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);
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);
80 mutex_unlock(&cgroup_mutex);
84 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
87 * cgroup_transfer_tasks - move tasks from one cgroup to another
88 * @to: cgroup to which the tasks will be moved
89 * @from: cgroup in which the tasks currently reside
91 * Locking rules between cgroup_post_fork() and the migration path
92 * guarantee that, if a task is forking while being migrated, the new child
93 * is guaranteed to be either visible in the source cgroup after the
94 * parent's migration is complete or put into the target cgroup. No task
95 * can slip out of migration through forking.
97 * Return: %0 on success or a negative errno code on failure
99 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
101 DEFINE_CGROUP_MGCTX(mgctx);
102 struct cgrp_cset_link *link;
103 struct css_task_iter it;
104 struct task_struct *task;
107 if (cgroup_on_dfl(to))
110 ret = cgroup_migrate_vet_dst(to);
114 mutex_lock(&cgroup_mutex);
116 percpu_down_write(&cgroup_threadgroup_rwsem);
118 /* all tasks in @from are being moved, all csets are source */
119 spin_lock_irq(&css_set_lock);
120 list_for_each_entry(link, &from->cset_links, cset_link)
121 cgroup_migrate_add_src(link->cset, to, &mgctx);
122 spin_unlock_irq(&css_set_lock);
124 ret = cgroup_migrate_prepare_dst(&mgctx);
129 * Migrate tasks one-by-one until @from is empty. This fails iff
130 * ->can_attach() fails.
133 css_task_iter_start(&from->self, 0, &it);
136 task = css_task_iter_next(&it);
137 } while (task && (task->flags & PF_EXITING));
140 get_task_struct(task);
141 css_task_iter_end(&it);
144 ret = cgroup_migrate(task, false, &mgctx);
146 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
147 put_task_struct(task);
149 } while (task && !ret);
151 cgroup_migrate_finish(&mgctx);
152 percpu_up_write(&cgroup_threadgroup_rwsem);
153 mutex_unlock(&cgroup_mutex);
158 * Stuff for reading the 'tasks'/'procs' files.
160 * Reading this file can return large amounts of data if a cgroup has
161 * *lots* of attached tasks. So it may need several calls to read(),
162 * but we cannot guarantee that the information we produce is correct
163 * unless we produce it entirely atomically.
167 /* which pidlist file are we talking about? */
168 enum cgroup_filetype {
174 * A pidlist is a list of pids that virtually represents the contents of one
175 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
176 * a pair (one each for procs, tasks) for each pid namespace that's relevant
179 struct cgroup_pidlist {
181 * used to find which pidlist is wanted. doesn't change as long as
182 * this particular list stays in the list.
184 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
187 /* how many elements the above list has */
189 /* each of these stored in a list by its cgroup */
190 struct list_head links;
191 /* pointer to the cgroup we belong to, for list removal purposes */
192 struct cgroup *owner;
193 /* for delayed destruction */
194 struct delayed_work destroy_dwork;
198 * Used to destroy all pidlists lingering waiting for destroy timer. None
199 * should be left afterwards.
201 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
203 struct cgroup_pidlist *l, *tmp_l;
205 mutex_lock(&cgrp->pidlist_mutex);
206 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
207 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
208 mutex_unlock(&cgrp->pidlist_mutex);
210 flush_workqueue(cgroup_pidlist_destroy_wq);
211 BUG_ON(!list_empty(&cgrp->pidlists));
214 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
216 struct delayed_work *dwork = to_delayed_work(work);
217 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
219 struct cgroup_pidlist *tofree = NULL;
221 mutex_lock(&l->owner->pidlist_mutex);
224 * Destroy iff we didn't get queued again. The state won't change
225 * as destroy_dwork can only be queued while locked.
227 if (!delayed_work_pending(dwork)) {
230 put_pid_ns(l->key.ns);
234 mutex_unlock(&l->owner->pidlist_mutex);
239 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
240 * Returns the number of unique elements.
