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 spin_lock_irq(&css_set_lock);
67 from_cgrp = task_cgroup_from_root(from, root);
68 spin_unlock_irq(&css_set_lock);
70 retval = cgroup_attach_task(from_cgrp, tsk, false);
74 percpu_up_write(&cgroup_threadgroup_rwsem);
75 mutex_unlock(&cgroup_mutex);
79 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
82 * cgroup_transfer_tasks - move tasks from one cgroup to another
83 * @to: cgroup to which the tasks will be moved
84 * @from: cgroup in which the tasks currently reside
86 * Locking rules between cgroup_post_fork() and the migration path
87 * guarantee that, if a task is forking while being migrated, the new child
88 * is guaranteed to be either visible in the source cgroup after the
89 * parent's migration is complete or put into the target cgroup. No task
90 * can slip out of migration through forking.
92 * Return: %0 on success or a negative errno code on failure
94 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
96 DEFINE_CGROUP_MGCTX(mgctx);
97 struct cgrp_cset_link *link;
98 struct css_task_iter it;
99 struct task_struct *task;
102 if (cgroup_on_dfl(to))
105 ret = cgroup_migrate_vet_dst(to);
109 mutex_lock(&cgroup_mutex);
111 percpu_down_write(&cgroup_threadgroup_rwsem);
113 /* all tasks in @from are being moved, all csets are source */
114 spin_lock_irq(&css_set_lock);
115 list_for_each_entry(link, &from->cset_links, cset_link)
116 cgroup_migrate_add_src(link->cset, to, &mgctx);
117 spin_unlock_irq(&css_set_lock);
119 ret = cgroup_migrate_prepare_dst(&mgctx);
124 * Migrate tasks one-by-one until @from is empty. This fails iff
125 * ->can_attach() fails.
128 css_task_iter_start(&from->self, 0, &it);
131 task = css_task_iter_next(&it);
132 } while (task && (task->flags & PF_EXITING));
135 get_task_struct(task);
136 css_task_iter_end(&it);
139 ret = cgroup_migrate(task, false, &mgctx);
141 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
142 put_task_struct(task);
144 } while (task && !ret);
146 cgroup_migrate_finish(&mgctx);
147 percpu_up_write(&cgroup_threadgroup_rwsem);
148 mutex_unlock(&cgroup_mutex);
153 * Stuff for reading the 'tasks'/'procs' files.
155 * Reading this file can return large amounts of data if a cgroup has
156 * *lots* of attached tasks. So it may need several calls to read(),
157 * but we cannot guarantee that the information we produce is correct
158 * unless we produce it entirely atomically.
162 /* which pidlist file are we talking about? */
163 enum cgroup_filetype {
169 * A pidlist is a list of pids that virtually represents the contents of one
170 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
171 * a pair (one each for procs, tasks) for each pid namespace that's relevant
174 struct cgroup_pidlist {
176 * used to find which pidlist is wanted. doesn't change as long as
177 * this particular list stays in the list.
179 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
182 /* how many elements the above list has */
184 /* each of these stored in a list by its cgroup */
185 struct list_head links;
186 /* pointer to the cgroup we belong to, for list removal purposes */
187 struct cgroup *owner;
188 /* for delayed destruction */
189 struct delayed_work destroy_dwork;
193 * Used to destroy all pidlists lingering waiting for destroy timer. None
194 * should be left afterwards.
196 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
198 struct cgroup_pidlist *l, *tmp_l;
200 mutex_lock(&cgrp->pidlist_mutex);
201 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
202 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
203 mutex_unlock(&cgrp->pidlist_mutex);
205 flush_workqueue(cgroup_pidlist_destroy_wq);
206 BUG_ON(!list_empty(&cgrp->pidlists));
209 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
211 struct delayed_work *dwork = to_delayed_work(work);
212 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
214 struct cgroup_pidlist *tofree = NULL;
216 mutex_lock(&l->owner->pidlist_mutex);
219 * Destroy iff we didn't get queued again. The state won't change
220 * as destroy_dwork can only be queued while locked.
