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 if (!cgroup_may_migrate_to(to
))
105 mutex_lock(&cgroup_mutex
);
107 percpu_down_write(&cgroup_threadgroup_rwsem
);
109 /* all tasks in @from are being moved, all csets are source */
110 spin_lock_irq(&css_set_lock
);
111 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
112 cgroup_migrate_add_src(link
->cset
, to
, &mgctx
);
113 spin_unlock_irq(&css_set_lock
);
115 ret
= cgroup_migrate_prepare_dst(&mgctx
);
120 * Migrate tasks one-by-one until @from is empty. This fails iff
121 * ->can_attach() fails.
124 css_task_iter_start(&from
->self
, &it
);
125 task
= css_task_iter_next(&it
);
127 get_task_struct(task
);
128 css_task_iter_end(&it
);
131 ret
= cgroup_migrate(task
, false, &mgctx
);
133 trace_cgroup_transfer_tasks(to
, task
, false);
134 put_task_struct(task
);
136 } while (task
&& !ret
);
138 cgroup_migrate_finish(&mgctx
);
139 percpu_up_write(&cgroup_threadgroup_rwsem
);
140 mutex_unlock(&cgroup_mutex
);
145 * Stuff for reading the 'tasks'/'procs' files.
147 * Reading this file can return large amounts of data if a cgroup has
148 * *lots* of attached tasks. So it may need several calls to read(),
149 * but we cannot guarantee that the information we produce is correct
150 * unless we produce it entirely atomically.
154 /* which pidlist file are we talking about? */
155 enum cgroup_filetype
{
161 * A pidlist is a list of pids that virtually represents the contents of one
162 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
163 * a pair (one each for procs, tasks) for each pid namespace that's relevant
166 struct cgroup_pidlist
{
168 * used to find which pidlist is wanted. doesn't change as long as
169 * this particular list stays in the list.
171 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
174 /* how many elements the above list has */
176 /* each of these stored in a list by its cgroup */
177 struct list_head links
;
178 /* pointer to the cgroup we belong to, for list removal purposes */
179 struct cgroup
*owner
;
180 /* for delayed destruction */
181 struct delayed_work destroy_dwork
;
185 * The following two functions "fix" the issue where there are more pids
186 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
187 * TODO: replace with a kernel-wide solution to this problem
189 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
190 static void *pidlist_allocate(int count
)
192 if (PIDLIST_TOO_LARGE(count
))
193 return vmalloc(count
* sizeof(pid_t
));
195 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
198 static void pidlist_free(void *p
)
204 * Used to destroy all pidlists lingering waiting for destroy timer. None
205 * should be left afterwards.
207 void cgroup1_pidlist_destroy_all(struct cgroup
*cgrp
)
209 struct cgroup_pidlist
*l
, *tmp_l
;
211 mutex_lock(&cgrp
->pidlist_mutex
);
212 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
213 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
214 mutex_unlock(&cgrp
->pidlist_mutex
);
216 flush_workqueue(cgroup_pidlist_destroy_wq
);
217 BUG_ON(!list_empty(&cgrp
->pidlists
));
220 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
222 struct delayed_work
*dwork
= to_delayed_work(work
);
223 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
225 struct cgroup_pidlist
*tofree
= NULL
;
227 mutex_lock(&l
->owner
->pidlist_mutex
);
230 * Destroy iff we didn't get queued again. The state won't change
231 * as destroy_dwork can only be queued while locked.
233 if (!delayed_work_pending(dwork
)) {
235 pidlist_free(l
->list
);
236 put_pid_ns(l
->key
.ns
);
240 mutex_unlock(&l
->owner
->pidlist_mutex
);
245 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
246 * Returns the number of unique elements.
248 static int pidlist_uniq(pid_t
*list
, int length
)
253 * we presume the 0th element is unique, so i starts at 1. trivial
254 * edge cases first; no work needs to be done for either
256 if (length
== 0 || length
== 1)
258 /* src and dest walk down the list; dest counts unique elements */
259 for (src
= 1; src
< length
; src
++) {
260 /* find next unique element */
261 while (list
[src
] == list
[src
-1]) {
266 /* dest always points to where the next unique element goes */
267 list
[dest
] = list
[src
];
275 * The two pid files - task and cgroup.procs - guaranteed that the result
276 * is sorted, which forced this whole pidlist fiasco. As pid order is
277 * different per namespace, each namespace needs differently sorted list,
278 * making it impossible to use, for example, single rbtree of member tasks
279 * sorted by task pointer. As pidlists can be fairly large, allocating one
280 * per open file is dangerous, so cgroup had to implement shared pool of
281 * pidlists keyed by cgroup and namespace.
