1 #include "cgroup-internal.h"
3 #include <linux/ctype.h>
4 #include <linux/kmod.h>
5 #include <linux/sort.h>
6 #include <linux/delay.h>
8 #include <linux/sched/signal.h>
9 #include <linux/sched/task.h>
10 #include <linux/magic.h>
11 #include <linux/slab.h>
12 #include <linux/vmalloc.h>
13 #include <linux/delayacct.h>
14 #include <linux/pid_namespace.h>
15 #include <linux/cgroupstats.h>
16 #include <linux/fs_parser.h>
18 #include <trace/events/cgroup.h>
20 #define cg_invalf(fc, fmt, ...) invalf(fc, fmt, ## __VA_ARGS__)
23 * pidlists linger the following amount before being destroyed. The goal
24 * is avoiding frequent destruction in the middle of consecutive read calls
25 * Expiring in the middle is a performance problem not a correctness one.
26 * 1 sec should be enough.
28 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
30 /* Controllers blocked by the commandline in v1 */
31 static u16 cgroup_no_v1_mask;
33 /* disable named v1 mounts */
34 static bool cgroup_no_v1_named;
37 * pidlist destructions need to be flushed on cgroup destruction. Use a
38 * separate workqueue as flush domain.
40 static struct workqueue_struct *cgroup_pidlist_destroy_wq;
43 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
44 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
46 static DEFINE_SPINLOCK(release_agent_path_lock);
48 bool cgroup1_ssid_disabled(int ssid)
50 return cgroup_no_v1_mask & (1 << ssid);
54 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
55 * @from: attach to all cgroups of a given task
56 * @tsk: the task to be attached
58 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
60 struct cgroup_root *root;
63 mutex_lock(&cgroup_mutex);
64 percpu_down_write(&cgroup_threadgroup_rwsem);
66 struct cgroup *from_cgrp;
68 if (root == &cgrp_dfl_root)
71 spin_lock_irq(&css_set_lock);
72 from_cgrp = task_cgroup_from_root(from, root);
73 spin_unlock_irq(&css_set_lock);
75 retval = cgroup_attach_task(from_cgrp, tsk, false);
79 percpu_up_write(&cgroup_threadgroup_rwsem);
80 mutex_unlock(&cgroup_mutex);
84 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
87 * cgroup_trasnsfer_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 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
99 DEFINE_CGROUP_MGCTX(mgctx);
100 struct cgrp_cset_link *link;
101 struct css_task_iter it;
102 struct task_struct *task;
105 if (cgroup_on_dfl(to))
108 ret = cgroup_migrate_vet_dst(to);
112 mutex_lock(&cgroup_mutex);
114 percpu_down_write(&cgroup_threadgroup_rwsem);
116 /* all tasks in @from are being moved, all csets are source */
117 spin_lock_irq(&css_set_lock);
118 list_for_each_entry(link, &from->cset_links, cset_link)
119 cgroup_migrate_add_src(link->cset, to, &mgctx);
120 spin_unlock_irq(&css_set_lock);
122 ret = cgroup_migrate_prepare_dst(&mgctx);
127 * Migrate tasks one-by-one until @from is empty. This fails iff
128 * ->can_attach() fails.
131 css_task_iter_start(&from->self, 0, &it);
134 task = css_task_iter_next(&it);
135 } while (task && (task->flags & PF_EXITING));
138 get_task_struct(task);
139 css_task_iter_end(&it);
142 ret = cgroup_migrate(task, false, &mgctx);
144 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
145 put_task_struct(task);
147 } while (task && !ret);
149 cgroup_migrate_finish(&mgctx);
150 percpu_up_write(&cgroup_threadgroup_rwsem);
151 mutex_unlock(&cgroup_mutex);
156 * Stuff for reading the 'tasks'/'procs' files.
158 * Reading this file can return large amounts of data if a cgroup has
159 * *lots* of attached tasks. So it may need several calls to read(),
160 * but we cannot guarantee that the information we produce is correct
161 * unless we produce it entirely atomically.
165 /* which pidlist file are we talking about? */
166 enum cgroup_filetype {
172 * A pidlist is a list of pids that virtually represents the contents of one
173 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
174 * a pair (one each for procs, tasks) for each pid namespace that's relevant
177 struct cgroup_pidlist {
179 * used to find which pidlist is wanted. doesn't change as long as
180 * this particular list stays in the list.
