 cgroup_finalize_control(cgrp
, ret
);
3473 kernfs_activate(cgrp
->kn
);
3476 cgroup_kn_unlock(of
->kn
);
3477 return ret
?: nbytes
;
3480 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3482 seq_printf(seq
, "populated %d\n",
3483 cgroup_is_populated(seq_css(seq
)->cgroup
));
3487 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3488 size_t nbytes
, loff_t off
)
3490 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3491 struct cftype
*cft
= of
->kn
->priv
;
3492 struct cgroup_subsys_state
*css
;
3496 return cft
->write(of
, buf
, nbytes
, off
);
3499 * kernfs guarantees that a file isn't deleted with operations in
3500 * flight, which means that the matching css is and stays alive and
3501 * doesn't need to be pinned. The RCU locking is not necessary
3502 * either. It's just for the convenience of using cgroup_css().
3505 css
= cgroup_css(cgrp
, cft
->ss
);
3508 if (cft
->write_u64
) {
3509 unsigned long long v
;
3510 ret
= kstrtoull(buf
, 0, &v
);
3512 ret
= cft
->write_u64(css
, cft
, v
);
3513 } else if (cft
->write_s64
) {
3515 ret
= kstrtoll(buf
, 0, &v
);
3517 ret
= cft
->write_s64(css
, cft
, v
);
3522 return ret
?: nbytes
;
3525 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3527 return seq_cft(seq
)->seq_start(seq
, ppos
);
3530 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3532 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3535 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3537 seq_cft(seq
)->seq_stop(seq
, v
);
3540 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3542 struct cftype
*cft
= seq_cft(m
);
3543 struct cgroup_subsys_state
*css
= seq_css(m
);
3546 return cft
->seq_show(m
, arg
);
3549 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3550 else if (cft
->read_s64
)
3551 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3557 static struct kernfs_ops cgroup_kf_single_ops
= {
3558 .atomic_write_len
= PAGE_SIZE
,
3559 .write
= cgroup_file_write
,
3560 .seq_show
= cgroup_seqfile_show
,
3563 static struct kernfs_ops cgroup_kf_ops
= {
3564 .atomic_write_len
= PAGE_SIZE
,
3565 .write
= cgroup_file_write
,
3566 .seq_start
= cgroup_seqfile_start
,
3567 .seq_next
= cgroup_seqfile_next
,
3568 .seq_stop
= cgroup_seqfile_stop
,
3569 .seq_show
= cgroup_seqfile_show
,
3573 * cgroup_rename - Only allow simple rename of directories in place.
3575 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
3576 const char *new_name_str
)
3578 struct cgroup
*cgrp
= kn
->priv
;
3581 if (kernfs_type(kn
) != KERNFS_DIR
)
3583 if (kn
->parent
!= new_parent
)
3587 * This isn't a proper migration and its usefulness is very
3588 * limited. Disallow on the default hierarchy.
3590 if (cgroup_on_dfl(cgrp
))
3594 * We're gonna grab cgroup_mutex which nests outside kernfs
3595 * active_ref. kernfs_rename() doesn't require active_ref
3596 * protection. Break them before grabbing cgroup_mutex.
3598 kernfs_break_active_protection(new_parent
);
3599 kernfs_break_active_protection(kn
);
3601 mutex_lock(&cgroup_mutex
);
3603 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3605 mutex_unlock(&cgroup_mutex
);
3607 kernfs_unbreak_active_protection(kn
);
3608 kernfs_unbreak_active_protection(new_parent
);
3612 /* set uid and gid of cgroup dirs and files to that of the creator */
3613 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3615 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3616 .ia_uid
= current_fsuid(),
3617 .ia_gid
= current_fsgid(), };
3619 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3620 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3623 return kernfs_setattr(kn
, &iattr
);
3626 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3629 char name
[CGROUP_FILE_NAME_MAX
];
3630 struct kernfs_node
*kn
;
3631 struct lock_class_key
*key
= NULL
;
3634 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3635 key
= &cft
->lockdep_key
;
3637 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3638 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3643 ret
= cgroup_kn_set_ugid(kn
);
3649 if (cft
->file_offset
) {
3650 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3652 spin_lock_irq(&cgroup_file_kn_lock
);
3654 spin_unlock_irq(&cgroup_file_kn_lock
);
3661 * cgroup_addrm_files - add or remove files to a cgroup directory
3662 * @css: the target css
3663 * @cgrp: the target cgroup (usually css->cgroup)
3664 * @cfts: array of cftypes to be added
3665 * @is_add: whether to add or remove
3667 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3668 * For removals, this function never fails.
3670 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3671 struct cgroup
*cgrp
, struct cftype cfts
[],
3674 struct cftype
*cft
, *cft_end
= NULL
;
3677 lockdep_assert_held(&cgroup_mutex
);
3680 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3681 /* does cft->flags tell us to skip this file on @cgrp? */
3682 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3684 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3686 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3688 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3692 ret
= cgroup_add_file(css
, cgrp
, cft
);
3694 pr_warn("%s: failed to add %s, err=%d\n",
3695 __func__
, cft
->name
, ret
);
3701 cgroup_rm_file(cgrp
, cft
);
3707 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3710 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3711 struct cgroup
*root
= &ss
->root
->cgrp
;
3712 struct cgroup_subsys_state
*css
;
3715 lockdep_assert_held(&cgroup_mutex
);
3717 /* add/rm files for all cgroups created before */
3718 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3719 struct cgroup
*cgrp
= css
->cgroup
;
3721 if (!(css
->flags
& CSS_VISIBLE
))
3724 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3730 kernfs_activate(root
->kn
);
3734 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3738 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3739 /* free copy for custom atomic_write_len, see init_cftypes() */
3740 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3745 /* revert flags set by cgroup core while adding @cfts */
3746 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3750 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3754 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3755 struct kernfs_ops
*kf_ops
;
3757 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3760 kf_ops
= &cgroup_kf_ops
;
3762 kf_ops
= &cgroup_kf_single_ops
;
3765 * Ugh... if @cft wants a custom max_write_len, we need to
3766 * make a copy of kf_ops to set its atomic_write_len.
3768 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3769 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3771 cgroup_exit_cftypes(cfts
);
3774 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3777 cft
->kf_ops
= kf_ops
;
3784 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3786 lockdep_assert_held(&cgroup_mutex
);
3788 if (!cfts
|| !cfts
[0].ss
)
3791 list_del(&cfts
->node
);
3792 cgroup_apply_cftypes(cfts
, false);
3793 cgroup_exit_cftypes(cfts
);
3798 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3799 * @cfts: zero-length name terminated array of cftypes
3801 * Unregister @cfts. Files described by @cfts are removed from all
3802 * existing cgroups and all future cgroups won't have them either. This
3803 * function can be called anytime whether @cfts' subsys is attached or not.
3805 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3808 int cgroup_rm_cftypes(struct cftype
*cfts
)
3812 mutex_lock(&cgroup_mutex
);
3813 ret
= cgroup_rm_cftypes_locked(cfts
);
3814 mutex_unlock(&cgroup_mutex
);
3819 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3820 * @ss: target cgroup subsystem
3821 * @cfts: zero-length name terminated array of cftypes
3823 * Register @cfts to @ss. Files described by @cfts are created for all
3824 * existing cgroups to which @ss is attached and all future cgroups will
3825 * have them too. This function can be called anytime whether @ss is
3828 * Returns 0 on successful registration, -errno on failure. Note that this
3829 * function currently returns 0 as long as @cfts registration is successful
3830 * even if some file creation attempts on existing cgroups fail.
3832 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3836 if (!cgroup_ssid_enabled(ss
->id
))
3839 if (!cfts
|| cfts
[0].name
[0] == '\0')
3842 ret
= cgroup_init_cftypes(ss
, cfts
);
3846 mutex_lock(&cgroup_mutex
);
3848 list_add_tail(&cfts
->node
, &ss
->cfts
);
3849 ret
= cgroup_apply_cftypes(cfts
, true);
3851 cgroup_rm_cftypes_locked(cfts
);
3853 mutex_unlock(&cgroup_mutex
);
3858 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3859 * @ss: target cgroup subsystem
3860 * @cfts: zero-length name terminated array of cftypes
3862 * Similar to cgroup_add_cftypes() but the added files are only used for
3863 * the default hierarchy.
3865 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3869 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3870 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3871 return cgroup_add_cftypes(ss
, cfts
);
3875 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3876 * @ss: target cgroup subsystem
3877 * @cfts: zero-length name terminated array of cftypes
3879 * Similar to cgroup_add_cftypes() but the added files are only used for
3880 * the legacy hierarchies.
3882 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3886 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3887 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3888 return cgroup_add_cftypes(ss
, cfts
);
3892 * cgroup_file_notify - generate a file modified event for a cgroup_file
3893 * @cfile: target cgroup_file
3895 * @cfile must have been obtained by setting cftype->file_offset.
3897 void cgroup_file_notify(struct cgroup_file
*cfile
)
3899 unsigned long flags
;
3901 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
3903 kernfs_notify(cfile
->kn
);
3904 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
3908 * cgroup_task_count - count the number of tasks in a cgroup.
3909 * @cgrp: the cgroup in question
3911 * Return the number of tasks in the cgroup.
