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 * The following two functions "fix" the issue where there are more pids
198 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
199 * TODO: replace with a kernel-wide solution to this problem
201 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
202 static void *pidlist_allocate(int count)
204 if (PIDLIST_TOO_LARGE(count))
205 return vmalloc(array_size(count, sizeof(pid_t)));
207 return kmalloc_array(count, sizeof(pid_t), GFP_KERNEL);
210 static void pidlist_free(void *p)
216 * Used to destroy all pidlists lingering waiting for destroy timer. None
217 * should be left afterwards.
219 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
221 struct cgroup_pidlist *l, *tmp_l;
223 mutex_lock(&cgrp->pidlist_mutex);
224 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
225 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
226 mutex_unlock(&cgrp->pidlist_mutex);
228 flush_workqueue(cgroup_pidlist_destroy_wq);
229 BUG_ON(!list_empty(&cgrp->pidlists));
232 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
234 struct delayed_work *dwork = to_delayed_work(work);
235 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
237 struct cgroup_pidlist *tofree = NULL;
239 mutex_lock(&l->owner->pidlist_mutex);
242 * Destroy iff we didn't get queued again. The state won't change
243 * as destroy_dwork can only be queued while locked.
245 if (!delayed_work_pending(dwork)) {
247 pidlist_free(l->list);
248 put_pid_ns(l->key.ns);
252 mutex_unlock(&l->owner->pidlist_mutex);
257 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
258 * Returns the number of unique elements.
260 static int pidlist_uniq(pid_t *list, int length)
265 * we presume the 0th element is unique, so i starts at 1. trivial
266 * edge cases first; no work needs to be done for either
268 if (length == 0 || length == 1)
270 /* src and dest walk down the list; dest counts unique elements */
271 for (src = 1; src < length; src++) {
272 /* find next unique element */
273 while (list[src] == list[src-1]) {
278 /* dest always points to where the next unique element goes */
279 list[dest] = list[src];
287 * The two pid files - task and cgroup.procs - guaranteed that the result
288 * is sorted, which forced this whole pidlist fiasco. As pid order is
289 * different per namespace, each namespace needs differently sorted list,
290 * making it impossible to use, for example, single rbtree of member tasks
291 * sorted by task pointer. As pidlists can be fairly large, allocating one
292 * per open file is dangerous, so cgroup had to implement shared pool of
293 * pidlists keyed by cgroup and namespace.
295 static int cmppid(const void *a, const void *b)
297 return *(pid_t *)a - *(pid_t *)b;
300 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
301 enum cgroup_filetype type)
303 struct cgroup_pidlist *l;
304 /* don't need task_nsproxy() if we're looking at ourself */
305 struct pid_namespace *ns = task_active_pid_ns(current);
307 lockdep_assert_held(&cgrp->pidlist_mutex);
309 list_for_each_entry(l, &cgrp->pidlists, links)
310 if (l->key.type == type && l->key.ns == ns)
316 * find the appropriate pidlist for our purpose (given procs vs tasks)
317 * returns with the lock on that pidlist already held, and takes care
318 * of the use count, or returns NULL with no locks held if we're out of
321 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
322 enum cgroup_filetype type)
324 struct cgroup_pidlist *l;
326 lockdep_assert_held(&cgrp->pidlist_mutex);
328 l = cgroup_pidlist_find(cgrp, type);
332 /* entry not found; create a new one */
333 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
337 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
339 /* don't need task_nsproxy() if we're looking at ourself */
340 l->key.ns = get_pid_ns(task_active_pid_ns(current));
342 list_add(&l->links, &cgrp->pidlists);
347 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
349 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
350 struct cgroup_pidlist **lp)
354 int pid, n = 0; /* used for populating the array */
355 struct css_task_iter it;
356 struct task_struct *tsk;
357 struct cgroup_pidlist *l;
359 lockdep_assert_held(&cgrp->pidlist_mutex);
362 * If cgroup gets more users after we read count, we won't have
363 * enough space - tough. This race is indistinguishable to the
364 * caller from the case that the additional cgroup users didn't
365 * show up until sometime later on.
