2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/cgroup.h>
62 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
66 * cgroup_mutex is the master lock. Any modification to cgroup or its
67 * hierarchy must be performed while holding it.
69 * css_set_lock protects task->cgroups pointer, the list of css_set
70 * objects, and the chain of tasks off each css_set.
72 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
73 * cgroup.h can use them for lockdep annotations.
75 DEFINE_MUTEX(cgroup_mutex
);
76 DEFINE_SPINLOCK(css_set_lock
);
78 #ifdef CONFIG_PROVE_RCU
79 EXPORT_SYMBOL_GPL(cgroup_mutex
);
80 EXPORT_SYMBOL_GPL(css_set_lock
);
84 * Protects cgroup_idr and css_idr so that IDs can be released without
85 * grabbing cgroup_mutex.
87 static DEFINE_SPINLOCK(cgroup_idr_lock
);
90 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
91 * against file removal/re-creation across css hiding.
93 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
95 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
97 #define cgroup_assert_mutex_or_rcu_locked() \
98 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
99 !lockdep_is_held(&cgroup_mutex), \
100 "cgroup_mutex or RCU read lock required");
103 * cgroup destruction makes heavy use of work items and there can be a lot
104 * of concurrent destructions. Use a separate workqueue so that cgroup
105 * destruction work items don't end up filling up max_active of system_wq
106 * which may lead to deadlock.
108 static struct workqueue_struct
*cgroup_destroy_wq
;
110 /* generate an array of cgroup subsystem pointers */
111 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
112 struct cgroup_subsys
*cgroup_subsys
[] = {
113 #include <linux/cgroup_subsys.h>
117 /* array of cgroup subsystem names */
118 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
119 static const char *cgroup_subsys_name
[] = {
120 #include <linux/cgroup_subsys.h>
124 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
126 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
127 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
128 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
129 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
130 #include <linux/cgroup_subsys.h>
133 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
134 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
135 #include <linux/cgroup_subsys.h>
139 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
140 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
141 #include <linux/cgroup_subsys.h>
146 * The default hierarchy, reserved for the subsystems that are otherwise
147 * unattached - it never has more than a single cgroup, and all tasks are
148 * part of that cgroup.
150 struct cgroup_root cgrp_dfl_root
;
151 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
154 * The default hierarchy always exists but is hidden until mounted for the
155 * first time. This is for backward compatibility.
157 static bool cgrp_dfl_visible
;
159 /* some controllers are not supported in the default hierarchy */
160 static u16 cgrp_dfl_inhibit_ss_mask
;
162 /* some controllers are implicitly enabled on the default hierarchy */
163 static u16 cgrp_dfl_implicit_ss_mask
;
165 /* some controllers can be threaded on the default hierarchy */
166 static u16 cgrp_dfl_threaded_ss_mask
;
168 /* The list of hierarchy roots */
169 LIST_HEAD(cgroup_roots
);
170 static int cgroup_root_count
;
172 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
173 static DEFINE_IDR(cgroup_hierarchy_idr
);
176 * Assign a monotonically increasing serial number to csses. It guarantees
177 * cgroups with bigger numbers are newer than those with smaller numbers.
178 * Also, as csses are always appended to the parent's ->children list, it
179 * guarantees that sibling csses are always sorted in the ascending serial
180 * number order on the list. Protected by cgroup_mutex.
182 static u64 css_serial_nr_next
= 1;
185 * These bitmasks identify subsystems with specific features to avoid
186 * having to do iterative checks repeatedly.
188 static u16 have_fork_callback __read_mostly
;
189 static u16 have_exit_callback __read_mostly
;
190 static u16 have_release_callback __read_mostly
;
191 static u16 have_canfork_callback __read_mostly
;
193 /* cgroup namespace for init task */
194 struct cgroup_namespace init_cgroup_ns
= {
195 .count
= REFCOUNT_INIT(2),
196 .user_ns
= &init_user_ns
,
197 .ns
.ops
= &cgroupns_operations
,
198 .ns
.inum
= PROC_CGROUP_INIT_INO
,
199 .root_cset
= &init_css_set
,
202 static struct file_system_type cgroup2_fs_type
;
203 static struct cftype cgroup_base_files
[];
205 static int cgroup_apply_control(struct cgroup
*cgrp
);
206 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
);
207 static void css_task_iter_advance(struct css_task_iter
*it
);
208 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
209 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
210 struct cgroup_subsys
*ss
);
211 static void css_release(struct percpu_ref
*ref
);
212 static void kill_css(struct cgroup_subsys_state
*css
);
213 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
214 struct cgroup
*cgrp
, struct cftype cfts
[],
218 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
219 * @ssid: subsys ID of interest
221 * cgroup_subsys_enabled() can only be used with literal subsys names which
222 * is fine for individual subsystems but unsuitable for cgroup core. This
223 * is slower static_key_enabled() based test indexed by @ssid.
225 bool cgroup_ssid_enabled(int ssid
)
227 if (CGROUP_SUBSYS_COUNT
== 0)
230 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
234 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
235 * @cgrp: the cgroup of interest
237 * The default hierarchy is the v2 interface of cgroup and this function
238 * can be used to test whether a cgroup is on the default hierarchy for
239 * cases where a subsystem should behave differnetly depending on the
242 * The set of behaviors which change on the default hierarchy are still
243 * being determined and the mount option is prefixed with __DEVEL__.
245 * List of changed behaviors:
247 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
248 * and "name" are disallowed.
250 * - When mounting an existing superblock, mount options should match.
252 * - Remount is disallowed.
254 * - rename(2) is disallowed.
256 * - "tasks" is removed. Everything should be at process granularity. Use
257 * "cgroup.procs" instead.
259 * - "cgroup.procs" is not sorted. pids will be unique unless they got
260 * recycled inbetween reads.
262 * - "release_agent" and "notify_on_release" are removed. Replacement
263 * notification mechanism will be implemented.
265 * - "cgroup.clone_children" is removed.
267 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
268 * and its descendants contain no task; otherwise, 1. The file also
269 * generates kernfs notification which can be monitored through poll and
270 * [di]notify when the value of the file changes.
272 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
273 * take masks of ancestors with non-empty cpus/mems, instead of being
274 * moved to an ancestor.
276 * - cpuset: a task can be moved into an empty cpuset, and again it takes
277 * masks of ancestors.
279 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
282 * - blkcg: blk-throttle becomes properly hierarchical.
284 * - debug: disallowed on the default hierarchy.
286 bool cgroup_on_dfl(const struct cgroup
*cgrp
)
288 return cgrp
->root
== &cgrp_dfl_root
;
291 /* IDR wrappers which synchronize using cgroup_idr_lock */
292 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
297 idr_preload(gfp_mask
);
298 spin_lock_bh(&cgroup_idr_lock
);
299 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
300 spin_unlock_bh(&cgroup_idr_lock
);
305 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
309 spin_lock_bh(&cgroup_idr_lock
);
310 ret
= idr_replace(idr
, ptr
, id
);
311 spin_unlock_bh(&cgroup_idr_lock
);
315 static void cgroup_idr_remove(struct idr
*idr
, int id
)
317 spin_lock_bh(&cgroup_idr_lock
);
319 spin_unlock_bh(&cgroup_idr_lock
);
322 static bool cgroup_has_tasks(struct cgroup
*cgrp
)
324 return cgrp
->nr_populated_csets
;
327 bool cgroup_is_threaded(struct cgroup
*cgrp
)
329 return cgrp
->dom_cgrp
!= cgrp
;
332 /* can @cgrp host both domain and threaded children? */
333 static bool cgroup_is_mixable(struct cgroup
*cgrp
)
336 * Root isn't under domain level resource control exempting it from
337 * the no-internal-process constraint, so it can serve as a thread
338 * root and a parent of resource domains at the same time.
340 return !cgroup_parent(cgrp
);
343 /* can @cgrp become a thread root? should always be true for a thread root */
344 static bool cgroup_can_be_thread_root(struct cgroup
*cgrp
)
346 /* mixables don't care */
347 if (cgroup_is_mixable(cgrp
))
350 /* domain roots can't be nested under threaded */
351 if (cgroup_is_threaded(cgrp
))
354 /* can only have either domain or threaded children */
355 if (cgrp
->nr_populated_domain_children
)
358 /* and no domain controllers can be enabled */
359 if (cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
365 /* is @cgrp root of a threaded subtree? */
366 bool cgroup_is_thread_root(struct cgroup
*cgrp
)
368 /* thread root should be a domain */
369 if (cgroup_is_threaded(cgrp
))
372 /* a domain w/ threaded children is a thread root */
373 if (cgrp
->nr_threaded_children
)
377 * A domain which has tasks and explicit threaded controllers
378 * enabled is a thread root.
380 if (cgroup_has_tasks(cgrp
) &&
381 (cgrp
->subtree_control
& cgrp_dfl_threaded_ss_mask
))
387 /* a domain which isn't connected to the root w/o brekage can't be used */
388 static bool cgroup_is_valid_domain(struct cgroup
*cgrp
)
390 /* the cgroup itself can be a thread root */
391 if (cgroup_is_threaded(cgrp
))
394 /* but the ancestors can't be unless mixable */
395 while ((cgrp
= cgroup_parent(cgrp
))) {
396 if (!cgroup_is_mixable(cgrp
) && cgroup_is_thread_root(cgrp
))
398 if (cgroup_is_threaded(cgrp
))
405 /* subsystems visibly enabled on a cgroup */
406 static u16
cgroup_control(struct cgroup
*cgrp
)
408 struct cgroup
*parent
= cgroup_parent(cgrp
);
409 u16 root_ss_mask
= cgrp
->root
->subsys_mask
;
412 u16 ss_mask
= parent
->subtree_control
;
414 /* threaded cgroups can only have threaded controllers */
415 if (cgroup_is_threaded(cgrp
))
416 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
420 if (cgroup_on_dfl(cgrp
))
421 root_ss_mask
&= ~(cgrp_dfl_inhibit_ss_mask
|
422 cgrp_dfl_implicit_ss_mask
);
426 /* subsystems enabled on a cgroup */
427 static u16
cgroup_ss_mask(struct cgroup
*cgrp
)
429 struct cgroup
*parent
= cgroup_parent(cgrp
);
432 u16 ss_mask
= parent
->subtree_ss_mask
;
434 /* threaded cgroups can only have threaded controllers */
435 if (cgroup_is_threaded(cgrp
))
436 ss_mask
&= cgrp_dfl_threaded_ss_mask
;
440 return cgrp
->root
->subsys_mask
;
444 * cgroup_css - obtain a cgroup's css for the specified subsystem
445 * @cgrp: the cgroup of interest
446 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
448 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
449 * function must be called either under cgroup_mutex or rcu_read_lock() and
450 * the caller is responsible for pinning the returned css if it wants to
451 * keep accessing it outside the said locks. This function may return
452 * %NULL if @cgrp doesn't have @subsys_id enabled.
454 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
455 struct cgroup_subsys
*ss
)
458 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
459 lockdep_is_held(&cgroup_mutex
));
465 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
466 * @cgrp: the cgroup of interest
467 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
469 * Similar to cgroup_css() but returns the effective css, which is defined
470 * as the matching css of the nearest ancestor including self which has @ss
471 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
472 * function is guaranteed to return non-NULL css.
474 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
475 struct cgroup_subsys
*ss
)
477 lockdep_assert_held(&cgroup_mutex
);
483 * This function is used while updating css associations and thus
484 * can't test the csses directly. Test ss_mask.
486 while (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
))) {
487 cgrp
= cgroup_parent(cgrp
);
492 return cgroup_css(cgrp
, ss
);
496 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
497 * @cgrp: the cgroup of interest
498 * @ss: the subsystem of interest
500 * Find and get the effective css of @cgrp for @ss. The effective css is
501 * defined as the matching css of the nearest ancestor including self which
502 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
503 * the root css is returned, so this function always returns a valid css.
504 * The returned css must be put using css_put().
506 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
507 struct cgroup_subsys
*ss
)
509 struct cgroup_subsys_state
*css
;
514 css
= cgroup_css(cgrp
, ss
);
516 if (css
&& css_tryget_online(css
))
518 cgrp
= cgroup_parent(cgrp
);
521 css
= init_css_set
.subsys
[ss
->id
];
528 static void cgroup_get_live(struct cgroup
*cgrp
)
530 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
531 css_get(&cgrp
->self
);
534 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
536 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
537 struct cftype
*cft
= of_cft(of
);
540 * This is open and unprotected implementation of cgroup_css().
541 * seq_css() is only called from a kernfs file operation which has
542 * an active reference on the file. Because all the subsystem
543 * files are drained before a css is disassociated with a cgroup,
544 * the matching css from the cgroup's subsys table is guaranteed to
545 * be and stay valid until the enclosing operation is complete.
548 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
552 EXPORT_SYMBOL_GPL(of_css
);
555 * for_each_css - iterate all css's of a cgroup
556 * @css: the iteration cursor
557 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
558 * @cgrp: the target cgroup to iterate css's of
560 * Should be called under cgroup_[tree_]mutex.
562 #define for_each_css(css, ssid, cgrp) \
563 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
564 if (!((css) = rcu_dereference_check( \
565 (cgrp)->subsys[(ssid)], \
566 lockdep_is_held(&cgroup_mutex)))) { } \
570 * for_each_e_css - iterate all effective css's of a cgroup
571 * @css: the iteration cursor
572 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
573 * @cgrp: the target cgroup to iterate css's of
575 * Should be called under cgroup_[tree_]mutex.
577 #define for_each_e_css(css, ssid, cgrp) \
578 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
579 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
584 * do_each_subsys_mask - filter for_each_subsys with a bitmask
585 * @ss: the iteration cursor
586 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
587 * @ss_mask: the bitmask
589 * The block will only run for cases where the ssid-th bit (1 << ssid) of
592 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
593 unsigned long __ss_mask = (ss_mask); \
594 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
598 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
599 (ss) = cgroup_subsys[ssid]; \
602 #define while_each_subsys_mask() \
607 /* iterate over child cgrps, lock should be held throughout iteration */
608 #define cgroup_for_each_live_child(child, cgrp) \
609 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
610 if (({ lockdep_assert_held(&cgroup_mutex); \
611 cgroup_is_dead(child); })) \
615 /* walk live descendants in preorder */
616 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
617 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
618 if (({ lockdep_assert_held(&cgroup_mutex); \
619 (dsct) = (d_css)->cgroup; \
620 cgroup_is_dead(dsct); })) \
624 /* walk live descendants in postorder */
625 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
626 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
627 if (({ lockdep_assert_held(&cgroup_mutex); \
628 (dsct) = (d_css)->cgroup; \
629 cgroup_is_dead(dsct); })) \
634 * The default css_set - used by init and its children prior to any
635 * hierarchies being mounted. It contains a pointer to the root state
636 * for each subsystem. Also used to anchor the list of css_sets. Not
637 * reference-counted, to improve performance when child cgroups
638 * haven't been created.
640 struct css_set init_css_set
= {
641 .refcount
= REFCOUNT_INIT(1),
642 .dom_cset
= &init_css_set
,
643 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
644 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
645 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
646 .threaded_csets
= LIST_HEAD_INIT(init_css_set
.threaded_csets
),
647 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
648 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
649 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
652 static int css_set_count
= 1; /* 1 for init_css_set */
654 static bool css_set_threaded(struct css_set
*cset
)
656 return cset
->dom_cset
!= cset
;
660 * css_set_populated - does a css_set contain any tasks?
661 * @cset: target css_set
663 * css_set_populated() should be the same as !!cset->nr_tasks at steady
664 * state. However, css_set_populated() can be called while a task is being
665 * added to or removed from the linked list before the nr_tasks is
666 * properly updated. Hence, we can't just look at ->nr_tasks here.
668 static bool css_set_populated(struct css_set
*cset
)
670 lockdep_assert_held(&css_set_lock
);
672 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
676 * cgroup_update_populated - update the populated count of a cgroup
677 * @cgrp: the target cgroup
678 * @populated: inc or dec populated count
680 * One of the css_sets associated with @cgrp is either getting its first
681 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
682 * count is propagated towards root so that a given cgroup's
683 * nr_populated_children is zero iff none of its descendants contain any
686 * @cgrp's interface file "cgroup.populated" is zero if both
687 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
688 * 1 otherwise. When the sum changes from or to zero, userland is notified
689 * that the content of the interface file has changed. This can be used to
690 * detect when @cgrp and its descendants become populated or empty.
692 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
694 struct cgroup
*child
= NULL
;
695 int adj
= populated
? 1 : -1;
697 lockdep_assert_held(&css_set_lock
);
700 bool was_populated
= cgroup_is_populated(cgrp
);
703 cgrp
->nr_populated_csets
+= adj
;
705 if (cgroup_is_threaded(child
))
706 cgrp
->nr_populated_threaded_children
+= adj
;
708 cgrp
->nr_populated_domain_children
+= adj
;
711 if (was_populated
== cgroup_is_populated(cgrp
))
714 cgroup1_check_for_release(cgrp
);
715 cgroup_file_notify(&cgrp
->events_file
);
718 cgrp
= cgroup_parent(cgrp
);
723 * css_set_update_populated - update populated state of a css_set
724 * @cset: target css_set
725 * @populated: whether @cset is populated or depopulated
727 * @cset is either getting the first task or losing the last. Update the
728 * populated counters of all associated cgroups accordingly.
730 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
732 struct cgrp_cset_link
*link
;
734 lockdep_assert_held(&css_set_lock
);
736 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
737 cgroup_update_populated(link
->cgrp
, populated
);
741 * css_set_move_task - move a task from one css_set to another
742 * @task: task being moved
743 * @from_cset: css_set @task currently belongs to (may be NULL)
744 * @to_cset: new css_set @task is being moved to (may be NULL)
745 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
747 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
748 * css_set, @from_cset can be NULL. If @task is being disassociated
749 * instead of moved, @to_cset can be NULL.
751 * This function automatically handles populated counter updates and
752 * css_task_iter adjustments but the caller is responsible for managing
753 * @from_cset and @to_cset's reference counts.
