1 // SPDX-License-Identifier: GPL-2.0-only
2 #include "cgroup-internal.h"
4 #include <linux/ctype.h>
5 #include <linux/kmod.h>
6 #include <linux/sort.h>
7 #include <linux/delay.h>
9 #include <linux/sched/signal.h>
10 #include <linux/sched/task.h>
11 #include <linux/magic.h>
12 #include <linux/slab.h>
13 #include <linux/vmalloc.h>
14 #include <linux/delayacct.h>
15 #include <linux/pid_namespace.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/fs_parser.h>
19 #include <trace/events/cgroup.h>
21 #define cg_invalf(fc, fmt, ...) invalf(fc, fmt, ## __VA_ARGS__)
24 * pidlists linger the following amount before being destroyed. The goal
25 * is avoiding frequent destruction in the middle of consecutive read calls
26 * Expiring in the middle is a performance problem not a correctness one.
27 * 1 sec should be enough.
29 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
31 /* Controllers blocked by the commandline in v1 */
32 static u16 cgroup_no_v1_mask
;
34 /* disable named v1 mounts */
35 static bool cgroup_no_v1_named
;
38 * pidlist destructions need to be flushed on cgroup destruction. Use a
39 * separate workqueue as flush domain.
41 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
44 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
45 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
47 static DEFINE_SPINLOCK(release_agent_path_lock
);
49 bool cgroup1_ssid_disabled(int ssid
)
51 return cgroup_no_v1_mask
& (1 << ssid
);
55 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
56 * @from: attach to all cgroups of a given task
57 * @tsk: the task to be attached
59 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
61 struct cgroup_root
*root
;
64 mutex_lock(&cgroup_mutex
);
65 percpu_down_write(&cgroup_threadgroup_rwsem
);
67 struct cgroup
*from_cgrp
;
69 if (root
== &cgrp_dfl_root
)
72 spin_lock_irq(&css_set_lock
);
73 from_cgrp
= task_cgroup_from_root(from
, root
);
74 spin_unlock_irq(&css_set_lock
);
76 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
80 percpu_up_write(&cgroup_threadgroup_rwsem
);
81 mutex_unlock(&cgroup_mutex
);
85 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
88 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
89 * @to: cgroup to which the tasks will be moved
90 * @from: cgroup in which the tasks currently reside
92 * Locking rules between cgroup_post_fork() and the migration path
93 * guarantee that, if a task is forking while being migrated, the new child
94 * is guaranteed to be either visible in the source cgroup after the
95 * parent's migration is complete or put into the target cgroup. No task
96 * can slip out of migration through forking.
98 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
100 DEFINE_CGROUP_MGCTX(mgctx
);
101 struct cgrp_cset_link
*link
;
102 struct css_task_iter it
;
103 struct task_struct
*task
;
106 if (cgroup_on_dfl(to
))
109 ret
= cgroup_migrate_vet_dst(to
);
113 mutex_lock(&cgroup_mutex
);
115 percpu_down_write(&cgroup_threadgroup_rwsem
);
117 /* all tasks in @from are being moved, all csets are source */
118 spin_lock_irq(&css_set_lock
);
119 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
120 cgroup_migrate_add_src(link
->cset
, to
, &mgctx
);
121 spin_unlock_irq(&css_set_lock
);
123 ret
= cgroup_migrate_prepare_dst(&mgctx
);
128 * Migrate tasks one-by-one until @from is empty. This fails iff
129 * ->can_attach() fails.
132 css_task_iter_start(&from
->self
, 0, &it
);
135 task
= css_task_iter_next(&it
);
136 } while (task
&& (task
->flags
& PF_EXITING
));
139 get_task_struct(task
);
140 css_task_iter_end(&it
);
143 ret
= cgroup_migrate(task
, false, &mgctx
);
145 TRACE_CGROUP_PATH(transfer_tasks
, to
, task
, false);
146 put_task_struct(task
);
148 } while (task
&& !ret
);
150 cgroup_migrate_finish(&mgctx
);
151 percpu_up_write(&cgroup_threadgroup_rwsem
);
152 mutex_unlock(&cgroup_mutex
);
157 * Stuff for reading the 'tasks'/'procs' files.