3913 static int cgroup_task_count(const struct cgroup
*cgrp
)
3916 struct cgrp_cset_link
*link
;
3918 spin_lock_irq(&css_set_lock
);
3919 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3920 count
+= atomic_read(&link
->cset
->refcount
);
3921 spin_unlock_irq(&css_set_lock
);
3926 * css_next_child - find the next child of a given css
3927 * @pos: the current position (%NULL to initiate traversal)
3928 * @parent: css whose children to walk
3930 * This function returns the next child of @parent and should be called
3931 * under either cgroup_mutex or RCU read lock. The only requirement is
3932 * that @parent and @pos are accessible. The next sibling is guaranteed to
3933 * be returned regardless of their states.
3935 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3936 * css which finished ->css_online() is guaranteed to be visible in the
3937 * future iterations and will stay visible until the last reference is put.
3938 * A css which hasn't finished ->css_online() or already finished
3939 * ->css_offline() may show up during traversal. It's each subsystem's
3940 * responsibility to synchronize against on/offlining.
3942 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3943 struct cgroup_subsys_state
*parent
)
3945 struct cgroup_subsys_state
*next
;
3947 cgroup_assert_mutex_or_rcu_locked();
3950 * @pos could already have been unlinked from the sibling list.
3951 * Once a cgroup is removed, its ->sibling.next is no longer
3952 * updated when its next sibling changes. CSS_RELEASED is set when
3953 * @pos is taken off list, at which time its next pointer is valid,
3954 * and, as releases are serialized, the one pointed to by the next
3955 * pointer is guaranteed to not have started release yet. This
3956 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3957 * critical section, the one pointed to by its next pointer is
3958 * guaranteed to not have finished its RCU grace period even if we
3959 * have dropped rcu_read_lock() inbetween iterations.
3961 * If @pos has CSS_RELEASED set, its next pointer can't be
3962 * dereferenced; however, as each css is given a monotonically
3963 * increasing unique serial number and always appended to the
3964 * sibling list, the next one can be found by walking the parent's
3965 * children until the first css with higher serial number than
3966 * @pos's. While this path can be slower, it happens iff iteration
3967 * races against release and the race window is very small.
3970 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3971 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3972 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3974 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3975 if (next
->serial_nr
> pos
->serial_nr
)
3980 * @next, if not pointing to the head, can be dereferenced and is
3983 if (&next
->sibling
!= &parent
->children
)
3989 * css_next_descendant_pre - find the next descendant for pre-order walk
3990 * @pos: the current position (%NULL to initiate traversal)
3991 * @root: css whose descendants to walk
3993 * To be used by css_for_each_descendant_pre(). Find the next descendant
3994 * to visit for pre-order traversal of @root's descendants. @root is
3995 * included in the iteration and the first node to be visited.
3997 * While this function requires cgroup_mutex or RCU read locking, it
3998 * doesn't require the whole traversal to be contained in a single critical
3999 * section. This function will return the correct next descendant as long
4000 * as both @pos and @root are accessible and @pos is a descendant of @root.
4002 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4003 * css which finished ->css_online() is guaranteed to be visible in the
4004 * future iterations and will stay visible until the last reference is put.
4005 * A css which hasn't finished ->css_online() or already finished
4006 * ->css_offline() may show up during traversal. It's each subsystem's
4007 * responsibility to synchronize against on/offlining.
4009 struct cgroup_subsys_state
*
4010 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
4011 struct cgroup_subsys_state
*root
)
4013 struct cgroup_subsys_state
*next
;
4015 cgroup_assert_mutex_or_rcu_locked();
4017 /* if first iteration, visit @root */
4021 /* visit the first child if exists */
4022 next
= css_next_child(NULL
, pos
);
4026 /* no child, visit my or the closest ancestor's next sibling */
4027 while (pos
!= root
) {
4028 next
= css_next_child(pos
, pos
->parent
);
4038 * css_rightmost_descendant - return the rightmost descendant of a css
4039 * @pos: css of interest
4041 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4042 * is returned. This can be used during pre-order traversal to skip
4045 * While this function requires cgroup_mutex or RCU read locking, it
4046 * doesn't require the whole traversal to be contained in a single critical
4047 * section. This function will return the correct rightmost descendant as
4048 * long as @pos is accessible.
4050 struct cgroup_subsys_state
*
4051 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
4053 struct cgroup_subsys_state
*last
, *tmp
;
4055 cgroup_assert_mutex_or_rcu_locked();
4059 /* ->prev isn't RCU safe, walk ->next till the end */
4061 css_for_each_child(tmp
, last
)
4068 static struct cgroup_subsys_state
*
4069 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
4071 struct cgroup_subsys_state
*last
;
4075 pos
= css_next_child(NULL
, pos
);
4082 * css_next_descendant_post - find the next descendant for post-order walk
4083 * @pos: the current position (%NULL to initiate traversal)
4084 * @root: css whose descendants to walk
4086 * To be used by css_for_each_descendant_post(). Find the next descendant
4087 * to visit for post-order traversal of @root's descendants. @root is
4088 * included in the iteration and the last node to be visited.
4090 * While this function requires cgroup_mutex or RCU read locking, it
4091 * doesn't require the whole traversal to be contained in a single critical
4092 * section. This function will return the correct next descendant as long
4093 * as both @pos and @cgroup are accessible and @pos is a descendant of
4096 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4097 * css which finished ->css_online() is guaranteed to be visible in the
4098 * future iterations and will stay visible until the last reference is put.
4099 * A css which hasn't finished ->css_online() or already finished
4100 * ->css_offline() may show up during traversal. It's each subsystem's
4101 * responsibility to synchronize against on/offlining.
4103 struct cgroup_subsys_state
*
4104 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
4105 struct cgroup_subsys_state
*root
)
4107 struct cgroup_subsys_state
*next
;
4109 cgroup_assert_mutex_or_rcu_locked();
4111 /* if first iteration, visit leftmost descendant which may be @root */
4113 return css_leftmost_descendant(root
);
4115 /* if we visited @root, we're done */
4119 /* if there's an unvisited sibling, visit its leftmost descendant */
4120 next
= css_next_child(pos
, pos
->parent
);
4122 return css_leftmost_descendant(next
);
4124 /* no sibling left, visit parent */
4129 * css_has_online_children - does a css have online children
4130 * @css: the target css
4132 * Returns %true if @css has any online children; otherwise, %false. This
4133 * function can be called from any context but the caller is responsible
4134 * for synchronizing against on/offlining as necessary.
4136 bool css_has_online_children(struct cgroup_subsys_state
*css
)
4138 struct cgroup_subsys_state
*child
;
4142 css_for_each_child(child
, css
) {
4143 if (child
->flags
& CSS_ONLINE
) {
4153 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4154 * @it: the iterator to advance
4156 * Advance @it to the next css_set to walk.
4158 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4160 struct list_head
*l
= it
->cset_pos
;
4161 struct cgrp_cset_link
*link
;
4162 struct css_set
*cset
;
4164 lockdep_assert_held(&css_set_lock
);
4166 /* Advance to the next non-empty css_set */
4169 if (l
== it
->cset_head
) {
4170 it
->cset_pos
= NULL
;
4171 it
->task_pos
= NULL
;
4176 cset
= container_of(l
, struct css_set
,
4177 e_cset_node
[it
->ss
->id
]);
4179 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4182 } while (!css_set_populated(cset
));
4186 if (!list_empty(&cset
->tasks
))
4187 it
->task_pos
= cset
->tasks
.next
;
4189 it
->task_pos
= cset
->mg_tasks
.next
;
4191 it
->tasks_head
= &cset
->tasks
;
4192 it
->mg_tasks_head
= &cset
->mg_tasks
;
4195 * We don't keep css_sets locked across iteration steps and thus
4196 * need to take steps to ensure that iteration can be resumed after
4197 * the lock is re-acquired. Iteration is performed at two levels -
4198 * css_sets and tasks in them.
4200 * Once created, a css_set never leaves its cgroup lists, so a
4201 * pinned css_set is guaranteed to stay put and we can resume
4202 * iteration afterwards.
4204 * Tasks may leave @cset across iteration steps. This is resolved
4205 * by registering each iterator with the css_set currently being
4206 * walked and making css_set_move_task() advance iterators whose
4207 * next task is leaving.