755 static void css_set_move_task(struct task_struct
*task
,
756 struct css_set
*from_cset
, struct css_set
*to_cset
,
759 lockdep_assert_held(&css_set_lock
);
761 if (to_cset
&& !css_set_populated(to_cset
))
762 css_set_update_populated(to_cset
, true);
765 struct css_task_iter
*it
, *pos
;
767 WARN_ON_ONCE(list_empty(&task
->cg_list
));
770 * @task is leaving, advance task iterators which are
771 * pointing to it so that they can resume at the next
772 * position. Advancing an iterator might remove it from
773 * the list, use safe walk. See css_task_iter_advance*()
776 list_for_each_entry_safe(it
, pos
, &from_cset
->task_iters
,
778 if (it
->task_pos
== &task
->cg_list
)
779 css_task_iter_advance(it
);
781 list_del_init(&task
->cg_list
);
782 if (!css_set_populated(from_cset
))
783 css_set_update_populated(from_cset
, false);
785 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
790 * We are synchronized through cgroup_threadgroup_rwsem
791 * against PF_EXITING setting such that we can't race
792 * against cgroup_exit() changing the css_set to
793 * init_css_set and dropping the old one.
795 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
797 rcu_assign_pointer(task
->cgroups
, to_cset
);
798 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
804 * hash table for cgroup groups. This improves the performance to find
805 * an existing css_set. This hash doesn't (currently) take into
806 * account cgroups in empty hierarchies.
808 #define CSS_SET_HASH_BITS 7
809 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
811 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
813 unsigned long key
= 0UL;
814 struct cgroup_subsys
*ss
;
817 for_each_subsys(ss
, i
)
818 key
+= (unsigned long)css
[i
];
819 key
= (key
>> 16) ^ key
;
824 void put_css_set_locked(struct css_set
*cset
)
826 struct cgrp_cset_link
*link
, *tmp_link
;
827 struct cgroup_subsys
*ss
;
830 lockdep_assert_held(&css_set_lock
);
832 if (!refcount_dec_and_test(&cset
->refcount
))
835 WARN_ON_ONCE(!list_empty(&cset
->threaded_csets
));
837 /* This css_set is dead. unlink it and release cgroup and css refs */
838 for_each_subsys(ss
, ssid
) {
839 list_del(&cset
->e_cset_node
[ssid
]);
840 css_put(cset
->subsys
[ssid
]);
842 hash_del(&cset
->hlist
);
845 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
846 list_del(&link
->cset_link
);
847 list_del(&link
->cgrp_link
);
848 if (cgroup_parent(link
->cgrp
))
849 cgroup_put(link
->cgrp
);
853 if (css_set_threaded(cset
)) {
854 list_del(&cset
->threaded_csets_node
);
855 put_css_set_locked(cset
->dom_cset
);
858 kfree_rcu(cset
, rcu_head
);
862 * compare_css_sets - helper function for find_existing_css_set().
863 * @cset: candidate css_set being tested
864 * @old_cset: existing css_set for a task
865 * @new_cgrp: cgroup that's being entered by the task
866 * @template: desired set of css pointers in css_set (pre-calculated)
868 * Returns true if "cset" matches "old_cset" except for the hierarchy
869 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
871 static bool compare_css_sets(struct css_set
*cset
,
872 struct css_set
*old_cset
,
873 struct cgroup
*new_cgrp
,
874 struct cgroup_subsys_state
*template[])
876 struct cgroup
*new_dfl_cgrp
;
877 struct list_head
*l1
, *l2
;
880 * On the default hierarchy, there can be csets which are
881 * associated with the same set of cgroups but different csses.
882 * Let's first ensure that csses match.
884 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
888 /* @cset's domain should match the default cgroup's */
889 if (cgroup_on_dfl(new_cgrp
))
890 new_dfl_cgrp
= new_cgrp
;
892 new_dfl_cgrp
= old_cset
->dfl_cgrp
;
894 if (new_dfl_cgrp
->dom_cgrp
!= cset
->dom_cset
->dfl_cgrp
)
898 * Compare cgroup pointers in order to distinguish between
899 * different cgroups in hierarchies. As different cgroups may
900 * share the same effective css, this comparison is always
903 l1
= &cset
->cgrp_links
;
904 l2
= &old_cset
->cgrp_links
;
906 struct cgrp_cset_link
*link1
, *link2
;
907 struct cgroup
*cgrp1
, *cgrp2
;
911 /* See if we reached the end - both lists are equal length. */
912 if (l1
== &cset
->cgrp_links
) {
913 BUG_ON(l2
!= &old_cset
->cgrp_links
);
916 BUG_ON(l2
== &old_cset
->cgrp_links
);
918 /* Locate the cgroups associated with these links. */
919 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
920 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
923 /* Hierarchies should be linked in the same order. */
924 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
927 * If this hierarchy is the hierarchy of the cgroup
928 * that's changing, then we need to check that this
929 * css_set points to the new cgroup; if it's any other
930 * hierarchy, then this css_set should point to the
931 * same cgroup as the old css_set.
933 if (cgrp1
->root
== new_cgrp
->root
) {
934 if (cgrp1
!= new_cgrp
)
945 * find_existing_css_set - init css array and find the matching css_set
946 * @old_cset: the css_set that we're using before the cgroup transition
947 * @cgrp: the cgroup that we're moving into
948 * @template: out param for the new set of csses, should be clear on entry
950 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
952 struct cgroup_subsys_state
*template[])
954 struct cgroup_root
*root
= cgrp
->root
;
955 struct cgroup_subsys
*ss
;
956 struct css_set
*cset
;
961 * Build the set of subsystem state objects that we want to see in the
962 * new css_set. while subsystems can change globally, the entries here
963 * won't change, so no need for locking.
965 for_each_subsys(ss
, i
) {
966 if (root
->subsys_mask
& (1UL << i
)) {
968 * @ss is in this hierarchy, so we want the
969 * effective css from @cgrp.
971 template[i
] = cgroup_e_css(cgrp
, ss
);
974 * @ss is not in this hierarchy, so we don't want
977 template[i
] = old_cset
->subsys
[i
];
981 key
= css_set_hash(template);
982 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
983 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
986 /* This css_set matches what we need */
990 /* No existing cgroup group matched */
994 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
996 struct cgrp_cset_link
*link
, *tmp_link
;
998 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
999 list_del(&link
->cset_link
);
1005 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1006 * @count: the number of links to allocate
1007 * @tmp_links: list_head the allocated links are put on
1009 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1010 * through ->cset_link. Returns 0 on success or -errno.
1012 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
1014 struct cgrp_cset_link
*link
;
1017 INIT_LIST_HEAD(tmp_links
);
1019 for (i
= 0; i
< count
; i
++) {
1020 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
1022 free_cgrp_cset_links(tmp_links
);
1025 list_add(&link
->cset_link
, tmp_links
);
1031 * link_css_set - a helper function to link a css_set to a cgroup
1032 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1033 * @cset: the css_set to be linked
1034 * @cgrp: the destination cgroup
1036 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
1037 struct cgroup
*cgrp
)
1039 struct cgrp_cset_link
*link
;
1041 BUG_ON(list_empty(tmp_links
));
1043 if (cgroup_on_dfl(cgrp
))
1044 cset
->dfl_cgrp
= cgrp
;
1046 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
1051 * Always add links to the tail of the lists so that the lists are
1052 * in choronological order.
1054 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
1055 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
1057 if (cgroup_parent(cgrp
))
1058 cgroup_get_live(cgrp
);
1062 * find_css_set - return a new css_set with one cgroup updated
1063 * @old_cset: the baseline css_set
1064 * @cgrp: the cgroup to be updated
1066 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1067 * substituted into the appropriate hierarchy.
1069 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1070 struct cgroup
*cgrp
)
1072 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1073 struct css_set
*cset
;
1074 struct list_head tmp_links
;
1075 struct cgrp_cset_link
*link
;
1076 struct cgroup_subsys
*ss
;
1080 lockdep_assert_held(&cgroup_mutex
);
1082 /* First see if we already have a cgroup group that matches
1083 * the desired set */
1084 spin_lock_irq(&css_set_lock
);
1085 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1088 spin_unlock_irq(&css_set_lock
);
1093 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1097 /* Allocate all the cgrp_cset_link objects that we'll need */
1098 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1103 refcount_set(&cset
->refcount
, 1);
1104 cset
->dom_cset
= cset
;
1105 INIT_LIST_HEAD(&cset
->tasks
);
1106 INIT_LIST_HEAD(&cset
->mg_tasks
);
1107 INIT_LIST_HEAD(&cset
->task_iters
);
1108 INIT_LIST_HEAD(&cset
->threaded_csets
);
1109 INIT_HLIST_NODE(&cset
->hlist
);
1110 INIT_LIST_HEAD(&cset
->cgrp_links
);
1111 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1112 INIT_LIST_HEAD(&cset
->mg_node
);
1114 /* Copy the set of subsystem state objects generated in
1115 * find_existing_css_set() */
1116 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1118 spin_lock_irq(&css_set_lock
);
1119 /* Add reference counts and links from the new css_set. */
1120 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1121 struct cgroup
*c
= link
->cgrp
;
1123 if (c
->root
== cgrp
->root
)
1125 link_css_set(&tmp_links
, cset
, c
);
1128 BUG_ON(!list_empty(&tmp_links
));
1132 /* Add @cset to the hash table */
1133 key
= css_set_hash(cset
->subsys
);
1134 hash_add(css_set_table
, &cset
->hlist
, key
);
1136 for_each_subsys(ss
, ssid
) {
1137 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1139 list_add_tail(&cset
->e_cset_node
[ssid
],
1140 &css
->cgroup
->e_csets
[ssid
]);
1144 spin_unlock_irq(&css_set_lock
);
1147 * If @cset should be threaded, look up the matching dom_cset and
1148 * link them up. We first fully initialize @cset then look for the
1149 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1150 * to stay empty until we return.
1152 if (cgroup_is_threaded(cset
->dfl_cgrp
)) {
1153 struct css_set
*dcset
;
1155 dcset
= find_css_set(cset
, cset
->dfl_cgrp
->dom_cgrp
);
1161 spin_lock_irq(&css_set_lock
);
1162 cset
->dom_cset
= dcset
;
1163 list_add_tail(&cset
->threaded_csets_node
,
1164 &dcset
->threaded_csets
);
1165 spin_unlock_irq(&css_set_lock
);
1171 struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1173 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1175 return root_cgrp
->root
;
1178 static int cgroup_init_root_id(struct cgroup_root
*root
)
1182 lockdep_assert_held(&cgroup_mutex
);
1184 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1188 root
->hierarchy_id
= id
;
1192 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1194 lockdep_assert_held(&cgroup_mutex
);
1196 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1199 void cgroup_free_root(struct cgroup_root
*root
)
1202 idr_destroy(&root
->cgroup_idr
);
1207 static void cgroup_destroy_root(struct cgroup_root
*root
)
1209 struct cgroup
*cgrp
= &root
->cgrp
;
1210 struct cgrp_cset_link
*link
, *tmp_link
;
1212 trace_cgroup_destroy_root(root
);
1214 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1216 BUG_ON(atomic_read(&root
->nr_cgrps
));
1217 BUG_ON(!list_empty(&cgrp
->self
.children
));
1219 /* Rebind all subsystems back to the default hierarchy */
1220 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
));
1223 * Release all the links from cset_links to this hierarchy's
1226 spin_lock_irq(&css_set_lock
);
1228 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1229 list_del(&link
->cset_link
);
1230 list_del(&link
->cgrp_link
);
1234 spin_unlock_irq(&css_set_lock
);
1236 if (!list_empty(&root
->root_list
)) {
1237 list_del(&root
->root_list
);
1238 cgroup_root_count
--;
1241 cgroup_exit_root_id(root
);
1243 mutex_unlock(&cgroup_mutex
);
1245 kernfs_destroy_root(root
->kf_root
);
1246 cgroup_free_root(root
);
1250 * look up cgroup associated with current task's cgroup namespace on the
1251 * specified hierarchy
1253 static struct cgroup
*
1254 current_cgns_cgroup_from_root(struct cgroup_root
*root
)
1256 struct cgroup
*res
= NULL
;
1257 struct css_set
*cset
;
1259 lockdep_assert_held(&css_set_lock
);
1263 cset
= current
->nsproxy
->cgroup_ns
->root_cset
;
1264 if (cset
== &init_css_set
) {
1267 struct cgrp_cset_link
*link
;
1269 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1270 struct cgroup
*c
= link
->cgrp
;
1272 if (c
->root
== root
) {
1284 /* look up cgroup associated with given css_set on the specified hierarchy */
1285 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1286 struct cgroup_root
*root
)
1288 struct cgroup
*res
= NULL
;
1290 lockdep_assert_held(&cgroup_mutex
);
1291 lockdep_assert_held(&css_set_lock
);
1293 if (cset
== &init_css_set
) {
1295 } else if (root
== &cgrp_dfl_root
) {
1296 res
= cset
->dfl_cgrp
;
1298 struct cgrp_cset_link
*link
;
1300 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1301 struct cgroup
*c
= link
->cgrp
;
1303 if (c
->root
== root
) {
1315 * Return the cgroup for "task" from the given hierarchy. Must be
1316 * called with cgroup_mutex and css_set_lock held.
1318 struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1319 struct cgroup_root
*root
)
1322 * No need to lock the task - since we hold cgroup_mutex the
1323 * task can't change groups, so the only thing that can happen
1324 * is that it exits and its css is set back to init_css_set.
1326 return cset_cgroup_from_root(task_css_set(task
), root
);
1330 * A task must hold cgroup_mutex to modify cgroups.
1332 * Any task can increment and decrement the count field without lock.
1333 * So in general, code holding cgroup_mutex can't rely on the count
1334 * field not changing. However, if the count goes to zero, then only
1335 * cgroup_attach_task() can increment it again. Because a count of zero
1336 * means that no tasks are currently attached, therefore there is no
1337 * way a task attached to that cgroup can fork (the other way to
1338 * increment the count). So code holding cgroup_mutex can safely
1339 * assume that if the count is zero, it will stay zero. Similarly, if
1340 * a task holds cgroup_mutex on a cgroup with zero count, it
1341 * knows that the cgroup won't be removed, as cgroup_rmdir()
1344 * A cgroup can only be deleted if both its 'count' of using tasks
1345 * is zero, and its list of 'children' cgroups is empty. Since all
1346 * tasks in the system use _some_ cgroup, and since there is always at
1347 * least one task in the system (init, pid == 1), therefore, root cgroup
1348 * always has either children cgroups and/or using tasks. So we don't
1349 * need a special hack to ensure that root cgroup cannot be deleted.
1351 * P.S. One more locking exception. RCU is used to guard the
1352 * update of a tasks cgroup pointer by cgroup_attach_task()
1355 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1357 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1360 struct cgroup_subsys
*ss
= cft
->ss
;
1362 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1363 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1364 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1365 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1368 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1373 * cgroup_file_mode - deduce file mode of a control file
1374 * @cft: the control file in question
1376 * S_IRUGO for read, S_IWUSR for write.
1378 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1382 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1385 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1386 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1396 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1397 * @subtree_control: the new subtree_control mask to consider
1398 * @this_ss_mask: available subsystems
1400 * On the default hierarchy, a subsystem may request other subsystems to be
1401 * enabled together through its ->depends_on mask. In such cases, more
1402 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1404 * This function calculates which subsystems need to be enabled if
1405 * @subtree_control is to be applied while restricted to @this_ss_mask.
1407 static u16
cgroup_calc_subtree_ss_mask(u16 subtree_control
, u16 this_ss_mask
)
1409 u16 cur_ss_mask
= subtree_control
;
1410 struct cgroup_subsys
*ss
;
1413 lockdep_assert_held(&cgroup_mutex
);
1415 cur_ss_mask
|= cgrp_dfl_implicit_ss_mask
;
1418 u16 new_ss_mask
= cur_ss_mask
;
1420 do_each_subsys_mask(ss
, ssid
, cur_ss_mask
) {
1421 new_ss_mask
|= ss
->depends_on
;
1422 } while_each_subsys_mask();
1425 * Mask out subsystems which aren't available. This can
1426 * happen only if some depended-upon subsystems were bound
1427 * to non-default hierarchies.
1429 new_ss_mask
&= this_ss_mask
;
1431 if (new_ss_mask
== cur_ss_mask
)
1433 cur_ss_mask
= new_ss_mask
;
1440 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1441 * @kn: the kernfs_node being serviced
1443 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1444 * the method finishes if locking succeeded. Note that once this function
1445 * returns the cgroup returned by cgroup_kn_lock_live() may become
1446 * inaccessible any time. If the caller intends to continue to access the
1447 * cgroup, it should pin it before invoking this function.
1449 void cgroup_kn_unlock(struct kernfs_node
*kn
)
1451 struct cgroup
*cgrp
;
1453 if (kernfs_type(kn
) == KERNFS_DIR
)
1456 cgrp
= kn
->parent
->priv
;
1458 mutex_unlock(&cgroup_mutex
);
1460 kernfs_unbreak_active_protection(kn
);
1465 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1466 * @kn: the kernfs_node being serviced
1467 * @drain_offline: perform offline draining on the cgroup
1469 * This helper is to be used by a cgroup kernfs method currently servicing
1470 * @kn. It breaks the active protection, performs cgroup locking and
1471 * verifies that the associated cgroup is alive. Returns the cgroup if
1472 * alive; otherwise, %NULL. A successful return should be undone by a
1473 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1474 * cgroup is drained of offlining csses before return.
1476 * Any cgroup kernfs method implementation which requires locking the
1477 * associated cgroup should use this helper. It avoids nesting cgroup
1478 * locking under kernfs active protection and allows all kernfs operations
1479 * including self-removal.
1481 struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
, bool drain_offline
)
1483 struct cgroup
*cgrp
;
1485 if (kernfs_type(kn
) == KERNFS_DIR
)
1488 cgrp
= kn
->parent
->priv
;
1491 * We're gonna grab cgroup_mutex which nests outside kernfs
1492 * active_ref. cgroup liveliness check alone provides enough
1493 * protection against removal. Ensure @cgrp stays accessible and
1494 * break the active_ref protection.
