Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
0a268dbd TH |
2 | #include "cgroup-internal.h" |
3 | ||
1592c9b2 | 4 | #include <linux/ctype.h> |
0a268dbd TH |
5 | #include <linux/kmod.h> |
6 | #include <linux/sort.h> | |
1592c9b2 | 7 | #include <linux/delay.h> |
0a268dbd | 8 | #include <linux/mm.h> |
c3edc401 | 9 | #include <linux/sched/signal.h> |
56cd6973 | 10 | #include <linux/sched/task.h> |
50ff9d13 | 11 | #include <linux/magic.h> |
0a268dbd TH |
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> | |
8d2451f4 | 17 | #include <linux/fs_parser.h> |
0a268dbd TH |
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 */ | |
30 | static u16 cgroup_no_v1_mask; | |
31 | ||
3fc9c12d TH |
32 | /* disable named v1 mounts */ |
33 | static bool cgroup_no_v1_named; | |
34 | ||
0a268dbd TH |
35 | /* |
36 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
37 | * separate workqueue as flush domain. | |
38 | */ | |
39 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
40 | ||
e7b20d97 | 41 | /* protects cgroup_subsys->release_agent_path */ |
1592c9b2 | 42 | static DEFINE_SPINLOCK(release_agent_path_lock); |
0a268dbd | 43 | |
d62beb7f | 44 | bool cgroup1_ssid_disabled(int ssid) |
0a268dbd TH |
45 | { |
46 | return cgroup_no_v1_mask & (1 << ssid); | |
47 | } | |
48 | ||
49 | /** | |
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 | |
b4cc6196 RD |
53 | * |
54 | * Return: %0 on success or a negative errno code on failure | |
0a268dbd TH |
55 | */ |
56 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
57 | { | |
58 | struct cgroup_root *root; | |
59 | int retval = 0; | |
60 | ||
61 | mutex_lock(&cgroup_mutex); | |
62 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
63 | for_each_root(root) { | |
64 | struct cgroup *from_cgrp; | |
65 | ||
0a268dbd TH |
66 | spin_lock_irq(&css_set_lock); |
67 | from_cgrp = task_cgroup_from_root(from, root); | |
68 | spin_unlock_irq(&css_set_lock); | |
69 | ||
70 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
71 | if (retval) | |
72 | break; | |
73 | } | |
74 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
75 | mutex_unlock(&cgroup_mutex); | |
76 | ||
77 | return retval; | |
78 | } | |
79 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
80 | ||
81 | /** | |
b4cc6196 | 82 | * cgroup_transfer_tasks - move tasks from one cgroup to another |
0a268dbd TH |
83 | * @to: cgroup to which the tasks will be moved |
84 | * @from: cgroup in which the tasks currently reside | |
85 | * | |
86 | * Locking rules between cgroup_post_fork() and the migration path | |
87 | * guarantee that, if a task is forking while being migrated, the new child | |
88 | * is guaranteed to be either visible in the source cgroup after the | |
89 | * parent's migration is complete or put into the target cgroup. No task | |
90 | * can slip out of migration through forking. | |
b4cc6196 RD |
91 | * |
92 | * Return: %0 on success or a negative errno code on failure | |
0a268dbd TH |
93 | */ |
94 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
95 | { | |
e595cd70 | 96 | DEFINE_CGROUP_MGCTX(mgctx); |
0a268dbd TH |
97 | struct cgrp_cset_link *link; |
98 | struct css_task_iter it; | |
99 | struct task_struct *task; | |
100 | int ret; | |
101 | ||
102 | if (cgroup_on_dfl(to)) | |
103 | return -EINVAL; | |
104 | ||
8cfd8147 TH |
105 | ret = cgroup_migrate_vet_dst(to); |
106 | if (ret) | |
107 | return ret; | |
0a268dbd TH |
108 | |
109 | mutex_lock(&cgroup_mutex); | |
110 | ||
111 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
112 | ||
113 | /* all tasks in @from are being moved, all csets are source */ | |
114 | spin_lock_irq(&css_set_lock); | |
115 | list_for_each_entry(link, &from->cset_links, cset_link) | |
e595cd70 | 116 | cgroup_migrate_add_src(link->cset, to, &mgctx); |
0a268dbd TH |
117 | spin_unlock_irq(&css_set_lock); |
118 | ||
e595cd70 | 119 | ret = cgroup_migrate_prepare_dst(&mgctx); |
0a268dbd TH |
120 | if (ret) |
121 | goto out_err; | |
122 | ||
123 | /* | |
124 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
125 | * ->can_attach() fails. | |
126 | */ | |
127 | do { | |
bc2fb7ed | 128 | css_task_iter_start(&from->self, 0, &it); |
116d2f74 PS |
129 | |
130 | do { | |
131 | task = css_task_iter_next(&it); | |
