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);
61 percpu_down_write(&cgroup_threadgroup_rwsem);
63 struct cgroup *from_cgrp;
65 if (root == &cgrp_dfl_root)
68 spin_lock_irq(&css_set_lock);
69 from_cgrp = task_cgroup_from_root(from, root);
70 spin_unlock_irq(&css_set_lock);
72 retval = cgroup_attach_task(from_cgrp, tsk, false);
76 percpu_up_write(&cgroup_threadgroup_rwsem);
78 mutex_unlock(&cgroup_mutex);
82 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
85 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
86 * @to: cgroup to which the tasks will be moved
87 * @from: cgroup in which the tasks currently reside
89 * Locking rules between cgroup_post_fork() and the migration path
90 * guarantee that, if a task is forking while being migrated, the new child
91 * is guaranteed to be either visible in the source cgroup after the
92 * parent's migration is complete or put into the target cgroup. No task
93 * can slip out of migration through forking.
95 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
97 DEFINE_CGROUP_MGCTX(mgctx);
98 struct cgrp_cset_link *link;
99 struct css_task_iter it;
100 struct task_struct *task;
103 if (cgroup_on_dfl(to))
106 ret = cgroup_migrate_vet_dst(to);
110 mutex_lock(&cgroup_mutex);
112 percpu_down_write(&cgroup_threadgroup_rwsem);
114 /* all tasks in @from are being moved, all csets are source */
115 spin_lock_irq(&css_set_lock);
116 list_for_each_entry(link, &from->cset_links, cset_link)
117 cgroup_migrate_add_src(link->cset, to, &mgctx);
118 spin_unlock_irq(&css_set_lock);
120 ret = cgroup_migrate_prepare_dst(&mgctx);
125 * Migrate tasks one-by-one until @from is empty. This fails iff
126 * ->can_attach() fails.
129 css_task_iter_start(&from->self, 0, &it);
132 task = css_task_iter_next(&it);
133 } while (task && (task->flags & PF_EXITING));
136 get_task_struct(task);
137 css_task_iter_end(&it);
140 ret = cgroup_migrate(task, false, &mgctx);
142 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
143 put_task_struct(task);
145 } while (task && !ret);
147 cgroup_migrate_finish(&mgctx);
148 percpu_up_write(&cgroup_threadgroup_rwsem);
149 mutex_unlock(&cgroup_mutex);
154 * Stuff for reading the 'tasks'/'procs' files.
156 * Reading this file can return large amounts of data if a cgroup has
157 * *lots* of attached tasks. So it may need several calls to read(),
158 * but we cannot guarantee that the information we produce is correct
159 * unless we produce it entirely atomically.
163 /* which pidlist file are we talking about? */
164 enum cgroup_filetype {
170 * A pidlist is a list of pids that virtually represents the contents of one
171 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
172 * a pair (one each for procs, tasks) for each pid namespace that's relevant
175 struct cgroup_pidlist {
177 * used to find which pidlist is wanted. doesn't change as long as
178 * this particular list stays in the list.
180 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
183 /* how many elements the above list has */
185 /* each of these stored in a list by its cgroup */
186 struct list_head links;
187 /* pointer to the cgroup we belong to, for list removal purposes */
188 struct cgroup *owner;
189 /* for delayed destruction */
190 struct delayed_work destroy_dwork;
194 * Used to destroy all pidlists lingering waiting for destroy timer. None
195 * should be left afterwards.
197 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
199 struct cgroup_pidlist *l, *tmp_l;
201 mutex_lock(&cgrp->pidlist_mutex);
202 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
203 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
204 mutex_unlock(&cgrp->pidlist_mutex);
206 flush_workqueue(cgroup_pidlist_destroy_wq);
207 BUG_ON(!list_empty(&cgrp->pidlists));
210 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
212 struct delayed_work *dwork = to_delayed_work(work);
213 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
215 struct cgroup_pidlist *tofree = NULL;
217 mutex_lock(&l->owner->pidlist_mutex);
220 * Destroy iff we didn't get queued again. The state won't change
221 * as destroy_dwork can only be queued while locked.