242 static int pidlist_uniq(pid_t *list, int length)
247 * we presume the 0th element is unique, so i starts at 1. trivial
248 * edge cases first; no work needs to be done for either
250 if (length == 0 || length == 1)
252 /* src and dest walk down the list; dest counts unique elements */
253 for (src = 1; src < length; src++) {
254 /* find next unique element */
255 while (list[src] == list[src-1]) {
260 /* dest always points to where the next unique element goes */
261 list[dest] = list[src];
269 * The two pid files - task and cgroup.procs - guaranteed that the result
270 * is sorted, which forced this whole pidlist fiasco. As pid order is
271 * different per namespace, each namespace needs differently sorted list,
272 * making it impossible to use, for example, single rbtree of member tasks
273 * sorted by task pointer. As pidlists can be fairly large, allocating one
274 * per open file is dangerous, so cgroup had to implement shared pool of
275 * pidlists keyed by cgroup and namespace.
277 static int cmppid(const void *a, const void *b)
279 return *(pid_t *)a - *(pid_t *)b;
282 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
283 enum cgroup_filetype type)
285 struct cgroup_pidlist *l;
286 /* don't need task_nsproxy() if we're looking at ourself */
287 struct pid_namespace *ns = task_active_pid_ns(current);
289 lockdep_assert_held(&cgrp->pidlist_mutex);
291 list_for_each_entry(l, &cgrp->pidlists, links)
292 if (l->key.type == type && l->key.ns == ns)
298 * find the appropriate pidlist for our purpose (given procs vs tasks)
299 * returns with the lock on that pidlist already held, and takes care
300 * of the use count, or returns NULL with no locks held if we're out of
303 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
304 enum cgroup_filetype type)
306 struct cgroup_pidlist *l;
308 lockdep_assert_held(&cgrp->pidlist_mutex);
310 l = cgroup_pidlist_find(cgrp, type);
314 /* entry not found; create a new one */
315 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
319 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
321 /* don't need task_nsproxy() if we're looking at ourself */
322 l->key.ns = get_pid_ns(task_active_pid_ns(current));
324 list_add(&l->links, &cgrp->pidlists);
329 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
331 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
332 struct cgroup_pidlist **lp)
336 int pid, n = 0; /* used for populating the array */
337 struct css_task_iter it;
338 struct task_struct *tsk;
339 struct cgroup_pidlist *l;
341 lockdep_assert_held(&cgrp->pidlist_mutex);
344 * If cgroup gets more users after we read count, we won't have
345 * enough space - tough. This race is indistinguishable to the
346 * caller from the case that the additional cgroup users didn't
347 * show up until sometime later on.
349 length = cgroup_task_count(cgrp);
350 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
353 /* now, populate the array */
354 css_task_iter_start(&cgrp->self, 0, &it);
355 while ((tsk = css_task_iter_next(&it))) {
356 if (unlikely(n == length))
358 /* get tgid or pid for procs or tasks file respectively */
359 if (type == CGROUP_FILE_PROCS)
360 pid = task_tgid_vnr(tsk);
362 pid = task_pid_vnr(tsk);
363 if (pid > 0) /* make sure to only use valid results */
366 css_task_iter_end(&it);
368 /* now sort & (if procs) strip out duplicates */
369 sort(array, length, sizeof(pid_t), cmppid, NULL);
370 if (type == CGROUP_FILE_PROCS)
371 length = pidlist_uniq(array, length);
373 l = cgroup_pidlist_find_create(cgrp, type);
379 /* store array, freeing old if necessary */
388 * seq_file methods for the tasks/procs files. The seq_file position is the
389 * next pid to display; the seq_file iterator is a pointer to the pid
390 * in the cgroup->l->list array.