159 * Reading this file can return large amounts of data if a cgroup has
160 * *lots* of attached tasks. So it may need several calls to read(),
161 * but we cannot guarantee that the information we produce is correct
162 * unless we produce it entirely atomically.
166 /* which pidlist file are we talking about? */
167 enum cgroup_filetype
{
173 * A pidlist is a list of pids that virtually represents the contents of one
174 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
175 * a pair (one each for procs, tasks) for each pid namespace that's relevant
178 struct cgroup_pidlist
{
180 * used to find which pidlist is wanted. doesn't change as long as
181 * this particular list stays in the list.
183 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
186 /* how many elements the above list has */
188 /* each of these stored in a list by its cgroup */
189 struct list_head links
;
190 /* pointer to the cgroup we belong to, for list removal purposes */
191 struct cgroup
*owner
;
192 /* for delayed destruction */
193 struct delayed_work destroy_dwork
;
197 * Used to destroy all pidlists lingering waiting for destroy timer. None
198 * should be left afterwards.
200 void cgroup1_pidlist_destroy_all(struct cgroup
*cgrp
)
202 struct cgroup_pidlist
*l
, *tmp_l
;
204 mutex_lock(&cgrp
->pidlist_mutex
);
205 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
206 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
207 mutex_unlock(&cgrp
->pidlist_mutex
);
209 flush_workqueue(cgroup_pidlist_destroy_wq
);
210 BUG_ON(!list_empty(&cgrp
->pidlists
));
213 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
215 struct delayed_work
*dwork
= to_delayed_work(work
);
216 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
218 struct cgroup_pidlist
*tofree
= NULL
;
220 mutex_lock(&l
->owner
->pidlist_mutex
);
223 * Destroy iff we didn't get queued again. The state won't change
224 * as destroy_dwork can only be queued while locked.
226 if (!delayed_work_pending(dwork
)) {
229 put_pid_ns(l
->key
.ns
);
233 mutex_unlock(&l
->owner
->pidlist_mutex
);
238 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
239 * Returns the number of unique elements.
241 static int pidlist_uniq(pid_t
*list
, int length
)
246 * we presume the 0th element is unique, so i starts at 1. trivial
247 * edge cases first; no work needs to be done for either
249 if (length
== 0 || length
== 1)
251 /* src and dest walk down the list; dest counts unique elements */
252 for (src
= 1; src
< length
; src
++) {
253 /* find next unique element */
254 while (list
[src
] == list
[src
-1]) {
259 /* dest always points to where the next unique element goes */
260 list
[dest
] = list
[src
];
268 * The two pid files - task and cgroup.procs - guaranteed that the result
269 * is sorted, which forced this whole pidlist fiasco. As pid order is
270 * different per namespace, each namespace needs differently sorted list,
271 * making it impossible to use, for example, single rbtree of member tasks
272 * sorted by task pointer. As pidlists can be fairly large, allocating one
273 * per open file is dangerous, so cgroup had to implement shared pool of
274 * pidlists keyed by cgroup and namespace.
276 static int cmppid(const void *a
, const void *b
)
278 return *(pid_t
*)a
- *(pid_t
*)b
;
281 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
282 enum cgroup_filetype type
)
284 struct cgroup_pidlist
*l
;
285 /* don't need task_nsproxy() if we're looking at ourself */
286 struct pid_namespace
*ns
= task_active_pid_ns(current
);
288 lockdep_assert_held(&cgrp
->pidlist_mutex
);
290 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
291 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
297 * find the appropriate pidlist for our purpose (given procs vs tasks)
298 * returns with the lock on that pidlist already held, and takes care
299 * of the use count, or returns NULL with no locks held if we're out of
302 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
303 enum cgroup_filetype type
)
305 struct cgroup_pidlist
*l
;
307 lockdep_assert_held(&cgrp
->pidlist_mutex
);
309 l
= cgroup_pidlist_find(cgrp
, type
);
313 /* entry not found; create a new one */
314 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
318 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
320 /* don't need task_nsproxy() if we're looking at ourself */
321 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
323 list_add(&l
->links
, &cgrp
->pidlists
);
328 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
330 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
331 struct cgroup_pidlist
**lp
)
335 int pid
, n
= 0; /* used for populating the array */
336 struct css_task_iter it
;
337 struct task_struct
*tsk
;
338 struct cgroup_pidlist
*l
;
340 lockdep_assert_held(&cgrp
->pidlist_mutex
);
343 * If cgroup gets more users after we read count, we won't have
344 * enough space - tough. This race is indistinguishable to the
345 * caller from the case that the additional cgroup users didn't
346 * show up until sometime later on.