1496 if (!cgroup_tryget(cgrp
))
1498 kernfs_break_active_protection(kn
);
1501 cgroup_lock_and_drain_offline(cgrp
);
1503 mutex_lock(&cgroup_mutex
);
1505 if (!cgroup_is_dead(cgrp
))
1508 cgroup_kn_unlock(kn
);
1512 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1514 char name
[CGROUP_FILE_NAME_MAX
];
1516 lockdep_assert_held(&cgroup_mutex
);
1518 if (cft
->file_offset
) {
1519 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1520 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1522 spin_lock_irq(&cgroup_file_kn_lock
);
1524 spin_unlock_irq(&cgroup_file_kn_lock
);
1527 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1531 * css_clear_dir - remove subsys files in a cgroup directory
1534 static void css_clear_dir(struct cgroup_subsys_state
*css
)
1536 struct cgroup
*cgrp
= css
->cgroup
;
1537 struct cftype
*cfts
;
1539 if (!(css
->flags
& CSS_VISIBLE
))
1542 css
->flags
&= ~CSS_VISIBLE
;
1544 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1545 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1549 * css_populate_dir - create subsys files in a cgroup directory
1552 * On failure, no file is added.
1554 static int css_populate_dir(struct cgroup_subsys_state
*css
)
1556 struct cgroup
*cgrp
= css
->cgroup
;
1557 struct cftype
*cfts
, *failed_cfts
;
1560 if ((css
->flags
& CSS_VISIBLE
) || !cgrp
->kn
)
1564 if (cgroup_on_dfl(cgrp
))
1565 cfts
= cgroup_base_files
;
1567 cfts
= cgroup1_base_files
;
1569 return cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1572 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1573 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1580 css
->flags
|= CSS_VISIBLE
;
1584 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1585 if (cfts
== failed_cfts
)
1587 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1592 int rebind_subsystems(struct cgroup_root
*dst_root
, u16 ss_mask
)
1594 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1595 struct cgroup_subsys
*ss
;
1598 lockdep_assert_held(&cgroup_mutex
);
1600 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1602 * If @ss has non-root csses attached to it, can't move.
1603 * If @ss is an implicit controller, it is exempt from this
1604 * rule and can be stolen.
1606 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)) &&
1607 !ss
->implicit_on_dfl
)
1610 /* can't move between two non-dummy roots either */
1611 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1613 } while_each_subsys_mask();
1615 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
1616 struct cgroup_root
*src_root
= ss
->root
;
1617 struct cgroup
*scgrp
= &src_root
->cgrp
;
1618 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1619 struct css_set
*cset
;
1621 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1623 /* disable from the source */
1624 src_root
->subsys_mask
&= ~(1 << ssid
);
1625 WARN_ON(cgroup_apply_control(scgrp
));
1626 cgroup_finalize_control(scgrp
, 0);
1629 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1630 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1631 ss
->root
= dst_root
;
1632 css
->cgroup
= dcgrp
;
1634 spin_lock_irq(&css_set_lock
);
1635 hash_for_each(css_set_table
, i
, cset
, hlist
)
1636 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1637 &dcgrp
->e_csets
[ss
->id
]);
1638 spin_unlock_irq(&css_set_lock
);
1640 /* default hierarchy doesn't enable controllers by default */
1641 dst_root
->subsys_mask
|= 1 << ssid
;
1642 if (dst_root
== &cgrp_dfl_root
) {
1643 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1645 dcgrp
->subtree_control
|= 1 << ssid
;
1646 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1649 ret
= cgroup_apply_control(dcgrp
);
1651 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1656 } while_each_subsys_mask();
1658 kernfs_activate(dcgrp
->kn
);
1662 int cgroup_show_path(struct seq_file
*sf
, struct kernfs_node
*kf_node
,
1663 struct kernfs_root
*kf_root
)
1667 struct cgroup_root
*kf_cgroot
= cgroup_root_from_kf(kf_root
);
1668 struct cgroup
*ns_cgroup
;
1670 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
1674 spin_lock_irq(&css_set_lock
);
1675 ns_cgroup
= current_cgns_cgroup_from_root(kf_cgroot
);
1676 len
= kernfs_path_from_node(kf_node
, ns_cgroup
->kn
, buf
, PATH_MAX
);
1677 spin_unlock_irq(&css_set_lock
);
1679 if (len
>= PATH_MAX
)
1682 seq_escape(sf
, buf
, " \t\n\\");
1689 static int parse_cgroup_root_flags(char *data
, unsigned int *root_flags
)
1695 if (!data
|| *data
== '\0')
1698 while ((token
= strsep(&data
, ",")) != NULL
) {
1699 if (!strcmp(token
, "nsdelegate")) {
1700 *root_flags
|= CGRP_ROOT_NS_DELEGATE
;
1704 pr_err("cgroup2: unknown option \"%s\"\n", token
);
1711 static void apply_cgroup_root_flags(unsigned int root_flags
)
1713 if (current
->nsproxy
->cgroup_ns
== &init_cgroup_ns
) {
1714 if (root_flags
& CGRP_ROOT_NS_DELEGATE
)
1715 cgrp_dfl_root
.flags
|= CGRP_ROOT_NS_DELEGATE
;
1717 cgrp_dfl_root
.flags
&= ~CGRP_ROOT_NS_DELEGATE
;
1721 static int cgroup_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
1723 if (cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
)
1724 seq_puts(seq
, ",nsdelegate");
1728 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1730 unsigned int root_flags
;
1733 ret
= parse_cgroup_root_flags(data
, &root_flags
);
1737 apply_cgroup_root_flags(root_flags
);
1742 * To reduce the fork() overhead for systems that are not actually using
1743 * their cgroups capability, we don't maintain the lists running through
1744 * each css_set to its tasks until we see the list actually used - in other
1745 * words after the first mount.
1747 static bool use_task_css_set_links __read_mostly
;
1749 static void cgroup_enable_task_cg_lists(void)
1751 struct task_struct
*p
, *g
;
1753 spin_lock_irq(&css_set_lock
);
1755 if (use_task_css_set_links
)
1758 use_task_css_set_links
= true;
1761 * We need tasklist_lock because RCU is not safe against
1762 * while_each_thread(). Besides, a forking task that has passed
1763 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1764 * is not guaranteed to have its child immediately visible in the
1765 * tasklist if we walk through it with RCU.
1767 read_lock(&tasklist_lock
);
1768 do_each_thread(g
, p
) {
1769 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1770 task_css_set(p
) != &init_css_set
);
1773 * We should check if the process is exiting, otherwise
1774 * it will race with cgroup_exit() in that the list
1775 * entry won't be deleted though the process has exited.
1776 * Do it while holding siglock so that we don't end up
1777 * racing against cgroup_exit().
1779 * Interrupts were already disabled while acquiring
1780 * the css_set_lock, so we do not need to disable it
1781 * again when acquiring the sighand->siglock here.
1783 spin_lock(&p
->sighand
->siglock
);
1784 if (!(p
->flags
& PF_EXITING
)) {
1785 struct css_set
*cset
= task_css_set(p
);
1787 if (!css_set_populated(cset
))
1788 css_set_update_populated(cset
, true);
1789 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1793 spin_unlock(&p
->sighand
->siglock
);
1794 } while_each_thread(g
, p
);
1795 read_unlock(&tasklist_lock
);
1797 spin_unlock_irq(&css_set_lock
);
1800 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1802 struct cgroup_subsys
*ss
;
1805 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1806 INIT_LIST_HEAD(&cgrp
->self
.children
);
1807 INIT_LIST_HEAD(&cgrp
->cset_links
);
1808 INIT_LIST_HEAD(&cgrp
->pidlists
);
1809 mutex_init(&cgrp
->pidlist_mutex
);
1810 cgrp
->self
.cgroup
= cgrp
;
1811 cgrp
->self
.flags
|= CSS_ONLINE
;
1812 cgrp
->dom_cgrp
= cgrp
;
1813 cgrp
->max_descendants
= INT_MAX
;
1814 cgrp
->max_depth
= INT_MAX
;
1816 for_each_subsys(ss
, ssid
)
1817 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1819 init_waitqueue_head(&cgrp
->offline_waitq
);
1820 INIT_WORK(&cgrp
->release_agent_work
, cgroup1_release_agent
);
1823 void init_cgroup_root(struct cgroup_root
*root
, struct cgroup_sb_opts
*opts
)
1825 struct cgroup
*cgrp
= &root
->cgrp
;
1827 INIT_LIST_HEAD(&root
->root_list
);
1828 atomic_set(&root
->nr_cgrps
, 1);
1830 init_cgroup_housekeeping(cgrp
);
1831 idr_init(&root
->cgroup_idr
);
1833 root
->flags
= opts
->flags
;
1834 if (opts
->release_agent
)
1835 strcpy(root
->release_agent_path
, opts
->release_agent
);
1837 strcpy(root
->name
, opts
->name
);
1838 if (opts
->cpuset_clone_children
)
1839 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1842 int cgroup_setup_root(struct cgroup_root
*root
, u16 ss_mask
, int ref_flags
)
1844 LIST_HEAD(tmp_links
);
1845 struct cgroup
*root_cgrp
= &root
->cgrp
;
1846 struct kernfs_syscall_ops
*kf_sops
;
1847 struct css_set
*cset
;
1850 lockdep_assert_held(&cgroup_mutex
);
1852 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1855 root_cgrp
->id
= ret
;
1856 root_cgrp
->ancestor_ids
[0] = ret
;
1858 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
,
1859 ref_flags
, GFP_KERNEL
);
1864 * We're accessing css_set_count without locking css_set_lock here,
1865 * but that's OK - it can only be increased by someone holding
1866 * cgroup_lock, and that's us. Later rebinding may disable
1867 * controllers on the default hierarchy and thus create new csets,
1868 * which can't be more than the existing ones. Allocate 2x.
1870 ret
= allocate_cgrp_cset_links(2 * css_set_count
, &tmp_links
);
1874 ret
= cgroup_init_root_id(root
);
1878 kf_sops
= root
== &cgrp_dfl_root
?
1879 &cgroup_kf_syscall_ops
: &cgroup1_kf_syscall_ops
;
1881 root
->kf_root
= kernfs_create_root(kf_sops
,
1882 KERNFS_ROOT_CREATE_DEACTIVATED
|
1883 KERNFS_ROOT_SUPPORT_EXPORTOP
,
1885 if (IS_ERR(root
->kf_root
)) {
1886 ret
= PTR_ERR(root
->kf_root
);
1889 root_cgrp
->kn
= root
->kf_root
->kn
;
1891 ret
= css_populate_dir(&root_cgrp
->self
);
1895 ret
= rebind_subsystems(root
, ss_mask
);
1899 trace_cgroup_setup_root(root
);
1902 * There must be no failure case after here, since rebinding takes
1903 * care of subsystems' refcounts, which are explicitly dropped in
1904 * the failure exit path.
1906 list_add(&root
->root_list
, &cgroup_roots
);
1907 cgroup_root_count
++;
1910 * Link the root cgroup in this hierarchy into all the css_set
1913 spin_lock_irq(&css_set_lock
);
1914 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1915 link_css_set(&tmp_links
, cset
, root_cgrp
);
1916 if (css_set_populated(cset
))
1917 cgroup_update_populated(root_cgrp
, true);
1919 spin_unlock_irq(&css_set_lock
);
1921 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1922 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1924 kernfs_activate(root_cgrp
->kn
);
1929 kernfs_destroy_root(root
->kf_root
);
1930 root
->kf_root
= NULL
;
1932 cgroup_exit_root_id(root
);
1934 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1936 free_cgrp_cset_links(&tmp_links
);
1940 struct dentry
*cgroup_do_mount(struct file_system_type
*fs_type
, int flags
,
1941 struct cgroup_root
*root
, unsigned long magic
,
1942 struct cgroup_namespace
*ns
)
1944 struct dentry
*dentry
;
1945 bool new_sb
= false;
1947 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
, magic
, &new_sb
);
1950 * In non-init cgroup namespace, instead of root cgroup's dentry,
1951 * we return the dentry corresponding to the cgroupns->root_cgrp.
1953 if (!IS_ERR(dentry
) && ns
!= &init_cgroup_ns
) {
1954 struct dentry
*nsdentry
;
1955 struct super_block
*sb
= dentry
->d_sb
;
1956 struct cgroup
*cgrp
;
1958 mutex_lock(&cgroup_mutex
);
1959 spin_lock_irq(&css_set_lock
);
1961 cgrp
= cset_cgroup_from_root(ns
->root_cset
, root
);
1963 spin_unlock_irq(&css_set_lock
);
1964 mutex_unlock(&cgroup_mutex
);
1966 nsdentry
= kernfs_node_dentry(cgrp
->kn
, sb
);
1968 if (IS_ERR(nsdentry
))
1969 deactivate_locked_super(sb
);
1974 cgroup_put(&root
->cgrp
);
1979 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1980 int flags
, const char *unused_dev_name
,
1983 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
1984 struct dentry
*dentry
;
1989 /* Check if the caller has permission to mount. */
1990 if (!ns_capable(ns
->user_ns
, CAP_SYS_ADMIN
)) {
1992 return ERR_PTR(-EPERM
);
1996 * The first time anyone tries to mount a cgroup, enable the list
1997 * linking each css_set to its tasks and fix up all existing tasks.
1999 if (!use_task_css_set_links
)
2000 cgroup_enable_task_cg_lists();
2002 if (fs_type
== &cgroup2_fs_type
) {
2003 unsigned int root_flags
;
2005 ret
= parse_cgroup_root_flags(data
, &root_flags
);
2008 return ERR_PTR(ret
);
2011 cgrp_dfl_visible
= true;
2012 cgroup_get_live(&cgrp_dfl_root
.cgrp
);
2014 dentry
= cgroup_do_mount(&cgroup2_fs_type
, flags
, &cgrp_dfl_root
,
2015 CGROUP2_SUPER_MAGIC
, ns
);
2016 if (!IS_ERR(dentry
))
2017 apply_cgroup_root_flags(root_flags
);
2019 dentry
= cgroup1_mount(&cgroup_fs_type
, flags
, data
,
2020 CGROUP_SUPER_MAGIC
, ns
);
2027 static void cgroup_kill_sb(struct super_block
*sb
)
2029 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2030 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2033 * If @root doesn't have any mounts or children, start killing it.
2034 * This prevents new mounts by disabling percpu_ref_tryget_live().
2035 * cgroup_mount() may wait for @root's release.
2037 * And don't kill the default root.
2039 if (!list_empty(&root
->cgrp
.self
.children
) ||
2040 root
== &cgrp_dfl_root
)
2041 cgroup_put(&root
->cgrp
);
2043 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2048 struct file_system_type cgroup_fs_type
= {
2050 .mount
= cgroup_mount
,
2051 .kill_sb
= cgroup_kill_sb
,
2052 .fs_flags
= FS_USERNS_MOUNT
,
2055 static struct file_system_type cgroup2_fs_type
= {
2057 .mount
= cgroup_mount
,
2058 .kill_sb
= cgroup_kill_sb
,
2059 .fs_flags
= FS_USERNS_MOUNT
,
2062 int cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2063 struct cgroup_namespace
*ns
)
2065 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2067 return kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2070 int cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2071 struct cgroup_namespace
*ns
)
2075 mutex_lock(&cgroup_mutex
);
2076 spin_lock_irq(&css_set_lock
);
2078 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2080 spin_unlock_irq(&css_set_lock
);
2081 mutex_unlock(&cgroup_mutex
);
2085 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2088 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2089 * @task: target task
2090 * @buf: the buffer to write the path into
2091 * @buflen: the length of the buffer
2093 * Determine @task's cgroup on the first (the one with the lowest non-zero
2094 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2095 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2096 * cgroup controller callbacks.
2098 * Return value is the same as kernfs_path().
2100 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2102 struct cgroup_root
*root
;
2103 struct cgroup
*cgrp
;
2104 int hierarchy_id
= 1;
2107 mutex_lock(&cgroup_mutex
);
2108 spin_lock_irq(&css_set_lock
);
2110 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2113 cgrp
= task_cgroup_from_root(task
, root
);
2114 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2116 /* if no hierarchy exists, everyone is in "/" */
2117 ret
= strlcpy(buf
, "/", buflen
);
2120 spin_unlock_irq(&css_set_lock
);
2121 mutex_unlock(&cgroup_mutex
);
2124 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2127 * cgroup_migrate_add_task - add a migration target task to a migration context
2128 * @task: target task
2129 * @mgctx: target migration context
2131 * Add @task, which is a migration target, to @mgctx->tset. This function
2132 * becomes noop if @task doesn't need to be migrated. @task's css_set
2133 * should have been added as a migration source and @task->cg_list will be
2134 * moved from the css_set's tasks list to mg_tasks one.
2136 static void cgroup_migrate_add_task(struct task_struct
*task
,
2137 struct cgroup_mgctx
*mgctx
)
2139 struct css_set
*cset
;
2141 lockdep_assert_held(&css_set_lock
);
2143 /* @task either already exited or can't exit until the end */
2144 if (task
->flags
& PF_EXITING
)
2147 /* leave @task alone if post_fork() hasn't linked it yet */
2148 if (list_empty(&task
->cg_list
))
2151 cset
= task_css_set(task
);
2152 if (!cset
->mg_src_cgrp
)
2155 mgctx
->tset
.nr_tasks
++;
2157 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2158 if (list_empty(&cset
->mg_node
))
2159 list_add_tail(&cset
->mg_node
,
2160 &mgctx
->tset
.src_csets
);
2161 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2162 list_add_tail(&cset
->mg_dst_cset
->mg_node
,
2163 &mgctx
->tset
.dst_csets
);
2167 * cgroup_taskset_first - reset taskset and return the first task
2168 * @tset: taskset of interest
2169 * @dst_cssp: output variable for the destination css
2171 * @tset iteration is initialized and the first task is returned.
2173 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2174 struct cgroup_subsys_state
**dst_cssp
)
2176 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2177 tset
->cur_task
= NULL
;
2179 return cgroup_taskset_next(tset
, dst_cssp
);
2183 * cgroup_taskset_next - iterate to the next task in taskset
2184 * @tset: taskset of interest
2185 * @dst_cssp: output variable for the destination css
2187 * Return the next task in @tset. Iteration must have been initialized
2188 * with cgroup_taskset_first().
