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 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
56 struct cgroup_root *root;
59 mutex_lock(&cgroup_mutex);
60 percpu_down_write(&cgroup_threadgroup_rwsem);
62 struct cgroup *from_cgrp;
64 if (root == &cgrp_dfl_root)
67 spin_lock_irq(&css_set_lock);
68 from_cgrp = task_cgroup_from_root(from, root);
69 spin_unlock_irq(&css_set_lock);
71 retval = cgroup_attach_task(from_cgrp, tsk, false);
75 percpu_up_write(&cgroup_threadgroup_rwsem);
76 mutex_unlock(&cgroup_mutex);
80 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
83 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
84 * @to: cgroup to which the tasks will be moved
85 * @from: cgroup in which the tasks currently reside
87 * Locking rules between cgroup_post_fork() and the migration path
88 * guarantee that, if a task is forking while being migrated, the new child
89 * is guaranteed to be either visible in the source cgroup after the
90 * parent's migration is complete or put into the target cgroup. No task
91 * can slip out of migration through forking.
93 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
95 DEFINE_CGROUP_MGCTX(mgctx);
96 struct cgrp_cset_link *link;
97 struct css_task_iter it;
98 struct task_struct *task;
101 if (cgroup_on_dfl(to))
104 ret = cgroup_migrate_vet_dst(to);
108 mutex_lock(&cgroup_mutex);
110 percpu_down_write(&cgroup_threadgroup_rwsem);
112 /* all tasks in @from are being moved, all csets are source */
113 spin_lock_irq(&css_set_lock);
114 list_for_each_entry(link, &from->cset_links, cset_link)
115 cgroup_migrate_add_src(link->cset, to, &mgctx);
116 spin_unlock_irq(&css_set_lock);
118 ret = cgroup_migrate_prepare_dst(&mgctx);
123 * Migrate tasks one-by-one until @from is empty. This fails iff
124 * ->can_attach() fails.
127 css_task_iter_start(&from->self, 0, &it);
130 task = css_task_iter_next(&it);
131 } while (task && (task->flags & PF_EXITING));
134 get_task_struct(task);
135 css_task_iter_end(&it);
138 ret = cgroup_migrate(task, false, &mgctx);
140 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
141 put_task_struct(task);
143 } while (task && !ret);
145 cgroup_migrate_finish(&mgctx);
146 percpu_up_write(&cgroup_threadgroup_rwsem);
147 mutex_unlock(&cgroup_mutex);
152 * Stuff for reading the 'tasks'/'procs' files.
154 * Reading this file can return large amounts of data if a cgroup has
155 * *lots* of attached tasks. So it may need several calls to read(),
156 * but we cannot guarantee that the information we produce is correct
157 * unless we produce it entirely atomically.
161 /* which pidlist file are we talking about? */
162 enum cgroup_filetype {
168 * A pidlist is a list of pids that virtually represents the contents of one
169 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
170 * a pair (one each for procs, tasks) for each pid namespace that's relevant
173 struct cgroup_pidlist {
175 * used to find which pidlist is wanted. doesn't change as long as
176 * this particular list stays in the list.
178 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
181 /* how many elements the above list has */
183 /* each of these stored in a list by its cgroup */
184 struct list_head links;
185 /* pointer to the cgroup we belong to, for list removal purposes */
186 struct cgroup *owner;
187 /* for delayed destruction */
188 struct delayed_work destroy_dwork;
192 * Used to destroy all pidlists lingering waiting for destroy timer. None
193 * should be left afterwards.
195 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
197 struct cgroup_pidlist *l, *tmp_l;
199 mutex_lock(&cgrp->pidlist_mutex);
200 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
201 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
202 mutex_unlock(&cgrp->pidlist_mutex);
204 flush_workqueue(cgroup_pidlist_destroy_wq);
205 BUG_ON(!list_empty(&cgrp->pidlists));
208 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
210 struct delayed_work *dwork = to_delayed_work(work);
211 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
213 struct cgroup_pidlist *tofree = NULL;
215 mutex_lock(&l->owner->pidlist_mutex);
218 * Destroy iff we didn't get queued again. The state won't change
219 * as destroy_dwork can only be queued while locked.
