2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/rwsem.h>
49 #include <linux/percpu-rwsem.h>
50 #include <linux/string.h>
51 #include <linux/sort.h>
52 #include <linux/kmod.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hashtable.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/kthread.h>
60 #include <linux/delay.h>
62 #include <linux/atomic.h>
65 * pidlists linger the following amount before being destroyed. The goal
66 * is avoiding frequent destruction in the middle of consecutive read calls
67 * Expiring in the middle is a performance problem not a correctness one.
68 * 1 sec should be enough.
70 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
72 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
76 * cgroup_mutex is the master lock. Any modification to cgroup or its
77 * hierarchy must be performed while holding it.
79 * css_set_rwsem protects task->cgroups pointer, the list of css_set
80 * objects, and the chain of tasks off each css_set.
82 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
83 * cgroup.h can use them for lockdep annotations.
85 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex
);
87 DECLARE_RWSEM(css_set_rwsem
);
88 EXPORT_SYMBOL_GPL(cgroup_mutex
);
89 EXPORT_SYMBOL_GPL(css_set_rwsem
);
91 static DEFINE_MUTEX(cgroup_mutex
);
92 static DECLARE_RWSEM(css_set_rwsem
);
96 * Protects cgroup_idr and css_idr so that IDs can be released without
97 * grabbing cgroup_mutex.
99 static DEFINE_SPINLOCK(cgroup_idr_lock
);
102 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
103 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
105 static DEFINE_SPINLOCK(release_agent_path_lock
);
107 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
109 #define cgroup_assert_mutex_or_rcu_locked() \
110 rcu_lockdep_assert(rcu_read_lock_held() || \
111 lockdep_is_held(&cgroup_mutex), \
112 "cgroup_mutex or RCU read lock required");
115 * cgroup destruction makes heavy use of work items and there can be a lot
116 * of concurrent destructions. Use a separate workqueue so that cgroup
117 * destruction work items don't end up filling up max_active of system_wq
118 * which may lead to deadlock.
120 static struct workqueue_struct
*cgroup_destroy_wq
;
123 * pidlist destructions need to be flushed on cgroup destruction. Use a
124 * separate workqueue as flush domain.
126 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
128 /* generate an array of cgroup subsystem pointers */
129 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
130 static struct cgroup_subsys
*cgroup_subsys
[] = {
131 #include <linux/cgroup_subsys.h>
135 /* array of cgroup subsystem names */
136 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
137 static const char *cgroup_subsys_name
[] = {
138 #include <linux/cgroup_subsys.h>
143 * The default hierarchy, reserved for the subsystems that are otherwise
144 * unattached - it never has more than a single cgroup, and all tasks are
145 * part of that cgroup.
147 struct cgroup_root cgrp_dfl_root
;
148 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
151 * The default hierarchy always exists but is hidden until mounted for the
152 * first time. This is for backward compatibility.
154 static bool cgrp_dfl_root_visible
;
157 * Set by the boot param of the same name and makes subsystems with NULL
158 * ->dfl_files to use ->legacy_files on the default hierarchy.
160 static bool cgroup_legacy_files_on_dfl
;
162 /* some controllers are not supported in the default hierarchy */
163 static unsigned long cgrp_dfl_root_inhibit_ss_mask
;
165 /* The list of hierarchy roots */
167 static LIST_HEAD(cgroup_roots
);
168 static int cgroup_root_count
;
170 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
171 static DEFINE_IDR(cgroup_hierarchy_idr
);
174 * Assign a monotonically increasing serial number to csses. It guarantees
175 * cgroups with bigger numbers are newer than those with smaller numbers.
176 * Also, as csses are always appended to the parent's ->children list, it
177 * guarantees that sibling csses are always sorted in the ascending serial
178 * number order on the list. Protected by cgroup_mutex.
180 static u64 css_serial_nr_next
= 1;
183 * These bitmask flags indicate whether tasks in the fork and exit paths have
184 * fork/exit handlers to call. This avoids us having to do extra work in the
185 * fork/exit path to check which subsystems have fork/exit callbacks.
187 static unsigned long have_fork_callback __read_mostly
;
188 static unsigned long have_exit_callback __read_mostly
;
190 /* Ditto for the can_fork callback. */
191 static unsigned long have_canfork_callback __read_mostly
;
193 static struct cftype cgroup_dfl_base_files
[];
194 static struct cftype cgroup_legacy_base_files
[];
196 static int rebind_subsystems(struct cgroup_root
*dst_root
,
197 unsigned long ss_mask
);
198 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
199 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
201 static void css_release(struct percpu_ref
*ref
);
202 static void kill_css(struct cgroup_subsys_state
*css
);
203 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
206 /* IDR wrappers which synchronize using cgroup_idr_lock */
207 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
212 idr_preload(gfp_mask
);
213 spin_lock_bh(&cgroup_idr_lock
);
214 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_WAIT
);
215 spin_unlock_bh(&cgroup_idr_lock
);
220 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
224 spin_lock_bh(&cgroup_idr_lock
);
225 ret
= idr_replace(idr
, ptr
, id
);
226 spin_unlock_bh(&cgroup_idr_lock
);
230 static void cgroup_idr_remove(struct idr
*idr
, int id
)
232 spin_lock_bh(&cgroup_idr_lock
);
234 spin_unlock_bh(&cgroup_idr_lock
);
237 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
239 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
242 return container_of(parent_css
, struct cgroup
, self
);
247 * cgroup_css - obtain a cgroup's css for the specified subsystem
248 * @cgrp: the cgroup of interest
249 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
251 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
252 * function must be called either under cgroup_mutex or rcu_read_lock() and
253 * the caller is responsible for pinning the returned css if it wants to
254 * keep accessing it outside the said locks. This function may return
255 * %NULL if @cgrp doesn't have @subsys_id enabled.
257 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
258 struct cgroup_subsys
*ss
)
261 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
262 lockdep_is_held(&cgroup_mutex
));
268 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
269 * @cgrp: the cgroup of interest
270 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
272 * Similar to cgroup_css() but returns the effective css, which is defined
273 * as the matching css of the nearest ancestor including self which has @ss
274 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
275 * function is guaranteed to return non-NULL css.
277 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
278 struct cgroup_subsys
*ss
)
280 lockdep_assert_held(&cgroup_mutex
);
285 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
289 * This function is used while updating css associations and thus
290 * can't test the csses directly. Use ->child_subsys_mask.
292 while (cgroup_parent(cgrp
) &&
293 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
294 cgrp
= cgroup_parent(cgrp
);
296 return cgroup_css(cgrp
, ss
);
300 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
301 * @cgrp: the cgroup of interest
302 * @ss: the subsystem of interest
304 * Find and get the effective css of @cgrp for @ss. The effective css is
305 * defined as the matching css of the nearest ancestor including self which
306 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
307 * the root css is returned, so this function always returns a valid css.
308 * The returned css must be put using css_put().
310 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
311 struct cgroup_subsys
*ss
)
313 struct cgroup_subsys_state
*css
;
318 css
= cgroup_css(cgrp
, ss
);
320 if (css
&& css_tryget_online(css
))
322 cgrp
= cgroup_parent(cgrp
);
325 css
= init_css_set
.subsys
[ss
->id
];
332 /* convenient tests for these bits */
333 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
335 return !(cgrp
->self
.flags
& CSS_ONLINE
);
338 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
340 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
341 struct cftype
*cft
= of_cft(of
);
344 * This is open and unprotected implementation of cgroup_css().
345 * seq_css() is only called from a kernfs file operation which has
346 * an active reference on the file. Because all the subsystem
347 * files are drained before a css is disassociated with a cgroup,
348 * the matching css from the cgroup's subsys table is guaranteed to
349 * be and stay valid until the enclosing operation is complete.
352 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
356 EXPORT_SYMBOL_GPL(of_css
);
359 * cgroup_is_descendant - test ancestry
360 * @cgrp: the cgroup to be tested
361 * @ancestor: possible ancestor of @cgrp
363 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
364 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
365 * and @ancestor are accessible.
367 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
370 if (cgrp
== ancestor
)
372 cgrp
= cgroup_parent(cgrp
);
377 static int notify_on_release(const struct cgroup
*cgrp
)
379 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
383 * for_each_css - iterate all css's of a cgroup
384 * @css: the iteration cursor
385 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
386 * @cgrp: the target cgroup to iterate css's of
388 * Should be called under cgroup_[tree_]mutex.
390 #define for_each_css(css, ssid, cgrp) \
391 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
392 if (!((css) = rcu_dereference_check( \
393 (cgrp)->subsys[(ssid)], \
394 lockdep_is_held(&cgroup_mutex)))) { } \
398 * for_each_e_css - iterate all effective css's of a cgroup
399 * @css: the iteration cursor
400 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
401 * @cgrp: the target cgroup to iterate css's of
403 * Should be called under cgroup_[tree_]mutex.
405 #define for_each_e_css(css, ssid, cgrp) \
406 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
407 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
412 * for_each_subsys - iterate all enabled cgroup subsystems
413 * @ss: the iteration cursor
414 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
416 #define for_each_subsys(ss, ssid) \
417 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
418 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
421 * for_each_subsys_which - filter for_each_subsys with a bitmask
422 * @ss: the iteration cursor
423 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
424 * @ss_maskp: a pointer to the bitmask
426 * The block will only run for cases where the ssid-th bit (1 << ssid) of
429 #define for_each_subsys_which(ss, ssid, ss_maskp) \
430 if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \
433 for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \
434 if (((ss) = cgroup_subsys[ssid]) && false) \
438 /* iterate across the hierarchies */
439 #define for_each_root(root) \
440 list_for_each_entry((root), &cgroup_roots, root_list)
442 /* iterate over child cgrps, lock should be held throughout iteration */
443 #define cgroup_for_each_live_child(child, cgrp) \
444 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
445 if (({ lockdep_assert_held(&cgroup_mutex); \
446 cgroup_is_dead(child); })) \
450 static void cgroup_release_agent(struct work_struct
*work
);
451 static void check_for_release(struct cgroup
*cgrp
);
454 * A cgroup can be associated with multiple css_sets as different tasks may
455 * belong to different cgroups on different hierarchies. In the other
456 * direction, a css_set is naturally associated with multiple cgroups.
457 * This M:N relationship is represented by the following link structure
458 * which exists for each association and allows traversing the associations
461 struct cgrp_cset_link
{
462 /* the cgroup and css_set this link associates */
464 struct css_set
*cset
;
466 /* list of cgrp_cset_links anchored at cgrp->cset_links */
467 struct list_head cset_link
;
469 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
470 struct list_head cgrp_link
;
474 * The default css_set - used by init and its children prior to any
475 * hierarchies being mounted. It contains a pointer to the root state
476 * for each subsystem. Also used to anchor the list of css_sets. Not
477 * reference-counted, to improve performance when child cgroups
478 * haven't been created.
480 struct css_set init_css_set
= {
481 .refcount
= ATOMIC_INIT(1),
482 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
483 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
484 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
485 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
486 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
489 static int css_set_count
= 1; /* 1 for init_css_set */
492 * cgroup_update_populated - updated populated count of a cgroup
493 * @cgrp: the target cgroup
494 * @populated: inc or dec populated count
496 * @cgrp is either getting the first task (css_set) or losing the last.
497 * Update @cgrp->populated_cnt accordingly. The count is propagated
498 * towards root so that a given cgroup's populated_cnt is zero iff the
499 * cgroup and all its descendants are empty.
501 * @cgrp's interface file "cgroup.populated" is zero if
502 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
503 * changes from or to zero, userland is notified that the content of the
504 * interface file has changed. This can be used to detect when @cgrp and
505 * its descendants become populated or empty.
507 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
509 lockdep_assert_held(&css_set_rwsem
);
515 trigger
= !cgrp
->populated_cnt
++;
517 trigger
= !--cgrp
->populated_cnt
;
522 if (cgrp
->populated_kn
)
523 kernfs_notify(cgrp
->populated_kn
);
524 cgrp
= cgroup_parent(cgrp
);
529 * hash table for cgroup groups. This improves the performance to find
530 * an existing css_set. This hash doesn't (currently) take into
531 * account cgroups in empty hierarchies.
533 #define CSS_SET_HASH_BITS 7
534 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
536 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
538 unsigned long key
= 0UL;
539 struct cgroup_subsys
*ss
;
542 for_each_subsys(ss
, i
)
543 key
+= (unsigned long)css
[i
];
544 key
= (key
>> 16) ^ key
;
549 static void put_css_set_locked(struct css_set
*cset
)
551 struct cgrp_cset_link
*link
, *tmp_link
;
552 struct cgroup_subsys
*ss
;
555 lockdep_assert_held(&css_set_rwsem
);
557 if (!atomic_dec_and_test(&cset
->refcount
))
560 /* This css_set is dead. unlink it and release cgroup refcounts */
561 for_each_subsys(ss
, ssid
)
562 list_del(&cset
->e_cset_node
[ssid
]);
563 hash_del(&cset
->hlist
);
566 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
567 struct cgroup
*cgrp
= link
->cgrp
;
569 list_del(&link
->cset_link
);
570 list_del(&link
->cgrp_link
);
572 /* @cgrp can't go away while we're holding css_set_rwsem */
573 if (list_empty(&cgrp
->cset_links
)) {
574 cgroup_update_populated(cgrp
, false);
575 check_for_release(cgrp
);
581 kfree_rcu(cset
, rcu_head
);
584 static void put_css_set(struct css_set
*cset
)
587 * Ensure that the refcount doesn't hit zero while any readers
588 * can see it. Similar to atomic_dec_and_lock(), but for an
591 if (atomic_add_unless(&cset
->refcount
, -1, 1))
594 down_write(&css_set_rwsem
);
595 put_css_set_locked(cset
);
596 up_write(&css_set_rwsem
);
600 * refcounted get/put for css_set objects
602 static inline void get_css_set(struct css_set
*cset
)
604 atomic_inc(&cset
->refcount
);
608 * compare_css_sets - helper function for find_existing_css_set().
609 * @cset: candidate css_set being tested
610 * @old_cset: existing css_set for a task
611 * @new_cgrp: cgroup that's being entered by the task
612 * @template: desired set of css pointers in css_set (pre-calculated)
614 * Returns true if "cset" matches "old_cset" except for the hierarchy
615 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
617 static bool compare_css_sets(struct css_set
*cset
,
618 struct css_set
*old_cset
,
619 struct cgroup
*new_cgrp
,
620 struct cgroup_subsys_state
*template[])
622 struct list_head
*l1
, *l2
;
625 * On the default hierarchy, there can be csets which are
626 * associated with the same set of cgroups but different csses.
627 * Let's first ensure that csses match.
629 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
633 * Compare cgroup pointers in order to distinguish between
634 * different cgroups in hierarchies. As different cgroups may
635 * share the same effective css, this comparison is always
638 l1
= &cset
->cgrp_links
;
639 l2
= &old_cset
->cgrp_links
;
641 struct cgrp_cset_link
*link1
, *link2
;
642 struct cgroup
*cgrp1
, *cgrp2
;
646 /* See if we reached the end - both lists are equal length. */
647 if (l1
== &cset
->cgrp_links
) {
648 BUG_ON(l2
!= &old_cset
->cgrp_links
);
651 BUG_ON(l2
== &old_cset
->cgrp_links
);
653 /* Locate the cgroups associated with these links. */
654 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
655 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
658 /* Hierarchies should be linked in the same order. */
659 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
662 * If this hierarchy is the hierarchy of the cgroup
663 * that's changing, then we need to check that this
664 * css_set points to the new cgroup; if it's any other
665 * hierarchy, then this css_set should point to the
666 * same cgroup as the old css_set.
668 if (cgrp1
->root
== new_cgrp
->root
) {
669 if (cgrp1
!= new_cgrp
)
680 * find_existing_css_set - init css array and find the matching css_set
681 * @old_cset: the css_set that we're using before the cgroup transition
682 * @cgrp: the cgroup that we're moving into
683 * @template: out param for the new set of csses, should be clear on entry
685 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
687 struct cgroup_subsys_state
*template[])
689 struct cgroup_root
*root
= cgrp
->root
;
690 struct cgroup_subsys
*ss
;
691 struct css_set
*cset
;
696 * Build the set of subsystem state objects that we want to see in the
697 * new css_set. while subsystems can change globally, the entries here
698 * won't change, so no need for locking.
700 for_each_subsys(ss
, i
) {
701 if (root
->subsys_mask
& (1UL << i
)) {
703 * @ss is in this hierarchy, so we want the
704 * effective css from @cgrp.
706 template[i
] = cgroup_e_css(cgrp
, ss
);
709 * @ss is not in this hierarchy, so we don't want
712 template[i
] = old_cset
->subsys
[i
];
716 key
= css_set_hash(template);
717 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
718 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
721 /* This css_set matches what we need */
725 /* No existing cgroup group matched */
729 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
731 struct cgrp_cset_link
*link
, *tmp_link
;
733 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
734 list_del(&link
->cset_link
);
740 * allocate_cgrp_cset_links - allocate cgrp_cset_links
741 * @count: the number of links to allocate
742 * @tmp_links: list_head the allocated links are put on
744 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
745 * through ->cset_link. Returns 0 on success or -errno.