367 length = cgroup_task_count(cgrp);
368 array = pidlist_allocate(length);
371 /* now, populate the array */
372 css_task_iter_start(&cgrp->self, 0, &it);
373 while ((tsk = css_task_iter_next(&it))) {
374 if (unlikely(n == length))
376 /* get tgid or pid for procs or tasks file respectively */
377 if (type == CGROUP_FILE_PROCS)
378 pid = task_tgid_vnr(tsk);
380 pid = task_pid_vnr(tsk);
381 if (pid > 0) /* make sure to only use valid results */
384 css_task_iter_end(&it);
386 /* now sort & (if procs) strip out duplicates */
387 sort(array, length, sizeof(pid_t), cmppid, NULL);
388 if (type == CGROUP_FILE_PROCS)
389 length = pidlist_uniq(array, length);
391 l = cgroup_pidlist_find_create(cgrp, type);
397 /* store array, freeing old if necessary */
398 pidlist_free(l->list);
406 * seq_file methods for the tasks/procs files. The seq_file position is the
407 * next pid to display; the seq_file iterator is a pointer to the pid
408 * in the cgroup->l->list array.
411 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
414 * Initially we receive a position value that corresponds to
415 * one more than the last pid shown (or 0 on the first call or
416 * after a seek to the start). Use a binary-search to find the
417 * next pid to display, if any
419 struct kernfs_open_file *of = s->private;
420 struct cgroup *cgrp = seq_css(s)->cgroup;
421 struct cgroup_pidlist *l;
422 enum cgroup_filetype type = seq_cft(s)->private;
423 int index = 0, pid = *pos;
426 mutex_lock(&cgrp->pidlist_mutex);
429 * !NULL @of->priv indicates that this isn't the first start()
430 * after open. If the matching pidlist is around, we can use that.
431 * Look for it. Note that @of->priv can't be used directly. It
432 * could already have been destroyed.
435 of->priv = cgroup_pidlist_find(cgrp, type);
438 * Either this is the first start() after open or the matching
439 * pidlist has been destroyed inbetween. Create a new one.
442 ret = pidlist_array_load(cgrp, type,
443 (struct cgroup_pidlist **)&of->priv);
452 while (index < end) {
453 int mid = (index + end) / 2;
454 if (l->list[mid] == pid) {
457 } else if (l->list[mid] <= pid)
463 /* If we're off the end of the array, we're done */
464 if (index >= l->length)
466 /* Update the abstract position to be the actual pid that we found */
467 iter = l->list + index;
472 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
474 struct kernfs_open_file *of = s->private;
475 struct cgroup_pidlist *l = of->priv;
478 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
479 CGROUP_PIDLIST_DESTROY_DELAY);
480 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
483 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
485 struct kernfs_open_file *of = s->private;
486 struct cgroup_pidlist *l = of->priv;
488 pid_t *end = l->list + l->length;
490 * Advance to the next pid in the array. If this goes off the
502 static int cgroup_pidlist_show(struct seq_file *s, void *v)
504 seq_printf(s, "%d\n", *(int *)v);
509 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
510 char *buf, size_t nbytes, loff_t off,
514 struct task_struct *task;
515 const struct cred *cred, *tcred;
518 cgrp = cgroup_kn_lock_live(of->kn, false);
522 task = cgroup_procs_write_start(buf, threadgroup);
523 ret = PTR_ERR_OR_ZERO(task);
528 * Even if we're attaching all tasks in the thread group, we only
529 * need to check permissions on one of them.