182 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
185 /* how many elements the above list has */
187 /* each of these stored in a list by its cgroup */
188 struct list_head links;
189 /* pointer to the cgroup we belong to, for list removal purposes */
190 struct cgroup *owner;
191 /* for delayed destruction */
192 struct delayed_work destroy_dwork;
196 * The following two functions "fix" the issue where there are more pids
197 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
198 * TODO: replace with a kernel-wide solution to this problem
200 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
201 static void *pidlist_allocate(int count)
203 if (PIDLIST_TOO_LARGE(count))
204 return vmalloc(array_size(count, sizeof(pid_t)));
206 return kmalloc_array(count, sizeof(pid_t), GFP_KERNEL);
209 static void pidlist_free(void *p)
215 * Used to destroy all pidlists lingering waiting for destroy timer. None
216 * should be left afterwards.
218 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
220 struct cgroup_pidlist *l, *tmp_l;
222 mutex_lock(&cgrp->pidlist_mutex);
223 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
224 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
225 mutex_unlock(&cgrp->pidlist_mutex);
227 flush_workqueue(cgroup_pidlist_destroy_wq);
228 BUG_ON(!list_empty(&cgrp->pidlists));
231 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
233 struct delayed_work *dwork = to_delayed_work(work);
234 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
236 struct cgroup_pidlist *tofree = NULL;
238 mutex_lock(&l->owner->pidlist_mutex);
241 * Destroy iff we didn't get queued again. The state won't change
242 * as destroy_dwork can only be queued while locked.
244 if (!delayed_work_pending(dwork)) {
246 pidlist_free(l->list);
247 put_pid_ns(l->key.ns);
251 mutex_unlock(&l->owner->pidlist_mutex);
256 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
257 * Returns the number of unique elements.
259 static int pidlist_uniq(pid_t *list, int length)
264 * we presume the 0th element is unique, so i starts at 1. trivial
265 * edge cases first; no work needs to be done for either
267 if (length == 0 || length == 1)
269 /* src and dest walk down the list; dest counts unique elements */
270 for (src = 1; src < length; src++) {
271 /* find next unique element */
272 while (list[src] == list[src-1]) {
277 /* dest always points to where the next unique element goes */
278 list[dest] = list[src];
286 * The two pid files - task and cgroup.procs - guaranteed that the result
287 * is sorted, which forced this whole pidlist fiasco. As pid order is
288 * different per namespace, each namespace needs differently sorted list,
289 * making it impossible to use, for example, single rbtree of member tasks
290 * sorted by task pointer. As pidlists can be fairly large, allocating one
291 * per open file is dangerous, so cgroup had to implement shared pool of
292 * pidlists keyed by cgroup and namespace.
294 static int cmppid(const void *a, const void *b)
296 return *(pid_t *)a - *(pid_t *)b;
299 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
300 enum cgroup_filetype type)
302 struct cgroup_pidlist *l;
303 /* don't need task_nsproxy() if we're looking at ourself */
304 struct pid_namespace *ns = task_active_pid_ns(current);
306 lockdep_assert_held(&cgrp->pidlist_mutex);
308 list_for_each_entry(l, &cgrp->pidlists, links)
309 if (l->key.type == type && l->key.ns == ns)
315 * find the appropriate pidlist for our purpose (given procs vs tasks)
316 * returns with the lock on that pidlist already held, and takes care
317 * of the use count, or returns NULL with no locks held if we're out of
320 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
321 enum cgroup_filetype type)
323 struct cgroup_pidlist *l;
325 lockdep_assert_held(&cgrp->pidlist_mutex);
327 l = cgroup_pidlist_find(cgrp, type);
331 /* entry not found; create a new one */
332 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
336 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
338 /* don't need task_nsproxy() if we're looking at ourself */
339 l->key.ns = get_pid_ns(task_active_pid_ns(current));
341 list_add(&l->links, &cgrp->pidlists);
346 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
348 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
349 struct cgroup_pidlist **lp)
353 int pid, n = 0; /* used for populating the array */
354 struct css_task_iter it;
355 struct task_struct *tsk;
356 struct cgroup_pidlist *l;
358 lockdep_assert_held(&cgrp->pidlist_mutex);
361 * If cgroup gets more users after we read count, we won't have
362 * enough space - tough. This race is indistinguishable to the
363 * caller from the case that the additional cgroup users didn't
364 * show up until sometime later on.