4210 list_del(&it
->iters_node
);
4211 put_css_set_locked(it
->cur_cset
);
4214 it
->cur_cset
= cset
;
4215 list_add(&it
->iters_node
, &cset
->task_iters
);
4218 static void css_task_iter_advance(struct css_task_iter
*it
)
4220 struct list_head
*l
= it
->task_pos
;
4222 lockdep_assert_held(&css_set_lock
);
4226 * Advance iterator to find next entry. cset->tasks is consumed
4227 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4232 if (l
== it
->tasks_head
)
4233 l
= it
->mg_tasks_head
->next
;
4235 if (l
== it
->mg_tasks_head
)
4236 css_task_iter_advance_css_set(it
);
4242 * css_task_iter_start - initiate task iteration
4243 * @css: the css to walk tasks of
4244 * @it: the task iterator to use
4246 * Initiate iteration through the tasks of @css. The caller can call
4247 * css_task_iter_next() to walk through the tasks until the function
4248 * returns NULL. On completion of iteration, css_task_iter_end() must be
4251 void css_task_iter_start(struct cgroup_subsys_state
*css
,
4252 struct css_task_iter
*it
)
4254 /* no one should try to iterate before mounting cgroups */
4255 WARN_ON_ONCE(!use_task_css_set_links
);
4257 memset(it
, 0, sizeof(*it
));
4259 spin_lock_irq(&css_set_lock
);
4264 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4266 it
->cset_pos
= &css
->cgroup
->cset_links
;
4268 it
->cset_head
= it
->cset_pos
;
4270 css_task_iter_advance_css_set(it
);
4272 spin_unlock_irq(&css_set_lock
);
4276 * css_task_iter_next - return the next task for the iterator
4277 * @it: the task iterator being iterated
4279 * The "next" function for task iteration. @it should have been
4280 * initialized via css_task_iter_start(). Returns NULL when the iteration
4283 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4286 put_task_struct(it
->cur_task
);
4287 it
->cur_task
= NULL
;
4290 spin_lock_irq(&css_set_lock
);
4293 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4295 get_task_struct(it
->cur_task
);
4296 css_task_iter_advance(it
);
4299 spin_unlock_irq(&css_set_lock
);
4301 return it
->cur_task
;
4305 * css_task_iter_end - finish task iteration
4306 * @it: the task iterator to finish
4308 * Finish task iteration started by css_task_iter_start().
4310 void css_task_iter_end(struct css_task_iter
*it
)
4313 spin_lock_irq(&css_set_lock
);
4314 list_del(&it
->iters_node
);
4315 put_css_set_locked(it
->cur_cset
);
4316 spin_unlock_irq(&css_set_lock
);
4320 put_task_struct(it
->cur_task
);
4324 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4325 * @to: cgroup to which the tasks will be moved
4326 * @from: cgroup in which the tasks currently reside
4328 * Locking rules between cgroup_post_fork() and the migration path
4329 * guarantee that, if a task is forking while being migrated, the new child
4330 * is guaranteed to be either visible in the source cgroup after the
4331 * parent's migration is complete or put into the target cgroup. No task
4332 * can slip out of migration through forking.
4334 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
4336 LIST_HEAD(preloaded_csets
);
4337 struct cgrp_cset_link
*link
;
4338 struct css_task_iter it
;
4339 struct task_struct
*task
;
4342 if (!cgroup_may_migrate_to(to
))
4345 mutex_lock(&cgroup_mutex
);
4347 /* all tasks in @from are being moved, all csets are source */
4348 spin_lock_irq(&css_set_lock
);
4349 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
4350 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
4351 spin_unlock_irq(&css_set_lock
);
4353 ret
= cgroup_migrate_prepare_dst(&preloaded_csets
);
4358 * Migrate tasks one-by-one until @from is empty. This fails iff
4359 * ->can_attach() fails.
4362 css_task_iter_start(&from
->self
, &it
);
4363 task
= css_task_iter_next(&it
);
4365 get_task_struct(task
);
4366 css_task_iter_end(&it
);
4369 ret
= cgroup_migrate(task
, false, to
->root
);
4370 put_task_struct(task
);
4372 } while (task
&& !ret
);
4374 cgroup_migrate_finish(&preloaded_csets
);
4375 mutex_unlock(&cgroup_mutex
);
4380 * Stuff for reading the 'tasks'/'procs' files.
4382 * Reading this file can return large amounts of data if a cgroup has
4383 * *lots* of attached tasks. So it may need several calls to read(),
4384 * but we cannot guarantee that the information we produce is correct
4385 * unless we produce it entirely atomically.
4389 /* which pidlist file are we talking about? */
4390 enum cgroup_filetype
{
4396 * A pidlist is a list of pids that virtually represents the contents of one
4397 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4398 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4401 struct cgroup_pidlist
{
4403 * used to find which pidlist is wanted. doesn't change as long as
4404 * this particular list stays in the list.
4406 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
4409 /* how many elements the above list has */
4411 /* each of these stored in a list by its cgroup */
4412 struct list_head links
;
4413 /* pointer to the cgroup we belong to, for list removal purposes */
4414 struct cgroup
*owner
;
4415 /* for delayed destruction */
4416 struct delayed_work destroy_dwork
;
4420 * The following two functions "fix" the issue where there are more pids
4421 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4422 * TODO: replace with a kernel-wide solution to this problem
4424 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4425 static void *pidlist_allocate(int count
)
4427 if (PIDLIST_TOO_LARGE(count
))
4428 return vmalloc(count
* sizeof(pid_t
));
4430 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
4433 static void pidlist_free(void *p
)
4439 * Used to destroy all pidlists lingering waiting for destroy timer. None
4440 * should be left afterwards.
4442 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
4444 struct cgroup_pidlist
*l
, *tmp_l
;
4446 mutex_lock(&cgrp
->pidlist_mutex
);
4447 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
4448 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
4449 mutex_unlock(&cgrp
->pidlist_mutex
);
4451 flush_workqueue(cgroup_pidlist_destroy_wq
);
4452 BUG_ON(!list_empty(&cgrp
->pidlists
));
4455 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
4457 struct delayed_work
*dwork
= to_delayed_work(work
);
4458 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
4460 struct cgroup_pidlist
*tofree
= NULL
;
4462 mutex_lock(&l
->owner
->pidlist_mutex
);
4465 * Destroy iff we didn't get queued again. The state won't change
4466 * as destroy_dwork can only be queued while locked.
4468 if (!delayed_work_pending(dwork
)) {
4469 list_del(&l
->links
);
4470 pidlist_free(l
->list
);
4471 put_pid_ns(l
->key
.ns
);
4475 mutex_unlock(&l
->owner
->pidlist_mutex
);
4480 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4481 * Returns the number of unique elements.
4483 static int pidlist_uniq(pid_t
*list
, int length
)
4488 * we presume the 0th element is unique, so i starts at 1. trivial
4489 * edge cases first; no work needs to be done for either
4491 if (length
== 0 || length
== 1)
4493 /* src and dest walk down the list; dest counts unique elements */
4494 for (src
= 1; src
< length
; src
++) {
4495 /* find next unique element */
4496 while (list
[src
] == list
[src
-1]) {
4501 /* dest always points to where the next unique element goes */
4502 list
[dest
] = list
[src
];
4510 * The two pid files - task and cgroup.procs - guaranteed that the result
4511 * is sorted, which forced this whole pidlist fiasco. As pid order is
4512 * different per namespace, each namespace needs differently sorted list,
4513 * making it impossible to use, for example, single rbtree of member tasks
4514 * sorted by task pointer. As pidlists can be fairly large, allocating one
4515 * per open file is dangerous, so cgroup had to implement shared pool of
4516 * pidlists keyed by cgroup and namespace.
4518 * All this extra complexity was caused by the original implementation
4519 * committing to an entirely unnecessary property. In the long term, we
4520 * want to do away with it. Explicitly scramble sort order if on the
4521 * default hierarchy so that no such expectation exists in the new
4524 * Scrambling is done by swapping every two consecutive bits, which is
4525 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4527 static pid_t
pid_fry(pid_t pid
)
4529 unsigned a
= pid
& 0x55555555;
4530 unsigned b
= pid
& 0xAAAAAAAA;
4532 return (a
<< 1) | (b
>> 1);
4535 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
4537 if (cgroup_on_dfl(cgrp
))
4538 return pid_fry(pid
);
4543 static int cmppid(const void *a
, const void *b
)
4545 return *(pid_t
*)a
- *(pid_t
*)b
;
4548 static int fried_cmppid(const void *a
, const void *b
)
4550 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
4553 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
4554 enum cgroup_filetype type
)
4556 struct cgroup_pidlist
*l
;
4557 /* don't need task_nsproxy() if we're looking at ourself */
4558 struct pid_namespace
*ns
= task_active_pid_ns(current
);
4560 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4562 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
4563 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
4569 * find the appropriate pidlist for our purpose (given procs vs tasks)
4570 * returns with the lock on that pidlist already held, and takes care
4571 * of the use count, or returns NULL with no locks held if we're out of
4574 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
4575 enum cgroup_filetype type
)
4577 struct cgroup_pidlist
*l
;
4579 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4581 l
= cgroup_pidlist_find(cgrp
, type
);
4585 /* entry not found; create a new one */
4586 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
4590 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
4592 /* don't need task_nsproxy() if we're looking at ourself */
4593 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
4595 list_add(&l
->links
, &cgrp
->pidlists
);
4600 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4602 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
4603 struct cgroup_pidlist
**lp
)
4607 int pid
, n
= 0; /* used for populating the array */
4608 struct css_task_iter it
;
4609 struct task_struct
*tsk
;
4610 struct cgroup_pidlist
*l
;
4612 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4615 * If cgroup gets more users after we read count, we won't have
4616 * enough space - tough. This race is indistinguishable to the
4617 * caller from the case that the additional cgroup users didn't
4618 * show up until sometime later on.