531 cred = current_cred();
532 tcred = get_task_cred(task);
533 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
534 !uid_eq(cred->euid, tcred->uid) &&
535 !uid_eq(cred->euid, tcred->suid))
541 ret = cgroup_attach_task(cgrp, task, threadgroup);
544 cgroup_procs_write_finish(task);
546 cgroup_kn_unlock(of->kn);
548 return ret ?: nbytes;
551 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
552 char *buf, size_t nbytes, loff_t off)
554 return __cgroup1_procs_write(of, buf, nbytes, off, true);
557 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
558 char *buf, size_t nbytes, loff_t off)
560 return __cgroup1_procs_write(of, buf, nbytes, off, false);
563 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
564 char *buf, size_t nbytes, loff_t off)
568 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
570 cgrp = cgroup_kn_lock_live(of->kn, false);
573 spin_lock(&release_agent_path_lock);
574 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
575 sizeof(cgrp->root->release_agent_path));
576 spin_unlock(&release_agent_path_lock);
577 cgroup_kn_unlock(of->kn);
581 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
583 struct cgroup *cgrp = seq_css(seq)->cgroup;
585 spin_lock(&release_agent_path_lock);
586 seq_puts(seq, cgrp->root->release_agent_path);
587 spin_unlock(&release_agent_path_lock);
592 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
594 seq_puts(seq, "0\n");
598 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
601 return notify_on_release(css->cgroup);
604 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
605 struct cftype *cft, u64 val)
608 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
610 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
614 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
617 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
620 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
621 struct cftype *cft, u64 val)
624 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
626 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
630 /* cgroup core interface files for the legacy hierarchies */
631 struct cftype cgroup1_base_files[] = {
633 .name = "cgroup.procs",
634 .seq_start = cgroup_pidlist_start,
635 .seq_next = cgroup_pidlist_next,
636 .seq_stop = cgroup_pidlist_stop,
637 .seq_show = cgroup_pidlist_show,
638 .private = CGROUP_FILE_PROCS,
639 .write = cgroup1_procs_write,
642 .name = "cgroup.clone_children",
643 .read_u64 = cgroup_clone_children_read,
644 .write_u64 = cgroup_clone_children_write,
647 .name = "cgroup.sane_behavior",
648 .flags = CFTYPE_ONLY_ON_ROOT,
649 .seq_show = cgroup_sane_behavior_show,
653 .seq_start = cgroup_pidlist_start,
654 .seq_next = cgroup_pidlist_next,
655 .seq_stop = cgroup_pidlist_stop,
656 .seq_show = cgroup_pidlist_show,
657 .private = CGROUP_FILE_TASKS,
658 .write = cgroup1_tasks_write,
661 .name = "notify_on_release",
662 .read_u64 = cgroup_read_notify_on_release,
663 .write_u64 = cgroup_write_notify_on_release,
666 .name = "release_agent",
667 .flags = CFTYPE_ONLY_ON_ROOT,
668 .seq_show = cgroup_release_agent_show,
669 .write = cgroup_release_agent_write,
670 .max_write_len = PATH_MAX - 1,
675 /* Display information about each subsystem and each hierarchy */
676 int proc_cgroupstats_show(struct seq_file *m, void *v)
678 struct cgroup_subsys *ss;
681 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
683 * ideally we don't want subsystems moving around while we do this.
684 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
685 * subsys/hierarchy state.
687 mutex_lock(&cgroup_mutex);
689 for_each_subsys(ss, i)
690 seq_printf(m, "%s\t%d\t%d\t%d\n",
691 ss->legacy_name, ss->root->hierarchy_id,
692 atomic_read(&ss->root->nr_cgrps),
693 cgroup_ssid_enabled(i));
695 mutex_unlock(&cgroup_mutex);
700 * cgroupstats_build - build and fill cgroupstats
701 * @stats: cgroupstats to fill information into
702 * @dentry: A dentry entry belonging to the cgroup for which stats have
705 * Build and fill cgroupstats so that taskstats can export it to user
708 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
710 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
712 struct css_task_iter it;
713 struct task_struct *tsk;
715 /* it should be kernfs_node belonging to cgroupfs and is a directory */
716 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
717 kernfs_type(kn) != KERNFS_DIR)
720 mutex_lock(&cgroup_mutex);
723 * We aren't being called from kernfs and there's no guarantee on
724 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
725 * @kn->priv is RCU safe. Let's do the RCU dancing.
728 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
729 if (!cgrp || cgroup_is_dead(cgrp)) {
731 mutex_unlock(&cgroup_mutex);
736 css_task_iter_start(&cgrp->self, 0, &it);
737 while ((tsk = css_task_iter_next(&it))) {
738 switch (tsk->state) {
742 case TASK_INTERRUPTIBLE:
743 stats->nr_sleeping++;
745 case TASK_UNINTERRUPTIBLE:
746 stats->nr_uninterruptible++;
752 if (delayacct_is_task_waiting_on_io(tsk))
757 css_task_iter_end(&it);
759 mutex_unlock(&cgroup_mutex);
763 void cgroup1_check_for_release(struct cgroup *cgrp)
765 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
766 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
767 schedule_work(&cgrp->release_agent_work);
771 * Notify userspace when a cgroup is released, by running the
772 * configured release agent with the name of the cgroup (path
773 * relative to the root of cgroup file system) as the argument.