223 if (!delayed_work_pending(dwork)) {
226 put_pid_ns(l->key.ns);
230 mutex_unlock(&l->owner->pidlist_mutex);
235 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
236 * Returns the number of unique elements.
238 static int pidlist_uniq(pid_t *list, int length)
243 * we presume the 0th element is unique, so i starts at 1. trivial
244 * edge cases first; no work needs to be done for either
246 if (length == 0 || length == 1)
248 /* src and dest walk down the list; dest counts unique elements */
249 for (src = 1; src < length; src++) {
250 /* find next unique element */
251 while (list[src] == list[src-1]) {
256 /* dest always points to where the next unique element goes */
257 list[dest] = list[src];
265 * The two pid files - task and cgroup.procs - guaranteed that the result
266 * is sorted, which forced this whole pidlist fiasco. As pid order is
267 * different per namespace, each namespace needs differently sorted list,
268 * making it impossible to use, for example, single rbtree of member tasks
269 * sorted by task pointer. As pidlists can be fairly large, allocating one
270 * per open file is dangerous, so cgroup had to implement shared pool of
271 * pidlists keyed by cgroup and namespace.
273 static int cmppid(const void *a, const void *b)
275 return *(pid_t *)a - *(pid_t *)b;
278 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
279 enum cgroup_filetype type)
281 struct cgroup_pidlist *l;
282 /* don't need task_nsproxy() if we're looking at ourself */
283 struct pid_namespace *ns = task_active_pid_ns(current);
285 lockdep_assert_held(&cgrp->pidlist_mutex);
287 list_for_each_entry(l, &cgrp->pidlists, links)
288 if (l->key.type == type && l->key.ns == ns)
294 * find the appropriate pidlist for our purpose (given procs vs tasks)
295 * returns with the lock on that pidlist already held, and takes care
296 * of the use count, or returns NULL with no locks held if we're out of
299 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
300 enum cgroup_filetype type)
302 struct cgroup_pidlist *l;
304 lockdep_assert_held(&cgrp->pidlist_mutex);
306 l = cgroup_pidlist_find(cgrp, type);
310 /* entry not found; create a new one */
311 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
315 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
317 /* don't need task_nsproxy() if we're looking at ourself */
318 l->key.ns = get_pid_ns(task_active_pid_ns(current));
320 list_add(&l->links, &cgrp->pidlists);
325 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
327 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
328 struct cgroup_pidlist **lp)
332 int pid, n = 0; /* used for populating the array */
333 struct css_task_iter it;
334 struct task_struct *tsk;
335 struct cgroup_pidlist *l;
337 lockdep_assert_held(&cgrp->pidlist_mutex);
340 * If cgroup gets more users after we read count, we won't have
341 * enough space - tough. This race is indistinguishable to the
342 * caller from the case that the additional cgroup users didn't
343 * show up until sometime later on.
345 length = cgroup_task_count(cgrp);
346 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
349 /* now, populate the array */
350 css_task_iter_start(&cgrp->self, 0, &it);
351 while ((tsk = css_task_iter_next(&it))) {
352 if (unlikely(n == length))
354 /* get tgid or pid for procs or tasks file respectively */
355 if (type == CGROUP_FILE_PROCS)
356 pid = task_tgid_vnr(tsk);
358 pid = task_pid_vnr(tsk);
359 if (pid > 0) /* make sure to only use valid results */
362 css_task_iter_end(&it);
364 /* now sort & strip out duplicates (tgids or recycled thread PIDs) */
365 sort(array, length, sizeof(pid_t), cmppid, NULL);
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_file_ctx *ctx = of->priv;
398 struct cgroup *cgrp = seq_css(s)->cgroup;
399 struct cgroup_pidlist *l;
400 enum cgroup_filetype type = seq_cft(s)->private;
401 int index = 0, pid = *pos;
404 mutex_lock(&cgrp->pidlist_mutex);
407 * !NULL @ctx->procs1.pidlist indicates that this isn't the first
408 * start() after open. If the matching pidlist is around, we can use
409 * that. Look for it. Note that @ctx->procs1.pidlist can't be used
410 * directly. It could already have been destroyed.