283 static int cmppid(const void *a
, const void *b
)
285 return *(pid_t
*)a
- *(pid_t
*)b
;
288 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
289 enum cgroup_filetype type
)
291 struct cgroup_pidlist
*l
;
292 /* don't need task_nsproxy() if we're looking at ourself */
293 struct pid_namespace
*ns
= task_active_pid_ns(current
);
295 lockdep_assert_held(&cgrp
->pidlist_mutex
);
297 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
298 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
304 * find the appropriate pidlist for our purpose (given procs vs tasks)
305 * returns with the lock on that pidlist already held, and takes care
306 * of the use count, or returns NULL with no locks held if we're out of
309 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
310 enum cgroup_filetype type
)
312 struct cgroup_pidlist
*l
;
314 lockdep_assert_held(&cgrp
->pidlist_mutex
);
316 l
= cgroup_pidlist_find(cgrp
, type
);
320 /* entry not found; create a new one */
321 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
325 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
327 /* don't need task_nsproxy() if we're looking at ourself */
328 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
330 list_add(&l
->links
, &cgrp
->pidlists
);
335 * cgroup_task_count - count the number of tasks in a cgroup.
336 * @cgrp: the cgroup in question
338 * Return the number of tasks in the cgroup. The returned number can be
339 * higher than the actual number of tasks due to css_set references from
340 * namespace roots and temporary usages.
342 static int cgroup_task_count(const struct cgroup
*cgrp
)
345 struct cgrp_cset_link
*link
;
347 spin_lock_irq(&css_set_lock
);
348 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
349 count
+= refcount_read(&link
->cset
->refcount
);
350 spin_unlock_irq(&css_set_lock
);
355 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
357 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
358 struct cgroup_pidlist
**lp
)
362 int pid
, n
= 0; /* used for populating the array */
363 struct css_task_iter it
;
364 struct task_struct
*tsk
;
365 struct cgroup_pidlist
*l
;
367 lockdep_assert_held(&cgrp
->pidlist_mutex
);
370 * If cgroup gets more users after we read count, we won't have
371 * enough space - tough. This race is indistinguishable to the
372 * caller from the case that the additional cgroup users didn't
373 * show up until sometime later on.
375 length
= cgroup_task_count(cgrp
);
376 array
= pidlist_allocate(length
);
379 /* now, populate the array */
380 css_task_iter_start(&cgrp
->self
, &it
);
381 while ((tsk
= css_task_iter_next(&it
))) {
382 if (unlikely(n
== length
))
384 /* get tgid or pid for procs or tasks file respectively */
385 if (type
== CGROUP_FILE_PROCS
)
386 pid
= task_tgid_vnr(tsk
);
388 pid
= task_pid_vnr(tsk
);
389 if (pid
> 0) /* make sure to only use valid results */
392 css_task_iter_end(&it
);
394 /* now sort & (if procs) strip out duplicates */
395 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
396 if (type
== CGROUP_FILE_PROCS
)
397 length
= pidlist_uniq(array
, length
);
399 l
= cgroup_pidlist_find_create(cgrp
, type
);
405 /* store array, freeing old if necessary */
406 pidlist_free(l
->list
);
414 * seq_file methods for the tasks/procs files. The seq_file position is the
415 * next pid to display; the seq_file iterator is a pointer to the pid
416 * in the cgroup->l->list array.
419 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
422 * Initially we receive a position value that corresponds to
423 * one more than the last pid shown (or 0 on the first call or
424 * after a seek to the start). Use a binary-search to find the
425 * next pid to display, if any
427 struct kernfs_open_file
*of
= s
->private;
428 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
429 struct cgroup_pidlist
*l
;
430 enum cgroup_filetype type
= seq_cft(s
)->private;
431 int index
= 0, pid
= *pos
;
434 mutex_lock(&cgrp
->pidlist_mutex
);
437 * !NULL @of->priv indicates that this isn't the first start()
438 * after open. If the matching pidlist is around, we can use that.
439 * Look for it. Note that @of->priv can't be used directly. It
440 * could already have been destroyed.
443 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
446 * Either this is the first start() after open or the matching
447 * pidlist has been destroyed inbetween. Create a new one.