775 * Most likely, this user command will try to rmdir this cgroup.
777 * This races with the possibility that some other task will be
778 * attached to this cgroup before it is removed, or that some other
779 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
780 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
781 * unused, and this cgroup will be reprieved from its death sentence,
782 * to continue to serve a useful existence. Next time it's released,
783 * we will get notified again, if it still has 'notify_on_release' set.
785 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
786 * means only wait until the task is successfully execve()'d. The
787 * separate release agent task is forked by call_usermodehelper(),
788 * then control in this thread returns here, without waiting for the
789 * release agent task. We don't bother to wait because the caller of
790 * this routine has no use for the exit status of the release agent
791 * task, so no sense holding our caller up for that.
793 void cgroup1_release_agent(struct work_struct *work)
795 struct cgroup *cgrp =
796 container_of(work, struct cgroup, release_agent_work);
797 char *pathbuf = NULL, *agentbuf = NULL;
798 char *argv[3], *envp[3];
801 mutex_lock(&cgroup_mutex);
803 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
804 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
805 if (!pathbuf || !agentbuf)
808 spin_lock_irq(&css_set_lock);
809 ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
810 spin_unlock_irq(&css_set_lock);
811 if (ret < 0 || ret >= PATH_MAX)
818 /* minimal command environment */
820 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
823 mutex_unlock(&cgroup_mutex);
824 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
827 mutex_unlock(&cgroup_mutex);
834 * cgroup_rename - Only allow simple rename of directories in place.
836 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
837 const char *new_name_str)
839 struct cgroup *cgrp = kn->priv;
842 if (kernfs_type(kn) != KERNFS_DIR)
844 if (kn->parent != new_parent)
848 * We're gonna grab cgroup_mutex which nests outside kernfs
849 * active_ref. kernfs_rename() doesn't require active_ref
850 * protection. Break them before grabbing cgroup_mutex.
852 kernfs_break_active_protection(new_parent);
853 kernfs_break_active_protection(kn);
855 mutex_lock(&cgroup_mutex);
857 ret = kernfs_rename(kn, new_parent, new_name_str);
859 TRACE_CGROUP_PATH(rename, cgrp);
861 mutex_unlock(&cgroup_mutex);
863 kernfs_unbreak_active_protection(kn);
864 kernfs_unbreak_active_protection(new_parent);
868 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
870 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
871 struct cgroup_subsys *ss;
874 for_each_subsys(ss, ssid)
875 if (root->subsys_mask & (1 << ssid))
876 seq_show_option(seq, ss->legacy_name, NULL);
877 if (root->flags & CGRP_ROOT_NOPREFIX)
878 seq_puts(seq, ",noprefix");
879 if (root->flags & CGRP_ROOT_XATTR)
880 seq_puts(seq, ",xattr");
881 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
882 seq_puts(seq, ",cpuset_v2_mode");
884 spin_lock(&release_agent_path_lock);
885 if (strlen(root->release_agent_path))
886 seq_show_option(seq, "release_agent",
887 root->release_agent_path);
888 spin_unlock(&release_agent_path_lock);
890 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
891 seq_puts(seq, ",clone_children");
892 if (strlen(root->name))
893 seq_show_option(seq, "name", root->name);
908 static const struct fs_parameter_spec cgroup1_param_specs[] = {
909 fsparam_flag ("all", Opt_all),
910 fsparam_flag ("clone_children", Opt_clone_children),
911 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
912 fsparam_string("name", Opt_name),
913 fsparam_flag ("none", Opt_none),
914 fsparam_flag ("noprefix", Opt_noprefix),
915 fsparam_string("release_agent", Opt_release_agent),
916 fsparam_flag ("xattr", Opt_xattr),
920 const struct fs_parameter_description cgroup1_fs_parameters = {
922 .specs = cgroup1_param_specs,
925 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
927 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
928 struct cgroup_subsys *ss;
929 struct fs_parse_result result;
932 opt = fs_parse(fc, &cgroup1_fs_parameters, param, &result);
933 if (opt == -ENOPARAM) {
934 if (strcmp(param->key, "source") == 0) {
935 fc->source = param->string;
936 param->string = NULL;
939 for_each_subsys(ss, i) {