4620 length
= cgroup_task_count(cgrp
);
4621 array
= pidlist_allocate(length
);
4624 /* now, populate the array */
4625 css_task_iter_start(&cgrp
->self
, &it
);
4626 while ((tsk
= css_task_iter_next(&it
))) {
4627 if (unlikely(n
== length
))
4629 /* get tgid or pid for procs or tasks file respectively */
4630 if (type
== CGROUP_FILE_PROCS
)
4631 pid
= task_tgid_vnr(tsk
);
4633 pid
= task_pid_vnr(tsk
);
4634 if (pid
> 0) /* make sure to only use valid results */
4637 css_task_iter_end(&it
);
4639 /* now sort & (if procs) strip out duplicates */
4640 if (cgroup_on_dfl(cgrp
))
4641 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
4643 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
4644 if (type
== CGROUP_FILE_PROCS
)
4645 length
= pidlist_uniq(array
, length
);
4647 l
= cgroup_pidlist_find_create(cgrp
, type
);
4649 pidlist_free(array
);
4653 /* store array, freeing old if necessary */
4654 pidlist_free(l
->list
);
4662 * cgroupstats_build - build and fill cgroupstats
4663 * @stats: cgroupstats to fill information into
4664 * @dentry: A dentry entry belonging to the cgroup for which stats have
4667 * Build and fill cgroupstats so that taskstats can export it to user
4670 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4672 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4673 struct cgroup
*cgrp
;
4674 struct css_task_iter it
;
4675 struct task_struct
*tsk
;
4677 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4678 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4679 kernfs_type(kn
) != KERNFS_DIR
)
4682 mutex_lock(&cgroup_mutex
);
4685 * We aren't being called from kernfs and there's no guarantee on
4686 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4687 * @kn->priv is RCU safe. Let's do the RCU dancing.
4690 cgrp
= rcu_dereference(kn
->priv
);
4691 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4693 mutex_unlock(&cgroup_mutex
);
4698 css_task_iter_start(&cgrp
->self
, &it
);
4699 while ((tsk
= css_task_iter_next(&it
))) {
4700 switch (tsk
->state
) {
4702 stats
->nr_running
++;
4704 case TASK_INTERRUPTIBLE
:
4705 stats
->nr_sleeping
++;
4707 case TASK_UNINTERRUPTIBLE
:
4708 stats
->nr_uninterruptible
++;
4711 stats
->nr_stopped
++;
4714 if (delayacct_is_task_waiting_on_io(tsk
))
4715 stats
->nr_io_wait
++;
4719 css_task_iter_end(&it
);
4721 mutex_unlock(&cgroup_mutex
);
4727 * seq_file methods for the tasks/procs files. The seq_file position is the
4728 * next pid to display; the seq_file iterator is a pointer to the pid
4729 * in the cgroup->l->list array.
4732 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4735 * Initially we receive a position value that corresponds to
4736 * one more than the last pid shown (or 0 on the first call or
4737 * after a seek to the start). Use a binary-search to find the
4738 * next pid to display, if any
4740 struct kernfs_open_file
*of
= s
->private;
4741 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4742 struct cgroup_pidlist
*l
;
4743 enum cgroup_filetype type
= seq_cft(s
)->private;
4744 int index
= 0, pid
= *pos
;
4747 mutex_lock(&cgrp
->pidlist_mutex
);
4750 * !NULL @of->priv indicates that this isn't the first start()
4751 * after open. If the matching pidlist is around, we can use that.
4752 * Look for it. Note that @of->priv can't be used directly. It
4753 * could already have been destroyed.
4756 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4759 * Either this is the first start() after open or the matching
4760 * pidlist has been destroyed inbetween. Create a new one.
4763 ret
= pidlist_array_load(cgrp
, type
,
4764 (struct cgroup_pidlist
**)&of
->priv
);
4766 return ERR_PTR(ret
);
4771 int end
= l
->length
;
4773 while (index
< end
) {
4774 int mid
= (index
+ end
) / 2;
4775 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4778 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4784 /* If we're off the end of the array, we're done */
4785 if (index
>= l
->length
)
4787 /* Update the abstract position to be the actual pid that we found */
4788 iter
= l
->list
+ index
;
4789 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4793 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4795 struct kernfs_open_file
*of
= s
->private;
4796 struct cgroup_pidlist
*l
= of
->priv
;
4799 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4800 CGROUP_PIDLIST_DESTROY_DELAY
);
4801 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4804 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4806 struct kernfs_open_file
*of
= s
->private;
4807 struct cgroup_pidlist
*l
= of
->priv
;
4809 pid_t
*end
= l
->list
+ l
->length
;
4811 * Advance to the next pid in the array. If this goes off the
4818 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4823 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4825 seq_printf(s
, "%d\n", *(int *)v
);
4830 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4833 return notify_on_release(css
->cgroup
);
4836 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4837 struct cftype
*cft
, u64 val
)
4840 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4842 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4846 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4849 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4852 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4853 struct cftype
*cft
, u64 val
)
4856 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4858 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4862 /* cgroup core interface files for the default hierarchy */
4863 static struct cftype cgroup_dfl_base_files
[] = {
4865 .name
= "cgroup.procs",
4866 .file_offset
= offsetof(struct cgroup
, procs_file
),
4867 .seq_start
= cgroup_pidlist_start
,
4868 .seq_next
= cgroup_pidlist_next
,
4869 .seq_stop
= cgroup_pidlist_stop
,
4870 .seq_show
= cgroup_pidlist_show
,
4871 .private = CGROUP_FILE_PROCS
,
4872 .write
= cgroup_procs_write
,
4875 .name
= "cgroup.controllers",
4876 .seq_show
= cgroup_controllers_show
,
4879 .name
= "cgroup.subtree_control",
4880 .seq_show
= cgroup_subtree_control_show
,
4881 .write
= cgroup_subtree_control_write
,
4884 .name
= "cgroup.events",
4885 .flags
= CFTYPE_NOT_ON_ROOT
,
4886 .file_offset
= offsetof(struct cgroup
, events_file
),
4887 .seq_show
= cgroup_events_show
,
4892 /* cgroup core interface files for the legacy hierarchies */
4893 static struct cftype cgroup_legacy_base_files
[] = {
4895 .name
= "cgroup.procs",
4896 .seq_start
= cgroup_pidlist_start
,
4897 .seq_next
= cgroup_pidlist_next
,
4898 .seq_stop
= cgroup_pidlist_stop
,
4899 .seq_show
= cgroup_pidlist_show
,
4900 .private = CGROUP_FILE_PROCS
,
4901 .write
= cgroup_procs_write
,
4904 .name
= "cgroup.clone_children",
4905 .read_u64
= cgroup_clone_children_read
,
4906 .write_u64
= cgroup_clone_children_write
,
4909 .name
= "cgroup.sane_behavior",
4910 .flags
= CFTYPE_ONLY_ON_ROOT
,
4911 .seq_show
= cgroup_sane_behavior_show
,
4915 .seq_start
= cgroup_pidlist_start
,
4916 .seq_next
= cgroup_pidlist_next
,
4917 .seq_stop
= cgroup_pidlist_stop
,
4918 .seq_show
= cgroup_pidlist_show
,
4919 .private = CGROUP_FILE_TASKS
,
4920 .write
= cgroup_tasks_write
,
4923 .name
= "notify_on_release",
4924 .read_u64
= cgroup_read_notify_on_release
,
4925 .write_u64
= cgroup_write_notify_on_release
,
4928 .name
= "release_agent",
4929 .flags
= CFTYPE_ONLY_ON_ROOT
,
4930 .seq_show
= cgroup_release_agent_show
,
4931 .write
= cgroup_release_agent_write
,
4932 .max_write_len
= PATH_MAX
- 1,
4938 * css destruction is four-stage process.
4940 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4941 * Implemented in kill_css().
4943 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4944 * and thus css_tryget_online() is guaranteed to fail, the css can be
4945 * offlined by invoking offline_css(). After offlining, the base ref is
4946 * put. Implemented in css_killed_work_fn().
4948 * 3. When the percpu_ref reaches zero, the only possible remaining
4949 * accessors are inside RCU read sections. css_release() schedules the
4952 * 4. After the grace period, the css can be freed. Implemented in
4953 * css_free_work_fn().
4955 * It is actually hairier because both step 2 and 4 require process context
4956 * and thus involve punting to css->destroy_work adding two additional
4957 * steps to the already complex sequence.
