1 #include "cgroup-internal.h"
3 #include <linux/ctype.h>
4 #include <linux/kmod.h>
5 #include <linux/sort.h>
6 #include <linux/delay.h>
8 #include <linux/sched/signal.h>
9 #include <linux/sched/task.h>
10 #include <linux/magic.h>
11 #include <linux/slab.h>
12 #include <linux/vmalloc.h>
13 #include <linux/delayacct.h>
14 #include <linux/pid_namespace.h>
15 #include <linux/cgroupstats.h>
17 #include <trace/events/cgroup.h>
20 * pidlists linger the following amount before being destroyed. The goal
21 * is avoiding frequent destruction in the middle of consecutive read calls
22 * Expiring in the middle is a performance problem not a correctness one.
23 * 1 sec should be enough.
25 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
27 /* Controllers blocked by the commandline in v1 */
28 static u16 cgroup_no_v1_mask
;
31 * pidlist destructions need to be flushed on cgroup destruction. Use a
32 * separate workqueue as flush domain.
34 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
37 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
38 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
40 static DEFINE_SPINLOCK(release_agent_path_lock
);
42 bool cgroup1_ssid_disabled(int ssid
)
44 return cgroup_no_v1_mask
& (1 << ssid
);
48 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
49 * @from: attach to all cgroups of a given task
50 * @tsk: the task to be attached
52 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
54 struct cgroup_root
*root
;
57 mutex_lock(&cgroup_mutex
);
58 percpu_down_write(&cgroup_threadgroup_rwsem
);
60 struct cgroup
*from_cgrp
;
62 if (root
== &cgrp_dfl_root
)
65 spin_lock_irq(&css_set_lock
);
66 from_cgrp
= task_cgroup_from_root(from
, root
);
67 spin_unlock_irq(&css_set_lock
);
69 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
73 percpu_up_write(&cgroup_threadgroup_rwsem
);
74 mutex_unlock(&cgroup_mutex
);
78 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
81 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
82 * @to: cgroup to which the tasks will be moved
83 * @from: cgroup in which the tasks currently reside
85 * Locking rules between cgroup_post_fork() and the migration path
86 * guarantee that, if a task is forking while being migrated, the new child
87 * is guaranteed to be either visible in the source cgroup after the
88 * parent's migration is complete or put into the target cgroup. No task
89 * can slip out of migration through forking.
91 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
93 DEFINE_CGROUP_MGCTX(mgctx
);
94 struct cgrp_cset_link
*link
;
95 struct css_task_iter it
;
96 struct task_struct
*task
;
99 if (cgroup_on_dfl(to
))
102 ret
= cgroup_migrate_vet_dst(to
);
106 mutex_lock(&cgroup_mutex
);
108 percpu_down_write(&cgroup_threadgroup_rwsem
);
110 /* all tasks in @from are being moved, all csets are source */
111 spin_lock_irq(&css_set_lock
);
112 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
113 cgroup_migrate_add_src(link
->cset
, to
, &mgctx
);
114 spin_unlock_irq(&css_set_lock
);
116 ret
= cgroup_migrate_prepare_dst(&mgctx
);
121 * Migrate tasks one-by-one until @from is empty. This fails iff
122 * ->can_attach() fails.
125 css_task_iter_start(&from
->self
, 0, &it
);
128 task
= css_task_iter_next(&it
);
129 } while (task
&& (task
->flags
& PF_EXITING
));
132 get_task_struct(task
);
133 css_task_iter_end(&it
);
136 ret
= cgroup_migrate(task
, false, &mgctx
);
138 trace_cgroup_transfer_tasks(to
, task
, false);
139 put_task_struct(task
);
141 } while (task
&& !ret
);
143 cgroup_migrate_finish(&mgctx
);
144 percpu_up_write(&cgroup_threadgroup_rwsem
);
145 mutex_unlock(&cgroup_mutex
);
150 * Stuff for reading the 'tasks'/'procs' files.
