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>
22 * pidlists linger the following amount before being destroyed. The goal
23 * is avoiding frequent destruction in the middle of consecutive read calls
24 * Expiring in the middle is a performance problem not a correctness one.
25 * 1 sec should be enough.
27 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
29 /* Controllers blocked by the commandline in v1 */
30 static u16 cgroup_no_v1_mask;
32 /* disable named v1 mounts */
33 static bool cgroup_no_v1_named;
36 * pidlist destructions need to be flushed on cgroup destruction. Use a
37 * separate workqueue as flush domain.
39 static struct workqueue_struct *cgroup_pidlist_destroy_wq;
41 /* protects cgroup_subsys->release_agent_path */
42 static DEFINE_SPINLOCK(release_agent_path_lock);
44 bool cgroup1_ssid_disabled(int ssid)
46 return cgroup_no_v1_mask & (1 << ssid);
50 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
51 * @from: attach to all cgroups of a given task
52 * @tsk: the task to be attached
54 * Return: %0 on success or a negative errno code on failure
56 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
58 struct cgroup_root *root;
61 mutex_lock(&cgroup_mutex);
62 percpu_down_write(&cgroup_threadgroup_rwsem);
64 struct cgroup *from_cgrp;
66 if (root == &cgrp_dfl_root)
69 spin_lock_irq(&css_set_lock);
70 from_cgrp = task_cgroup_from_root(from, root);
71 spin_unlock_irq(&css_set_lock);
73 retval = cgroup_attach_task(from_cgrp, tsk, false);
77 percpu_up_write(&cgroup_threadgroup_rwsem);
78 mutex_unlock(&cgroup_mutex);
82 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
85 * cgroup_transfer_tasks - move tasks from one cgroup to another
86 * @to: cgroup to which the tasks will be moved
87 * @from: cgroup in which the tasks currently reside
89 * Locking rules between cgroup_post_fork() and the migration path
90 * guarantee that, if a task is forking while being migrated, the new child
91 * is guaranteed to be either visible in the source cgroup after the
92 * parent's migration is complete or put into the target cgroup. No task
93 * can slip out of migration through forking.
95 * Return: %0 on success or a negative errno code on failure
97 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
99 DEFINE_CGROUP_MGCTX(mgctx);
100 struct cgrp_cset_link *link;
101 struct css_task_iter it;
102 struct task_struct *task;
105 if (cgroup_on_dfl(to))
108 ret = cgroup_migrate_vet_dst(to);
112 mutex_lock(&cgroup_mutex);
114 percpu_down_write(&cgroup_threadgroup_rwsem);
116 /* all tasks in @from are being moved, all csets are source */
117 spin_lock_irq(&css_set_lock);
118 list_for_each_entry(link, &from->cset_links, cset_link)
119 cgroup_migrate_add_src(link->cset, to, &mgctx);
120 spin_unlock_irq(&css_set_lock);
122 ret = cgroup_migrate_prepare_dst(&mgctx);
127 * Migrate tasks one-by-one until @from is empty. This fails iff
128 * ->can_attach() fails.
131 css_task_iter_start(&from->self, 0, &it);
134 task = css_task_iter_next(&it);
135 } while (task && (task->flags & PF_EXITING));
138 get_task_struct(task);
139 css_task_iter_end(&it);
142 ret = cgroup_migrate(task, false, &mgctx);
144 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
145 put_task_struct(task);
147 } while (task && !ret);
149 cgroup_migrate_finish(&mgctx);
150 percpu_up_write(&cgroup_threadgroup_rwsem);
151 mutex_unlock(&cgroup_mutex);
156 * Stuff for reading the 'tasks'/'procs' files.
158 * Reading this file can return large amounts of data if a cgroup has
159 * *lots* of attached tasks. So it may need several calls to read(),
160 * but we cannot guarantee that the information we produce is correct
161 * unless we produce it entirely atomically.
165 /* which pidlist file are we talking about? */
166 enum cgroup_filetype {
172 * A pidlist is a list of pids that virtually represents the contents of one
173 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
174 * a pair (one each for procs, tasks) for each pid namespace that's relevant
177 struct cgroup_pidlist {
179 * used to find which pidlist is wanted. doesn't change as long as
180 * this particular list stays in the list.
