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;
42 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
43 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
45 static DEFINE_SPINLOCK(release_agent_path_lock);
47 bool cgroup1_ssid_disabled(int ssid)
49 return cgroup_no_v1_mask & (1 << ssid);
53 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
54 * @from: attach to all cgroups of a given task
55 * @tsk: the task to be attached
57 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
59 struct cgroup_root *root;
62 mutex_lock(&cgroup_mutex);
63 percpu_down_write(&cgroup_threadgroup_rwsem);
65 struct cgroup *from_cgrp;
67 if (root == &cgrp_dfl_root)
70 spin_lock_irq(&css_set_lock);
71 from_cgrp = task_cgroup_from_root(from, root);
72 spin_unlock_irq(&css_set_lock);
74 retval = cgroup_attach_task(from_cgrp, tsk, false);
78 percpu_up_write(&cgroup_threadgroup_rwsem);
79 mutex_unlock(&cgroup_mutex);
83 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
86 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
87 * @to: cgroup to which the tasks will be moved
88 * @from: cgroup in which the tasks currently reside
90 * Locking rules between cgroup_post_fork() and the migration path
91 * guarantee that, if a task is forking while being migrated, the new child
92 * is guaranteed to be either visible in the source cgroup after the
93 * parent's migration is complete or put into the target cgroup. No task
94 * can slip out of migration through forking.
96 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
98 DEFINE_CGROUP_MGCTX(mgctx);
99 struct cgrp_cset_link *link;
100 struct css_task_iter it;
101 struct task_struct *task;
104 if (cgroup_on_dfl(to))
107 ret = cgroup_migrate_vet_dst(to);
111 mutex_lock(&cgroup_mutex);
113 percpu_down_write(&cgroup_threadgroup_rwsem);
115 /* all tasks in @from are being moved, all csets are source */
116 spin_lock_irq(&css_set_lock);
117 list_for_each_entry(link, &from->cset_links, cset_link)
118 cgroup_migrate_add_src(link->cset, to, &mgctx);
119 spin_unlock_irq(&css_set_lock);
121 ret = cgroup_migrate_prepare_dst(&mgctx);
126 * Migrate tasks one-by-one until @from is empty. This fails iff
127 * ->can_attach() fails.
130 css_task_iter_start(&from->self, 0, &it);
133 task = css_task_iter_next(&it);
134 } while (task && (task->flags & PF_EXITING));
137 get_task_struct(task);
138 css_task_iter_end(&it);
141 ret = cgroup_migrate(task, false, &mgctx);
143 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
144 put_task_struct(task);
146 } while (task && !ret);
148 cgroup_migrate_finish(&mgctx);
149 percpu_up_write(&cgroup_threadgroup_rwsem);
150 mutex_unlock(&cgroup_mutex);
155 * Stuff for reading the 'tasks'/'procs' files.
157 * Reading this file can return large amounts of data if a cgroup has
158 * *lots* of attached tasks. So it may need several calls to read(),
159 * but we cannot guarantee that the information we produce is correct
160 * unless we produce it entirely atomically.
164 /* which pidlist file are we talking about? */
165 enum cgroup_filetype {
171 * A pidlist is a list of pids that virtually represents the contents of one
172 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
173 * a pair (one each for procs, tasks) for each pid namespace that's relevant
176 struct cgroup_pidlist {
178 * used to find which pidlist is wanted. doesn't change as long as
179 * this particular list stays in the list.
181 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
184 /* how many elements the above list has */
186 /* each of these stored in a list by its cgroup */
187 struct list_head links;
188 /* pointer to the cgroup we belong to, for list removal purposes */
189 struct cgroup *owner;
190 /* for delayed destruction */
191 struct delayed_work destroy_dwork;
195 * Used to destroy all pidlists lingering waiting for destroy timer. None
196 * should be left afterwards.