222 if (!delayed_work_pending(dwork)) {
225 put_pid_ns(l->key.ns);
229 mutex_unlock(&l->owner->pidlist_mutex);
234 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
235 * Returns the number of unique elements.
237 static int pidlist_uniq(pid_t *list, int length)
242 * we presume the 0th element is unique, so i starts at 1. trivial
243 * edge cases first; no work needs to be done for either
245 if (length == 0 || length == 1)
247 /* src and dest walk down the list; dest counts unique elements */
248 for (src = 1; src < length; src++) {
249 /* find next unique element */
250 while (list[src] == list[src-1]) {
255 /* dest always points to where the next unique element goes */
256 list[dest] = list[src];
264 * The two pid files - task and cgroup.procs - guaranteed that the result
265 * is sorted, which forced this whole pidlist fiasco. As pid order is
266 * different per namespace, each namespace needs differently sorted list,
267 * making it impossible to use, for example, single rbtree of member tasks
268 * sorted by task pointer. As pidlists can be fairly large, allocating one
269 * per open file is dangerous, so cgroup had to implement shared pool of
270 * pidlists keyed by cgroup and namespace.
272 static int cmppid(const void *a, const void *b)
274 return *(pid_t *)a - *(pid_t *)b;
277 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
278 enum cgroup_filetype type)
280 struct cgroup_pidlist *l;
281 /* don't need task_nsproxy() if we're looking at ourself */
282 struct pid_namespace *ns = task_active_pid_ns(current);
284 lockdep_assert_held(&cgrp->pidlist_mutex);
286 list_for_each_entry(l, &cgrp->pidlists, links)
287 if (l->key.type == type && l->key.ns == ns)
293 * find the appropriate pidlist for our purpose (given procs vs tasks)
294 * returns with the lock on that pidlist already held, and takes care
295 * of the use count, or returns NULL with no locks held if we're out of
298 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
299 enum cgroup_filetype type)
301 struct cgroup_pidlist *l;
303 lockdep_assert_held(&cgrp->pidlist_mutex);
305 l = cgroup_pidlist_find(cgrp, type);
309 /* entry not found; create a new one */
310 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
314 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
316 /* don't need task_nsproxy() if we're looking at ourself */
317 l->key.ns = get_pid_ns(task_active_pid_ns(current));
319 list_add(&l->links, &cgrp->pidlists);
324 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
326 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
327 struct cgroup_pidlist **lp)
331 int pid, n = 0; /* used for populating the array */
332 struct css_task_iter it;
333 struct task_struct *tsk;
334 struct cgroup_pidlist *l;
336 lockdep_assert_held(&cgrp->pidlist_mutex);
339 * If cgroup gets more users after we read count, we won't have
340 * enough space - tough. This race is indistinguishable to the
341 * caller from the case that the additional cgroup users didn't
342 * show up until sometime later on.
344 length = cgroup_task_count(cgrp);
345 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
348 /* now, populate the array */
349 css_task_iter_start(&cgrp->self, 0, &it);
350 while ((tsk = css_task_iter_next(&it))) {
351 if (unlikely(n == length))
353 /* get tgid or pid for procs or tasks file respectively */
354 if (type == CGROUP_FILE_PROCS)
355 pid = task_tgid_vnr(tsk);
357 pid = task_pid_vnr(tsk);
358 if (pid > 0) /* make sure to only use valid results */
361 css_task_iter_end(&it);
363 /* now sort & (if procs) strip out duplicates */
364 sort(array, length, sizeof(pid_t), cmppid, NULL);
365 if (type == CGROUP_FILE_PROCS)
366 length = pidlist_uniq(array, length);
368 l = cgroup_pidlist_find_create(cgrp, type);
374 /* store array, freeing old if necessary */
383 * seq_file methods for the tasks/procs files. The seq_file position is the
384 * next pid to display; the seq_file iterator is a pointer to the pid
385 * in the cgroup->l->list array.