393 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
396 * Initially we receive a position value that corresponds to
397 * one more than the last pid shown (or 0 on the first call or
398 * after a seek to the start). Use a binary-search to find the
399 * next pid to display, if any
401 struct kernfs_open_file *of = s->private;
402 struct cgroup_file_ctx *ctx = of->priv;
403 struct cgroup *cgrp = seq_css(s)->cgroup;
404 struct cgroup_pidlist *l;
405 enum cgroup_filetype type = seq_cft(s)->private;
406 int index = 0, pid = *pos;
409 mutex_lock(&cgrp->pidlist_mutex);
412 * !NULL @ctx->procs1.pidlist indicates that this isn't the first
413 * start() after open. If the matching pidlist is around, we can use
414 * that. Look for it. Note that @ctx->procs1.pidlist can't be used
415 * directly. It could already have been destroyed.
417 if (ctx->procs1.pidlist)
418 ctx->procs1.pidlist = cgroup_pidlist_find(cgrp, type);
421 * Either this is the first start() after open or the matching
422 * pidlist has been destroyed inbetween. Create a new one.
424 if (!ctx->procs1.pidlist) {
425 ret = pidlist_array_load(cgrp, type, &ctx->procs1.pidlist);
429 l = ctx->procs1.pidlist;
434 while (index < end) {
435 int mid = (index + end) / 2;
436 if (l->list[mid] == pid) {
439 } else if (l->list[mid] <= pid)
445 /* If we're off the end of the array, we're done */
446 if (index >= l->length)
448 /* Update the abstract position to be the actual pid that we found */
449 iter = l->list + index;
454 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
456 struct kernfs_open_file *of = s->private;
457 struct cgroup_file_ctx *ctx = of->priv;
458 struct cgroup_pidlist *l = ctx->procs1.pidlist;
461 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
462 CGROUP_PIDLIST_DESTROY_DELAY);
463 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
466 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
468 struct kernfs_open_file *of = s->private;
469 struct cgroup_file_ctx *ctx = of->priv;
470 struct cgroup_pidlist *l = ctx->procs1.pidlist;
472 pid_t *end = l->list + l->length;
474 * Advance to the next pid in the array. If this goes off the
487 static int cgroup_pidlist_show(struct seq_file *s, void *v)
489 seq_printf(s, "%d\n", *(int *)v);
494 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
495 char *buf, size_t nbytes, loff_t off,
499 struct task_struct *task;
500 const struct cred *cred, *tcred;
504 cgrp = cgroup_kn_lock_live(of->kn, false);
508 task = cgroup_procs_write_start(buf, threadgroup, &locked);
509 ret = PTR_ERR_OR_ZERO(task);
514 * Even if we're attaching all tasks in the thread group, we only need
515 * to check permissions on one of them. Check permissions using the
516 * credentials from file open to protect against inherited fd attacks.
518 cred = of->file->f_cred;
519 tcred = get_task_cred(task);
520 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
521 !uid_eq(cred->euid, tcred->uid) &&
522 !uid_eq(cred->euid, tcred->suid))
528 ret = cgroup_attach_task(cgrp, task, threadgroup);
531 cgroup_procs_write_finish(task, locked);
533 cgroup_kn_unlock(of->kn);
535 return ret ?: nbytes;
538 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
539 char *buf, size_t nbytes, loff_t off)
541 return __cgroup1_procs_write(of, buf, nbytes, off, true);
544 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
545 char *buf, size_t nbytes, loff_t off)
547 return __cgroup1_procs_write(of, buf, nbytes, off, false);
550 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
551 char *buf, size_t nbytes, loff_t off)
554 struct cgroup_file_ctx *ctx;
556 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
559 * Release agent gets called with all capabilities,
560 * require capabilities to set release agent.