4959 static void css_free_work_fn(struct work_struct
*work
)
4961 struct cgroup_subsys_state
*css
=
4962 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4963 struct cgroup_subsys
*ss
= css
->ss
;
4964 struct cgroup
*cgrp
= css
->cgroup
;
4966 percpu_ref_exit(&css
->refcnt
);
4970 struct cgroup_subsys_state
*parent
= css
->parent
;
4974 cgroup_idr_remove(&ss
->css_idr
, id
);
4980 /* cgroup free path */
4981 atomic_dec(&cgrp
->root
->nr_cgrps
);
4982 cgroup_pidlist_destroy_all(cgrp
);
4983 cancel_work_sync(&cgrp
->release_agent_work
);
4985 if (cgroup_parent(cgrp
)) {
4987 * We get a ref to the parent, and put the ref when
4988 * this cgroup is being freed, so it's guaranteed
4989 * that the parent won't be destroyed before its
4992 cgroup_put(cgroup_parent(cgrp
));
4993 kernfs_put(cgrp
->kn
);
4997 * This is root cgroup's refcnt reaching zero,
4998 * which indicates that the root should be
5001 cgroup_destroy_root(cgrp
->root
);
5006 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
5008 struct cgroup_subsys_state
*css
=
5009 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
5011 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
5012 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5015 static void css_release_work_fn(struct work_struct
*work
)
5017 struct cgroup_subsys_state
*css
=
5018 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5019 struct cgroup_subsys
*ss
= css
->ss
;
5020 struct cgroup
*cgrp
= css
->cgroup
;
5022 mutex_lock(&cgroup_mutex
);
5024 css
->flags
|= CSS_RELEASED
;
5025 list_del_rcu(&css
->sibling
);
5028 /* css release path */
5029 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
5030 if (ss
->css_released
)
5031 ss
->css_released(css
);
5033 /* cgroup release path */
5034 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
5038 * There are two control paths which try to determine
5039 * cgroup from dentry without going through kernfs -
5040 * cgroupstats_build() and css_tryget_online_from_dir().
5041 * Those are supported by RCU protecting clearing of
5042 * cgrp->kn->priv backpointer.
5045 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
5049 mutex_unlock(&cgroup_mutex
);
5051 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
5054 static void css_release(struct percpu_ref
*ref
)
5056 struct cgroup_subsys_state
*css
=
5057 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5059 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
5060 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5063 static void init_and_link_css(struct cgroup_subsys_state
*css
,
5064 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
5066 lockdep_assert_held(&cgroup_mutex
);
5070 memset(css
, 0, sizeof(*css
));
5074 INIT_LIST_HEAD(&css
->sibling
);
5075 INIT_LIST_HEAD(&css
->children
);
5076 css
->serial_nr
= css_serial_nr_next
++;
5077 atomic_set(&css
->online_cnt
, 0);
5079 if (cgroup_parent(cgrp
)) {
5080 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
5081 css_get(css
->parent
);
5084 BUG_ON(cgroup_css(cgrp
, ss
));
5087 /* invoke ->css_online() on a new CSS and mark it online if successful */
5088 static int online_css(struct cgroup_subsys_state
*css
)
5090 struct cgroup_subsys
*ss
= css
->ss
;
5093 lockdep_assert_held(&cgroup_mutex
);
5096 ret
= ss
->css_online(css
);
5098 css
->flags
|= CSS_ONLINE
;
5099 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
5101 atomic_inc(&css
->online_cnt
);
5103 atomic_inc(&css
->parent
->online_cnt
);
5108 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5109 static void offline_css(struct cgroup_subsys_state
*css
)
5111 struct cgroup_subsys
*ss
= css
->ss
;
5113 lockdep_assert_held(&cgroup_mutex
);
5115 if (!(css
->flags
& CSS_ONLINE
))
5121 if (ss
->css_offline
)
5122 ss
->css_offline(css
);
5124 css
->flags
&= ~CSS_ONLINE
;
5125 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
5127 wake_up_all(&css
->cgroup
->offline_waitq
);
5131 * css_create - create a cgroup_subsys_state
5132 * @cgrp: the cgroup new css will be associated with
5133 * @ss: the subsys of new css
5135 * Create a new css associated with @cgrp - @ss pair. On success, the new
5136 * css is online and installed in @cgrp. This function doesn't create the
5137 * interface files. Returns 0 on success, -errno on failure.
5139 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
5140 struct cgroup_subsys
*ss
)
5142 struct cgroup
*parent
= cgroup_parent(cgrp
);
5143 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
5144 struct cgroup_subsys_state
*css
;
5147 lockdep_assert_held(&cgroup_mutex
);
5149 css
= ss
->css_alloc(parent_css
);
5153 init_and_link_css(css
, ss
, cgrp
);
5155 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
5159 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
5164 /* @css is ready to be brought online now, make it visible */
5165 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
5166 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
5168 err
= online_css(css
);
5172 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
5173 cgroup_parent(parent
)) {
5174 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5175 current
->comm
, current
->pid
, ss
->name
);
5176 if (!strcmp(ss
->name
, "memory"))
5177 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5178 ss
->warned_broken_hierarchy
= true;
5184 list_del_rcu(&css
->sibling
);
5186 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
5187 return ERR_PTR(err
);
5190 static struct cgroup
*cgroup_create(struct cgroup
*parent
)
5192 struct cgroup_root
*root
= parent
->root
;
5193 struct cgroup
*cgrp
, *tcgrp
;
5194 int level
= parent
->level
+ 1;
5197 /* allocate the cgroup and its ID, 0 is reserved for the root */
5198 cgrp
= kzalloc(sizeof(*cgrp
) +
5199 sizeof(cgrp
->ancestor_ids
[0]) * (level
+ 1), GFP_KERNEL
);
5201 return ERR_PTR(-ENOMEM
);
5203 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5208 * Temporarily set the pointer to NULL, so idr_find() won't return
5209 * a half-baked cgroup.
5211 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
5214 goto out_cancel_ref
;
5217 init_cgroup_housekeeping(cgrp
);
5219 cgrp
->self
.parent
= &parent
->self
;
5221 cgrp
->level
= level
;
5223 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
))
5224 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
5226 if (notify_on_release(parent
))
5227 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5229 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5230 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5232 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5234 /* allocation complete, commit to creation */
5235 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5236 atomic_inc(&root
->nr_cgrps
);
5240 * @cgrp is now fully operational. If something fails after this
5241 * point, it'll be released via the normal destruction path.
5243 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
5246 * On the default hierarchy, a child doesn't automatically inherit
5247 * subtree_control from the parent. Each is configured manually.
5249 if (!cgroup_on_dfl(cgrp
))
5250 cgrp
->subtree_control
= cgroup_control(cgrp
);
5252 cgroup_propagate_control(cgrp
);
5254 /* @cgrp doesn't have dir yet so the following will only create csses */
5255 ret
= cgroup_apply_control_enable(cgrp
);
5262 percpu_ref_exit(&cgrp
->self
.refcnt
);
5265 return ERR_PTR(ret
);
5267 cgroup_destroy_locked(cgrp
);
5268 return ERR_PTR(ret
);
5271 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
5274 struct cgroup
*parent
, *cgrp
;
5275 struct kernfs_node
*kn
;
5278 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5279 if (strchr(name
, '\n'))
5282 parent
= cgroup_kn_lock_live(parent_kn
, false);
5286 cgrp
= cgroup_create(parent
);
5288 ret
= PTR_ERR(cgrp
);
5292 /* create the directory */
5293 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5301 * This extra ref will be put in cgroup_free_fn() and guarantees
5302 * that @cgrp->kn is always accessible.
5306 ret
= cgroup_kn_set_ugid(kn
);
5310 ret
= css_populate_dir(&cgrp
->self
);
5314 ret
= cgroup_apply_control_enable(cgrp
);
5318 /* let's create and online css's */
5319 kernfs_activate(kn
);
5325 cgroup_destroy_locked(cgrp
);
5327 cgroup_kn_unlock(parent_kn
);
5332 * This is called when the refcnt of a css is confirmed to be killed.
5333 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5334 * initate destruction and put the css ref from kill_css().
5336 static void css_killed_work_fn(struct work_struct
*work
)
5338 struct cgroup_subsys_state
*css
=
5339 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5341 mutex_lock(&cgroup_mutex
);
5346 /* @css can't go away while we're holding cgroup_mutex */
5348 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5350 mutex_unlock(&cgroup_mutex
);
5353 /* css kill confirmation processing requires process context, bounce */
5354 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5356 struct cgroup_subsys_state
*css
=
5357 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5359 if (atomic_dec_and_test(&css
->online_cnt
)) {
5360 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5361 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5366 * kill_css - destroy a css
5367 * @css: css to destroy
5369 * This function initiates destruction of @css by removing cgroup interface
5370 * files and putting its base reference. ->css_offline() will be invoked
5371 * asynchronously once css_tryget_online() is guaranteed to fail and when
5372 * the reference count reaches zero, @css will be released.
5374 static void kill_css(struct cgroup_subsys_state
*css
)
5376 lockdep_assert_held(&cgroup_mutex
);
5379 * This must happen before css is disassociated with its cgroup.
5380 * See seq_css() for details.
5385 * Killing would put the base ref, but we need to keep it alive
5386 * until after ->css_offline().
5391 * cgroup core guarantees that, by the time ->css_offline() is
5392 * invoked, no new css reference will be given out via
5393 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5394 * proceed to offlining css's because percpu_ref_kill() doesn't
5395 * guarantee that the ref is seen as killed on all CPUs on return.
5397 * Use percpu_ref_kill_and_confirm() to get notifications as each
5398 * css is confirmed to be seen as killed on all CPUs.
5400 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5404 * cgroup_destroy_locked - the first stage of cgroup destruction
5405 * @cgrp: cgroup to be destroyed
5407 * css's make use of percpu refcnts whose killing latency shouldn't be
5408 * exposed to userland and are RCU protected. Also, cgroup core needs to
5409 * guarantee that css_tryget_online() won't succeed by the time
5410 * ->css_offline() is invoked. To satisfy all the requirements,
5411 * destruction is implemented in the following two steps.