412 if (ctx->procs1.pidlist)
413 ctx->procs1.pidlist = cgroup_pidlist_find(cgrp, type);
416 * Either this is the first start() after open or the matching
417 * pidlist has been destroyed inbetween. Create a new one.
419 if (!ctx->procs1.pidlist) {
420 ret = pidlist_array_load(cgrp, type, &ctx->procs1.pidlist);
424 l = ctx->procs1.pidlist;
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_file_ctx *ctx = of->priv;
453 struct cgroup_pidlist *l = ctx->procs1.pidlist;
456 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
457 CGROUP_PIDLIST_DESTROY_DELAY);
458 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
461 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
463 struct kernfs_open_file *of = s->private;
464 struct cgroup_file_ctx *ctx = of->priv;
465 struct cgroup_pidlist *l = ctx->procs1.pidlist;
467 pid_t *end = l->list + l->length;
469 * Advance to the next pid in the array. If this goes off the
482 static int cgroup_pidlist_show(struct seq_file *s, void *v)
484 seq_printf(s, "%d\n", *(int *)v);
489 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
490 char *buf, size_t nbytes, loff_t off,
494 struct task_struct *task;
495 const struct cred *cred, *tcred;
499 cgrp = cgroup_kn_lock_live(of->kn, false);
503 task = cgroup_procs_write_start(buf, threadgroup, &locked);
504 ret = PTR_ERR_OR_ZERO(task);
509 * Even if we're attaching all tasks in the thread group, we only need
510 * to check permissions on one of them. Check permissions using the
511 * credentials from file open to protect against inherited fd attacks.
513 cred = of->file->f_cred;
514 tcred = get_task_cred(task);
515 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
516 !uid_eq(cred->euid, tcred->uid) &&
517 !uid_eq(cred->euid, tcred->suid))
523 ret = cgroup_attach_task(cgrp, task, threadgroup);
526 cgroup_procs_write_finish(task, locked);
528 cgroup_kn_unlock(of->kn);
530 return ret ?: nbytes;
533 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
534 char *buf, size_t nbytes, loff_t off)
536 return __cgroup1_procs_write(of, buf, nbytes, off, true);
539 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
540 char *buf, size_t nbytes, loff_t off)
542 return __cgroup1_procs_write(of, buf, nbytes, off, false);
545 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
546 char *buf, size_t nbytes, loff_t off)
549 struct cgroup_file_ctx *ctx;
551 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
554 * Release agent gets called with all capabilities,
555 * require capabilities to set release agent.