366 length = cgroup_task_count(cgrp);
367 array = pidlist_allocate(length);
370 /* now, populate the array */
371 css_task_iter_start(&cgrp->self, 0, &it);
372 while ((tsk = css_task_iter_next(&it))) {
373 if (unlikely(n == length))
375 /* get tgid or pid for procs or tasks file respectively */
376 if (type == CGROUP_FILE_PROCS)
377 pid = task_tgid_vnr(tsk);
379 pid = task_pid_vnr(tsk);
380 if (pid > 0) /* make sure to only use valid results */
383 css_task_iter_end(&it);
385 /* now sort & (if procs) strip out duplicates */
386 sort(array, length, sizeof(pid_t), cmppid, NULL);
387 if (type == CGROUP_FILE_PROCS)
388 length = pidlist_uniq(array, length);
390 l = cgroup_pidlist_find_create(cgrp, type);
396 /* store array, freeing old if necessary */
397 pidlist_free(l->list);
405 * seq_file methods for the tasks/procs files. The seq_file position is the
406 * next pid to display; the seq_file iterator is a pointer to the pid
407 * in the cgroup->l->list array.
410 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
413 * Initially we receive a position value that corresponds to
414 * one more than the last pid shown (or 0 on the first call or
415 * after a seek to the start). Use a binary-search to find the
416 * next pid to display, if any
418 struct kernfs_open_file *of = s->private;
419 struct cgroup *cgrp = seq_css(s)->cgroup;
420 struct cgroup_pidlist *l;
421 enum cgroup_filetype type = seq_cft(s)->private;
422 int index = 0, pid = *pos;
425 mutex_lock(&cgrp->pidlist_mutex);
428 * !NULL @of->priv indicates that this isn't the first start()
429 * after open. If the matching pidlist is around, we can use that.
430 * Look for it. Note that @of->priv can't be used directly. It
431 * could already have been destroyed.
434 of->priv = cgroup_pidlist_find(cgrp, type);
437 * Either this is the first start() after open or the matching
438 * pidlist has been destroyed inbetween. Create a new one.
441 ret = pidlist_array_load(cgrp, type,
442 (struct cgroup_pidlist **)&of->priv);
451 while (index < end) {
452 int mid = (index + end) / 2;
453 if (l->list[mid] == pid) {
456 } else if (l->list[mid] <= pid)
462 /* If we're off the end of the array, we're done */
463 if (index >= l->length)
465 /* Update the abstract position to be the actual pid that we found */
466 iter = l->list + index;
471 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
473 struct kernfs_open_file *of = s->private;
474 struct cgroup_pidlist *l = of->priv;
477 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
478 CGROUP_PIDLIST_DESTROY_DELAY);
479 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
482 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
484 struct kernfs_open_file *of = s->private;
485 struct cgroup_pidlist *l = of->priv;
487 pid_t *end = l->list + l->length;
489 * Advance to the next pid in the array. If this goes off the
501 static int cgroup_pidlist_show(struct seq_file *s, void *v)
503 seq_printf(s, "%d\n", *(int *)v);
508 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
509 char *buf, size_t nbytes, loff_t off,
513 struct task_struct *task;
514 const struct cred *cred, *tcred;
517 cgrp = cgroup_kn_lock_live(of->kn, false);
521 task = cgroup_procs_write_start(buf, threadgroup);
522 ret = PTR_ERR_OR_ZERO(task);
527 * Even if we're attaching all tasks in the thread group, we only
528 * need to check permissions on one of them.