940 if (strcmp(param->key, ss->legacy_name))
942 ctx->subsys_mask |= (1 << i);
945 return cg_invalf(fc, "cgroup1: Unknown subsys name '%s'", param->key);
952 /* Explicitly have no subsystems */
959 ctx->flags |= CGRP_ROOT_NOPREFIX;
961 case Opt_clone_children:
962 ctx->cpuset_clone_children = true;
964 case Opt_cpuset_v2_mode:
965 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
968 ctx->flags |= CGRP_ROOT_XATTR;
970 case Opt_release_agent:
971 /* Specifying two release agents is forbidden */
972 if (ctx->release_agent)
973 return cg_invalf(fc, "cgroup1: release_agent respecified");
974 ctx->release_agent = param->string;
975 param->string = NULL;
978 /* blocked by boot param? */
979 if (cgroup_no_v1_named)
981 /* Can't specify an empty name */
983 return cg_invalf(fc, "cgroup1: Empty name");
984 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
985 return cg_invalf(fc, "cgroup1: Name too long");
986 /* Must match [\w.-]+ */
987 for (i = 0; i < param->size; i++) {
988 char c = param->string[i];
991 if ((c == '.') || (c == '-') || (c == '_'))
993 return cg_invalf(fc, "cgroup1: Invalid name");
995 /* Specifying two names is forbidden */
997 return cg_invalf(fc, "cgroup1: name respecified");
998 ctx->name = param->string;
999 param->string = NULL;
1005 static int check_cgroupfs_options(struct fs_context *fc)
1007 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1010 struct cgroup_subsys *ss;
1013 #ifdef CONFIG_CPUSETS
1014 mask = ~((u16)1 << cpuset_cgrp_id);
1016 for_each_subsys(ss, i)
1017 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1020 ctx->subsys_mask &= enabled;
1023 * In absense of 'none', 'name=' or subsystem name options,
1024 * let's default to 'all'.
1026 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1030 /* Mutually exclusive option 'all' + subsystem name */
1031 if (ctx->subsys_mask)
1032 return cg_invalf(fc, "cgroup1: subsys name conflicts with all");
1033 /* 'all' => select all the subsystems */
1034 ctx->subsys_mask = enabled;
1038 * We either have to specify by name or by subsystems. (So all
1039 * empty hierarchies must have a name).
1041 if (!ctx->subsys_mask && !ctx->name)
1042 return cg_invalf(fc, "cgroup1: Need name or subsystem set");
1045 * Option noprefix was introduced just for backward compatibility
1046 * with the old cpuset, so we allow noprefix only if mounting just
1047 * the cpuset subsystem.
1049 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1050 return cg_invalf(fc, "cgroup1: noprefix used incorrectly");
1052 /* Can't specify "none" and some subsystems */
1053 if (ctx->subsys_mask && ctx->none)
1054 return cg_invalf(fc, "cgroup1: none used incorrectly");
1059 int cgroup1_reconfigure(struct fs_context *fc)
1061 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1062 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1063 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1065 u16 added_mask, removed_mask;
1067 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1069 /* See what subsystems are wanted */
1070 ret = check_cgroupfs_options(fc);
1074 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1075 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1076 task_tgid_nr(current), current->comm);
1078 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1079 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1081 /* Don't allow flags or name to change at remount */
1082 if ((ctx->flags ^ root->flags) ||
1083 (ctx->name && strcmp(ctx->name, root->name))) {
1084 cg_invalf(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1085 ctx->flags, ctx->name ?: "", root->flags, root->name);
1090 /* remounting is not allowed for populated hierarchies */
1091 if (!list_empty(&root->cgrp.self.children)) {
1096 ret = rebind_subsystems(root, added_mask);
1100 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1102 if (ctx->release_agent) {
1103 spin_lock(&release_agent_path_lock);
1104 strcpy(root->release_agent_path, ctx->release_agent);
1105 spin_unlock(&release_agent_path_lock);
1108 trace_cgroup_remount(root);
1111 mutex_unlock(&cgroup_mutex);
1115 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1116 .rename = cgroup1_rename,
1117 .show_options = cgroup1_show_options,
1118 .mkdir = cgroup_mkdir,
1119 .rmdir = cgroup_rmdir,
1120 .show_path = cgroup_show_path,
1124 * The guts of cgroup1 mount - find or create cgroup_root to use.