152 * Reading this file can return large amounts of data if a cgroup has
153 * *lots* of attached tasks. So it may need several calls to read(),
154 * but we cannot guarantee that the information we produce is correct
155 * unless we produce it entirely atomically.
159 /* which pidlist file are we talking about? */
160 enum cgroup_filetype
{
166 * A pidlist is a list of pids that virtually represents the contents of one
167 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
168 * a pair (one each for procs, tasks) for each pid namespace that's relevant
171 struct cgroup_pidlist
{
173 * used to find which pidlist is wanted. doesn't change as long as
174 * this particular list stays in the list.
176 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
179 /* how many elements the above list has */
181 /* each of these stored in a list by its cgroup */
182 struct list_head links
;
183 /* pointer to the cgroup we belong to, for list removal purposes */
184 struct cgroup
*owner
;
185 /* for delayed destruction */
186 struct delayed_work destroy_dwork
;
190 * The following two functions "fix" the issue where there are more pids
191 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
192 * TODO: replace with a kernel-wide solution to this problem
194 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
195 static void *pidlist_allocate(int count
)
197 if (PIDLIST_TOO_LARGE(count
))
198 return vmalloc(array_size(count
, sizeof(pid_t
)));
200 return kmalloc_array(count
, sizeof(pid_t
), GFP_KERNEL
);
203 static void pidlist_free(void *p
)
209 * Used to destroy all pidlists lingering waiting for destroy timer. None
210 * should be left afterwards.
212 void cgroup1_pidlist_destroy_all(struct cgroup
*cgrp
)
214 struct cgroup_pidlist
*l
, *tmp_l
;
216 mutex_lock(&cgrp
->pidlist_mutex
);
217 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
218 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
219 mutex_unlock(&cgrp
->pidlist_mutex
);
221 flush_workqueue(cgroup_pidlist_destroy_wq
);
222 BUG_ON(!list_empty(&cgrp
->pidlists
));
225 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
227 struct delayed_work
*dwork
= to_delayed_work(work
);
228 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
230 struct cgroup_pidlist
*tofree
= NULL
;
232 mutex_lock(&l
->owner
->pidlist_mutex
);
235 * Destroy iff we didn't get queued again. The state won't change
236 * as destroy_dwork can only be queued while locked.
238 if (!delayed_work_pending(dwork
)) {
240 pidlist_free(l
->list
);
241 put_pid_ns(l
->key
.ns
);
245 mutex_unlock(&l
->owner
->pidlist_mutex
);
250 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
251 * Returns the number of unique elements.
253 static int pidlist_uniq(pid_t
*list
, int length
)
258 * we presume the 0th element is unique, so i starts at 1. trivial
259 * edge cases first; no work needs to be done for either
261 if (length
== 0 || length
== 1)
263 /* src and dest walk down the list; dest counts unique elements */
264 for (src
= 1; src
< length
; src
++) {
265 /* find next unique element */
266 while (list
[src
] == list
[src
-1]) {
271 /* dest always points to where the next unique element goes */
272 list
[dest
] = list
[src
];
280 * The two pid files - task and cgroup.procs - guaranteed that the result
281 * is sorted, which forced this whole pidlist fiasco. As pid order is
282 * different per namespace, each namespace needs differently sorted list,
283 * making it impossible to use, for example, single rbtree of member tasks
284 * sorted by task pointer. As pidlists can be fairly large, allocating one
285 * per open file is dangerous, so cgroup had to implement shared pool of
286 * pidlists keyed by cgroup and namespace.
