Loading...
Note: File does not exist in v3.5.6.
1// SPDX-License-Identifier: GPL-2.0-only
2#include "cgroup-internal.h"
3
4#include <linux/ctype.h>
5#include <linux/kmod.h>
6#include <linux/sort.h>
7#include <linux/delay.h>
8#include <linux/mm.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>
18
19#include <trace/events/cgroup.h>
20
21/*
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.
26 */
27#define CGROUP_PIDLIST_DESTROY_DELAY HZ
28
29/* Controllers blocked by the commandline in v1 */
30static u16 cgroup_no_v1_mask;
31
32/* disable named v1 mounts */
33static bool cgroup_no_v1_named;
34
35/*
36 * pidlist destructions need to be flushed on cgroup destruction. Use a
37 * separate workqueue as flush domain.
38 */
39static struct workqueue_struct *cgroup_pidlist_destroy_wq;
40
41/* protects cgroup_subsys->release_agent_path */
42static DEFINE_SPINLOCK(release_agent_path_lock);
43
44bool cgroup1_ssid_disabled(int ssid)
45{
46 return cgroup_no_v1_mask & (1 << ssid);
47}
48
49static bool cgroup1_subsys_absent(struct cgroup_subsys *ss)
50{
51 /* Check also dfl_cftypes for file-less controllers, i.e. perf_event */
52 return ss->legacy_cftypes == NULL && ss->dfl_cftypes;
53}
54
55/**
56 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
57 * @from: attach to all cgroups of a given task
58 * @tsk: the task to be attached
59 *
60 * Return: %0 on success or a negative errno code on failure
61 */
62int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
63{
64 struct cgroup_root *root;
65 int retval = 0;
66
67 cgroup_lock();
68 cgroup_attach_lock(true);
69 for_each_root(root) {
70 struct cgroup *from_cgrp;
71
72 spin_lock_irq(&css_set_lock);
73 from_cgrp = task_cgroup_from_root(from, root);
74 spin_unlock_irq(&css_set_lock);
75
76 retval = cgroup_attach_task(from_cgrp, tsk, false);
77 if (retval)
78 break;
79 }
80 cgroup_attach_unlock(true);
81 cgroup_unlock();
82
83 return retval;
84}
85EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
86
87/**
88 * cgroup_transfer_tasks - move tasks from one cgroup to another
89 * @to: cgroup to which the tasks will be moved
90 * @from: cgroup in which the tasks currently reside
91 *
92 * Locking rules between cgroup_post_fork() and the migration path
93 * guarantee that, if a task is forking while being migrated, the new child
94 * is guaranteed to be either visible in the source cgroup after the
95 * parent's migration is complete or put into the target cgroup. No task
96 * can slip out of migration through forking.
97 *
98 * Return: %0 on success or a negative errno code on failure
99 */
100int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
101{
102 DEFINE_CGROUP_MGCTX(mgctx);
103 struct cgrp_cset_link *link;
104 struct css_task_iter it;
105 struct task_struct *task;
106 int ret;
107
108 if (cgroup_on_dfl(to))
109 return -EINVAL;
110
111 ret = cgroup_migrate_vet_dst(to);
112 if (ret)
113 return ret;
114
115 cgroup_lock();
116
117 cgroup_attach_lock(true);
118
119 /* all tasks in @from are being moved, all csets are source */
120 spin_lock_irq(&css_set_lock);
121 list_for_each_entry(link, &from->cset_links, cset_link)
122 cgroup_migrate_add_src(link->cset, to, &mgctx);
123 spin_unlock_irq(&css_set_lock);
124
125 ret = cgroup_migrate_prepare_dst(&mgctx);
126 if (ret)
127 goto out_err;
128
129 /*
130 * Migrate tasks one-by-one until @from is empty. This fails iff
131 * ->can_attach() fails.
132 */
133 do {
134 css_task_iter_start(&from->self, 0, &it);
135
136 do {
137 task = css_task_iter_next(&it);
138 } while (task && (task->flags & PF_EXITING));
139
140 if (task)
141 get_task_struct(task);
142 css_task_iter_end(&it);
143
144 if (task) {
145 ret = cgroup_migrate(task, false, &mgctx);
146 if (!ret)
147 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
148 put_task_struct(task);
149 }
150 } while (task && !ret);
151out_err:
152 cgroup_migrate_finish(&mgctx);
153 cgroup_attach_unlock(true);
154 cgroup_unlock();
155 return ret;
156}
157
158/*
159 * Stuff for reading the 'tasks'/'procs' files.
160 *
161 * Reading this file can return large amounts of data if a cgroup has
162 * *lots* of attached tasks. So it may need several calls to read(),
163 * but we cannot guarantee that the information we produce is correct
164 * unless we produce it entirely atomically.