450 ret
= pidlist_array_load(cgrp
, type
,
451 (struct cgroup_pidlist
**)&of
->priv
);
460 while (index
< end
) {
461 int mid
= (index
+ end
) / 2;
462 if (l
->list
[mid
] == pid
) {
465 } else if (l
->list
[mid
] <= pid
)
471 /* If we're off the end of the array, we're done */
472 if (index
>= l
->length
)
474 /* Update the abstract position to be the actual pid that we found */
475 iter
= l
->list
+ index
;
480 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
482 struct kernfs_open_file
*of
= s
->private;
483 struct cgroup_pidlist
*l
= of
->priv
;
486 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
487 CGROUP_PIDLIST_DESTROY_DELAY
);
488 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
491 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
493 struct kernfs_open_file
*of
= s
->private;
494 struct cgroup_pidlist
*l
= of
->priv
;
496 pid_t
*end
= l
->list
+ l
->length
;
498 * Advance to the next pid in the array. If this goes off the
510 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
512 seq_printf(s
, "%d\n", *(int *)v
);
517 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
518 char *buf
, size_t nbytes
, loff_t off
)
520 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
523 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
524 char *buf
, size_t nbytes
, loff_t off
)
528 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
530 cgrp
= cgroup_kn_lock_live(of
->kn
, false);
533 spin_lock(&release_agent_path_lock
);
534 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
535 sizeof(cgrp
->root
->release_agent_path
));
536 spin_unlock(&release_agent_path_lock
);
537 cgroup_kn_unlock(of
->kn
);
541 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
543 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
545 spin_lock(&release_agent_path_lock
);
546 seq_puts(seq
, cgrp
->root
->release_agent_path
);
547 spin_unlock(&release_agent_path_lock
);
552 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
554 seq_puts(seq
, "0\n");
558 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
561 return notify_on_release(css
->cgroup
);
564 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
565 struct cftype
*cft
, u64 val
)
568 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
570 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
574 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
577 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
580 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
581 struct cftype
*cft
, u64 val
)
584 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
586 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
590 /* cgroup core interface files for the legacy hierarchies */
591 struct cftype cgroup1_base_files
[] = {
593 .name
= "cgroup.procs",
594 .seq_start
= cgroup_pidlist_start
,
595 .seq_next
= cgroup_pidlist_next
,
596 .seq_stop
= cgroup_pidlist_stop
,
597 .seq_show
= cgroup_pidlist_show
,
598 .private = CGROUP_FILE_PROCS
,
599 .write
= cgroup_procs_write
,
602 .name
= "cgroup.clone_children",
603 .read_u64
= cgroup_clone_children_read
,
604 .write_u64
= cgroup_clone_children_write
,
607 .name
= "cgroup.sane_behavior",
608 .flags
= CFTYPE_ONLY_ON_ROOT
,
609 .seq_show
= cgroup_sane_behavior_show
,
613 .seq_start
= cgroup_pidlist_start
,
614 .seq_next
= cgroup_pidlist_next
,
615 .seq_stop
= cgroup_pidlist_stop
,
616 .seq_show
= cgroup_pidlist_show
,
617 .private = CGROUP_FILE_TASKS
,
618 .write
= cgroup_tasks_write
,
621 .name
= "notify_on_release",
622 .read_u64
= cgroup_read_notify_on_release
,
623 .write_u64
= cgroup_write_notify_on_release
,
626 .name
= "release_agent",
627 .flags
= CFTYPE_ONLY_ON_ROOT
,
628 .seq_show
= cgroup_release_agent_show
,
629 .write
= cgroup_release_agent_write
,
630 .max_write_len
= PATH_MAX
- 1,
635 /* Display information about each subsystem and each hierarchy */
636 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
638 struct cgroup_subsys
*ss
;
641 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
643 * ideally we don't want subsystems moving around while we do this.
644 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
645 * subsys/hierarchy state.