2190 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2191 struct cgroup_subsys_state
**dst_cssp
)
2193 struct css_set
*cset
= tset
->cur_cset
;
2194 struct task_struct
*task
= tset
->cur_task
;
2196 while (&cset
->mg_node
!= tset
->csets
) {
2198 task
= list_first_entry(&cset
->mg_tasks
,
2199 struct task_struct
, cg_list
);
2201 task
= list_next_entry(task
, cg_list
);
2203 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2204 tset
->cur_cset
= cset
;
2205 tset
->cur_task
= task
;
2208 * This function may be called both before and
2209 * after cgroup_taskset_migrate(). The two cases
2210 * can be distinguished by looking at whether @cset
2211 * has its ->mg_dst_cset set.
2213 if (cset
->mg_dst_cset
)
2214 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2216 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2221 cset
= list_next_entry(cset
, mg_node
);
2229 * cgroup_taskset_migrate - migrate a taskset
2230 * @mgctx: migration context
2232 * Migrate tasks in @mgctx as setup by migration preparation functions.
2233 * This function fails iff one of the ->can_attach callbacks fails and
2234 * guarantees that either all or none of the tasks in @mgctx are migrated.
2235 * @mgctx is consumed regardless of success.
2237 static int cgroup_migrate_execute(struct cgroup_mgctx
*mgctx
)
2239 struct cgroup_taskset
*tset
= &mgctx
->tset
;
2240 struct cgroup_subsys
*ss
;
2241 struct task_struct
*task
, *tmp_task
;
2242 struct css_set
*cset
, *tmp_cset
;
2243 int ssid
, failed_ssid
, ret
;
2245 /* check that we can legitimately attach to the cgroup */
2246 if (tset
->nr_tasks
) {
2247 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2248 if (ss
->can_attach
) {
2250 ret
= ss
->can_attach(tset
);
2253 goto out_cancel_attach
;
2256 } while_each_subsys_mask();
2260 * Now that we're guaranteed success, proceed to move all tasks to
2261 * the new cgroup. There are no failure cases after here, so this
2262 * is the commit point.
2264 spin_lock_irq(&css_set_lock
);
2265 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2266 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2267 struct css_set
*from_cset
= task_css_set(task
);
2268 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2270 get_css_set(to_cset
);
2271 to_cset
->nr_tasks
++;
2272 css_set_move_task(task
, from_cset
, to_cset
, true);
2273 put_css_set_locked(from_cset
);
2274 from_cset
->nr_tasks
--;
2277 spin_unlock_irq(&css_set_lock
);
2280 * Migration is committed, all target tasks are now on dst_csets.
2281 * Nothing is sensitive to fork() after this point. Notify
2282 * controllers that migration is complete.
2284 tset
->csets
= &tset
->dst_csets
;
2286 if (tset
->nr_tasks
) {
2287 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2292 } while_each_subsys_mask();
2296 goto out_release_tset
;
2299 if (tset
->nr_tasks
) {
2300 do_each_subsys_mask(ss
, ssid
, mgctx
->ss_mask
) {
2301 if (ssid
== failed_ssid
)
2303 if (ss
->cancel_attach
) {
2305 ss
->cancel_attach(tset
);
2307 } while_each_subsys_mask();
2310 spin_lock_irq(&css_set_lock
);
2311 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2312 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2313 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2314 list_del_init(&cset
->mg_node
);
2316 spin_unlock_irq(&css_set_lock
);
2319 * Re-initialize the cgroup_taskset structure in case it is reused
2320 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2324 tset
->csets
= &tset
->src_csets
;
2329 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2330 * @dst_cgrp: destination cgroup to test
2332 * On the default hierarchy, except for the mixable, (possible) thread root
2333 * and threaded cgroups, subtree_control must be zero for migration
2334 * destination cgroups with tasks so that child cgroups don't compete
2337 int cgroup_migrate_vet_dst(struct cgroup
*dst_cgrp
)
2339 /* v1 doesn't have any restriction */
2340 if (!cgroup_on_dfl(dst_cgrp
))
2343 /* verify @dst_cgrp can host resources */
2344 if (!cgroup_is_valid_domain(dst_cgrp
->dom_cgrp
))
2347 /* mixables don't care */
2348 if (cgroup_is_mixable(dst_cgrp
))
2352 * If @dst_cgrp is already or can become a thread root or is
2353 * threaded, it doesn't matter.
2355 if (cgroup_can_be_thread_root(dst_cgrp
) || cgroup_is_threaded(dst_cgrp
))
2358 /* apply no-internal-process constraint */
2359 if (dst_cgrp
->subtree_control
)
2366 * cgroup_migrate_finish - cleanup after attach
2367 * @mgctx: migration context
2369 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2370 * those functions for details.
2372 void cgroup_migrate_finish(struct cgroup_mgctx
*mgctx
)
2374 LIST_HEAD(preloaded
);
2375 struct css_set
*cset
, *tmp_cset
;
2377 lockdep_assert_held(&cgroup_mutex
);
2379 spin_lock_irq(&css_set_lock
);
2381 list_splice_tail_init(&mgctx
->preloaded_src_csets
, &preloaded
);
2382 list_splice_tail_init(&mgctx
->preloaded_dst_csets
, &preloaded
);
2384 list_for_each_entry_safe(cset
, tmp_cset
, &preloaded
, mg_preload_node
) {
2385 cset
->mg_src_cgrp
= NULL
;
2386 cset
->mg_dst_cgrp
= NULL
;
2387 cset
->mg_dst_cset
= NULL
;
2388 list_del_init(&cset
->mg_preload_node
);
2389 put_css_set_locked(cset
);
2392 spin_unlock_irq(&css_set_lock
);
2396 * cgroup_migrate_add_src - add a migration source css_set
2397 * @src_cset: the source css_set to add
2398 * @dst_cgrp: the destination cgroup
2399 * @mgctx: migration context
2401 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2402 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2403 * up by cgroup_migrate_finish().
2405 * This function may be called without holding cgroup_threadgroup_rwsem
2406 * even if the target is a process. Threads may be created and destroyed
2407 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2408 * into play and the preloaded css_sets are guaranteed to cover all
2411 void cgroup_migrate_add_src(struct css_set
*src_cset
,
2412 struct cgroup
*dst_cgrp
,
2413 struct cgroup_mgctx
*mgctx
)
2415 struct cgroup
*src_cgrp
;
2417 lockdep_assert_held(&cgroup_mutex
);
2418 lockdep_assert_held(&css_set_lock
);
2421 * If ->dead, @src_set is associated with one or more dead cgroups
2422 * and doesn't contain any migratable tasks. Ignore it early so
2423 * that the rest of migration path doesn't get confused by it.
2428 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2430 if (!list_empty(&src_cset
->mg_preload_node
))
2433 WARN_ON(src_cset
->mg_src_cgrp
);
2434 WARN_ON(src_cset
->mg_dst_cgrp
);
2435 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2436 WARN_ON(!list_empty(&src_cset
->mg_node
));
2438 src_cset
->mg_src_cgrp
= src_cgrp
;
2439 src_cset
->mg_dst_cgrp
= dst_cgrp
;
2440 get_css_set(src_cset
);
2441 list_add_tail(&src_cset
->mg_preload_node
, &mgctx
->preloaded_src_csets
);
2445 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2446 * @mgctx: migration context
2448 * Tasks are about to be moved and all the source css_sets have been
2449 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2450 * pins all destination css_sets, links each to its source, and append them
2451 * to @mgctx->preloaded_dst_csets.
2453 * This function must be called after cgroup_migrate_add_src() has been
2454 * called on each migration source css_set. After migration is performed
2455 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2458 int cgroup_migrate_prepare_dst(struct cgroup_mgctx
*mgctx
)
2460 struct css_set
*src_cset
, *tmp_cset
;
2462 lockdep_assert_held(&cgroup_mutex
);
2464 /* look up the dst cset for each src cset and link it to src */
2465 list_for_each_entry_safe(src_cset
, tmp_cset
, &mgctx
->preloaded_src_csets
,
2467 struct css_set
*dst_cset
;
2468 struct cgroup_subsys
*ss
;
2471 dst_cset
= find_css_set(src_cset
, src_cset
->mg_dst_cgrp
);
2475 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2478 * If src cset equals dst, it's noop. Drop the src.
2479 * cgroup_migrate() will skip the cset too. Note that we
2480 * can't handle src == dst as some nodes are used by both.
2482 if (src_cset
== dst_cset
) {
2483 src_cset
->mg_src_cgrp
= NULL
;
2484 src_cset
->mg_dst_cgrp
= NULL
;
2485 list_del_init(&src_cset
->mg_preload_node
);
2486 put_css_set(src_cset
);
2487 put_css_set(dst_cset
);
2491 src_cset
->mg_dst_cset
= dst_cset
;
2493 if (list_empty(&dst_cset
->mg_preload_node
))
2494 list_add_tail(&dst_cset
->mg_preload_node
,
2495 &mgctx
->preloaded_dst_csets
);
2497 put_css_set(dst_cset
);
2499 for_each_subsys(ss
, ssid
)
2500 if (src_cset
->subsys
[ssid
] != dst_cset
->subsys
[ssid
])
2501 mgctx
->ss_mask
|= 1 << ssid
;
2506 cgroup_migrate_finish(mgctx
);
2511 * cgroup_migrate - migrate a process or task to a cgroup
2512 * @leader: the leader of the process or the task to migrate
2513 * @threadgroup: whether @leader points to the whole process or a single task
2514 * @mgctx: migration context
2516 * Migrate a process or task denoted by @leader. If migrating a process,
2517 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2518 * responsible for invoking cgroup_migrate_add_src() and
2519 * cgroup_migrate_prepare_dst() on the targets before invoking this
2520 * function and following up with cgroup_migrate_finish().
2522 * As long as a controller's ->can_attach() doesn't fail, this function is
2523 * guaranteed to succeed. This means that, excluding ->can_attach()
2524 * failure, when migrating multiple targets, the success or failure can be
2525 * decided for all targets by invoking group_migrate_prepare_dst() before
2526 * actually starting migrating.
2528 int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2529 struct cgroup_mgctx
*mgctx
)
2531 struct task_struct
*task
;
2534 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2535 * already PF_EXITING could be freed from underneath us unless we
2536 * take an rcu_read_lock.
2538 spin_lock_irq(&css_set_lock
);
2542 cgroup_migrate_add_task(task
, mgctx
);
2545 } while_each_thread(leader
, task
);
2547 spin_unlock_irq(&css_set_lock
);
2549 return cgroup_migrate_execute(mgctx
);
2553 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2554 * @dst_cgrp: the cgroup to attach to
2555 * @leader: the task or the leader of the threadgroup to be attached
2556 * @threadgroup: attach the whole threadgroup?
2558 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2560 int cgroup_attach_task(struct cgroup
*dst_cgrp
, struct task_struct
*leader
,
2563 DEFINE_CGROUP_MGCTX(mgctx
);
2564 struct task_struct
*task
;
2567 ret
= cgroup_migrate_vet_dst(dst_cgrp
);
2571 /* look up all src csets */
2572 spin_lock_irq(&css_set_lock
);
2576 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
, &mgctx
);
2579 } while_each_thread(leader
, task
);
2581 spin_unlock_irq(&css_set_lock
);
2583 /* prepare dst csets and commit */
2584 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2586 ret
= cgroup_migrate(leader
, threadgroup
, &mgctx
);
2588 cgroup_migrate_finish(&mgctx
);
2591 trace_cgroup_attach_task(dst_cgrp
, leader
, threadgroup
);
2596 struct task_struct
*cgroup_procs_write_start(char *buf
, bool threadgroup
)
2597 __acquires(&cgroup_threadgroup_rwsem
)
2599 struct task_struct
*tsk
;
2602 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2603 return ERR_PTR(-EINVAL
);
2605 percpu_down_write(&cgroup_threadgroup_rwsem
);
2609 tsk
= find_task_by_vpid(pid
);
2611 tsk
= ERR_PTR(-ESRCH
);
2612 goto out_unlock_threadgroup
;
2619 tsk
= tsk
->group_leader
;
2622 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2623 * If userland migrates such a kthread to a non-root cgroup, it can
2624 * become trapped in a cpuset, or RT kthread may be born in a
2625 * cgroup with no rt_runtime allocated. Just say no.
2627 if (tsk
->no_cgroup_migration
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2628 tsk
= ERR_PTR(-EINVAL
);
2629 goto out_unlock_threadgroup
;
2632 get_task_struct(tsk
);
2633 goto out_unlock_rcu
;
2635 out_unlock_threadgroup
:
2636 percpu_up_write(&cgroup_threadgroup_rwsem
);
2642 void cgroup_procs_write_finish(struct task_struct
*task
)
2643 __releases(&cgroup_threadgroup_rwsem
)
2645 struct cgroup_subsys
*ss
;
2648 /* release reference from cgroup_procs_write_start() */
2649 put_task_struct(task
);
2651 percpu_up_write(&cgroup_threadgroup_rwsem
);
2652 for_each_subsys(ss
, ssid
)
2653 if (ss
->post_attach
)
2657 static void cgroup_print_ss_mask(struct seq_file
*seq
, u16 ss_mask
)
2659 struct cgroup_subsys
*ss
;
2660 bool printed
= false;
2663 do_each_subsys_mask(ss
, ssid
, ss_mask
) {
2666 seq_printf(seq
, "%s", ss
->name
);
2668 } while_each_subsys_mask();
2670 seq_putc(seq
, '\n');
2673 /* show controllers which are enabled from the parent */
2674 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2676 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2678 cgroup_print_ss_mask(seq
, cgroup_control(cgrp
));
2682 /* show controllers which are enabled for a given cgroup's children */
2683 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2685 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2687 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2692 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2693 * @cgrp: root of the subtree to update csses for
2695 * @cgrp's control masks have changed and its subtree's css associations
2696 * need to be updated accordingly. This function looks up all css_sets
2697 * which are attached to the subtree, creates the matching updated css_sets
2698 * and migrates the tasks to the new ones.
2700 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2702 DEFINE_CGROUP_MGCTX(mgctx
);
2703 struct cgroup_subsys_state
*d_css
;
2704 struct cgroup
*dsct
;
2705 struct css_set
*src_cset
;
2708 lockdep_assert_held(&cgroup_mutex
);
2710 percpu_down_write(&cgroup_threadgroup_rwsem
);
2712 /* look up all csses currently attached to @cgrp's subtree */
2713 spin_lock_irq(&css_set_lock
);
2714 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2715 struct cgrp_cset_link
*link
;
2717 list_for_each_entry(link
, &dsct
->cset_links
, cset_link
)
2718 cgroup_migrate_add_src(link
->cset
, dsct
, &mgctx
);
2720 spin_unlock_irq(&css_set_lock
);
2722 /* NULL dst indicates self on default hierarchy */
2723 ret
= cgroup_migrate_prepare_dst(&mgctx
);
2727 spin_lock_irq(&css_set_lock
);
2728 list_for_each_entry(src_cset
, &mgctx
.preloaded_src_csets
, mg_preload_node
) {
2729 struct task_struct
*task
, *ntask
;
2731 /* all tasks in src_csets need to be migrated */
2732 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2733 cgroup_migrate_add_task(task
, &mgctx
);
2735 spin_unlock_irq(&css_set_lock
);
2737 ret
= cgroup_migrate_execute(&mgctx
);
2739 cgroup_migrate_finish(&mgctx
);
2740 percpu_up_write(&cgroup_threadgroup_rwsem
);
2745 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2746 * @cgrp: root of the target subtree
2748 * Because css offlining is asynchronous, userland may try to re-enable a
2749 * controller while the previous css is still around. This function grabs
2750 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2752 void cgroup_lock_and_drain_offline(struct cgroup
*cgrp
)
2753 __acquires(&cgroup_mutex
)
2755 struct cgroup
*dsct
;
2756 struct cgroup_subsys_state
*d_css
;
2757 struct cgroup_subsys
*ss
;
2761 mutex_lock(&cgroup_mutex
);
2763 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2764 for_each_subsys(ss
, ssid
) {
2765 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2768 if (!css
|| !percpu_ref_is_dying(&css
->refcnt
))
2771 cgroup_get_live(dsct
);
2772 prepare_to_wait(&dsct
->offline_waitq
, &wait
,
2773 TASK_UNINTERRUPTIBLE
);
2775 mutex_unlock(&cgroup_mutex
);
2777 finish_wait(&dsct
->offline_waitq
, &wait
);
2786 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2787 * @cgrp: root of the target subtree
2789 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2790 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2793 static void cgroup_save_control(struct cgroup
*cgrp
)
2795 struct cgroup
*dsct
;
2796 struct cgroup_subsys_state
*d_css
;
2798 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2799 dsct
->old_subtree_control
= dsct
->subtree_control
;
2800 dsct
->old_subtree_ss_mask
= dsct
->subtree_ss_mask
;
2801 dsct
->old_dom_cgrp
= dsct
->dom_cgrp
;
2806 * cgroup_propagate_control - refresh control masks of a subtree
2807 * @cgrp: root of the target subtree
2809 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2810 * ->subtree_control and propagate controller availability through the
2811 * subtree so that descendants don't have unavailable controllers enabled.
2813 static void cgroup_propagate_control(struct cgroup
*cgrp
)
2815 struct cgroup
*dsct
;
2816 struct cgroup_subsys_state
*d_css
;
2818 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2819 dsct
->subtree_control
&= cgroup_control(dsct
);
2820 dsct
->subtree_ss_mask
=
2821 cgroup_calc_subtree_ss_mask(dsct
->subtree_control
,
2822 cgroup_ss_mask(dsct
));
2827 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
2828 * @cgrp: root of the target subtree
2830 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
2831 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2834 static void cgroup_restore_control(struct cgroup
*cgrp
)
2836 struct cgroup
*dsct
;
2837 struct cgroup_subsys_state
*d_css
;
2839 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2840 dsct
->subtree_control
= dsct
->old_subtree_control
;
2841 dsct
->subtree_ss_mask
= dsct
->old_subtree_ss_mask
;
2842 dsct
->dom_cgrp
= dsct
->old_dom_cgrp
;
2846 static bool css_visible(struct cgroup_subsys_state
*css
)
2848 struct cgroup_subsys
*ss
= css
->ss
;
2849 struct cgroup
*cgrp
= css
->cgroup
;
2851 if (cgroup_control(cgrp
) & (1 << ss
->id
))
2853 if (!(cgroup_ss_mask(cgrp
) & (1 << ss
->id
)))
2855 return cgroup_on_dfl(cgrp
) && ss
->implicit_on_dfl
;
2859 * cgroup_apply_control_enable - enable or show csses according to control
2860 * @cgrp: root of the target subtree
2862 * Walk @cgrp's subtree and create new csses or make the existing ones
2863 * visible. A css is created invisible if it's being implicitly enabled
2864 * through dependency. An invisible css is made visible when the userland
2865 * explicitly enables it.