165 *
166 */
167
168/* which pidlist file are we talking about? */
169enum cgroup_filetype {
170 CGROUP_FILE_PROCS,
171 CGROUP_FILE_TASKS,
172};
173
174/*
175 * A pidlist is a list of pids that virtually represents the contents of one
176 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
177 * a pair (one each for procs, tasks) for each pid namespace that's relevant
178 * to the cgroup.
179 */
180struct cgroup_pidlist {
181 /*
182 * used to find which pidlist is wanted. doesn't change as long as
183 * this particular list stays in the list.
184 */
185 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
186 /* array of xids */
187 pid_t *list;
188 /* how many elements the above list has */
189 int length;
190 /* each of these stored in a list by its cgroup */
191 struct list_head links;
192 /* pointer to the cgroup we belong to, for list removal purposes */
193 struct cgroup *owner;
194 /* for delayed destruction */
195 struct delayed_work destroy_dwork;
196};
197
198/*
199 * Used to destroy all pidlists lingering waiting for destroy timer. None
200 * should be left afterwards.
201 */
202void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
203{
204 struct cgroup_pidlist *l, *tmp_l;
205
206 mutex_lock(&cgrp->pidlist_mutex);
207 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
208 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
209 mutex_unlock(&cgrp->pidlist_mutex);
210
211 flush_workqueue(cgroup_pidlist_destroy_wq);
212 BUG_ON(!list_empty(&cgrp->pidlists));
213}
214
215static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
216{
217 struct delayed_work *dwork = to_delayed_work(work);
218 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
219 destroy_dwork);
220 struct cgroup_pidlist *tofree = NULL;
221
222 mutex_lock(&l->owner->pidlist_mutex);
223
224 /*
225 * Destroy iff we didn't get queued again. The state won't change
226 * as destroy_dwork can only be queued while locked.
227 */
228 if (!delayed_work_pending(dwork)) {
229 list_del(&l->links);
230 kvfree(l->list);
231 put_pid_ns(l->key.ns);
232 tofree = l;
233 }
234
235 mutex_unlock(&l->owner->pidlist_mutex);
236 kfree(tofree);
237}
238
239/*
240 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
241 * Returns the number of unique elements.
242 */
243static int pidlist_uniq(pid_t *list, int length)
244{
245 int src, dest = 1;
246
247 /*
248 * we presume the 0th element is unique, so i starts at 1. trivial
249 * edge cases first; no work needs to be done for either
250 */
251 if (length == 0 || length == 1)
252 return length;
253 /* src and dest walk down the list; dest counts unique elements */
254 for (src = 1; src < length; src++) {
255 /* find next unique element */
256 while (list[src] == list[src-1]) {
257 src++;
258 if (src == length)
259 goto after;
260 }
261 /* dest always points to where the next unique element goes */
262 list[dest] = list[src];
263 dest++;
264 }
265after:
266 return dest;
267}
268
269/*
270 * The two pid files - task and cgroup.procs - guaranteed that the result
271 * is sorted, which forced this whole pidlist fiasco. As pid order is
272 * different per namespace, each namespace needs differently sorted list,
273 * making it impossible to use, for example, single rbtree of member tasks
274 * sorted by task pointer. As pidlists can be fairly large, allocating one
275 * per open file is dangerous, so cgroup had to implement shared pool of
276 * pidlists keyed by cgroup and namespace.
277 */
278static int cmppid(const void *a, const void *b)
279{
280 return *(pid_t *)a - *(pid_t *)b;
281}
282
283static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
284 enum cgroup_filetype type)
285{
286 struct cgroup_pidlist *l;
287 /* don't need task_nsproxy() if we're looking at ourself */
288 struct pid_namespace *ns = task_active_pid_ns(current);
289
290 lockdep_assert_held(&cgrp->pidlist_mutex);
291
292 list_for_each_entry(l, &cgrp->pidlists, links)
293 if (l->key.type == type && l->key.ns == ns)
294 return l;
295 return NULL;
296}
297
298/*
299 * find the appropriate pidlist for our purpose (given procs vs tasks)
300 * returns with the lock on that pidlist already held, and takes care
301 * of the use count, or returns NULL with no locks held if we're out of
302 * memory.
303 */
304static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
305 enum cgroup_filetype type)
306{
307 struct cgroup_pidlist *l;
308
309 lockdep_assert_held(&cgrp->pidlist_mutex);
310
311 l = cgroup_pidlist_find(cgrp, type);
312 if (l)
313 return l;
314
315 /* entry not found; create a new one */
316 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
317 if (!l)
318 return l;
319
320 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
321 l->key.type = type;
322 /* don't need task_nsproxy() if we're looking at ourself */
323 l->key.ns = get_pid_ns(task_active_pid_ns(current));
324 l->owner = cgrp;
325 list_add(&l->links, &cgrp->pidlists);
326 return l;
327}
328
329/*
330 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
331 */
332static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
333 struct cgroup_pidlist **lp)
334{
335 pid_t *array;
336 int length;
337 int pid, n = 0; /* used for populating the array */
338 struct css_task_iter it;
339 struct task_struct *tsk;
340 struct cgroup_pidlist *l;
341
342 lockdep_assert_held(&cgrp->pidlist_mutex);
343
344 /*
345 * If cgroup gets more users after we read count, we won't have
346 * enough space - tough. This race is indistinguishable to the
347 * caller from the case that the additional cgroup users didn't
348 * show up until sometime later on.