5413 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5414 * userland visible parts and start killing the percpu refcnts of
5415 * css's. Set up so that the next stage will be kicked off once all
5416 * the percpu refcnts are confirmed to be killed.
5418 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5419 * rest of destruction. Once all cgroup references are gone, the
5420 * cgroup is RCU-freed.
5422 * This function implements s1. After this step, @cgrp is gone as far as
5423 * the userland is concerned and a new cgroup with the same name may be
5424 * created. As cgroup doesn't care about the names internally, this
5425 * doesn't cause any problem.
5427 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5428 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5430 struct cgroup_subsys_state
*css
;
5431 struct cgrp_cset_link
*link
;
5434 lockdep_assert_held(&cgroup_mutex
);
5437 * Only migration can raise populated from zero and we're already
5438 * holding cgroup_mutex.
5440 if (cgroup_is_populated(cgrp
))
5444 * Make sure there's no live children. We can't test emptiness of
5445 * ->self.children as dead children linger on it while being
5446 * drained; otherwise, "rmdir parent/child parent" may fail.
5448 if (css_has_online_children(&cgrp
->self
))
5452 * Mark @cgrp and the associated csets dead. The former prevents
5453 * further task migration and child creation by disabling
5454 * cgroup_lock_live_group(). The latter makes the csets ignored by
5455 * the migration path.
5457 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5459 spin_lock_irq(&css_set_lock
);
5460 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5461 link
->cset
->dead
= true;
5462 spin_unlock_irq(&css_set_lock
);
5464 /* initiate massacre of all css's */
5465 for_each_css(css
, ssid
, cgrp
)
5469 * Remove @cgrp directory along with the base files. @cgrp has an
5470 * extra ref on its kn.
5472 kernfs_remove(cgrp
->kn
);
5474 check_for_release(cgroup_parent(cgrp
));
5476 /* put the base reference */
5477 percpu_ref_kill(&cgrp
->self
.refcnt
);
5482 static int cgroup_rmdir(struct kernfs_node
*kn
)
5484 struct cgroup
*cgrp
;
5487 cgrp
= cgroup_kn_lock_live(kn
, false);
5491 ret
= cgroup_destroy_locked(cgrp
);
5493 cgroup_kn_unlock(kn
);
5497 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5498 .remount_fs
= cgroup_remount
,
5499 .show_options
= cgroup_show_options
,
5500 .mkdir
= cgroup_mkdir
,
5501 .rmdir
= cgroup_rmdir
,
5502 .rename
= cgroup_rename
,
5503 .show_path
= cgroup_show_path
,
5506 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5508 struct cgroup_subsys_state
*css
;
5510 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5512 mutex_lock(&cgroup_mutex
);
5514 idr_init(&ss
->css_idr
);
5515 INIT_LIST_HEAD(&ss
->cfts
);
5517 /* Create the root cgroup state for this subsystem */
5518 ss
->root
= &cgrp_dfl_root
;
5519 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5520 /* We don't handle early failures gracefully */
5521 BUG_ON(IS_ERR(css
));
5522 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5525 * Root csses are never destroyed and we can't initialize
5526 * percpu_ref during early init. Disable refcnting.
5528 css
->flags
|= CSS_NO_REF
;
5531 /* allocation can't be done safely during early init */
5534 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5535 BUG_ON(css
->id
< 0);
5538 /* Update the init_css_set to contain a subsys
5539 * pointer to this state - since the subsystem is
5540 * newly registered, all tasks and hence the
5541 * init_css_set is in the subsystem's root cgroup. */
5542 init_css_set
.subsys
[ss
->id
] = css
;
5544 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5545 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5546 have_free_callback
|= (bool)ss
->free
<< ss
->id
;
5547 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5549 /* At system boot, before all subsystems have been
5550 * registered, no tasks have been forked, so we don't
5551 * need to invoke fork callbacks here. */
5552 BUG_ON(!list_empty(&init_task
.tasks
));
5554 BUG_ON(online_css(css
));
5556 mutex_unlock(&cgroup_mutex
);
5560 * cgroup_init_early - cgroup initialization at system boot
5562 * Initialize cgroups at system boot, and initialize any
5563 * subsystems that request early init.
5565 int __init
cgroup_init_early(void)
5567 static struct cgroup_sb_opts __initdata opts
;
5568 struct cgroup_subsys
*ss
;
5571 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5572 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5574 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5576 for_each_subsys(ss
, i
) {
5577 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5578 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5579 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5581 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5582 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5585 ss
->name
= cgroup_subsys_name
[i
];
5586 if (!ss
->legacy_name
)
5587 ss
->legacy_name
= cgroup_subsys_name
[i
];
5590 cgroup_init_subsys(ss
, true);
5595 static u16 cgroup_disable_mask __initdata
;
5598 * cgroup_init - cgroup initialization
5600 * Register cgroup filesystem and /proc file, and initialize
5601 * any subsystems that didn't request early init.
5603 int __init
cgroup_init(void)
5605 struct cgroup_subsys
*ss
;
5608 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5609 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5610 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
5611 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
5613 get_user_ns(init_cgroup_ns
.user_ns
);
5615 mutex_lock(&cgroup_mutex
);
5618 * Add init_css_set to the hash table so that dfl_root can link to
5621 hash_add(css_set_table
, &init_css_set
.hlist
,
5622 css_set_hash(init_css_set
.subsys
));
5624 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5626 mutex_unlock(&cgroup_mutex
);
5628 for_each_subsys(ss
, ssid
) {
5629 if (ss
->early_init
) {
5630 struct cgroup_subsys_state
*css
=
5631 init_css_set
.subsys
[ss
->id
];
5633 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5635 BUG_ON(css
->id
< 0);
5637 cgroup_init_subsys(ss
, false);
5640 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5641 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5644 * Setting dfl_root subsys_mask needs to consider the
5645 * disabled flag and cftype registration needs kmalloc,
5646 * both of which aren't available during early_init.
5648 if (cgroup_disable_mask
& (1 << ssid
)) {
5649 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5650 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5655 if (cgroup_ssid_no_v1(ssid
))
5656 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5659 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5661 if (ss
->implicit_on_dfl
)
5662 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5663 else if (!ss
->dfl_cftypes
)
5664 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5666 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5667 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5669 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5670 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5674 ss
->bind(init_css_set
.subsys
[ssid
]);
5677 /* init_css_set.subsys[] has been updated, re-hash */
5678 hash_del(&init_css_set
.hlist
);
5679 hash_add(css_set_table
, &init_css_set
.hlist
,
5680 css_set_hash(init_css_set
.subsys
));
5682 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5683 WARN_ON(register_filesystem(&cgroup_fs_type
));
5684 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5685 WARN_ON(!proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
));
5690 static int __init
cgroup_wq_init(void)
5693 * There isn't much point in executing destruction path in
5694 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5695 * Use 1 for @max_active.
5697 * We would prefer to do this in cgroup_init() above, but that
5698 * is called before init_workqueues(): so leave this until after.
5700 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5701 BUG_ON(!cgroup_destroy_wq
);
5704 * Used to destroy pidlists and separate to serve as flush domain.
5705 * Cap @max_active to 1 too.
5707 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5709 BUG_ON(!cgroup_pidlist_destroy_wq
);
5713 core_initcall(cgroup_wq_init
);
5716 * proc_cgroup_show()
5717 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5718 * - Used for /proc/<pid>/cgroup.
5720 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5721 struct pid
*pid
, struct task_struct
*tsk
)
5725 struct cgroup_root
*root
;
5728 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5732 mutex_lock(&cgroup_mutex
);
5733 spin_lock_irq(&css_set_lock
);
5735 for_each_root(root
) {
5736 struct cgroup_subsys
*ss
;
5737 struct cgroup
*cgrp
;
5738 int ssid
, count
= 0;
5740 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5743 seq_printf(m
, "%d:", root
->hierarchy_id
);
5744 if (root
!= &cgrp_dfl_root
)
5745 for_each_subsys(ss
, ssid
)
5746 if (root
->subsys_mask
& (1 << ssid
))
5747 seq_printf(m
, "%s%s", count
++ ? "," : "",
5749 if (strlen(root
->name
))
5750 seq_printf(m
, "%sname=%s", count
? "," : "",
5754 cgrp
= task_cgroup_from_root(tsk
, root
);
5757 * On traditional hierarchies, all zombie tasks show up as
5758 * belonging to the root cgroup. On the default hierarchy,
5759 * while a zombie doesn't show up in "cgroup.procs" and
5760 * thus can't be migrated, its /proc/PID/cgroup keeps
5761 * reporting the cgroup it belonged to before exiting. If
5762 * the cgroup is removed before the zombie is reaped,
5763 * " (deleted)" is appended to the cgroup path.
5765 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5766 path
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5767 current
->nsproxy
->cgroup_ns
);
5769 retval
= -ENAMETOOLONG
;
5778 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5779 seq_puts(m
, " (deleted)\n");
5786 spin_unlock_irq(&css_set_lock
);
5787 mutex_unlock(&cgroup_mutex
);
5793 /* Display information about each subsystem and each hierarchy */
5794 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5796 struct cgroup_subsys
*ss
;
5799 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5801 * ideally we don't want subsystems moving around while we do this.