747 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
749 struct cgrp_cset_link
*link
;
752 INIT_LIST_HEAD(tmp_links
);
754 for (i
= 0; i
< count
; i
++) {
755 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
757 free_cgrp_cset_links(tmp_links
);
760 list_add(&link
->cset_link
, tmp_links
);
766 * link_css_set - a helper function to link a css_set to a cgroup
767 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
768 * @cset: the css_set to be linked
769 * @cgrp: the destination cgroup
771 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
774 struct cgrp_cset_link
*link
;
776 BUG_ON(list_empty(tmp_links
));
778 if (cgroup_on_dfl(cgrp
))
779 cset
->dfl_cgrp
= cgrp
;
781 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
785 if (list_empty(&cgrp
->cset_links
))
786 cgroup_update_populated(cgrp
, true);
787 list_move(&link
->cset_link
, &cgrp
->cset_links
);
790 * Always add links to the tail of the list so that the list
791 * is sorted by order of hierarchy creation
793 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
797 * find_css_set - return a new css_set with one cgroup updated
798 * @old_cset: the baseline css_set
799 * @cgrp: the cgroup to be updated
801 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
802 * substituted into the appropriate hierarchy.
804 static struct css_set
*find_css_set(struct css_set
*old_cset
,
807 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
808 struct css_set
*cset
;
809 struct list_head tmp_links
;
810 struct cgrp_cset_link
*link
;
811 struct cgroup_subsys
*ss
;
815 lockdep_assert_held(&cgroup_mutex
);
817 /* First see if we already have a cgroup group that matches
819 down_read(&css_set_rwsem
);
820 cset
= find_existing_css_set(old_cset
, cgrp
, template);
823 up_read(&css_set_rwsem
);
828 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
832 /* Allocate all the cgrp_cset_link objects that we'll need */
833 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
838 atomic_set(&cset
->refcount
, 1);
839 INIT_LIST_HEAD(&cset
->cgrp_links
);
840 INIT_LIST_HEAD(&cset
->tasks
);
841 INIT_LIST_HEAD(&cset
->mg_tasks
);
842 INIT_LIST_HEAD(&cset
->mg_preload_node
);
843 INIT_LIST_HEAD(&cset
->mg_node
);
844 INIT_HLIST_NODE(&cset
->hlist
);
846 /* Copy the set of subsystem state objects generated in
847 * find_existing_css_set() */
848 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
850 down_write(&css_set_rwsem
);
851 /* Add reference counts and links from the new css_set. */
852 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
853 struct cgroup
*c
= link
->cgrp
;
855 if (c
->root
== cgrp
->root
)
857 link_css_set(&tmp_links
, cset
, c
);
860 BUG_ON(!list_empty(&tmp_links
));
864 /* Add @cset to the hash table */
865 key
= css_set_hash(cset
->subsys
);
866 hash_add(css_set_table
, &cset
->hlist
, key
);
868 for_each_subsys(ss
, ssid
)
869 list_add_tail(&cset
->e_cset_node
[ssid
],
870 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
872 up_write(&css_set_rwsem
);
877 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
879 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
881 return root_cgrp
->root
;
884 static int cgroup_init_root_id(struct cgroup_root
*root
)
888 lockdep_assert_held(&cgroup_mutex
);
890 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
894 root
->hierarchy_id
= id
;
898 static void cgroup_exit_root_id(struct cgroup_root
*root
)
900 lockdep_assert_held(&cgroup_mutex
);
902 if (root
->hierarchy_id
) {
903 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
904 root
->hierarchy_id
= 0;
908 static void cgroup_free_root(struct cgroup_root
*root
)
911 /* hierarchy ID should already have been released */
912 WARN_ON_ONCE(root
->hierarchy_id
);
914 idr_destroy(&root
->cgroup_idr
);
919 static void cgroup_destroy_root(struct cgroup_root
*root
)
921 struct cgroup
*cgrp
= &root
->cgrp
;
922 struct cgrp_cset_link
*link
, *tmp_link
;
924 mutex_lock(&cgroup_mutex
);
926 BUG_ON(atomic_read(&root
->nr_cgrps
));
927 BUG_ON(!list_empty(&cgrp
->self
.children
));
929 /* Rebind all subsystems back to the default hierarchy */
930 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
933 * Release all the links from cset_links to this hierarchy's
936 down_write(&css_set_rwsem
);
938 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
939 list_del(&link
->cset_link
);
940 list_del(&link
->cgrp_link
);
943 up_write(&css_set_rwsem
);
945 if (!list_empty(&root
->root_list
)) {
946 list_del(&root
->root_list
);
950 cgroup_exit_root_id(root
);
952 mutex_unlock(&cgroup_mutex
);
954 kernfs_destroy_root(root
->kf_root
);
955 cgroup_free_root(root
);
958 /* look up cgroup associated with given css_set on the specified hierarchy */
959 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
960 struct cgroup_root
*root
)
962 struct cgroup
*res
= NULL
;
964 lockdep_assert_held(&cgroup_mutex
);
965 lockdep_assert_held(&css_set_rwsem
);
967 if (cset
== &init_css_set
) {
970 struct cgrp_cset_link
*link
;
972 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
973 struct cgroup
*c
= link
->cgrp
;
975 if (c
->root
== root
) {
987 * Return the cgroup for "task" from the given hierarchy. Must be
988 * called with cgroup_mutex and css_set_rwsem held.
990 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
991 struct cgroup_root
*root
)
994 * No need to lock the task - since we hold cgroup_mutex the
995 * task can't change groups, so the only thing that can happen
996 * is that it exits and its css is set back to init_css_set.
998 return cset_cgroup_from_root(task_css_set(task
), root
);
1002 * A task must hold cgroup_mutex to modify cgroups.
1004 * Any task can increment and decrement the count field without lock.
1005 * So in general, code holding cgroup_mutex can't rely on the count
1006 * field not changing. However, if the count goes to zero, then only
1007 * cgroup_attach_task() can increment it again. Because a count of zero
1008 * means that no tasks are currently attached, therefore there is no
1009 * way a task attached to that cgroup can fork (the other way to
1010 * increment the count). So code holding cgroup_mutex can safely
1011 * assume that if the count is zero, it will stay zero. Similarly, if
1012 * a task holds cgroup_mutex on a cgroup with zero count, it
1013 * knows that the cgroup won't be removed, as cgroup_rmdir()
1016 * A cgroup can only be deleted if both its 'count' of using tasks
1017 * is zero, and its list of 'children' cgroups is empty. Since all
1018 * tasks in the system use _some_ cgroup, and since there is always at
1019 * least one task in the system (init, pid == 1), therefore, root cgroup
1020 * always has either children cgroups and/or using tasks. So we don't
1021 * need a special hack to ensure that root cgroup cannot be deleted.
1023 * P.S. One more locking exception. RCU is used to guard the
1024 * update of a tasks cgroup pointer by cgroup_attach_task()
1027 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
1028 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1029 static const struct file_operations proc_cgroupstats_operations
;
1031 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1034 struct cgroup_subsys
*ss
= cft
->ss
;
1036 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1037 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1038 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1039 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1042 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1047 * cgroup_file_mode - deduce file mode of a control file
1048 * @cft: the control file in question
1050 * returns cft->mode if ->mode is not 0
1051 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1052 * returns S_IRUGO if it has only a read handler
1053 * returns S_IWUSR if it has only a write hander
1055 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1062 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1065 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
)
1071 static void cgroup_get(struct cgroup
*cgrp
)
1073 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
1074 css_get(&cgrp
->self
);
1077 static bool cgroup_tryget(struct cgroup
*cgrp
)
1079 return css_tryget(&cgrp
->self
);
1082 static void cgroup_put(struct cgroup
*cgrp
)
1084 css_put(&cgrp
->self
);
1088 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
1089 * @cgrp: the target cgroup
1090 * @subtree_control: the new subtree_control mask to consider
1092 * On the default hierarchy, a subsystem may request other subsystems to be
1093 * enabled together through its ->depends_on mask. In such cases, more
1094 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1096 * This function calculates which subsystems need to be enabled if
1097 * @subtree_control is to be applied to @cgrp. The returned mask is always
1098 * a superset of @subtree_control and follows the usual hierarchy rules.
1100 static unsigned long cgroup_calc_child_subsys_mask(struct cgroup
*cgrp
,
1101 unsigned long subtree_control
)
1103 struct cgroup
*parent
= cgroup_parent(cgrp
);
1104 unsigned long cur_ss_mask
= subtree_control
;
1105 struct cgroup_subsys
*ss
;
1108 lockdep_assert_held(&cgroup_mutex
);
1110 if (!cgroup_on_dfl(cgrp
))
1114 unsigned long new_ss_mask
= cur_ss_mask
;
1116 for_each_subsys_which(ss
, ssid
, &cur_ss_mask
)
1117 new_ss_mask
|= ss
->depends_on
;
1120 * Mask out subsystems which aren't available. This can
1121 * happen only if some depended-upon subsystems were bound
1122 * to non-default hierarchies.
1125 new_ss_mask
&= parent
->child_subsys_mask
;
1127 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1129 if (new_ss_mask
== cur_ss_mask
)
1131 cur_ss_mask
= new_ss_mask
;
1138 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1139 * @cgrp: the target cgroup
1141 * Update @cgrp->child_subsys_mask according to the current
1142 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1144 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1146 cgrp
->child_subsys_mask
=
1147 cgroup_calc_child_subsys_mask(cgrp
, cgrp
->subtree_control
);
1151 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1152 * @kn: the kernfs_node being serviced
1154 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1155 * the method finishes if locking succeeded. Note that once this function
1156 * returns the cgroup returned by cgroup_kn_lock_live() may become
1157 * inaccessible any time. If the caller intends to continue to access the
1158 * cgroup, it should pin it before invoking this function.
1160 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1162 struct cgroup
*cgrp
;
1164 if (kernfs_type(kn
) == KERNFS_DIR
)
1167 cgrp
= kn
->parent
->priv
;
1169 mutex_unlock(&cgroup_mutex
);
1171 kernfs_unbreak_active_protection(kn
);
1176 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1177 * @kn: the kernfs_node being serviced
1179 * This helper is to be used by a cgroup kernfs method currently servicing
1180 * @kn. It breaks the active protection, performs cgroup locking and
1181 * verifies that the associated cgroup is alive. Returns the cgroup if
1182 * alive; otherwise, %NULL. A successful return should be undone by a
1183 * matching cgroup_kn_unlock() invocation.
1185 * Any cgroup kernfs method implementation which requires locking the
1186 * associated cgroup should use this helper. It avoids nesting cgroup
1187 * locking under kernfs active protection and allows all kernfs operations
1188 * including self-removal.
1190 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1192 struct cgroup
*cgrp
;
1194 if (kernfs_type(kn
) == KERNFS_DIR
)
1197 cgrp
= kn
->parent
->priv
;
1200 * We're gonna grab cgroup_mutex which nests outside kernfs
1201 * active_ref. cgroup liveliness check alone provides enough
1202 * protection against removal. Ensure @cgrp stays accessible and
1203 * break the active_ref protection.
1205 if (!cgroup_tryget(cgrp
))
1207 kernfs_break_active_protection(kn
);
1209 mutex_lock(&cgroup_mutex
);
1211 if (!cgroup_is_dead(cgrp
))
1214 cgroup_kn_unlock(kn
);
1218 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1220 char name
[CGROUP_FILE_NAME_MAX
];
1222 lockdep_assert_held(&cgroup_mutex
);
1223 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1227 * cgroup_clear_dir - remove subsys files in a cgroup directory
1228 * @cgrp: target cgroup
1229 * @subsys_mask: mask of the subsystem ids whose files should be removed
1231 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
1233 struct cgroup_subsys
*ss
;
1236 for_each_subsys(ss
, i
) {
1237 struct cftype
*cfts
;
1239 if (!(subsys_mask
& (1 << i
)))
1241 list_for_each_entry(cfts
, &ss
->cfts
, node
)
1242 cgroup_addrm_files(cgrp
, cfts
, false);
1246 static int rebind_subsystems(struct cgroup_root
*dst_root
,
1247 unsigned long ss_mask
)
1249 struct cgroup_subsys
*ss
;
1250 unsigned long tmp_ss_mask
;
1253 lockdep_assert_held(&cgroup_mutex
);
1255 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1256 /* if @ss has non-root csses attached to it, can't move */
1257 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1260 /* can't move between two non-dummy roots either */
1261 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1265 /* skip creating root files on dfl_root for inhibited subsystems */
1266 tmp_ss_mask
= ss_mask
;
1267 if (dst_root
== &cgrp_dfl_root
)
1268 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1270 ret
= cgroup_populate_dir(&dst_root
->cgrp
, tmp_ss_mask
);
1272 if (dst_root
!= &cgrp_dfl_root
)
1276 * Rebinding back to the default root is not allowed to
1277 * fail. Using both default and non-default roots should
1278 * be rare. Moving subsystems back and forth even more so.
1279 * Just warn about it and continue.
1281 if (cgrp_dfl_root_visible
) {
1282 pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n",
1284 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1289 * Nothing can fail from this point on. Remove files for the
1290 * removed subsystems and rebind each subsystem.
1292 for_each_subsys_which(ss
, ssid
, &ss_mask
)
1293 cgroup_clear_dir(&ss
->root
->cgrp
, 1 << ssid
);
1295 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1296 struct cgroup_root
*src_root
;
1297 struct cgroup_subsys_state
*css
;
1298 struct css_set
*cset
;
1300 src_root
= ss
->root
;
1301 css
= cgroup_css(&src_root
->cgrp
, ss
);
1303 WARN_ON(!css
|| cgroup_css(&dst_root
->cgrp
, ss
));
1305 RCU_INIT_POINTER(src_root
->cgrp
.subsys
[ssid
], NULL
);
1306 rcu_assign_pointer(dst_root
->cgrp
.subsys
[ssid
], css
);
1307 ss
->root
= dst_root
;
1308 css
->cgroup
= &dst_root
->cgrp
;
1310 down_write(&css_set_rwsem
);
1311 hash_for_each(css_set_table
, i
, cset
, hlist
)
1312 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1313 &dst_root
->cgrp
.e_csets
[ss
->id
]);
1314 up_write(&css_set_rwsem
);
1316 src_root
->subsys_mask
&= ~(1 << ssid
);
1317 src_root
->cgrp
.subtree_control
&= ~(1 << ssid
);
1318 cgroup_refresh_child_subsys_mask(&src_root
->cgrp
);
1320 /* default hierarchy doesn't enable controllers by default */
1321 dst_root
->subsys_mask
|= 1 << ssid
;
1322 if (dst_root
!= &cgrp_dfl_root
) {
1323 dst_root
->cgrp
.subtree_control
|= 1 << ssid
;
1324 cgroup_refresh_child_subsys_mask(&dst_root
->cgrp
);
1331 kernfs_activate(dst_root
->cgrp
.kn
);
1335 static int cgroup_show_options(struct seq_file
*seq
,
1336 struct kernfs_root
*kf_root
)
1338 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1339 struct cgroup_subsys
*ss
;
1342 if (root
!= &cgrp_dfl_root
)
1343 for_each_subsys(ss
, ssid
)
1344 if (root
->subsys_mask
& (1 << ssid
))
1345 seq_printf(seq
, ",%s", ss
->legacy_name
);
1346 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1347 seq_puts(seq
, ",noprefix");
1348 if (root
->flags
& CGRP_ROOT_XATTR
)
1349 seq_puts(seq
, ",xattr");
1351 spin_lock(&release_agent_path_lock
);
1352 if (strlen(root
->release_agent_path
))
1353 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1354 spin_unlock(&release_agent_path_lock
);
1356 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1357 seq_puts(seq
, ",clone_children");
1358 if (strlen(root
->name
))
1359 seq_printf(seq
, ",name=%s", root
->name
);
1363 struct cgroup_sb_opts
{
1364 unsigned long subsys_mask
;
1366 char *release_agent
;
1367 bool cpuset_clone_children
;
1369 /* User explicitly requested empty subsystem */
1373 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1375 char *token
, *o
= data
;
1376 bool all_ss
= false, one_ss
= false;
1377 unsigned long mask
= -1UL;
1378 struct cgroup_subsys
*ss
;
1382 #ifdef CONFIG_CPUSETS
1383 mask
= ~(1U << cpuset_cgrp_id
);
1386 memset(opts
, 0, sizeof(*opts
));
1388 while ((token
= strsep(&o
, ",")) != NULL
) {
1393 if (!strcmp(token
, "none")) {
1394 /* Explicitly have no subsystems */
1398 if (!strcmp(token
, "all")) {
1399 /* Mutually exclusive option 'all' + subsystem name */
1405 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1406 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1409 if (!strcmp(token
, "noprefix")) {
1410 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1413 if (!strcmp(token
, "clone_children")) {
1414 opts
->cpuset_clone_children
= true;
1417 if (!strcmp(token
, "xattr")) {
1418 opts
->flags
|= CGRP_ROOT_XATTR
;
1421 if (!strncmp(token
, "release_agent=", 14)) {
1422 /* Specifying two release agents is forbidden */
1423 if (opts
->release_agent
)
1425 opts
->release_agent
=
1426 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1427 if (!opts
->release_agent
)
1431 if (!strncmp(token
, "name=", 5)) {
1432 const char *name
= token
+ 5;
1433 /* Can't specify an empty name */
1436 /* Must match [\w.-]+ */
1437 for (i
= 0; i
< strlen(name
); i
++) {
1441 if ((c
== '.') || (c
== '-') || (c
== '_'))
1445 /* Specifying two names is forbidden */
1448 opts
->name
= kstrndup(name
,
1449 MAX_CGROUP_ROOT_NAMELEN
- 1,
1457 for_each_subsys(ss
, i
) {
1458 if (strcmp(token
, ss
->legacy_name
))
1463 /* Mutually exclusive option 'all' + subsystem name */
1466 opts
->subsys_mask
|= (1 << i
);
1471 if (i
== CGROUP_SUBSYS_COUNT
)
1475 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1476 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1478 pr_err("sane_behavior: no other mount options allowed\n");
1485 * If the 'all' option was specified select all the subsystems,
1486 * otherwise if 'none', 'name=' and a subsystem name options were
1487 * not specified, let's default to 'all'
1489 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1490 for_each_subsys(ss
, i
)
1492 opts
->subsys_mask
|= (1 << i
);
1495 * We either have to specify by name or by subsystems. (So all
1496 * empty hierarchies must have a name).