288 static int cmppid(const void *a
, const void *b
)
290 return *(pid_t
*)a
- *(pid_t
*)b
;
293 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
294 enum cgroup_filetype type
)
296 struct cgroup_pidlist
*l
;
297 /* don't need task_nsproxy() if we're looking at ourself */
298 struct pid_namespace
*ns
= task_active_pid_ns(current
);
300 lockdep_assert_held(&cgrp
->pidlist_mutex
);
302 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
303 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
309 * find the appropriate pidlist for our purpose (given procs vs tasks)
310 * returns with the lock on that pidlist already held, and takes care
311 * of the use count, or returns NULL with no locks held if we're out of
314 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
315 enum cgroup_filetype type
)
317 struct cgroup_pidlist
*l
;
319 lockdep_assert_held(&cgrp
->pidlist_mutex
);
321 l
= cgroup_pidlist_find(cgrp
, type
);
325 /* entry not found; create a new one */
326 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
330 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
332 /* don't need task_nsproxy() if we're looking at ourself */
333 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
335 list_add(&l
->links
, &cgrp
->pidlists
);
340 * cgroup_task_count - count the number of tasks in a cgroup.
341 * @cgrp: the cgroup in question
343 int cgroup_task_count(const struct cgroup
*cgrp
)
346 struct cgrp_cset_link
*link
;
348 spin_lock_irq(&css_set_lock
);
349 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
350 count
+= link
->cset
->nr_tasks
;
351 spin_unlock_irq(&css_set_lock
);
356 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
358 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
359 struct cgroup_pidlist
**lp
)
363 int pid
, n
= 0; /* used for populating the array */
364 struct css_task_iter it
;
365 struct task_struct
*tsk
;
366 struct cgroup_pidlist
*l
;
368 lockdep_assert_held(&cgrp
->pidlist_mutex
);
371 * If cgroup gets more users after we read count, we won't have
372 * enough space - tough. This race is indistinguishable to the
373 * caller from the case that the additional cgroup users didn't
374 * show up until sometime later on.
376 length
= cgroup_task_count(cgrp
);
377 array
= pidlist_allocate(length
);
380 /* now, populate the array */
381 css_task_iter_start(&cgrp
->self
, 0, &it
);
382 while ((tsk
= css_task_iter_next(&it
))) {
383 if (unlikely(n
== length
))
385 /* get tgid or pid for procs or tasks file respectively */
386 if (type
== CGROUP_FILE_PROCS
)
387 pid
= task_tgid_vnr(tsk
);
389 pid
= task_pid_vnr(tsk
);
390 if (pid
> 0) /* make sure to only use valid results */
393 css_task_iter_end(&it
);
395 /* now sort & (if procs) strip out duplicates */
396 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
397 if (type
== CGROUP_FILE_PROCS
)
398 length
= pidlist_uniq(array
, length
);
400 l
= cgroup_pidlist_find_create(cgrp
, type
);
406 /* store array, freeing old if necessary */
407 pidlist_free(l
->list
);
415 * seq_file methods for the tasks/procs files. The seq_file position is the
416 * next pid to display; the seq_file iterator is a pointer to the pid
417 * in the cgroup->l->list array.
420 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
423 * Initially we receive a position value that corresponds to
424 * one more than the last pid shown (or 0 on the first call or
425 * after a seek to the start). Use a binary-search to find the
426 * next pid to display, if any
428 struct kernfs_open_file
*of
= s
->private;
429 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
430 struct cgroup_pidlist
*l
;
431 enum cgroup_filetype type
= seq_cft(s
)->private;
432 int index
= 0, pid
= *pos
;
435 mutex_lock(&cgrp
->pidlist_mutex
);
438 * !NULL @of->priv indicates that this isn't the first start()
439 * after open. If the matching pidlist is around, we can use that.
440 * Look for it. Note that @of->priv can't be used directly. It
441 * could already have been destroyed.
444 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
447 * Either this is the first start() after open or the matching
448 * pidlist has been destroyed inbetween. Create a new one.