182 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
185 /* how many elements the above list has */
187 /* each of these stored in a list by its cgroup */
188 struct list_head links;
189 /* pointer to the cgroup we belong to, for list removal purposes */
190 struct cgroup *owner;
191 /* for delayed destruction */
192 struct delayed_work destroy_dwork;
196 * Used to destroy all pidlists lingering waiting for destroy timer. None
197 * should be left afterwards.
199 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
201 struct cgroup_pidlist *l, *tmp_l;
203 mutex_lock(&cgrp->pidlist_mutex);
204 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
205 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
206 mutex_unlock(&cgrp->pidlist_mutex);
208 flush_workqueue(cgroup_pidlist_destroy_wq);
209 BUG_ON(!list_empty(&cgrp->pidlists));
212 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
214 struct delayed_work *dwork = to_delayed_work(work);
215 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
217 struct cgroup_pidlist *tofree = NULL;
219 mutex_lock(&l->owner->pidlist_mutex);
222 * Destroy iff we didn't get queued again. The state won't change
223 * as destroy_dwork can only be queued while locked.
225 if (!delayed_work_pending(dwork)) {
228 put_pid_ns(l->key.ns);
232 mutex_unlock(&l->owner->pidlist_mutex);
237 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
238 * Returns the number of unique elements.
240 static int pidlist_uniq(pid_t *list, int length)
245 * we presume the 0th element is unique, so i starts at 1. trivial
246 * edge cases first; no work needs to be done for either
248 if (length == 0 || length == 1)
250 /* src and dest walk down the list; dest counts unique elements */
251 for (src = 1; src < length; src++) {
252 /* find next unique element */
253 while (list[src] == list[src-1]) {
258 /* dest always points to where the next unique element goes */
259 list[dest] = list[src];
267 * The two pid files - task and cgroup.procs - guaranteed that the result
268 * is sorted, which forced this whole pidlist fiasco. As pid order is
269 * different per namespace, each namespace needs differently sorted list,
270 * making it impossible to use, for example, single rbtree of member tasks
271 * sorted by task pointer. As pidlists can be fairly large, allocating one
272 * per open file is dangerous, so cgroup had to implement shared pool of
273 * pidlists keyed by cgroup and namespace.
275 static int cmppid(const void *a, const void *b)
277 return *(pid_t *)a - *(pid_t *)b;
280 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
281 enum cgroup_filetype type)
283 struct cgroup_pidlist *l;
284 /* don't need task_nsproxy() if we're looking at ourself */
285 struct pid_namespace *ns = task_active_pid_ns(current);
287 lockdep_assert_held(&cgrp->pidlist_mutex);
289 list_for_each_entry(l, &cgrp->pidlists, links)
290 if (l->key.type == type && l->key.ns == ns)
296 * find the appropriate pidlist for our purpose (given procs vs tasks)
297 * returns with the lock on that pidlist already held, and takes care
298 * of the use count, or returns NULL with no locks held if we're out of
301 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
302 enum cgroup_filetype type)
304 struct cgroup_pidlist *l;
306 lockdep_assert_held(&cgrp->pidlist_mutex);
308 l = cgroup_pidlist_find(cgrp, type);
312 /* entry not found; create a new one */
313 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
317 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
319 /* don't need task_nsproxy() if we're looking at ourself */
320 l->key.ns = get_pid_ns(task_active_pid_ns(current));
322 list_add(&l->links, &cgrp->pidlists);
327 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
329 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
330 struct cgroup_pidlist **lp)
334 int pid, n = 0; /* used for populating the array */
335 struct css_task_iter it;
336 struct task_struct *tsk;
337 struct cgroup_pidlist *l;
339 lockdep_assert_held(&cgrp->pidlist_mutex);
342 * If cgroup gets more users after we read count, we won't have
343 * enough space - tough. This race is indistinguishable to the
344 * caller from the case that the additional cgroup users didn't
345 * show up until sometime later on.