349 */
350 length = cgroup_task_count(cgrp);
351 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
352 if (!array)
353 return -ENOMEM;
354 /* now, populate the array */
355 css_task_iter_start(&cgrp->self, 0, &it);
356 while ((tsk = css_task_iter_next(&it))) {
357 if (unlikely(n == length))
358 break;
359 /* get tgid or pid for procs or tasks file respectively */
360 if (type == CGROUP_FILE_PROCS)
361 pid = task_tgid_vnr(tsk);
362 else
363 pid = task_pid_vnr(tsk);
364 if (pid > 0) /* make sure to only use valid results */
365 array[n++] = pid;
366 }
367 css_task_iter_end(&it);
368 length = n;
369 /* now sort & strip out duplicates (tgids or recycled thread PIDs) */
370 sort(array, length, sizeof(pid_t), cmppid, NULL);
371 length = pidlist_uniq(array, length);
372
373 l = cgroup_pidlist_find_create(cgrp, type);
374 if (!l) {
375 kvfree(array);
376 return -ENOMEM;
377 }
378
379 /* store array, freeing old if necessary */
380 kvfree(l->list);
381 l->list = array;
382 l->length = length;
383 *lp = l;
384 return 0;
385}
386
387/*
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.
391 */
392
393static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
394{
395 /*
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
400 */
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;
407 int *iter, ret;
408
409 mutex_lock(&cgrp->pidlist_mutex);
410
411 /*
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.
416 */
417 if (ctx->procs1.pidlist)
418 ctx->procs1.pidlist = cgroup_pidlist_find(cgrp, type);
419
420 /*
421 * Either this is the first start() after open or the matching
422 * pidlist has been destroyed inbetween. Create a new one.
423 */
424 if (!ctx->procs1.pidlist) {
425 ret = pidlist_array_load(cgrp, type, &ctx->procs1.pidlist);
426 if (ret)
427 return ERR_PTR(ret);
428 }
429 l = ctx->procs1.pidlist;
430
431 if (pid) {
432 int end = l->length;
433
434 while (index < end) {
435 int mid = (index + end) / 2;
436 if (l->list[mid] == pid) {
437 index = mid;
438 break;
439 } else if (l->list[mid] < pid)
440 index = mid + 1;
441 else
442 end = mid;
443 }
444 }
445 /* If we're off the end of the array, we're done */
446 if (index >= l->length)
447 return NULL;
448 /* Update the abstract position to be the actual pid that we found */
449 iter = l->list + index;
450 *pos = *iter;
451 return iter;
452}
453
454static void cgroup_pidlist_stop(struct seq_file *s, void *v)
455{
456 struct kernfs_open_file *of = s->private;
457 struct cgroup_file_ctx *ctx = of->priv;
458 struct cgroup_pidlist *l = ctx->procs1.pidlist;
459
460 if (l)
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);
464}
465
466static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
467{
468 struct kernfs_open_file *of = s->private;
469 struct cgroup_file_ctx *ctx = of->priv;
470 struct cgroup_pidlist *l = ctx->procs1.pidlist;
471 pid_t *p = v;
472 pid_t *end = l->list + l->length;
473 /*
474 * Advance to the next pid in the array. If this goes off the
475 * end, we're done
476 */
477 p++;
478 if (p >= end) {
479 (*pos)++;
480 return NULL;
481 } else {
482 *pos = *p;
483 return p;
484 }
485}
486
487static int cgroup_pidlist_show(struct seq_file *s, void *v)
488{
489 seq_printf(s, "%d\n", *(int *)v);
490
491 return 0;
492}
493
494static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
495 char *buf, size_t nbytes, loff_t off,
496 bool threadgroup)
497{
498 struct cgroup *cgrp;
499 struct task_struct *task;
500 const struct cred *cred, *tcred;
501 ssize_t ret;
502 bool locked;
503
504 cgrp = cgroup_kn_lock_live(of->kn, false);
505 if (!cgrp)
506 return -ENODEV;
507
508 task = cgroup_procs_write_start(buf, threadgroup, &locked);
509 ret = PTR_ERR_OR_ZERO(task);
510 if (ret)
511 goto out_unlock;
512
513 /*
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.