388 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
391 * Initially we receive a position value that corresponds to
392 * one more than the last pid shown (or 0 on the first call or
393 * after a seek to the start). Use a binary-search to find the
394 * next pid to display, if any
396 struct kernfs_open_file *of = s->private;
397 struct cgroup *cgrp = seq_css(s)->cgroup;
398 struct cgroup_pidlist *l;
399 enum cgroup_filetype type = seq_cft(s)->private;
400 int index = 0, pid = *pos;
403 mutex_lock(&cgrp->pidlist_mutex);
406 * !NULL @of->priv indicates that this isn't the first start()
407 * after open. If the matching pidlist is around, we can use that.
408 * Look for it. Note that @of->priv can't be used directly. It
409 * could already have been destroyed.
412 of->priv = cgroup_pidlist_find(cgrp, type);
415 * Either this is the first start() after open or the matching
416 * pidlist has been destroyed inbetween. Create a new one.
419 ret = pidlist_array_load(cgrp, type,
420 (struct cgroup_pidlist **)&of->priv);
429 while (index < end) {
430 int mid = (index + end) / 2;
431 if (l->list[mid] == pid) {
434 } else if (l->list[mid] <= pid)
440 /* If we're off the end of the array, we're done */
441 if (index >= l->length)
443 /* Update the abstract position to be the actual pid that we found */
444 iter = l->list + index;
449 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
451 struct kernfs_open_file *of = s->private;
452 struct cgroup_pidlist *l = of->priv;
455 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
456 CGROUP_PIDLIST_DESTROY_DELAY);
457 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
460 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
462 struct kernfs_open_file *of = s->private;
463 struct cgroup_pidlist *l = of->priv;
465 pid_t *end = l->list + l->length;
467 * Advance to the next pid in the array. If this goes off the
480 static int cgroup_pidlist_show(struct seq_file *s, void *v)
482 seq_printf(s, "%d\n", *(int *)v);
487 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
488 char *buf, size_t nbytes, loff_t off,
492 struct task_struct *task;
493 const struct cred *cred, *tcred;
497 cgrp = cgroup_kn_lock_live(of->kn, false);
501 task = cgroup_procs_write_start(buf, threadgroup, &locked);
502 ret = PTR_ERR_OR_ZERO(task);
507 * Even if we're attaching all tasks in the thread group, we only
508 * need to check permissions on one of them.
510 cred = current_cred();
511 tcred = get_task_cred(task);
512 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
513 !uid_eq(cred->euid, tcred->uid) &&
514 !uid_eq(cred->euid, tcred->suid))
520 ret = cgroup_attach_task(cgrp, task, threadgroup);
523 cgroup_procs_write_finish(task, locked);
525 cgroup_kn_unlock(of->kn);
527 return ret ?: nbytes;
530 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
531 char *buf, size_t nbytes, loff_t off)
533 return __cgroup1_procs_write(of, buf, nbytes, off, true);
536 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
537 char *buf, size_t nbytes, loff_t off)
539 return __cgroup1_procs_write(of, buf, nbytes, off, false);
542 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
543 char *buf, size_t nbytes, loff_t off)
547 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
549 cgrp = cgroup_kn_lock_live(of->kn, false);
552 spin_lock(&release_agent_path_lock);
553 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
554 sizeof(cgrp->root->release_agent_path));
555 spin_unlock(&release_agent_path_lock);
556 cgroup_kn_unlock(of->kn);
560 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
562 struct cgroup *cgrp = seq_css(seq)->cgroup;
564 spin_lock(&release_agent_path_lock);
565 seq_puts(seq, cgrp->root->release_agent_path);
566 spin_unlock(&release_agent_path_lock);
571 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
573 seq_puts(seq, "0\n");
577 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
580 return notify_on_release(css->cgroup);
583 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