451 ret
= pidlist_array_load(cgrp
, type
,
452 (struct cgroup_pidlist
**)&of
->priv
);
461 while (index
< end
) {
462 int mid
= (index
+ end
) / 2;
463 if (l
->list
[mid
] == pid
) {
466 } else if (l
->list
[mid
] <= pid
)
472 /* If we're off the end of the array, we're done */
473 if (index
>= l
->length
)
475 /* Update the abstract position to be the actual pid that we found */
476 iter
= l
->list
+ index
;
481 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
483 struct kernfs_open_file
*of
= s
->private;
484 struct cgroup_pidlist
*l
= of
->priv
;
487 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
488 CGROUP_PIDLIST_DESTROY_DELAY
);
489 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
492 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
494 struct kernfs_open_file
*of
= s
->private;
495 struct cgroup_pidlist
*l
= of
->priv
;
497 pid_t
*end
= l
->list
+ l
->length
;
499 * Advance to the next pid in the array. If this goes off the
511 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
513 seq_printf(s
, "%d\n", *(int *)v
);
518 static ssize_t
__cgroup1_procs_write(struct kernfs_open_file
*of
,
519 char *buf
, size_t nbytes
, loff_t off
,
523 struct task_struct
*task
;
524 const struct cred
*cred
, *tcred
;
527 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
531 task
= cgroup_procs_write_start(buf
, threadgroup
);
532 ret
= PTR_ERR_OR_ZERO(task
);
537 * Even if we're attaching all tasks in the thread group, we only
538 * need to check permissions on one of them.
540 cred
= current_cred();
541 tcred
= get_task_cred(task
);
542 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
543 !uid_eq(cred
->euid
, tcred
->uid
) &&
544 !uid_eq(cred
->euid
, tcred
->suid
))
550 ret
= cgroup_attach_task(cgrp
, task
, threadgroup
);
553 cgroup_procs_write_finish(task
);
555 cgroup_kn_unlock(of
->kn
);
557 return ret
?: nbytes
;
560 static ssize_t
cgroup1_procs_write(struct kernfs_open_file
*of
,
561 char *buf
, size_t nbytes
, loff_t off
)
563 return __cgroup1_procs_write(of
, buf
, nbytes
, off
, true);
566 static ssize_t
cgroup1_tasks_write(struct kernfs_open_file
*of
,
567 char *buf
, size_t nbytes
, loff_t off
)
569 return __cgroup1_procs_write(of
, buf
, nbytes
, off
, false);
572 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
573 char *buf
, size_t nbytes
, loff_t off
)
577 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
579 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
582 spin_lock(&release_agent_path_lock
);
583 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
584 sizeof(cgrp
->root
->release_agent_path
));
585 spin_unlock(&release_agent_path_lock
);
586 cgroup_kn_unlock(of
->kn
);
590 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
592 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
594 spin_lock(&release_agent_path_lock
);
595 seq_puts(seq
, cgrp
->root
->release_agent_path
);
596 spin_unlock(&release_agent_path_lock
);
601 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
603 seq_puts(seq
, "0\n");
607 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
610 return notify_on_release(css
->cgroup
);
613 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
614 struct cftype
*cft
, u64 val
)
617 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
619 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
623 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
626 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
629 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
630 struct cftype
*cft
, u64 val
)
633 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
635 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
639 /* cgroup core interface files for the legacy hierarchies */
640 struct cftype cgroup1_base_files
[] = {
642 .name
= "cgroup.procs",
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_PROCS
,
648 .write
= cgroup1_procs_write
,
651 .name
= "cgroup.clone_children",
652 .read_u64
= cgroup_clone_children_read
,
653 .write_u64
= cgroup_clone_children_write
,
656 .name
= "cgroup.sane_behavior",
657 .flags
= CFTYPE_ONLY_ON_ROOT
,
658 .seq_show
= cgroup_sane_behavior_show
,
662 .seq_start
= cgroup_pidlist_start
,
663 .seq_next
= cgroup_pidlist_next
,
664 .seq_stop
= cgroup_pidlist_stop
,
665 .seq_show
= cgroup_pidlist_show
,
666 .private = CGROUP_FILE_TASKS
,
667 .write
= cgroup1_tasks_write
,
670 .name
= "notify_on_release",
671 .read_u64
= cgroup_read_notify_on_release
,
672 .write_u64
= cgroup_write_notify_on_release
,
675 .name
= "release_agent",
676 .flags
= CFTYPE_ONLY_ON_ROOT
,
677 .seq_show
= cgroup_release_agent_show
,
678 .write
= cgroup_release_agent_write
,
679 .max_write_len
= PATH_MAX
- 1,
684 /* Display information about each subsystem and each hierarchy */
685 int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
687 struct cgroup_subsys
*ss
;
690 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
692 * ideally we don't want subsystems moving around while we do this.