1498 if (!opts
->subsys_mask
&& !opts
->name
)
1502 * Option noprefix was introduced just for backward compatibility
1503 * with the old cpuset, so we allow noprefix only if mounting just
1504 * the cpuset subsystem.
1506 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1509 /* Can't specify "none" and some subsystems */
1510 if (opts
->subsys_mask
&& opts
->none
)
1516 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1519 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1520 struct cgroup_sb_opts opts
;
1521 unsigned long added_mask
, removed_mask
;
1523 if (root
== &cgrp_dfl_root
) {
1524 pr_err("remount is not allowed\n");
1528 mutex_lock(&cgroup_mutex
);
1530 /* See what subsystems are wanted */
1531 ret
= parse_cgroupfs_options(data
, &opts
);
1535 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1536 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1537 task_tgid_nr(current
), current
->comm
);
1539 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1540 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1542 /* Don't allow flags or name to change at remount */
1543 if ((opts
.flags
^ root
->flags
) ||
1544 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1545 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1546 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1551 /* remounting is not allowed for populated hierarchies */
1552 if (!list_empty(&root
->cgrp
.self
.children
)) {
1557 ret
= rebind_subsystems(root
, added_mask
);
1561 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1563 if (opts
.release_agent
) {
1564 spin_lock(&release_agent_path_lock
);
1565 strcpy(root
->release_agent_path
, opts
.release_agent
);
1566 spin_unlock(&release_agent_path_lock
);
1569 kfree(opts
.release_agent
);
1571 mutex_unlock(&cgroup_mutex
);
1576 * To reduce the fork() overhead for systems that are not actually using
1577 * their cgroups capability, we don't maintain the lists running through
1578 * each css_set to its tasks until we see the list actually used - in other
1579 * words after the first mount.
1581 static bool use_task_css_set_links __read_mostly
;
1583 static void cgroup_enable_task_cg_lists(void)
1585 struct task_struct
*p
, *g
;
1587 down_write(&css_set_rwsem
);
1589 if (use_task_css_set_links
)
1592 use_task_css_set_links
= true;
1595 * We need tasklist_lock because RCU is not safe against
1596 * while_each_thread(). Besides, a forking task that has passed
1597 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1598 * is not guaranteed to have its child immediately visible in the
1599 * tasklist if we walk through it with RCU.
1601 read_lock(&tasklist_lock
);
1602 do_each_thread(g
, p
) {
1603 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1604 task_css_set(p
) != &init_css_set
);
1607 * We should check if the process is exiting, otherwise
1608 * it will race with cgroup_exit() in that the list
1609 * entry won't be deleted though the process has exited.
1610 * Do it while holding siglock so that we don't end up
1611 * racing against cgroup_exit().
1613 spin_lock_irq(&p
->sighand
->siglock
);
1614 if (!(p
->flags
& PF_EXITING
)) {
1615 struct css_set
*cset
= task_css_set(p
);
1617 list_add(&p
->cg_list
, &cset
->tasks
);
1620 spin_unlock_irq(&p
->sighand
->siglock
);
1621 } while_each_thread(g
, p
);
1622 read_unlock(&tasklist_lock
);
1624 up_write(&css_set_rwsem
);
1627 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1629 struct cgroup_subsys
*ss
;
1632 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1633 INIT_LIST_HEAD(&cgrp
->self
.children
);
1634 INIT_LIST_HEAD(&cgrp
->cset_links
);
1635 INIT_LIST_HEAD(&cgrp
->pidlists
);
1636 mutex_init(&cgrp
->pidlist_mutex
);
1637 cgrp
->self
.cgroup
= cgrp
;
1638 cgrp
->self
.flags
|= CSS_ONLINE
;
1640 for_each_subsys(ss
, ssid
)
1641 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1643 init_waitqueue_head(&cgrp
->offline_waitq
);
1644 INIT_WORK(&cgrp
->release_agent_work
, cgroup_release_agent
);
1647 static void init_cgroup_root(struct cgroup_root
*root
,
1648 struct cgroup_sb_opts
*opts
)
1650 struct cgroup
*cgrp
= &root
->cgrp
;
1652 INIT_LIST_HEAD(&root
->root_list
);
1653 atomic_set(&root
->nr_cgrps
, 1);
1655 init_cgroup_housekeeping(cgrp
);
1656 idr_init(&root
->cgroup_idr
);
1658 root
->flags
= opts
->flags
;
1659 if (opts
->release_agent
)
1660 strcpy(root
->release_agent_path
, opts
->release_agent
);
1662 strcpy(root
->name
, opts
->name
);
1663 if (opts
->cpuset_clone_children
)
1664 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1667 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned long ss_mask
)
1669 LIST_HEAD(tmp_links
);
1670 struct cgroup
*root_cgrp
= &root
->cgrp
;
1671 struct cftype
*base_files
;
1672 struct css_set
*cset
;
1675 lockdep_assert_held(&cgroup_mutex
);
1677 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1680 root_cgrp
->id
= ret
;
1682 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
, 0,
1688 * We're accessing css_set_count without locking css_set_rwsem here,
1689 * but that's OK - it can only be increased by someone holding
1690 * cgroup_lock, and that's us. The worst that can happen is that we
1691 * have some link structures left over
1693 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1697 ret
= cgroup_init_root_id(root
);
1701 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1702 KERNFS_ROOT_CREATE_DEACTIVATED
,
1704 if (IS_ERR(root
->kf_root
)) {
1705 ret
= PTR_ERR(root
->kf_root
);
1708 root_cgrp
->kn
= root
->kf_root
->kn
;
1710 if (root
== &cgrp_dfl_root
)
1711 base_files
= cgroup_dfl_base_files
;
1713 base_files
= cgroup_legacy_base_files
;
1715 ret
= cgroup_addrm_files(root_cgrp
, base_files
, true);
1719 ret
= rebind_subsystems(root
, ss_mask
);
1724 * There must be no failure case after here, since rebinding takes
1725 * care of subsystems' refcounts, which are explicitly dropped in
1726 * the failure exit path.
1728 list_add(&root
->root_list
, &cgroup_roots
);
1729 cgroup_root_count
++;
1732 * Link the root cgroup in this hierarchy into all the css_set
1735 down_write(&css_set_rwsem
);
1736 hash_for_each(css_set_table
, i
, cset
, hlist
)
1737 link_css_set(&tmp_links
, cset
, root_cgrp
);
1738 up_write(&css_set_rwsem
);
1740 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1741 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1743 kernfs_activate(root_cgrp
->kn
);
1748 kernfs_destroy_root(root
->kf_root
);
1749 root
->kf_root
= NULL
;
1751 cgroup_exit_root_id(root
);
1753 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1755 free_cgrp_cset_links(&tmp_links
);
1759 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1760 int flags
, const char *unused_dev_name
,
1763 struct super_block
*pinned_sb
= NULL
;
1764 struct cgroup_subsys
*ss
;
1765 struct cgroup_root
*root
;
1766 struct cgroup_sb_opts opts
;
1767 struct dentry
*dentry
;
1773 * The first time anyone tries to mount a cgroup, enable the list
1774 * linking each css_set to its tasks and fix up all existing tasks.
1776 if (!use_task_css_set_links
)
1777 cgroup_enable_task_cg_lists();
1779 mutex_lock(&cgroup_mutex
);
1781 /* First find the desired set of subsystems */
1782 ret
= parse_cgroupfs_options(data
, &opts
);
1786 /* look for a matching existing root */
1787 if (opts
.flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1788 cgrp_dfl_root_visible
= true;
1789 root
= &cgrp_dfl_root
;
1790 cgroup_get(&root
->cgrp
);
1796 * Destruction of cgroup root is asynchronous, so subsystems may
1797 * still be dying after the previous unmount. Let's drain the
1798 * dying subsystems. We just need to ensure that the ones
1799 * unmounted previously finish dying and don't care about new ones
1800 * starting. Testing ref liveliness is good enough.
1802 for_each_subsys(ss
, i
) {
1803 if (!(opts
.subsys_mask
& (1 << i
)) ||
1804 ss
->root
== &cgrp_dfl_root
)
1807 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
1808 mutex_unlock(&cgroup_mutex
);
1810 ret
= restart_syscall();
1813 cgroup_put(&ss
->root
->cgrp
);
1816 for_each_root(root
) {
1817 bool name_match
= false;
1819 if (root
== &cgrp_dfl_root
)
1823 * If we asked for a name then it must match. Also, if
1824 * name matches but sybsys_mask doesn't, we should fail.
1825 * Remember whether name matched.
1828 if (strcmp(opts
.name
, root
->name
))
1834 * If we asked for subsystems (or explicitly for no
1835 * subsystems) then they must match.
1837 if ((opts
.subsys_mask
|| opts
.none
) &&
1838 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1845 if (root
->flags
^ opts
.flags
)
1846 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1849 * We want to reuse @root whose lifetime is governed by its
1850 * ->cgrp. Let's check whether @root is alive and keep it
1851 * that way. As cgroup_kill_sb() can happen anytime, we
1852 * want to block it by pinning the sb so that @root doesn't
1853 * get killed before mount is complete.
1855 * With the sb pinned, tryget_live can reliably indicate
1856 * whether @root can be reused. If it's being killed,
1857 * drain it. We can use wait_queue for the wait but this
1858 * path is super cold. Let's just sleep a bit and retry.
1860 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
1861 if (IS_ERR(pinned_sb
) ||
1862 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1863 mutex_unlock(&cgroup_mutex
);
1864 if (!IS_ERR_OR_NULL(pinned_sb
))
1865 deactivate_super(pinned_sb
);
1867 ret
= restart_syscall();
1876 * No such thing, create a new one. name= matching without subsys
1877 * specification is allowed for already existing hierarchies but we
1878 * can't create new one without subsys specification.
1880 if (!opts
.subsys_mask
&& !opts
.none
) {
1885 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1891 init_cgroup_root(root
, &opts
);
1893 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
1895 cgroup_free_root(root
);
1898 mutex_unlock(&cgroup_mutex
);
1900 kfree(opts
.release_agent
);
1904 return ERR_PTR(ret
);
1906 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
1907 CGROUP_SUPER_MAGIC
, &new_sb
);
1908 if (IS_ERR(dentry
) || !new_sb
)
1909 cgroup_put(&root
->cgrp
);
1912 * If @pinned_sb, we're reusing an existing root and holding an
1913 * extra ref on its sb. Mount is complete. Put the extra ref.
1917 deactivate_super(pinned_sb
);
1923 static void cgroup_kill_sb(struct super_block
*sb
)
1925 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
1926 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1929 * If @root doesn't have any mounts or children, start killing it.
1930 * This prevents new mounts by disabling percpu_ref_tryget_live().
1931 * cgroup_mount() may wait for @root's release.
1933 * And don't kill the default root.
1935 if (!list_empty(&root
->cgrp
.self
.children
) ||
1936 root
== &cgrp_dfl_root
)
1937 cgroup_put(&root
->cgrp
);
1939 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
1944 static struct file_system_type cgroup_fs_type
= {
1946 .mount
= cgroup_mount
,
1947 .kill_sb
= cgroup_kill_sb
,
1951 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1952 * @task: target task
1953 * @buf: the buffer to write the path into
1954 * @buflen: the length of the buffer
1956 * Determine @task's cgroup on the first (the one with the lowest non-zero
1957 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1958 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1959 * cgroup controller callbacks.
1961 * Return value is the same as kernfs_path().
1963 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1965 struct cgroup_root
*root
;
1966 struct cgroup
*cgrp
;
1967 int hierarchy_id
= 1;
1970 mutex_lock(&cgroup_mutex
);
1971 down_read(&css_set_rwsem
);
1973 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1976 cgrp
= task_cgroup_from_root(task
, root
);
1977 path
= cgroup_path(cgrp
, buf
, buflen
);
1979 /* if no hierarchy exists, everyone is in "/" */
1980 if (strlcpy(buf
, "/", buflen
) < buflen
)
1984 up_read(&css_set_rwsem
);
1985 mutex_unlock(&cgroup_mutex
);
1988 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1990 /* used to track tasks and other necessary states during migration */
1991 struct cgroup_taskset
{
1992 /* the src and dst cset list running through cset->mg_node */
1993 struct list_head src_csets
;
1994 struct list_head dst_csets
;
1997 * Fields for cgroup_taskset_*() iteration.
1999 * Before migration is committed, the target migration tasks are on
2000 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2001 * the csets on ->dst_csets. ->csets point to either ->src_csets
2002 * or ->dst_csets depending on whether migration is committed.
2004 * ->cur_csets and ->cur_task point to the current task position
2007 struct list_head
*csets
;
2008 struct css_set
*cur_cset
;
2009 struct task_struct
*cur_task
;
2013 * cgroup_taskset_first - reset taskset and return the first task
2014 * @tset: taskset of interest
2016 * @tset iteration is initialized and the first task is returned.
2018 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
2020 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2021 tset
->cur_task
= NULL
;
2023 return cgroup_taskset_next(tset
);
2027 * cgroup_taskset_next - iterate to the next task in taskset
2028 * @tset: taskset of interest
2030 * Return the next task in @tset. Iteration must have been initialized
2031 * with cgroup_taskset_first().
2033 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
2035 struct css_set
*cset
= tset
->cur_cset
;
2036 struct task_struct
*task
= tset
->cur_task
;
2038 while (&cset
->mg_node
!= tset
->csets
) {
2040 task
= list_first_entry(&cset
->mg_tasks
,
2041 struct task_struct
, cg_list
);
2043 task
= list_next_entry(task
, cg_list
);
2045 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2046 tset
->cur_cset
= cset
;
2047 tset
->cur_task
= task
;
2051 cset
= list_next_entry(cset
, mg_node
);
2059 * cgroup_task_migrate - move a task from one cgroup to another.
2060 * @old_cgrp: the cgroup @tsk is being migrated from
2061 * @tsk: the task being migrated
2062 * @new_cset: the new css_set @tsk is being attached to
2064 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
2066 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
2067 struct task_struct
*tsk
,
2068 struct css_set
*new_cset
)
2070 struct css_set
*old_cset
;
2072 lockdep_assert_held(&cgroup_mutex
);
2073 lockdep_assert_held(&css_set_rwsem
);
2076 * We are synchronized through cgroup_threadgroup_rwsem against
2077 * PF_EXITING setting such that we can't race against cgroup_exit()
2078 * changing the css_set to init_css_set and dropping the old one.
2080 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
2081 old_cset
= task_css_set(tsk
);
2083 get_css_set(new_cset
);
2084 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
2087 * Use move_tail so that cgroup_taskset_first() still returns the
2088 * leader after migration. This works because cgroup_migrate()
2089 * ensures that the dst_cset of the leader is the first on the
2090 * tset's dst_csets list.
2092 list_move_tail(&tsk
->cg_list
, &new_cset
->mg_tasks
);
2095 * We just gained a reference on old_cset by taking it from the
2096 * task. As trading it for new_cset is protected by cgroup_mutex,
2097 * we're safe to drop it here; it will be freed under RCU.
2099 put_css_set_locked(old_cset
);
2103 * cgroup_migrate_finish - cleanup after attach
2104 * @preloaded_csets: list of preloaded css_sets
2106 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2107 * those functions for details.