348 length
= cgroup_task_count(cgrp
);
349 array
= kvmalloc_array(length
, sizeof(pid_t
), GFP_KERNEL
);
352 /* now, populate the array */
353 css_task_iter_start(&cgrp
->self
, 0, &it
);
354 while ((tsk
= css_task_iter_next(&it
))) {
355 if (unlikely(n
== length
))
357 /* get tgid or pid for procs or tasks file respectively */
358 if (type
== CGROUP_FILE_PROCS
)
359 pid
= task_tgid_vnr(tsk
);
361 pid
= task_pid_vnr(tsk
);
362 if (pid
> 0) /* make sure to only use valid results */
365 css_task_iter_end(&it
);
367 /* now sort & (if procs) strip out duplicates */
368 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
369 if (type
== CGROUP_FILE_PROCS
)
370 length
= pidlist_uniq(array
, length
);
372 l
= cgroup_pidlist_find_create(cgrp
, type
);
378 /* store array, freeing old if necessary */
387 * seq_file methods for the tasks/procs files. The seq_file position is the
388 * next pid to display; the seq_file iterator is a pointer to the pid
389 * in the cgroup->l->list array.
392 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
395 * Initially we receive a position value that corresponds to
396 * one more than the last pid shown (or 0 on the first call or
397 * after a seek to the start). Use a binary-search to find the
398 * next pid to display, if any
400 struct kernfs_open_file
*of
= s
->private;
401 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
402 struct cgroup_pidlist
*l
;
403 enum cgroup_filetype type
= seq_cft(s
)->private;
404 int index
= 0, pid
= *pos
;
407 mutex_lock(&cgrp
->pidlist_mutex
);
410 * !NULL @of->priv indicates that this isn't the first start()
411 * after open. If the matching pidlist is around, we can use that.
412 * Look for it. Note that @of->priv can't be used directly. It
413 * could already have been destroyed.
416 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
419 * Either this is the first start() after open or the matching
420 * pidlist has been destroyed inbetween. Create a new one.
423 ret
= pidlist_array_load(cgrp
, type
,
424 (struct cgroup_pidlist
**)&of
->priv
);
433 while (index
< end
) {
434 int mid
= (index
+ end
) / 2;
435 if (l
->list
[mid
] == pid
) {
438 } else if (l
->list
[mid
] <= pid
)
444 /* If we're off the end of the array, we're done */
445 if (index
>= l
->length
)
447 /* Update the abstract position to be the actual pid that we found */
448 iter
= l
->list
+ index
;
453 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
455 struct kernfs_open_file
*of
= s
->private;
456 struct cgroup_pidlist
*l
= of
->priv
;
459 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
460 CGROUP_PIDLIST_DESTROY_DELAY
);
461 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
464 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
466 struct kernfs_open_file
*of
= s
->private;
467 struct cgroup_pidlist
*l
= of
->priv
;
469 pid_t
*end
= l
->list
+ l
->length
;
471 * Advance to the next pid in the array. If this goes off the
484 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
486 seq_printf(s
, "%d\n", *(int *)v
);
491 static ssize_t
__cgroup1_procs_write(struct kernfs_open_file
*of
,
492 char *buf
, size_t nbytes
, loff_t off
,
496 struct task_struct
*task
;
497 const struct cred
*cred
, *tcred
;
500 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
504 task
= cgroup_procs_write_start(buf
, threadgroup
);
505 ret
= PTR_ERR_OR_ZERO(task
);
510 * Even if we're attaching all tasks in the thread group, we only
511 * need to check permissions on one of them.