693 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
694 * subsys/hierarchy state.
696 mutex_lock(&cgroup_mutex
);
698 for_each_subsys(ss
, i
)
699 seq_printf(m
, "%s\t%d\t%d\t%d\n",
700 ss
->legacy_name
, ss
->root
->hierarchy_id
,
701 atomic_read(&ss
->root
->nr_cgrps
),
702 cgroup_ssid_enabled(i
));
704 mutex_unlock(&cgroup_mutex
);
709 * cgroupstats_build - build and fill cgroupstats
710 * @stats: cgroupstats to fill information into
711 * @dentry: A dentry entry belonging to the cgroup for which stats have
714 * Build and fill cgroupstats so that taskstats can export it to user
717 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
719 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
721 struct css_task_iter it
;
722 struct task_struct
*tsk
;
724 /* it should be kernfs_node belonging to cgroupfs and is a directory */
725 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
726 kernfs_type(kn
) != KERNFS_DIR
)
729 mutex_lock(&cgroup_mutex
);
732 * We aren't being called from kernfs and there's no guarantee on
733 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
734 * @kn->priv is RCU safe. Let's do the RCU dancing.
737 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
738 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
740 mutex_unlock(&cgroup_mutex
);
745 css_task_iter_start(&cgrp
->self
, 0, &it
);
746 while ((tsk
= css_task_iter_next(&it
))) {
747 switch (tsk
->state
) {
751 case TASK_INTERRUPTIBLE
:
752 stats
->nr_sleeping
++;
754 case TASK_UNINTERRUPTIBLE
:
755 stats
->nr_uninterruptible
++;
761 if (delayacct_is_task_waiting_on_io(tsk
))
766 css_task_iter_end(&it
);
768 mutex_unlock(&cgroup_mutex
);
772 void cgroup1_check_for_release(struct cgroup
*cgrp
)
774 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
775 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
776 schedule_work(&cgrp
->release_agent_work
);
780 * Notify userspace when a cgroup is released, by running the
781 * configured release agent with the name of the cgroup (path
782 * relative to the root of cgroup file system) as the argument.
784 * Most likely, this user command will try to rmdir this cgroup.
786 * This races with the possibility that some other task will be
787 * attached to this cgroup before it is removed, or that some other
788 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
789 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
790 * unused, and this cgroup will be reprieved from its death sentence,
791 * to continue to serve a useful existence. Next time it's released,
792 * we will get notified again, if it still has 'notify_on_release' set.
794 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
795 * means only wait until the task is successfully execve()'d. The
796 * separate release agent task is forked by call_usermodehelper(),
797 * then control in this thread returns here, without waiting for the
798 * release agent task. We don't bother to wait because the caller of
799 * this routine has no use for the exit status of the release agent
800 * task, so no sense holding our caller up for that.
802 void cgroup1_release_agent(struct work_struct
*work
)
804 struct cgroup
*cgrp
=
805 container_of(work
, struct cgroup
, release_agent_work
);
806 char *pathbuf
= NULL
, *agentbuf
= NULL
;
807 char *argv
[3], *envp
[3];
810 mutex_lock(&cgroup_mutex
);
812 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
813 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
814 if (!pathbuf
|| !agentbuf
)
817 spin_lock_irq(&css_set_lock
);
818 ret
= cgroup_path_ns_locked(cgrp
, pathbuf
, PATH_MAX
, &init_cgroup_ns
);
819 spin_unlock_irq(&css_set_lock
);
820 if (ret
< 0 || ret
>= PATH_MAX
)
827 /* minimal command environment */
829 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
832 mutex_unlock(&cgroup_mutex
);
833 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
836 mutex_unlock(&cgroup_mutex
);
843 * cgroup_rename - Only allow simple rename of directories in place.