198 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
200 struct cgroup_pidlist *l, *tmp_l;
202 mutex_lock(&cgrp->pidlist_mutex);
203 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
204 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
205 mutex_unlock(&cgrp->pidlist_mutex);
207 flush_workqueue(cgroup_pidlist_destroy_wq);
208 BUG_ON(!list_empty(&cgrp->pidlists));
211 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
213 struct delayed_work *dwork = to_delayed_work(work);
214 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
216 struct cgroup_pidlist *tofree = NULL;
218 mutex_lock(&l->owner->pidlist_mutex);
221 * Destroy iff we didn't get queued again. The state won't change
222 * as destroy_dwork can only be queued while locked.
224 if (!delayed_work_pending(dwork)) {
227 put_pid_ns(l->key.ns);
231 mutex_unlock(&l->owner->pidlist_mutex);
236 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
237 * Returns the number of unique elements.
239 static int pidlist_uniq(pid_t *list, int length)
244 * we presume the 0th element is unique, so i starts at 1. trivial
245 * edge cases first; no work needs to be done for either
247 if (length == 0 || length == 1)
249 /* src and dest walk down the list; dest counts unique elements */
250 for (src = 1; src < length; src++) {
251 /* find next unique element */
252 while (list[src] == list[src-1]) {
257 /* dest always points to where the next unique element goes */
258 list[dest] = list[src];
266 * The two pid files - task and cgroup.procs - guaranteed that the result
267 * is sorted, which forced this whole pidlist fiasco. As pid order is
268 * different per namespace, each namespace needs differently sorted list,
269 * making it impossible to use, for example, single rbtree of member tasks
270 * sorted by task pointer. As pidlists can be fairly large, allocating one
271 * per open file is dangerous, so cgroup had to implement shared pool of
272 * pidlists keyed by cgroup and namespace.
274 static int cmppid(const void *a, const void *b)
276 return *(pid_t *)a - *(pid_t *)b;
279 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
280 enum cgroup_filetype type)
282 struct cgroup_pidlist *l;
283 /* don't need task_nsproxy() if we're looking at ourself */
284 struct pid_namespace *ns = task_active_pid_ns(current);
286 lockdep_assert_held(&cgrp->pidlist_mutex);
288 list_for_each_entry(l, &cgrp->pidlists, links)
289 if (l->key.type == type && l->key.ns == ns)
295 * find the appropriate pidlist for our purpose (given procs vs tasks)
296 * returns with the lock on that pidlist already held, and takes care
297 * of the use count, or returns NULL with no locks held if we're out of
300 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
301 enum cgroup_filetype type)
303 struct cgroup_pidlist *l;
305 lockdep_assert_held(&cgrp->pidlist_mutex);
307 l = cgroup_pidlist_find(cgrp, type);
311 /* entry not found; create a new one */
312 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
316 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
318 /* don't need task_nsproxy() if we're looking at ourself */
319 l->key.ns = get_pid_ns(task_active_pid_ns(current));
321 list_add(&l->links, &cgrp->pidlists);
326 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
328 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
329 struct cgroup_pidlist **lp)
333 int pid, n = 0; /* used for populating the array */
334 struct css_task_iter it;
335 struct task_struct *tsk;
336 struct cgroup_pidlist *l;
338 lockdep_assert_held(&cgrp->pidlist_mutex);
341 * If cgroup gets more users after we read count, we won't have
342 * enough space - tough. This race is indistinguishable to the
343 * caller from the case that the additional cgroup users didn't
344 * show up until sometime later on.
346 length = cgroup_task_count(cgrp);
347 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
350 /* now, populate the array */
351 css_task_iter_start(&cgrp->self, 0, &it);
352 while ((tsk = css_task_iter_next(&it))) {
353 if (unlikely(n == length))
355 /* get tgid or pid for procs or tasks file respectively */
356 if (type == CGROUP_FILE_PROCS)
357 pid = task_tgid_vnr(tsk);
359 pid = task_pid_vnr(tsk);
360 if (pid > 0) /* make sure to only use valid results */
363 css_task_iter_end(&it);
365 /* now sort & (if procs) strip out duplicates */
366 sort(array, length, sizeof(pid_t), cmppid, NULL);
367 if (type == CGROUP_FILE_PROCS)
368 length = pidlist_uniq(array, length);
370 l = cgroup_pidlist_find_create(cgrp, type);
376 /* store array, freeing old if necessary */
385 * seq_file methods for the tasks/procs files. The seq_file position is the
386 * next pid to display; the seq_file iterator is a pointer to the pid
387 * in the cgroup->l->list array.