513 cred
= current_cred();
514 tcred
= get_task_cred(task
);
515 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
516 !uid_eq(cred
->euid
, tcred
->uid
) &&
517 !uid_eq(cred
->euid
, tcred
->suid
))
523 ret
= cgroup_attach_task(cgrp
, task
, threadgroup
);
526 cgroup_procs_write_finish(task
);
528 cgroup_kn_unlock(of
->kn
);
530 return ret
?: nbytes
;
533 static ssize_t
cgroup1_procs_write(struct kernfs_open_file
*of
,
534 char *buf
, size_t nbytes
, loff_t off
)
536 return __cgroup1_procs_write(of
, buf
, nbytes
, off
, true);
539 static ssize_t
cgroup1_tasks_write(struct kernfs_open_file
*of
,
540 char *buf
, size_t nbytes
, loff_t off
)
542 return __cgroup1_procs_write(of
, buf
, nbytes
, off
, false);
545 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
546 char *buf
, size_t nbytes
, loff_t off
)
550 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
553 * Release agent gets called with all capabilities,
554 * require capabilities to set release agent.
556 if ((of
->file
->f_cred
->user_ns
!= &init_user_ns
) ||
557 !capable(CAP_SYS_ADMIN
))
560 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
563 spin_lock(&release_agent_path_lock
);
564 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
565 sizeof(cgrp
->root
->release_agent_path
));
566 spin_unlock(&release_agent_path_lock
);
567 cgroup_kn_unlock(of
->kn
);
571 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
573 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
575 spin_lock(&release_agent_path_lock
);
576 seq_puts(seq
, cgrp
->root
->release_agent_path
);
577 spin_unlock(&release_agent_path_lock
);
582 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
584 seq_puts(seq
, "0\n");
588 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
591 return notify_on_release(css
->cgroup
);
594 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
595 struct cftype
*cft
, u64 val
)
598 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
600 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
604 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
607 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
610 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
611 struct cftype
*cft
, u64 val
)
614 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
616 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
620 /* cgroup core interface files for the legacy hierarchies */
621 struct cftype cgroup1_base_files
[] = {
623 .name
= "cgroup.procs",
624 .seq_start
= cgroup_pidlist_start
,
625 .seq_next
= cgroup_pidlist_next
,
626 .seq_stop
= cgroup_pidlist_stop
,
627 .seq_show
= cgroup_pidlist_show
,
628 .private = CGROUP_FILE_PROCS
,
629 .write
= cgroup1_procs_write
,
632 .name
= "cgroup.clone_children",
633 .read_u64
= cgroup_clone_children_read
,
634 .write_u64
= cgroup_clone_children_write
,
637 .name
= "cgroup.sane_behavior",
638 .flags
= CFTYPE_ONLY_ON_ROOT
,
639 .seq_show
= cgroup_sane_behavior_show
,
643 .seq_start
= cgroup_pidlist_start
,
644 .seq_next
= cgroup_pidlist_next
,
645 .seq_stop
= cgroup_pidlist_stop
,
646 .seq_show
= cgroup_pidlist_show
,
647 .private = CGROUP_FILE_TASKS
,
648 .write
= cgroup1_tasks_write
,
651 .name
= "notify_on_release",
652 .read_u64
= cgroup_read_notify_on_release
,
653 .write_u64
= cgroup_write_notify_on_release
,
656 .name
= "release_agent",
657 .flags
= CFTYPE_ONLY_ON_ROOT
,
658 .seq_show
= cgroup_release_agent_show
,
659 .write
= cgroup_release_agent_write
,
660 .max_write_len
= PATH_MAX
- 1,
665 /* Display information about each subsystem and each hierarchy */
666 int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
668 struct cgroup_subsys
*ss
;
671 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
673 * ideally we don't want subsystems moving around while we do this.
674 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
675 * subsys/hierarchy state.