530 cred = current_cred();
531 tcred = get_task_cred(task);
532 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
533 !uid_eq(cred->euid, tcred->uid) &&
534 !uid_eq(cred->euid, tcred->suid))
540 ret = cgroup_attach_task(cgrp, task, threadgroup);
543 cgroup_procs_write_finish(task);
545 cgroup_kn_unlock(of->kn);
547 return ret ?: nbytes;
550 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
551 char *buf, size_t nbytes, loff_t off)
553 return __cgroup1_procs_write(of, buf, nbytes, off, true);
556 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
557 char *buf, size_t nbytes, loff_t off)
559 return __cgroup1_procs_write(of, buf, nbytes, off, false);
562 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
563 char *buf, size_t nbytes, loff_t off)
567 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
569 cgrp = cgroup_kn_lock_live(of->kn, false);
572 spin_lock(&release_agent_path_lock);
573 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
574 sizeof(cgrp->root->release_agent_path));
575 spin_unlock(&release_agent_path_lock);
576 cgroup_kn_unlock(of->kn);
580 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
582 struct cgroup *cgrp = seq_css(seq)->cgroup;
584 spin_lock(&release_agent_path_lock);
585 seq_puts(seq, cgrp->root->release_agent_path);
586 spin_unlock(&release_agent_path_lock);
591 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
593 seq_puts(seq, "0\n");
597 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
600 return notify_on_release(css->cgroup);
603 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
604 struct cftype *cft, u64 val)
607 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
609 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
613 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
616 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
619 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
620 struct cftype *cft, u64 val)
623 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
625 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
629 /* cgroup core interface files for the legacy hierarchies */
630 struct cftype cgroup1_base_files[] = {
632 .name = "cgroup.procs",
633 .seq_start = cgroup_pidlist_start,
634 .seq_next = cgroup_pidlist_next,
635 .seq_stop = cgroup_pidlist_stop,
636 .seq_show = cgroup_pidlist_show,
637 .private = CGROUP_FILE_PROCS,
638 .write = cgroup1_procs_write,
641 .name = "cgroup.clone_children",
642 .read_u64 = cgroup_clone_children_read,
643 .write_u64 = cgroup_clone_children_write,
646 .name = "cgroup.sane_behavior",
647 .flags = CFTYPE_ONLY_ON_ROOT,
648 .seq_show = cgroup_sane_behavior_show,
652 .seq_start = cgroup_pidlist_start,
653 .seq_next = cgroup_pidlist_next,
654 .seq_stop = cgroup_pidlist_stop,
655 .seq_show = cgroup_pidlist_show,
656 .private = CGROUP_FILE_TASKS,
657 .write = cgroup1_tasks_write,
660 .name = "notify_on_release",
661 .read_u64 = cgroup_read_notify_on_release,
662 .write_u64 = cgroup_write_notify_on_release,
665 .name = "release_agent",
666 .flags = CFTYPE_ONLY_ON_ROOT,
667 .seq_show = cgroup_release_agent_show,
668 .write = cgroup_release_agent_write,
669 .max_write_len = PATH_MAX - 1,
674 /* Display information about each subsystem and each hierarchy */
675 int proc_cgroupstats_show(struct seq_file *m, void *v)
677 struct cgroup_subsys *ss;
680 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
682 * ideally we don't want subsystems moving around while we do this.
683 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
684 * subsys/hierarchy state.
686 mutex_lock(&cgroup_mutex);
688 for_each_subsys(ss, i)
689 seq_printf(m, "%s\t%d\t%d\t%d\n",
690 ss->legacy_name, ss->root->hierarchy_id,
691 atomic_read(&ss->root->nr_cgrps),
692 cgroup_ssid_enabled(i));
694 mutex_unlock(&cgroup_mutex);
699 * cgroupstats_build - build and fill cgroupstats
700 * @stats: cgroupstats to fill information into
701 * @dentry: A dentry entry belonging to the cgroup for which stats have
704 * Build and fill cgroupstats so that taskstats can export it to user
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 (tsk->state) {
741 case TASK_INTERRUPTIBLE:
742 stats->nr_sleeping++;
744 case TASK_UNINTERRUPTIBLE:
745 stats->nr_uninterruptible++;
751 if (delayacct_is_task_waiting_on_io(tsk))
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 = NULL, *agentbuf = NULL;
797 char *argv[3], *envp[3];
800 mutex_lock(&cgroup_mutex);
802 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
803 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
804 if (!pathbuf || !agentbuf)
807 spin_lock_irq(&css_set_lock);
808 ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
809 spin_unlock_irq(&css_set_lock);
810 if (ret < 0 || ret >= PATH_MAX)
817 /* minimal command environment */
819 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
822 mutex_unlock(&cgroup_mutex);
823 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
826 mutex_unlock(&cgroup_mutex);
833 * cgroup_rename - Only allow simple rename of directories in place.