390 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
393 * Initially we receive a position value that corresponds to
394 * one more than the last pid shown (or 0 on the first call or
395 * after a seek to the start). Use a binary-search to find the
396 * next pid to display, if any
398 struct kernfs_open_file *of = s->private;
399 struct cgroup *cgrp = seq_css(s)->cgroup;
400 struct cgroup_pidlist *l;
401 enum cgroup_filetype type = seq_cft(s)->private;
402 int index = 0, pid = *pos;
405 mutex_lock(&cgrp->pidlist_mutex);
408 * !NULL @of->priv indicates that this isn't the first start()
409 * after open. If the matching pidlist is around, we can use that.
410 * Look for it. Note that @of->priv can't be used directly. It
411 * could already have been destroyed.
414 of->priv = cgroup_pidlist_find(cgrp, type);
417 * Either this is the first start() after open or the matching
418 * pidlist has been destroyed inbetween. Create a new one.
421 ret = pidlist_array_load(cgrp, type,
422 (struct cgroup_pidlist **)&of->priv);
431 while (index < end) {
432 int mid = (index + end) / 2;
433 if (l->list[mid] == pid) {
436 } else if (l->list[mid] <= pid)
442 /* If we're off the end of the array, we're done */
443 if (index >= l->length)
445 /* Update the abstract position to be the actual pid that we found */
446 iter = l->list + index;
451 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
453 struct kernfs_open_file *of = s->private;
454 struct cgroup_pidlist *l = of->priv;
457 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
458 CGROUP_PIDLIST_DESTROY_DELAY);
459 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
462 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
464 struct kernfs_open_file *of = s->private;
465 struct cgroup_pidlist *l = of->priv;
467 pid_t *end = l->list + l->length;
469 * Advance to the next pid in the array. If this goes off the
482 static int cgroup_pidlist_show(struct seq_file *s, void *v)
484 seq_printf(s, "%d\n", *(int *)v);
489 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
490 char *buf, size_t nbytes, loff_t off,
494 struct task_struct *task;
495 const struct cred *cred, *tcred;
499 cgrp = cgroup_kn_lock_live(of->kn, false);
503 task = cgroup_procs_write_start(buf, threadgroup, &locked);
504 ret = PTR_ERR_OR_ZERO(task);
509 * Even if we're attaching all tasks in the thread group, we only
510 * need to check permissions on one of them.