647 mutex_lock(&cgroup_mutex
);
649 for_each_subsys(ss
, i
)
650 seq_printf(m
, "%s\t%d\t%d\t%d\n",
651 ss
->legacy_name
, ss
->root
->hierarchy_id
,
652 atomic_read(&ss
->root
->nr_cgrps
),
653 cgroup_ssid_enabled(i
));
655 mutex_unlock(&cgroup_mutex
);
659 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
661 return single_open(file
, proc_cgroupstats_show
, NULL
);
664 const struct file_operations proc_cgroupstats_operations
= {
665 .open
= cgroupstats_open
,
668 .release
= single_release
,
672 * cgroupstats_build - build and fill cgroupstats
673 * @stats: cgroupstats to fill information into
674 * @dentry: A dentry entry belonging to the cgroup for which stats have
677 * Build and fill cgroupstats so that taskstats can export it to user
680 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
682 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
684 struct css_task_iter it
;
685 struct task_struct
*tsk
;
687 /* it should be kernfs_node belonging to cgroupfs and is a directory */
688 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
689 kernfs_type(kn
) != KERNFS_DIR
)
692 mutex_lock(&cgroup_mutex
);
695 * We aren't being called from kernfs and there's no guarantee on
696 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
697 * @kn->priv is RCU safe. Let's do the RCU dancing.
700 cgrp
= rcu_dereference(*(void __rcu __force
**)&kn
->priv
);
701 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
703 mutex_unlock(&cgroup_mutex
);
708 css_task_iter_start(&cgrp
->self
, &it
);
709 while ((tsk
= css_task_iter_next(&it
))) {
710 switch (tsk
->state
) {
714 case TASK_INTERRUPTIBLE
:
715 stats
->nr_sleeping
++;
717 case TASK_UNINTERRUPTIBLE
:
718 stats
->nr_uninterruptible
++;
724 if (delayacct_is_task_waiting_on_io(tsk
))
729 css_task_iter_end(&it
);
731 mutex_unlock(&cgroup_mutex
);
735 void cgroup1_check_for_release(struct cgroup
*cgrp
)
737 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
738 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
739 schedule_work(&cgrp
->release_agent_work
);
743 * Notify userspace when a cgroup is released, by running the
744 * configured release agent with the name of the cgroup (path
745 * relative to the root of cgroup file system) as the argument.
747 * Most likely, this user command will try to rmdir this cgroup.
749 * This races with the possibility that some other task will be
750 * attached to this cgroup before it is removed, or that some other
751 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
752 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
753 * unused, and this cgroup will be reprieved from its death sentence,
754 * to continue to serve a useful existence. Next time it's released,
755 * we will get notified again, if it still has 'notify_on_release' set.
757 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
758 * means only wait until the task is successfully execve()'d. The
759 * separate release agent task is forked by call_usermodehelper(),
760 * then control in this thread returns here, without waiting for the
761 * release agent task. We don't bother to wait because the caller of
762 * this routine has no use for the exit status of the release agent
763 * task, so no sense holding our caller up for that.
765 void cgroup1_release_agent(struct work_struct
*work
)
767 struct cgroup
*cgrp
=
768 container_of(work
, struct cgroup
, release_agent_work
);
769 char *pathbuf
= NULL
, *agentbuf
= NULL
;
770 char *argv
[3], *envp
[3];
773 mutex_lock(&cgroup_mutex
);
775 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
776 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
777 if (!pathbuf
|| !agentbuf
)
780 spin_lock_irq(&css_set_lock
);
781 ret
= cgroup_path_ns_locked(cgrp
, pathbuf
, PATH_MAX
, &init_cgroup_ns
);
782 spin_unlock_irq(&css_set_lock
);
783 if (ret
< 0 || ret
>= PATH_MAX
)
790 /* minimal command environment */
792 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
795 mutex_unlock(&cgroup_mutex
);
796 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
799 mutex_unlock(&cgroup_mutex
);
806 * cgroup_rename - Only allow simple rename of directories in place.
808 static int cgroup1_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
809 const char *new_name_str
)
811 struct cgroup
*cgrp
= kn
->priv
;
814 if (kernfs_type(kn
) != KERNFS_DIR
)
816 if (kn
->parent
!= new_parent
)
820 * We're gonna grab cgroup_mutex which nests outside kernfs
821 * active_ref. kernfs_rename() doesn't require active_ref
822 * protection. Break them before grabbing cgroup_mutex.