132 | } while (task && (task->flags & PF_EXITING)); | |
133 | ||
0a268dbd TH |
134 | if (task) |
135 | get_task_struct(task); | |
136 | css_task_iter_end(&it); | |
137 | ||
138 | if (task) { | |
bfc2cf6f | 139 | ret = cgroup_migrate(task, false, &mgctx); |
0a268dbd | 140 | if (!ret) |
e4f8d81c | 141 | TRACE_CGROUP_PATH(transfer_tasks, to, task, false); |
0a268dbd TH |
142 | put_task_struct(task); |
143 | } | |
144 | } while (task && !ret); | |
145 | out_err: | |
e595cd70 | 146 | cgroup_migrate_finish(&mgctx); |
0a268dbd TH |
147 | percpu_up_write(&cgroup_threadgroup_rwsem); |
148 | mutex_unlock(&cgroup_mutex); | |
149 | return ret; | |
150 | } | |
151 | ||
152 | /* | |
153 | * Stuff for reading the 'tasks'/'procs' files. | |
154 | * | |
155 | * Reading this file can return large amounts of data if a cgroup has | |
156 | * *lots* of attached tasks. So it may need several calls to read(), | |
157 | * but we cannot guarantee that the information we produce is correct | |
158 | * unless we produce it entirely atomically. | |
159 | * | |
160 | */ | |
161 | ||
162 | /* which pidlist file are we talking about? */ | |
163 | enum cgroup_filetype { | |
164 | CGROUP_FILE_PROCS, | |
165 | CGROUP_FILE_TASKS, | |
166 | }; | |
167 | ||
168 | /* | |
169 | * A pidlist is a list of pids that virtually represents the contents of one | |
170 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
171 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
172 | * to the cgroup. | |
173 | */ | |
174 | struct cgroup_pidlist { | |
175 | /* | |
176 | * used to find which pidlist is wanted. doesn't change as long as | |
177 | * this particular list stays in the list. | |
178 | */ | |
179 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
180 | /* array of xids */ | |
181 | pid_t *list; | |
182 | /* how many elements the above list has */ | |
183 | int length; | |
184 | /* each of these stored in a list by its cgroup */ | |
185 | struct list_head links; | |
186 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
187 | struct cgroup *owner; | |
188 | /* for delayed destruction */ | |
189 | struct delayed_work destroy_dwork; | |
190 | }; | |
191 | ||
0a268dbd TH |
192 | /* |
193 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
194 | * should be left afterwards. | |
195 | */ | |
d62beb7f | 196 | void cgroup1_pidlist_destroy_all(struct cgroup *cgrp) |
0a268dbd TH |
197 | { |
198 | struct cgroup_pidlist *l, *tmp_l; | |
199 | ||
200 | mutex_lock(&cgrp->pidlist_mutex); | |
201 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
202 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
203 | mutex_unlock(&cgrp->pidlist_mutex); | |
204 | ||
205 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
206 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
207 | } | |
208 | ||
209 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
210 | { | |
211 | struct delayed_work *dwork = to_delayed_work(work); | |
212 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
213 | destroy_dwork); | |
214 | struct cgroup_pidlist *tofree = NULL; | |
215 | ||
216 | mutex_lock(&l->owner->pidlist_mutex); | |
217 | ||
218 | /* | |
219 | * Destroy iff we didn't get queued again. The state won't change | |
220 | * as destroy_dwork can only be queued while locked. | |
221 | */ | |
222 | if (!delayed_work_pending(dwork)) { | |
223 | list_del(&l->links); | |
653a23ca | 224 | kvfree(l->list); |
0a268dbd TH |
225 | put_pid_ns(l->key.ns); |
226 | tofree = l; | |
227 | } | |
228 | ||
229 | mutex_unlock(&l->owner->pidlist_mutex); | |
230 | kfree(tofree); | |
231 | } | |
232 | ||
233 | /* | |
234 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
235 | * Returns the number of unique elements. | |
236 | */ | |
237 | static int pidlist_uniq(pid_t *list, int length) | |
238 | { | |
239 | int src, dest = 1; | |
240 | ||
241 | /* | |
242 | * we presume the 0th element is unique, so i starts at 1. trivial | |
243 | * edge cases first; no work needs to be done for either | |
244 | */ | |
245 | if (length == 0 || length == 1) | |
246 | return length; | |
247 | /* src and dest walk down the list; dest counts unique elements */ | |
248 | for (src = 1; src < length; src++) { | |