512 cred = current_cred();
513 tcred = get_task_cred(task);
514 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
515 !uid_eq(cred->euid, tcred->uid) &&
516 !uid_eq(cred->euid, tcred->suid))
522 ret = cgroup_attach_task(cgrp, task, threadgroup);
525 cgroup_procs_write_finish(task, locked);
527 cgroup_kn_unlock(of->kn);
529 return ret ?: nbytes;
532 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
533 char *buf, size_t nbytes, loff_t off)
535 return __cgroup1_procs_write(of, buf, nbytes, off, true);
538 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
539 char *buf, size_t nbytes, loff_t off)
541 return __cgroup1_procs_write(of, buf, nbytes, off, false);
544 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
545 char *buf, size_t nbytes, loff_t off)
549 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
551 cgrp = cgroup_kn_lock_live(of->kn, false);
554 spin_lock(&release_agent_path_lock);
555 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
556 sizeof(cgrp->root->release_agent_path));
557 spin_unlock(&release_agent_path_lock);
558 cgroup_kn_unlock(of->kn);
562 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
564 struct cgroup *cgrp = seq_css(seq)->cgroup;
566 spin_lock(&release_agent_path_lock);
567 seq_puts(seq, cgrp->root->release_agent_path);
568 spin_unlock(&release_agent_path_lock);
573 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
575 seq_puts(seq, "0\n");
579 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
582 return notify_on_release(css->cgroup);
585 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
586 struct cftype *cft, u64 val)
589 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
591 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
595 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
598 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
601 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
602 struct cftype *cft, u64 val)
605 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
607 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
611 /* cgroup core interface files for the legacy hierarchies */
612 struct cftype cgroup1_base_files[] = {
614 .name = "cgroup.procs",
615 .seq_start = cgroup_pidlist_start,
616 .seq_next = cgroup_pidlist_next,
617 .seq_stop = cgroup_pidlist_stop,
618 .seq_show = cgroup_pidlist_show,
619 .private = CGROUP_FILE_PROCS,
620 .write = cgroup1_procs_write,
623 .name = "cgroup.clone_children",
624 .read_u64 = cgroup_clone_children_read,
625 .write_u64 = cgroup_clone_children_write,
628 .name = "cgroup.sane_behavior",
629 .flags = CFTYPE_ONLY_ON_ROOT,
630 .seq_show = cgroup_sane_behavior_show,
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_TASKS,
639 .write = cgroup1_tasks_write,
642 .name = "notify_on_release",
643 .read_u64 = cgroup_read_notify_on_release,
644 .write_u64 = cgroup_write_notify_on_release,
647 .name = "release_agent",
648 .flags = CFTYPE_ONLY_ON_ROOT,
649 .seq_show = cgroup_release_agent_show,
650 .write = cgroup_release_agent_write,
651 .max_write_len = PATH_MAX - 1,
656 /* Display information about each subsystem and each hierarchy */
657 int proc_cgroupstats_show(struct seq_file *m, void *v)
659 struct cgroup_subsys *ss;
662 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
664 * ideally we don't want subsystems moving around while we do this.
665 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
666 * subsys/hierarchy state.
668 mutex_lock(&cgroup_mutex);
670 for_each_subsys(ss, i)
671 seq_printf(m, "%s\t%d\t%d\t%d\n",
672 ss->legacy_name, ss->root->hierarchy_id,
673 atomic_read(&ss->root->nr_cgrps),
674 cgroup_ssid_enabled(i));
676 mutex_unlock(&cgroup_mutex);
681 * cgroupstats_build - build and fill cgroupstats
682 * @stats: cgroupstats to fill information into
683 * @dentry: A dentry entry belonging to the cgroup for which stats have
686 * Build and fill cgroupstats so that taskstats can export it to user
689 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
691 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
693 struct css_task_iter it;
694 struct task_struct *tsk;
696 /* it should be kernfs_node belonging to cgroupfs and is a directory */
697 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
698 kernfs_type(kn) != KERNFS_DIR)
701 mutex_lock(&cgroup_mutex);
704 * We aren't being called from kernfs and there's no guarantee on
705 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
706 * @kn->priv is RCU safe. Let's do the RCU dancing.
709 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
710 if (!cgrp || cgroup_is_dead(cgrp)) {
712 mutex_unlock(&cgroup_mutex);
717 css_task_iter_start(&cgrp->self, 0, &it);
718 while ((tsk = css_task_iter_next(&it))) {
719 switch (tsk->state) {
723 case TASK_INTERRUPTIBLE:
724 stats->nr_sleeping++;
726 case TASK_UNINTERRUPTIBLE:
727 stats->nr_uninterruptible++;
733 if (delayacct_is_task_waiting_on_io(tsk))
738 css_task_iter_end(&it);
740 mutex_unlock(&cgroup_mutex);
744 void cgroup1_check_for_release(struct cgroup *cgrp)
746 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
747 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
748 schedule_work(&cgrp->release_agent_work);
752 * Notify userspace when a cgroup is released, by running the
753 * configured release agent with the name of the cgroup (path
754 * relative to the root of cgroup file system) as the argument.