2109 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2111 struct css_set
*cset
, *tmp_cset
;
2113 lockdep_assert_held(&cgroup_mutex
);
2115 down_write(&css_set_rwsem
);
2116 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2117 cset
->mg_src_cgrp
= NULL
;
2118 cset
->mg_dst_cset
= NULL
;
2119 list_del_init(&cset
->mg_preload_node
);
2120 put_css_set_locked(cset
);
2122 up_write(&css_set_rwsem
);
2126 * cgroup_migrate_add_src - add a migration source css_set
2127 * @src_cset: the source css_set to add
2128 * @dst_cgrp: the destination cgroup
2129 * @preloaded_csets: list of preloaded css_sets
2131 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2132 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2133 * up by cgroup_migrate_finish().
2135 * This function may be called without holding cgroup_threadgroup_rwsem
2136 * even if the target is a process. Threads may be created and destroyed
2137 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2138 * into play and the preloaded css_sets are guaranteed to cover all
2141 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2142 struct cgroup
*dst_cgrp
,
2143 struct list_head
*preloaded_csets
)
2145 struct cgroup
*src_cgrp
;
2147 lockdep_assert_held(&cgroup_mutex
);
2148 lockdep_assert_held(&css_set_rwsem
);
2150 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2152 if (!list_empty(&src_cset
->mg_preload_node
))
2155 WARN_ON(src_cset
->mg_src_cgrp
);
2156 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2157 WARN_ON(!list_empty(&src_cset
->mg_node
));
2159 src_cset
->mg_src_cgrp
= src_cgrp
;
2160 get_css_set(src_cset
);
2161 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2165 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2166 * @dst_cgrp: the destination cgroup (may be %NULL)
2167 * @preloaded_csets: list of preloaded source css_sets
2169 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2170 * have been preloaded to @preloaded_csets. This function looks up and
2171 * pins all destination css_sets, links each to its source, and append them
2172 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2173 * source css_set is assumed to be its cgroup on the default hierarchy.
2175 * This function must be called after cgroup_migrate_add_src() has been
2176 * called on each migration source css_set. After migration is performed
2177 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2180 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2181 struct list_head
*preloaded_csets
)
2184 struct css_set
*src_cset
, *tmp_cset
;
2186 lockdep_assert_held(&cgroup_mutex
);
2189 * Except for the root, child_subsys_mask must be zero for a cgroup
2190 * with tasks so that child cgroups don't compete against tasks.
2192 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2193 dst_cgrp
->child_subsys_mask
)
2196 /* look up the dst cset for each src cset and link it to src */
2197 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2198 struct css_set
*dst_cset
;
2200 dst_cset
= find_css_set(src_cset
,
2201 dst_cgrp
?: src_cset
->dfl_cgrp
);
2205 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2208 * If src cset equals dst, it's noop. Drop the src.
2209 * cgroup_migrate() will skip the cset too. Note that we
2210 * can't handle src == dst as some nodes are used by both.
2212 if (src_cset
== dst_cset
) {
2213 src_cset
->mg_src_cgrp
= NULL
;
2214 list_del_init(&src_cset
->mg_preload_node
);
2215 put_css_set(src_cset
);
2216 put_css_set(dst_cset
);
2220 src_cset
->mg_dst_cset
= dst_cset
;
2222 if (list_empty(&dst_cset
->mg_preload_node
))
2223 list_add(&dst_cset
->mg_preload_node
, &csets
);
2225 put_css_set(dst_cset
);
2228 list_splice_tail(&csets
, preloaded_csets
);
2231 cgroup_migrate_finish(&csets
);
2236 * cgroup_migrate - migrate a process or task to a cgroup
2237 * @cgrp: the destination cgroup
2238 * @leader: the leader of the process or the task to migrate
2239 * @threadgroup: whether @leader points to the whole process or a single task
2241 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2242 * process, the caller must be holding cgroup_threadgroup_rwsem. The
2243 * caller is also responsible for invoking cgroup_migrate_add_src() and
2244 * cgroup_migrate_prepare_dst() on the targets before invoking this
2245 * function and following up with cgroup_migrate_finish().
2247 * As long as a controller's ->can_attach() doesn't fail, this function is
2248 * guaranteed to succeed. This means that, excluding ->can_attach()
2249 * failure, when migrating multiple targets, the success or failure can be
2250 * decided for all targets by invoking group_migrate_prepare_dst() before
2251 * actually starting migrating.
2253 static int cgroup_migrate(struct cgroup
*cgrp
, struct task_struct
*leader
,
2256 struct cgroup_taskset tset
= {
2257 .src_csets
= LIST_HEAD_INIT(tset
.src_csets
),
2258 .dst_csets
= LIST_HEAD_INIT(tset
.dst_csets
),
2259 .csets
= &tset
.src_csets
,
2261 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2262 struct css_set
*cset
, *tmp_cset
;
2263 struct task_struct
*task
, *tmp_task
;
2267 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2268 * already PF_EXITING could be freed from underneath us unless we
2269 * take an rcu_read_lock.
2271 down_write(&css_set_rwsem
);
2275 /* @task either already exited or can't exit until the end */
2276 if (task
->flags
& PF_EXITING
)
2279 /* leave @task alone if post_fork() hasn't linked it yet */
2280 if (list_empty(&task
->cg_list
))
2283 cset
= task_css_set(task
);
2284 if (!cset
->mg_src_cgrp
)
2288 * cgroup_taskset_first() must always return the leader.
2289 * Take care to avoid disturbing the ordering.
2291 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2292 if (list_empty(&cset
->mg_node
))
2293 list_add_tail(&cset
->mg_node
, &tset
.src_csets
);
2294 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2295 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2300 } while_each_thread(leader
, task
);
2302 up_write(&css_set_rwsem
);
2304 /* methods shouldn't be called if no task is actually migrating */
2305 if (list_empty(&tset
.src_csets
))
2308 /* check that we can legitimately attach to the cgroup */
2309 for_each_e_css(css
, i
, cgrp
) {
2310 if (css
->ss
->can_attach
) {
2311 ret
= css
->ss
->can_attach(css
, &tset
);
2314 goto out_cancel_attach
;
2320 * Now that we're guaranteed success, proceed to move all tasks to
2321 * the new cgroup. There are no failure cases after here, so this
2322 * is the commit point.
2324 down_write(&css_set_rwsem
);
2325 list_for_each_entry(cset
, &tset
.src_csets
, mg_node
) {
2326 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
)
2327 cgroup_task_migrate(cset
->mg_src_cgrp
, task
,
2330 up_write(&css_set_rwsem
);
2333 * Migration is committed, all target tasks are now on dst_csets.
2334 * Nothing is sensitive to fork() after this point. Notify
2335 * controllers that migration is complete.
2337 tset
.csets
= &tset
.dst_csets
;
2339 for_each_e_css(css
, i
, cgrp
)
2340 if (css
->ss
->attach
)
2341 css
->ss
->attach(css
, &tset
);
2344 goto out_release_tset
;
2347 for_each_e_css(css
, i
, cgrp
) {
2348 if (css
== failed_css
)
2350 if (css
->ss
->cancel_attach
)
2351 css
->ss
->cancel_attach(css
, &tset
);
2354 down_write(&css_set_rwsem
);
2355 list_splice_init(&tset
.dst_csets
, &tset
.src_csets
);
2356 list_for_each_entry_safe(cset
, tmp_cset
, &tset
.src_csets
, mg_node
) {
2357 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2358 list_del_init(&cset
->mg_node
);
2360 up_write(&css_set_rwsem
);
2365 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2366 * @dst_cgrp: the cgroup to attach to
2367 * @leader: the task or the leader of the threadgroup to be attached
2368 * @threadgroup: attach the whole threadgroup?
2370 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2372 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2373 struct task_struct
*leader
, bool threadgroup
)
2375 LIST_HEAD(preloaded_csets
);
2376 struct task_struct
*task
;
2379 /* look up all src csets */
2380 down_read(&css_set_rwsem
);
2384 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2388 } while_each_thread(leader
, task
);
2390 up_read(&css_set_rwsem
);
2392 /* prepare dst csets and commit */
2393 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2395 ret
= cgroup_migrate(dst_cgrp
, leader
, threadgroup
);
2397 cgroup_migrate_finish(&preloaded_csets
);
2401 static int cgroup_procs_write_permission(struct task_struct
*task
,
2402 struct cgroup
*dst_cgrp
,
2403 struct kernfs_open_file
*of
)
2405 const struct cred
*cred
= current_cred();
2406 const struct cred
*tcred
= get_task_cred(task
);
2410 * even if we're attaching all tasks in the thread group, we only
2411 * need to check permissions on one of them.
2413 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2414 !uid_eq(cred
->euid
, tcred
->uid
) &&
2415 !uid_eq(cred
->euid
, tcred
->suid
))
2418 if (!ret
&& cgroup_on_dfl(dst_cgrp
)) {
2419 struct super_block
*sb
= of
->file
->f_path
.dentry
->d_sb
;
2420 struct cgroup
*cgrp
;
2421 struct inode
*inode
;
2423 down_read(&css_set_rwsem
);
2424 cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
2425 up_read(&css_set_rwsem
);
2427 while (!cgroup_is_descendant(dst_cgrp
, cgrp
))
2428 cgrp
= cgroup_parent(cgrp
);
2431 inode
= kernfs_get_inode(sb
, cgrp
->procs_kn
);
2433 ret
= inode_permission(inode
, MAY_WRITE
);
2443 * Find the task_struct of the task to attach by vpid and pass it along to the
2444 * function to attach either it or all tasks in its threadgroup. Will lock
2445 * cgroup_mutex and threadgroup.
2447 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2448 size_t nbytes
, loff_t off
, bool threadgroup
)
2450 struct task_struct
*tsk
;
2451 struct cgroup
*cgrp
;
2455 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2458 cgrp
= cgroup_kn_lock_live(of
->kn
);
2462 percpu_down_write(&cgroup_threadgroup_rwsem
);
2465 tsk
= find_task_by_vpid(pid
);
2468 goto out_unlock_rcu
;
2475 tsk
= tsk
->group_leader
;
2478 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2479 * trapped in a cpuset, or RT worker may be born in a cgroup
2480 * with no rt_runtime allocated. Just say no.
2482 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2484 goto out_unlock_rcu
;
2487 get_task_struct(tsk
);
2490 ret
= cgroup_procs_write_permission(tsk
, cgrp
, of
);
2492 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2494 put_task_struct(tsk
);
2495 goto out_unlock_threadgroup
;
2499 out_unlock_threadgroup
:
2500 percpu_up_write(&cgroup_threadgroup_rwsem
);
2501 cgroup_kn_unlock(of
->kn
);
2502 return ret
?: nbytes
;
2506 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2507 * @from: attach to all cgroups of a given task
2508 * @tsk: the task to be attached
2510 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2512 struct cgroup_root
*root
;
2515 mutex_lock(&cgroup_mutex
);
2516 for_each_root(root
) {
2517 struct cgroup
*from_cgrp
;
2519 if (root
== &cgrp_dfl_root
)
2522 down_read(&css_set_rwsem
);
2523 from_cgrp
= task_cgroup_from_root(from
, root
);
2524 up_read(&css_set_rwsem
);
2526 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2530 mutex_unlock(&cgroup_mutex
);
2534 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2536 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2537 char *buf
, size_t nbytes
, loff_t off
)
2539 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2542 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2543 char *buf
, size_t nbytes
, loff_t off
)
2545 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2548 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2549 char *buf
, size_t nbytes
, loff_t off
)
2551 struct cgroup
*cgrp
;
2553 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2555 cgrp
= cgroup_kn_lock_live(of
->kn
);
2558 spin_lock(&release_agent_path_lock
);
2559 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2560 sizeof(cgrp
->root
->release_agent_path
));
2561 spin_unlock(&release_agent_path_lock
);
2562 cgroup_kn_unlock(of
->kn
);
2566 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2568 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2570 spin_lock(&release_agent_path_lock
);
2571 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2572 spin_unlock(&release_agent_path_lock
);
2573 seq_putc(seq
, '\n');
2577 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2579 seq_puts(seq
, "0\n");
2583 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned long ss_mask
)
2585 struct cgroup_subsys
*ss
;
2586 bool printed
= false;
2589 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
2592 seq_printf(seq
, "%s", ss
->name
);
2596 seq_putc(seq
, '\n');
2599 /* show controllers which are currently attached to the default hierarchy */
2600 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2602 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2604 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2605 ~cgrp_dfl_root_inhibit_ss_mask
);
2609 /* show controllers which are enabled from the parent */
2610 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2612 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2614 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2618 /* show controllers which are enabled for a given cgroup's children */
2619 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2621 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2623 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2628 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2629 * @cgrp: root of the subtree to update csses for
2631 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2632 * css associations need to be updated accordingly. This function looks up
2633 * all css_sets which are attached to the subtree, creates the matching
2634 * updated css_sets and migrates the tasks to the new ones.
2636 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2638 LIST_HEAD(preloaded_csets
);
2639 struct cgroup_subsys_state
*css
;
2640 struct css_set
*src_cset
;
2643 lockdep_assert_held(&cgroup_mutex
);
2645 percpu_down_write(&cgroup_threadgroup_rwsem
);
2647 /* look up all csses currently attached to @cgrp's subtree */
2648 down_read(&css_set_rwsem
);
2649 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2650 struct cgrp_cset_link
*link
;
2652 /* self is not affected by child_subsys_mask change */
2653 if (css
->cgroup
== cgrp
)
2656 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2657 cgroup_migrate_add_src(link
->cset
, cgrp
,
2660 up_read(&css_set_rwsem
);
2662 /* NULL dst indicates self on default hierarchy */
2663 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2667 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2668 struct task_struct
*last_task
= NULL
, *task
;
2670 /* src_csets precede dst_csets, break on the first dst_cset */
2671 if (!src_cset
->mg_src_cgrp
)
2675 * All tasks in src_cset need to be migrated to the
2676 * matching dst_cset. Empty it process by process. We
2677 * walk tasks but migrate processes. The leader might even
2678 * belong to a different cset but such src_cset would also
2679 * be among the target src_csets because the default
2680 * hierarchy enforces per-process membership.
2683 down_read(&css_set_rwsem
);
2684 task
= list_first_entry_or_null(&src_cset
->tasks
,
2685 struct task_struct
, cg_list
);
2687 task
= task
->group_leader
;
2688 WARN_ON_ONCE(!task_css_set(task
)->mg_src_cgrp
);
2689 get_task_struct(task
);
2691 up_read(&css_set_rwsem
);
2696 /* guard against possible infinite loop */
2697 if (WARN(last_task
== task
,
2698 "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n"))
2702 ret
= cgroup_migrate(src_cset
->dfl_cgrp
, task
, true);
2704 put_task_struct(task
);
2706 if (WARN(ret
, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret
))
2712 cgroup_migrate_finish(&preloaded_csets
);
2713 percpu_up_write(&cgroup_threadgroup_rwsem
);
2717 /* change the enabled child controllers for a cgroup in the default hierarchy */
2718 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2719 char *buf
, size_t nbytes
,
2722 unsigned long enable
= 0, disable
= 0;
2723 unsigned long css_enable
, css_disable
, old_sc
, new_sc
, old_ss
, new_ss
;
2724 struct cgroup
*cgrp
, *child
;
2725 struct cgroup_subsys
*ss
;
2730 * Parse input - space separated list of subsystem names prefixed
2731 * with either + or -.
2733 buf
= strstrip(buf
);
2734 while ((tok
= strsep(&buf
, " "))) {
2735 unsigned long tmp_ss_mask
= ~cgrp_dfl_root_inhibit_ss_mask
;
2739 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
2740 if (ss
->disabled
|| strcmp(tok
+ 1, ss
->name
))
2744 enable
|= 1 << ssid
;
2745 disable
&= ~(1 << ssid
);
2746 } else if (*tok
== '-') {
2747 disable
|= 1 << ssid
;
2748 enable
&= ~(1 << ssid
);
2754 if (ssid
== CGROUP_SUBSYS_COUNT
)
2758 cgrp
= cgroup_kn_lock_live(of
->kn
);
2762 for_each_subsys(ss
, ssid
) {
2763 if (enable
& (1 << ssid
)) {
2764 if (cgrp
->subtree_control
& (1 << ssid
)) {
2765 enable
&= ~(1 << ssid
);
2769 /* unavailable or not enabled on the parent? */
2770 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2771 (cgroup_parent(cgrp
) &&
2772 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2776 } else if (disable
& (1 << ssid
)) {
2777 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2778 disable
&= ~(1 << ssid
);
2782 /* a child has it enabled? */
2783 cgroup_for_each_live_child(child
, cgrp
) {
2784 if (child
->subtree_control
& (1 << ssid
)) {
2792 if (!enable
&& !disable
) {
2798 * Except for the root, subtree_control must be zero for a cgroup
2799 * with tasks so that child cgroups don't compete against tasks.
2801 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2807 * Update subsys masks and calculate what needs to be done. More
2808 * subsystems than specified may need to be enabled or disabled
2809 * depending on subsystem dependencies.