835 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
836 const char *new_name_str)
838 struct cgroup *cgrp = kn->priv;
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 static const struct fs_parameter_spec cgroup1_param_specs[] = {
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 const struct fs_parameter_description cgroup1_fs_parameters = {
921 .specs = cgroup1_param_specs,
924 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
926 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
927 struct cgroup_subsys *ss;
928 struct fs_parse_result result;
931 opt = fs_parse(fc, &cgroup1_fs_parameters, param, &result);
932 if (opt == -ENOPARAM) {
933 if (strcmp(param->key, "source") == 0) {
934 fc->source = param->string;
935 param->string = NULL;
938 for_each_subsys(ss, i) {
939 if (strcmp(param->key, ss->legacy_name))
941 ctx->subsys_mask |= (1 << i);
944 return cg_invalf(fc, "cgroup1: Unknown subsys name '%s'", param->key);
951 /* Explicitly have no subsystems */
958 ctx->flags |= CGRP_ROOT_NOPREFIX;
960 case Opt_clone_children:
961 ctx->cpuset_clone_children = true;
963 case Opt_cpuset_v2_mode:
964 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
967 ctx->flags |= CGRP_ROOT_XATTR;
969 case Opt_release_agent:
970 /* Specifying two release agents is forbidden */
971 if (ctx->release_agent)
972 return cg_invalf(fc, "cgroup1: release_agent respecified");
973 ctx->release_agent = param->string;
974 param->string = NULL;
977 /* blocked by boot param? */
978 if (cgroup_no_v1_named)
980 /* Can't specify an empty name */
982 return cg_invalf(fc, "cgroup1: Empty name");
983 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
984 return cg_invalf(fc, "cgroup1: Name too long");
985 /* Must match [\w.-]+ */
986 for (i = 0; i < param->size; i++) {
987 char c = param->string[i];
990 if ((c == '.') || (c == '-') || (c == '_'))
992 return cg_invalf(fc, "cgroup1: Invalid name");
994 /* Specifying two names is forbidden */
996 return cg_invalf(fc, "cgroup1: name respecified");
997 ctx->name = param->string;
998 param->string = NULL;
1004 static int check_cgroupfs_options(struct fs_context *fc)
1006 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1009 struct cgroup_subsys *ss;
1012 #ifdef CONFIG_CPUSETS
1013 mask = ~((u16)1 << cpuset_cgrp_id);
1015 for_each_subsys(ss, i)
1016 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
1019 ctx->subsys_mask &= enabled;
1022 * In absense of 'none', 'name=' or subsystem name options,
1023 * let's default to 'all'.
1025 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1029 /* Mutually exclusive option 'all' + subsystem name */
1030 if (ctx->subsys_mask)
1031 return cg_invalf(fc, "cgroup1: subsys name conflicts with all");
1032 /* 'all' => select all the subsystems */
1033 ctx->subsys_mask = enabled;
1037 * We either have to specify by name or by subsystems. (So all
1038 * empty hierarchies must have a name).
1040 if (!ctx->subsys_mask && !ctx->name)
1041 return cg_invalf(fc, "cgroup1: Need name or subsystem set");
1044 * Option noprefix was introduced just for backward compatibility
1045 * with the old cpuset, so we allow noprefix only if mounting just
1046 * the cpuset subsystem.
1048 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1049 return cg_invalf(fc, "cgroup1: noprefix used incorrectly");
1051 /* Can't specify "none" and some subsystems */
1052 if (ctx->subsys_mask && ctx->none)
1053 return cg_invalf(fc, "cgroup1: none used incorrectly");
1058 int cgroup1_reconfigure(struct fs_context *fc)
1060 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1061 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1062 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1064 u16 added_mask, removed_mask;
1066 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1068 /* See what subsystems are wanted */
1069 ret = check_cgroupfs_options(fc);
1073 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1074 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1075 task_tgid_nr(current), current->comm);
1077 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1078 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1080 /* Don't allow flags or name to change at remount */
1081 if ((ctx->flags ^ root->flags) ||
1082 (ctx->name && strcmp(ctx->name, root->name))) {
1083 cg_invalf(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1084 ctx->flags, ctx->name ?: "", root->flags, root->name);
1089 /* remounting is not allowed for populated hierarchies */
1090 if (!list_empty(&root->cgrp.self.children)) {
1095 ret = rebind_subsystems(root, added_mask);
1099 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1101 if (ctx->release_agent) {
1102 spin_lock(&release_agent_path_lock);
1103 strcpy(root->release_agent_path, ctx->release_agent);
1104 spin_unlock(&release_agent_path_lock);
1107 trace_cgroup_remount(root);
1110 mutex_unlock(&cgroup_mutex);
1114 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1115 .rename = cgroup1_rename,
1116 .show_options = cgroup1_show_options,
1117 .mkdir = cgroup_mkdir,
1118 .rmdir = cgroup_rmdir,
1119 .show_path = cgroup_show_path,
1123 * The guts of cgroup1 mount - find or create cgroup_root to use.