221 if (!delayed_work_pending(dwork)) {
224 put_pid_ns(l->key.ns);
228 mutex_unlock(&l->owner->pidlist_mutex);
233 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
234 * Returns the number of unique elements.
236 static int pidlist_uniq(pid_t *list, int length)
241 * we presume the 0th element is unique, so i starts at 1. trivial
242 * edge cases first; no work needs to be done for either
244 if (length == 0 || length == 1)
246 /* src and dest walk down the list; dest counts unique elements */
247 for (src = 1; src < length; src++) {
248 /* find next unique element */
249 while (list[src] == list[src-1]) {
254 /* dest always points to where the next unique element goes */
255 list[dest] = list[src];
263 * The two pid files - task and cgroup.procs - guaranteed that the result
264 * is sorted, which forced this whole pidlist fiasco. As pid order is
265 * different per namespace, each namespace needs differently sorted list,
266 * making it impossible to use, for example, single rbtree of member tasks
267 * sorted by task pointer. As pidlists can be fairly large, allocating one
268 * per open file is dangerous, so cgroup had to implement shared pool of
269 * pidlists keyed by cgroup and namespace.
271 static int cmppid(const void *a, const void *b)
273 return *(pid_t *)a - *(pid_t *)b;
276 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
277 enum cgroup_filetype type)
279 struct cgroup_pidlist *l;
280 /* don't need task_nsproxy() if we're looking at ourself */
281 struct pid_namespace *ns = task_active_pid_ns(current);
283 lockdep_assert_held(&cgrp->pidlist_mutex);
285 list_for_each_entry(l, &cgrp->pidlists, links)
286 if (l->key.type == type && l->key.ns == ns)
292 * find the appropriate pidlist for our purpose (given procs vs tasks)
293 * returns with the lock on that pidlist already held, and takes care
294 * of the use count, or returns NULL with no locks held if we're out of
297 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
298 enum cgroup_filetype type)
300 struct cgroup_pidlist *l;
302 lockdep_assert_held(&cgrp->pidlist_mutex);
304 l = cgroup_pidlist_find(cgrp, type);
308 /* entry not found; create a new one */
309 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
313 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
315 /* don't need task_nsproxy() if we're looking at ourself */
316 l->key.ns = get_pid_ns(task_active_pid_ns(current));
318 list_add(&l->links, &cgrp->pidlists);
323 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
325 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
326 struct cgroup_pidlist **lp)
330 int pid, n = 0; /* used for populating the array */
331 struct css_task_iter it;
332 struct task_struct *tsk;
333 struct cgroup_pidlist *l;
335 lockdep_assert_held(&cgrp->pidlist_mutex);
338 * If cgroup gets more users after we read count, we won't have
339 * enough space - tough. This race is indistinguishable to the
340 * caller from the case that the additional cgroup users didn't
341 * show up until sometime later on.
343 length = cgroup_task_count(cgrp);
344 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
347 /* now, populate the array */
348 css_task_iter_start(&cgrp->self, 0, &it);
349 while ((tsk = css_task_iter_next(&it))) {
350 if (unlikely(n == length))
352 /* get tgid or pid for procs or tasks file respectively */
353 if (type == CGROUP_FILE_PROCS)
354 pid = task_tgid_vnr(tsk);
356 pid = task_pid_vnr(tsk);
357 if (pid > 0) /* make sure to only use valid results */
360 css_task_iter_end(&it);
362 /* now sort & (if procs) strip out duplicates */
363 sort(array, length, sizeof(pid_t), cmppid, NULL);
364 if (type == CGROUP_FILE_PROCS)
365 length = pidlist_uniq(array, length);
367 l = cgroup_pidlist_find_create(cgrp, type);
373 /* store array, freeing old if necessary */
382 * seq_file methods for the tasks/procs files. The seq_file position is the
383 * next pid to display; the seq_file iterator is a pointer to the pid
384 * in the cgroup->l->list array.
387 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
390 * Initially we receive a position value that corresponds to
391 * one more than the last pid shown (or 0 on the first call or
392 * after a seek to the start). Use a binary-search to find the
393 * next pid to display, if any
395 struct kernfs_open_file *of = s->private;
396 struct cgroup *cgrp = seq_css(s)->cgroup;
397 struct cgroup_pidlist *l;
398 enum cgroup_filetype type = seq_cft(s)->private;
399 int index = 0, pid = *pos;
402 mutex_lock(&cgrp->pidlist_mutex);
405 * !NULL @of->priv indicates that this isn't the first start()
406 * after open. If the matching pidlist is around, we can use that.