845 static int cgroup1_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
846 const char *new_name_str
)
848 struct cgroup
*cgrp
= kn
->priv
;
851 if (kernfs_type(kn
) != KERNFS_DIR
)
853 if (kn
->parent
!= new_parent
)
857 * We're gonna grab cgroup_mutex which nests outside kernfs
858 * active_ref. kernfs_rename() doesn't require active_ref
859 * protection. Break them before grabbing cgroup_mutex.
861 kernfs_break_active_protection(new_parent
);
862 kernfs_break_active_protection(kn
);
864 mutex_lock(&cgroup_mutex
);
866 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
868 trace_cgroup_rename(cgrp
);
870 mutex_unlock(&cgroup_mutex
);
872 kernfs_unbreak_active_protection(kn
);
873 kernfs_unbreak_active_protection(new_parent
);
877 static int cgroup1_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
879 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
880 struct cgroup_subsys
*ss
;
883 for_each_subsys(ss
, ssid
)
884 if (root
->subsys_mask
& (1 << ssid
))
885 seq_show_option(seq
, ss
->legacy_name
, NULL
);
886 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
887 seq_puts(seq
, ",noprefix");
888 if (root
->flags
& CGRP_ROOT_XATTR
)
889 seq_puts(seq
, ",xattr");
890 if (root
->flags
& CGRP_ROOT_CPUSET_V2_MODE
)
891 seq_puts(seq
, ",cpuset_v2_mode");
893 spin_lock(&release_agent_path_lock
);
894 if (strlen(root
->release_agent_path
))
895 seq_show_option(seq
, "release_agent",
896 root
->release_agent_path
);
897 spin_unlock(&release_agent_path_lock
);
899 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
900 seq_puts(seq
, ",clone_children");
901 if (strlen(root
->name
))
902 seq_show_option(seq
, "name", root
->name
);
906 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
908 char *token
, *o
= data
;
909 bool all_ss
= false, one_ss
= false;
911 struct cgroup_subsys
*ss
;
915 #ifdef CONFIG_CPUSETS
916 mask
= ~((u16
)1 << cpuset_cgrp_id
);
919 memset(opts
, 0, sizeof(*opts
));
921 while ((token
= strsep(&o
, ",")) != NULL
) {
926 if (!strcmp(token
, "none")) {
927 /* Explicitly have no subsystems */
931 if (!strcmp(token
, "all")) {
932 /* Mutually exclusive option 'all' + subsystem name */
938 if (!strcmp(token
, "noprefix")) {
939 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
942 if (!strcmp(token
, "clone_children")) {
943 opts
->cpuset_clone_children
= true;
946 if (!strcmp(token
, "cpuset_v2_mode")) {
947 opts
->flags
|= CGRP_ROOT_CPUSET_V2_MODE
;
950 if (!strcmp(token
, "xattr")) {
951 opts
->flags
|= CGRP_ROOT_XATTR
;
954 if (!strncmp(token
, "release_agent=", 14)) {
955 /* Specifying two release agents is forbidden */
956 if (opts
->release_agent
)
958 opts
->release_agent
=
959 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
960 if (!opts
->release_agent
)
964 if (!strncmp(token
, "name=", 5)) {
965 const char *name
= token
+ 5;
966 /* Can't specify an empty name */
969 /* Must match [\w.-]+ */
970 for (i
= 0; i
< strlen(name
); i
++) {
974 if ((c
== '.') || (c
== '-') || (c
== '_'))
978 /* Specifying two names is forbidden */
981 opts
->name
= kstrndup(name
,
982 MAX_CGROUP_ROOT_NAMELEN
- 1,
990 for_each_subsys(ss
, i
) {
991 if (strcmp(token
, ss
->legacy_name
))
993 if (!cgroup_ssid_enabled(i
))
995 if (cgroup1_ssid_disabled(i
))
998 /* Mutually exclusive option 'all' + subsystem name */
1001 opts
->subsys_mask
|= (1 << i
);
1006 if (i
== CGROUP_SUBSYS_COUNT
)
1011 * If the 'all' option was specified select all the subsystems,
1012 * otherwise if 'none', 'name=' and a subsystem name options were
1013 * not specified, let's default to 'all'
1015 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1016 for_each_subsys(ss
, i
)
1017 if (cgroup_ssid_enabled(i
) && !cgroup1_ssid_disabled(i
))
1018 opts
->subsys_mask
|= (1 << i
);
1021 * We either have to specify by name or by subsystems. (So all
1022 * empty hierarchies must have a name).