347 length = cgroup_task_count(cgrp);
348 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
351 /* now, populate the array */
352 css_task_iter_start(&cgrp->self, 0, &it);
353 while ((tsk = css_task_iter_next(&it))) {
354 if (unlikely(n == length))
356 /* get tgid or pid for procs or tasks file respectively */
357 if (type == CGROUP_FILE_PROCS)
358 pid = task_tgid_vnr(tsk);
360 pid = task_pid_vnr(tsk);
361 if (pid > 0) /* make sure to only use valid results */
364 css_task_iter_end(&it);
366 /* now sort & (if procs) strip out duplicates */
367 sort(array, length, sizeof(pid_t), cmppid, NULL);
368 if (type == CGROUP_FILE_PROCS)
369 length = pidlist_uniq(array, length);
371 l = cgroup_pidlist_find_create(cgrp, type);
377 /* store array, freeing old if necessary */
386 * seq_file methods for the tasks/procs files. The seq_file position is the
387 * next pid to display; the seq_file iterator is a pointer to the pid
388 * in the cgroup->l->list array.
391 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
394 * Initially we receive a position value that corresponds to
395 * one more than the last pid shown (or 0 on the first call or
396 * after a seek to the start). Use a binary-search to find the
397 * next pid to display, if any
399 struct kernfs_open_file *of = s->private;
400 struct cgroup *cgrp = seq_css(s)->cgroup;
401 struct cgroup_pidlist *l;
402 enum cgroup_filetype type = seq_cft(s)->private;
403 int index = 0, pid = *pos;
406 mutex_lock(&cgrp->pidlist_mutex);
409 * !NULL @of->priv indicates that this isn't the first start()
410 * after open. If the matching pidlist is around, we can use that.
411 * Look for it. Note that @of->priv can't be used directly. It
412 * could already have been destroyed.
415 of->priv = cgroup_pidlist_find(cgrp, type);
418 * Either this is the first start() after open or the matching
419 * pidlist has been destroyed inbetween. Create a new one.
422 ret = pidlist_array_load(cgrp, type,
423 (struct cgroup_pidlist **)&of->priv);
432 while (index < end) {
433 int mid = (index + end) / 2;
434 if (l->list[mid] == pid) {
437 } else if (l->list[mid] <= pid)
443 /* If we're off the end of the array, we're done */
444 if (index >= l->length)
446 /* Update the abstract position to be the actual pid that we found */
447 iter = l->list + index;
452 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
454 struct kernfs_open_file *of = s->private;
455 struct cgroup_pidlist *l = of->priv;
458 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
459 CGROUP_PIDLIST_DESTROY_DELAY);
460 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
463 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
465 struct kernfs_open_file *of = s->private;
466 struct cgroup_pidlist *l = of->priv;
468 pid_t *end = l->list + l->length;
470 * Advance to the next pid in the array. If this goes off the
483 static int cgroup_pidlist_show(struct seq_file *s, void *v)
485 seq_printf(s, "%d\n", *(int *)v);
490 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
491 char *buf, size_t nbytes, loff_t off,
495 struct task_struct *task;
496 const struct cred *cred, *tcred;
500 cgrp = cgroup_kn_lock_live(of->kn, false);
504 task = cgroup_procs_write_start(buf, threadgroup, &locked);
505 ret = PTR_ERR_OR_ZERO(task);
510 * Even if we're attaching all tasks in the thread group, we only
511 * need to check permissions on one of them.