584 struct cftype *cft, u64 val)
587 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
589 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
593 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
596 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
599 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
600 struct cftype *cft, u64 val)
603 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
605 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
609 /* cgroup core interface files for the legacy hierarchies */
610 struct cftype cgroup1_base_files[] = {
612 .name = "cgroup.procs",
613 .seq_start = cgroup_pidlist_start,
614 .seq_next = cgroup_pidlist_next,
615 .seq_stop = cgroup_pidlist_stop,
616 .seq_show = cgroup_pidlist_show,
617 .private = CGROUP_FILE_PROCS,
618 .write = cgroup1_procs_write,
621 .name = "cgroup.clone_children",
622 .read_u64 = cgroup_clone_children_read,
623 .write_u64 = cgroup_clone_children_write,
626 .name = "cgroup.sane_behavior",
627 .flags = CFTYPE_ONLY_ON_ROOT,
628 .seq_show = cgroup_sane_behavior_show,
632 .seq_start = cgroup_pidlist_start,
633 .seq_next = cgroup_pidlist_next,
634 .seq_stop = cgroup_pidlist_stop,
635 .seq_show = cgroup_pidlist_show,
636 .private = CGROUP_FILE_TASKS,
637 .write = cgroup1_tasks_write,
640 .name = "notify_on_release",
641 .read_u64 = cgroup_read_notify_on_release,
642 .write_u64 = cgroup_write_notify_on_release,
645 .name = "release_agent",
646 .flags = CFTYPE_ONLY_ON_ROOT,
647 .seq_show = cgroup_release_agent_show,
648 .write = cgroup_release_agent_write,
649 .max_write_len = PATH_MAX - 1,
654 /* Display information about each subsystem and each hierarchy */
655 int proc_cgroupstats_show(struct seq_file *m, void *v)
657 struct cgroup_subsys *ss;
660 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
662 * Grab the subsystems state racily. No need to add avenue to
663 * cgroup_mutex contention.
666 for_each_subsys(ss, i)
667 seq_printf(m, "%s\t%d\t%d\t%d\n",
668 ss->legacy_name, ss->root->hierarchy_id,
669 atomic_read(&ss->root->nr_cgrps),
670 cgroup_ssid_enabled(i));
676 * cgroupstats_build - build and fill cgroupstats
677 * @stats: cgroupstats to fill information into
678 * @dentry: A dentry entry belonging to the cgroup for which stats have
681 * Build and fill cgroupstats so that taskstats can export it to user
684 * Return: %0 on success or a negative errno code on failure
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)
699 * We aren't being called from kernfs and there's no guarantee on
700 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
701 * @kn->priv is RCU safe. Let's do the RCU dancing.
704 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
705 if (!cgrp || !cgroup_tryget(cgrp)) {
711 css_task_iter_start(&cgrp->self, 0, &it);
712 while ((tsk = css_task_iter_next(&it))) {
713 switch (READ_ONCE(tsk->__state)) {
717 case TASK_INTERRUPTIBLE:
718 stats->nr_sleeping++;
720 case TASK_UNINTERRUPTIBLE:
721 stats->nr_uninterruptible++;
732 css_task_iter_end(&it);
738 void cgroup1_check_for_release(struct cgroup *cgrp)
740 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
741 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
742 schedule_work(&cgrp->release_agent_work);
746 * Notify userspace when a cgroup is released, by running the
747 * configured release agent with the name of the cgroup (path
748 * relative to the root of cgroup file system) as the argument.
750 * Most likely, this user command will try to rmdir this cgroup.
752 * This races with the possibility that some other task will be
753 * attached to this cgroup before it is removed, or that some other
754 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
755 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
756 * unused, and this cgroup will be reprieved from its death sentence,
757 * to continue to serve a useful existence. Next time it's released,
758 * we will get notified again, if it still has 'notify_on_release' set.