756 * Most likely, this user command will try to rmdir this cgroup.
758 * This races with the possibility that some other task will be
759 * attached to this cgroup before it is removed, or that some other
760 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
761 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
762 * unused, and this cgroup will be reprieved from its death sentence,
763 * to continue to serve a useful existence. Next time it's released,
764 * we will get notified again, if it still has 'notify_on_release' set.
766 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
767 * means only wait until the task is successfully execve()'d. The
768 * separate release agent task is forked by call_usermodehelper(),
769 * then control in this thread returns here, without waiting for the
770 * release agent task. We don't bother to wait because the caller of
771 * this routine has no use for the exit status of the release agent
772 * task, so no sense holding our caller up for that.
774 void cgroup1_release_agent(struct work_struct *work)
776 struct cgroup *cgrp =
777 container_of(work, struct cgroup, release_agent_work);
778 char *pathbuf = NULL, *agentbuf = NULL;
779 char *argv[3], *envp[3];
782 mutex_lock(&cgroup_mutex);
784 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
785 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
786 if (!pathbuf || !agentbuf || !strlen(agentbuf))
789 spin_lock_irq(&css_set_lock);
790 ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
791 spin_unlock_irq(&css_set_lock);
792 if (ret < 0 || ret >= PATH_MAX)
799 /* minimal command environment */
801 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
804 mutex_unlock(&cgroup_mutex);
805 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
808 mutex_unlock(&cgroup_mutex);
815 * cgroup_rename - Only allow simple rename of directories in place.
817 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
818 const char *new_name_str)
820 struct cgroup *cgrp = kn->priv;
823 if (kernfs_type(kn) != KERNFS_DIR)
825 if (kn->parent != new_parent)
829 * We're gonna grab cgroup_mutex which nests outside kernfs
830 * active_ref. kernfs_rename() doesn't require active_ref
831 * protection. Break them before grabbing cgroup_mutex.
833 kernfs_break_active_protection(new_parent);
834 kernfs_break_active_protection(kn);
836 mutex_lock(&cgroup_mutex);
838 ret = kernfs_rename(kn, new_parent, new_name_str);
840 TRACE_CGROUP_PATH(rename, cgrp);
842 mutex_unlock(&cgroup_mutex);
844 kernfs_unbreak_active_protection(kn);
845 kernfs_unbreak_active_protection(new_parent);
849 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
851 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
852 struct cgroup_subsys *ss;
855 for_each_subsys(ss, ssid)
856 if (root->subsys_mask & (1 << ssid))
857 seq_show_option(seq, ss->legacy_name, NULL);
858 if (root->flags & CGRP_ROOT_NOPREFIX)
859 seq_puts(seq, ",noprefix");
860 if (root->flags & CGRP_ROOT_XATTR)
861 seq_puts(seq, ",xattr");
862 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
863 seq_puts(seq, ",cpuset_v2_mode");
865 spin_lock(&release_agent_path_lock);
866 if (strlen(root->release_agent_path))
867 seq_show_option(seq, "release_agent",
868 root->release_agent_path);
869 spin_unlock(&release_agent_path_lock);
871 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
872 seq_puts(seq, ",clone_children");
873 if (strlen(root->name))
874 seq_show_option(seq, "name", root->name);
889 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
890 fsparam_flag ("all", Opt_all),
891 fsparam_flag ("clone_children", Opt_clone_children),
892 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
893 fsparam_string("name", Opt_name),
894 fsparam_flag ("none", Opt_none),
895 fsparam_flag ("noprefix", Opt_noprefix),
896 fsparam_string("release_agent", Opt_release_agent),
897 fsparam_flag ("xattr", Opt_xattr),
901 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
903 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
904 struct cgroup_subsys *ss;
905 struct fs_parse_result result;
908 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
909 if (opt == -ENOPARAM) {
910 if (strcmp(param->key, "source") == 0) {
911 fc->source = param->string;
912 param->string = NULL;