249 | /* find next unique element */ | |
250 | while (list[src] == list[src-1]) { | |
251 | src++; | |
252 | if (src == length) | |
253 | goto after; | |
254 | } | |
255 | /* dest always points to where the next unique element goes */ | |
256 | list[dest] = list[src]; | |
257 | dest++; | |
258 | } | |
259 | after: | |
260 | return dest; | |
261 | } | |
262 | ||
263 | /* | |
264 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
265 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
266 | * different per namespace, each namespace needs differently sorted list, | |
267 | * making it impossible to use, for example, single rbtree of member tasks | |
268 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
269 | * per open file is dangerous, so cgroup had to implement shared pool of | |
270 | * pidlists keyed by cgroup and namespace. | |
271 | */ | |
272 | static int cmppid(const void *a, const void *b) | |
273 | { | |
274 | return *(pid_t *)a - *(pid_t *)b; | |
275 | } | |
276 | ||
277 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
278 | enum cgroup_filetype type) | |
279 | { | |
280 | struct cgroup_pidlist *l; | |
281 | /* don't need task_nsproxy() if we're looking at ourself */ | |
282 | struct pid_namespace *ns = task_active_pid_ns(current); | |
283 | ||
284 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
285 | ||
286 | list_for_each_entry(l, &cgrp->pidlists, links) | |
287 | if (l->key.type == type && l->key.ns == ns) | |
288 | return l; | |
289 | return NULL; | |
290 | } | |
291 | ||
292 | /* | |
293 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
294 | * returns with the lock on that pidlist already held, and takes care | |
295 | * of the use count, or returns NULL with no locks held if we're out of | |
296 | * memory. | |
297 | */ | |
298 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
299 | enum cgroup_filetype type) | |
300 | { | |
301 | struct cgroup_pidlist *l; | |
302 | ||
303 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
304 | ||
305 | l = cgroup_pidlist_find(cgrp, type); | |
306 | if (l) | |
307 | return l; | |
308 | ||
309 | /* entry not found; create a new one */ | |
310 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
311 | if (!l) | |
312 | return l; | |
313 | ||
314 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
315 | l->key.type = type; | |
316 | /* don't need task_nsproxy() if we're looking at ourself */ | |
317 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
318 | l->owner = cgrp; | |
319 | list_add(&l->links, &cgrp->pidlists); | |
320 | return l; | |
321 | } | |
322 | ||
0a268dbd TH |
323 | /* |
324 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
325 | */ | |
326 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
327 | struct cgroup_pidlist **lp) | |
328 | { | |
329 | pid_t *array; | |
330 | int length; | |
331 | int pid, n = 0; /* used for populating the array */ | |
332 | struct css_task_iter it; | |
333 | struct task_struct *tsk; | |
334 | struct cgroup_pidlist *l; | |
335 | ||
336 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
337 | ||
338 | /* | |
339 | * If cgroup gets more users after we read count, we won't have | |
340 | * enough space - tough. This race is indistinguishable to the | |
341 | * caller from the case that the additional cgroup users didn't | |
342 | * show up until sometime later on. | |
343 | */ | |
344 | length = cgroup_task_count(cgrp); | |
653a23ca | 345 | array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL); |
0a268dbd TH |
346 | if (!array) |
347 | return -ENOMEM; | |
348 | /* now, populate the array */ | |
bc2fb7ed | 349 | css_task_iter_start(&cgrp->self, 0, &it); |
0a268dbd TH |
350 | while ((tsk = css_task_iter_next(&it))) { |
351 | if (unlikely(n == length)) | |
352 | break; | |
353 | /* get tgid or pid for procs or tasks file respectively */ | |
354 | if (type == CGROUP_FILE_PROCS) | |
355 | pid = task_tgid_vnr(tsk); | |
356 | else | |
357 | pid = task_pid_vnr(tsk); | |
358 | if (pid > 0) /* make sure to only use valid results */ | |
359 | array[n++] = pid; | |
360 | } | |
361 | css_task_iter_end(&it); | |
362 | length = n; | |
363 | /* now sort & (if procs) strip out duplicates */ | |
364 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