407 * Look for it. Note that @of->priv can't be used directly. It
408 * could already have been destroyed.
411 of->priv = cgroup_pidlist_find(cgrp, type);
414 * Either this is the first start() after open or the matching
415 * pidlist has been destroyed inbetween. Create a new one.
418 ret = pidlist_array_load(cgrp, type,
419 (struct cgroup_pidlist **)&of->priv);
428 while (index < end) {
429 int mid = (index + end) / 2;
430 if (l->list[mid] == pid) {
433 } else if (l->list[mid] <= pid)
439 /* If we're off the end of the array, we're done */
440 if (index >= l->length)
442 /* Update the abstract position to be the actual pid that we found */
443 iter = l->list + index;
448 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
450 struct kernfs_open_file *of = s->private;
451 struct cgroup_pidlist *l = of->priv;
454 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
455 CGROUP_PIDLIST_DESTROY_DELAY);
456 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
459 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
461 struct kernfs_open_file *of = s->private;
462 struct cgroup_pidlist *l = of->priv;
464 pid_t *end = l->list + l->length;
466 * Advance to the next pid in the array. If this goes off the
479 static int cgroup_pidlist_show(struct seq_file *s, void *v)
481 seq_printf(s, "%d\n", *(int *)v);
486 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
487 char *buf, size_t nbytes, loff_t off,
491 struct task_struct *task;
492 const struct cred *cred, *tcred;
496 cgrp = cgroup_kn_lock_live(of->kn, false);
500 task = cgroup_procs_write_start(buf, threadgroup, &locked);
501 ret = PTR_ERR_OR_ZERO(task);
506 * Even if we're attaching all tasks in the thread group, we only
507 * need to check permissions on one of them.
509 cred = current_cred();
510 tcred = get_task_cred(task);
511 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
512 !uid_eq(cred->euid, tcred->uid) &&
513 !uid_eq(cred->euid, tcred->suid))
519 ret = cgroup_attach_task(cgrp, task, threadgroup);
522 cgroup_procs_write_finish(task, locked);
524 cgroup_kn_unlock(of->kn);
526 return ret ?: nbytes;
529 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
530 char *buf, size_t nbytes, loff_t off)
532 return __cgroup1_procs_write(of, buf, nbytes, off, true);
535 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
536 char *buf, size_t nbytes, loff_t off)
538 return __cgroup1_procs_write(of, buf, nbytes, off, false);
541 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
542 char *buf, size_t nbytes, loff_t off)
546 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
548 cgrp = cgroup_kn_lock_live(of->kn, false);
551 spin_lock(&release_agent_path_lock);
552 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
553 sizeof(cgrp->root->release_agent_path));
554 spin_unlock(&release_agent_path_lock);
555 cgroup_kn_unlock(of->kn);
559 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
561 struct cgroup *cgrp = seq_css(seq)->cgroup;
563 spin_lock(&release_agent_path_lock);
564 seq_puts(seq, cgrp->root->release_agent_path);
565 spin_unlock(&release_agent_path_lock);
570 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
572 seq_puts(seq, "0\n");
576 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
579 return notify_on_release(css->cgroup);
582 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
583 struct cftype *cft, u64 val)
586 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
588 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
592 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
595 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
598 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
599 struct cftype *cft, u64 val)
602 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
604 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
608 /* cgroup core interface files for the legacy hierarchies */
609 struct cftype cgroup1_base_files[] = {
611 .name = "cgroup.procs",
612 .seq_start = cgroup_pidlist_start,
613 .seq_next = cgroup_pidlist_next,
614 .seq_stop = cgroup_pidlist_stop,
615 .seq_show = cgroup_pidlist_show,
616 .private = CGROUP_FILE_PROCS,
617 .write = cgroup1_procs_write,
620 .name = "cgroup.clone_children",
621 .read_u64 = cgroup_clone_children_read,
622 .write_u64 = cgroup_clone_children_write,
625 .name = "cgroup.sane_behavior",
626 .flags = CFTYPE_ONLY_ON_ROOT,
627 .seq_show = cgroup_sane_behavior_show,
631 .seq_start = cgroup_pidlist_start,
632 .seq_next = cgroup_pidlist_next,
633 .seq_stop = cgroup_pidlist_stop,
634 .seq_show = cgroup_pidlist_show,
635 .private = CGROUP_FILE_TASKS,
636 .write = cgroup1_tasks_write,
639 .name = "notify_on_release",
640 .read_u64 = cgroup_read_notify_on_release,
641 .write_u64 = cgroup_write_notify_on_release,
644 .name = "release_agent",
645 .flags = CFTYPE_ONLY_ON_ROOT,
646 .seq_show = cgroup_release_agent_show,
647 .write = cgroup_release_agent_write,
648 .max_write_len = PATH_MAX - 1,
653 /* Display information about each subsystem and each hierarchy */
654 int proc_cgroupstats_show(struct seq_file *m, void *v)
656 struct cgroup_subsys *ss;
659 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
661 * ideally we don't want subsystems moving around while we do this.