760 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
761 * means only wait until the task is successfully execve()'d. The
762 * separate release agent task is forked by call_usermodehelper(),
763 * then control in this thread returns here, without waiting for the
764 * release agent task. We don't bother to wait because the caller of
765 * this routine has no use for the exit status of the release agent
766 * task, so no sense holding our caller up for that.
768 void cgroup1_release_agent(struct work_struct *work)
770 struct cgroup *cgrp =
771 container_of(work, struct cgroup, release_agent_work);
772 char *pathbuf, *agentbuf;
773 char *argv[3], *envp[3];
776 /* snoop agent path and exit early if empty */
777 if (!cgrp->root->release_agent_path[0])
780 /* prepare argument buffers */
781 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
782 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
783 if (!pathbuf || !agentbuf)
786 spin_lock(&release_agent_path_lock);
787 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
788 spin_unlock(&release_agent_path_lock);
792 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
793 if (ret < 0 || ret >= PATH_MAX)
800 /* minimal command environment */
802 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
805 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
812 * cgroup_rename - Only allow simple rename of directories in place.
814 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
815 const char *new_name_str)
817 struct cgroup *cgrp = kn->priv;
820 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
821 if (strchr(new_name_str, '\n'))
824 if (kernfs_type(kn) != KERNFS_DIR)
826 if (kn->parent != new_parent)
830 * We're gonna grab cgroup_mutex which nests outside kernfs
831 * active_ref. kernfs_rename() doesn't require active_ref
832 * protection. Break them before grabbing cgroup_mutex.
834 kernfs_break_active_protection(new_parent);
835 kernfs_break_active_protection(kn);
837 mutex_lock(&cgroup_mutex);
839 ret = kernfs_rename(kn, new_parent, new_name_str);
841 TRACE_CGROUP_PATH(rename, cgrp);
843 mutex_unlock(&cgroup_mutex);
845 kernfs_unbreak_active_protection(kn);
846 kernfs_unbreak_active_protection(new_parent);
850 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
852 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
853 struct cgroup_subsys *ss;
856 for_each_subsys(ss, ssid)
857 if (root->subsys_mask & (1 << ssid))
858 seq_show_option(seq, ss->legacy_name, NULL);
859 if (root->flags & CGRP_ROOT_NOPREFIX)
860 seq_puts(seq, ",noprefix");
861 if (root->flags & CGRP_ROOT_XATTR)
862 seq_puts(seq, ",xattr");
863 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
864 seq_puts(seq, ",cpuset_v2_mode");
866 spin_lock(&release_agent_path_lock);
867 if (strlen(root->release_agent_path))
868 seq_show_option(seq, "release_agent",
869 root->release_agent_path);
870 spin_unlock(&release_agent_path_lock);
872 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
873 seq_puts(seq, ",clone_children");
874 if (strlen(root->name))
875 seq_show_option(seq, "name", root->name);
890 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
891 fsparam_flag ("all", Opt_all),
892 fsparam_flag ("clone_children", Opt_clone_children),
893 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
894 fsparam_string("name", Opt_name),
895 fsparam_flag ("none", Opt_none),
896 fsparam_flag ("noprefix", Opt_noprefix),
897 fsparam_string("release_agent", Opt_release_agent),
898 fsparam_flag ("xattr", Opt_xattr),
902 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
904 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
905 struct cgroup_subsys *ss;
906 struct fs_parse_result result;
909 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
910 if (opt == -ENOPARAM) {
913 ret = vfs_parse_fs_param_source(fc, param);
914 if (ret != -ENOPARAM)
916 for_each_subsys(ss, i) {
917 if (strcmp(param->key, ss->legacy_name))
919 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
920 return invalfc(fc, "Disabled controller '%s'",
922 ctx->subsys_mask |= (1 << i);
925 return invalfc(fc, "Unknown subsys name '%s'", param->key);