677 mutex_lock(&cgroup_mutex
);
679 for_each_subsys(ss
, i
)
680 seq_printf(m
, "%s\t%d\t%d\t%d\n",
681 ss
->legacy_name
, ss
->root
->hierarchy_id
,
682 atomic_read(&ss
->root
->nr_cgrps
),
683 cgroup_ssid_enabled(i
));
685 mutex_unlock(&cgroup_mutex
);
690 * cgroupstats_build - build and fill cgroupstats
691 * @stats: cgroupstats to fill information into
692 * @dentry: A dentry entry belonging to the cgroup for which stats have
695 * Build and fill cgroupstats so that taskstats can export it to user
698 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
700 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
702 struct css_task_iter it
;
703 struct task_struct
*tsk
;
705 /* it should be kernfs_node belonging to cgroupfs and is a directory */
706 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
707 kernfs_type(kn
) != KERNFS_DIR
)
710 mutex_lock(&cgroup_mutex
);
713 * We aren't being called from kernfs and there's no guarantee on
714 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
715 * @kn->priv is RCU safe. Let's do the RCU dancing.
718 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
719 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
721 mutex_unlock(&cgroup_mutex
);
726 css_task_iter_start(&cgrp
->self
, 0, &it
);
727 while ((tsk
= css_task_iter_next(&it
))) {
728 switch (tsk
->state
) {
732 case TASK_INTERRUPTIBLE
:
733 stats
->nr_sleeping
++;
735 case TASK_UNINTERRUPTIBLE
:
736 stats
->nr_uninterruptible
++;
742 if (delayacct_is_task_waiting_on_io(tsk
))
747 css_task_iter_end(&it
);
749 mutex_unlock(&cgroup_mutex
);
753 void cgroup1_check_for_release(struct cgroup
*cgrp
)
755 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
756 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
757 schedule_work(&cgrp
->release_agent_work
);
761 * Notify userspace when a cgroup is released, by running the
762 * configured release agent with the name of the cgroup (path
763 * relative to the root of cgroup file system) as the argument.
765 * Most likely, this user command will try to rmdir this cgroup.
767 * This races with the possibility that some other task will be
768 * attached to this cgroup before it is removed, or that some other
769 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
770 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
771 * unused, and this cgroup will be reprieved from its death sentence,
772 * to continue to serve a useful existence. Next time it's released,
773 * we will get notified again, if it still has 'notify_on_release' set.
775 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
776 * means only wait until the task is successfully execve()'d. The
777 * separate release agent task is forked by call_usermodehelper(),
778 * then control in this thread returns here, without waiting for the
779 * release agent task. We don't bother to wait because the caller of
780 * this routine has no use for the exit status of the release agent
781 * task, so no sense holding our caller up for that.
783 void cgroup1_release_agent(struct work_struct
*work
)
785 struct cgroup
*cgrp
=
786 container_of(work
, struct cgroup
, release_agent_work
);
787 char *pathbuf
= NULL
, *agentbuf
= NULL
;
788 char *argv
[3], *envp
[3];
791 mutex_lock(&cgroup_mutex
);
793 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
794 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
795 if (!pathbuf
|| !agentbuf
|| !strlen(agentbuf
))
798 spin_lock_irq(&css_set_lock
);
799 ret
= cgroup_path_ns_locked(cgrp
, pathbuf
, PATH_MAX
, &init_cgroup_ns
);
800 spin_unlock_irq(&css_set_lock
);
801 if (ret
< 0 || ret
>= PATH_MAX
)
808 /* minimal command environment */
810 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
813 mutex_unlock(&cgroup_mutex
);
814 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
817 mutex_unlock(&cgroup_mutex
);
824 * cgroup_rename - Only allow simple rename of directories in place.
826 static int cgroup1_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
827 const char *new_name_str
)
829 struct cgroup
*cgrp
= kn
->priv
;
832 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
833 if (strchr(new_name_str
, '\n'))
836 if (kernfs_type(kn
) != KERNFS_DIR
)
838 if (kn
->parent
!= new_parent
)
842 * We're gonna grab cgroup_mutex which nests outside kernfs
843 * active_ref. kernfs_rename() doesn't require active_ref
844 * protection. Break them before grabbing cgroup_mutex.