662 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
663 * subsys/hierarchy state.
665 mutex_lock(&cgroup_mutex);
667 for_each_subsys(ss, i)
668 seq_printf(m, "%s\t%d\t%d\t%d\n",
669 ss->legacy_name, ss->root->hierarchy_id,
670 atomic_read(&ss->root->nr_cgrps),
671 cgroup_ssid_enabled(i));
673 mutex_unlock(&cgroup_mutex);
678 * cgroupstats_build - build and fill cgroupstats
679 * @stats: cgroupstats to fill information into
680 * @dentry: A dentry entry belonging to the cgroup for which stats have
683 * Build and fill cgroupstats so that taskstats can export it to user
686 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
688 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
690 struct css_task_iter it;
691 struct task_struct *tsk;
693 /* it should be kernfs_node belonging to cgroupfs and is a directory */
694 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
695 kernfs_type(kn) != KERNFS_DIR)
698 mutex_lock(&cgroup_mutex);
701 * We aren't being called from kernfs and there's no guarantee on
702 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
703 * @kn->priv is RCU safe. Let's do the RCU dancing.
706 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
707 if (!cgrp || cgroup_is_dead(cgrp)) {
709 mutex_unlock(&cgroup_mutex);
714 css_task_iter_start(&cgrp->self, 0, &it);
715 while ((tsk = css_task_iter_next(&it))) {
716 switch (tsk->state) {
720 case TASK_INTERRUPTIBLE:
721 stats->nr_sleeping++;
723 case TASK_UNINTERRUPTIBLE:
724 stats->nr_uninterruptible++;
735 css_task_iter_end(&it);
737 mutex_unlock(&cgroup_mutex);
741 void cgroup1_check_for_release(struct cgroup *cgrp)
743 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
744 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
745 schedule_work(&cgrp->release_agent_work);
749 * Notify userspace when a cgroup is released, by running the
750 * configured release agent with the name of the cgroup (path
751 * relative to the root of cgroup file system) as the argument.
753 * Most likely, this user command will try to rmdir this cgroup.
755 * This races with the possibility that some other task will be
756 * attached to this cgroup before it is removed, or that some other
757 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
758 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
759 * unused, and this cgroup will be reprieved from its death sentence,
760 * to continue to serve a useful existence. Next time it's released,
761 * we will get notified again, if it still has 'notify_on_release' set.
763 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
764 * means only wait until the task is successfully execve()'d. The
765 * separate release agent task is forked by call_usermodehelper(),
766 * then control in this thread returns here, without waiting for the
767 * release agent task. We don't bother to wait because the caller of
768 * this routine has no use for the exit status of the release agent
769 * task, so no sense holding our caller up for that.