558 if ((ctx->ns->user_ns != &init_user_ns) ||
559 !file_ns_capable(of->file, &init_user_ns, CAP_SYS_ADMIN))
562 cgrp = cgroup_kn_lock_live(of->kn, false);
565 spin_lock(&release_agent_path_lock);
566 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
567 sizeof(cgrp->root->release_agent_path));
568 spin_unlock(&release_agent_path_lock);
569 cgroup_kn_unlock(of->kn);
573 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
575 struct cgroup *cgrp = seq_css(seq)->cgroup;
577 spin_lock(&release_agent_path_lock);
578 seq_puts(seq, cgrp->root->release_agent_path);
579 spin_unlock(&release_agent_path_lock);
584 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
586 seq_puts(seq, "0\n");
590 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
593 return notify_on_release(css->cgroup);
596 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
597 struct cftype *cft, u64 val)
600 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
602 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
606 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
609 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
612 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
613 struct cftype *cft, u64 val)
616 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
618 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
622 /* cgroup core interface files for the legacy hierarchies */
623 struct cftype cgroup1_base_files[] = {
625 .name = "cgroup.procs",
626 .seq_start = cgroup_pidlist_start,
627 .seq_next = cgroup_pidlist_next,
628 .seq_stop = cgroup_pidlist_stop,
629 .seq_show = cgroup_pidlist_show,
630 .private = CGROUP_FILE_PROCS,
631 .write = cgroup1_procs_write,
634 .name = "cgroup.clone_children",
635 .read_u64 = cgroup_clone_children_read,
636 .write_u64 = cgroup_clone_children_write,
639 .name = "cgroup.sane_behavior",
640 .flags = CFTYPE_ONLY_ON_ROOT,
641 .seq_show = cgroup_sane_behavior_show,
645 .seq_start = cgroup_pidlist_start,
646 .seq_next = cgroup_pidlist_next,
647 .seq_stop = cgroup_pidlist_stop,
648 .seq_show = cgroup_pidlist_show,
649 .private = CGROUP_FILE_TASKS,
650 .write = cgroup1_tasks_write,
653 .name = "notify_on_release",
654 .read_u64 = cgroup_read_notify_on_release,
655 .write_u64 = cgroup_write_notify_on_release,
658 .name = "release_agent",
659 .flags = CFTYPE_ONLY_ON_ROOT,
660 .seq_show = cgroup_release_agent_show,
661 .write = cgroup_release_agent_write,
662 .max_write_len = PATH_MAX - 1,
667 /* Display information about each subsystem and each hierarchy */
668 int proc_cgroupstats_show(struct seq_file *m, void *v)
670 struct cgroup_subsys *ss;
673 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
675 * ideally we don't want subsystems moving around while we do this.
676 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
677 * subsys/hierarchy state.
679 mutex_lock(&cgroup_mutex);
681 for_each_subsys(ss, i)
682 seq_printf(m, "%s\t%d\t%d\t%d\n",
683 ss->legacy_name, ss->root->hierarchy_id,
684 atomic_read(&ss->root->nr_cgrps),
685 cgroup_ssid_enabled(i));
687 mutex_unlock(&cgroup_mutex);
692 * cgroupstats_build - build and fill cgroupstats
693 * @stats: cgroupstats to fill information into
694 * @dentry: A dentry entry belonging to the cgroup for which stats have
697 * Build and fill cgroupstats so that taskstats can export it to user
700 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
702 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
704 struct css_task_iter it;
705 struct task_struct *tsk;
707 /* it should be kernfs_node belonging to cgroupfs and is a directory */
708 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
709 kernfs_type(kn) != KERNFS_DIR)
712 mutex_lock(&cgroup_mutex);
715 * We aren't being called from kernfs and there's no guarantee on
716 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
717 * @kn->priv is RCU safe. Let's do the RCU dancing.
720 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
721 if (!cgrp || cgroup_is_dead(cgrp)) {
723 mutex_unlock(&cgroup_mutex);
728 css_task_iter_start(&cgrp->self, 0, &it);
729 while ((tsk = css_task_iter_next(&it))) {
730 switch (tsk->state) {
734 case TASK_INTERRUPTIBLE:
735 stats->nr_sleeping++;
737 case TASK_UNINTERRUPTIBLE:
738 stats->nr_uninterruptible++;
744 if (delayacct_is_task_waiting_on_io(tsk))
749 css_task_iter_end(&it);
751 mutex_unlock(&cgroup_mutex);
755 void cgroup1_check_for_release(struct cgroup *cgrp)
757 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
758 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
759 schedule_work(&cgrp->release_agent_work);
763 * Notify userspace when a cgroup is released, by running the
764 * configured release agent with the name of the cgroup (path
765 * relative to the root of cgroup file system) as the argument.
767 * Most likely, this user command will try to rmdir this cgroup.
769 * This races with the possibility that some other task will be
770 * attached to this cgroup before it is removed, or that some other
771 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
772 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
773 * unused, and this cgroup will be reprieved from its death sentence,
774 * to continue to serve a useful existence. Next time it's released,
775 * we will get notified again, if it still has 'notify_on_release' set.