932 /* Explicitly have no subsystems */
939 ctx->flags |= CGRP_ROOT_NOPREFIX;
941 case Opt_clone_children:
942 ctx->cpuset_clone_children = true;
944 case Opt_cpuset_v2_mode:
945 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
948 ctx->flags |= CGRP_ROOT_XATTR;
950 case Opt_release_agent:
951 /* Specifying two release agents is forbidden */
952 if (ctx->release_agent)
953 return invalfc(fc, "release_agent respecified");
954 ctx->release_agent = param->string;
955 param->string = NULL;
958 /* blocked by boot param? */
959 if (cgroup_no_v1_named)
961 /* Can't specify an empty name */
963 return invalfc(fc, "Empty name");
964 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
965 return invalfc(fc, "Name too long");
966 /* Must match [\w.-]+ */
967 for (i = 0; i < param->size; i++) {
968 char c = param->string[i];
971 if ((c == '.') || (c == '-') || (c == '_'))
973 return invalfc(fc, "Invalid name");
975 /* Specifying two names is forbidden */
977 return invalfc(fc, "name respecified");
978 ctx->name = param->string;
979 param->string = NULL;
985 static int check_cgroupfs_options(struct fs_context *fc)
987 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
990 struct cgroup_subsys *ss;
993 #ifdef CONFIG_CPUSETS
994 mask = ~((u16)1 << cpuset_cgrp_id);
996 for_each_subsys(ss, i)
997 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1000 ctx->subsys_mask &= enabled;
1003 * In absence of 'none', 'name=' and subsystem name options,
1004 * let's default to 'all'.
1006 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1010 /* Mutually exclusive option 'all' + subsystem name */
1011 if (ctx->subsys_mask)
1012 return invalfc(fc, "subsys name conflicts with all");
1013 /* 'all' => select all the subsystems */
1014 ctx->subsys_mask = enabled;
1018 * We either have to specify by name or by subsystems. (So all
1019 * empty hierarchies must have a name).
1021 if (!ctx->subsys_mask && !ctx->name)
1022 return invalfc(fc, "Need name or subsystem set");
1025 * Option noprefix was introduced just for backward compatibility
1026 * with the old cpuset, so we allow noprefix only if mounting just
1027 * the cpuset subsystem.
1029 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1030 return invalfc(fc, "noprefix used incorrectly");
1032 /* Can't specify "none" and some subsystems */
1033 if (ctx->subsys_mask && ctx->none)
1034 return invalfc(fc, "none used incorrectly");
1039 int cgroup1_reconfigure(struct fs_context *fc)
1041 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1042 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1043 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1045 u16 added_mask, removed_mask;
1047 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1049 /* See what subsystems are wanted */
1050 ret = check_cgroupfs_options(fc);
1054 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1055 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1056 task_tgid_nr(current), current->comm);
1058 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1059 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1061 /* Don't allow flags or name to change at remount */
1062 if ((ctx->flags ^ root->flags) ||
1063 (ctx->name && strcmp(ctx->name, root->name))) {
1064 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1065 ctx->flags, ctx->name ?: "", root->flags, root->name);
1070 /* remounting is not allowed for populated hierarchies */
1071 if (!list_empty(&root->cgrp.self.children)) {
1076 ret = rebind_subsystems(root, added_mask);
1080 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1082 if (ctx->release_agent) {
1083 spin_lock(&release_agent_path_lock);
1084 strcpy(root->release_agent_path, ctx->release_agent);
1085 spin_unlock(&release_agent_path_lock);
1088 trace_cgroup_remount(root);
1091 mutex_unlock(&cgroup_mutex);
1095 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1096 .rename = cgroup1_rename,
1097 .show_options = cgroup1_show_options,
1098 .mkdir = cgroup_mkdir,
1099 .rmdir = cgroup_rmdir,
1100 .show_path = cgroup_show_path,
1104 * The guts of cgroup1 mount - find or create cgroup_root to use.