5802 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5803 * subsys/hierarchy state.
5805 mutex_lock(&cgroup_mutex
);
5807 for_each_subsys(ss
, i
)
5808 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5809 ss
->legacy_name
, ss
->root
->hierarchy_id
,
5810 atomic_read(&ss
->root
->nr_cgrps
),
5811 cgroup_ssid_enabled(i
));
5813 mutex_unlock(&cgroup_mutex
);
5817 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5819 return single_open(file
, proc_cgroupstats_show
, NULL
);
5822 static const struct file_operations proc_cgroupstats_operations
= {
5823 .open
= cgroupstats_open
,
5825 .llseek
= seq_lseek
,
5826 .release
= single_release
,
5830 * cgroup_fork - initialize cgroup related fields during copy_process()
5831 * @child: pointer to task_struct of forking parent process.
5833 * A task is associated with the init_css_set until cgroup_post_fork()
5834 * attaches it to the parent's css_set. Empty cg_list indicates that
5835 * @child isn't holding reference to its css_set.
5837 void cgroup_fork(struct task_struct
*child
)
5839 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5840 INIT_LIST_HEAD(&child
->cg_list
);
5844 * cgroup_can_fork - called on a new task before the process is exposed
5845 * @child: the task in question.
5847 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5848 * returns an error, the fork aborts with that error code. This allows for
5849 * a cgroup subsystem to conditionally allow or deny new forks.
5851 int cgroup_can_fork(struct task_struct
*child
)
5853 struct cgroup_subsys
*ss
;
5856 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
5857 ret
= ss
->can_fork(child
);
5860 } while_each_subsys_mask();
5865 for_each_subsys(ss
, j
) {
5868 if (ss
->cancel_fork
)
5869 ss
->cancel_fork(child
);
5876 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5877 * @child: the task in question
5879 * This calls the cancel_fork() callbacks if a fork failed *after*
5880 * cgroup_can_fork() succeded.
5882 void cgroup_cancel_fork(struct task_struct
*child
)
5884 struct cgroup_subsys
*ss
;
5887 for_each_subsys(ss
, i
)
5888 if (ss
->cancel_fork
)
5889 ss
->cancel_fork(child
);
5893 * cgroup_post_fork - called on a new task after adding it to the task list
5894 * @child: the task in question
5896 * Adds the task to the list running through its css_set if necessary and
5897 * call the subsystem fork() callbacks. Has to be after the task is
5898 * visible on the task list in case we race with the first call to
5899 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5902 void cgroup_post_fork(struct task_struct
*child
)
5904 struct cgroup_subsys
*ss
;
5908 * This may race against cgroup_enable_task_cg_lists(). As that
5909 * function sets use_task_css_set_links before grabbing
5910 * tasklist_lock and we just went through tasklist_lock to add
5911 * @child, it's guaranteed that either we see the set
5912 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5913 * @child during its iteration.
5915 * If we won the race, @child is associated with %current's
5916 * css_set. Grabbing css_set_lock guarantees both that the
5917 * association is stable, and, on completion of the parent's
5918 * migration, @child is visible in the source of migration or
5919 * already in the destination cgroup. This guarantee is necessary
5920 * when implementing operations which need to migrate all tasks of
5921 * a cgroup to another.
5923 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5924 * will remain in init_css_set. This is safe because all tasks are
5925 * in the init_css_set before cg_links is enabled and there's no
5926 * operation which transfers all tasks out of init_css_set.
5928 if (use_task_css_set_links
) {
5929 struct css_set
*cset
;
5931 spin_lock_irq(&css_set_lock
);
5932 cset
= task_css_set(current
);
5933 if (list_empty(&child
->cg_list
)) {
5935 css_set_move_task(child
, NULL
, cset
, false);
5937 spin_unlock_irq(&css_set_lock
);
5941 * Call ss->fork(). This must happen after @child is linked on
5942 * css_set; otherwise, @child might change state between ->fork()
5943 * and addition to css_set.
5945 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
5947 } while_each_subsys_mask();
5951 * cgroup_exit - detach cgroup from exiting task
5952 * @tsk: pointer to task_struct of exiting process
5954 * Description: Detach cgroup from @tsk and release it.
5956 * Note that cgroups marked notify_on_release force every task in
5957 * them to take the global cgroup_mutex mutex when exiting.
5958 * This could impact scaling on very large systems. Be reluctant to
5959 * use notify_on_release cgroups where very high task exit scaling
5960 * is required on large systems.
5962 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5963 * call cgroup_exit() while the task is still competent to handle
5964 * notify_on_release(), then leave the task attached to the root cgroup in
5965 * each hierarchy for the remainder of its exit. No need to bother with
5966 * init_css_set refcnting. init_css_set never goes away and we can't race
5967 * with migration path - PF_EXITING is visible to migration path.
5969 void cgroup_exit(struct task_struct
*tsk
)
5971 struct cgroup_subsys
*ss
;
5972 struct css_set
*cset
;
5976 * Unlink from @tsk from its css_set. As migration path can't race
5977 * with us, we can check css_set and cg_list without synchronization.
5979 cset
= task_css_set(tsk
);
5981 if (!list_empty(&tsk
->cg_list
)) {
5982 spin_lock_irq(&css_set_lock
);
5983 css_set_move_task(tsk
, cset
, NULL
, false);
5984 spin_unlock_irq(&css_set_lock
);
5989 /* see cgroup_post_fork() for details */
5990 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
5992 } while_each_subsys_mask();
5995 void cgroup_free(struct task_struct
*task
)
5997 struct css_set
*cset
= task_css_set(task
);
5998 struct cgroup_subsys
*ss
;
6001 do_each_subsys_mask(ss
, ssid
, have_free_callback
) {
6003 } while_each_subsys_mask();
6008 static void check_for_release(struct cgroup
*cgrp
)
6010 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
6011 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
6012 schedule_work(&cgrp
->release_agent_work
);
6016 * Notify userspace when a cgroup is released, by running the
6017 * configured release agent with the name of the cgroup (path
6018 * relative to the root of cgroup file system) as the argument.
6020 * Most likely, this user command will try to rmdir this cgroup.
6022 * This races with the possibility that some other task will be
6023 * attached to this cgroup before it is removed, or that some other
6024 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
6025 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
6026 * unused, and this cgroup will be reprieved from its death sentence,
6027 * to continue to serve a useful existence. Next time it's released,
6028 * we will get notified again, if it still has 'notify_on_release' set.
6030 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
6031 * means only wait until the task is successfully execve()'d. The
6032 * separate release agent task is forked by call_usermodehelper(),
6033 * then control in this thread returns here, without waiting for the
6034 * release agent task. We don't bother to wait because the caller of
6035 * this routine has no use for the exit status of the release agent
6036 * task, so no sense holding our caller up for that.
6038 static void cgroup_release_agent(struct work_struct
*work
)
6040 struct cgroup
*cgrp
=
6041 container_of(work
, struct cgroup
, release_agent_work
);
6042 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
6043 char *argv
[3], *envp
[3];
6045 mutex_lock(&cgroup_mutex
);
6047 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
6048 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
6049 if (!pathbuf
|| !agentbuf
)
6052 spin_lock_irq(&css_set_lock
);
6053 path
= cgroup_path_ns_locked(cgrp
, pathbuf
, PATH_MAX
, &init_cgroup_ns
);
6054 spin_unlock_irq(&css_set_lock
);
6062 /* minimal command environment */
6064 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
6067 mutex_unlock(&cgroup_mutex
);
6068 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
6071 mutex_unlock(&cgroup_mutex
);
6077 static int __init
cgroup_disable(char *str
)
6079 struct cgroup_subsys
*ss
;
6083 while ((token
= strsep(&str
, ",")) != NULL
) {
6087 for_each_subsys(ss
, i
) {
6088 if (strcmp(token
, ss
->name
) &&
6089 strcmp(token
, ss
->legacy_name
))
6091 cgroup_disable_mask
|= 1 << i
;
6096 __setup("cgroup_disable=", cgroup_disable
);
6098 static int __init
cgroup_no_v1(char *str
)
6100 struct cgroup_subsys
*ss
;
6104 while ((token
= strsep(&str
, ",")) != NULL
) {
6108 if (!strcmp(token
, "all")) {
6109 cgroup_no_v1_mask
= U16_MAX
;
6113 for_each_subsys(ss
, i
) {
6114 if (strcmp(token
, ss
->name
) &&
6115 strcmp(token
, ss
->legacy_name
))
6118 cgroup_no_v1_mask
|= 1 << i
;
6123 __setup("cgroup_no_v1=", cgroup_no_v1
);
6126 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6127 * @dentry: directory dentry of interest
6128 * @ss: subsystem of interest
6130 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6131 * to get the corresponding css and return it. If such css doesn't exist
6132 * or can't be pinned, an ERR_PTR value is returned.
6134 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
6135 struct cgroup_subsys
*ss
)
6137 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
6138 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
6139 struct cgroup_subsys_state
*css
= NULL
;
6140 struct cgroup
*cgrp
;
6142 /* is @dentry a cgroup dir? */
6143 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
6144 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
6145 return ERR_PTR(-EBADF
);
6150 * This path doesn't originate from kernfs and @kn could already
6151 * have been or be removed at any point. @kn->priv is RCU
6152 * protected for this access. See css_release_work_fn() for details.