824 kernfs_break_active_protection(new_parent
);
825 kernfs_break_active_protection(kn
);
827 mutex_lock(&cgroup_mutex
);
829 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
831 trace_cgroup_rename(cgrp
);
833 mutex_unlock(&cgroup_mutex
);
835 kernfs_unbreak_active_protection(kn
);
836 kernfs_unbreak_active_protection(new_parent
);
840 static int cgroup1_show_options(struct seq_file
*seq
, struct kernfs_root
*kf_root
)
842 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
843 struct cgroup_subsys
*ss
;
846 for_each_subsys(ss
, ssid
)
847 if (root
->subsys_mask
& (1 << ssid
))
848 seq_show_option(seq
, ss
->legacy_name
, NULL
);
849 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
850 seq_puts(seq
, ",noprefix");
851 if (root
->flags
& CGRP_ROOT_XATTR
)
852 seq_puts(seq
, ",xattr");
854 spin_lock(&release_agent_path_lock
);
855 if (strlen(root
->release_agent_path
))
856 seq_show_option(seq
, "release_agent",
857 root
->release_agent_path
);
858 spin_unlock(&release_agent_path_lock
);
860 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
861 seq_puts(seq
, ",clone_children");
862 if (strlen(root
->name
))
863 seq_show_option(seq
, "name", root
->name
);
867 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
869 char *token
, *o
= data
;
870 bool all_ss
= false, one_ss
= false;
872 struct cgroup_subsys
*ss
;
876 #ifdef CONFIG_CPUSETS
877 mask
= ~((u16
)1 << cpuset_cgrp_id
);
880 memset(opts
, 0, sizeof(*opts
));
882 while ((token
= strsep(&o
, ",")) != NULL
) {
887 if (!strcmp(token
, "none")) {
888 /* Explicitly have no subsystems */
892 if (!strcmp(token
, "all")) {
893 /* Mutually exclusive option 'all' + subsystem name */
899 if (!strcmp(token
, "noprefix")) {
900 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
903 if (!strcmp(token
, "clone_children")) {
904 opts
->cpuset_clone_children
= true;
907 if (!strcmp(token
, "xattr")) {
908 opts
->flags
|= CGRP_ROOT_XATTR
;
911 if (!strncmp(token
, "release_agent=", 14)) {
912 /* Specifying two release agents is forbidden */
913 if (opts
->release_agent
)
915 opts
->release_agent
=
916 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
917 if (!opts
->release_agent
)
921 if (!strncmp(token
, "name=", 5)) {
922 const char *name
= token
+ 5;
923 /* Can't specify an empty name */
926 /* Must match [\w.-]+ */
927 for (i
= 0; i
< strlen(name
); i
++) {
931 if ((c
== '.') || (c
== '-') || (c
== '_'))
935 /* Specifying two names is forbidden */
938 opts
->name
= kstrndup(name
,
939 MAX_CGROUP_ROOT_NAMELEN
- 1,
947 for_each_subsys(ss
, i
) {
948 if (strcmp(token
, ss
->legacy_name
))
950 if (!cgroup_ssid_enabled(i
))
952 if (cgroup1_ssid_disabled(i
))
955 /* Mutually exclusive option 'all' + subsystem name */
958 opts
->subsys_mask
|= (1 << i
);
963 if (i
== CGROUP_SUBSYS_COUNT
)
968 * If the 'all' option was specified select all the subsystems,
969 * otherwise if 'none', 'name=' and a subsystem name options were
970 * not specified, let's default to 'all'
972 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
973 for_each_subsys(ss
, i
)
974 if (cgroup_ssid_enabled(i
) && !cgroup1_ssid_disabled(i
))
975 opts
->subsys_mask
|= (1 << i
);
978 * We either have to specify by name or by subsystems. (So all
979 * empty hierarchies must have a name).
981 if (!opts
->subsys_mask
&& !opts
->name
)
985 * Option noprefix was introduced just for backward compatibility
986 * with the old cpuset, so we allow noprefix only if mounting just
987 * the cpuset subsystem.