517 */
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))
523 ret = -EACCES;
524 put_cred(tcred);
525 if (ret)
526 goto out_finish;
527
528 ret = cgroup_attach_task(cgrp, task, threadgroup);
529
530out_finish:
531 cgroup_procs_write_finish(task, locked);
532out_unlock:
533 cgroup_kn_unlock(of->kn);
534
535 return ret ?: nbytes;
536}
537
538static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
539 char *buf, size_t nbytes, loff_t off)
540{
541 return __cgroup1_procs_write(of, buf, nbytes, off, true);
542}
543
544static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
545 char *buf, size_t nbytes, loff_t off)
546{
547 return __cgroup1_procs_write(of, buf, nbytes, off, false);
548}
549
550static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
551 char *buf, size_t nbytes, loff_t off)
552{
553 struct cgroup *cgrp;
554 struct cgroup_file_ctx *ctx;
555
556 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
557
558 /*
559 * Release agent gets called with all capabilities,
560 * require capabilities to set release agent.
561 */
562 ctx = of->priv;
563 if ((ctx->ns->user_ns != &init_user_ns) ||
564 !file_ns_capable(of->file, &init_user_ns, CAP_SYS_ADMIN))
565 return -EPERM;
566
567 cgrp = cgroup_kn_lock_live(of->kn, false);
568 if (!cgrp)
569 return -ENODEV;
570 spin_lock(&release_agent_path_lock);
571 strscpy(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);
575 return nbytes;
576}
577
578static int cgroup_release_agent_show(struct seq_file *seq, void *v)
579{
580 struct cgroup *cgrp = seq_css(seq)->cgroup;
581
582 spin_lock(&release_agent_path_lock);
583 seq_puts(seq, cgrp->root->release_agent_path);
584 spin_unlock(&release_agent_path_lock);
585 seq_putc(seq, '\n');
586 return 0;
587}
588
589static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
590{
591 seq_puts(seq, "0\n");
592 return 0;
593}
594
595static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
596 struct cftype *cft)
597{
598 return notify_on_release(css->cgroup);
599}
600
601static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
602 struct cftype *cft, u64 val)
603{
604 if (val)
605 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
606 else
607 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
608 return 0;
609}
610
611static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
612 struct cftype *cft)
613{
614 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
615}
616
617static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
618 struct cftype *cft, u64 val)
619{
620 if (val)
621 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
622 else
623 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
624 return 0;
625}
626
627/* cgroup core interface files for the legacy hierarchies */
628struct cftype cgroup1_base_files[] = {
629 {
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,
637 },
638 {
639 .name = "cgroup.clone_children",
640 .read_u64 = cgroup_clone_children_read,
641 .write_u64 = cgroup_clone_children_write,
642 },
643 {
644 .name = "cgroup.sane_behavior",
645 .flags = CFTYPE_ONLY_ON_ROOT,
646 .seq_show = cgroup_sane_behavior_show,
647 },
648 {
649 .name = "tasks",
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,
656 },
657 {
658 .name = "notify_on_release",
659 .read_u64 = cgroup_read_notify_on_release,
660 .write_u64 = cgroup_write_notify_on_release,
661 },
662 {
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,
668 },
669 { } /* terminate */
670};
671
672/* Display information about each subsystem and each hierarchy */
673int proc_cgroupstats_show(struct seq_file *m, void *v)
674{
675 struct cgroup_subsys *ss;
676 int i;
677
678 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
679 /*
680 * Grab the subsystems state racily. No need to add avenue to
681 * cgroup_mutex contention.
682 */
683
684 for_each_subsys(ss, i) {
685 if (cgroup1_subsys_absent(ss))
686 continue;
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));
691 }
692
693 return 0;
694}
695
696/**
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
700 * been requested.
701 *
702 * Build and fill cgroupstats so that taskstats can export it to user
703 * space.
704 *
705 * Return: %0 on success or a negative errno code on failure
706 */
707int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
708{
709 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
710 struct cgroup *cgrp;
711 struct css_task_iter it;
712 struct task_struct *tsk;
713
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)
717 return -EINVAL;
718
719 /*
720 * We aren't being called from kernfs and there's no guarantee on
721 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
722 * @kn->priv is RCU safe. Let's do the RCU dancing.