2867 * Returns 0 on success, -errno on failure. On failure, csses which have
2868 * been processed already aren't cleaned up. The caller is responsible for
2869 * cleaning up with cgroup_apply_control_disable().
2871 static int cgroup_apply_control_enable(struct cgroup
*cgrp
)
2873 struct cgroup
*dsct
;
2874 struct cgroup_subsys_state
*d_css
;
2875 struct cgroup_subsys
*ss
;
2878 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
) {
2879 for_each_subsys(ss
, ssid
) {
2880 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2882 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
2884 if (!(cgroup_ss_mask(dsct
) & (1 << ss
->id
)))
2888 css
= css_create(dsct
, ss
);
2890 return PTR_ERR(css
);
2893 if (css_visible(css
)) {
2894 ret
= css_populate_dir(css
);
2905 * cgroup_apply_control_disable - kill or hide csses according to control
2906 * @cgrp: root of the target subtree
2908 * Walk @cgrp's subtree and kill and hide csses so that they match
2909 * cgroup_ss_mask() and cgroup_visible_mask().
2911 * A css is hidden when the userland requests it to be disabled while other
2912 * subsystems are still depending on it. The css must not actively control
2913 * resources and be in the vanilla state if it's made visible again later.
2914 * Controllers which may be depended upon should provide ->css_reset() for
2917 static void cgroup_apply_control_disable(struct cgroup
*cgrp
)
2919 struct cgroup
*dsct
;
2920 struct cgroup_subsys_state
*d_css
;
2921 struct cgroup_subsys
*ss
;
2924 cgroup_for_each_live_descendant_post(dsct
, d_css
, cgrp
) {
2925 for_each_subsys(ss
, ssid
) {
2926 struct cgroup_subsys_state
*css
= cgroup_css(dsct
, ss
);
2928 WARN_ON_ONCE(css
&& percpu_ref_is_dying(&css
->refcnt
));
2934 !(cgroup_ss_mask(dsct
) & (1 << ss
->id
))) {
2936 } else if (!css_visible(css
)) {
2946 * cgroup_apply_control - apply control mask updates to the subtree
2947 * @cgrp: root of the target subtree
2949 * subsystems can be enabled and disabled in a subtree using the following
2952 * 1. Call cgroup_save_control() to stash the current state.
2953 * 2. Update ->subtree_control masks in the subtree as desired.
2954 * 3. Call cgroup_apply_control() to apply the changes.
2955 * 4. Optionally perform other related operations.
2956 * 5. Call cgroup_finalize_control() to finish up.
2958 * This function implements step 3 and propagates the mask changes
2959 * throughout @cgrp's subtree, updates csses accordingly and perform
2960 * process migrations.
2962 static int cgroup_apply_control(struct cgroup
*cgrp
)
2966 cgroup_propagate_control(cgrp
);
2968 ret
= cgroup_apply_control_enable(cgrp
);
2973 * At this point, cgroup_e_css() results reflect the new csses
2974 * making the following cgroup_update_dfl_csses() properly update
2975 * css associations of all tasks in the subtree.
2977 ret
= cgroup_update_dfl_csses(cgrp
);
2985 * cgroup_finalize_control - finalize control mask update
2986 * @cgrp: root of the target subtree
2987 * @ret: the result of the update
2989 * Finalize control mask update. See cgroup_apply_control() for more info.
2991 static void cgroup_finalize_control(struct cgroup
*cgrp
, int ret
)
2994 cgroup_restore_control(cgrp
);
2995 cgroup_propagate_control(cgrp
);
2998 cgroup_apply_control_disable(cgrp
);
3001 static int cgroup_vet_subtree_control_enable(struct cgroup
*cgrp
, u16 enable
)
3003 u16 domain_enable
= enable
& ~cgrp_dfl_threaded_ss_mask
;
3005 /* if nothing is getting enabled, nothing to worry about */
3009 /* can @cgrp host any resources? */
3010 if (!cgroup_is_valid_domain(cgrp
->dom_cgrp
))
3013 /* mixables don't care */
3014 if (cgroup_is_mixable(cgrp
))
3017 if (domain_enable
) {
3018 /* can't enable domain controllers inside a thread subtree */
3019 if (cgroup_is_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3023 * Threaded controllers can handle internal competitions
3024 * and are always allowed inside a (prospective) thread
3027 if (cgroup_can_be_thread_root(cgrp
) || cgroup_is_threaded(cgrp
))
3032 * Controllers can't be enabled for a cgroup with tasks to avoid
3033 * child cgroups competing against tasks.
3035 if (cgroup_has_tasks(cgrp
))
3041 /* change the enabled child controllers for a cgroup in the default hierarchy */
3042 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3043 char *buf
, size_t nbytes
,
3046 u16 enable
= 0, disable
= 0;
3047 struct cgroup
*cgrp
, *child
;
3048 struct cgroup_subsys
*ss
;
3053 * Parse input - space separated list of subsystem names prefixed
3054 * with either + or -.
3056 buf
= strstrip(buf
);
3057 while ((tok
= strsep(&buf
, " "))) {
3060 do_each_subsys_mask(ss
, ssid
, ~cgrp_dfl_inhibit_ss_mask
) {
3061 if (!cgroup_ssid_enabled(ssid
) ||
3062 strcmp(tok
+ 1, ss
->name
))
3066 enable
|= 1 << ssid
;
3067 disable
&= ~(1 << ssid
);
3068 } else if (*tok
== '-') {
3069 disable
|= 1 << ssid
;
3070 enable
&= ~(1 << ssid
);
3075 } while_each_subsys_mask();
3076 if (ssid
== CGROUP_SUBSYS_COUNT
)
3080 cgrp
= cgroup_kn_lock_live(of
->kn
, true);
3084 for_each_subsys(ss
, ssid
) {
3085 if (enable
& (1 << ssid
)) {
3086 if (cgrp
->subtree_control
& (1 << ssid
)) {
3087 enable
&= ~(1 << ssid
);
3091 if (!(cgroup_control(cgrp
) & (1 << ssid
))) {
3095 } else if (disable
& (1 << ssid
)) {
3096 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3097 disable
&= ~(1 << ssid
);
3101 /* a child has it enabled? */
3102 cgroup_for_each_live_child(child
, cgrp
) {
3103 if (child
->subtree_control
& (1 << ssid
)) {
3111 if (!enable
&& !disable
) {
3116 ret
= cgroup_vet_subtree_control_enable(cgrp
, enable
);
3120 /* save and update control masks and prepare csses */
3121 cgroup_save_control(cgrp
);
3123 cgrp
->subtree_control
|= enable
;
3124 cgrp
->subtree_control
&= ~disable
;
3126 ret
= cgroup_apply_control(cgrp
);
3127 cgroup_finalize_control(cgrp
, ret
);
3131 kernfs_activate(cgrp
->kn
);
3133 cgroup_kn_unlock(of
->kn
);
3134 return ret
?: nbytes
;
3138 * cgroup_enable_threaded - make @cgrp threaded
3139 * @cgrp: the target cgroup
3141 * Called when "threaded" is written to the cgroup.type interface file and
3142 * tries to make @cgrp threaded and join the parent's resource domain.
3143 * This function is never called on the root cgroup as cgroup.type doesn't
3146 static int cgroup_enable_threaded(struct cgroup
*cgrp
)
3148 struct cgroup
*parent
= cgroup_parent(cgrp
);
3149 struct cgroup
*dom_cgrp
= parent
->dom_cgrp
;
3150 struct cgroup
*dsct
;
3151 struct cgroup_subsys_state
*d_css
;
3154 lockdep_assert_held(&cgroup_mutex
);
3156 /* noop if already threaded */
3157 if (cgroup_is_threaded(cgrp
))
3161 * If @cgroup is populated or has domain controllers enabled, it
3162 * can't be switched. While the below cgroup_can_be_thread_root()
3163 * test can catch the same conditions, that's only when @parent is
3164 * not mixable, so let's check it explicitly.
3166 if (cgroup_is_populated(cgrp
) ||
3167 cgrp
->subtree_control
& ~cgrp_dfl_threaded_ss_mask
)
3170 /* we're joining the parent's domain, ensure its validity */
3171 if (!cgroup_is_valid_domain(dom_cgrp
) ||
3172 !cgroup_can_be_thread_root(dom_cgrp
))
3176 * The following shouldn't cause actual migrations and should
3179 cgroup_save_control(cgrp
);
3181 cgroup_for_each_live_descendant_pre(dsct
, d_css
, cgrp
)
3182 if (dsct
== cgrp
|| cgroup_is_threaded(dsct
))
3183 dsct
->dom_cgrp
= dom_cgrp
;
3185 ret
= cgroup_apply_control(cgrp
);
3187 parent
->nr_threaded_children
++;
3189 cgroup_finalize_control(cgrp
, ret
);
3193 static int cgroup_type_show(struct seq_file
*seq
, void *v
)
3195 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3197 if (cgroup_is_threaded(cgrp
))
3198 seq_puts(seq
, "threaded\n");
3199 else if (!cgroup_is_valid_domain(cgrp
))
3200 seq_puts(seq
, "domain invalid\n");
3201 else if (cgroup_is_thread_root(cgrp
))
3202 seq_puts(seq
, "domain threaded\n");
3204 seq_puts(seq
, "domain\n");
3209 static ssize_t
cgroup_type_write(struct kernfs_open_file
*of
, char *buf
,
3210 size_t nbytes
, loff_t off
)
3212 struct cgroup
*cgrp
;
3215 /* only switching to threaded mode is supported */
3216 if (strcmp(strstrip(buf
), "threaded"))
3219 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3223 /* threaded can only be enabled */
3224 ret
= cgroup_enable_threaded(cgrp
);
3226 cgroup_kn_unlock(of
->kn
);
3227 return ret
?: nbytes
;
3230 static int cgroup_max_descendants_show(struct seq_file
*seq
, void *v
)
3232 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3233 int descendants
= READ_ONCE(cgrp
->max_descendants
);
3235 if (descendants
== INT_MAX
)
3236 seq_puts(seq
, "max\n");
3238 seq_printf(seq
, "%d\n", descendants
);
3243 static ssize_t
cgroup_max_descendants_write(struct kernfs_open_file
*of
,
3244 char *buf
, size_t nbytes
, loff_t off
)
3246 struct cgroup
*cgrp
;
3250 buf
= strstrip(buf
);
3251 if (!strcmp(buf
, "max")) {
3252 descendants
= INT_MAX
;
3254 ret
= kstrtoint(buf
, 0, &descendants
);
3259 if (descendants
< 0)
3262 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3266 cgrp
->max_descendants
= descendants
;
3268 cgroup_kn_unlock(of
->kn
);
3273 static int cgroup_max_depth_show(struct seq_file
*seq
, void *v
)
3275 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
3276 int depth
= READ_ONCE(cgrp
->max_depth
);
3278 if (depth
== INT_MAX
)
3279 seq_puts(seq
, "max\n");
3281 seq_printf(seq
, "%d\n", depth
);
3286 static ssize_t
cgroup_max_depth_write(struct kernfs_open_file
*of
,
3287 char *buf
, size_t nbytes
, loff_t off
)
3289 struct cgroup
*cgrp
;
3293 buf
= strstrip(buf
);
3294 if (!strcmp(buf
, "max")) {
3297 ret
= kstrtoint(buf
, 0, &depth
);
3305 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
3309 cgrp
->max_depth
= depth
;
3311 cgroup_kn_unlock(of
->kn
);
3316 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3318 seq_printf(seq
, "populated %d\n",
3319 cgroup_is_populated(seq_css(seq
)->cgroup
));
3323 static int cgroup_stat_show(struct seq_file
*seq
, void *v
)
3325 struct cgroup
*cgroup
= seq_css(seq
)->cgroup
;
3327 seq_printf(seq
, "nr_descendants %d\n",
3328 cgroup
->nr_descendants
);
3329 seq_printf(seq
, "nr_dying_descendants %d\n",
3330 cgroup
->nr_dying_descendants
);
3335 static int cgroup_file_open(struct kernfs_open_file
*of
)
3337 struct cftype
*cft
= of
->kn
->priv
;
3340 return cft
->open(of
);
3344 static void cgroup_file_release(struct kernfs_open_file
*of
)
3346 struct cftype
*cft
= of
->kn
->priv
;
3352 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3353 size_t nbytes
, loff_t off
)
3355 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
3356 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3357 struct cftype
*cft
= of
->kn
->priv
;
3358 struct cgroup_subsys_state
*css
;
3362 * If namespaces are delegation boundaries, disallow writes to
3363 * files in an non-init namespace root from inside the namespace
3364 * except for the files explicitly marked delegatable -
3365 * cgroup.procs and cgroup.subtree_control.
3367 if ((cgrp
->root
->flags
& CGRP_ROOT_NS_DELEGATE
) &&
3368 !(cft
->flags
& CFTYPE_NS_DELEGATABLE
) &&
3369 ns
!= &init_cgroup_ns
&& ns
->root_cset
->dfl_cgrp
== cgrp
)
3373 return cft
->write(of
, buf
, nbytes
, off
);
3376 * kernfs guarantees that a file isn't deleted with operations in
3377 * flight, which means that the matching css is and stays alive and
3378 * doesn't need to be pinned. The RCU locking is not necessary
3379 * either. It's just for the convenience of using cgroup_css().
3382 css
= cgroup_css(cgrp
, cft
->ss
);
3385 if (cft
->write_u64
) {
3386 unsigned long long v
;
3387 ret
= kstrtoull(buf
, 0, &v
);
3389 ret
= cft
->write_u64(css
, cft
, v
);
3390 } else if (cft
->write_s64
) {
3392 ret
= kstrtoll(buf
, 0, &v
);
3394 ret
= cft
->write_s64(css
, cft
, v
);
3399 return ret
?: nbytes
;
3402 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3404 return seq_cft(seq
)->seq_start(seq
, ppos
);
3407 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3409 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3412 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3414 if (seq_cft(seq
)->seq_stop
)
3415 seq_cft(seq
)->seq_stop(seq
, v
);
3418 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3420 struct cftype
*cft
= seq_cft(m
);
3421 struct cgroup_subsys_state
*css
= seq_css(m
);
3424 return cft
->seq_show(m
, arg
);
3427 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3428 else if (cft
->read_s64
)
3429 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3435 static struct kernfs_ops cgroup_kf_single_ops
= {
3436 .atomic_write_len
= PAGE_SIZE
,
3437 .open
= cgroup_file_open
,
3438 .release
= cgroup_file_release
,
3439 .write
= cgroup_file_write
,
3440 .seq_show
= cgroup_seqfile_show
,
3443 static struct kernfs_ops cgroup_kf_ops
= {
3444 .atomic_write_len
= PAGE_SIZE
,
3445 .open
= cgroup_file_open
,
3446 .release
= cgroup_file_release
,
3447 .write
= cgroup_file_write
,
3448 .seq_start
= cgroup_seqfile_start
,
3449 .seq_next
= cgroup_seqfile_next
,
3450 .seq_stop
= cgroup_seqfile_stop
,
3451 .seq_show
= cgroup_seqfile_show
,
3454 /* set uid and gid of cgroup dirs and files to that of the creator */
3455 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3457 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3458 .ia_uid
= current_fsuid(),
3459 .ia_gid
= current_fsgid(), };
3461 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3462 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3465 return kernfs_setattr(kn
, &iattr
);
3468 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3471 char name
[CGROUP_FILE_NAME_MAX
];
3472 struct kernfs_node
*kn
;
3473 struct lock_class_key
*key
= NULL
;
3476 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3477 key
= &cft
->lockdep_key
;
3479 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3480 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3485 ret
= cgroup_kn_set_ugid(kn
);
3491 if (cft
->file_offset
) {
3492 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3494 spin_lock_irq(&cgroup_file_kn_lock
);
3496 spin_unlock_irq(&cgroup_file_kn_lock
);
3503 * cgroup_addrm_files - add or remove files to a cgroup directory
3504 * @css: the target css
3505 * @cgrp: the target cgroup (usually css->cgroup)
3506 * @cfts: array of cftypes to be added
3507 * @is_add: whether to add or remove
3509 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3510 * For removals, this function never fails.
3512 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3513 struct cgroup
*cgrp
, struct cftype cfts
[],
3516 struct cftype
*cft
, *cft_end
= NULL
;
3519 lockdep_assert_held(&cgroup_mutex
);
3522 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3523 /* does cft->flags tell us to skip this file on @cgrp? */
3524 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3526 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3528 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3530 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3534 ret
= cgroup_add_file(css
, cgrp
, cft
);
3536 pr_warn("%s: failed to add %s, err=%d\n",
3537 __func__
, cft
->name
, ret
);
3543 cgroup_rm_file(cgrp
, cft
);
3549 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3551 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3552 struct cgroup
*root
= &ss
->root
->cgrp
;
3553 struct cgroup_subsys_state
*css
;
3556 lockdep_assert_held(&cgroup_mutex
);
3558 /* add/rm files for all cgroups created before */
3559 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3560 struct cgroup
*cgrp
= css
->cgroup
;
3562 if (!(css
->flags
& CSS_VISIBLE
))
3565 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3571 kernfs_activate(root
->kn
);
3575 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3579 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3580 /* free copy for custom atomic_write_len, see init_cftypes() */
3581 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3586 /* revert flags set by cgroup core while adding @cfts */
3587 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3591 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3595 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3596 struct kernfs_ops
*kf_ops
;
3598 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3601 kf_ops
= &cgroup_kf_ops
;
3603 kf_ops
= &cgroup_kf_single_ops
;
3606 * Ugh... if @cft wants a custom max_write_len, we need to
3607 * make a copy of kf_ops to set its atomic_write_len.