989 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
992 /* Can't specify "none" and some subsystems */
993 if (opts
->subsys_mask
&& opts
->none
)
999 static int cgroup1_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1002 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1003 struct cgroup_sb_opts opts
;
1004 u16 added_mask
, removed_mask
;
1006 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1008 /* See what subsystems are wanted */
1009 ret
= parse_cgroupfs_options(data
, &opts
);
1013 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1014 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1015 task_tgid_nr(current
), current
->comm
);
1017 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1018 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1020 /* Don't allow flags or name to change at remount */
1021 if ((opts
.flags
^ root
->flags
) ||
1022 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1023 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1024 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1029 /* remounting is not allowed for populated hierarchies */
1030 if (!list_empty(&root
->cgrp
.self
.children
)) {
1035 ret
= rebind_subsystems(root
, added_mask
);
1039 WARN_ON(rebind_subsystems(&cgrp_dfl_root
, removed_mask
));
1041 if (opts
.release_agent
) {
1042 spin_lock(&release_agent_path_lock
);
1043 strcpy(root
->release_agent_path
, opts
.release_agent
);
1044 spin_unlock(&release_agent_path_lock
);
1047 trace_cgroup_remount(root
);
1050 kfree(opts
.release_agent
);
1052 mutex_unlock(&cgroup_mutex
);
1056 struct kernfs_syscall_ops cgroup1_kf_syscall_ops
= {
1057 .rename
= cgroup1_rename
,
1058 .show_options
= cgroup1_show_options
,
1059 .remount_fs
= cgroup1_remount
,
1060 .mkdir
= cgroup_mkdir
,
1061 .rmdir
= cgroup_rmdir
,
1062 .show_path
= cgroup_show_path
,
1065 struct dentry
*cgroup1_mount(struct file_system_type
*fs_type
, int flags
,
1066 void *data
, unsigned long magic
,
1067 struct cgroup_namespace
*ns
)
1069 struct super_block
*pinned_sb
= NULL
;
1070 struct cgroup_sb_opts opts
;
1071 struct cgroup_root
*root
;
1072 struct cgroup_subsys
*ss
;
1073 struct dentry
*dentry
;
1075 bool new_root
= false;
1077 cgroup_lock_and_drain_offline(&cgrp_dfl_root
.cgrp
);
1079 /* First find the desired set of subsystems */
1080 ret
= parse_cgroupfs_options(data
, &opts
);
1085 * Destruction of cgroup root is asynchronous, so subsystems may
1086 * still be dying after the previous unmount. Let's drain the
1087 * dying subsystems. We just need to ensure that the ones
1088 * unmounted previously finish dying and don't care about new ones
1089 * starting. Testing ref liveliness is good enough.
1091 for_each_subsys(ss
, i
) {
1092 if (!(opts
.subsys_mask
& (1 << i
)) ||
1093 ss
->root
== &cgrp_dfl_root
)
1096 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
1097 mutex_unlock(&cgroup_mutex
);
1099 ret
= restart_syscall();
1102 cgroup_put(&ss
->root
->cgrp
);
1105 for_each_root(root
) {
1106 bool name_match
= false;
1108 if (root
== &cgrp_dfl_root
)
1112 * If we asked for a name then it must match. Also, if
1113 * name matches but sybsys_mask doesn't, we should fail.
1114 * Remember whether name matched.
1117 if (strcmp(opts
.name
, root
->name
))
1123 * If we asked for subsystems (or explicitly for no
1124 * subsystems) then they must match.
1126 if ((opts
.subsys_mask
|| opts
.none
) &&
1127 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1134 if (root
->flags
^ opts
.flags
)
1135 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1138 * We want to reuse @root whose lifetime is governed by its
1139 * ->cgrp. Let's check whether @root is alive and keep it
1140 * that way. As cgroup_kill_sb() can happen anytime, we
1141 * want to block it by pinning the sb so that @root doesn't
1142 * get killed before mount is complete.
1144 * With the sb pinned, tryget_live can reliably indicate
1145 * whether @root can be reused. If it's being killed,
1146 * drain it. We can use wait_queue for the wait but this
1147 * path is super cold. Let's just sleep a bit and retry.
1149 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
1150 if (IS_ERR(pinned_sb
) ||
1151 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1152 mutex_unlock(&cgroup_mutex
);
1153 if (!IS_ERR_OR_NULL(pinned_sb
))
1154 deactivate_super(pinned_sb
);
1156 ret
= restart_syscall();
1165 * No such thing, create a new one. name= matching without subsys
1166 * specification is allowed for already existing hierarchies but we
1167 * can't create new one without subsys specification.
1169 if (!opts
.subsys_mask
&& !opts
.none
) {
1174 /* Hierarchies may only be created in the initial cgroup namespace. */
1175 if (ns
!= &init_cgroup_ns
) {
1180 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1187 init_cgroup_root(root
, &opts
);
1189 ret
= cgroup_setup_root(root
, opts
.subsys_mask
, PERCPU_REF_INIT_DEAD
);
1191 cgroup_free_root(root
);
1194 mutex_unlock(&cgroup_mutex
);
1196 kfree(opts
.release_agent
);
1200 return ERR_PTR(ret
);
1202 dentry
= cgroup_do_mount(&cgroup_fs_type
, flags
, root
,
1203 CGROUP_SUPER_MAGIC
, ns
);
1206 * There's a race window after we release cgroup_mutex and before
1207 * allocating a superblock. Make sure a concurrent process won't
1208 * be able to re-use the root during this window by delaying the
1209 * initialization of root refcnt.