846 kernfs_break_active_protection(new_parent
);
847 kernfs_break_active_protection(kn
);
849 mutex_lock(&cgroup_mutex
);
851 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
853 TRACE_CGROUP_PATH(rename
, cgrp
);
855 mutex_unlock(&cgroup_mutex
);
857 kernfs_unbreak_active_protection(kn
);
858 kernfs_unbreak_active_protection(new_parent
);
862 static int cgroup1_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
864 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
865 struct cgroup_subsys
*ss
;
868 for_each_subsys(ss
, ssid
)
869 if (root
->subsys_mask
& (1 << ssid
))
870 seq_show_option(seq
, ss
->legacy_name
, NULL
);
871 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
872 seq_puts(seq
, ",noprefix");
873 if (root
->flags
& CGRP_ROOT_XATTR
)
874 seq_puts(seq
, ",xattr");
875 if (root
->flags
& CGRP_ROOT_CPUSET_V2_MODE
)
876 seq_puts(seq
, ",cpuset_v2_mode");
878 spin_lock(&release_agent_path_lock
);
879 if (strlen(root
->release_agent_path
))
880 seq_show_option(seq
, "release_agent",
881 root
->release_agent_path
);
882 spin_unlock(&release_agent_path_lock
);
884 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
885 seq_puts(seq
, ",clone_children");
886 if (strlen(root
->name
))
887 seq_show_option(seq
, "name", root
->name
);
902 static const struct fs_parameter_spec cgroup1_param_specs
[] = {
903 fsparam_flag ("all", Opt_all
),
904 fsparam_flag ("clone_children", Opt_clone_children
),
905 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode
),
906 fsparam_string("name", Opt_name
),
907 fsparam_flag ("none", Opt_none
),
908 fsparam_flag ("noprefix", Opt_noprefix
),
909 fsparam_string("release_agent", Opt_release_agent
),
910 fsparam_flag ("xattr", Opt_xattr
),
914 const struct fs_parameter_description cgroup1_fs_parameters
= {
916 .specs
= cgroup1_param_specs
,
919 int cgroup1_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
921 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
922 struct cgroup_subsys
*ss
;
923 struct fs_parse_result result
;
926 opt
= fs_parse(fc
, &cgroup1_fs_parameters
, param
, &result
);
927 if (opt
== -ENOPARAM
) {
928 if (strcmp(param
->key
, "source") == 0) {
929 if (param
->type
!= fs_value_is_string
)
930 return invalf(fc
, "Non-string source");
932 return invalf(fc
, "Multiple sources not supported");
933 fc
->source
= param
->string
;
934 param
->string
= NULL
;
937 for_each_subsys(ss
, i
) {
938 if (strcmp(param
->key
, ss
->legacy_name
))
940 ctx
->subsys_mask
|= (1 << i
);
943 return cg_invalf(fc
, "cgroup1: Unknown subsys name '%s'", param
->key
);
950 /* Explicitly have no subsystems */
957 ctx
->flags
|= CGRP_ROOT_NOPREFIX
;
959 case Opt_clone_children
:
960 ctx
->cpuset_clone_children
= true;
962 case Opt_cpuset_v2_mode
:
963 ctx
->flags
|= CGRP_ROOT_CPUSET_V2_MODE
;
966 ctx
->flags
|= CGRP_ROOT_XATTR
;
968 case Opt_release_agent
:
969 /* Specifying two release agents is forbidden */
970 if (ctx
->release_agent
)
971 return cg_invalf(fc
, "cgroup1: release_agent respecified");
973 * Release agent gets called with all capabilities,
974 * require capabilities to set release agent.
976 if ((fc
->user_ns
!= &init_user_ns
) || !capable(CAP_SYS_ADMIN
))
977 return cg_invalf(fc
, "cgroup1: Setting release_agent not allowed");
978 ctx
->release_agent
= param
->string
;
979 param
->string
= NULL
;
982 /* blocked by boot param? */
983 if (cgroup_no_v1_named
)
985 /* Can't specify an empty name */
987 return cg_invalf(fc
, "cgroup1: Empty name");
988 if (param
->size
> MAX_CGROUP_ROOT_NAMELEN
- 1)
989 return cg_invalf(fc
, "cgroup1: Name too long");
990 /* Must match [\w.-]+ */
991 for (i
= 0; i
< param
->size
; i
++) {
992 char c
= param
->string
[i
];
995 if ((c
== '.') || (c
== '-') || (c
== '_'))
997 return cg_invalf(fc
, "cgroup1: Invalid name");
999 /* Specifying two names is forbidden */
1001 return cg_invalf(fc
, "cgroup1: name respecified");
1002 ctx
->name
= param
->string
;
1003 param
->string
= NULL
;
1009 static int check_cgroupfs_options(struct fs_context
*fc
)
1011 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1014 struct cgroup_subsys
*ss
;
1017 #ifdef CONFIG_CPUSETS
1018 mask
= ~((u16
)1 << cpuset_cgrp_id
);
1020 for_each_subsys(ss
, i
)
1021 if (cgroup_ssid_enabled(i
) && !cgroup1_ssid_disabled(i
))
1024 ctx
->subsys_mask
&= enabled
;
1027 * In absense of 'none', 'name=' or subsystem name options,
1028 * let's default to 'all'.