723 */
724 rcu_read_lock();
725 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
726 if (!cgrp || !cgroup_tryget(cgrp)) {
727 rcu_read_unlock();
728 return -ENOENT;
729 }
730 rcu_read_unlock();
731
732 css_task_iter_start(&cgrp->self, 0, &it);
733 while ((tsk = css_task_iter_next(&it))) {
734 switch (READ_ONCE(tsk->__state)) {
735 case TASK_RUNNING:
736 stats->nr_running++;
737 break;
738 case TASK_INTERRUPTIBLE:
739 stats->nr_sleeping++;
740 break;
741 case TASK_UNINTERRUPTIBLE:
742 stats->nr_uninterruptible++;
743 break;
744 case TASK_STOPPED:
745 stats->nr_stopped++;
746 break;
747 default:
748 if (tsk->in_iowait)
749 stats->nr_io_wait++;
750 break;
751 }
752 }
753 css_task_iter_end(&it);
754
755 cgroup_put(cgrp);
756 return 0;
757}
758
759void cgroup1_check_for_release(struct cgroup *cgrp)
760{
761 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
762 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
763 schedule_work(&cgrp->release_agent_work);
764}
765
766/*
767 * Notify userspace when a cgroup is released, by running the
768 * configured release agent with the name of the cgroup (path
769 * relative to the root of cgroup file system) as the argument.
770 *
771 * Most likely, this user command will try to rmdir this cgroup.
772 *
773 * This races with the possibility that some other task will be
774 * attached to this cgroup before it is removed, or that some other
775 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
776 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
777 * unused, and this cgroup will be reprieved from its death sentence,
778 * to continue to serve a useful existence. Next time it's released,
779 * we will get notified again, if it still has 'notify_on_release' set.
780 *
781 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
782 * means only wait until the task is successfully execve()'d. The
783 * separate release agent task is forked by call_usermodehelper(),
784 * then control in this thread returns here, without waiting for the
785 * release agent task. We don't bother to wait because the caller of
786 * this routine has no use for the exit status of the release agent
787 * task, so no sense holding our caller up for that.
788 */
789void cgroup1_release_agent(struct work_struct *work)
790{
791 struct cgroup *cgrp =
792 container_of(work, struct cgroup, release_agent_work);
793 char *pathbuf, *agentbuf;
794 char *argv[3], *envp[3];
795 int ret;
796
797 /* snoop agent path and exit early if empty */
798 if (!cgrp->root->release_agent_path[0])
799 return;
800
801 /* prepare argument buffers */
802 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
803 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
804 if (!pathbuf || !agentbuf)
805 goto out_free;
806
807 spin_lock(&release_agent_path_lock);
808 strscpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
809 spin_unlock(&release_agent_path_lock);
810 if (!agentbuf[0])
811 goto out_free;
812
813 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
814 if (ret < 0)
815 goto out_free;
816
817 argv[0] = agentbuf;
818 argv[1] = pathbuf;
819 argv[2] = NULL;
820
821 /* minimal command environment */
822 envp[0] = "HOME=/";
823 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
824 envp[2] = NULL;
825
826 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
827out_free:
828 kfree(agentbuf);
829 kfree(pathbuf);
830}
831
832/*
833 * cgroup_rename - Only allow simple rename of directories in place.
834 */
835static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
836 const char *new_name_str)
837{
838 struct cgroup *cgrp = kn->priv;
839 int ret;
840
841 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
842 if (strchr(new_name_str, '\n'))
843 return -EINVAL;
844
845 if (kernfs_type(kn) != KERNFS_DIR)
846 return -ENOTDIR;
847 if (kn->parent != new_parent)
848 return -EIO;
849
850 /*
851 * We're gonna grab cgroup_mutex which nests outside kernfs
852 * active_ref. kernfs_rename() doesn't require active_ref
853 * protection. Break them before grabbing cgroup_mutex.
854 */
855 kernfs_break_active_protection(new_parent);
856 kernfs_break_active_protection(kn);
857
858 cgroup_lock();
859
860 ret = kernfs_rename(kn, new_parent, new_name_str);
861 if (!ret)
862 TRACE_CGROUP_PATH(rename, cgrp);
863
864 cgroup_unlock();
865
866 kernfs_unbreak_active_protection(kn);
867 kernfs_unbreak_active_protection(new_parent);
868 return ret;
869}
870
871static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
872{
873 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
874 struct cgroup_subsys *ss;
875 int ssid;
876
877 for_each_subsys(ss, ssid)
878 if (root->subsys_mask & (1 << ssid))
879 seq_show_option(seq, ss->legacy_name, NULL);
880 if (root->flags & CGRP_ROOT_NOPREFIX)