771 void cgroup1_release_agent(struct work_struct *work)
773 struct cgroup *cgrp =
774 container_of(work, struct cgroup, release_agent_work);
775 char *pathbuf, *agentbuf;
776 char *argv[3], *envp[3];
779 /* snoop agent path and exit early if empty */
780 if (!cgrp->root->release_agent_path[0])
783 /* prepare argument buffers */
784 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
785 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
786 if (!pathbuf || !agentbuf)
789 spin_lock(&release_agent_path_lock);
790 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
791 spin_unlock(&release_agent_path_lock);
795 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
796 if (ret < 0 || ret >= PATH_MAX)
803 /* minimal command environment */
805 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
808 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
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) {
912 return invalf(fc, "Multiple sources not supported");
913 fc->source = param->string;
914 param->string = NULL;
917 for_each_subsys(ss, i) {
918 if (strcmp(param->key, ss->legacy_name))
920 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
921 return invalfc(fc, "Disabled controller '%s'",
923 ctx->subsys_mask |= (1 << i);
926 return invalfc(fc, "Unknown subsys name '%s'", param->key);
933 /* Explicitly have no subsystems */
940 ctx->flags |= CGRP_ROOT_NOPREFIX;
942 case Opt_clone_children:
943 ctx->cpuset_clone_children = true;
945 case Opt_cpuset_v2_mode:
946 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
949 ctx->flags |= CGRP_ROOT_XATTR;
951 case Opt_release_agent:
952 /* Specifying two release agents is forbidden */
953 if (ctx->release_agent)
954 return invalfc(fc, "release_agent respecified");
955 ctx->release_agent = param->string;
956 param->string = NULL;
959 /* blocked by boot param? */
960 if (cgroup_no_v1_named)
962 /* Can't specify an empty name */
964 return invalfc(fc, "Empty name");
965 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
966 return invalfc(fc, "Name too long");
967 /* Must match [\w.-]+ */
968 for (i = 0; i < param->size; i++) {
969 char c = param->string[i];
972 if ((c == '.') || (c == '-') || (c == '_'))
974 return invalfc(fc, "Invalid name");
976 /* Specifying two names is forbidden */
978 return invalfc(fc, "name respecified");
979 ctx->name = param->string;
980 param->string = NULL;
986 static int check_cgroupfs_options(struct fs_context *fc)
988 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
991 struct cgroup_subsys *ss;
994 #ifdef CONFIG_CPUSETS
995 mask = ~((u16)1 << cpuset_cgrp_id);
997 for_each_subsys(ss, i)
998 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1001 ctx->subsys_mask &= enabled;
1004 * In absence of 'none', 'name=' and subsystem name options,
1005 * let's default to 'all'.
1007 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1011 /* Mutually exclusive option 'all' + subsystem name */
1012 if (ctx->subsys_mask)
1013 return invalfc(fc, "subsys name conflicts with all");
1014 /* 'all' => select all the subsystems */
1015 ctx->subsys_mask = enabled;
1019 * We either have to specify by name or by subsystems. (So all
1020 * empty hierarchies must have a name).
1022 if (!ctx->subsys_mask && !ctx->name)
1023 return invalfc(fc, "Need name or subsystem set");
1026 * Option noprefix was introduced just for backward compatibility
1027 * with the old cpuset, so we allow noprefix only if mounting just
1028 * the cpuset subsystem.
1030 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1031 return invalfc(fc, "noprefix used incorrectly");
1033 /* Can't specify "none" and some subsystems */
1034 if (ctx->subsys_mask && ctx->none)
1035 return invalfc(fc, "none used incorrectly");
1040 int cgroup1_reconfigure(struct fs_context *fc)
1042 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1043 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1044 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1046 u16 added_mask, removed_mask;
1048 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1050 /* See what subsystems are wanted */
1051 ret = check_cgroupfs_options(fc);
1055 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1056 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1057 task_tgid_nr(current), current->comm);
1059 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1060 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1062 /* Don't allow flags or name to change at remount */
1063 if ((ctx->flags ^ root->flags) ||
1064 (ctx->name && strcmp(ctx->name, root->name))) {
1065 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1066 ctx->flags, ctx->name ?: "", root->flags, root->name);
1071 /* remounting is not allowed for populated hierarchies */
1072 if (!list_empty(&root->cgrp.self.children)) {
1077 ret = rebind_subsystems(root, added_mask);
1081 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1083 if (ctx->release_agent) {
1084 spin_lock(&release_agent_path_lock);
1085 strcpy(root->release_agent_path, ctx->release_agent);
1086 spin_unlock(&release_agent_path_lock);
1089 trace_cgroup_remount(root);
1092 mutex_unlock(&cgroup_mutex);
1096 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1097 .rename = cgroup1_rename,
1098 .show_options = cgroup1_show_options,
1099 .mkdir = cgroup_mkdir,
1100 .rmdir = cgroup_rmdir,
1101 .show_path = cgroup_show_path,
1105 * The guts of cgroup1 mount - find or create cgroup_root to use.