513 cred = current_cred();
514 tcred = get_task_cred(task);
515 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
516 !uid_eq(cred->euid, tcred->uid) &&
517 !uid_eq(cred->euid, tcred->suid))
523 ret = cgroup_attach_task(cgrp, task, threadgroup);
526 cgroup_procs_write_finish(task, locked);
528 cgroup_kn_unlock(of->kn);
530 return ret ?: nbytes;
533 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
534 char *buf, size_t nbytes, loff_t off)
536 return __cgroup1_procs_write(of, buf, nbytes, off, true);
539 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
540 char *buf, size_t nbytes, loff_t off)
542 return __cgroup1_procs_write(of, buf, nbytes, off, false);
545 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
546 char *buf, size_t nbytes, loff_t off)
550 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
552 cgrp = cgroup_kn_lock_live(of->kn, false);
555 spin_lock(&release_agent_path_lock);
556 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
557 sizeof(cgrp->root->release_agent_path));
558 spin_unlock(&release_agent_path_lock);
559 cgroup_kn_unlock(of->kn);
563 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
565 struct cgroup *cgrp = seq_css(seq)->cgroup;
567 spin_lock(&release_agent_path_lock);
568 seq_puts(seq, cgrp->root->release_agent_path);
569 spin_unlock(&release_agent_path_lock);
574 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
576 seq_puts(seq, "0\n");
580 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
583 return notify_on_release(css->cgroup);
586 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
587 struct cftype *cft, u64 val)
590 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
592 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
596 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
599 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
602 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
603 struct cftype *cft, u64 val)
606 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
608 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
612 /* cgroup core interface files for the legacy hierarchies */
613 struct cftype cgroup1_base_files[] = {
615 .name = "cgroup.procs",
616 .seq_start = cgroup_pidlist_start,
617 .seq_next = cgroup_pidlist_next,
618 .seq_stop = cgroup_pidlist_stop,
619 .seq_show = cgroup_pidlist_show,
620 .private = CGROUP_FILE_PROCS,
621 .write = cgroup1_procs_write,
624 .name = "cgroup.clone_children",
625 .read_u64 = cgroup_clone_children_read,
626 .write_u64 = cgroup_clone_children_write,
629 .name = "cgroup.sane_behavior",
630 .flags = CFTYPE_ONLY_ON_ROOT,
631 .seq_show = cgroup_sane_behavior_show,
635 .seq_start = cgroup_pidlist_start,
636 .seq_next = cgroup_pidlist_next,
637 .seq_stop = cgroup_pidlist_stop,
638 .seq_show = cgroup_pidlist_show,
639 .private = CGROUP_FILE_TASKS,
640 .write = cgroup1_tasks_write,
643 .name = "notify_on_release",
644 .read_u64 = cgroup_read_notify_on_release,
645 .write_u64 = cgroup_write_notify_on_release,
648 .name = "release_agent",
649 .flags = CFTYPE_ONLY_ON_ROOT,
650 .seq_show = cgroup_release_agent_show,
651 .write = cgroup_release_agent_write,
652 .max_write_len = PATH_MAX - 1,
657 /* Display information about each subsystem and each hierarchy */
658 int proc_cgroupstats_show(struct seq_file *m, void *v)
660 struct cgroup_subsys *ss;
663 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
665 * ideally we don't want subsystems moving around while we do this.
666 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
667 * subsys/hierarchy state.
669 mutex_lock(&cgroup_mutex);
671 for_each_subsys(ss, i)
672 seq_printf(m, "%s\t%d\t%d\t%d\n",
673 ss->legacy_name, ss->root->hierarchy_id,
674 atomic_read(&ss->root->nr_cgrps),
675 cgroup_ssid_enabled(i));
677 mutex_unlock(&cgroup_mutex);
682 * cgroupstats_build - build and fill cgroupstats
683 * @stats: cgroupstats to fill information into
684 * @dentry: A dentry entry belonging to the cgroup for which stats have
687 * Build and fill cgroupstats so that taskstats can export it to user
690 * Return: %0 on success or a negative errno code on failure
692 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
694 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
696 struct css_task_iter it;
697 struct task_struct *tsk;
699 /* it should be kernfs_node belonging to cgroupfs and is a directory */
700 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
701 kernfs_type(kn) != KERNFS_DIR)
704 mutex_lock(&cgroup_mutex);
707 * We aren't being called from kernfs and there's no guarantee on
708 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
709 * @kn->priv is RCU safe. Let's do the RCU dancing.