3609 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3610 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3612 cgroup_exit_cftypes(cfts
);
3615 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3618 cft
->kf_ops
= kf_ops
;
3625 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3627 lockdep_assert_held(&cgroup_mutex
);
3629 if (!cfts
|| !cfts
[0].ss
)
3632 list_del(&cfts
->node
);
3633 cgroup_apply_cftypes(cfts
, false);
3634 cgroup_exit_cftypes(cfts
);
3639 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3640 * @cfts: zero-length name terminated array of cftypes
3642 * Unregister @cfts. Files described by @cfts are removed from all
3643 * existing cgroups and all future cgroups won't have them either. This
3644 * function can be called anytime whether @cfts' subsys is attached or not.
3646 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3649 int cgroup_rm_cftypes(struct cftype
*cfts
)
3653 mutex_lock(&cgroup_mutex
);
3654 ret
= cgroup_rm_cftypes_locked(cfts
);
3655 mutex_unlock(&cgroup_mutex
);
3660 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3661 * @ss: target cgroup subsystem
3662 * @cfts: zero-length name terminated array of cftypes
3664 * Register @cfts to @ss. Files described by @cfts are created for all
3665 * existing cgroups to which @ss is attached and all future cgroups will
3666 * have them too. This function can be called anytime whether @ss is
3669 * Returns 0 on successful registration, -errno on failure. Note that this
3670 * function currently returns 0 as long as @cfts registration is successful
3671 * even if some file creation attempts on existing cgroups fail.
3673 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3677 if (!cgroup_ssid_enabled(ss
->id
))
3680 if (!cfts
|| cfts
[0].name
[0] == '\0')
3683 ret
= cgroup_init_cftypes(ss
, cfts
);
3687 mutex_lock(&cgroup_mutex
);
3689 list_add_tail(&cfts
->node
, &ss
->cfts
);
3690 ret
= cgroup_apply_cftypes(cfts
, true);
3692 cgroup_rm_cftypes_locked(cfts
);
3694 mutex_unlock(&cgroup_mutex
);
3699 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3700 * @ss: target cgroup subsystem
3701 * @cfts: zero-length name terminated array of cftypes
3703 * Similar to cgroup_add_cftypes() but the added files are only used for
3704 * the default hierarchy.
3706 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3710 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3711 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3712 return cgroup_add_cftypes(ss
, cfts
);
3716 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3717 * @ss: target cgroup subsystem
3718 * @cfts: zero-length name terminated array of cftypes
3720 * Similar to cgroup_add_cftypes() but the added files are only used for
3721 * the legacy hierarchies.
3723 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3727 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3728 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3729 return cgroup_add_cftypes(ss
, cfts
);
3733 * cgroup_file_notify - generate a file modified event for a cgroup_file
3734 * @cfile: target cgroup_file
3736 * @cfile must have been obtained by setting cftype->file_offset.
3738 void cgroup_file_notify(struct cgroup_file
*cfile
)
3740 unsigned long flags
;
3742 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
3744 kernfs_notify(cfile
->kn
);
3745 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
3749 * css_next_child - find the next child of a given css
3750 * @pos: the current position (%NULL to initiate traversal)
3751 * @parent: css whose children to walk
3753 * This function returns the next child of @parent and should be called
3754 * under either cgroup_mutex or RCU read lock. The only requirement is
3755 * that @parent and @pos are accessible. The next sibling is guaranteed to
3756 * be returned regardless of their states.
3758 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3759 * css which finished ->css_online() is guaranteed to be visible in the
3760 * future iterations and will stay visible until the last reference is put.
3761 * A css which hasn't finished ->css_online() or already finished
3762 * ->css_offline() may show up during traversal. It's each subsystem's
3763 * responsibility to synchronize against on/offlining.
3765 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3766 struct cgroup_subsys_state
*parent
)
3768 struct cgroup_subsys_state
*next
;
3770 cgroup_assert_mutex_or_rcu_locked();
3773 * @pos could already have been unlinked from the sibling list.
3774 * Once a cgroup is removed, its ->sibling.next is no longer
3775 * updated when its next sibling changes. CSS_RELEASED is set when
3776 * @pos is taken off list, at which time its next pointer is valid,
3777 * and, as releases are serialized, the one pointed to by the next
3778 * pointer is guaranteed to not have started release yet. This
3779 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3780 * critical section, the one pointed to by its next pointer is
3781 * guaranteed to not have finished its RCU grace period even if we
3782 * have dropped rcu_read_lock() inbetween iterations.
3784 * If @pos has CSS_RELEASED set, its next pointer can't be
3785 * dereferenced; however, as each css is given a monotonically
3786 * increasing unique serial number and always appended to the
3787 * sibling list, the next one can be found by walking the parent's
3788 * children until the first css with higher serial number than
3789 * @pos's. While this path can be slower, it happens iff iteration
3790 * races against release and the race window is very small.
3793 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3794 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3795 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3797 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3798 if (next
->serial_nr
> pos
->serial_nr
)
3803 * @next, if not pointing to the head, can be dereferenced and is
3806 if (&next
->sibling
!= &parent
->children
)
3812 * css_next_descendant_pre - find the next descendant for pre-order walk
3813 * @pos: the current position (%NULL to initiate traversal)
3814 * @root: css whose descendants to walk
3816 * To be used by css_for_each_descendant_pre(). Find the next descendant
3817 * to visit for pre-order traversal of @root's descendants. @root is
3818 * included in the iteration and the first node to be visited.
3820 * While this function requires cgroup_mutex or RCU read locking, it
3821 * doesn't require the whole traversal to be contained in a single critical
3822 * section. This function will return the correct next descendant as long
3823 * as both @pos and @root are accessible and @pos is a descendant of @root.
3825 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3826 * css which finished ->css_online() is guaranteed to be visible in the
3827 * future iterations and will stay visible until the last reference is put.
3828 * A css which hasn't finished ->css_online() or already finished
3829 * ->css_offline() may show up during traversal. It's each subsystem's
3830 * responsibility to synchronize against on/offlining.
3832 struct cgroup_subsys_state
*
3833 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3834 struct cgroup_subsys_state
*root
)
3836 struct cgroup_subsys_state
*next
;
3838 cgroup_assert_mutex_or_rcu_locked();
3840 /* if first iteration, visit @root */
3844 /* visit the first child if exists */
3845 next
= css_next_child(NULL
, pos
);
3849 /* no child, visit my or the closest ancestor's next sibling */
3850 while (pos
!= root
) {
3851 next
= css_next_child(pos
, pos
->parent
);
3861 * css_rightmost_descendant - return the rightmost descendant of a css
3862 * @pos: css of interest
3864 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3865 * is returned. This can be used during pre-order traversal to skip
3868 * While this function requires cgroup_mutex or RCU read locking, it
3869 * doesn't require the whole traversal to be contained in a single critical
3870 * section. This function will return the correct rightmost descendant as
3871 * long as @pos is accessible.
3873 struct cgroup_subsys_state
*
3874 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3876 struct cgroup_subsys_state
*last
, *tmp
;
3878 cgroup_assert_mutex_or_rcu_locked();
3882 /* ->prev isn't RCU safe, walk ->next till the end */
3884 css_for_each_child(tmp
, last
)
3891 static struct cgroup_subsys_state
*
3892 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3894 struct cgroup_subsys_state
*last
;
3898 pos
= css_next_child(NULL
, pos
);
3905 * css_next_descendant_post - find the next descendant for post-order walk
3906 * @pos: the current position (%NULL to initiate traversal)
3907 * @root: css whose descendants to walk
3909 * To be used by css_for_each_descendant_post(). Find the next descendant
3910 * to visit for post-order traversal of @root's descendants. @root is
3911 * included in the iteration and the last node to be visited.
3913 * While this function requires cgroup_mutex or RCU read locking, it
3914 * doesn't require the whole traversal to be contained in a single critical
3915 * section. This function will return the correct next descendant as long
3916 * as both @pos and @cgroup are accessible and @pos is a descendant of
3919 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3920 * css which finished ->css_online() is guaranteed to be visible in the
3921 * future iterations and will stay visible until the last reference is put.
3922 * A css which hasn't finished ->css_online() or already finished
3923 * ->css_offline() may show up during traversal. It's each subsystem's
3924 * responsibility to synchronize against on/offlining.
3926 struct cgroup_subsys_state
*
3927 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3928 struct cgroup_subsys_state
*root
)
3930 struct cgroup_subsys_state
*next
;
3932 cgroup_assert_mutex_or_rcu_locked();
3934 /* if first iteration, visit leftmost descendant which may be @root */
3936 return css_leftmost_descendant(root
);
3938 /* if we visited @root, we're done */
3942 /* if there's an unvisited sibling, visit its leftmost descendant */
3943 next
= css_next_child(pos
, pos
->parent
);
3945 return css_leftmost_descendant(next
);
3947 /* no sibling left, visit parent */
3952 * css_has_online_children - does a css have online children
3953 * @css: the target css
3955 * Returns %true if @css has any online children; otherwise, %false. This
3956 * function can be called from any context but the caller is responsible
3957 * for synchronizing against on/offlining as necessary.
3959 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3961 struct cgroup_subsys_state
*child
;
3965 css_for_each_child(child
, css
) {
3966 if (child
->flags
& CSS_ONLINE
) {
3975 static struct css_set
*css_task_iter_next_css_set(struct css_task_iter
*it
)
3977 struct list_head
*l
;
3978 struct cgrp_cset_link
*link
;
3979 struct css_set
*cset
;
3981 lockdep_assert_held(&css_set_lock
);
3983 /* find the next threaded cset */
3984 if (it
->tcset_pos
) {
3985 l
= it
->tcset_pos
->next
;
3987 if (l
!= it
->tcset_head
) {
3989 return container_of(l
, struct css_set
,
3990 threaded_csets_node
);
3993 it
->tcset_pos
= NULL
;
3996 /* find the next cset */
3999 if (l
== it
->cset_head
) {
4000 it
->cset_pos
= NULL
;
4005 cset
= container_of(l
, struct css_set
, e_cset_node
[it
->ss
->id
]);
4007 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
4013 /* initialize threaded css_set walking */
4014 if (it
->flags
& CSS_TASK_ITER_THREADED
) {
4016 put_css_set_locked(it
->cur_dcset
);
4017 it
->cur_dcset
= cset
;
4020 it
->tcset_head
= &cset
->threaded_csets
;
4021 it
->tcset_pos
= &cset
->threaded_csets
;
4028 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4029 * @it: the iterator to advance
4031 * Advance @it to the next css_set to walk.
4033 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
4035 struct css_set
*cset
;
4037 lockdep_assert_held(&css_set_lock
);
4039 /* Advance to the next non-empty css_set */
4041 cset
= css_task_iter_next_css_set(it
);
4043 it
->task_pos
= NULL
;
4046 } while (!css_set_populated(cset
));
4048 if (!list_empty(&cset
->tasks
))
4049 it
->task_pos
= cset
->tasks
.next
;
4051 it
->task_pos
= cset
->mg_tasks
.next
;
4053 it
->tasks_head
= &cset
->tasks
;
4054 it
->mg_tasks_head
= &cset
->mg_tasks
;
4057 * We don't keep css_sets locked across iteration steps and thus
4058 * need to take steps to ensure that iteration can be resumed after
4059 * the lock is re-acquired. Iteration is performed at two levels -
4060 * css_sets and tasks in them.
4062 * Once created, a css_set never leaves its cgroup lists, so a
4063 * pinned css_set is guaranteed to stay put and we can resume
4064 * iteration afterwards.
4066 * Tasks may leave @cset across iteration steps. This is resolved
4067 * by registering each iterator with the css_set currently being
4068 * walked and making css_set_move_task() advance iterators whose
4069 * next task is leaving.
4072 list_del(&it
->iters_node
);
4073 put_css_set_locked(it
->cur_cset
);
4076 it
->cur_cset
= cset
;
4077 list_add(&it
->iters_node
, &cset
->task_iters
);
4080 static void css_task_iter_advance(struct css_task_iter
*it
)
4082 struct list_head
*next
;
4084 lockdep_assert_held(&css_set_lock
);
4088 * Advance iterator to find next entry. cset->tasks is
4089 * consumed first and then ->mg_tasks. After ->mg_tasks,
4090 * we move onto the next cset.
4092 next
= it
->task_pos
->next
;
4094 if (next
== it
->tasks_head
)
4095 next
= it
->mg_tasks_head
->next
;
4097 if (next
== it
->mg_tasks_head
)
4098 css_task_iter_advance_css_set(it
);
4100 it
->task_pos
= next
;
4102 /* called from start, proceed to the first cset */
4103 css_task_iter_advance_css_set(it
);
4106 /* if PROCS, skip over tasks which aren't group leaders */
4107 if ((it
->flags
& CSS_TASK_ITER_PROCS
) && it
->task_pos
&&
4108 !thread_group_leader(list_entry(it
->task_pos
, struct task_struct
,
4114 * css_task_iter_start - initiate task iteration
4115 * @css: the css to walk tasks of
4116 * @flags: CSS_TASK_ITER_* flags
4117 * @it: the task iterator to use
4119 * Initiate iteration through the tasks of @css. The caller can call
4120 * css_task_iter_next() to walk through the tasks until the function
4121 * returns NULL. On completion of iteration, css_task_iter_end() must be
4124 void css_task_iter_start(struct cgroup_subsys_state
*css
, unsigned int flags
,
4125 struct css_task_iter
*it
)
4127 /* no one should try to iterate before mounting cgroups */
4128 WARN_ON_ONCE(!use_task_css_set_links
);
4130 memset(it
, 0, sizeof(*it
));
4132 spin_lock_irq(&css_set_lock
);
4138 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4140 it
->cset_pos
= &css
->cgroup
->cset_links
;
4142 it
->cset_head
= it
->cset_pos
;
4144 css_task_iter_advance(it
);
4146 spin_unlock_irq(&css_set_lock
);
4150 * css_task_iter_next - return the next task for the iterator
4151 * @it: the task iterator being iterated
4153 * The "next" function for task iteration. @it should have been
4154 * initialized via css_task_iter_start(). Returns NULL when the iteration
4157 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4160 put_task_struct(it
->cur_task
);
4161 it
->cur_task
= NULL
;
4164 spin_lock_irq(&css_set_lock
);
4167 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4169 get_task_struct(it
->cur_task
);
4170 css_task_iter_advance(it
);
4173 spin_unlock_irq(&css_set_lock
);
4175 return it
->cur_task
;
4179 * css_task_iter_end - finish task iteration
4180 * @it: the task iterator to finish
4182 * Finish task iteration started by css_task_iter_start().
4184 void css_task_iter_end(struct css_task_iter
*it
)
4187 spin_lock_irq(&css_set_lock
);
4188 list_del(&it
->iters_node
);
4189 put_css_set_locked(it
->cur_cset
);
4190 spin_unlock_irq(&css_set_lock
);
4194 put_css_set(it
->cur_dcset
);
4197 put_task_struct(it
->cur_task
);
4200 static void cgroup_procs_release(struct kernfs_open_file
*of
)
4203 css_task_iter_end(of
->priv
);
4208 static void *cgroup_procs_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4210 struct kernfs_open_file
*of
= s
->private;
4211 struct css_task_iter
*it
= of
->priv
;
4213 return css_task_iter_next(it
);
4216 static void *__cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
,
4217 unsigned int iter_flags
)
4219 struct kernfs_open_file
*of
= s
->private;
4220 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4221 struct css_task_iter
*it
= of
->priv
;
4224 * When a seq_file is seeked, it's always traversed sequentially
4225 * from position 0, so we can simply keep iterating on !0 *pos.
4228 if (WARN_ON_ONCE((*pos
)++))
4229 return ERR_PTR(-EINVAL
);
4231 it
= kzalloc(sizeof(*it
), GFP_KERNEL
);
4233 return ERR_PTR(-ENOMEM
);
4235 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4236 } else if (!(*pos
)++) {
4237 css_task_iter_end(it
);
4238 css_task_iter_start(&cgrp
->self
, iter_flags
, it
);
4241 return cgroup_procs_next(s
, NULL
, NULL
);
4244 static void *cgroup_procs_start(struct seq_file
*s
, loff_t
*pos
)
4246 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4249 * All processes of a threaded subtree belong to the domain cgroup
4250 * of the subtree. Only threads can be distributed across the
4251 * subtree. Reject reads on cgroup.procs in the subtree proper.
4252 * They're always empty anyway.
4254 if (cgroup_is_threaded(cgrp
))
4255 return ERR_PTR(-EOPNOTSUPP
);
4257 return __cgroup_procs_start(s
, pos
, CSS_TASK_ITER_PROCS
|
4258 CSS_TASK_ITER_THREADED
);
4261 static int cgroup_procs_show(struct seq_file
*s
, void *v
)
4263 seq_printf(s
, "%d\n", task_pid_vnr(v
));
4267 static int cgroup_procs_write_permission(struct cgroup
*src_cgrp
,
4268 struct cgroup
*dst_cgrp
,
4269 struct super_block
*sb
)
4271 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
4272 struct cgroup
*com_cgrp
= src_cgrp
;
4273 struct inode
*inode
;
4276 lockdep_assert_held(&cgroup_mutex
);
4278 /* find the common ancestor */
4279 while (!cgroup_is_descendant(dst_cgrp
, com_cgrp
))
4280 com_cgrp
= cgroup_parent(com_cgrp
);
4282 /* %current should be authorized to migrate to the common ancestor */
4283 inode
= kernfs_get_inode(sb
, com_cgrp
->procs_file
.kn
);
4287 ret
= inode_permission(inode
, MAY_WRITE
);
4293 * If namespaces are delegation boundaries, %current must be able
4294 * to see both source and destination cgroups from its namespace.