2811 old_sc
= cgrp
->subtree_control
;
2812 old_ss
= cgrp
->child_subsys_mask
;
2813 new_sc
= (old_sc
| enable
) & ~disable
;
2814 new_ss
= cgroup_calc_child_subsys_mask(cgrp
, new_sc
);
2816 css_enable
= ~old_ss
& new_ss
;
2817 css_disable
= old_ss
& ~new_ss
;
2818 enable
|= css_enable
;
2819 disable
|= css_disable
;
2822 * Because css offlining is asynchronous, userland might try to
2823 * re-enable the same controller while the previous instance is
2824 * still around. In such cases, wait till it's gone using
2827 for_each_subsys_which(ss
, ssid
, &css_enable
) {
2828 cgroup_for_each_live_child(child
, cgrp
) {
2831 if (!cgroup_css(child
, ss
))
2835 prepare_to_wait(&child
->offline_waitq
, &wait
,
2836 TASK_UNINTERRUPTIBLE
);
2837 cgroup_kn_unlock(of
->kn
);
2839 finish_wait(&child
->offline_waitq
, &wait
);
2842 return restart_syscall();
2846 cgrp
->subtree_control
= new_sc
;
2847 cgrp
->child_subsys_mask
= new_ss
;
2850 * Create new csses or make the existing ones visible. A css is
2851 * created invisible if it's being implicitly enabled through
2852 * dependency. An invisible css is made visible when the userland
2853 * explicitly enables it.
2855 for_each_subsys(ss
, ssid
) {
2856 if (!(enable
& (1 << ssid
)))
2859 cgroup_for_each_live_child(child
, cgrp
) {
2860 if (css_enable
& (1 << ssid
))
2861 ret
= create_css(child
, ss
,
2862 cgrp
->subtree_control
& (1 << ssid
));
2864 ret
= cgroup_populate_dir(child
, 1 << ssid
);
2871 * At this point, cgroup_e_css() results reflect the new csses
2872 * making the following cgroup_update_dfl_csses() properly update
2873 * css associations of all tasks in the subtree.
2875 ret
= cgroup_update_dfl_csses(cgrp
);
2880 * All tasks are migrated out of disabled csses. Kill or hide
2881 * them. A css is hidden when the userland requests it to be
2882 * disabled while other subsystems are still depending on it. The
2883 * css must not actively control resources and be in the vanilla
2884 * state if it's made visible again later. Controllers which may
2885 * be depended upon should provide ->css_reset() for this purpose.
2887 for_each_subsys(ss
, ssid
) {
2888 if (!(disable
& (1 << ssid
)))
2891 cgroup_for_each_live_child(child
, cgrp
) {
2892 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2894 if (css_disable
& (1 << ssid
)) {
2897 cgroup_clear_dir(child
, 1 << ssid
);
2905 * The effective csses of all the descendants (excluding @cgrp) may
2906 * have changed. Subsystems can optionally subscribe to this event
2907 * by implementing ->css_e_css_changed() which is invoked if any of
2908 * the effective csses seen from the css's cgroup may have changed.
2910 for_each_subsys(ss
, ssid
) {
2911 struct cgroup_subsys_state
*this_css
= cgroup_css(cgrp
, ss
);
2912 struct cgroup_subsys_state
*css
;
2914 if (!ss
->css_e_css_changed
|| !this_css
)
2917 css_for_each_descendant_pre(css
, this_css
)
2918 if (css
!= this_css
)
2919 ss
->css_e_css_changed(css
);
2922 kernfs_activate(cgrp
->kn
);
2925 cgroup_kn_unlock(of
->kn
);
2926 return ret
?: nbytes
;
2929 cgrp
->subtree_control
= old_sc
;
2930 cgrp
->child_subsys_mask
= old_ss
;
2932 for_each_subsys(ss
, ssid
) {
2933 if (!(enable
& (1 << ssid
)))
2936 cgroup_for_each_live_child(child
, cgrp
) {
2937 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2942 if (css_enable
& (1 << ssid
))
2945 cgroup_clear_dir(child
, 1 << ssid
);
2951 static int cgroup_populated_show(struct seq_file
*seq
, void *v
)
2953 seq_printf(seq
, "%d\n", (bool)seq_css(seq
)->cgroup
->populated_cnt
);
2957 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
2958 size_t nbytes
, loff_t off
)
2960 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
2961 struct cftype
*cft
= of
->kn
->priv
;
2962 struct cgroup_subsys_state
*css
;
2966 return cft
->write(of
, buf
, nbytes
, off
);
2969 * kernfs guarantees that a file isn't deleted with operations in
2970 * flight, which means that the matching css is and stays alive and
2971 * doesn't need to be pinned. The RCU locking is not necessary
2972 * either. It's just for the convenience of using cgroup_css().
2975 css
= cgroup_css(cgrp
, cft
->ss
);
2978 if (cft
->write_u64
) {
2979 unsigned long long v
;
2980 ret
= kstrtoull(buf
, 0, &v
);
2982 ret
= cft
->write_u64(css
, cft
, v
);
2983 } else if (cft
->write_s64
) {
2985 ret
= kstrtoll(buf
, 0, &v
);
2987 ret
= cft
->write_s64(css
, cft
, v
);
2992 return ret
?: nbytes
;
2995 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
2997 return seq_cft(seq
)->seq_start(seq
, ppos
);
3000 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3002 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3005 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3007 seq_cft(seq
)->seq_stop(seq
, v
);
3010 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3012 struct cftype
*cft
= seq_cft(m
);
3013 struct cgroup_subsys_state
*css
= seq_css(m
);
3016 return cft
->seq_show(m
, arg
);
3019 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3020 else if (cft
->read_s64
)
3021 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3027 static struct kernfs_ops cgroup_kf_single_ops
= {
3028 .atomic_write_len
= PAGE_SIZE
,
3029 .write
= cgroup_file_write
,
3030 .seq_show
= cgroup_seqfile_show
,
3033 static struct kernfs_ops cgroup_kf_ops
= {
3034 .atomic_write_len
= PAGE_SIZE
,
3035 .write
= cgroup_file_write
,
3036 .seq_start
= cgroup_seqfile_start
,
3037 .seq_next
= cgroup_seqfile_next
,
3038 .seq_stop
= cgroup_seqfile_stop
,
3039 .seq_show
= cgroup_seqfile_show
,
3043 * cgroup_rename - Only allow simple rename of directories in place.
3045 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
3046 const char *new_name_str
)
3048 struct cgroup
*cgrp
= kn
->priv
;
3051 if (kernfs_type(kn
) != KERNFS_DIR
)
3053 if (kn
->parent
!= new_parent
)
3057 * This isn't a proper migration and its usefulness is very
3058 * limited. Disallow on the default hierarchy.
3060 if (cgroup_on_dfl(cgrp
))
3064 * We're gonna grab cgroup_mutex which nests outside kernfs
3065 * active_ref. kernfs_rename() doesn't require active_ref
3066 * protection. Break them before grabbing cgroup_mutex.
3068 kernfs_break_active_protection(new_parent
);
3069 kernfs_break_active_protection(kn
);
3071 mutex_lock(&cgroup_mutex
);
3073 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3075 mutex_unlock(&cgroup_mutex
);
3077 kernfs_unbreak_active_protection(kn
);
3078 kernfs_unbreak_active_protection(new_parent
);
3082 /* set uid and gid of cgroup dirs and files to that of the creator */
3083 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3085 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3086 .ia_uid
= current_fsuid(),
3087 .ia_gid
= current_fsgid(), };
3089 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3090 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3093 return kernfs_setattr(kn
, &iattr
);
3096 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
3098 char name
[CGROUP_FILE_NAME_MAX
];
3099 struct kernfs_node
*kn
;
3100 struct lock_class_key
*key
= NULL
;
3103 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3104 key
= &cft
->lockdep_key
;
3106 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3107 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3112 ret
= cgroup_kn_set_ugid(kn
);
3118 if (cft
->write
== cgroup_procs_write
)
3119 cgrp
->procs_kn
= kn
;
3120 else if (cft
->seq_show
== cgroup_populated_show
)
3121 cgrp
->populated_kn
= kn
;
3126 * cgroup_addrm_files - add or remove files to a cgroup directory
3127 * @cgrp: the target cgroup
3128 * @cfts: array of cftypes to be added
3129 * @is_add: whether to add or remove
3131 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3132 * For removals, this function never fails. If addition fails, this
3133 * function doesn't remove files already added. The caller is responsible
3136 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
3142 lockdep_assert_held(&cgroup_mutex
);
3144 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3145 /* does cft->flags tell us to skip this file on @cgrp? */
3146 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3148 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3150 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3152 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3156 ret
= cgroup_add_file(cgrp
, cft
);
3158 pr_warn("%s: failed to add %s, err=%d\n",
3159 __func__
, cft
->name
, ret
);
3163 cgroup_rm_file(cgrp
, cft
);
3169 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3172 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3173 struct cgroup
*root
= &ss
->root
->cgrp
;
3174 struct cgroup_subsys_state
*css
;
3177 lockdep_assert_held(&cgroup_mutex
);
3179 /* add/rm files for all cgroups created before */
3180 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3181 struct cgroup
*cgrp
= css
->cgroup
;
3183 if (cgroup_is_dead(cgrp
))
3186 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
3192 kernfs_activate(root
->kn
);
3196 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3200 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3201 /* free copy for custom atomic_write_len, see init_cftypes() */
3202 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3207 /* revert flags set by cgroup core while adding @cfts */
3208 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3212 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3216 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3217 struct kernfs_ops
*kf_ops
;
3219 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3222 kf_ops
= &cgroup_kf_ops
;
3224 kf_ops
= &cgroup_kf_single_ops
;
3227 * Ugh... if @cft wants a custom max_write_len, we need to
3228 * make a copy of kf_ops to set its atomic_write_len.
3230 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3231 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3233 cgroup_exit_cftypes(cfts
);
3236 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3239 cft
->kf_ops
= kf_ops
;
3246 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3248 lockdep_assert_held(&cgroup_mutex
);
3250 if (!cfts
|| !cfts
[0].ss
)
3253 list_del(&cfts
->node
);
3254 cgroup_apply_cftypes(cfts
, false);
3255 cgroup_exit_cftypes(cfts
);
3260 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3261 * @cfts: zero-length name terminated array of cftypes
3263 * Unregister @cfts. Files described by @cfts are removed from all
3264 * existing cgroups and all future cgroups won't have them either. This
3265 * function can be called anytime whether @cfts' subsys is attached or not.
3267 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3270 int cgroup_rm_cftypes(struct cftype
*cfts
)
3274 mutex_lock(&cgroup_mutex
);
3275 ret
= cgroup_rm_cftypes_locked(cfts
);
3276 mutex_unlock(&cgroup_mutex
);
3281 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3282 * @ss: target cgroup subsystem
3283 * @cfts: zero-length name terminated array of cftypes
3285 * Register @cfts to @ss. Files described by @cfts are created for all
3286 * existing cgroups to which @ss is attached and all future cgroups will
3287 * have them too. This function can be called anytime whether @ss is
3290 * Returns 0 on successful registration, -errno on failure. Note that this
3291 * function currently returns 0 as long as @cfts registration is successful
3292 * even if some file creation attempts on existing cgroups fail.
3294 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3301 if (!cfts
|| cfts
[0].name
[0] == '\0')
3304 ret
= cgroup_init_cftypes(ss
, cfts
);
3308 mutex_lock(&cgroup_mutex
);
3310 list_add_tail(&cfts
->node
, &ss
->cfts
);
3311 ret
= cgroup_apply_cftypes(cfts
, true);
3313 cgroup_rm_cftypes_locked(cfts
);
3315 mutex_unlock(&cgroup_mutex
);
3320 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3321 * @ss: target cgroup subsystem
3322 * @cfts: zero-length name terminated array of cftypes
3324 * Similar to cgroup_add_cftypes() but the added files are only used for
3325 * the default hierarchy.
3327 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3331 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3332 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3333 return cgroup_add_cftypes(ss
, cfts
);
3337 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3338 * @ss: target cgroup subsystem
3339 * @cfts: zero-length name terminated array of cftypes
3341 * Similar to cgroup_add_cftypes() but the added files are only used for
3342 * the legacy hierarchies.
3344 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3349 * If legacy_flies_on_dfl, we want to show the legacy files on the
3350 * dfl hierarchy but iff the target subsystem hasn't been updated
3351 * for the dfl hierarchy yet.
3353 if (!cgroup_legacy_files_on_dfl
||
3354 ss
->dfl_cftypes
!= ss
->legacy_cftypes
) {
3355 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3356 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3359 return cgroup_add_cftypes(ss
, cfts
);
3363 * cgroup_task_count - count the number of tasks in a cgroup.
3364 * @cgrp: the cgroup in question
3366 * Return the number of tasks in the cgroup.
3368 static int cgroup_task_count(const struct cgroup
*cgrp
)
3371 struct cgrp_cset_link
*link
;
3373 down_read(&css_set_rwsem
);
3374 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3375 count
+= atomic_read(&link
->cset
->refcount
);
3376 up_read(&css_set_rwsem
);
3381 * css_next_child - find the next child of a given css
3382 * @pos: the current position (%NULL to initiate traversal)
3383 * @parent: css whose children to walk
3385 * This function returns the next child of @parent and should be called
3386 * under either cgroup_mutex or RCU read lock. The only requirement is
3387 * that @parent and @pos are accessible. The next sibling is guaranteed to
3388 * be returned regardless of their states.
3390 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3391 * css which finished ->css_online() is guaranteed to be visible in the
3392 * future iterations and will stay visible until the last reference is put.
3393 * A css which hasn't finished ->css_online() or already finished
3394 * ->css_offline() may show up during traversal. It's each subsystem's
3395 * responsibility to synchronize against on/offlining.
3397 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3398 struct cgroup_subsys_state
*parent
)
3400 struct cgroup_subsys_state
*next
;
3402 cgroup_assert_mutex_or_rcu_locked();
3405 * @pos could already have been unlinked from the sibling list.
3406 * Once a cgroup is removed, its ->sibling.next is no longer
3407 * updated when its next sibling changes. CSS_RELEASED is set when
3408 * @pos is taken off list, at which time its next pointer is valid,
3409 * and, as releases are serialized, the one pointed to by the next
3410 * pointer is guaranteed to not have started release yet. This
3411 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3412 * critical section, the one pointed to by its next pointer is
3413 * guaranteed to not have finished its RCU grace period even if we
3414 * have dropped rcu_read_lock() inbetween iterations.
3416 * If @pos has CSS_RELEASED set, its next pointer can't be
3417 * dereferenced; however, as each css is given a monotonically
3418 * increasing unique serial number and always appended to the
3419 * sibling list, the next one can be found by walking the parent's
3420 * children until the first css with higher serial number than
3421 * @pos's. While this path can be slower, it happens iff iteration
3422 * races against release and the race window is very small.
3425 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3426 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3427 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3429 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3430 if (next
->serial_nr
> pos
->serial_nr
)
3435 * @next, if not pointing to the head, can be dereferenced and is
3438 if (&next
->sibling
!= &parent
->children
)
3444 * css_next_descendant_pre - find the next descendant for pre-order walk
3445 * @pos: the current position (%NULL to initiate traversal)
3446 * @root: css whose descendants to walk
3448 * To be used by css_for_each_descendant_pre(). Find the next descendant
3449 * to visit for pre-order traversal of @root's descendants. @root is
3450 * included in the iteration and the first node to be visited.
3452 * While this function requires cgroup_mutex or RCU read locking, it
3453 * doesn't require the whole traversal to be contained in a single critical
3454 * section. This function will return the correct next descendant as long
3455 * as both @pos and @root are accessible and @pos is a descendant of @root.
3457 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3458 * css which finished ->css_online() is guaranteed to be visible in the
3459 * future iterations and will stay visible until the last reference is put.
3460 * A css which hasn't finished ->css_online() or already finished
3461 * ->css_offline() may show up during traversal. It's each subsystem's
3462 * responsibility to synchronize against on/offlining.
3464 struct cgroup_subsys_state
*
3465 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3466 struct cgroup_subsys_state
*root
)
3468 struct cgroup_subsys_state
*next
;
3470 cgroup_assert_mutex_or_rcu_locked();
3472 /* if first iteration, visit @root */
3476 /* visit the first child if exists */
3477 next
= css_next_child(NULL
, pos
);
3481 /* no child, visit my or the closest ancestor's next sibling */
3482 while (pos
!= root
) {
3483 next
= css_next_child(pos
, pos
->parent
);
3493 * css_rightmost_descendant - return the rightmost descendant of a css
3494 * @pos: css of interest
3496 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3497 * is returned. This can be used during pre-order traversal to skip
3500 * While this function requires cgroup_mutex or RCU read locking, it
3501 * doesn't require the whole traversal to be contained in a single critical
3502 * section. This function will return the correct rightmost descendant as
3503 * long as @pos is accessible.