6154 cgrp
= rcu_dereference(kn
->priv
);
6156 css
= cgroup_css(cgrp
, ss
);
6158 if (!css
|| !css_tryget_online(css
))
6159 css
= ERR_PTR(-ENOENT
);
6166 * css_from_id - lookup css by id
6167 * @id: the cgroup id
6168 * @ss: cgroup subsys to be looked into
6170 * Returns the css if there's valid one with @id, otherwise returns NULL.
6171 * Should be called under rcu_read_lock().
6173 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
6175 WARN_ON_ONCE(!rcu_read_lock_held());
6176 return id
> 0 ? idr_find(&ss
->css_idr
, id
) : NULL
;
6180 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6181 * @path: path on the default hierarchy
6183 * Find the cgroup at @path on the default hierarchy, increment its
6184 * reference count and return it. Returns pointer to the found cgroup on
6185 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6186 * if @path points to a non-directory.
6188 struct cgroup
*cgroup_get_from_path(const char *path
)
6190 struct kernfs_node
*kn
;
6191 struct cgroup
*cgrp
;
6193 mutex_lock(&cgroup_mutex
);
6195 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
6197 if (kernfs_type(kn
) == KERNFS_DIR
) {
6201 cgrp
= ERR_PTR(-ENOTDIR
);
6205 cgrp
= ERR_PTR(-ENOENT
);
6208 mutex_unlock(&cgroup_mutex
);
6211 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
6214 * cgroup_get_from_fd - get a cgroup pointer from a fd
6215 * @fd: fd obtained by open(cgroup2_dir)
6217 * Find the cgroup from a fd which should be obtained
6218 * by opening a cgroup directory. Returns a pointer to the
6219 * cgroup on success. ERR_PTR is returned if the cgroup
6222 struct cgroup
*cgroup_get_from_fd(int fd
)
6224 struct cgroup_subsys_state
*css
;
6225 struct cgroup
*cgrp
;
6230 return ERR_PTR(-EBADF
);
6232 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
6235 return ERR_CAST(css
);
6238 if (!cgroup_on_dfl(cgrp
)) {
6240 return ERR_PTR(-EBADF
);
6245 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
6248 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6249 * definition in cgroup-defs.h.
6251 #ifdef CONFIG_SOCK_CGROUP_DATA
6253 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6255 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
6256 static bool cgroup_sk_alloc_disabled __read_mostly
;
6258 void cgroup_sk_alloc_disable(void)
6260 if (cgroup_sk_alloc_disabled
)
6262 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6263 cgroup_sk_alloc_disabled
= true;
6268 #define cgroup_sk_alloc_disabled false
6272 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6274 if (cgroup_sk_alloc_disabled
)
6280 struct css_set
*cset
;
6282 cset
= task_css_set(current
);
6283 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6284 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
6293 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6295 cgroup_put(sock_cgroup_ptr(skcd
));
6298 #endif /* CONFIG_SOCK_CGROUP_DATA */
6300 /* cgroup namespaces */
6302 static struct cgroup_namespace
*alloc_cgroup_ns(void)
6304 struct cgroup_namespace
*new_ns
;
6307 new_ns
= kzalloc(sizeof(struct cgroup_namespace
), GFP_KERNEL
);
6309 return ERR_PTR(-ENOMEM
);
6310 ret
= ns_alloc_inum(&new_ns
->ns
);
6313 return ERR_PTR(ret
);
6315 atomic_set(&new_ns
->count
, 1);
6316 new_ns
->ns
.ops
= &cgroupns_operations
;
6320 void free_cgroup_ns(struct cgroup_namespace
*ns
)
6322 put_css_set(ns
->root_cset
);
6323 put_user_ns(ns
->user_ns
);
6324 ns_free_inum(&ns
->ns
);
6327 EXPORT_SYMBOL(free_cgroup_ns
);
6329 struct cgroup_namespace
*copy_cgroup_ns(unsigned long flags
,
6330 struct user_namespace
*user_ns
,
6331 struct cgroup_namespace
*old_ns
)
6333 struct cgroup_namespace
*new_ns
;
6334 struct css_set
*cset
;
6338 if (!(flags
& CLONE_NEWCGROUP
)) {
6339 get_cgroup_ns(old_ns
);
6343 /* Allow only sysadmin to create cgroup namespace. */
6344 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
6345 return ERR_PTR(-EPERM
);
6347 mutex_lock(&cgroup_mutex
);
6348 spin_lock_irq(&css_set_lock
);
6350 cset
= task_css_set(current
);
6353 spin_unlock_irq(&css_set_lock
);
6354 mutex_unlock(&cgroup_mutex
);
6356 new_ns
= alloc_cgroup_ns();
6357 if (IS_ERR(new_ns
)) {
6362 new_ns
->user_ns
= get_user_ns(user_ns
);
6363 new_ns
->root_cset
= cset
;
6368 static inline struct cgroup_namespace
*to_cg_ns(struct ns_common
*ns
)
6370 return container_of(ns
, struct cgroup_namespace
, ns
);
6373 static int cgroupns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
6375 struct cgroup_namespace
*cgroup_ns
= to_cg_ns(ns
);
6377 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN
) ||
6378 !ns_capable(cgroup_ns
->user_ns
, CAP_SYS_ADMIN
))
6381 /* Don't need to do anything if we are attaching to our own cgroupns. */
6382 if (cgroup_ns
== nsproxy
->cgroup_ns
)
6385 get_cgroup_ns(cgroup_ns
);
6386 put_cgroup_ns(nsproxy
->cgroup_ns
);
6387 nsproxy
->cgroup_ns
= cgroup_ns
;
6392 static struct ns_common
*cgroupns_get(struct task_struct
*task
)
6394 struct cgroup_namespace
*ns
= NULL
;
6395 struct nsproxy
*nsproxy
;
6398 nsproxy
= task
->nsproxy
;
6400 ns
= nsproxy
->cgroup_ns
;
6405 return ns
? &ns
->ns
: NULL
;
6408 static void cgroupns_put(struct ns_common
*ns
)
6410 put_cgroup_ns(to_cg_ns(ns
));
6413 const struct proc_ns_operations cgroupns_operations
= {
6415 .type
= CLONE_NEWCGROUP
,
6416 .get
= cgroupns_get
,
6417 .put
= cgroupns_put
,
6418 .install
= cgroupns_install
,
6421 static __init
int cgroup_namespaces_init(void)
6425 subsys_initcall(cgroup_namespaces_init
);
6427 #ifdef CONFIG_CGROUP_DEBUG
6428 static struct cgroup_subsys_state
*
6429 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
6431 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
6434 return ERR_PTR(-ENOMEM
);
6439 static void debug_css_free(struct cgroup_subsys_state
*css
)
6444 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
6447 return cgroup_task_count(css
->cgroup
);
6450 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
6453 return (u64
)(unsigned long)current
->cgroups
;
6456 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
6462 count
= atomic_read(&task_css_set(current
)->refcount
);
6467 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
6469 struct cgrp_cset_link
*link
;
6470 struct css_set
*cset
;
6473 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
6477 spin_lock_irq(&css_set_lock
);
6479 cset
= rcu_dereference(current
->cgroups
);
6480 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
6481 struct cgroup
*c
= link
->cgrp
;
6483 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
6484 seq_printf(seq
, "Root %d group %s\n",
6485 c
->root
->hierarchy_id
, name_buf
);
6488 spin_unlock_irq(&css_set_lock
);
6493 #define MAX_TASKS_SHOWN_PER_CSS 25
6494 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
6496 struct cgroup_subsys_state
*css
= seq_css(seq
);
6497 struct cgrp_cset_link
*link
;
6499 spin_lock_irq(&css_set_lock
);
6500 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
6501 struct css_set
*cset
= link
->cset
;
6502 struct task_struct
*task
;
6505 seq_printf(seq
, "css_set %p\n", cset
);
6507 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
6508 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
6510 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
6513 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
6514 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
6516 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
6520 seq_puts(seq
, " ...\n");
6522 spin_unlock_irq(&css_set_lock
);
6526 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
6528 return (!cgroup_is_populated(css
->cgroup
) &&
6529 !css_has_online_children(&css
->cgroup
->self
));
6532 static struct cftype debug_files
[] = {
6534 .name
= "taskcount",
6535 .read_u64
= debug_taskcount_read
,
6539 .name
= "current_css_set",
6540 .read_u64
= current_css_set_read
,
6544 .name
= "current_css_set_refcount",
6545 .read_u64
= current_css_set_refcount_read
,
6549 .name
= "current_css_set_cg_links",
6550 .seq_show
= current_css_set_cg_links_read
,
6554 .name
= "cgroup_css_links",
6555 .seq_show
= cgroup_css_links_read
,
6559 .name
= "releasable",
6560 .read_u64
= releasable_read
,
6566 struct cgroup_subsys debug_cgrp_subsys
= {
6567 .css_alloc
= debug_css_alloc
,
6568 .css_free
= debug_css_free
,
6569 .legacy_cftypes
= debug_files
,
6571 #endif /* CONFIG_CGROUP_DEBUG */