365 | if (type == CGROUP_FILE_PROCS) | |
366 | length = pidlist_uniq(array, length); | |
367 | ||
368 | l = cgroup_pidlist_find_create(cgrp, type); | |
369 | if (!l) { | |
653a23ca | 370 | kvfree(array); |
0a268dbd TH |
371 | return -ENOMEM; |
372 | } | |
373 | ||
374 | /* store array, freeing old if necessary */ | |
653a23ca | 375 | kvfree(l->list); |
0a268dbd TH |
376 | l->list = array; |
377 | l->length = length; | |
378 | *lp = l; | |
379 | return 0; | |
380 | } | |
381 | ||
382 | /* | |
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. | |
386 | */ | |
387 | ||
388 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
389 | { | |
390 | /* | |
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 | |
395 | */ | |
396 | struct kernfs_open_file *of = s->private; | |
0d2b5955 | 397 | struct cgroup_file_ctx *ctx = of->priv; |
0a268dbd TH |
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; | |
402 | int *iter, ret; | |
403 | ||
404 | mutex_lock(&cgrp->pidlist_mutex); | |
405 | ||
406 | /* | |
0d2b5955 TH |
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. | |
0a268dbd | 411 | */ |
0d2b5955 TH |
412 | if (ctx->procs1.pidlist) |
413 | ctx->procs1.pidlist = cgroup_pidlist_find(cgrp, type); | |
0a268dbd TH |
414 | |
415 | /* | |
416 | * Either this is the first start() after open or the matching | |
417 | * pidlist has been destroyed inbetween. Create a new one. | |
418 | */ | |
0d2b5955 TH |
419 | if (!ctx->procs1.pidlist) { |
420 | ret = pidlist_array_load(cgrp, type, &ctx->procs1.pidlist); | |
0a268dbd TH |
421 | if (ret) |
422 | return ERR_PTR(ret); | |
423 | } | |
0d2b5955 | 424 | l = ctx->procs1.pidlist; |
0a268dbd TH |
425 | |
426 | if (pid) { | |
427 | int end = l->length; | |
428 | ||
429 | while (index < end) { | |
430 | int mid = (index + end) / 2; | |
431 | if (l->list[mid] == pid) { | |
432 | index = mid; | |
433 | break; | |
434 | } else if (l->list[mid] <= pid) | |
435 | index = mid + 1; | |
436 | else | |
437 | end = mid; | |
438 | } | |
439 | } | |
440 | /* If we're off the end of the array, we're done */ | |
441 | if (index >= l->length) | |
442 | return NULL; | |
443 | /* Update the abstract position to be the actual pid that we found */ | |
444 | iter = l->list + index; | |
445 | *pos = *iter; | |
446 | return iter; | |
447 | } | |
448 | ||
449 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
450 | { | |
451 | struct kernfs_open_file *of = s->private; | |
0d2b5955 TH |
452 | struct cgroup_file_ctx *ctx = of->priv; |
453 | struct cgroup_pidlist *l = ctx->procs1.pidlist; | |
0a268dbd TH |
454 | |
455 | if (l) | |
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); | |
459 | } | |
460 | ||
461 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
462 | { | |
463 | struct kernfs_open_file *of = s->private; | |
0d2b5955 TH |
464 | struct cgroup_file_ctx *ctx = of->priv; |
465 | struct cgroup_pidlist *l = ctx->procs1.pidlist; | |
0a268dbd TH |
466 | pid_t *p = v; |
467 | pid_t *end = l->list + l->length; | |
468 | /* | |
469 | * Advance to the next pid in the array. If this goes off the | |
470 | * end, we're done | |
471 | */ | |
472 | p++; | |
473 | if (p >= end) { | |
db8dd969 | 474 | (*pos)++; |
0a268dbd TH |
475 | return NULL; |
476 | } else { | |
477 | *pos = *p; | |
478 | return p; | |
479 | } | |
480 | } | |
481 | ||
482 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
483 | { | |
484 | seq_printf(s, "%d\n", *(int *)v); | |
485 | ||
486 | return 0; | |
487 | } | |
488 | ||
715c809d TH |
489 | static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of, |
490 | char *buf, size_t nbytes, loff_t off, | |
491 | bool threadgroup) | |
0a268dbd | 492 | { |
715c809d TH |
493 | struct cgroup *cgrp; |
494 | struct task_struct *task; | |
495 | const struct cred *cred, *tcred; | |
496 | ssize_t ret; | |
9a3284fa | 497 | bool locked; |
715c809d TH |
498 | |
499 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
500 | if (!cgrp) | |
501 | return -ENODEV; | |
502 | ||
9a3284fa | 503 | task = cgroup_procs_write_start(buf, threadgroup, &locked); |
715c809d TH |
504 | ret = PTR_ERR_OR_ZERO(task); |
505 | if (ret) | |
506 | goto out_unlock; | |
507 | ||
508 | /* | |
1756d799 TH |