1125 * Called with cgroup_mutex held; returns 0 on success, -E... on
1126 * error and positive - in case when the candidate is busy dying.
1127 * On success it stashes a reference to cgroup_root into given
1128 * cgroup_fs_context; that reference is *NOT* counting towards the
1129 * cgroup_root refcount.
1131 static int cgroup1_root_to_use(struct fs_context *fc)
1133 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1134 struct cgroup_root *root;
1135 struct cgroup_subsys *ss;
1138 /* First find the desired set of subsystems */
1139 ret = check_cgroupfs_options(fc);
1144 * Destruction of cgroup root is asynchronous, so subsystems may
1145 * still be dying after the previous unmount. Let's drain the
1146 * dying subsystems. We just need to ensure that the ones
1147 * unmounted previously finish dying and don't care about new ones
1148 * starting. Testing ref liveliness is good enough.
1150 for_each_subsys(ss, i) {
1151 if (!(ctx->subsys_mask & (1 << i)) ||
1152 ss->root == &cgrp_dfl_root)
1155 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1156 return 1; /* restart */
1157 cgroup_put(&ss->root->cgrp);
1160 for_each_root(root) {
1161 bool name_match = false;
1163 if (root == &cgrp_dfl_root)
1167 * If we asked for a name then it must match. Also, if
1168 * name matches but sybsys_mask doesn't, we should fail.
1169 * Remember whether name matched.
1172 if (strcmp(ctx->name, root->name))
1178 * If we asked for subsystems (or explicitly for no
1179 * subsystems) then they must match.
1181 if ((ctx->subsys_mask || ctx->none) &&
1182 (ctx->subsys_mask != root->subsys_mask)) {
1188 if (root->flags ^ ctx->flags)
1189 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1196 * No such thing, create a new one. name= matching without subsys
1197 * specification is allowed for already existing hierarchies but we
1198 * can't create new one without subsys specification.
1200 if (!ctx->subsys_mask && !ctx->none)
1201 return cg_invalf(fc, "cgroup1: No subsys list or none specified");
1203 /* Hierarchies may only be created in the initial cgroup namespace. */
1204 if (ctx->ns != &init_cgroup_ns)
1207 root = kzalloc(sizeof(*root), GFP_KERNEL);
1212 init_cgroup_root(ctx);
1214 ret = cgroup_setup_root(root, ctx->subsys_mask);
1216 cgroup_free_root(root);
1220 int cgroup1_get_tree(struct fs_context *fc)
1222 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1225 /* Check if the caller has permission to mount. */
1226 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1229 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1231 ret = cgroup1_root_to_use(fc);
1232 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1233 ret = 1; /* restart */
1235 mutex_unlock(&cgroup_mutex);
1238 ret = cgroup_do_get_tree(fc);
1240 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1241 struct super_block *sb = fc->root->d_sb;
1243 deactivate_locked_super(sb);
1247 if (unlikely(ret > 0)) {
1249 return restart_syscall();
1254 static int __init cgroup1_wq_init(void)
1257 * Used to destroy pidlists and separate to serve as flush domain.
1258 * Cap @max_active to 1 too.
1260 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1262 BUG_ON(!cgroup_pidlist_destroy_wq);
1265 core_initcall(cgroup1_wq_init);
1267 static int __init cgroup_no_v1(char *str)
1269 struct cgroup_subsys *ss;
1273 while ((token = strsep(&str, ",")) != NULL) {
1277 if (!strcmp(token, "all")) {
1278 cgroup_no_v1_mask = U16_MAX;
1282 if (!strcmp(token, "named")) {
1283 cgroup_no_v1_named = true;
1287 for_each_subsys(ss, i) {
1288 if (strcmp(token, ss->name) &&
1289 strcmp(token, ss->legacy_name))
1292 cgroup_no_v1_mask |= 1 << i;
1297 __setup("cgroup_no_v1=", cgroup_no_v1);