881 seq_puts(seq, ",noprefix");
882 if (root->flags & CGRP_ROOT_XATTR)
883 seq_puts(seq, ",xattr");
884 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
885 seq_puts(seq, ",cpuset_v2_mode");
886 if (root->flags & CGRP_ROOT_FAVOR_DYNMODS)
887 seq_puts(seq, ",favordynmods");
888
889 spin_lock(&release_agent_path_lock);
890 if (strlen(root->release_agent_path))
891 seq_show_option(seq, "release_agent",
892 root->release_agent_path);
893 spin_unlock(&release_agent_path_lock);
894
895 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
896 seq_puts(seq, ",clone_children");
897 if (strlen(root->name))
898 seq_show_option(seq, "name", root->name);
899 return 0;
900}
901
902enum cgroup1_param {
903 Opt_all,
904 Opt_clone_children,
905 Opt_cpuset_v2_mode,
906 Opt_name,
907 Opt_none,
908 Opt_noprefix,
909 Opt_release_agent,
910 Opt_xattr,
911 Opt_favordynmods,
912 Opt_nofavordynmods,
913};
914
915const struct fs_parameter_spec cgroup1_fs_parameters[] = {
916 fsparam_flag ("all", Opt_all),
917 fsparam_flag ("clone_children", Opt_clone_children),
918 fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
919 fsparam_string("name", Opt_name),
920 fsparam_flag ("none", Opt_none),
921 fsparam_flag ("noprefix", Opt_noprefix),
922 fsparam_string("release_agent", Opt_release_agent),
923 fsparam_flag ("xattr", Opt_xattr),
924 fsparam_flag ("favordynmods", Opt_favordynmods),
925 fsparam_flag ("nofavordynmods", Opt_nofavordynmods),
926 {}
927};
928
929int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
930{
931 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
932 struct cgroup_subsys *ss;
933 struct fs_parse_result result;
934 int opt, i;
935
936 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
937 if (opt == -ENOPARAM) {
938 int ret;
939
940 ret = vfs_parse_fs_param_source(fc, param);
941 if (ret != -ENOPARAM)
942 return ret;
943 for_each_subsys(ss, i) {
944 if (strcmp(param->key, ss->legacy_name) ||
945 cgroup1_subsys_absent(ss))
946 continue;
947 if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i))
948 return invalfc(fc, "Disabled controller '%s'",
949 param->key);
950 ctx->subsys_mask |= (1 << i);
951 return 0;
952 }
953 return invalfc(fc, "Unknown subsys name '%s'", param->key);
954 }
955 if (opt < 0)
956 return opt;
957
958 switch (opt) {
959 case Opt_none:
960 /* Explicitly have no subsystems */
961 ctx->none = true;
962 break;
963 case Opt_all:
964 ctx->all_ss = true;
965 break;
966 case Opt_noprefix:
967 ctx->flags |= CGRP_ROOT_NOPREFIX;
968 break;
969 case Opt_clone_children:
970 ctx->cpuset_clone_children = true;
971 break;
972 case Opt_cpuset_v2_mode:
973 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
974 break;
975 case Opt_xattr:
976 ctx->flags |= CGRP_ROOT_XATTR;
977 break;
978 case Opt_favordynmods:
979 ctx->flags |= CGRP_ROOT_FAVOR_DYNMODS;
980 break;
981 case Opt_nofavordynmods:
982 ctx->flags &= ~CGRP_ROOT_FAVOR_DYNMODS;
983 break;
984 case Opt_release_agent:
985 /* Specifying two release agents is forbidden */
986 if (ctx->release_agent)
987 return invalfc(fc, "release_agent respecified");
988 /*
989 * Release agent gets called with all capabilities,
990 * require capabilities to set release agent.
991 */
992 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN))
993 return invalfc(fc, "Setting release_agent not allowed");
994 ctx->release_agent = param->string;
995 param->string = NULL;
996 break;
997 case Opt_name:
998 /* blocked by boot param? */
999 if (cgroup_no_v1_named)
1000 return -ENOENT;
1001 /* Can't specify an empty name */
1002 if (!param->size)
1003 return invalfc(fc, "Empty name");
1004 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
1005 return invalfc(fc, "Name too long");
1006 /* Must match [\w.-]+ */
1007 for (i = 0; i < param->size; i++) {
1008 char c = param->string[i];
1009 if (isalnum(c))
1010 continue;
1011 if ((c == '.') || (c == '-') || (c == '_'))
1012 continue;
1013 return invalfc(fc, "Invalid name");
1014 }
1015 /* Specifying two names is forbidden */
1016 if (ctx->name)
1017 return invalfc(fc, "name respecified");
1018 ctx->name = param->string;
1019 param->string = NULL;
1020 break;
1021 }
1022 return 0;
1023}
1024
1025static int check_cgroupfs_options(struct fs_context *fc)
1026{
1027 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1028 u16 mask = U16_MAX;
1029 u16 enabled = 0;
1030 struct cgroup_subsys *ss;
1031 int i;
1032
1033#ifdef CONFIG_CPUSETS
1034 mask = ~((u16)1 << cpuset_cgrp_id);
1035#endif
1036 for_each_subsys(ss, i)
1037 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i) &&
1038 !cgroup1_subsys_absent(ss))
1039 enabled |= 1 << i;
1040
1041 ctx->subsys_mask &= enabled;
1042
1043 /*
1044 * In absence of 'none', 'name=' and subsystem name options,
1045 * let's default to 'all'.