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. | |
715c809d | 512 | */ |
1756d799 | 513 | cred = of->file->f_cred; |
715c809d TH |
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)) | |
518 | ret = -EACCES; | |
519 | put_cred(tcred); | |
520 | if (ret) | |
521 | goto out_finish; | |
522 | ||
523 | ret = cgroup_attach_task(cgrp, task, threadgroup); | |
524 | ||
525 | out_finish: | |
9a3284fa | 526 | cgroup_procs_write_finish(task, locked); |
715c809d TH |
527 | out_unlock: |
528 | cgroup_kn_unlock(of->kn); | |
529 | ||
530 | return ret ?: nbytes; | |
531 | } | |
532 | ||
533 | static ssize_t cgroup1_procs_write(struct kernfs_open_file *of, | |
534 | char *buf, size_t nbytes, loff_t off) | |
535 | { | |
536 | return __cgroup1_procs_write(of, buf, nbytes, off, true); | |
537 | } | |
538 | ||
539 | static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of, | |
540 | char *buf, size_t nbytes, loff_t off) | |
541 | { | |
542 | return __cgroup1_procs_write(of, buf, nbytes, off, false); | |
0a268dbd TH |
543 | } |
544 | ||
545 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
546 | char *buf, size_t nbytes, loff_t off) | |
547 | { | |
548 | struct cgroup *cgrp; | |
549 | ||
550 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
551 | ||
24f60085 EB |
552 | /* |
553 | * Release agent gets called with all capabilities, | |
554 | * require capabilities to set release agent. | |
555 | */ | |
556 | if ((of->file->f_cred->user_ns != &init_user_ns) || | |
557 | !capable(CAP_SYS_ADMIN)) | |
558 | return -EPERM; | |
559 | ||
0a268dbd TH |
560 | cgrp = cgroup_kn_lock_live(of->kn, false); |
561 | if (!cgrp) | |
562 | return -ENODEV; | |
563 | spin_lock(&release_agent_path_lock); | |
564 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
565 | sizeof(cgrp->root->release_agent_path)); | |
566 | spin_unlock(&release_agent_path_lock); | |
567 | cgroup_kn_unlock(of->kn); | |
568 | return nbytes; | |
569 | } | |
570 | ||
571 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
572 | { | |
573 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
574 | ||
575 | spin_lock(&release_agent_path_lock); | |
576 | seq_puts(seq, cgrp->root->release_agent_path); | |
577 | spin_unlock(&release_agent_path_lock); | |
578 | seq_putc(seq, '\n'); | |
579 | return 0; | |
580 | } | |
581 | ||
582 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
583 | { | |
584 | seq_puts(seq, "0\n"); | |
585 | return 0; | |
586 | } | |
587 | ||
588 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
589 | struct cftype *cft) | |
590 | { | |
591 | return notify_on_release(css->cgroup); | |
592 | } | |
593 | ||
594 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
595 | struct cftype *cft, u64 val) | |
596 | { | |
597 | if (val) | |
598 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
599 | else | |
600 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
601 | return 0; | |
602 | } | |
603 | ||
604 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
605 | struct cftype *cft) | |
606 | { | |
607 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
608 | } | |
609 | ||
610 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
611 | struct cftype *cft, u64 val) | |
612 | { | |
613 | if (val) | |
614 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
615 | else | |
616 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
617 | return 0; | |
618 | } | |
619 | ||
620 | /* cgroup core interface files for the legacy hierarchies */ | |
d62beb7f | 621 | struct cftype cgroup1_base_files[] = { |
0a268dbd TH |
622 | { |
623 | .name = "cgroup.procs", | |
624 | .seq_start = cgroup_pidlist_start, | |
625 | .seq_next = cgroup_pidlist_next, | |
626 | .seq_stop = cgroup_pidlist_stop, | |
627 | .seq_show = cgroup_pidlist_show, | |
628 | .private = CGROUP_FILE_PROCS, | |
715c809d | 629 | .write = cgroup1_procs_write, |
0a268dbd TH |
630 | }, |
631 | { | |
632 | .name = "cgroup.clone_children", | |
633 | .read_u64 = cgroup_clone_children_read, | |
634 | .write_u64 = cgroup_clone_children_write, | |
635 | }, | |
636 | { | |
637 | .name = "cgroup.sane_behavior", | |
638 | .flags = CFTYPE_ONLY_ON_ROOT, | |
639 | .seq_show = cgroup_sane_behavior_show, | |
640 | }, | |
641 | { | |
642 | .name = "tasks", | |