563 if ((ctx->ns->user_ns != &init_user_ns) ||
564 !file_ns_capable(of->file, &init_user_ns, CAP_SYS_ADMIN))
567 cgrp = cgroup_kn_lock_live(of->kn, false);
570 spin_lock(&release_agent_path_lock);
571 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
572 sizeof(cgrp->root->release_agent_path));
573 spin_unlock(&release_agent_path_lock);
574 cgroup_kn_unlock(of->kn);
578 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
580 struct cgroup *cgrp = seq_css(seq)->cgroup;
582 spin_lock(&release_agent_path_lock);
583 seq_puts(seq, cgrp->root->release_agent_path);
584 spin_unlock(&release_agent_path_lock);
589 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
591 seq_puts(seq, "0\n");
595 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
598 return notify_on_release(css->cgroup);
601 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
602 struct cftype *cft, u64 val)
605 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
607 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
611 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
614 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
617 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
618 struct cftype *cft, u64 val)
621 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
623 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
627 /* cgroup core interface files for the legacy hierarchies */
628 struct cftype cgroup1_base_files[] = {
630 .name = "cgroup.procs",
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_PROCS,
636 .write = cgroup1_procs_write,
639 .name = "cgroup.clone_children",
640 .read_u64 = cgroup_clone_children_read,
641 .write_u64 = cgroup_clone_children_write,
644 .name = "cgroup.sane_behavior",
645 .flags = CFTYPE_ONLY_ON_ROOT,
646 .seq_show = cgroup_sane_behavior_show,
650 .seq_start = cgroup_pidlist_start,
651 .seq_next = cgroup_pidlist_next,
652 .seq_stop = cgroup_pidlist_stop,
653 .seq_show = cgroup_pidlist_show,
654 .private = CGROUP_FILE_TASKS,
655 .write = cgroup1_tasks_write,
658 .name = "notify_on_release",
659 .read_u64 = cgroup_read_notify_on_release,
660 .write_u64 = cgroup_write_notify_on_release,
663 .name = "release_agent",
664 .flags = CFTYPE_ONLY_ON_ROOT,
665 .seq_show = cgroup_release_agent_show,
666 .write = cgroup_release_agent_write,
667 .max_write_len = PATH_MAX - 1,
672 /* Display information about each subsystem and each hierarchy */
673 int proc_cgroupstats_show(struct seq_file *m, void *v)
675 struct cgroup_subsys *ss;
678 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
680 * ideally we don't want subsystems moving around while we do this.
681 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
682 * subsys/hierarchy state.
684 mutex_lock(&cgroup_mutex);
686 for_each_subsys(ss, i)
687 seq_printf(m, "%s\t%d\t%d\t%d\n",
688 ss->legacy_name, ss->root->hierarchy_id,
689 atomic_read(&ss->root->nr_cgrps),
690 cgroup_ssid_enabled(i));
692 mutex_unlock(&cgroup_mutex);
697 * cgroupstats_build - build and fill cgroupstats
698 * @stats: cgroupstats to fill information into
699 * @dentry: A dentry entry belonging to the cgroup for which stats have
702 * Build and fill cgroupstats so that taskstats can export it to user
705 * Return: %0 on success or a negative errno code on failure
707 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
709 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
711 struct css_task_iter it;
712 struct task_struct *tsk;
714 /* it should be kernfs_node belonging to cgroupfs and is a directory */
715 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
716 kernfs_type(kn) != KERNFS_DIR)
719 mutex_lock(&cgroup_mutex);
722 * We aren't being called from kernfs and there's no guarantee on
723 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
724 * @kn->priv is RCU safe. Let's do the RCU dancing.
727 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
728 if (!cgrp || cgroup_is_dead(cgrp)) {
730 mutex_unlock(&cgroup_mutex);
735 css_task_iter_start(&cgrp->self, 0, &it);
736 while ((tsk = css_task_iter_next(&it))) {
737 switch (READ_ONCE(tsk->__state)) {
741 case TASK_INTERRUPTIBLE:
742 stats->nr_sleeping++;
744 case TASK_UNINTERRUPTIBLE:
745 stats->nr_uninterruptible++;
756 css_task_iter_end(&it);
758 mutex_unlock(&cgroup_mutex);
762 void cgroup1_check_for_release(struct cgroup *cgrp)
764 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
765 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
766 schedule_work(&cgrp->release_agent_work);
770 * Notify userspace when a cgroup is released, by running the
771 * configured release agent with the name of the cgroup (path
772 * relative to the root of cgroup file system) as the argument.
774 * Most likely, this user command will try to rmdir this cgroup.