3505 struct cgroup_subsys_state
*
3506 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3508 struct cgroup_subsys_state
*last
, *tmp
;
3510 cgroup_assert_mutex_or_rcu_locked();
3514 /* ->prev isn't RCU safe, walk ->next till the end */
3516 css_for_each_child(tmp
, last
)
3523 static struct cgroup_subsys_state
*
3524 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3526 struct cgroup_subsys_state
*last
;
3530 pos
= css_next_child(NULL
, pos
);
3537 * css_next_descendant_post - find the next descendant for post-order walk
3538 * @pos: the current position (%NULL to initiate traversal)
3539 * @root: css whose descendants to walk
3541 * To be used by css_for_each_descendant_post(). Find the next descendant
3542 * to visit for post-order traversal of @root's descendants. @root is
3543 * included in the iteration and the last node to be visited.
3545 * While this function requires cgroup_mutex or RCU read locking, it
3546 * doesn't require the whole traversal to be contained in a single critical
3547 * section. This function will return the correct next descendant as long
3548 * as both @pos and @cgroup are accessible and @pos is a descendant of
3551 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3552 * css which finished ->css_online() is guaranteed to be visible in the
3553 * future iterations and will stay visible until the last reference is put.
3554 * A css which hasn't finished ->css_online() or already finished
3555 * ->css_offline() may show up during traversal. It's each subsystem's
3556 * responsibility to synchronize against on/offlining.
3558 struct cgroup_subsys_state
*
3559 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3560 struct cgroup_subsys_state
*root
)
3562 struct cgroup_subsys_state
*next
;
3564 cgroup_assert_mutex_or_rcu_locked();
3566 /* if first iteration, visit leftmost descendant which may be @root */
3568 return css_leftmost_descendant(root
);
3570 /* if we visited @root, we're done */
3574 /* if there's an unvisited sibling, visit its leftmost descendant */
3575 next
= css_next_child(pos
, pos
->parent
);
3577 return css_leftmost_descendant(next
);
3579 /* no sibling left, visit parent */
3584 * css_has_online_children - does a css have online children
3585 * @css: the target css
3587 * Returns %true if @css has any online children; otherwise, %false. This
3588 * function can be called from any context but the caller is responsible
3589 * for synchronizing against on/offlining as necessary.
3591 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3593 struct cgroup_subsys_state
*child
;
3597 css_for_each_child(child
, css
) {
3598 if (child
->flags
& CSS_ONLINE
) {
3608 * css_advance_task_iter - advance a task itererator to the next css_set
3609 * @it: the iterator to advance
3611 * Advance @it to the next css_set to walk.
3613 static void css_advance_task_iter(struct css_task_iter
*it
)
3615 struct list_head
*l
= it
->cset_pos
;
3616 struct cgrp_cset_link
*link
;
3617 struct css_set
*cset
;
3619 /* Advance to the next non-empty css_set */
3622 if (l
== it
->cset_head
) {
3623 it
->cset_pos
= NULL
;
3628 cset
= container_of(l
, struct css_set
,
3629 e_cset_node
[it
->ss
->id
]);
3631 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3634 } while (list_empty(&cset
->tasks
) && list_empty(&cset
->mg_tasks
));
3638 if (!list_empty(&cset
->tasks
))
3639 it
->task_pos
= cset
->tasks
.next
;
3641 it
->task_pos
= cset
->mg_tasks
.next
;
3643 it
->tasks_head
= &cset
->tasks
;
3644 it
->mg_tasks_head
= &cset
->mg_tasks
;
3648 * css_task_iter_start - initiate task iteration
3649 * @css: the css to walk tasks of
3650 * @it: the task iterator to use
3652 * Initiate iteration through the tasks of @css. The caller can call
3653 * css_task_iter_next() to walk through the tasks until the function
3654 * returns NULL. On completion of iteration, css_task_iter_end() must be
3657 * Note that this function acquires a lock which is released when the
3658 * iteration finishes. The caller can't sleep while iteration is in
3661 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3662 struct css_task_iter
*it
)
3663 __acquires(css_set_rwsem
)
3665 /* no one should try to iterate before mounting cgroups */
3666 WARN_ON_ONCE(!use_task_css_set_links
);
3668 down_read(&css_set_rwsem
);
3673 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3675 it
->cset_pos
= &css
->cgroup
->cset_links
;
3677 it
->cset_head
= it
->cset_pos
;
3679 css_advance_task_iter(it
);
3683 * css_task_iter_next - return the next task for the iterator
3684 * @it: the task iterator being iterated
3686 * The "next" function for task iteration. @it should have been
3687 * initialized via css_task_iter_start(). Returns NULL when the iteration
3690 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3692 struct task_struct
*res
;
3693 struct list_head
*l
= it
->task_pos
;
3695 /* If the iterator cg is NULL, we have no tasks */
3698 res
= list_entry(l
, struct task_struct
, cg_list
);
3701 * Advance iterator to find next entry. cset->tasks is consumed
3702 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3707 if (l
== it
->tasks_head
)
3708 l
= it
->mg_tasks_head
->next
;
3710 if (l
== it
->mg_tasks_head
)
3711 css_advance_task_iter(it
);
3719 * css_task_iter_end - finish task iteration
3720 * @it: the task iterator to finish
3722 * Finish task iteration started by css_task_iter_start().
3724 void css_task_iter_end(struct css_task_iter
*it
)
3725 __releases(css_set_rwsem
)
3727 up_read(&css_set_rwsem
);
3731 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3732 * @to: cgroup to which the tasks will be moved
3733 * @from: cgroup in which the tasks currently reside
3735 * Locking rules between cgroup_post_fork() and the migration path
3736 * guarantee that, if a task is forking while being migrated, the new child
3737 * is guaranteed to be either visible in the source cgroup after the
3738 * parent's migration is complete or put into the target cgroup. No task
3739 * can slip out of migration through forking.
3741 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3743 LIST_HEAD(preloaded_csets
);
3744 struct cgrp_cset_link
*link
;
3745 struct css_task_iter it
;
3746 struct task_struct
*task
;
3749 mutex_lock(&cgroup_mutex
);
3751 /* all tasks in @from are being moved, all csets are source */
3752 down_read(&css_set_rwsem
);
3753 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3754 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3755 up_read(&css_set_rwsem
);
3757 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
3762 * Migrate tasks one-by-one until @form is empty. This fails iff
3763 * ->can_attach() fails.
3766 css_task_iter_start(&from
->self
, &it
);
3767 task
= css_task_iter_next(&it
);
3769 get_task_struct(task
);
3770 css_task_iter_end(&it
);
3773 ret
= cgroup_migrate(to
, task
, false);
3774 put_task_struct(task
);
3776 } while (task
&& !ret
);
3778 cgroup_migrate_finish(&preloaded_csets
);
3779 mutex_unlock(&cgroup_mutex
);
3784 * Stuff for reading the 'tasks'/'procs' files.
3786 * Reading this file can return large amounts of data if a cgroup has
3787 * *lots* of attached tasks. So it may need several calls to read(),
3788 * but we cannot guarantee that the information we produce is correct
3789 * unless we produce it entirely atomically.
3793 /* which pidlist file are we talking about? */
3794 enum cgroup_filetype
{
3800 * A pidlist is a list of pids that virtually represents the contents of one
3801 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3802 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3805 struct cgroup_pidlist
{
3807 * used to find which pidlist is wanted. doesn't change as long as
3808 * this particular list stays in the list.
3810 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3813 /* how many elements the above list has */
3815 /* each of these stored in a list by its cgroup */
3816 struct list_head links
;
3817 /* pointer to the cgroup we belong to, for list removal purposes */
3818 struct cgroup
*owner
;
3819 /* for delayed destruction */
3820 struct delayed_work destroy_dwork
;
3824 * The following two functions "fix" the issue where there are more pids
3825 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3826 * TODO: replace with a kernel-wide solution to this problem
3828 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3829 static void *pidlist_allocate(int count
)
3831 if (PIDLIST_TOO_LARGE(count
))
3832 return vmalloc(count
* sizeof(pid_t
));
3834 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3837 static void pidlist_free(void *p
)
3843 * Used to destroy all pidlists lingering waiting for destroy timer. None
3844 * should be left afterwards.
3846 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
3848 struct cgroup_pidlist
*l
, *tmp_l
;
3850 mutex_lock(&cgrp
->pidlist_mutex
);
3851 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
3852 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
3853 mutex_unlock(&cgrp
->pidlist_mutex
);
3855 flush_workqueue(cgroup_pidlist_destroy_wq
);
3856 BUG_ON(!list_empty(&cgrp
->pidlists
));
3859 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
3861 struct delayed_work
*dwork
= to_delayed_work(work
);
3862 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
3864 struct cgroup_pidlist
*tofree
= NULL
;
3866 mutex_lock(&l
->owner
->pidlist_mutex
);
3869 * Destroy iff we didn't get queued again. The state won't change
3870 * as destroy_dwork can only be queued while locked.
3872 if (!delayed_work_pending(dwork
)) {
3873 list_del(&l
->links
);
3874 pidlist_free(l
->list
);
3875 put_pid_ns(l
->key
.ns
);
3879 mutex_unlock(&l
->owner
->pidlist_mutex
);
3884 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3885 * Returns the number of unique elements.
3887 static int pidlist_uniq(pid_t
*list
, int length
)
3892 * we presume the 0th element is unique, so i starts at 1. trivial
3893 * edge cases first; no work needs to be done for either
3895 if (length
== 0 || length
== 1)
3897 /* src and dest walk down the list; dest counts unique elements */
3898 for (src
= 1; src
< length
; src
++) {
3899 /* find next unique element */
3900 while (list
[src
] == list
[src
-1]) {
3905 /* dest always points to where the next unique element goes */
3906 list
[dest
] = list
[src
];
3914 * The two pid files - task and cgroup.procs - guaranteed that the result
3915 * is sorted, which forced this whole pidlist fiasco. As pid order is
3916 * different per namespace, each namespace needs differently sorted list,
3917 * making it impossible to use, for example, single rbtree of member tasks
3918 * sorted by task pointer. As pidlists can be fairly large, allocating one
3919 * per open file is dangerous, so cgroup had to implement shared pool of
3920 * pidlists keyed by cgroup and namespace.
3922 * All this extra complexity was caused by the original implementation
3923 * committing to an entirely unnecessary property. In the long term, we
3924 * want to do away with it. Explicitly scramble sort order if on the
3925 * default hierarchy so that no such expectation exists in the new
3928 * Scrambling is done by swapping every two consecutive bits, which is
3929 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3931 static pid_t
pid_fry(pid_t pid
)
3933 unsigned a
= pid
& 0x55555555;
3934 unsigned b
= pid
& 0xAAAAAAAA;
3936 return (a
<< 1) | (b
>> 1);
3939 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
3941 if (cgroup_on_dfl(cgrp
))
3942 return pid_fry(pid
);
3947 static int cmppid(const void *a
, const void *b
)
3949 return *(pid_t
*)a
- *(pid_t
*)b
;
3952 static int fried_cmppid(const void *a
, const void *b
)
3954 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
3957 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3958 enum cgroup_filetype type
)
3960 struct cgroup_pidlist
*l
;
3961 /* don't need task_nsproxy() if we're looking at ourself */
3962 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3964 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3966 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
3967 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
3973 * find the appropriate pidlist for our purpose (given procs vs tasks)
3974 * returns with the lock on that pidlist already held, and takes care
3975 * of the use count, or returns NULL with no locks held if we're out of
3978 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
3979 enum cgroup_filetype type
)
3981 struct cgroup_pidlist
*l
;
3983 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3985 l
= cgroup_pidlist_find(cgrp
, type
);
3989 /* entry not found; create a new one */
3990 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3994 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
3996 /* don't need task_nsproxy() if we're looking at ourself */
3997 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
3999 list_add(&l
->links
, &cgrp
->pidlists
);
4004 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4006 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
4007 struct cgroup_pidlist
**lp
)
4011 int pid
, n
= 0; /* used for populating the array */
4012 struct css_task_iter it
;
4013 struct task_struct
*tsk
;
4014 struct cgroup_pidlist
*l
;
4016 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4019 * If cgroup gets more users after we read count, we won't have
4020 * enough space - tough. This race is indistinguishable to the
4021 * caller from the case that the additional cgroup users didn't
4022 * show up until sometime later on.
4024 length
= cgroup_task_count(cgrp
);
4025 array
= pidlist_allocate(length
);
4028 /* now, populate the array */
4029 css_task_iter_start(&cgrp
->self
, &it
);
4030 while ((tsk
= css_task_iter_next(&it
))) {
4031 if (unlikely(n
== length
))
4033 /* get tgid or pid for procs or tasks file respectively */
4034 if (type
== CGROUP_FILE_PROCS
)
4035 pid
= task_tgid_vnr(tsk
);
4037 pid
= task_pid_vnr(tsk
);
4038 if (pid
> 0) /* make sure to only use valid results */
4041 css_task_iter_end(&it
);
4043 /* now sort & (if procs) strip out duplicates */
4044 if (cgroup_on_dfl(cgrp
))
4045 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
4047 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
4048 if (type
== CGROUP_FILE_PROCS
)
4049 length
= pidlist_uniq(array
, length
);
4051 l
= cgroup_pidlist_find_create(cgrp
, type
);
4053 pidlist_free(array
);
4057 /* store array, freeing old if necessary */
4058 pidlist_free(l
->list
);
4066 * cgroupstats_build - build and fill cgroupstats
4067 * @stats: cgroupstats to fill information into
4068 * @dentry: A dentry entry belonging to the cgroup for which stats have
4071 * Build and fill cgroupstats so that taskstats can export it to user
4074 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4076 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4077 struct cgroup
*cgrp
;
4078 struct css_task_iter it
;
4079 struct task_struct
*tsk
;
4081 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4082 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4083 kernfs_type(kn
) != KERNFS_DIR
)
4086 mutex_lock(&cgroup_mutex
);
4089 * We aren't being called from kernfs and there's no guarantee on
4090 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4091 * @kn->priv is RCU safe. Let's do the RCU dancing.
4094 cgrp
= rcu_dereference(kn
->priv
);
4095 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4097 mutex_unlock(&cgroup_mutex
);
4102 css_task_iter_start(&cgrp
->self
, &it
);
4103 while ((tsk
= css_task_iter_next(&it
))) {
4104 switch (tsk
->state
) {
4106 stats
->nr_running
++;
4108 case TASK_INTERRUPTIBLE
:
4109 stats
->nr_sleeping
++;
4111 case TASK_UNINTERRUPTIBLE
:
4112 stats
->nr_uninterruptible
++;
4115 stats
->nr_stopped
++;
4118 if (delayacct_is_task_waiting_on_io(tsk
))
4119 stats
->nr_io_wait
++;
4123 css_task_iter_end(&it
);
4125 mutex_unlock(&cgroup_mutex
);
4131 * seq_file methods for the tasks/procs files. The seq_file position is the
4132 * next pid to display; the seq_file iterator is a pointer to the pid
4133 * in the cgroup->l->list array.
4136 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4139 * Initially we receive a position value that corresponds to
4140 * one more than the last pid shown (or 0 on the first call or
4141 * after a seek to the start). Use a binary-search to find the
4142 * next pid to display, if any
4144 struct kernfs_open_file
*of
= s
->private;
4145 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4146 struct cgroup_pidlist
*l
;
4147 enum cgroup_filetype type
= seq_cft(s
)->private;
4148 int index
= 0, pid
= *pos
;
4151 mutex_lock(&cgrp
->pidlist_mutex
);
4154 * !NULL @of->priv indicates that this isn't the first start()
4155 * after open. If the matching pidlist is around, we can use that.
4156 * Look for it. Note that @of->priv can't be used directly. It
4157 * could already have been destroyed.
4160 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4163 * Either this is the first start() after open or the matching
4164 * pidlist has been destroyed inbetween. Create a new one.