643 | .seq_start = cgroup_pidlist_start, | |
644 | .seq_next = cgroup_pidlist_next, | |
645 | .seq_stop = cgroup_pidlist_stop, | |
646 | .seq_show = cgroup_pidlist_show, | |
647 | .private = CGROUP_FILE_TASKS, | |
715c809d | 648 | .write = cgroup1_tasks_write, |
0a268dbd TH |
649 | }, |
650 | { | |
651 | .name = "notify_on_release", | |
652 | .read_u64 = cgroup_read_notify_on_release, | |
653 | .write_u64 = cgroup_write_notify_on_release, | |
654 | }, | |
655 | { | |
656 | .name = "release_agent", | |
657 | .flags = CFTYPE_ONLY_ON_ROOT, | |
658 | .seq_show = cgroup_release_agent_show, | |
659 | .write = cgroup_release_agent_write, | |
660 | .max_write_len = PATH_MAX - 1, | |
661 | }, | |
662 | { } /* terminate */ | |
663 | }; | |
664 | ||
665 | /* Display information about each subsystem and each hierarchy */ | |
3f3942ac | 666 | int proc_cgroupstats_show(struct seq_file *m, void *v) |
0a268dbd TH |
667 | { |
668 | struct cgroup_subsys *ss; | |
669 | int i; | |
670 | ||
671 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
672 | /* | |
822bc9ba SB |
673 | * Grab the subsystems state racily. No need to add avenue to |
674 | * cgroup_mutex contention. | |
0a268dbd | 675 | */ |
0a268dbd TH |
676 | |
677 | for_each_subsys(ss, i) | |
678 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
679 | ss->legacy_name, ss->root->hierarchy_id, | |
680 | atomic_read(&ss->root->nr_cgrps), | |
681 | cgroup_ssid_enabled(i)); | |
682 | ||
0a268dbd TH |
683 | return 0; |
684 | } | |
685 | ||
0a268dbd TH |
686 | /** |
687 | * cgroupstats_build - build and fill cgroupstats | |
688 | * @stats: cgroupstats to fill information into | |
689 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
690 | * been requested. | |
691 | * | |
692 | * Build and fill cgroupstats so that taskstats can export it to user | |
693 | * space. | |
b4cc6196 RD |
694 | * |
695 | * Return: %0 on success or a negative errno code on failure | |
0a268dbd TH |
696 | */ |
697 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
698 | { | |
699 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
700 | struct cgroup *cgrp; | |
701 | struct css_task_iter it; | |
702 | struct task_struct *tsk; | |
703 | ||
704 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
705 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
706 | kernfs_type(kn) != KERNFS_DIR) | |
707 | return -EINVAL; | |
708 | ||
0a268dbd TH |
709 | /* |
710 | * We aren't being called from kernfs and there's no guarantee on | |
711 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
712 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
713 | */ | |
714 | rcu_read_lock(); | |
e0aed7c7 | 715 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
bb758421 | 716 | if (!cgrp || !cgroup_tryget(cgrp)) { |
0a268dbd | 717 | rcu_read_unlock(); |
0a268dbd TH |
718 | return -ENOENT; |
719 | } | |
720 | rcu_read_unlock(); | |
721 | ||
bc2fb7ed | 722 | css_task_iter_start(&cgrp->self, 0, &it); |
0a268dbd | 723 | while ((tsk = css_task_iter_next(&it))) { |
2f064a59 | 724 | switch (READ_ONCE(tsk->__state)) { |
0a268dbd TH |
725 | case TASK_RUNNING: |
726 | stats->nr_running++; | |
727 | break; | |
728 | case TASK_INTERRUPTIBLE: | |
729 | stats->nr_sleeping++; | |
730 | break; | |
731 | case TASK_UNINTERRUPTIBLE: | |
732 | stats->nr_uninterruptible++; | |
733 | break; | |
734 | case TASK_STOPPED: | |
735 | stats->nr_stopped++; | |
736 | break; | |
737 | default: | |
ffeee417 | 738 | if (tsk->in_iowait) |
0a268dbd TH |
739 | stats->nr_io_wait++; |
740 | break; | |
741 | } | |
742 | } | |
743 | css_task_iter_end(&it); | |
744 | ||
bb758421 | 745 | cgroup_put(cgrp); |
0a268dbd TH |
746 | return 0; |
747 | } | |
748 | ||
d62beb7f | 749 | void cgroup1_check_for_release(struct cgroup *cgrp) |
0a268dbd TH |
750 | { |
751 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
752 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
753 | schedule_work(&cgrp->release_agent_work); | |
754 | } | |
755 | ||
756 | /* | |
757 | * Notify userspace when a cgroup is released, by running the | |
758 | * configured release agent with the name of the cgroup (path | |
759 | * relative to the root of cgroup file system) as the argument. | |
760 | * | |