776 * This races with the possibility that some other task will be
777 * attached to this cgroup before it is removed, or that some other
778 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
779 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
780 * unused, and this cgroup will be reprieved from its death sentence,
781 * to continue to serve a useful existence. Next time it's released,
782 * we will get notified again, if it still has 'notify_on_release' set.
784 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
785 * means only wait until the task is successfully execve()'d. The
786 * separate release agent task is forked by call_usermodehelper(),
787 * then control in this thread returns here, without waiting for the
788 * release agent task. We don't bother to wait because the caller of
789 * this routine has no use for the exit status of the release agent
790 * task, so no sense holding our caller up for that.
792 void cgroup1_release_agent(struct work_struct *work)
794 struct cgroup *cgrp =
795 container_of(work, struct cgroup, release_agent_work);
796 char *pathbuf, *agentbuf;
797 char *argv[3], *envp[3];
800 /* snoop agent path and exit early if empty */
801 if (!cgrp->root->release_agent_path[0])
804 /* prepare argument buffers */
805 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
806 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
807 if (!pathbuf || !agentbuf)
810 spin_lock(&release_agent_path_lock);
811 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
812 spin_unlock(&release_agent_path_lock);
816 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
817 if (ret < 0 || ret >= PATH_MAX)
824 /* minimal command environment */
826 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
829 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
836 * cgroup_rename - Only allow simple rename of directories in place.
838 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
839 const char *new_name_str)
841 struct cgroup *cgrp = kn->priv;
844 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
845 if (strchr(new_name_str, '\n'))
848 if (kernfs_type(kn) != KERNFS_DIR)
850 if (kn->parent != new_parent)
854 * We're gonna grab cgroup_mutex which nests outside kernfs
855 * active_ref. kernfs_rename() doesn't require active_ref
856 * protection. Break them before grabbing cgroup_mutex.
858 kernfs_break_active_protection(new_parent);
859 kernfs_break_active_protection(kn);
861 mutex_lock(&cgroup_mutex);
863 ret = kernfs_rename(kn, new_parent, new_name_str);
865 TRACE_CGROUP_PATH(rename, cgrp);
867 mutex_unlock(&cgroup_mutex);
869 kernfs_unbreak_active_protection(kn);
870 kernfs_unbreak_active_protection(new_parent);
874 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
876 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
877 struct cgroup_subsys *ss;
880 for_each_subsys(ss, ssid)
881 if (root->subsys_mask & (1 << ssid))
882 seq_show_option(seq, ss->legacy_name, NULL);
883 if (root->flags & CGRP_ROOT_NOPREFIX)
884 seq_puts(seq, ",noprefix");
885 if (root->flags & CGRP_ROOT_XATTR)
886 seq_puts(seq, ",xattr");
887 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
888 seq_puts(seq, ",cpuset_v2_mode");
890 spin_lock(&release_agent_path_lock);
891 if (strlen(root->release_agent_path))
892 seq_show_option(seq, "release_agent",
893 root->release_agent_path);
894 spin_unlock(&release_agent_path_lock);
896 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
897 seq_puts(seq, ",clone_children");
898 if (strlen(root->name))
899 seq_show_option(seq, "name", root->name);
914 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
915 fsparam_flag ("all", Opt_all),
916 fsparam_flag ("clone_children", Opt_clone_children),
917 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
918 fsparam_string("name", Opt_name),
919 fsparam_flag ("none", Opt_none),
920 fsparam_flag ("noprefix", Opt_noprefix),
921 fsparam_string("release_agent", Opt_release_agent),
922 fsparam_flag ("xattr", Opt_xattr),
926 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
928 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
929 struct cgroup_subsys *ss;
930 struct fs_parse_result result;
933 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
934 if (opt == -ENOPARAM) {
937 ret = vfs_parse_fs_param_source(fc, param);
938 if (ret != -ENOPARAM)
940 for_each_subsys(ss, i) {
941 if (strcmp(param->key, ss->legacy_name))
943 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
944 return invalfc(fc, "Disabled controller '%s'",
946 ctx->subsys_mask |= (1 << i);
949 return invalfc(fc, "Unknown subsys name '%s'", param->key);
956 /* Explicitly have no subsystems */
963 ctx->flags |= CGRP_ROOT_NOPREFIX;
965 case Opt_clone_children:
966 ctx->cpuset_clone_children = true;
968 case Opt_cpuset_v2_mode:
969 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
972 ctx->flags |= CGRP_ROOT_XATTR;
974 case Opt_release_agent:
975 /* Specifying two release agents is forbidden */
976 if (ctx->release_agent)
977 return invalfc(fc, "release_agent respecified");
979 * Release agent gets called with all capabilities,
980 * require capabilities to set release agent.