1046 */
1047 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1048 ctx->all_ss = true;
1049
1050 if (ctx->all_ss) {
1051 /* Mutually exclusive option 'all' + subsystem name */
1052 if (ctx->subsys_mask)
1053 return invalfc(fc, "subsys name conflicts with all");
1054 /* 'all' => select all the subsystems */
1055 ctx->subsys_mask = enabled;
1056 }
1057
1058 /*
1059 * We either have to specify by name or by subsystems. (So all
1060 * empty hierarchies must have a name).
1061 */
1062 if (!ctx->subsys_mask && !ctx->name)
1063 return invalfc(fc, "Need name or subsystem set");
1064
1065 /*
1066 * Option noprefix was introduced just for backward compatibility
1067 * with the old cpuset, so we allow noprefix only if mounting just
1068 * the cpuset subsystem.
1069 */
1070 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1071 return invalfc(fc, "noprefix used incorrectly");
1072
1073 /* Can't specify "none" and some subsystems */
1074 if (ctx->subsys_mask && ctx->none)
1075 return invalfc(fc, "none used incorrectly");
1076
1077 return 0;
1078}
1079
1080int cgroup1_reconfigure(struct fs_context *fc)
1081{
1082 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1083 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1084 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1085 int ret = 0;
1086 u16 added_mask, removed_mask;
1087
1088 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1089
1090 /* See what subsystems are wanted */
1091 ret = check_cgroupfs_options(fc);
1092 if (ret)
1093 goto out_unlock;
1094
1095 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1096 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1097 task_tgid_nr(current), current->comm);
1098
1099 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1100 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1101
1102 /* Don't allow flags or name to change at remount */
1103 if ((ctx->flags ^ root->flags) ||
1104 (ctx->name && strcmp(ctx->name, root->name))) {
1105 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1106 ctx->flags, ctx->name ?: "", root->flags, root->name);
1107 ret = -EINVAL;
1108 goto out_unlock;
1109 }
1110
1111 /* remounting is not allowed for populated hierarchies */
1112 if (!list_empty(&root->cgrp.self.children)) {
1113 ret = -EBUSY;
1114 goto out_unlock;
1115 }
1116
1117 ret = rebind_subsystems(root, added_mask);
1118 if (ret)
1119 goto out_unlock;
1120
1121 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1122
1123 if (ctx->release_agent) {
1124 spin_lock(&release_agent_path_lock);
1125 strcpy(root->release_agent_path, ctx->release_agent);
1126 spin_unlock(&release_agent_path_lock);
1127 }
1128
1129 trace_cgroup_remount(root);
1130
1131 out_unlock:
1132 cgroup_unlock();
1133 return ret;
1134}
1135
1136struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1137 .rename = cgroup1_rename,
1138 .show_options = cgroup1_show_options,
1139 .mkdir = cgroup_mkdir,
1140 .rmdir = cgroup_rmdir,
1141 .show_path = cgroup_show_path,
1142};
1143
1144/*
1145 * The guts of cgroup1 mount - find or create cgroup_root to use.
1146 * Called with cgroup_mutex held; returns 0 on success, -E... on
1147 * error and positive - in case when the candidate is busy dying.
1148 * On success it stashes a reference to cgroup_root into given
1149 * cgroup_fs_context; that reference is *NOT* counting towards the
1150 * cgroup_root refcount.
1151 */
1152static int cgroup1_root_to_use(struct fs_context *fc)
1153{
1154 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1155 struct cgroup_root *root;
1156 struct cgroup_subsys *ss;
1157 int i, ret;
1158
1159 /* First find the desired set of subsystems */
1160 ret = check_cgroupfs_options(fc);
1161 if (ret)
1162 return ret;
1163
1164 /*
1165 * Destruction of cgroup root is asynchronous, so subsystems may
1166 * still be dying after the previous unmount. Let's drain the
1167 * dying subsystems. We just need to ensure that the ones
1168 * unmounted previously finish dying and don't care about new ones
1169 * starting. Testing ref liveliness is good enough.
1170 */
1171 for_each_subsys(ss, i) {
1172 if (!(ctx->subsys_mask & (1 << i)) ||
1173 ss->root == &cgrp_dfl_root)
1174 continue;
1175
1176 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1177 return 1; /* restart */
1178 cgroup_put(&ss->root->cgrp);
1179 }
1180
1181 for_each_root(root) {
1182 bool name_match = false;
1183
1184 if (root == &cgrp_dfl_root)
1185 continue;
1186
1187 /*
1188 * If we asked for a name then it must match. Also, if
1189 * name matches but sybsys_mask doesn't, we should fail.
1190 * Remember whether name matched.
1191 */
1192 if (ctx->name) {
1193 if (strcmp(ctx->name, root->name))
1194 continue;
1195 name_match = true;
1196 }
1197
1198 /*
1199 * If we asked for subsystems (or explicitly for no
1200 * subsystems) then they must match.