761 | * Most likely, this user command will try to rmdir this cgroup. | |
762 | * | |
763 | * This races with the possibility that some other task will be | |
764 | * attached to this cgroup before it is removed, or that some other | |
765 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
766 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
767 | * unused, and this cgroup will be reprieved from its death sentence, | |
768 | * to continue to serve a useful existence. Next time it's released, | |
769 | * we will get notified again, if it still has 'notify_on_release' set. | |
770 | * | |
771 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
772 | * means only wait until the task is successfully execve()'d. The | |
773 | * separate release agent task is forked by call_usermodehelper(), | |
774 | * then control in this thread returns here, without waiting for the | |
775 | * release agent task. We don't bother to wait because the caller of | |
776 | * this routine has no use for the exit status of the release agent | |
777 | * task, so no sense holding our caller up for that. | |
778 | */ | |
d62beb7f | 779 | void cgroup1_release_agent(struct work_struct *work) |
0a268dbd TH |
780 | { |
781 | struct cgroup *cgrp = | |
782 | container_of(work, struct cgroup, release_agent_work); | |
e7b20d97 | 783 | char *pathbuf, *agentbuf; |
0a268dbd TH |
784 | char *argv[3], *envp[3]; |
785 | int ret; | |
786 | ||
e7b20d97 TH |
787 | /* snoop agent path and exit early if empty */ |
788 | if (!cgrp->root->release_agent_path[0]) | |
789 | return; | |
0a268dbd | 790 | |
e7b20d97 | 791 | /* prepare argument buffers */ |
0a268dbd | 792 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e7b20d97 TH |
793 | agentbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
794 | if (!pathbuf || !agentbuf) | |
795 | goto out_free; | |
0a268dbd | 796 | |
e7b20d97 TH |
797 | spin_lock(&release_agent_path_lock); |
798 | strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX); | |
799 | spin_unlock(&release_agent_path_lock); | |
800 | if (!agentbuf[0]) | |
801 | goto out_free; | |
802 | ||
803 | ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
0a268dbd | 804 | if (ret < 0 || ret >= PATH_MAX) |
e7b20d97 | 805 | goto out_free; |
0a268dbd TH |
806 | |
807 | argv[0] = agentbuf; | |
808 | argv[1] = pathbuf; | |
809 | argv[2] = NULL; | |
810 | ||
811 | /* minimal command environment */ | |
812 | envp[0] = "HOME=/"; | |
813 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
814 | envp[2] = NULL; | |
815 | ||
0a268dbd | 816 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
0a268dbd TH |
817 | out_free: |
818 | kfree(agentbuf); | |
819 | kfree(pathbuf); | |
820 | } | |
821 | ||
822 | /* | |
823 | * cgroup_rename - Only allow simple rename of directories in place. | |
824 | */ | |
1592c9b2 TH |
825 | static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
826 | const char *new_name_str) | |
0a268dbd TH |
827 | { |
828 | struct cgroup *cgrp = kn->priv; | |
829 | int ret; | |
830 | ||
b7e24eb1 AK |
831 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ |
832 | if (strchr(new_name_str, '\n')) | |
833 | return -EINVAL; | |
834 | ||
0a268dbd TH |
835 | if (kernfs_type(kn) != KERNFS_DIR) |
836 | return -ENOTDIR; | |
837 | if (kn->parent != new_parent) | |
838 | return -EIO; | |
839 | ||
0a268dbd TH |
840 | /* |
841 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
842 | * active_ref. kernfs_rename() doesn't require active_ref | |
843 | * protection. Break them before grabbing cgroup_mutex. | |
844 | */ | |
845 | kernfs_break_active_protection(new_parent); | |
846 | kernfs_break_active_protection(kn); | |
847 | ||
848 | mutex_lock(&cgroup_mutex); | |
849 | ||
850 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
851 | if (!ret) | |
e4f8d81c | 852 | TRACE_CGROUP_PATH(rename, cgrp); |
0a268dbd TH |
853 | |
854 | mutex_unlock(&cgroup_mutex); | |
855 | ||
856 | kernfs_unbreak_active_protection(kn); | |
857 | kernfs_unbreak_active_protection(new_parent); | |
858 | return ret; | |
859 | } | |
860 | ||
1592c9b2 TH |
861 | static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root) |
862 | { | |
863 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