777 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
778 * means only wait until the task is successfully execve()'d. The
779 * separate release agent task is forked by call_usermodehelper(),
780 * then control in this thread returns here, without waiting for the
781 * release agent task. We don't bother to wait because the caller of
782 * this routine has no use for the exit status of the release agent
783 * task, so no sense holding our caller up for that.
785 void cgroup1_release_agent(struct work_struct *work)
787 struct cgroup *cgrp =
788 container_of(work, struct cgroup, release_agent_work);
789 char *pathbuf, *agentbuf;
790 char *argv[3], *envp[3];
793 /* snoop agent path and exit early if empty */
794 if (!cgrp->root->release_agent_path[0])
797 /* prepare argument buffers */
798 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
799 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
800 if (!pathbuf || !agentbuf)
803 spin_lock(&release_agent_path_lock);
804 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
805 spin_unlock(&release_agent_path_lock);
809 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
810 if (ret < 0 || ret >= PATH_MAX)
817 /* minimal command environment */
819 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
822 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
829 * cgroup_rename - Only allow simple rename of directories in place.
831 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
832 const char *new_name_str)
834 struct cgroup *cgrp = kn->priv;
837 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
838 if (strchr(new_name_str, '\n'))
841 if (kernfs_type(kn) != KERNFS_DIR)
843 if (kn->parent != new_parent)
847 * We're gonna grab cgroup_mutex which nests outside kernfs
848 * active_ref. kernfs_rename() doesn't require active_ref
849 * protection. Break them before grabbing cgroup_mutex.
851 kernfs_break_active_protection(new_parent);
852 kernfs_break_active_protection(kn);
854 mutex_lock(&cgroup_mutex);
856 ret = kernfs_rename(kn, new_parent, new_name_str);
858 TRACE_CGROUP_PATH(rename, cgrp);
860 mutex_unlock(&cgroup_mutex);
862 kernfs_unbreak_active_protection(kn);
863 kernfs_unbreak_active_protection(new_parent);
867 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
869 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
870 struct cgroup_subsys *ss;
873 for_each_subsys(ss, ssid)
874 if (root->subsys_mask & (1 << ssid))
875 seq_show_option(seq, ss->legacy_name, NULL);
876 if (root->flags & CGRP_ROOT_NOPREFIX)
877 seq_puts(seq, ",noprefix");
878 if (root->flags & CGRP_ROOT_XATTR)
879 seq_puts(seq, ",xattr");
880 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
881 seq_puts(seq, ",cpuset_v2_mode");
883 spin_lock(&release_agent_path_lock);
884 if (strlen(root->release_agent_path))
885 seq_show_option(seq, "release_agent",
886 root->release_agent_path);
887 spin_unlock(&release_agent_path_lock);
889 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
890 seq_puts(seq, ",clone_children");
891 if (strlen(root->name))
892 seq_show_option(seq, "name", root->name);
907 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
908 fsparam_flag ("all", Opt_all),
909 fsparam_flag ("clone_children", Opt_clone_children),
910 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
911 fsparam_string("name", Opt_name),
912 fsparam_flag ("none", Opt_none),
913 fsparam_flag ("noprefix", Opt_noprefix),
914 fsparam_string("release_agent", Opt_release_agent),
915 fsparam_flag ("xattr", Opt_xattr),
919 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
921 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
922 struct cgroup_subsys *ss;
923 struct fs_parse_result result;
926 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
927 if (opt == -ENOPARAM) {
928 if (strcmp(param->key, "source") == 0) {
929 if (param->type != fs_value_is_string)
930 return invalf(fc, "Non-string source");
932 return invalf(fc, "Multiple sources not supported");
933 fc->source = param->string;
934 param->string = NULL;
937 for_each_subsys(ss, i) {
938 if (strcmp(param->key, ss->legacy_name))
940 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
941 return invalfc(fc, "Disabled controller '%s'",
943 ctx->subsys_mask |= (1 << i);
946 return invalfc(fc, "Unknown subsys name '%s'", param->key);
953 /* Explicitly have no subsystems */
960 ctx->flags |= CGRP_ROOT_NOPREFIX;
962 case Opt_clone_children:
963 ctx->cpuset_clone_children = true;
965 case Opt_cpuset_v2_mode:
966 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
969 ctx->flags |= CGRP_ROOT_XATTR;
971 case Opt_release_agent:
972 /* Specifying two release agents is forbidden */
973 if (ctx->release_agent)
974 return invalfc(fc, "release_agent respecified");
976 * Release agent gets called with all capabilities,
977 * require capabilities to set release agent.