1124 * Called with cgroup_mutex held; returns 0 on success, -E... on
1125 * error and positive - in case when the candidate is busy dying.
1126 * On success it stashes a reference to cgroup_root into given
1127 * cgroup_fs_context; that reference is *NOT* counting towards the
1128 * cgroup_root refcount.
1130 static int cgroup1_root_to_use(struct fs_context *fc)
1132 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1133 struct cgroup_root *root;
1134 struct cgroup_subsys *ss;
1137 /* First find the desired set of subsystems */
1138 ret = check_cgroupfs_options(fc);
1143 * Destruction of cgroup root is asynchronous, so subsystems may
1144 * still be dying after the previous unmount. Let's drain the
1145 * dying subsystems. We just need to ensure that the ones
1146 * unmounted previously finish dying and don't care about new ones
1147 * starting. Testing ref liveliness is good enough.
1149 for_each_subsys(ss, i) {
1150 if (!(ctx->subsys_mask & (1 << i)) ||
1151 ss->root == &cgrp_dfl_root)
1154 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1155 return 1; /* restart */
1156 cgroup_put(&ss->root->cgrp);
1159 for_each_root(root) {
1160 bool name_match = false;
1162 if (root == &cgrp_dfl_root)
1166 * If we asked for a name then it must match. Also, if
1167 * name matches but sybsys_mask doesn't, we should fail.
1168 * Remember whether name matched.
1171 if (strcmp(ctx->name, root->name))
1177 * If we asked for subsystems (or explicitly for no
1178 * subsystems) then they must match.
1180 if ((ctx->subsys_mask || ctx->none) &&
1181 (ctx->subsys_mask != root->subsys_mask)) {
1187 if (root->flags ^ ctx->flags)
1188 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1195 * No such thing, create a new one. name= matching without subsys
1196 * specification is allowed for already existing hierarchies but we
1197 * can't create new one without subsys specification.
1199 if (!ctx->subsys_mask && !ctx->none)
1200 return cg_invalf(fc, "cgroup1: No subsys list or none specified");
1202 /* Hierarchies may only be created in the initial cgroup namespace. */
1203 if (ctx->ns != &init_cgroup_ns)
1206 root = kzalloc(sizeof(*root), GFP_KERNEL);
1211 init_cgroup_root(ctx);
1213 ret = cgroup_setup_root(root, ctx->subsys_mask);
1215 cgroup_free_root(root);
1219 int cgroup1_get_tree(struct fs_context *fc)
1221 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1224 /* Check if the caller has permission to mount. */
1225 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1228 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1230 ret = cgroup1_root_to_use(fc);
1231 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1232 ret = 1; /* restart */
1234 mutex_unlock(&cgroup_mutex);
1237 ret = cgroup_do_get_tree(fc);
1239 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1240 struct super_block *sb = fc->root->d_sb;
1242 deactivate_locked_super(sb);
1246 if (unlikely(ret > 0)) {
1248 return restart_syscall();
1253 static int __init cgroup1_wq_init(void)
1256 * Used to destroy pidlists and separate to serve as flush domain.
1257 * Cap @max_active to 1 too.
1259 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1261 BUG_ON(!cgroup_pidlist_destroy_wq);
1264 core_initcall(cgroup1_wq_init);
1266 static int __init cgroup_no_v1(char *str)
1268 struct cgroup_subsys *ss;
1272 while ((token = strsep(&str, ",")) != NULL) {
1276 if (!strcmp(token, "all")) {
1277 cgroup_no_v1_mask = U16_MAX;
1281 if (!strcmp(token, "named")) {
1282 cgroup_no_v1_named = true;
1286 for_each_subsys(ss, i) {
1287 if (strcmp(token, ss->name) &&
1288 strcmp(token, ss->legacy_name))
1291 cgroup_no_v1_mask |= 1 << i;
1296 __setup("cgroup_no_v1=", cgroup_no_v1);