4296 if ((cgrp_dfl_root
.flags
& CGRP_ROOT_NS_DELEGATE
) &&
4297 (!cgroup_is_descendant(src_cgrp
, ns
->root_cset
->dfl_cgrp
) ||
4298 !cgroup_is_descendant(dst_cgrp
, ns
->root_cset
->dfl_cgrp
)))
4304 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
4305 char *buf
, size_t nbytes
, loff_t off
)
4307 struct cgroup
*src_cgrp
, *dst_cgrp
;
4308 struct task_struct
*task
;
4311 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4315 task
= cgroup_procs_write_start(buf
, true);
4316 ret
= PTR_ERR_OR_ZERO(task
);
4320 /* find the source cgroup */
4321 spin_lock_irq(&css_set_lock
);
4322 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4323 spin_unlock_irq(&css_set_lock
);
4325 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4326 of
->file
->f_path
.dentry
->d_sb
);
4330 ret
= cgroup_attach_task(dst_cgrp
, task
, true);
4333 cgroup_procs_write_finish(task
);
4335 cgroup_kn_unlock(of
->kn
);
4337 return ret
?: nbytes
;
4340 static void *cgroup_threads_start(struct seq_file
*s
, loff_t
*pos
)
4342 return __cgroup_procs_start(s
, pos
, 0);
4345 static ssize_t
cgroup_threads_write(struct kernfs_open_file
*of
,
4346 char *buf
, size_t nbytes
, loff_t off
)
4348 struct cgroup
*src_cgrp
, *dst_cgrp
;
4349 struct task_struct
*task
;
4352 buf
= strstrip(buf
);
4354 dst_cgrp
= cgroup_kn_lock_live(of
->kn
, false);
4358 task
= cgroup_procs_write_start(buf
, false);
4359 ret
= PTR_ERR_OR_ZERO(task
);
4363 /* find the source cgroup */
4364 spin_lock_irq(&css_set_lock
);
4365 src_cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
4366 spin_unlock_irq(&css_set_lock
);
4368 /* thread migrations follow the cgroup.procs delegation rule */
4369 ret
= cgroup_procs_write_permission(src_cgrp
, dst_cgrp
,
4370 of
->file
->f_path
.dentry
->d_sb
);
4374 /* and must be contained in the same domain */
4376 if (src_cgrp
->dom_cgrp
!= dst_cgrp
->dom_cgrp
)
4379 ret
= cgroup_attach_task(dst_cgrp
, task
, false);
4382 cgroup_procs_write_finish(task
);
4384 cgroup_kn_unlock(of
->kn
);
4386 return ret
?: nbytes
;
4389 /* cgroup core interface files for the default hierarchy */
4390 static struct cftype cgroup_base_files
[] = {
4392 .name
= "cgroup.type",
4393 .flags
= CFTYPE_NOT_ON_ROOT
,
4394 .seq_show
= cgroup_type_show
,
4395 .write
= cgroup_type_write
,
4398 .name
= "cgroup.procs",
4399 .flags
= CFTYPE_NS_DELEGATABLE
,
4400 .file_offset
= offsetof(struct cgroup
, procs_file
),
4401 .release
= cgroup_procs_release
,
4402 .seq_start
= cgroup_procs_start
,
4403 .seq_next
= cgroup_procs_next
,
4404 .seq_show
= cgroup_procs_show
,
4405 .write
= cgroup_procs_write
,
4408 .name
= "cgroup.threads",
4409 .release
= cgroup_procs_release
,
4410 .seq_start
= cgroup_threads_start
,
4411 .seq_next
= cgroup_procs_next
,
4412 .seq_show
= cgroup_procs_show
,
4413 .write
= cgroup_threads_write
,
4416 .name
= "cgroup.controllers",
4417 .seq_show
= cgroup_controllers_show
,
4420 .name
= "cgroup.subtree_control",
4421 .flags
= CFTYPE_NS_DELEGATABLE
,
4422 .seq_show
= cgroup_subtree_control_show
,
4423 .write
= cgroup_subtree_control_write
,
4426 .name
= "cgroup.events",
4427 .flags
= CFTYPE_NOT_ON_ROOT
,
4428 .file_offset
= offsetof(struct cgroup
, events_file
),
4429 .seq_show
= cgroup_events_show
,
4432 .name
= "cgroup.max.descendants",
4433 .seq_show
= cgroup_max_descendants_show
,
4434 .write
= cgroup_max_descendants_write
,
4437 .name
= "cgroup.max.depth",
4438 .seq_show
= cgroup_max_depth_show
,
4439 .write
= cgroup_max_depth_write
,
4442 .name
= "cgroup.stat",
4443 .seq_show
= cgroup_stat_show
,
4449 * css destruction is four-stage process.
4451 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4452 * Implemented in kill_css().
4454 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4455 * and thus css_tryget_online() is guaranteed to fail, the css can be
4456 * offlined by invoking offline_css(). After offlining, the base ref is
4457 * put. Implemented in css_killed_work_fn().
4459 * 3. When the percpu_ref reaches zero, the only possible remaining
4460 * accessors are inside RCU read sections. css_release() schedules the
4463 * 4. After the grace period, the css can be freed. Implemented in
4464 * css_free_work_fn().
4466 * It is actually hairier because both step 2 and 4 require process context
4467 * and thus involve punting to css->destroy_work adding two additional
4468 * steps to the already complex sequence.
4470 static void css_free_work_fn(struct work_struct
*work
)
4472 struct cgroup_subsys_state
*css
=
4473 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4474 struct cgroup_subsys
*ss
= css
->ss
;
4475 struct cgroup
*cgrp
= css
->cgroup
;
4477 percpu_ref_exit(&css
->refcnt
);
4481 struct cgroup_subsys_state
*parent
= css
->parent
;
4485 cgroup_idr_remove(&ss
->css_idr
, id
);
4491 /* cgroup free path */
4492 atomic_dec(&cgrp
->root
->nr_cgrps
);
4493 cgroup1_pidlist_destroy_all(cgrp
);
4494 cancel_work_sync(&cgrp
->release_agent_work
);
4496 if (cgroup_parent(cgrp
)) {
4498 * We get a ref to the parent, and put the ref when
4499 * this cgroup is being freed, so it's guaranteed
4500 * that the parent won't be destroyed before its
4503 cgroup_put(cgroup_parent(cgrp
));
4504 kernfs_put(cgrp
->kn
);
4508 * This is root cgroup's refcnt reaching zero,
4509 * which indicates that the root should be
4512 cgroup_destroy_root(cgrp
->root
);
4517 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4519 struct cgroup_subsys_state
*css
=
4520 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4522 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4523 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4526 static void css_release_work_fn(struct work_struct
*work
)
4528 struct cgroup_subsys_state
*css
=
4529 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4530 struct cgroup_subsys
*ss
= css
->ss
;
4531 struct cgroup
*cgrp
= css
->cgroup
;
4533 mutex_lock(&cgroup_mutex
);
4535 css
->flags
|= CSS_RELEASED
;
4536 list_del_rcu(&css
->sibling
);
4539 /* css release path */
4540 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4541 if (ss
->css_released
)
4542 ss
->css_released(css
);
4544 struct cgroup
*tcgrp
;
4546 /* cgroup release path */
4547 trace_cgroup_release(cgrp
);
4549 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
;
4550 tcgrp
= cgroup_parent(tcgrp
))
4551 tcgrp
->nr_dying_descendants
--;
4553 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4557 * There are two control paths which try to determine
4558 * cgroup from dentry without going through kernfs -
4559 * cgroupstats_build() and css_tryget_online_from_dir().
4560 * Those are supported by RCU protecting clearing of
4561 * cgrp->kn->priv backpointer.
4564 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
,
4567 cgroup_bpf_put(cgrp
);
4570 mutex_unlock(&cgroup_mutex
);
4572 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4575 static void css_release(struct percpu_ref
*ref
)
4577 struct cgroup_subsys_state
*css
=
4578 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4580 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4581 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4584 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4585 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4587 lockdep_assert_held(&cgroup_mutex
);
4589 cgroup_get_live(cgrp
);
4591 memset(css
, 0, sizeof(*css
));
4595 INIT_LIST_HEAD(&css
->sibling
);
4596 INIT_LIST_HEAD(&css
->children
);
4597 css
->serial_nr
= css_serial_nr_next
++;
4598 atomic_set(&css
->online_cnt
, 0);
4600 if (cgroup_parent(cgrp
)) {
4601 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4602 css_get(css
->parent
);
4605 BUG_ON(cgroup_css(cgrp
, ss
));
4608 /* invoke ->css_online() on a new CSS and mark it online if successful */
4609 static int online_css(struct cgroup_subsys_state
*css
)
4611 struct cgroup_subsys
*ss
= css
->ss
;
4614 lockdep_assert_held(&cgroup_mutex
);
4617 ret
= ss
->css_online(css
);
4619 css
->flags
|= CSS_ONLINE
;
4620 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4622 atomic_inc(&css
->online_cnt
);
4624 atomic_inc(&css
->parent
->online_cnt
);
4629 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4630 static void offline_css(struct cgroup_subsys_state
*css
)
4632 struct cgroup_subsys
*ss
= css
->ss
;
4634 lockdep_assert_held(&cgroup_mutex
);
4636 if (!(css
->flags
& CSS_ONLINE
))
4639 if (ss
->css_offline
)
4640 ss
->css_offline(css
);
4642 css
->flags
&= ~CSS_ONLINE
;
4643 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4645 wake_up_all(&css
->cgroup
->offline_waitq
);
4649 * css_create - create a cgroup_subsys_state
4650 * @cgrp: the cgroup new css will be associated with
4651 * @ss: the subsys of new css
4653 * Create a new css associated with @cgrp - @ss pair. On success, the new
4654 * css is online and installed in @cgrp. This function doesn't create the
4655 * interface files. Returns 0 on success, -errno on failure.
4657 static struct cgroup_subsys_state
*css_create(struct cgroup
*cgrp
,
4658 struct cgroup_subsys
*ss
)
4660 struct cgroup
*parent
= cgroup_parent(cgrp
);
4661 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4662 struct cgroup_subsys_state
*css
;
4665 lockdep_assert_held(&cgroup_mutex
);
4667 css
= ss
->css_alloc(parent_css
);
4669 css
= ERR_PTR(-ENOMEM
);
4673 init_and_link_css(css
, ss
, cgrp
);
4675 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4679 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4684 /* @css is ready to be brought online now, make it visible */
4685 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4686 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4688 err
= online_css(css
);
4692 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4693 cgroup_parent(parent
)) {
4694 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4695 current
->comm
, current
->pid
, ss
->name
);
4696 if (!strcmp(ss
->name
, "memory"))
4697 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4698 ss
->warned_broken_hierarchy
= true;
4704 list_del_rcu(&css
->sibling
);
4706 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4707 return ERR_PTR(err
);
4711 * The returned cgroup is fully initialized including its control mask, but
4712 * it isn't associated with its kernfs_node and doesn't have the control
4715 static struct cgroup
*cgroup_create(struct cgroup
*parent
)
4717 struct cgroup_root
*root
= parent
->root
;
4718 struct cgroup
*cgrp
, *tcgrp
;
4719 int level
= parent
->level
+ 1;
4722 /* allocate the cgroup and its ID, 0 is reserved for the root */
4723 cgrp
= kzalloc(sizeof(*cgrp
) +
4724 sizeof(cgrp
->ancestor_ids
[0]) * (level
+ 1), GFP_KERNEL
);
4726 return ERR_PTR(-ENOMEM
);
4728 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
4733 * Temporarily set the pointer to NULL, so idr_find() won't return
4734 * a half-baked cgroup.
4736 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
4739 goto out_cancel_ref
;
4742 init_cgroup_housekeeping(cgrp
);
4744 cgrp
->self
.parent
= &parent
->self
;
4746 cgrp
->level
= level
;
4748 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
4749 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
4752 tcgrp
->nr_descendants
++;
4755 if (notify_on_release(parent
))
4756 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4758 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4759 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4761 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4763 /* allocation complete, commit to creation */
4764 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4765 atomic_inc(&root
->nr_cgrps
);
4766 cgroup_get_live(parent
);
4769 * @cgrp is now fully operational. If something fails after this
4770 * point, it'll be released via the normal destruction path.
4772 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4775 * On the default hierarchy, a child doesn't automatically inherit
4776 * subtree_control from the parent. Each is configured manually.
4778 if (!cgroup_on_dfl(cgrp
))
4779 cgrp
->subtree_control
= cgroup_control(cgrp
);
4782 cgroup_bpf_inherit(cgrp
, parent
);
4784 cgroup_propagate_control(cgrp
);
4789 percpu_ref_exit(&cgrp
->self
.refcnt
);
4792 return ERR_PTR(ret
);
4795 static bool cgroup_check_hierarchy_limits(struct cgroup
*parent
)
4797 struct cgroup
*cgroup
;
4801 lockdep_assert_held(&cgroup_mutex
);
4803 for (cgroup
= parent
; cgroup
; cgroup
= cgroup_parent(cgroup
)) {
4804 if (cgroup
->nr_descendants
>= cgroup
->max_descendants
)
4807 if (level
> cgroup
->max_depth
)
4818 int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
, umode_t mode
)
4820 struct cgroup
*parent
, *cgrp
;
4821 struct kernfs_node
*kn
;
4824 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4825 if (strchr(name
, '\n'))
4828 parent
= cgroup_kn_lock_live(parent_kn
, false);
4832 if (!cgroup_check_hierarchy_limits(parent
)) {
4837 cgrp
= cgroup_create(parent
);
4839 ret
= PTR_ERR(cgrp
);
4843 /* create the directory */
4844 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4852 * This extra ref will be put in cgroup_free_fn() and guarantees
4853 * that @cgrp->kn is always accessible.
4857 ret
= cgroup_kn_set_ugid(kn
);
4861 ret
= css_populate_dir(&cgrp
->self
);
4865 ret
= cgroup_apply_control_enable(cgrp
);
4869 trace_cgroup_mkdir(cgrp
);
4871 /* let's create and online css's */
4872 kernfs_activate(kn
);
4878 cgroup_destroy_locked(cgrp
);
4880 cgroup_kn_unlock(parent_kn
);
4885 * This is called when the refcnt of a css is confirmed to be killed.
4886 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4887 * initate destruction and put the css ref from kill_css().
4889 static void css_killed_work_fn(struct work_struct
*work
)
4891 struct cgroup_subsys_state
*css
=
4892 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4894 mutex_lock(&cgroup_mutex
);
4899 /* @css can't go away while we're holding cgroup_mutex */
4901 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
4903 mutex_unlock(&cgroup_mutex
);
4906 /* css kill confirmation processing requires process context, bounce */
4907 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4909 struct cgroup_subsys_state
*css
=
4910 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4912 if (atomic_dec_and_test(&css
->online_cnt
)) {
4913 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4914 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4919 * kill_css - destroy a css
4920 * @css: css to destroy
4922 * This function initiates destruction of @css by removing cgroup interface
4923 * files and putting its base reference. ->css_offline() will be invoked
4924 * asynchronously once css_tryget_online() is guaranteed to fail and when
4925 * the reference count reaches zero, @css will be released.
4927 static void kill_css(struct cgroup_subsys_state
*css
)
4929 lockdep_assert_held(&cgroup_mutex
);
4931 if (css
->flags
& CSS_DYING
)
4934 css
->flags
|= CSS_DYING
;
4937 * This must happen before css is disassociated with its cgroup.
4938 * See seq_css() for details.
4943 * Killing would put the base ref, but we need to keep it alive
4944 * until after ->css_offline().
4949 * cgroup core guarantees that, by the time ->css_offline() is
4950 * invoked, no new css reference will be given out via
4951 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4952 * proceed to offlining css's because percpu_ref_kill() doesn't
4953 * guarantee that the ref is seen as killed on all CPUs on return.
4955 * Use percpu_ref_kill_and_confirm() to get notifications as each
4956 * css is confirmed to be seen as killed on all CPUs.
4958 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4962 * cgroup_destroy_locked - the first stage of cgroup destruction
4963 * @cgrp: cgroup to be destroyed
4965 * css's make use of percpu refcnts whose killing latency shouldn't be
4966 * exposed to userland and are RCU protected. Also, cgroup core needs to
4967 * guarantee that css_tryget_online() won't succeed by the time
4968 * ->css_offline() is invoked. To satisfy all the requirements,
4969 * destruction is implemented in the following two steps.
4971 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4972 * userland visible parts and start killing the percpu refcnts of
4973 * css's. Set up so that the next stage will be kicked off once all
4974 * the percpu refcnts are confirmed to be killed.
4976 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4977 * rest of destruction. Once all cgroup references are gone, the
4978 * cgroup is RCU-freed.
4980 * This function implements s1. After this step, @cgrp is gone as far as
4981 * the userland is concerned and a new cgroup with the same name may be
4982 * created. As cgroup doesn't care about the names internally, this
4983 * doesn't cause any problem.
4985 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4986 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4988 struct cgroup
*tcgrp
, *parent
= cgroup_parent(cgrp
);
4989 struct cgroup_subsys_state
*css
;
4990 struct cgrp_cset_link
*link
;
4993 lockdep_assert_held(&cgroup_mutex
);
4996 * Only migration can raise populated from zero and we're already
4997 * holding cgroup_mutex.
4999 if (cgroup_is_populated(cgrp
))
5003 * Make sure there's no live children. We can't test emptiness of
5004 * ->self.children as dead children linger on it while being
5005 * drained; otherwise, "rmdir parent/child parent" may fail.
5007 if (css_has_online_children(&cgrp
->self
))
5011 * Mark @cgrp and the associated csets dead. The former prevents
5012 * further task migration and child creation by disabling
5013 * cgroup_lock_live_group(). The latter makes the csets ignored by
5014 * the migration path.
5016 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5018 spin_lock_irq(&css_set_lock
);
5019 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
5020 link
->cset
->dead
= true;
5021 spin_unlock_irq(&css_set_lock
);
5023 /* initiate massacre of all css's */
5024 for_each_css(css
, ssid
, cgrp
)
5028 * Remove @cgrp directory along with the base files. @cgrp has an
5029 * extra ref on its kn.
5031 kernfs_remove(cgrp
->kn
);
5033 if (parent
&& cgroup_is_threaded(cgrp
))
5034 parent
->nr_threaded_children
--;
5036 for (tcgrp
= cgroup_parent(cgrp
); tcgrp
; tcgrp
= cgroup_parent(tcgrp
)) {
5037 tcgrp
->nr_descendants
--;
5038 tcgrp
->nr_dying_descendants
++;
5041 cgroup1_check_for_release(parent
);
5043 /* put the base reference */
5044 percpu_ref_kill(&cgrp
->self
.refcnt
);
5049 int cgroup_rmdir(struct kernfs_node
*kn
)
5051 struct cgroup
*cgrp
;
5054 cgrp
= cgroup_kn_lock_live(kn
, false);
5058 ret
= cgroup_destroy_locked(cgrp
);
5061 trace_cgroup_rmdir(cgrp
);
5063 cgroup_kn_unlock(kn
);
5067 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5068 .show_options
= cgroup_show_options
,
5069 .remount_fs
= cgroup_remount
,
5070 .mkdir
= cgroup_mkdir
,
5071 .rmdir
= cgroup_rmdir
,
5072 .show_path
= cgroup_show_path
,
5075 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5077 struct cgroup_subsys_state
*css
;
5079 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5081 mutex_lock(&cgroup_mutex
);
5083 idr_init(&ss
->css_idr
);
5084 INIT_LIST_HEAD(&ss
->cfts
);
5086 /* Create the root cgroup state for this subsystem */
5087 ss
->root
= &cgrp_dfl_root
;
5088 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5089 /* We don't handle early failures gracefully */
5090 BUG_ON(IS_ERR(css
));
5091 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5094 * Root csses are never destroyed and we can't initialize
5095 * percpu_ref during early init. Disable refcnting.