915 for_each_subsys(ss, i) {
916 if (strcmp(param->key, ss->legacy_name))
918 ctx->subsys_mask |= (1 << i);
921 return invalfc(fc, "Unknown subsys name '%s'", param->key);
928 /* Explicitly have no subsystems */
935 ctx->flags |= CGRP_ROOT_NOPREFIX;
937 case Opt_clone_children:
938 ctx->cpuset_clone_children = true;
940 case Opt_cpuset_v2_mode:
941 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
944 ctx->flags |= CGRP_ROOT_XATTR;
946 case Opt_release_agent:
947 /* Specifying two release agents is forbidden */
948 if (ctx->release_agent)
949 return invalfc(fc, "release_agent respecified");
950 ctx->release_agent = param->string;
951 param->string = NULL;
954 /* blocked by boot param? */
955 if (cgroup_no_v1_named)
957 /* Can't specify an empty name */
959 return invalfc(fc, "Empty name");
960 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
961 return invalfc(fc, "Name too long");
962 /* Must match [\w.-]+ */
963 for (i = 0; i < param->size; i++) {
964 char c = param->string[i];
967 if ((c == '.') || (c == '-') || (c == '_'))
969 return invalfc(fc, "Invalid name");
971 /* Specifying two names is forbidden */
973 return invalfc(fc, "name respecified");
974 ctx->name = param->string;
975 param->string = NULL;
981 static int check_cgroupfs_options(struct fs_context *fc)
983 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
986 struct cgroup_subsys *ss;
989 #ifdef CONFIG_CPUSETS
990 mask = ~((u16)1 << cpuset_cgrp_id);
992 for_each_subsys(ss, i)
993 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
996 ctx->subsys_mask &= enabled;
999 * In absense of 'none', 'name=' or subsystem name options,
1000 * let's default to 'all'.
1002 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1006 /* Mutually exclusive option 'all' + subsystem name */
1007 if (ctx->subsys_mask)
1008 return invalfc(fc, "subsys name conflicts with all");
1009 /* 'all' => select all the subsystems */
1010 ctx->subsys_mask = enabled;
1014 * We either have to specify by name or by subsystems. (So all
1015 * empty hierarchies must have a name).
1017 if (!ctx->subsys_mask && !ctx->name)
1018 return invalfc(fc, "Need name or subsystem set");
1021 * Option noprefix was introduced just for backward compatibility
1022 * with the old cpuset, so we allow noprefix only if mounting just
1023 * the cpuset subsystem.
1025 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1026 return invalfc(fc, "noprefix used incorrectly");
1028 /* Can't specify "none" and some subsystems */
1029 if (ctx->subsys_mask && ctx->none)
1030 return invalfc(fc, "none used incorrectly");
1035 int cgroup1_reconfigure(struct fs_context *fc)
1037 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1038 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1039 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1041 u16 added_mask, removed_mask;
1043 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1045 /* See what subsystems are wanted */
1046 ret = check_cgroupfs_options(fc);
1050 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1051 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1052 task_tgid_nr(current), current->comm);
1054 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1055 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1057 /* Don't allow flags or name to change at remount */
1058 if ((ctx->flags ^ root->flags) ||
1059 (ctx->name && strcmp(ctx->name, root->name))) {
1060 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1061 ctx->flags, ctx->name ?: "", root->flags, root->name);
1066 /* remounting is not allowed for populated hierarchies */
1067 if (!list_empty(&root->cgrp.self.children)) {
1072 ret = rebind_subsystems(root, added_mask);
1076 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1078 if (ctx->release_agent) {
1079 spin_lock(&release_agent_path_lock);
1080 strcpy(root->release_agent_path, ctx->release_agent);
1081 spin_unlock(&release_agent_path_lock);
1084 trace_cgroup_remount(root);
1087 mutex_unlock(&cgroup_mutex);
1091 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1092 .rename = cgroup1_rename,
1093 .show_options = cgroup1_show_options,
1094 .mkdir = cgroup_mkdir,
1095 .rmdir = cgroup_rmdir,
1096 .show_path = cgroup_show_path,
1100 * The guts of cgroup1 mount - find or create cgroup_root to use.