4167 ret
= pidlist_array_load(cgrp
, type
,
4168 (struct cgroup_pidlist
**)&of
->priv
);
4170 return ERR_PTR(ret
);
4175 int end
= l
->length
;
4177 while (index
< end
) {
4178 int mid
= (index
+ end
) / 2;
4179 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4182 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4188 /* If we're off the end of the array, we're done */
4189 if (index
>= l
->length
)
4191 /* Update the abstract position to be the actual pid that we found */
4192 iter
= l
->list
+ index
;
4193 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4197 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4199 struct kernfs_open_file
*of
= s
->private;
4200 struct cgroup_pidlist
*l
= of
->priv
;
4203 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4204 CGROUP_PIDLIST_DESTROY_DELAY
);
4205 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4208 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4210 struct kernfs_open_file
*of
= s
->private;
4211 struct cgroup_pidlist
*l
= of
->priv
;
4213 pid_t
*end
= l
->list
+ l
->length
;
4215 * Advance to the next pid in the array. If this goes off the
4222 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4227 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4229 seq_printf(s
, "%d\n", *(int *)v
);
4234 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4237 return notify_on_release(css
->cgroup
);
4240 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4241 struct cftype
*cft
, u64 val
)
4244 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4246 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4250 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4253 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4256 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4257 struct cftype
*cft
, u64 val
)
4260 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4262 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4266 /* cgroup core interface files for the default hierarchy */
4267 static struct cftype cgroup_dfl_base_files
[] = {
4269 .name
= "cgroup.procs",
4270 .seq_start
= cgroup_pidlist_start
,
4271 .seq_next
= cgroup_pidlist_next
,
4272 .seq_stop
= cgroup_pidlist_stop
,
4273 .seq_show
= cgroup_pidlist_show
,
4274 .private = CGROUP_FILE_PROCS
,
4275 .write
= cgroup_procs_write
,
4276 .mode
= S_IRUGO
| S_IWUSR
,
4279 .name
= "cgroup.controllers",
4280 .flags
= CFTYPE_ONLY_ON_ROOT
,
4281 .seq_show
= cgroup_root_controllers_show
,
4284 .name
= "cgroup.controllers",
4285 .flags
= CFTYPE_NOT_ON_ROOT
,
4286 .seq_show
= cgroup_controllers_show
,
4289 .name
= "cgroup.subtree_control",
4290 .seq_show
= cgroup_subtree_control_show
,
4291 .write
= cgroup_subtree_control_write
,
4294 .name
= "cgroup.populated",
4295 .flags
= CFTYPE_NOT_ON_ROOT
,
4296 .seq_show
= cgroup_populated_show
,
4301 /* cgroup core interface files for the legacy hierarchies */
4302 static struct cftype cgroup_legacy_base_files
[] = {
4304 .name
= "cgroup.procs",
4305 .seq_start
= cgroup_pidlist_start
,
4306 .seq_next
= cgroup_pidlist_next
,
4307 .seq_stop
= cgroup_pidlist_stop
,
4308 .seq_show
= cgroup_pidlist_show
,
4309 .private = CGROUP_FILE_PROCS
,
4310 .write
= cgroup_procs_write
,
4311 .mode
= S_IRUGO
| S_IWUSR
,
4314 .name
= "cgroup.clone_children",
4315 .read_u64
= cgroup_clone_children_read
,
4316 .write_u64
= cgroup_clone_children_write
,
4319 .name
= "cgroup.sane_behavior",
4320 .flags
= CFTYPE_ONLY_ON_ROOT
,
4321 .seq_show
= cgroup_sane_behavior_show
,
4325 .seq_start
= cgroup_pidlist_start
,
4326 .seq_next
= cgroup_pidlist_next
,
4327 .seq_stop
= cgroup_pidlist_stop
,
4328 .seq_show
= cgroup_pidlist_show
,
4329 .private = CGROUP_FILE_TASKS
,
4330 .write
= cgroup_tasks_write
,
4331 .mode
= S_IRUGO
| S_IWUSR
,
4334 .name
= "notify_on_release",
4335 .read_u64
= cgroup_read_notify_on_release
,
4336 .write_u64
= cgroup_write_notify_on_release
,
4339 .name
= "release_agent",
4340 .flags
= CFTYPE_ONLY_ON_ROOT
,
4341 .seq_show
= cgroup_release_agent_show
,
4342 .write
= cgroup_release_agent_write
,
4343 .max_write_len
= PATH_MAX
- 1,
4349 * cgroup_populate_dir - create subsys files in a cgroup directory
4350 * @cgrp: target cgroup
4351 * @subsys_mask: mask of the subsystem ids whose files should be added
4353 * On failure, no file is added.
4355 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4357 struct cgroup_subsys
*ss
;
4360 /* process cftsets of each subsystem */
4361 for_each_subsys(ss
, i
) {
4362 struct cftype
*cfts
;
4364 if (!(subsys_mask
& (1 << i
)))
4367 list_for_each_entry(cfts
, &ss
->cfts
, node
) {
4368 ret
= cgroup_addrm_files(cgrp
, cfts
, true);
4375 cgroup_clear_dir(cgrp
, subsys_mask
);
4380 * css destruction is four-stage process.
4382 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4383 * Implemented in kill_css().
4385 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4386 * and thus css_tryget_online() is guaranteed to fail, the css can be
4387 * offlined by invoking offline_css(). After offlining, the base ref is
4388 * put. Implemented in css_killed_work_fn().
4390 * 3. When the percpu_ref reaches zero, the only possible remaining
4391 * accessors are inside RCU read sections. css_release() schedules the
4394 * 4. After the grace period, the css can be freed. Implemented in
4395 * css_free_work_fn().
4397 * It is actually hairier because both step 2 and 4 require process context
4398 * and thus involve punting to css->destroy_work adding two additional
4399 * steps to the already complex sequence.
4401 static void css_free_work_fn(struct work_struct
*work
)
4403 struct cgroup_subsys_state
*css
=
4404 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4405 struct cgroup_subsys
*ss
= css
->ss
;
4406 struct cgroup
*cgrp
= css
->cgroup
;
4408 percpu_ref_exit(&css
->refcnt
);
4415 css_put(css
->parent
);
4418 cgroup_idr_remove(&ss
->css_idr
, id
);
4421 /* cgroup free path */
4422 atomic_dec(&cgrp
->root
->nr_cgrps
);
4423 cgroup_pidlist_destroy_all(cgrp
);
4424 cancel_work_sync(&cgrp
->release_agent_work
);
4426 if (cgroup_parent(cgrp
)) {
4428 * We get a ref to the parent, and put the ref when
4429 * this cgroup is being freed, so it's guaranteed
4430 * that the parent won't be destroyed before its
4433 cgroup_put(cgroup_parent(cgrp
));
4434 kernfs_put(cgrp
->kn
);
4438 * This is root cgroup's refcnt reaching zero,
4439 * which indicates that the root should be
4442 cgroup_destroy_root(cgrp
->root
);
4447 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4449 struct cgroup_subsys_state
*css
=
4450 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4452 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4453 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4456 static void css_release_work_fn(struct work_struct
*work
)
4458 struct cgroup_subsys_state
*css
=
4459 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4460 struct cgroup_subsys
*ss
= css
->ss
;
4461 struct cgroup
*cgrp
= css
->cgroup
;
4463 mutex_lock(&cgroup_mutex
);
4465 css
->flags
|= CSS_RELEASED
;
4466 list_del_rcu(&css
->sibling
);
4469 /* css release path */
4470 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4471 if (ss
->css_released
)
4472 ss
->css_released(css
);
4474 /* cgroup release path */
4475 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4479 * There are two control paths which try to determine
4480 * cgroup from dentry without going through kernfs -
4481 * cgroupstats_build() and css_tryget_online_from_dir().
4482 * Those are supported by RCU protecting clearing of
4483 * cgrp->kn->priv backpointer.
4485 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4488 mutex_unlock(&cgroup_mutex
);
4490 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4493 static void css_release(struct percpu_ref
*ref
)
4495 struct cgroup_subsys_state
*css
=
4496 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4498 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4499 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4502 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4503 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4505 lockdep_assert_held(&cgroup_mutex
);
4509 memset(css
, 0, sizeof(*css
));
4512 INIT_LIST_HEAD(&css
->sibling
);
4513 INIT_LIST_HEAD(&css
->children
);
4514 css
->serial_nr
= css_serial_nr_next
++;
4516 if (cgroup_parent(cgrp
)) {
4517 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4518 css_get(css
->parent
);
4521 BUG_ON(cgroup_css(cgrp
, ss
));
4524 /* invoke ->css_online() on a new CSS and mark it online if successful */
4525 static int online_css(struct cgroup_subsys_state
*css
)
4527 struct cgroup_subsys
*ss
= css
->ss
;
4530 lockdep_assert_held(&cgroup_mutex
);
4533 ret
= ss
->css_online(css
);
4535 css
->flags
|= CSS_ONLINE
;
4536 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4541 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4542 static void offline_css(struct cgroup_subsys_state
*css
)
4544 struct cgroup_subsys
*ss
= css
->ss
;
4546 lockdep_assert_held(&cgroup_mutex
);
4548 if (!(css
->flags
& CSS_ONLINE
))
4551 if (ss
->css_offline
)
4552 ss
->css_offline(css
);
4554 css
->flags
&= ~CSS_ONLINE
;
4555 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4557 wake_up_all(&css
->cgroup
->offline_waitq
);
4561 * create_css - create a cgroup_subsys_state
4562 * @cgrp: the cgroup new css will be associated with
4563 * @ss: the subsys of new css
4564 * @visible: whether to create control knobs for the new css or not
4566 * Create a new css associated with @cgrp - @ss pair. On success, the new
4567 * css is online and installed in @cgrp with all interface files created if
4568 * @visible. Returns 0 on success, -errno on failure.
4570 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4573 struct cgroup
*parent
= cgroup_parent(cgrp
);
4574 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4575 struct cgroup_subsys_state
*css
;
4578 lockdep_assert_held(&cgroup_mutex
);
4580 css
= ss
->css_alloc(parent_css
);
4582 return PTR_ERR(css
);
4584 init_and_link_css(css
, ss
, cgrp
);
4586 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4590 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4592 goto err_free_percpu_ref
;
4596 err
= cgroup_populate_dir(cgrp
, 1 << ss
->id
);
4601 /* @css is ready to be brought online now, make it visible */
4602 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4603 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4605 err
= online_css(css
);
4609 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4610 cgroup_parent(parent
)) {
4611 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4612 current
->comm
, current
->pid
, ss
->name
);
4613 if (!strcmp(ss
->name
, "memory"))
4614 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4615 ss
->warned_broken_hierarchy
= true;
4621 list_del_rcu(&css
->sibling
);
4622 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4624 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4625 err_free_percpu_ref
:
4626 percpu_ref_exit(&css
->refcnt
);
4628 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4632 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4635 struct cgroup
*parent
, *cgrp
;
4636 struct cgroup_root
*root
;
4637 struct cgroup_subsys
*ss
;
4638 struct kernfs_node
*kn
;
4639 struct cftype
*base_files
;
4642 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4644 if (strchr(name
, '\n'))
4647 parent
= cgroup_kn_lock_live(parent_kn
);
4650 root
= parent
->root
;
4652 /* allocate the cgroup and its ID, 0 is reserved for the root */
4653 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4659 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
4664 * Temporarily set the pointer to NULL, so idr_find() won't return
4665 * a half-baked cgroup.
4667 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
4670 goto out_cancel_ref
;
4673 init_cgroup_housekeeping(cgrp
);
4675 cgrp
->self
.parent
= &parent
->self
;
4678 if (notify_on_release(parent
))
4679 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4681 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4682 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4684 /* create the directory */
4685 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4693 * This extra ref will be put in cgroup_free_fn() and guarantees
4694 * that @cgrp->kn is always accessible.
4698 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4700 /* allocation complete, commit to creation */
4701 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4702 atomic_inc(&root
->nr_cgrps
);
4706 * @cgrp is now fully operational. If something fails after this
4707 * point, it'll be released via the normal destruction path.
4709 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4711 ret
= cgroup_kn_set_ugid(kn
);
4715 if (cgroup_on_dfl(cgrp
))
4716 base_files
= cgroup_dfl_base_files
;
4718 base_files
= cgroup_legacy_base_files
;
4720 ret
= cgroup_addrm_files(cgrp
, base_files
, true);
4724 /* let's create and online css's */
4725 for_each_subsys(ss
, ssid
) {
4726 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4727 ret
= create_css(cgrp
, ss
,
4728 parent
->subtree_control
& (1 << ssid
));
4735 * On the default hierarchy, a child doesn't automatically inherit
4736 * subtree_control from the parent. Each is configured manually.
4738 if (!cgroup_on_dfl(cgrp
)) {
4739 cgrp
->subtree_control
= parent
->subtree_control
;
4740 cgroup_refresh_child_subsys_mask(cgrp
);
4743 kernfs_activate(kn
);
4749 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4751 percpu_ref_exit(&cgrp
->self
.refcnt
);
4755 cgroup_kn_unlock(parent_kn
);
4759 cgroup_destroy_locked(cgrp
);
4764 * This is called when the refcnt of a css is confirmed to be killed.
4765 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4766 * initate destruction and put the css ref from kill_css().
4768 static void css_killed_work_fn(struct work_struct
*work
)
4770 struct cgroup_subsys_state
*css
=
4771 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4773 mutex_lock(&cgroup_mutex
);
4775 mutex_unlock(&cgroup_mutex
);
4780 /* css kill confirmation processing requires process context, bounce */
4781 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4783 struct cgroup_subsys_state
*css
=
4784 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4786 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4787 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4791 * kill_css - destroy a css
4792 * @css: css to destroy
4794 * This function initiates destruction of @css by removing cgroup interface
4795 * files and putting its base reference. ->css_offline() will be invoked
4796 * asynchronously once css_tryget_online() is guaranteed to fail and when
4797 * the reference count reaches zero, @css will be released.
4799 static void kill_css(struct cgroup_subsys_state
*css
)
4801 lockdep_assert_held(&cgroup_mutex
);
4804 * This must happen before css is disassociated with its cgroup.
4805 * See seq_css() for details.
4807 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4810 * Killing would put the base ref, but we need to keep it alive
4811 * until after ->css_offline().
4816 * cgroup core guarantees that, by the time ->css_offline() is
4817 * invoked, no new css reference will be given out via
4818 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4819 * proceed to offlining css's because percpu_ref_kill() doesn't
4820 * guarantee that the ref is seen as killed on all CPUs on return.
4822 * Use percpu_ref_kill_and_confirm() to get notifications as each
4823 * css is confirmed to be seen as killed on all CPUs.
4825 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4829 * cgroup_destroy_locked - the first stage of cgroup destruction
4830 * @cgrp: cgroup to be destroyed
4832 * css's make use of percpu refcnts whose killing latency shouldn't be
4833 * exposed to userland and are RCU protected. Also, cgroup core needs to
4834 * guarantee that css_tryget_online() won't succeed by the time
4835 * ->css_offline() is invoked. To satisfy all the requirements,
4836 * destruction is implemented in the following two steps.
4838 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4839 * userland visible parts and start killing the percpu refcnts of
4840 * css's. Set up so that the next stage will be kicked off once all
4841 * the percpu refcnts are confirmed to be killed.
4843 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4844 * rest of destruction. Once all cgroup references are gone, the
4845 * cgroup is RCU-freed.
4847 * This function implements s1. After this step, @cgrp is gone as far as
4848 * the userland is concerned and a new cgroup with the same name may be
4849 * created. As cgroup doesn't care about the names internally, this
4850 * doesn't cause any problem.
4852 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4853 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4855 struct cgroup_subsys_state
*css
;
4859 lockdep_assert_held(&cgroup_mutex
);
4862 * css_set_rwsem synchronizes access to ->cset_links and prevents
4863 * @cgrp from being removed while put_css_set() is in progress.
4865 down_read(&css_set_rwsem
);
4866 empty
= list_empty(&cgrp
->cset_links
);
4867 up_read(&css_set_rwsem
);
4872 * Make sure there's no live children. We can't test emptiness of
4873 * ->self.children as dead children linger on it while being
4874 * drained; otherwise, "rmdir parent/child parent" may fail.
4876 if (css_has_online_children(&cgrp
->self
))
4880 * Mark @cgrp dead. This prevents further task migration and child
4881 * creation by disabling cgroup_lock_live_group().
4883 cgrp
->self
.flags
&= ~CSS_ONLINE
;
4885 /* initiate massacre of all css's */
4886 for_each_css(css
, ssid
, cgrp
)
4890 * Remove @cgrp directory along with the base files. @cgrp has an
4891 * extra ref on its kn.
4893 kernfs_remove(cgrp
->kn
);
4895 check_for_release(cgroup_parent(cgrp
));
4897 /* put the base reference */
4898 percpu_ref_kill(&cgrp
->self
.refcnt
);
4903 static int cgroup_rmdir(struct kernfs_node
*kn
)
4905 struct cgroup
*cgrp
;
4908 cgrp
= cgroup_kn_lock_live(kn
);
4912 ret
= cgroup_destroy_locked(cgrp
);
4914 cgroup_kn_unlock(kn
);
4918 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
4919 .remount_fs
= cgroup_remount
,
4920 .show_options
= cgroup_show_options
,
4921 .mkdir
= cgroup_mkdir
,
4922 .rmdir
= cgroup_rmdir
,
4923 .rename
= cgroup_rename
,
4926 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
4928 struct cgroup_subsys_state
*css
;
4930 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4932 mutex_lock(&cgroup_mutex
);
4934 idr_init(&ss
->css_idr
);
4935 INIT_LIST_HEAD(&ss
->cfts
);
4937 /* Create the root cgroup state for this subsystem */
4938 ss
->root
= &cgrp_dfl_root
;
4939 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
4940 /* We don't handle early failures gracefully */
4941 BUG_ON(IS_ERR(css
));
4942 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
4945 * Root csses are never destroyed and we can't initialize
4946 * percpu_ref during early init. Disable refcnting.
4948 css
->flags
|= CSS_NO_REF
;
4951 /* allocation can't be done safely during early init */
4954 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
4955 BUG_ON(css
->id
< 0);
4958 /* Update the init_css_set to contain a subsys
4959 * pointer to this state - since the subsystem is
4960 * newly registered, all tasks and hence the
4961 * init_css_set is in the subsystem's root cgroup. */
4962 init_css_set
.subsys
[ss
->id
] = css
;
4964 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
4965 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
4966 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
4968 /* At system boot, before all subsystems have been
4969 * registered, no tasks have been forked, so we don't
4970 * need to invoke fork callbacks here. */
4971 BUG_ON(!list_empty(&init_task
.tasks
));
4973 BUG_ON(online_css(css
));
4975 mutex_unlock(&cgroup_mutex
);
4979 * cgroup_init_early - cgroup initialization at system boot
4981 * Initialize cgroups at system boot, and initialize any
4982 * subsystems that request early init.