1030 if (!ctx
->subsys_mask
&& !ctx
->none
&& !ctx
->name
)
1034 /* Mutually exclusive option 'all' + subsystem name */
1035 if (ctx
->subsys_mask
)
1036 return cg_invalf(fc
, "cgroup1: subsys name conflicts with all");
1037 /* 'all' => select all the subsystems */
1038 ctx
->subsys_mask
= enabled
;
1042 * We either have to specify by name or by subsystems. (So all
1043 * empty hierarchies must have a name).
1045 if (!ctx
->subsys_mask
&& !ctx
->name
)
1046 return cg_invalf(fc
, "cgroup1: Need name or subsystem set");
1049 * Option noprefix was introduced just for backward compatibility
1050 * with the old cpuset, so we allow noprefix only if mounting just
1051 * the cpuset subsystem.
1053 if ((ctx
->flags
& CGRP_ROOT_NOPREFIX
) && (ctx
->subsys_mask
& mask
))
1054 return cg_invalf(fc
, "cgroup1: noprefix used incorrectly");
1056 /* Can't specify "none" and some subsystems */
1057 if (ctx
->subsys_mask
&& ctx
->none
)
1058 return cg_invalf(fc
, "cgroup1: none used incorrectly");
1063 int cgroup1_reconfigure(struct fs_context
*fc
)
1065 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1066 struct kernfs_root
*kf_root
= kernfs_root_from_sb(fc
->root
->d_sb
);
1067 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1069 u16 added_mask
, removed_mask
;
1071 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1073 /* See what subsystems are wanted */
1074 ret
= check_cgroupfs_options(fc
);
1078 if (ctx
->subsys_mask
!= root
->subsys_mask
|| ctx
->release_agent
)
1079 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1080 task_tgid_nr(current
), current
->comm
);
1082 added_mask
= ctx
->subsys_mask
& ~root
->subsys_mask
;
1083 removed_mask
= root
->subsys_mask
& ~ctx
->subsys_mask
;
1085 /* Don't allow flags or name to change at remount */
1086 if ((ctx
->flags
^ root
->flags
) ||
1087 (ctx
->name
&& strcmp(ctx
->name
, root
->name
))) {
1088 cg_invalf(fc
, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1089 ctx
->flags
, ctx
->name
?: "", root
->flags
, root
->name
);
1094 /* remounting is not allowed for populated hierarchies */
1095 if (!list_empty(&root
->cgrp
.self
.children
)) {
1100 ret
= rebind_subsystems(root
, added_mask
);
1104 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, removed_mask
));
1106 if (ctx
->release_agent
) {
1107 spin_lock(&release_agent_path_lock
);
1108 strcpy(root
->release_agent_path
, ctx
->release_agent
);
1109 spin_unlock(&release_agent_path_lock
);
1112 trace_cgroup_remount(root
);
1115 mutex_unlock(&cgroup_mutex
);
1119 struct kernfs_syscall_ops cgroup1_kf_syscall_ops
= {
1120 .rename
= cgroup1_rename
,
1121 .show_options
= cgroup1_show_options
,
1122 .mkdir
= cgroup_mkdir
,
1123 .rmdir
= cgroup_rmdir
,
1124 .show_path
= cgroup_show_path
,
1128 * The guts of cgroup1 mount - find or create cgroup_root to use.
1129 * Called with cgroup_mutex held; returns 0 on success, -E... on
1130 * error and positive - in case when the candidate is busy dying.