1024 if (!opts
->subsys_mask
&& !opts
->name
)
1028 * Option noprefix was introduced just for backward compatibility
1029 * with the old cpuset, so we allow noprefix only if mounting just
1030 * the cpuset subsystem.
1032 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1035 /* Can't specify "none" and some subsystems */
1036 if (opts
->subsys_mask
&& opts
->none
)
1042 static int cgroup1_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1045 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1046 struct cgroup_sb_opts opts
;
1047 u16 added_mask
, removed_mask
;
1049 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1051 /* See what subsystems are wanted */
1052 ret
= parse_cgroupfs_options(data
, &opts
);
1056 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1057 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1058 task_tgid_nr(current
), current
->comm
);
1060 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1061 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1063 /* Don't allow flags or name to change at remount */
1064 if ((opts
.flags
^ root
->flags
) ||
1065 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1066 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1067 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1072 /* remounting is not allowed for populated hierarchies */
1073 if (!list_empty(&root
->cgrp
.self
.children
)) {
1078 ret
= rebind_subsystems(root
, added_mask
);
1082 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, removed_mask
));
1084 if (opts
.release_agent
) {
1085 spin_lock(&release_agent_path_lock
);
1086 strcpy(root
->release_agent_path
, opts
.release_agent
);
1087 spin_unlock(&release_agent_path_lock
);
1090 trace_cgroup_remount(root
);
1093 kfree(opts
.release_agent
);
1095 mutex_unlock(&cgroup_mutex
);
1099 struct kernfs_syscall_ops cgroup1_kf_syscall_ops
= {
1100 .rename
= cgroup1_rename
,
1101 .show_options
= cgroup1_show_options
,
1102 .remount_fs
= cgroup1_remount
,
1103 .mkdir
= cgroup_mkdir
,
1104 .rmdir
= cgroup_rmdir
,
1105 .show_path
= cgroup_show_path
,
1108 struct dentry
*cgroup1_mount(struct file_system_type
*fs_type
, int flags
,
1109 void *data
, unsigned long magic
,
1110 struct cgroup_namespace
*ns
)
1112 struct super_block
*pinned_sb
= NULL
;
1113 struct cgroup_sb_opts opts
;
1114 struct cgroup_root
*root
;
1115 struct cgroup_subsys
*ss
;
1116 struct dentry
*dentry
;
1118 bool new_root
= false;
1120 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1122 /* First find the desired set of subsystems */
1123 ret
= parse_cgroupfs_options(data
, &opts
);
1128 * Destruction of cgroup root is asynchronous, so subsystems may
1129 * still be dying after the previous unmount. Let's drain the
1130 * dying subsystems. We just need to ensure that the ones
1131 * unmounted previously finish dying and don't care about new ones
1132 * starting. Testing ref liveliness is good enough.