4984 int __init
cgroup_init_early(void)
4986 static struct cgroup_sb_opts __initdata opts
;
4987 struct cgroup_subsys
*ss
;
4990 init_cgroup_root(&cgrp_dfl_root
, &opts
);
4991 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
4993 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4995 for_each_subsys(ss
, i
) {
4996 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
4997 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4998 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5000 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5001 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5004 ss
->name
= cgroup_subsys_name
[i
];
5005 if (!ss
->legacy_name
)
5006 ss
->legacy_name
= cgroup_subsys_name
[i
];
5009 cgroup_init_subsys(ss
, true);
5015 * cgroup_init - cgroup initialization
5017 * Register cgroup filesystem and /proc file, and initialize
5018 * any subsystems that didn't request early init.
5020 int __init
cgroup_init(void)
5022 struct cgroup_subsys
*ss
;
5026 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5027 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
5028 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
5030 mutex_lock(&cgroup_mutex
);
5032 /* Add init_css_set to the hash table */
5033 key
= css_set_hash(init_css_set
.subsys
);
5034 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5036 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5038 mutex_unlock(&cgroup_mutex
);
5040 for_each_subsys(ss
, ssid
) {
5041 if (ss
->early_init
) {
5042 struct cgroup_subsys_state
*css
=
5043 init_css_set
.subsys
[ss
->id
];
5045 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5047 BUG_ON(css
->id
< 0);
5049 cgroup_init_subsys(ss
, false);
5052 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5053 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5056 * Setting dfl_root subsys_mask needs to consider the
5057 * disabled flag and cftype registration needs kmalloc,
5058 * both of which aren't available during early_init.
5063 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5065 if (cgroup_legacy_files_on_dfl
&& !ss
->dfl_cftypes
)
5066 ss
->dfl_cftypes
= ss
->legacy_cftypes
;
5068 if (!ss
->dfl_cftypes
)
5069 cgrp_dfl_root_inhibit_ss_mask
|= 1 << ss
->id
;
5071 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5072 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5074 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5075 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5079 ss
->bind(init_css_set
.subsys
[ssid
]);
5082 err
= sysfs_create_mount_point(fs_kobj
, "cgroup");
5086 err
= register_filesystem(&cgroup_fs_type
);
5088 sysfs_remove_mount_point(fs_kobj
, "cgroup");
5092 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5096 static int __init
cgroup_wq_init(void)
5099 * There isn't much point in executing destruction path in
5100 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5101 * Use 1 for @max_active.
5103 * We would prefer to do this in cgroup_init() above, but that
5104 * is called before init_workqueues(): so leave this until after.
5106 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5107 BUG_ON(!cgroup_destroy_wq
);
5110 * Used to destroy pidlists and separate to serve as flush domain.
5111 * Cap @max_active to 1 too.
5113 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5115 BUG_ON(!cgroup_pidlist_destroy_wq
);
5119 core_initcall(cgroup_wq_init
);
5122 * proc_cgroup_show()
5123 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5124 * - Used for /proc/<pid>/cgroup.
5126 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5127 struct pid
*pid
, struct task_struct
*tsk
)
5131 struct cgroup_root
*root
;
5134 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5138 mutex_lock(&cgroup_mutex
);
5139 down_read(&css_set_rwsem
);
5141 for_each_root(root
) {
5142 struct cgroup_subsys
*ss
;
5143 struct cgroup
*cgrp
;
5144 int ssid
, count
= 0;
5146 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5149 seq_printf(m
, "%d:", root
->hierarchy_id
);
5150 if (root
!= &cgrp_dfl_root
)
5151 for_each_subsys(ss
, ssid
)
5152 if (root
->subsys_mask
& (1 << ssid
))
5153 seq_printf(m
, "%s%s", count
++ ? "," : "",
5155 if (strlen(root
->name
))
5156 seq_printf(m
, "%sname=%s", count
? "," : "",
5159 cgrp
= task_cgroup_from_root(tsk
, root
);
5160 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
5162 retval
= -ENAMETOOLONG
;
5171 up_read(&css_set_rwsem
);
5172 mutex_unlock(&cgroup_mutex
);
5178 /* Display information about each subsystem and each hierarchy */
5179 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5181 struct cgroup_subsys
*ss
;
5184 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5186 * ideally we don't want subsystems moving around while we do this.
5187 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5188 * subsys/hierarchy state.
5190 mutex_lock(&cgroup_mutex
);
5192 for_each_subsys(ss
, i
)
5193 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5194 ss
->legacy_name
, ss
->root
->hierarchy_id
,
5195 atomic_read(&ss
->root
->nr_cgrps
), !ss
->disabled
);
5197 mutex_unlock(&cgroup_mutex
);
5201 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5203 return single_open(file
, proc_cgroupstats_show
, NULL
);
5206 static const struct file_operations proc_cgroupstats_operations
= {
5207 .open
= cgroupstats_open
,
5209 .llseek
= seq_lseek
,
5210 .release
= single_release
,
5213 static void **subsys_canfork_priv_p(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5215 if (CGROUP_CANFORK_START
<= i
&& i
< CGROUP_CANFORK_END
)
5216 return &ss_priv
[i
- CGROUP_CANFORK_START
];
5220 static void *subsys_canfork_priv(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5222 void **private = subsys_canfork_priv_p(ss_priv
, i
);
5223 return private ? *private : NULL
;
5227 * cgroup_fork - initialize cgroup related fields during copy_process()
5228 * @child: pointer to task_struct of forking parent process.
5230 * A task is associated with the init_css_set until cgroup_post_fork()
5231 * attaches it to the parent's css_set. Empty cg_list indicates that
5232 * @child isn't holding reference to its css_set.
5234 void cgroup_fork(struct task_struct
*child
)
5236 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5237 INIT_LIST_HEAD(&child
->cg_list
);
5241 * cgroup_can_fork - called on a new task before the process is exposed
5242 * @child: the task in question.
5244 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5245 * returns an error, the fork aborts with that error code. This allows for
5246 * a cgroup subsystem to conditionally allow or deny new forks.
5248 int cgroup_can_fork(struct task_struct
*child
,
5249 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5251 struct cgroup_subsys
*ss
;
5254 for_each_subsys_which(ss
, i
, &have_canfork_callback
) {
5255 ret
= ss
->can_fork(child
, subsys_canfork_priv_p(ss_priv
, i
));
5263 for_each_subsys(ss
, j
) {
5266 if (ss
->cancel_fork
)
5267 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, j
));
5274 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5275 * @child: the task in question
5277 * This calls the cancel_fork() callbacks if a fork failed *after*
5278 * cgroup_can_fork() succeded.
5280 void cgroup_cancel_fork(struct task_struct
*child
,
5281 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5283 struct cgroup_subsys
*ss
;
5286 for_each_subsys(ss
, i
)
5287 if (ss
->cancel_fork
)
5288 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, i
));
5292 * cgroup_post_fork - called on a new task after adding it to the task list
5293 * @child: the task in question
5295 * Adds the task to the list running through its css_set if necessary and
5296 * call the subsystem fork() callbacks. Has to be after the task is
5297 * visible on the task list in case we race with the first call to
5298 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5301 void cgroup_post_fork(struct task_struct
*child
,
5302 void *old_ss_priv
[CGROUP_CANFORK_COUNT
])
5304 struct cgroup_subsys
*ss
;
5308 * This may race against cgroup_enable_task_cg_lists(). As that
5309 * function sets use_task_css_set_links before grabbing
5310 * tasklist_lock and we just went through tasklist_lock to add
5311 * @child, it's guaranteed that either we see the set
5312 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5313 * @child during its iteration.
5315 * If we won the race, @child is associated with %current's
5316 * css_set. Grabbing css_set_rwsem guarantees both that the
5317 * association is stable, and, on completion of the parent's
5318 * migration, @child is visible in the source of migration or
5319 * already in the destination cgroup. This guarantee is necessary
5320 * when implementing operations which need to migrate all tasks of
5321 * a cgroup to another.
5323 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5324 * will remain in init_css_set. This is safe because all tasks are
5325 * in the init_css_set before cg_links is enabled and there's no
5326 * operation which transfers all tasks out of init_css_set.
5328 if (use_task_css_set_links
) {
5329 struct css_set
*cset
;
5331 down_write(&css_set_rwsem
);
5332 cset
= task_css_set(current
);
5333 if (list_empty(&child
->cg_list
)) {
5334 rcu_assign_pointer(child
->cgroups
, cset
);
5335 list_add(&child
->cg_list
, &cset
->tasks
);
5338 up_write(&css_set_rwsem
);
5342 * Call ss->fork(). This must happen after @child is linked on
5343 * css_set; otherwise, @child might change state between ->fork()
5344 * and addition to css_set.
5346 for_each_subsys_which(ss
, i
, &have_fork_callback
)
5347 ss
->fork(child
, subsys_canfork_priv(old_ss_priv
, i
));
5351 * cgroup_exit - detach cgroup from exiting task
5352 * @tsk: pointer to task_struct of exiting process
5354 * Description: Detach cgroup from @tsk and release it.
5356 * Note that cgroups marked notify_on_release force every task in
5357 * them to take the global cgroup_mutex mutex when exiting.
5358 * This could impact scaling on very large systems. Be reluctant to
5359 * use notify_on_release cgroups where very high task exit scaling
5360 * is required on large systems.
5362 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5363 * call cgroup_exit() while the task is still competent to handle
5364 * notify_on_release(), then leave the task attached to the root cgroup in
5365 * each hierarchy for the remainder of its exit. No need to bother with
5366 * init_css_set refcnting. init_css_set never goes away and we can't race
5367 * with migration path - PF_EXITING is visible to migration path.
5369 void cgroup_exit(struct task_struct
*tsk
)
5371 struct cgroup_subsys
*ss
;
5372 struct css_set
*cset
;
5373 bool put_cset
= false;
5377 * Unlink from @tsk from its css_set. As migration path can't race
5378 * with us, we can check cg_list without grabbing css_set_rwsem.
5380 if (!list_empty(&tsk
->cg_list
)) {
5381 down_write(&css_set_rwsem
);
5382 list_del_init(&tsk
->cg_list
);
5383 up_write(&css_set_rwsem
);
5387 /* Reassign the task to the init_css_set. */
5388 cset
= task_css_set(tsk
);
5389 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5391 /* see cgroup_post_fork() for details */
5392 for_each_subsys_which(ss
, i
, &have_exit_callback
) {
5393 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5394 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5396 ss
->exit(css
, old_css
, tsk
);
5403 static void check_for_release(struct cgroup
*cgrp
)
5405 if (notify_on_release(cgrp
) && !cgroup_has_tasks(cgrp
) &&
5406 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
5407 schedule_work(&cgrp
->release_agent_work
);
5411 * Notify userspace when a cgroup is released, by running the
5412 * configured release agent with the name of the cgroup (path
5413 * relative to the root of cgroup file system) as the argument.
5415 * Most likely, this user command will try to rmdir this cgroup.
5417 * This races with the possibility that some other task will be
5418 * attached to this cgroup before it is removed, or that some other
5419 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5420 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5421 * unused, and this cgroup will be reprieved from its death sentence,
5422 * to continue to serve a useful existence. Next time it's released,
5423 * we will get notified again, if it still has 'notify_on_release' set.
5425 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5426 * means only wait until the task is successfully execve()'d. The
5427 * separate release agent task is forked by call_usermodehelper(),
5428 * then control in this thread returns here, without waiting for the
5429 * release agent task. We don't bother to wait because the caller of
5430 * this routine has no use for the exit status of the release agent
5431 * task, so no sense holding our caller up for that.
5433 static void cgroup_release_agent(struct work_struct
*work
)
5435 struct cgroup
*cgrp
=
5436 container_of(work
, struct cgroup
, release_agent_work
);
5437 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5438 char *argv
[3], *envp
[3];
5440 mutex_lock(&cgroup_mutex
);
5442 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5443 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5444 if (!pathbuf
|| !agentbuf
)
5447 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5455 /* minimal command environment */
5457 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5460 mutex_unlock(&cgroup_mutex
);
5461 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5464 mutex_unlock(&cgroup_mutex
);
5470 static int __init
cgroup_disable(char *str
)
5472 struct cgroup_subsys
*ss
;
5476 while ((token
= strsep(&str
, ",")) != NULL
) {
5480 for_each_subsys(ss
, i
) {
5481 if (strcmp(token
, ss
->name
) &&
5482 strcmp(token
, ss
->legacy_name
))
5486 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5493 __setup("cgroup_disable=", cgroup_disable
);
5495 static int __init
cgroup_set_legacy_files_on_dfl(char *str
)
5497 printk("cgroup: using legacy files on the default hierarchy\n");
5498 cgroup_legacy_files_on_dfl
= true;
5501 __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl
);
5504 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5505 * @dentry: directory dentry of interest
5506 * @ss: subsystem of interest
5508 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5509 * to get the corresponding css and return it. If such css doesn't exist
5510 * or can't be pinned, an ERR_PTR value is returned.
5512 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5513 struct cgroup_subsys
*ss
)
5515 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5516 struct cgroup_subsys_state
*css
= NULL
;
5517 struct cgroup
*cgrp
;
5519 /* is @dentry a cgroup dir? */
5520 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5521 kernfs_type(kn
) != KERNFS_DIR
)
5522 return ERR_PTR(-EBADF
);
5527 * This path doesn't originate from kernfs and @kn could already
5528 * have been or be removed at any point. @kn->priv is RCU
5529 * protected for this access. See css_release_work_fn() for details.
5531 cgrp
= rcu_dereference(kn
->priv
);
5533 css
= cgroup_css(cgrp
, ss
);
5535 if (!css
|| !css_tryget_online(css
))
5536 css
= ERR_PTR(-ENOENT
);
5543 * css_from_id - lookup css by id
5544 * @id: the cgroup id
5545 * @ss: cgroup subsys to be looked into
5547 * Returns the css if there's valid one with @id, otherwise returns NULL.
5548 * Should be called under rcu_read_lock().
5550 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5552 WARN_ON_ONCE(!rcu_read_lock_held());
5553 return id
> 0 ? idr_find(&ss
->css_idr
, id
) : NULL
;
5556 #ifdef CONFIG_CGROUP_DEBUG
5557 static struct cgroup_subsys_state
*
5558 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5560 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5563 return ERR_PTR(-ENOMEM
);
5568 static void debug_css_free(struct cgroup_subsys_state
*css
)
5573 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5576 return cgroup_task_count(css
->cgroup
);
5579 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5582 return (u64
)(unsigned long)current
->cgroups
;
5585 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5591 count
= atomic_read(&task_css_set(current
)->refcount
);
5596 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5598 struct cgrp_cset_link
*link
;
5599 struct css_set
*cset
;
5602 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5606 down_read(&css_set_rwsem
);
5608 cset
= rcu_dereference(current
->cgroups
);
5609 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5610 struct cgroup
*c
= link
->cgrp
;
5612 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5613 seq_printf(seq
, "Root %d group %s\n",
5614 c
->root
->hierarchy_id
, name_buf
);
5617 up_read(&css_set_rwsem
);
5622 #define MAX_TASKS_SHOWN_PER_CSS 25
5623 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5625 struct cgroup_subsys_state
*css
= seq_css(seq
);
5626 struct cgrp_cset_link
*link
;
5628 down_read(&css_set_rwsem
);
5629 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5630 struct css_set
*cset
= link
->cset
;
5631 struct task_struct
*task
;
5634 seq_printf(seq
, "css_set %p\n", cset
);
5636 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5637 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5639 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5642 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5643 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5645 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5649 seq_puts(seq
, " ...\n");
5651 up_read(&css_set_rwsem
);
5655 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5657 return (!cgroup_has_tasks(css
->cgroup
) &&
5658 !css_has_online_children(&css
->cgroup
->self
));
5661 static struct cftype debug_files
[] = {
5663 .name
= "taskcount",
5664 .read_u64
= debug_taskcount_read
,
5668 .name
= "current_css_set",
5669 .read_u64
= current_css_set_read
,
5673 .name
= "current_css_set_refcount",
5674 .read_u64
= current_css_set_refcount_read
,
5678 .name
= "current_css_set_cg_links",
5679 .seq_show
= current_css_set_cg_links_read
,
5683 .name
= "cgroup_css_links",
5684 .seq_show
= cgroup_css_links_read
,
5688 .name
= "releasable",
5689 .read_u64
= releasable_read
,
5695 struct cgroup_subsys debug_cgrp_subsys
= {
5696 .css_alloc
= debug_css_alloc
,
5697 .css_free
= debug_css_free
,
5698 .legacy_cftypes
= debug_files
,
5700 #endif /* CONFIG_CGROUP_DEBUG */