1131 * On success it stashes a reference to cgroup_root into given
1132 * cgroup_fs_context; that reference is *NOT* counting towards the
1133 * cgroup_root refcount.
1135 static int cgroup1_root_to_use(struct fs_context
*fc
)
1137 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1138 struct cgroup_root
*root
;
1139 struct cgroup_subsys
*ss
;
1142 /* First find the desired set of subsystems */
1143 ret
= check_cgroupfs_options(fc
);
1148 * Destruction of cgroup root is asynchronous, so subsystems may
1149 * still be dying after the previous unmount. Let's drain the
1150 * dying subsystems. We just need to ensure that the ones
1151 * unmounted previously finish dying and don't care about new ones
1152 * starting. Testing ref liveliness is good enough.
1154 for_each_subsys(ss
, i
) {
1155 if (!(ctx
->subsys_mask
& (1 << i
)) ||
1156 ss
->root
== &cgrp_dfl_root
)
1159 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
))
1160 return 1; /* restart */
1161 cgroup_put(&ss
->root
->cgrp
);
1164 for_each_root(root
) {
1165 bool name_match
= false;
1167 if (root
== &cgrp_dfl_root
)
1171 * If we asked for a name then it must match. Also, if
1172 * name matches but sybsys_mask doesn't, we should fail.
1173 * Remember whether name matched.
1176 if (strcmp(ctx
->name
, root
->name
))
1182 * If we asked for subsystems (or explicitly for no
1183 * subsystems) then they must match.
1185 if ((ctx
->subsys_mask
|| ctx
->none
) &&
1186 (ctx
->subsys_mask
!= root
->subsys_mask
)) {
1192 if (root
->flags
^ ctx
->flags
)
1193 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1200 * No such thing, create a new one. name= matching without subsys
1201 * specification is allowed for already existing hierarchies but we
1202 * can't create new one without subsys specification.
1204 if (!ctx
->subsys_mask
&& !ctx
->none
)
1205 return cg_invalf(fc
, "cgroup1: No subsys list or none specified");
1207 /* Hierarchies may only be created in the initial cgroup namespace. */
1208 if (ctx
->ns
!= &init_cgroup_ns
)
1211 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1216 init_cgroup_root(ctx
);
1218 ret
= cgroup_setup_root(root
, ctx
->subsys_mask
);
1220 cgroup_free_root(root
);
1224 int cgroup1_get_tree(struct fs_context
*fc
)
1226 struct cgroup_fs_context
*ctx
= cgroup_fc2context(fc
);
1229 /* Check if the caller has permission to mount. */
1230 if (!ns_capable(ctx
->ns
->user_ns
, CAP_SYS_ADMIN
))
1233 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1235 ret
= cgroup1_root_to_use(fc
);
1236 if (!ret
&& !percpu_ref_tryget_live(&ctx
->root
->cgrp
.self
.refcnt
))
1237 ret
= 1; /* restart */
1239 mutex_unlock(&cgroup_mutex
);
1242 ret
= cgroup_do_get_tree(fc
);
1244 if (!ret
&& percpu_ref_is_dying(&ctx
->root
->cgrp
.self
.refcnt
)) {
1249 if (unlikely(ret
> 0)) {
1251 return restart_syscall();
1256 static int __init
cgroup1_wq_init(void)
1259 * Used to destroy pidlists and separate to serve as flush domain.
1260 * Cap @max_active to 1 too.
1262 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
1264 BUG_ON(!cgroup_pidlist_destroy_wq
);
1267 core_initcall(cgroup1_wq_init
);
1269 static int __init
cgroup_no_v1(char *str
)
1271 struct cgroup_subsys
*ss
;
1275 while ((token
= strsep(&str
, ",")) != NULL
) {
1279 if (!strcmp(token
, "all")) {
1280 cgroup_no_v1_mask
= U16_MAX
;
1284 if (!strcmp(token
, "named")) {
1285 cgroup_no_v1_named
= true;
1289 for_each_subsys(ss
, i
) {
1290 if (strcmp(token
, ss
->name
) &&
1291 strcmp(token
, ss
->legacy_name
))
1294 cgroup_no_v1_mask
|= 1 << i
;
1299 __setup("cgroup_no_v1=", cgroup_no_v1
);