5097 css
->flags
|= CSS_NO_REF
;
5100 /* allocation can't be done safely during early init */
5103 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5104 BUG_ON(css
->id
< 0);
5107 /* Update the init_css_set to contain a subsys
5108 * pointer to this state - since the subsystem is
5109 * newly registered, all tasks and hence the
5110 * init_css_set is in the subsystem's root cgroup. */
5111 init_css_set
.subsys
[ss
->id
] = css
;
5113 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5114 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5115 have_release_callback
|= (bool)ss
->release
<< ss
->id
;
5116 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5118 /* At system boot, before all subsystems have been
5119 * registered, no tasks have been forked, so we don't
5120 * need to invoke fork callbacks here. */
5121 BUG_ON(!list_empty(&init_task
.tasks
));
5123 BUG_ON(online_css(css
));
5125 mutex_unlock(&cgroup_mutex
);
5129 * cgroup_init_early - cgroup initialization at system boot
5131 * Initialize cgroups at system boot, and initialize any
5132 * subsystems that request early init.
5134 int __init
cgroup_init_early(void)
5136 static struct cgroup_sb_opts __initdata opts
;
5137 struct cgroup_subsys
*ss
;
5140 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5141 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5143 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5145 for_each_subsys(ss
, i
) {
5146 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5147 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5148 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5150 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5151 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5154 ss
->name
= cgroup_subsys_name
[i
];
5155 if (!ss
->legacy_name
)
5156 ss
->legacy_name
= cgroup_subsys_name
[i
];
5159 cgroup_init_subsys(ss
, true);
5164 static u16 cgroup_disable_mask __initdata
;
5167 * cgroup_init - cgroup initialization
5169 * Register cgroup filesystem and /proc file, and initialize
5170 * any subsystems that didn't request early init.
5172 int __init
cgroup_init(void)
5174 struct cgroup_subsys
*ss
;
5177 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT
> 16);
5178 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5179 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
5180 BUG_ON(cgroup_init_cftypes(NULL
, cgroup1_base_files
));
5183 * The latency of the synchronize_sched() is too high for cgroups,
5184 * avoid it at the cost of forcing all readers into the slow path.
5186 rcu_sync_enter_start(&cgroup_threadgroup_rwsem
.rss
);
5188 get_user_ns(init_cgroup_ns
.user_ns
);
5190 mutex_lock(&cgroup_mutex
);
5193 * Add init_css_set to the hash table so that dfl_root can link to
5196 hash_add(css_set_table
, &init_css_set
.hlist
,
5197 css_set_hash(init_css_set
.subsys
));
5199 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0, 0));
5201 mutex_unlock(&cgroup_mutex
);
5203 for_each_subsys(ss
, ssid
) {
5204 if (ss
->early_init
) {
5205 struct cgroup_subsys_state
*css
=
5206 init_css_set
.subsys
[ss
->id
];
5208 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5210 BUG_ON(css
->id
< 0);
5212 cgroup_init_subsys(ss
, false);
5215 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5216 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5219 * Setting dfl_root subsys_mask needs to consider the
5220 * disabled flag and cftype registration needs kmalloc,
5221 * both of which aren't available during early_init.
5223 if (cgroup_disable_mask
& (1 << ssid
)) {
5224 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5225 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5230 if (cgroup1_ssid_disabled(ssid
))
5231 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5234 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5236 /* implicit controllers must be threaded too */
5237 WARN_ON(ss
->implicit_on_dfl
&& !ss
->threaded
);
5239 if (ss
->implicit_on_dfl
)
5240 cgrp_dfl_implicit_ss_mask
|= 1 << ss
->id
;
5241 else if (!ss
->dfl_cftypes
)
5242 cgrp_dfl_inhibit_ss_mask
|= 1 << ss
->id
;
5245 cgrp_dfl_threaded_ss_mask
|= 1 << ss
->id
;
5247 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5248 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5250 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5251 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5255 ss
->bind(init_css_set
.subsys
[ssid
]);
5257 mutex_lock(&cgroup_mutex
);
5258 css_populate_dir(init_css_set
.subsys
[ssid
]);
5259 mutex_unlock(&cgroup_mutex
);
5262 /* init_css_set.subsys[] has been updated, re-hash */
5263 hash_del(&init_css_set
.hlist
);
5264 hash_add(css_set_table
, &init_css_set
.hlist
,
5265 css_set_hash(init_css_set
.subsys
));
5267 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5268 WARN_ON(register_filesystem(&cgroup_fs_type
));
5269 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5270 WARN_ON(!proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
));
5275 static int __init
cgroup_wq_init(void)
5278 * There isn't much point in executing destruction path in
5279 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5280 * Use 1 for @max_active.
5282 * We would prefer to do this in cgroup_init() above, but that
5283 * is called before init_workqueues(): so leave this until after.
5285 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5286 BUG_ON(!cgroup_destroy_wq
);
5289 core_initcall(cgroup_wq_init
);
5291 void cgroup_path_from_kernfs_id(const union kernfs_node_id
*id
,
5292 char *buf
, size_t buflen
)
5294 struct kernfs_node
*kn
;
5296 kn
= kernfs_get_node_by_id(cgrp_dfl_root
.kf_root
, id
);
5299 kernfs_path(kn
, buf
, buflen
);
5304 * proc_cgroup_show()
5305 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5306 * - Used for /proc/<pid>/cgroup.
5308 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5309 struct pid
*pid
, struct task_struct
*tsk
)
5313 struct cgroup_root
*root
;
5316 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5320 mutex_lock(&cgroup_mutex
);
5321 spin_lock_irq(&css_set_lock
);
5323 for_each_root(root
) {
5324 struct cgroup_subsys
*ss
;
5325 struct cgroup
*cgrp
;
5326 int ssid
, count
= 0;
5328 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_visible
)
5331 seq_printf(m
, "%d:", root
->hierarchy_id
);
5332 if (root
!= &cgrp_dfl_root
)
5333 for_each_subsys(ss
, ssid
)
5334 if (root
->subsys_mask
& (1 << ssid
))
5335 seq_printf(m
, "%s%s", count
++ ? "," : "",
5337 if (strlen(root
->name
))
5338 seq_printf(m
, "%sname=%s", count
? "," : "",
5342 cgrp
= task_cgroup_from_root(tsk
, root
);
5345 * On traditional hierarchies, all zombie tasks show up as
5346 * belonging to the root cgroup. On the default hierarchy,
5347 * while a zombie doesn't show up in "cgroup.procs" and
5348 * thus can't be migrated, its /proc/PID/cgroup keeps
5349 * reporting the cgroup it belonged to before exiting. If
5350 * the cgroup is removed before the zombie is reaped,
5351 * " (deleted)" is appended to the cgroup path.
5353 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5354 retval
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5355 current
->nsproxy
->cgroup_ns
);
5356 if (retval
>= PATH_MAX
)
5357 retval
= -ENAMETOOLONG
;
5366 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5367 seq_puts(m
, " (deleted)\n");
5374 spin_unlock_irq(&css_set_lock
);
5375 mutex_unlock(&cgroup_mutex
);
5382 * cgroup_fork - initialize cgroup related fields during copy_process()
5383 * @child: pointer to task_struct of forking parent process.
5385 * A task is associated with the init_css_set until cgroup_post_fork()
5386 * attaches it to the parent's css_set. Empty cg_list indicates that
5387 * @child isn't holding reference to its css_set.
5389 void cgroup_fork(struct task_struct
*child
)
5391 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5392 INIT_LIST_HEAD(&child
->cg_list
);
5396 * cgroup_can_fork - called on a new task before the process is exposed
5397 * @child: the task in question.
5399 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5400 * returns an error, the fork aborts with that error code. This allows for
5401 * a cgroup subsystem to conditionally allow or deny new forks.
5403 int cgroup_can_fork(struct task_struct
*child
)
5405 struct cgroup_subsys
*ss
;
5408 do_each_subsys_mask(ss
, i
, have_canfork_callback
) {
5409 ret
= ss
->can_fork(child
);
5412 } while_each_subsys_mask();
5417 for_each_subsys(ss
, j
) {
5420 if (ss
->cancel_fork
)
5421 ss
->cancel_fork(child
);
5428 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5429 * @child: the task in question
5431 * This calls the cancel_fork() callbacks if a fork failed *after*
5432 * cgroup_can_fork() succeded.
5434 void cgroup_cancel_fork(struct task_struct
*child
)
5436 struct cgroup_subsys
*ss
;
5439 for_each_subsys(ss
, i
)
5440 if (ss
->cancel_fork
)
5441 ss
->cancel_fork(child
);
5445 * cgroup_post_fork - called on a new task after adding it to the task list
5446 * @child: the task in question
5448 * Adds the task to the list running through its css_set if necessary and
5449 * call the subsystem fork() callbacks. Has to be after the task is
5450 * visible on the task list in case we race with the first call to
5451 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5454 void cgroup_post_fork(struct task_struct
*child
)
5456 struct cgroup_subsys
*ss
;
5460 * This may race against cgroup_enable_task_cg_lists(). As that
5461 * function sets use_task_css_set_links before grabbing
5462 * tasklist_lock and we just went through tasklist_lock to add
5463 * @child, it's guaranteed that either we see the set
5464 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5465 * @child during its iteration.
5467 * If we won the race, @child is associated with %current's
5468 * css_set. Grabbing css_set_lock guarantees both that the
5469 * association is stable, and, on completion of the parent's
5470 * migration, @child is visible in the source of migration or
5471 * already in the destination cgroup. This guarantee is necessary
5472 * when implementing operations which need to migrate all tasks of
5473 * a cgroup to another.
5475 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5476 * will remain in init_css_set. This is safe because all tasks are
5477 * in the init_css_set before cg_links is enabled and there's no
5478 * operation which transfers all tasks out of init_css_set.
5480 if (use_task_css_set_links
) {
5481 struct css_set
*cset
;
5483 spin_lock_irq(&css_set_lock
);
5484 cset
= task_css_set(current
);
5485 if (list_empty(&child
->cg_list
)) {
5488 css_set_move_task(child
, NULL
, cset
, false);
5490 spin_unlock_irq(&css_set_lock
);
5494 * Call ss->fork(). This must happen after @child is linked on
5495 * css_set; otherwise, @child might change state between ->fork()
5496 * and addition to css_set.
5498 do_each_subsys_mask(ss
, i
, have_fork_callback
) {
5500 } while_each_subsys_mask();
5504 * cgroup_exit - detach cgroup from exiting task
5505 * @tsk: pointer to task_struct of exiting process
5507 * Description: Detach cgroup from @tsk and release it.
5509 * Note that cgroups marked notify_on_release force every task in
5510 * them to take the global cgroup_mutex mutex when exiting.
5511 * This could impact scaling on very large systems. Be reluctant to
5512 * use notify_on_release cgroups where very high task exit scaling
5513 * is required on large systems.
5515 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5516 * call cgroup_exit() while the task is still competent to handle
5517 * notify_on_release(), then leave the task attached to the root cgroup in
5518 * each hierarchy for the remainder of its exit. No need to bother with
5519 * init_css_set refcnting. init_css_set never goes away and we can't race
5520 * with migration path - PF_EXITING is visible to migration path.
5522 void cgroup_exit(struct task_struct
*tsk
)
5524 struct cgroup_subsys
*ss
;
5525 struct css_set
*cset
;
5529 * Unlink from @tsk from its css_set. As migration path can't race
5530 * with us, we can check css_set and cg_list without synchronization.
5532 cset
= task_css_set(tsk
);
5534 if (!list_empty(&tsk
->cg_list
)) {
5535 spin_lock_irq(&css_set_lock
);
5536 css_set_move_task(tsk
, cset
, NULL
, false);
5538 spin_unlock_irq(&css_set_lock
);
5543 /* see cgroup_post_fork() for details */
5544 do_each_subsys_mask(ss
, i
, have_exit_callback
) {
5546 } while_each_subsys_mask();
5549 void cgroup_release(struct task_struct
*task
)
5551 struct cgroup_subsys
*ss
;
5554 do_each_subsys_mask(ss
, ssid
, have_release_callback
) {
5556 } while_each_subsys_mask();
5559 void cgroup_free(struct task_struct
*task
)
5561 struct css_set
*cset
= task_css_set(task
);
5565 static int __init
cgroup_disable(char *str
)
5567 struct cgroup_subsys
*ss
;
5571 while ((token
= strsep(&str
, ",")) != NULL
) {
5575 for_each_subsys(ss
, i
) {
5576 if (strcmp(token
, ss
->name
) &&
5577 strcmp(token
, ss
->legacy_name
))
5579 cgroup_disable_mask
|= 1 << i
;
5584 __setup("cgroup_disable=", cgroup_disable
);
5587 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5588 * @dentry: directory dentry of interest
5589 * @ss: subsystem of interest
5591 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5592 * to get the corresponding css and return it. If such css doesn't exist
5593 * or can't be pinned, an ERR_PTR value is returned.
5595 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5596 struct cgroup_subsys
*ss
)
5598 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5599 struct file_system_type
*s_type
= dentry
->d_sb
->s_type
;
5600 struct cgroup_subsys_state
*css
= NULL
;
5601 struct cgroup
*cgrp
;
5603 /* is @dentry a cgroup dir? */
5604 if ((s_type
!= &cgroup_fs_type
&& s_type
!= &cgroup2_fs_type
) ||
5605 !kn
|| kernfs_type(kn
) != KERNFS_DIR
)
5606 return ERR_PTR(-EBADF
);
5611 * This path doesn't originate from kernfs and @kn could already
5612 * have been or be removed at any point. @kn->priv is RCU
5613 * protected for this access. See css_release_work_fn() for details.
5615 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
5617 css
= cgroup_css(cgrp
, ss
);
5619 if (!css
|| !css_tryget_online(css
))
5620 css
= ERR_PTR(-ENOENT
);
5627 * css_from_id - lookup css by id
5628 * @id: the cgroup id
5629 * @ss: cgroup subsys to be looked into
5631 * Returns the css if there's valid one with @id, otherwise returns NULL.
5632 * Should be called under rcu_read_lock().
5634 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5636 WARN_ON_ONCE(!rcu_read_lock_held());
5637 return idr_find(&ss
->css_idr
, id
);
5641 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5642 * @path: path on the default hierarchy
5644 * Find the cgroup at @path on the default hierarchy, increment its
5645 * reference count and return it. Returns pointer to the found cgroup on
5646 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5647 * if @path points to a non-directory.
5649 struct cgroup
*cgroup_get_from_path(const char *path
)
5651 struct kernfs_node
*kn
;
5652 struct cgroup
*cgrp
;
5654 mutex_lock(&cgroup_mutex
);
5656 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
5658 if (kernfs_type(kn
) == KERNFS_DIR
) {
5660 cgroup_get_live(cgrp
);
5662 cgrp
= ERR_PTR(-ENOTDIR
);
5666 cgrp
= ERR_PTR(-ENOENT
);
5669 mutex_unlock(&cgroup_mutex
);
5672 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
5675 * cgroup_get_from_fd - get a cgroup pointer from a fd
5676 * @fd: fd obtained by open(cgroup2_dir)
5678 * Find the cgroup from a fd which should be obtained
5679 * by opening a cgroup directory. Returns a pointer to the
5680 * cgroup on success. ERR_PTR is returned if the cgroup
5683 struct cgroup
*cgroup_get_from_fd(int fd
)
5685 struct cgroup_subsys_state
*css
;
5686 struct cgroup
*cgrp
;
5691 return ERR_PTR(-EBADF
);
5693 css
= css_tryget_online_from_dir(f
->f_path
.dentry
, NULL
);
5696 return ERR_CAST(css
);
5699 if (!cgroup_on_dfl(cgrp
)) {
5701 return ERR_PTR(-EBADF
);
5706 EXPORT_SYMBOL_GPL(cgroup_get_from_fd
);
5709 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5710 * definition in cgroup-defs.h.
5712 #ifdef CONFIG_SOCK_CGROUP_DATA
5714 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5716 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
5717 static bool cgroup_sk_alloc_disabled __read_mostly
;
5719 void cgroup_sk_alloc_disable(void)
5721 if (cgroup_sk_alloc_disabled
)
5723 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5724 cgroup_sk_alloc_disabled
= true;
5729 #define cgroup_sk_alloc_disabled false
5733 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
5735 if (cgroup_sk_alloc_disabled
)
5738 /* Socket clone path */
5741 * We might be cloning a socket which is left in an empty
5742 * cgroup and the cgroup might have already been rmdir'd.
5743 * Don't use cgroup_get_live().
5745 cgroup_get(sock_cgroup_ptr(skcd
));
5752 struct css_set
*cset
;
5754 cset
= task_css_set(current
);
5755 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
5756 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
5765 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
5767 cgroup_put(sock_cgroup_ptr(skcd
));
5770 #endif /* CONFIG_SOCK_CGROUP_DATA */
5772 #ifdef CONFIG_CGROUP_BPF
5773 int cgroup_bpf_update(struct cgroup
*cgrp
, struct bpf_prog
*prog
,
5774 enum bpf_attach_type type
, bool overridable
)
5776 struct cgroup
*parent
= cgroup_parent(cgrp
);
5779 mutex_lock(&cgroup_mutex
);
5780 ret
= __cgroup_bpf_update(cgrp
, parent
, prog
, type
, overridable
);
5781 mutex_unlock(&cgroup_mutex
);
5784 #endif /* CONFIG_CGROUP_BPF */