1134 for_each_subsys(ss
, i
) {
1135 if (!(opts
.subsys_mask
& (1 << i
)) ||
1136 ss
->root
== &cgrp_dfl_root
)
1139 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
1140 mutex_unlock(&cgroup_mutex
);
1142 ret
= restart_syscall();
1145 cgroup_put(&ss
->root
->cgrp
);
1148 for_each_root(root
) {
1149 bool name_match
= false;
1151 if (root
== &cgrp_dfl_root
)
1155 * If we asked for a name then it must match. Also, if
1156 * name matches but sybsys_mask doesn't, we should fail.
1157 * Remember whether name matched.
1160 if (strcmp(opts
.name
, root
->name
))
1166 * If we asked for subsystems (or explicitly for no
1167 * subsystems) then they must match.
1169 if ((opts
.subsys_mask
|| opts
.none
) &&
1170 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1177 if (root
->flags
^ opts
.flags
)
1178 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1181 * We want to reuse @root whose lifetime is governed by its
1182 * ->cgrp. Let's check whether @root is alive and keep it
1183 * that way. As cgroup_kill_sb() can happen anytime, we
1184 * want to block it by pinning the sb so that @root doesn't
1185 * get killed before mount is complete.
1187 * With the sb pinned, tryget_live can reliably indicate
1188 * whether @root can be reused. If it's being killed,
1189 * drain it. We can use wait_queue for the wait but this
1190 * path is super cold. Let's just sleep a bit and retry.
1192 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
1193 if (IS_ERR(pinned_sb
) ||
1194 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1195 mutex_unlock(&cgroup_mutex
);
1196 if (!IS_ERR_OR_NULL(pinned_sb
))
1197 deactivate_super(pinned_sb
);
1199 ret
= restart_syscall();
1208 * No such thing, create a new one. name= matching without subsys
1209 * specification is allowed for already existing hierarchies but we
1210 * can't create new one without subsys specification.
1212 if (!opts
.subsys_mask
&& !opts
.none
) {
1217 /* Hierarchies may only be created in the initial cgroup namespace. */
1218 if (ns
!= &init_cgroup_ns
) {
1223 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1230 init_cgroup_root(root
, &opts
);
1232 ret
= cgroup_setup_root(root
, opts
.subsys_mask
, PERCPU_REF_INIT_DEAD
);
1234 cgroup_free_root(root
);
1237 mutex_unlock(&cgroup_mutex
);
1239 kfree(opts
.release_agent
);
1243 return ERR_PTR(ret
);
1245 dentry
= cgroup_do_mount(&cgroup_fs_type
, flags
, root
,
1246 CGROUP_SUPER_MAGIC
, ns
);
1249 * There's a race window after we release cgroup_mutex and before
1250 * allocating a superblock. Make sure a concurrent process won't
1251 * be able to re-use the root during this window by delaying the
1252 * initialization of root refcnt.
1255 mutex_lock(&cgroup_mutex
);
1256 percpu_ref_reinit(&root
->cgrp
.self
.refcnt
);
1257 mutex_unlock(&cgroup_mutex
);
1261 * If @pinned_sb, we're reusing an existing root and holding an
1262 * extra ref on its sb. Mount is complete. Put the extra ref.
1265 deactivate_super(pinned_sb
);
1270 static int __init
cgroup1_wq_init(void)
1273 * Used to destroy pidlists and separate to serve as flush domain.
1274 * Cap @max_active to 1 too.
1276 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
1278 BUG_ON(!cgroup_pidlist_destroy_wq
);
1281 core_initcall(cgroup1_wq_init
);
1283 static int __init
cgroup_no_v1(char *str
)
1285 struct cgroup_subsys
*ss
;
1289 while ((token
= strsep(&str
, ",")) != NULL
) {
1293 if (!strcmp(token
, "all")) {
1294 cgroup_no_v1_mask
= U16_MAX
;
1298 for_each_subsys(ss
, i
) {
1299 if (strcmp(token
, ss
->name
) &&
1300 strcmp(token
, ss
->legacy_name
))
1303 cgroup_no_v1_mask
|= 1 << i
;
1308 __setup("cgroup_no_v1=", cgroup_no_v1
);