1212 mutex_lock(&cgroup_mutex
);
1213 percpu_ref_reinit(&root
->cgrp
.self
.refcnt
);
1214 mutex_unlock(&cgroup_mutex
);
1218 * If @pinned_sb, we're reusing an existing root and holding an
1219 * extra ref on its sb. Mount is complete. Put the extra ref.
1222 deactivate_super(pinned_sb
);
1227 static int __init
cgroup1_wq_init(void)
1230 * Used to destroy pidlists and separate to serve as flush domain.
1231 * Cap @max_active to 1 too.
1233 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
1235 BUG_ON(!cgroup_pidlist_destroy_wq
);
1238 core_initcall(cgroup1_wq_init
);
1240 static int __init
cgroup_no_v1(char *str
)
1242 struct cgroup_subsys
*ss
;
1246 while ((token
= strsep(&str
, ",")) != NULL
) {
1250 if (!strcmp(token
, "all")) {
1251 cgroup_no_v1_mask
= U16_MAX
;
1255 for_each_subsys(ss
, i
) {
1256 if (strcmp(token
, ss
->name
) &&
1257 strcmp(token
, ss
->legacy_name
))
1260 cgroup_no_v1_mask
|= 1 << i
;
1265 __setup("cgroup_no_v1=", cgroup_no_v1
);
1268 #ifdef CONFIG_CGROUP_DEBUG
1269 static struct cgroup_subsys_state
*
1270 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
1272 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
1275 return ERR_PTR(-ENOMEM
);
1280 static void debug_css_free(struct cgroup_subsys_state
*css
)
1285 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
1288 return cgroup_task_count(css
->cgroup
);
1291 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
1294 return (u64
)(unsigned long)current
->cgroups
;
1297 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
1303 count
= refcount_read(&task_css_set(current
)->refcount
);
1308 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
1310 struct cgrp_cset_link
*link
;
1311 struct css_set
*cset
;
1314 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
1318 spin_lock_irq(&css_set_lock
);
1320 cset
= rcu_dereference(current
->cgroups
);
1321 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1322 struct cgroup
*c
= link
->cgrp
;
1324 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
1325 seq_printf(seq
, "Root %d group %s\n",
1326 c
->root
->hierarchy_id
, name_buf
);
1329 spin_unlock_irq(&css_set_lock
);
1334 #define MAX_TASKS_SHOWN_PER_CSS 25
1335 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
1337 struct cgroup_subsys_state
*css
= seq_css(seq
);
1338 struct cgrp_cset_link
*link
;
1340 spin_lock_irq(&css_set_lock
);
1341 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
1342 struct css_set
*cset
= link
->cset
;
1343 struct task_struct
*task
;
1346 seq_printf(seq
, "css_set %pK\n", cset
);
1348 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
1349 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
1351 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
1354 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
1355 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
1357 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
1361 seq_puts(seq
, " ...\n");
1363 spin_unlock_irq(&css_set_lock
);
1367 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
1369 return (!cgroup_is_populated(css
->cgroup
) &&
1370 !css_has_online_children(&css
->cgroup
->self
));
1373 static struct cftype debug_files
[] = {
1375 .name
= "taskcount",
1376 .read_u64
= debug_taskcount_read
,
1380 .name
= "current_css_set",
1381 .read_u64
= current_css_set_read
,
1385 .name
= "current_css_set_refcount",
1386 .read_u64
= current_css_set_refcount_read
,
1390 .name
= "current_css_set_cg_links",
1391 .seq_show
= current_css_set_cg_links_read
,
1395 .name
= "cgroup_css_links",
1396 .seq_show
= cgroup_css_links_read
,
1400 .name
= "releasable",
1401 .read_u64
= releasable_read
,
1407 struct cgroup_subsys debug_cgrp_subsys
= {
1408 .css_alloc
= debug_css_alloc
,
1409 .css_free
= debug_css_free
,
1410 .legacy_cftypes
= debug_files
,
1412 #endif /* CONFIG_CGROUP_DEBUG */