979 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN))
980 return invalfc(fc, "Setting release_agent not allowed");
981 ctx->release_agent = param->string;
982 param->string = NULL;
985 /* blocked by boot param? */
986 if (cgroup_no_v1_named)
988 /* Can't specify an empty name */
990 return invalfc(fc, "Empty name");
991 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
992 return invalfc(fc, "Name too long");
993 /* Must match [\w.-]+ */
994 for (i = 0; i < param->size; i++) {
995 char c = param->string[i];
998 if ((c == '.') || (c == '-') || (c == '_'))
1000 return invalfc(fc, "Invalid name");
1002 /* Specifying two names is forbidden */
1004 return invalfc(fc, "name respecified");
1005 ctx->name = param->string;
1006 param->string = NULL;
1012 static int check_cgroupfs_options(struct fs_context *fc)
1014 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1017 struct cgroup_subsys *ss;
1020 #ifdef CONFIG_CPUSETS
1021 mask = ~((u16)1 << cpuset_cgrp_id);
1023 for_each_subsys(ss, i)
1024 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1027 ctx->subsys_mask &= enabled;
1030 * In absense of 'none', 'name=' or subsystem name options,
1031 * let's default to 'all'.
1033 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1037 /* Mutually exclusive option 'all' + subsystem name */
1038 if (ctx->subsys_mask)
1039 return invalfc(fc, "subsys name conflicts with all");
1040 /* 'all' => select all the subsystems */
1041 ctx->subsys_mask = enabled;
1045 * We either have to specify by name or by subsystems. (So all
1046 * empty hierarchies must have a name).
1048 if (!ctx->subsys_mask && !ctx->name)
1049 return invalfc(fc, "Need name or subsystem set");
1052 * Option noprefix was introduced just for backward compatibility
1053 * with the old cpuset, so we allow noprefix only if mounting just
1054 * the cpuset subsystem.
1056 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1057 return invalfc(fc, "noprefix used incorrectly");
1059 /* Can't specify "none" and some subsystems */
1060 if (ctx->subsys_mask && ctx->none)
1061 return invalfc(fc, "none used incorrectly");
1066 int cgroup1_reconfigure(struct fs_context *fc)
1068 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1069 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1070 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1072 u16 added_mask, removed_mask;
1074 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1076 /* See what subsystems are wanted */
1077 ret = check_cgroupfs_options(fc);
1081 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1082 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1083 task_tgid_nr(current), current->comm);
1085 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1086 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1088 /* Don't allow flags or name to change at remount */
1089 if ((ctx->flags ^ root->flags) ||
1090 (ctx->name && strcmp(ctx->name, root->name))) {
1091 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1092 ctx->flags, ctx->name ?: "", root->flags, root->name);
1097 /* remounting is not allowed for populated hierarchies */
1098 if (!list_empty(&root->cgrp.self.children)) {
1103 ret = rebind_subsystems(root, added_mask);
1107 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1109 if (ctx->release_agent) {
1110 spin_lock(&release_agent_path_lock);
1111 strcpy(root->release_agent_path, ctx->release_agent);
1112 spin_unlock(&release_agent_path_lock);
1115 trace_cgroup_remount(root);
1118 mutex_unlock(&cgroup_mutex);
1122 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1123 .rename = cgroup1_rename,
1124 .show_options = cgroup1_show_options,
1125 .mkdir = cgroup_mkdir,
1126 .rmdir = cgroup_rmdir,
1127 .show_path = cgroup_show_path,
1131 * The guts of cgroup1 mount - find or create cgroup_root to use.