982 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN))
983 return invalfc(fc, "Setting release_agent not allowed");
984 ctx->release_agent = param->string;
985 param->string = NULL;
988 /* blocked by boot param? */
989 if (cgroup_no_v1_named)
991 /* Can't specify an empty name */
993 return invalfc(fc, "Empty name");
994 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
995 return invalfc(fc, "Name too long");
996 /* Must match [\w.-]+ */
997 for (i = 0; i < param->size; i++) {
998 char c = param->string[i];
1001 if ((c == '.') || (c == '-') || (c == '_'))
1003 return invalfc(fc, "Invalid name");
1005 /* Specifying two names is forbidden */
1007 return invalfc(fc, "name respecified");
1008 ctx->name = param->string;
1009 param->string = NULL;
1015 static int check_cgroupfs_options(struct fs_context *fc)
1017 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1020 struct cgroup_subsys *ss;
1023 #ifdef CONFIG_CPUSETS
1024 mask = ~((u16)1 << cpuset_cgrp_id);
1026 for_each_subsys(ss, i)
1027 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1030 ctx->subsys_mask &= enabled;
1033 * In absence of 'none', 'name=' and subsystem name options,
1034 * let's default to 'all'.
1036 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1040 /* Mutually exclusive option 'all' + subsystem name */
1041 if (ctx->subsys_mask)
1042 return invalfc(fc, "subsys name conflicts with all");
1043 /* 'all' => select all the subsystems */
1044 ctx->subsys_mask = enabled;
1048 * We either have to specify by name or by subsystems. (So all
1049 * empty hierarchies must have a name).
1051 if (!ctx->subsys_mask && !ctx->name)
1052 return invalfc(fc, "Need name or subsystem set");
1055 * Option noprefix was introduced just for backward compatibility
1056 * with the old cpuset, so we allow noprefix only if mounting just
1057 * the cpuset subsystem.
1059 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1060 return invalfc(fc, "noprefix used incorrectly");
1062 /* Can't specify "none" and some subsystems */
1063 if (ctx->subsys_mask && ctx->none)
1064 return invalfc(fc, "none used incorrectly");
1069 int cgroup1_reconfigure(struct fs_context *fc)
1071 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1072 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1073 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1075 u16 added_mask, removed_mask;
1077 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1079 /* See what subsystems are wanted */
1080 ret = check_cgroupfs_options(fc);
1084 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1085 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1086 task_tgid_nr(current), current->comm);
1088 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1089 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1091 /* Don't allow flags or name to change at remount */
1092 if ((ctx->flags ^ root->flags) ||
1093 (ctx->name && strcmp(ctx->name, root->name))) {
1094 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1095 ctx->flags, ctx->name ?: "", root->flags, root->name);
1100 /* remounting is not allowed for populated hierarchies */
1101 if (!list_empty(&root->cgrp.self.children)) {
1106 ret = rebind_subsystems(root, added_mask);
1110 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1112 if (ctx->release_agent) {
1113 spin_lock(&release_agent_path_lock);
1114 strcpy(root->release_agent_path, ctx->release_agent);
1115 spin_unlock(&release_agent_path_lock);
1118 trace_cgroup_remount(root);
1121 mutex_unlock(&cgroup_mutex);
1125 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1126 .rename = cgroup1_rename,
1127 .show_options = cgroup1_show_options,
1128 .mkdir = cgroup_mkdir,
1129 .rmdir = cgroup_rmdir,
1130 .show_path = cgroup_show_path,
1134 * The guts of cgroup1 mount - find or create cgroup_root to use.