712 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
713 if (!cgrp || cgroup_is_dead(cgrp)) {
715 mutex_unlock(&cgroup_mutex);
720 css_task_iter_start(&cgrp->self, 0, &it);
721 while ((tsk = css_task_iter_next(&it))) {
722 switch (READ_ONCE(tsk->__state)) {
726 case TASK_INTERRUPTIBLE:
727 stats->nr_sleeping++;
729 case TASK_UNINTERRUPTIBLE:
730 stats->nr_uninterruptible++;
741 css_task_iter_end(&it);
743 mutex_unlock(&cgroup_mutex);
747 void cgroup1_check_for_release(struct cgroup *cgrp)
749 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
750 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
751 schedule_work(&cgrp->release_agent_work);
755 * Notify userspace when a cgroup is released, by running the
756 * configured release agent with the name of the cgroup (path
757 * relative to the root of cgroup file system) as the argument.
759 * Most likely, this user command will try to rmdir this cgroup.
761 * This races with the possibility that some other task will be
762 * attached to this cgroup before it is removed, or that some other
763 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
764 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
765 * unused, and this cgroup will be reprieved from its death sentence,
766 * to continue to serve a useful existence. Next time it's released,
767 * we will get notified again, if it still has 'notify_on_release' set.
769 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
770 * means only wait until the task is successfully execve()'d. The
771 * separate release agent task is forked by call_usermodehelper(),
772 * then control in this thread returns here, without waiting for the
773 * release agent task. We don't bother to wait because the caller of
774 * this routine has no use for the exit status of the release agent
775 * task, so no sense holding our caller up for that.
777 void cgroup1_release_agent(struct work_struct *work)
779 struct cgroup *cgrp =
780 container_of(work, struct cgroup, release_agent_work);
781 char *pathbuf, *agentbuf;
782 char *argv[3], *envp[3];
785 /* snoop agent path and exit early if empty */
786 if (!cgrp->root->release_agent_path[0])
789 /* prepare argument buffers */
790 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
791 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
792 if (!pathbuf || !agentbuf)
795 spin_lock(&release_agent_path_lock);
796 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
797 spin_unlock(&release_agent_path_lock);
801 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
802 if (ret < 0 || ret >= PATH_MAX)
809 /* minimal command environment */
811 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
814 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
821 * cgroup_rename - Only allow simple rename of directories in place.
823 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
824 const char *new_name_str)
826 struct cgroup *cgrp = kn->priv;
829 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
830 if (strchr(new_name_str, '\n'))
833 if (kernfs_type(kn) != KERNFS_DIR)
835 if (kn->parent != new_parent)
839 * We're gonna grab cgroup_mutex which nests outside kernfs
840 * active_ref. kernfs_rename() doesn't require active_ref
841 * protection. Break them before grabbing cgroup_mutex.
843 kernfs_break_active_protection(new_parent);
844 kernfs_break_active_protection(kn);
846 mutex_lock(&cgroup_mutex);
848 ret = kernfs_rename(kn, new_parent, new_name_str);
850 TRACE_CGROUP_PATH(rename, cgrp);
852 mutex_unlock(&cgroup_mutex);
854 kernfs_unbreak_active_protection(kn);
855 kernfs_unbreak_active_protection(new_parent);
859 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
861 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
862 struct cgroup_subsys *ss;
865 for_each_subsys(ss, ssid)
866 if (root->subsys_mask & (1 << ssid))
867 seq_show_option(seq, ss->legacy_name, NULL);
868 if (root->flags & CGRP_ROOT_NOPREFIX)
869 seq_puts(seq, ",noprefix");
870 if (root->flags & CGRP_ROOT_XATTR)
871 seq_puts(seq, ",xattr");
872 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
873 seq_puts(seq, ",cpuset_v2_mode");
875 spin_lock(&release_agent_path_lock);
876 if (strlen(root->release_agent_path))
877 seq_show_option(seq, "release_agent",
878 root->release_agent_path);
879 spin_unlock(&release_agent_path_lock);
881 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
882 seq_puts(seq, ",clone_children");
883 if (strlen(root->name))
884 seq_show_option(seq, "name", root->name);
899 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
900 fsparam_flag ("all", Opt_all),
901 fsparam_flag ("clone_children", Opt_clone_children),
902 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
903 fsparam_string("name", Opt_name),
904 fsparam_flag ("none", Opt_none),
905 fsparam_flag ("noprefix", Opt_noprefix),