1132 * Called with cgroup_mutex held; returns 0 on success, -E... on
1133 * error and positive - in case when the candidate is busy dying.
1134 * On success it stashes a reference to cgroup_root into given
1135 * cgroup_fs_context; that reference is *NOT* counting towards the
1136 * cgroup_root refcount.
1138 static int cgroup1_root_to_use(struct fs_context *fc)
1140 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1141 struct cgroup_root *root;
1142 struct cgroup_subsys *ss;
1145 /* First find the desired set of subsystems */
1146 ret = check_cgroupfs_options(fc);
1151 * Destruction of cgroup root is asynchronous, so subsystems may
1152 * still be dying after the previous unmount. Let's drain the
1153 * dying subsystems. We just need to ensure that the ones
1154 * unmounted previously finish dying and don't care about new ones
1155 * starting. Testing ref liveliness is good enough.
1157 for_each_subsys(ss, i) {
1158 if (!(ctx->subsys_mask & (1 << i)) ||
1159 ss->root == &cgrp_dfl_root)
1162 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1163 return 1; /* restart */
1164 cgroup_put(&ss->root->cgrp);
1167 for_each_root(root) {
1168 bool name_match = false;
1170 if (root == &cgrp_dfl_root)
1174 * If we asked for a name then it must match. Also, if
1175 * name matches but sybsys_mask doesn't, we should fail.
1176 * Remember whether name matched.
1179 if (strcmp(ctx->name, root->name))
1185 * If we asked for subsystems (or explicitly for no
1186 * subsystems) then they must match.
1188 if ((ctx->subsys_mask || ctx->none) &&
1189 (ctx->subsys_mask != root->subsys_mask)) {
1195 if (root->flags ^ ctx->flags)
1196 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1203 * No such thing, create a new one. name= matching without subsys
1204 * specification is allowed for already existing hierarchies but we
1205 * can't create new one without subsys specification.
1207 if (!ctx->subsys_mask && !ctx->none)
1208 return invalfc(fc, "No subsys list or none specified");
1210 /* Hierarchies may only be created in the initial cgroup namespace. */
1211 if (ctx->ns != &init_cgroup_ns)
1214 root = kzalloc(sizeof(*root), GFP_KERNEL);
1219 init_cgroup_root(ctx);
1221 ret = cgroup_setup_root(root, ctx->subsys_mask);
1223 cgroup_free_root(root);
1227 int cgroup1_get_tree(struct fs_context *fc)
1229 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1232 /* Check if the caller has permission to mount. */
1233 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1236 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1238 ret = cgroup1_root_to_use(fc);
1239 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1240 ret = 1; /* restart */
1242 mutex_unlock(&cgroup_mutex);
1245 ret = cgroup_do_get_tree(fc);
1247 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1252 if (unlikely(ret > 0)) {
1254 return restart_syscall();
1259 static int __init cgroup1_wq_init(void)
1262 * Used to destroy pidlists and separate to serve as flush domain.
1263 * Cap @max_active to 1 too.
1265 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1267 BUG_ON(!cgroup_pidlist_destroy_wq);
1270 core_initcall(cgroup1_wq_init);
1272 static int __init cgroup_no_v1(char *str)
1274 struct cgroup_subsys *ss;
1278 while ((token = strsep(&str, ",")) != NULL) {
1282 if (!strcmp(token, "all")) {
1283 cgroup_no_v1_mask = U16_MAX;
1287 if (!strcmp(token, "named")) {
1288 cgroup_no_v1_named = true;
1292 for_each_subsys(ss, i) {
1293 if (strcmp(token, ss->name) &&
1294 strcmp(token, ss->legacy_name))
1297 cgroup_no_v1_mask |= 1 << i;
1302 __setup("cgroup_no_v1=", cgroup_no_v1);