1101 * Called with cgroup_mutex held; returns 0 on success, -E... on
1102 * error and positive - in case when the candidate is busy dying.
1103 * On success it stashes a reference to cgroup_root into given
1104 * cgroup_fs_context; that reference is *NOT* counting towards the
1105 * cgroup_root refcount.
1107 static int cgroup1_root_to_use(struct fs_context *fc)
1109 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1110 struct cgroup_root *root;
1111 struct cgroup_subsys *ss;
1114 /* First find the desired set of subsystems */
1115 ret = check_cgroupfs_options(fc);
1120 * Destruction of cgroup root is asynchronous, so subsystems may
1121 * still be dying after the previous unmount. Let's drain the
1122 * dying subsystems. We just need to ensure that the ones
1123 * unmounted previously finish dying and don't care about new ones
1124 * starting. Testing ref liveliness is good enough.
1126 for_each_subsys(ss, i) {
1127 if (!(ctx->subsys_mask & (1 << i)) ||
1128 ss->root == &cgrp_dfl_root)
1131 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1132 return 1; /* restart */
1133 cgroup_put(&ss->root->cgrp);
1136 for_each_root(root) {
1137 bool name_match = false;
1139 if (root == &cgrp_dfl_root)
1143 * If we asked for a name then it must match. Also, if
1144 * name matches but sybsys_mask doesn't, we should fail.
1145 * Remember whether name matched.
1148 if (strcmp(ctx->name, root->name))
1154 * If we asked for subsystems (or explicitly for no
1155 * subsystems) then they must match.
1157 if ((ctx->subsys_mask || ctx->none) &&
1158 (ctx->subsys_mask != root->subsys_mask)) {
1164 if (root->flags ^ ctx->flags)
1165 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1172 * No such thing, create a new one. name= matching without subsys
1173 * specification is allowed for already existing hierarchies but we
1174 * can't create new one without subsys specification.
1176 if (!ctx->subsys_mask && !ctx->none)
1177 return invalfc(fc, "No subsys list or none specified");
1179 /* Hierarchies may only be created in the initial cgroup namespace. */
1180 if (ctx->ns != &init_cgroup_ns)
1183 root = kzalloc(sizeof(*root), GFP_KERNEL);
1188 init_cgroup_root(ctx);
1190 ret = cgroup_setup_root(root, ctx->subsys_mask);
1192 cgroup_free_root(root);
1196 int cgroup1_get_tree(struct fs_context *fc)
1198 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1201 /* Check if the caller has permission to mount. */
1202 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1205 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1207 ret = cgroup1_root_to_use(fc);
1208 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1209 ret = 1; /* restart */
1211 mutex_unlock(&cgroup_mutex);
1214 ret = cgroup_do_get_tree(fc);
1216 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1217 struct super_block *sb = fc->root->d_sb;
1219 deactivate_locked_super(sb);
1223 if (unlikely(ret > 0)) {
1225 return restart_syscall();
1230 static int __init cgroup1_wq_init(void)
1233 * Used to destroy pidlists and separate to serve as flush domain.
1234 * Cap @max_active to 1 too.
1236 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1238 BUG_ON(!cgroup_pidlist_destroy_wq);
1241 core_initcall(cgroup1_wq_init);
1243 static int __init cgroup_no_v1(char *str)
1245 struct cgroup_subsys *ss;
1249 while ((token = strsep(&str, ",")) != NULL) {
1253 if (!strcmp(token, "all")) {
1254 cgroup_no_v1_mask = U16_MAX;
1258 if (!strcmp(token, "named")) {
1259 cgroup_no_v1_named = true;
1263 for_each_subsys(ss, i) {
1264 if (strcmp(token, ss->name) &&
1265 strcmp(token, ss->legacy_name))
1268 cgroup_no_v1_mask |= 1 << i;
1273 __setup("cgroup_no_v1=", cgroup_no_v1);