906 fsparam_string("release_agent", Opt_release_agent),
907 fsparam_flag ("xattr", Opt_xattr),
911 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
913 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
914 struct cgroup_subsys *ss;
915 struct fs_parse_result result;
918 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
919 if (opt == -ENOPARAM) {
922 ret = vfs_parse_fs_param_source(fc, param);
923 if (ret != -ENOPARAM)
925 for_each_subsys(ss, i) {
926 if (strcmp(param->key, ss->legacy_name))
928 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
929 return invalfc(fc, "Disabled controller '%s'",
931 ctx->subsys_mask |= (1 << i);
934 return invalfc(fc, "Unknown subsys name '%s'", param->key);
941 /* Explicitly have no subsystems */
948 ctx->flags |= CGRP_ROOT_NOPREFIX;
950 case Opt_clone_children:
951 ctx->cpuset_clone_children = true;
953 case Opt_cpuset_v2_mode:
954 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
957 ctx->flags |= CGRP_ROOT_XATTR;
959 case Opt_release_agent:
960 /* Specifying two release agents is forbidden */
961 if (ctx->release_agent)
962 return invalfc(fc, "release_agent respecified");
963 ctx->release_agent = param->string;
964 param->string = NULL;
967 /* blocked by boot param? */
968 if (cgroup_no_v1_named)
970 /* Can't specify an empty name */
972 return invalfc(fc, "Empty name");
973 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
974 return invalfc(fc, "Name too long");
975 /* Must match [\w.-]+ */
976 for (i = 0; i < param->size; i++) {
977 char c = param->string[i];
980 if ((c == '.') || (c == '-') || (c == '_'))
982 return invalfc(fc, "Invalid name");
984 /* Specifying two names is forbidden */
986 return invalfc(fc, "name respecified");
987 ctx->name = param->string;
988 param->string = NULL;
994 static int check_cgroupfs_options(struct fs_context *fc)
996 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
999 struct cgroup_subsys *ss;
1002 #ifdef CONFIG_CPUSETS
1003 mask = ~((u16)1 << cpuset_cgrp_id);
1005 for_each_subsys(ss, i)
1006 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1009 ctx->subsys_mask &= enabled;
1012 * In absence of 'none', 'name=' and subsystem name options,
1013 * let's default to 'all'.
1015 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1019 /* Mutually exclusive option 'all' + subsystem name */
1020 if (ctx->subsys_mask)
1021 return invalfc(fc, "subsys name conflicts with all");
1022 /* 'all' => select all the subsystems */
1023 ctx->subsys_mask = enabled;
1027 * We either have to specify by name or by subsystems. (So all
1028 * empty hierarchies must have a name).
1030 if (!ctx->subsys_mask && !ctx->name)
1031 return invalfc(fc, "Need name or subsystem set");
1034 * Option noprefix was introduced just for backward compatibility
1035 * with the old cpuset, so we allow noprefix only if mounting just
1036 * the cpuset subsystem.
1038 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1039 return invalfc(fc, "noprefix used incorrectly");
1041 /* Can't specify "none" and some subsystems */
1042 if (ctx->subsys_mask && ctx->none)
1043 return invalfc(fc, "none used incorrectly");
1048 int cgroup1_reconfigure(struct fs_context *fc)
1050 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1051 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1052 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1054 u16 added_mask, removed_mask;
1056 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1058 /* See what subsystems are wanted */
1059 ret = check_cgroupfs_options(fc);
1063 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1064 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1065 task_tgid_nr(current), current->comm);
1067 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1068 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1070 /* Don't allow flags or name to change at remount */
1071 if ((ctx->flags ^ root->flags) ||
1072 (ctx->name && strcmp(ctx->name, root->name))) {
1073 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1074 ctx->flags, ctx->name ?: "", root->flags, root->name);
1079 /* remounting is not allowed for populated hierarchies */
1080 if (!list_empty(&root->cgrp.self.children)) {
1085 ret = rebind_subsystems(root, added_mask);
1089 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1091 if (ctx->release_agent) {
1092 spin_lock(&release_agent_path_lock);
1093 strcpy(root->release_agent_path, ctx->release_agent);
1094 spin_unlock(&release_agent_path_lock);
1097 trace_cgroup_remount(root);
1100 mutex_unlock(&cgroup_mutex);
1104 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1105 .rename = cgroup1_rename,
1106 .show_options = cgroup1_show_options,
1107 .mkdir = cgroup_mkdir,
1108 .rmdir = cgroup_rmdir,
1109 .show_path = cgroup_show_path,
1113 * The guts of cgroup1 mount - find or create cgroup_root to use.