1106 * Called with cgroup_mutex held; returns 0 on success, -E... on
1107 * error and positive - in case when the candidate is busy dying.
1108 * On success it stashes a reference to cgroup_root into given
1109 * cgroup_fs_context; that reference is *NOT* counting towards the
1110 * cgroup_root refcount.
1112 static int cgroup1_root_to_use(struct fs_context *fc)
1114 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1115 struct cgroup_root *root;
1116 struct cgroup_subsys *ss;
1119 /* First find the desired set of subsystems */
1120 ret = check_cgroupfs_options(fc);
1125 * Destruction of cgroup root is asynchronous, so subsystems may
1126 * still be dying after the previous unmount. Let's drain the
1127 * dying subsystems. We just need to ensure that the ones
1128 * unmounted previously finish dying and don't care about new ones
1129 * starting. Testing ref liveliness is good enough.
1131 for_each_subsys(ss, i) {
1132 if (!(ctx->subsys_mask & (1 << i)) ||
1133 ss->root == &cgrp_dfl_root)
1136 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1137 return 1; /* restart */
1138 cgroup_put(&ss->root->cgrp);
1141 for_each_root(root) {
1142 bool name_match = false;
1144 if (root == &cgrp_dfl_root)
1148 * If we asked for a name then it must match. Also, if
1149 * name matches but sybsys_mask doesn't, we should fail.
1150 * Remember whether name matched.
1153 if (strcmp(ctx->name, root->name))
1159 * If we asked for subsystems (or explicitly for no
1160 * subsystems) then they must match.
1162 if ((ctx->subsys_mask || ctx->none) &&
1163 (ctx->subsys_mask != root->subsys_mask)) {
1169 if (root->flags ^ ctx->flags)
1170 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1177 * No such thing, create a new one. name= matching without subsys
1178 * specification is allowed for already existing hierarchies but we
1179 * can't create new one without subsys specification.
1181 if (!ctx->subsys_mask && !ctx->none)
1182 return invalfc(fc, "No subsys list or none specified");
1184 /* Hierarchies may only be created in the initial cgroup namespace. */
1185 if (ctx->ns != &init_cgroup_ns)
1188 root = kzalloc(sizeof(*root), GFP_KERNEL);
1193 init_cgroup_root(ctx);
1195 ret = cgroup_setup_root(root, ctx->subsys_mask);
1197 cgroup_free_root(root);
1201 int cgroup1_get_tree(struct fs_context *fc)
1203 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1206 /* Check if the caller has permission to mount. */
1207 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1210 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1212 ret = cgroup1_root_to_use(fc);
1213 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1214 ret = 1; /* restart */
1216 mutex_unlock(&cgroup_mutex);
1219 ret = cgroup_do_get_tree(fc);
1221 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1222 struct super_block *sb = fc->root->d_sb;
1224 deactivate_locked_super(sb);
1228 if (unlikely(ret > 0)) {
1230 return restart_syscall();
1235 static int __init cgroup1_wq_init(void)
1238 * Used to destroy pidlists and separate to serve as flush domain.
1239 * Cap @max_active to 1 too.
1241 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1243 BUG_ON(!cgroup_pidlist_destroy_wq);
1246 core_initcall(cgroup1_wq_init);
1248 static int __init cgroup_no_v1(char *str)
1250 struct cgroup_subsys *ss;
1254 while ((token = strsep(&str, ",")) != NULL) {
1258 if (!strcmp(token, "all")) {
1259 cgroup_no_v1_mask = U16_MAX;
1263 if (!strcmp(token, "named")) {
1264 cgroup_no_v1_named = true;
1268 for_each_subsys(ss, i) {
1269 if (strcmp(token, ss->name) &&
1270 strcmp(token, ss->legacy_name))
1273 cgroup_no_v1_mask |= 1 << i;
1278 __setup("cgroup_no_v1=", cgroup_no_v1);