1105 * Called with cgroup_mutex held; returns 0 on success, -E... on
1106 * error and positive - in case when the candidate is busy dying.
1107 * On success it stashes a reference to cgroup_root into given
1108 * cgroup_fs_context; that reference is *NOT* counting towards the
1109 * cgroup_root refcount.
1111 static int cgroup1_root_to_use(struct fs_context *fc)
1113 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1114 struct cgroup_root *root;
1115 struct cgroup_subsys *ss;
1118 /* First find the desired set of subsystems */
1119 ret = check_cgroupfs_options(fc);
1124 * Destruction of cgroup root is asynchronous, so subsystems may
1125 * still be dying after the previous unmount. Let's drain the
1126 * dying subsystems. We just need to ensure that the ones
1127 * unmounted previously finish dying and don't care about new ones
1128 * starting. Testing ref liveliness is good enough.
1130 for_each_subsys(ss, i) {
1131 if (!(ctx->subsys_mask & (1 << i)) ||
1132 ss->root == &cgrp_dfl_root)
1135 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1136 return 1; /* restart */
1137 cgroup_put(&ss->root->cgrp);
1140 for_each_root(root) {
1141 bool name_match = false;
1143 if (root == &cgrp_dfl_root)
1147 * If we asked for a name then it must match. Also, if
1148 * name matches but sybsys_mask doesn't, we should fail.
1149 * Remember whether name matched.
1152 if (strcmp(ctx->name, root->name))
1158 * If we asked for subsystems (or explicitly for no
1159 * subsystems) then they must match.
1161 if ((ctx->subsys_mask || ctx->none) &&
1162 (ctx->subsys_mask != root->subsys_mask)) {
1168 if (root->flags ^ ctx->flags)
1169 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1176 * No such thing, create a new one. name= matching without subsys
1177 * specification is allowed for already existing hierarchies but we
1178 * can't create new one without subsys specification.
1180 if (!ctx->subsys_mask && !ctx->none)
1181 return invalfc(fc, "No subsys list or none specified");
1183 /* Hierarchies may only be created in the initial cgroup namespace. */
1184 if (ctx->ns != &init_cgroup_ns)
1187 root = kzalloc(sizeof(*root), GFP_KERNEL);
1192 init_cgroup_root(ctx);
1194 ret = cgroup_setup_root(root, ctx->subsys_mask);
1196 cgroup_free_root(root);
1200 int cgroup1_get_tree(struct fs_context *fc)
1202 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1205 /* Check if the caller has permission to mount. */
1206 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1209 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1211 ret = cgroup1_root_to_use(fc);
1212 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1213 ret = 1; /* restart */
1215 mutex_unlock(&cgroup_mutex);
1218 ret = cgroup_do_get_tree(fc);
1220 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1225 if (unlikely(ret > 0)) {
1227 return restart_syscall();
1232 static int __init cgroup1_wq_init(void)
1235 * Used to destroy pidlists and separate to serve as flush domain.
1236 * Cap @max_active to 1 too.
1238 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1240 BUG_ON(!cgroup_pidlist_destroy_wq);
1243 core_initcall(cgroup1_wq_init);
1245 static int __init cgroup_no_v1(char *str)
1247 struct cgroup_subsys *ss;
1251 while ((token = strsep(&str, ",")) != NULL) {
1255 if (!strcmp(token, "all")) {
1256 cgroup_no_v1_mask = U16_MAX;
1260 if (!strcmp(token, "named")) {
1261 cgroup_no_v1_named = true;
1265 for_each_subsys(ss, i) {
1266 if (strcmp(token, ss->name) &&
1267 strcmp(token, ss->legacy_name))
1270 cgroup_no_v1_mask |= 1 << i;
1275 __setup("cgroup_no_v1=", cgroup_no_v1);