1135 * Called with cgroup_mutex held; returns 0 on success, -E... on
1136 * error and positive - in case when the candidate is busy dying.
1137 * On success it stashes a reference to cgroup_root into given
1138 * cgroup_fs_context; that reference is *NOT* counting towards the
1139 * cgroup_root refcount.
1141 static int cgroup1_root_to_use(struct fs_context *fc)
1143 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1144 struct cgroup_root *root;
1145 struct cgroup_subsys *ss;
1148 /* First find the desired set of subsystems */
1149 ret = check_cgroupfs_options(fc);
1154 * Destruction of cgroup root is asynchronous, so subsystems may
1155 * still be dying after the previous unmount. Let's drain the
1156 * dying subsystems. We just need to ensure that the ones
1157 * unmounted previously finish dying and don't care about new ones
1158 * starting. Testing ref liveliness is good enough.
1160 for_each_subsys(ss, i) {
1161 if (!(ctx->subsys_mask & (1 << i)) ||
1162 ss->root == &cgrp_dfl_root)
1165 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1166 return 1; /* restart */
1167 cgroup_put(&ss->root->cgrp);
1170 for_each_root(root) {
1171 bool name_match = false;
1173 if (root == &cgrp_dfl_root)
1177 * If we asked for a name then it must match. Also, if
1178 * name matches but sybsys_mask doesn't, we should fail.
1179 * Remember whether name matched.
1182 if (strcmp(ctx->name, root->name))
1188 * If we asked for subsystems (or explicitly for no
1189 * subsystems) then they must match.
1191 if ((ctx->subsys_mask || ctx->none) &&
1192 (ctx->subsys_mask != root->subsys_mask)) {
1198 if (root->flags ^ ctx->flags)
1199 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1206 * No such thing, create a new one. name= matching without subsys
1207 * specification is allowed for already existing hierarchies but we
1208 * can't create new one without subsys specification.
1210 if (!ctx->subsys_mask && !ctx->none)
1211 return invalfc(fc, "No subsys list or none specified");
1213 /* Hierarchies may only be created in the initial cgroup namespace. */
1214 if (ctx->ns != &init_cgroup_ns)
1217 root = kzalloc(sizeof(*root), GFP_KERNEL);
1222 init_cgroup_root(ctx);
1224 ret = cgroup_setup_root(root, ctx->subsys_mask);
1226 cgroup_free_root(root);
1230 int cgroup1_get_tree(struct fs_context *fc)
1232 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1235 /* Check if the caller has permission to mount. */
1236 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1239 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1241 ret = cgroup1_root_to_use(fc);
1242 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1243 ret = 1; /* restart */
1245 mutex_unlock(&cgroup_mutex);
1248 ret = cgroup_do_get_tree(fc);
1250 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1255 if (unlikely(ret > 0)) {
1257 return restart_syscall();
1262 static int __init cgroup1_wq_init(void)
1265 * Used to destroy pidlists and separate to serve as flush domain.
1266 * Cap @max_active to 1 too.
1268 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1270 BUG_ON(!cgroup_pidlist_destroy_wq);
1273 core_initcall(cgroup1_wq_init);
1275 static int __init cgroup_no_v1(char *str)
1277 struct cgroup_subsys *ss;
1281 while ((token = strsep(&str, ",")) != NULL) {
1285 if (!strcmp(token, "all")) {
1286 cgroup_no_v1_mask = U16_MAX;
1290 if (!strcmp(token, "named")) {
1291 cgroup_no_v1_named = true;
1295 for_each_subsys(ss, i) {
1296 if (strcmp(token, ss->name) &&
1297 strcmp(token, ss->legacy_name))
1300 cgroup_no_v1_mask |= 1 << i;
1305 __setup("cgroup_no_v1=", cgroup_no_v1);