1114 * Called with cgroup_mutex held; returns 0 on success, -E... on
1115 * error and positive - in case when the candidate is busy dying.
1116 * On success it stashes a reference to cgroup_root into given
1117 * cgroup_fs_context; that reference is *NOT* counting towards the
1118 * cgroup_root refcount.
1120 static int cgroup1_root_to_use(struct fs_context *fc)
1122 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1123 struct cgroup_root *root;
1124 struct cgroup_subsys *ss;
1127 /* First find the desired set of subsystems */
1128 ret = check_cgroupfs_options(fc);
1133 * Destruction of cgroup root is asynchronous, so subsystems may
1134 * still be dying after the previous unmount. Let's drain the
1135 * dying subsystems. We just need to ensure that the ones
1136 * unmounted previously finish dying and don't care about new ones
1137 * starting. Testing ref liveliness is good enough.
1139 for_each_subsys(ss, i) {
1140 if (!(ctx->subsys_mask & (1 << i)) ||
1141 ss->root == &cgrp_dfl_root)
1144 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1145 return 1; /* restart */
1146 cgroup_put(&ss->root->cgrp);
1149 for_each_root(root) {
1150 bool name_match = false;
1152 if (root == &cgrp_dfl_root)
1156 * If we asked for a name then it must match. Also, if
1157 * name matches but sybsys_mask doesn't, we should fail.
1158 * Remember whether name matched.
1161 if (strcmp(ctx->name, root->name))
1167 * If we asked for subsystems (or explicitly for no
1168 * subsystems) then they must match.
1170 if ((ctx->subsys_mask || ctx->none) &&
1171 (ctx->subsys_mask != root->subsys_mask)) {
1177 if (root->flags ^ ctx->flags)
1178 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1185 * No such thing, create a new one. name= matching without subsys
1186 * specification is allowed for already existing hierarchies but we
1187 * can't create new one without subsys specification.
1189 if (!ctx->subsys_mask && !ctx->none)
1190 return invalfc(fc, "No subsys list or none specified");
1192 /* Hierarchies may only be created in the initial cgroup namespace. */
1193 if (ctx->ns != &init_cgroup_ns)
1196 root = kzalloc(sizeof(*root), GFP_KERNEL);
1201 init_cgroup_root(ctx);
1203 ret = cgroup_setup_root(root, ctx->subsys_mask);
1205 cgroup_free_root(root);
1209 int cgroup1_get_tree(struct fs_context *fc)
1211 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1214 /* Check if the caller has permission to mount. */
1215 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1218 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1220 ret = cgroup1_root_to_use(fc);
1221 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1222 ret = 1; /* restart */
1224 mutex_unlock(&cgroup_mutex);
1227 ret = cgroup_do_get_tree(fc);
1229 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1234 if (unlikely(ret > 0)) {
1236 return restart_syscall();
1241 static int __init cgroup1_wq_init(void)
1244 * Used to destroy pidlists and separate to serve as flush domain.
1245 * Cap @max_active to 1 too.
1247 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1249 BUG_ON(!cgroup_pidlist_destroy_wq);
1252 core_initcall(cgroup1_wq_init);
1254 static int __init cgroup_no_v1(char *str)
1256 struct cgroup_subsys *ss;
1260 while ((token = strsep(&str, ",")) != NULL) {
1264 if (!strcmp(token, "all")) {
1265 cgroup_no_v1_mask = U16_MAX;
1269 if (!strcmp(token, "named")) {
1270 cgroup_no_v1_named = true;
1274 for_each_subsys(ss, i) {
1275 if (strcmp(token, ss->name) &&
1276 strcmp(token, ss->legacy_name))
1279 cgroup_no_v1_mask |= 1 << i;
1284 __setup("cgroup_no_v1=", cgroup_no_v1);