1201 */
1202 if ((ctx->subsys_mask || ctx->none) &&
1203 (ctx->subsys_mask != root->subsys_mask)) {
1204 if (!name_match)
1205 continue;
1206 return -EBUSY;
1207 }
1208
1209 if (root->flags ^ ctx->flags)
1210 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1211
1212 ctx->root = root;
1213 return 0;
1214 }
1215
1216 /*
1217 * No such thing, create a new one. name= matching without subsys
1218 * specification is allowed for already existing hierarchies but we
1219 * can't create new one without subsys specification.
1220 */
1221 if (!ctx->subsys_mask && !ctx->none)
1222 return invalfc(fc, "No subsys list or none specified");
1223
1224 /* Hierarchies may only be created in the initial cgroup namespace. */
1225 if (ctx->ns != &init_cgroup_ns)
1226 return -EPERM;
1227
1228 root = kzalloc(sizeof(*root), GFP_KERNEL);
1229 if (!root)
1230 return -ENOMEM;
1231
1232 ctx->root = root;
1233 init_cgroup_root(ctx);
1234
1235 ret = cgroup_setup_root(root, ctx->subsys_mask);
1236 if (!ret)
1237 cgroup_favor_dynmods(root, ctx->flags & CGRP_ROOT_FAVOR_DYNMODS);
1238 else
1239 cgroup_free_root(root);
1240
1241 return ret;
1242}
1243
1244int cgroup1_get_tree(struct fs_context *fc)
1245{
1246 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1247 int ret;
1248
1249 /* Check if the caller has permission to mount. */
1250 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1251 return -EPERM;
1252
1253 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1254
1255 ret = cgroup1_root_to_use(fc);
1256 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1257 ret = 1; /* restart */
1258
1259 cgroup_unlock();
1260
1261 if (!ret)
1262 ret = cgroup_do_get_tree(fc);
1263
1264 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1265 fc_drop_locked(fc);
1266 ret = 1;
1267 }
1268
1269 if (unlikely(ret > 0)) {
1270 msleep(10);
1271 return restart_syscall();
1272 }
1273 return ret;
1274}
1275
1276/**
1277 * task_get_cgroup1 - Acquires the associated cgroup of a task within a
1278 * specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
1279 * hierarchy ID.
1280 * @tsk: The target task
1281 * @hierarchy_id: The ID of a cgroup1 hierarchy
1282 *
1283 * On success, the cgroup is returned. On failure, ERR_PTR is returned.
1284 * We limit it to cgroup1 only.
1285 */
1286struct cgroup *task_get_cgroup1(struct task_struct *tsk, int hierarchy_id)
1287{
1288 struct cgroup *cgrp = ERR_PTR(-ENOENT);
1289 struct cgroup_root *root;
1290 unsigned long flags;
1291
1292 rcu_read_lock();
1293 for_each_root(root) {
1294 /* cgroup1 only*/
1295 if (root == &cgrp_dfl_root)
1296 continue;
1297 if (root->hierarchy_id != hierarchy_id)
1298 continue;
1299 spin_lock_irqsave(&css_set_lock, flags);
1300 cgrp = task_cgroup_from_root(tsk, root);
1301 if (!cgrp || !cgroup_tryget(cgrp))
1302 cgrp = ERR_PTR(-ENOENT);
1303 spin_unlock_irqrestore(&css_set_lock, flags);
1304 break;
1305 }
1306 rcu_read_unlock();
1307 return cgrp;
1308}
1309
1310static int __init cgroup1_wq_init(void)
1311{
1312 /*
1313 * Used to destroy pidlists and separate to serve as flush domain.
1314 * Cap @max_active to 1 too.
1315 */
1316 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1317 0, 1);
1318 BUG_ON(!cgroup_pidlist_destroy_wq);
1319 return 0;
1320}
1321core_initcall(cgroup1_wq_init);
1322
1323static int __init cgroup_no_v1(char *str)
1324{
1325 struct cgroup_subsys *ss;
1326 char *token;
1327 int i;
1328
1329 while ((token = strsep(&str, ",")) != NULL) {
1330 if (!*token)
1331 continue;
1332
1333 if (!strcmp(token, "all")) {
1334 cgroup_no_v1_mask = U16_MAX;
1335 continue;
1336 }
1337
1338 if (!strcmp(token, "named")) {
1339 cgroup_no_v1_named = true;
1340 continue;
1341 }
1342
1343 for_each_subsys(ss, i) {
1344 if (strcmp(token, ss->name) &&
1345 strcmp(token, ss->legacy_name))
1346 continue;
1347
1348 cgroup_no_v1_mask |= 1 << i;
1349 break;
1350 }
1351 }
1352 return 1;
1353}
1354__setup("cgroup_no_v1=", cgroup_no_v1);