864 | struct cgroup_subsys *ss; | |
865 | int ssid; | |
866 | ||
867 | for_each_subsys(ss, ssid) | |
868 | if (root->subsys_mask & (1 << ssid)) | |
869 | seq_show_option(seq, ss->legacy_name, NULL); | |
870 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
871 | seq_puts(seq, ",noprefix"); | |
872 | if (root->flags & CGRP_ROOT_XATTR) | |
873 | seq_puts(seq, ",xattr"); | |
e1cba4b8 WL |
874 | if (root->flags & CGRP_ROOT_CPUSET_V2_MODE) |
875 | seq_puts(seq, ",cpuset_v2_mode"); | |
1592c9b2 TH |
876 | |
877 | spin_lock(&release_agent_path_lock); | |
878 | if (strlen(root->release_agent_path)) | |
879 | seq_show_option(seq, "release_agent", | |
880 | root->release_agent_path); | |
881 | spin_unlock(&release_agent_path_lock); | |
882 | ||
883 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
884 | seq_puts(seq, ",clone_children"); | |
885 | if (strlen(root->name)) | |
886 | seq_show_option(seq, "name", root->name); | |
887 | return 0; | |
888 | } | |
889 | ||
8d2451f4 AV |
890 | enum cgroup1_param { |
891 | Opt_all, | |
892 | Opt_clone_children, | |
893 | Opt_cpuset_v2_mode, | |
894 | Opt_name, | |
895 | Opt_none, | |
896 | Opt_noprefix, | |
897 | Opt_release_agent, | |
898 | Opt_xattr, | |
899 | }; | |
1592c9b2 | 900 | |
d7167b14 | 901 | const struct fs_parameter_spec cgroup1_fs_parameters[] = { |
8d2451f4 AV |
902 | fsparam_flag ("all", Opt_all), |
903 | fsparam_flag ("clone_children", Opt_clone_children), | |
904 | fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode), | |
905 | fsparam_string("name", Opt_name), | |
906 | fsparam_flag ("none", Opt_none), | |
907 | fsparam_flag ("noprefix", Opt_noprefix), | |
908 | fsparam_string("release_agent", Opt_release_agent), | |
909 | fsparam_flag ("xattr", Opt_xattr), | |
910 | {} | |
911 | }; | |
1592c9b2 | 912 | |
8d2451f4 AV |
913 | int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param) |
914 | { | |
915 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); | |
916 | struct cgroup_subsys *ss; | |
917 | struct fs_parse_result result; | |
918 | int opt, i; | |
919 | ||
d7167b14 | 920 | opt = fs_parse(fc, cgroup1_fs_parameters, param, &result); |
8d2451f4 | 921 | if (opt == -ENOPARAM) { |
d1d488d8 CB |
922 | int ret; |
923 | ||
924 | ret = vfs_parse_fs_param_source(fc, param); | |
925 | if (ret != -ENOPARAM) | |
926 | return ret; | |
1592c9b2 | 927 | for_each_subsys(ss, i) { |
8d2451f4 | 928 | if (strcmp(param->key, ss->legacy_name)) |
1592c9b2 | 929 | continue; |
61e960b0 CZ |
930 | if (!cgroup_ssid_enabled(i) || cgroup1_ssid_disabled(i)) |
931 | return invalfc(fc, "Disabled controller '%s'", | |
932 | param->key); | |
f5dfb531 | 933 | ctx->subsys_mask |= (1 << i); |
8d2451f4 | 934 | return 0; |
1592c9b2 | 935 | } |
58c025f0 | 936 | return invalfc(fc, "Unknown subsys name '%s'", param->key); |
8d2451f4 AV |
937 | } |
938 | if (opt < 0) | |
939 | return opt; | |
940 | ||
941 | switch (opt) { | |
942 | case Opt_none: | |
943 | /* Explicitly have no subsystems */ | |
944 | ctx->none = true; | |
945 | break; | |
946 | case Opt_all: | |
947 | ctx->all_ss = true; | |
948 | break; | |
949 | case Opt_noprefix: | |
950 | ctx->flags |= CGRP_ROOT_NOPREFIX; | |
951 | break; | |
952 | case Opt_clone_children: | |
953 | ctx->cpuset_clone_children = true; | |
954 | break; | |
955 | case Opt_cpuset_v2_mode: | |
956 | ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE; | |
957 | break; | |
958 | case Opt_xattr: | |
959 | ctx->flags |= CGRP_ROOT_XATTR; | |
960 | break; | |
961 | case Opt_release_agent: | |
962 | /* Specifying two release agents is forbidden */ | |
963 | if (ctx->release_agent) | |
58c025f0 | 964 | return invalfc(fc, "release_agent respecified"); |
24f60085 EB |
965 | /* |
966 | * Release agent gets called with all capabilities, | |
967 | * require capabilities to set release agent. | |
968 | */ | |
969 | if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) | |
970 | return invalfc(fc, "Setting release_agent not allowed"); | |
8d2451f4 AV |
971 | ctx->release_agent = param->string; |
972 | param->string = NULL; | |
973 | break; | |
974 | case Opt_name: | |
975 | /* blocked by boot param? */ | |
976 | if (cgroup_no_v1_named) | |
1592c9b2 | 977 | return -ENOENT; |
8d2451f4 AV |
978 | /* Can't specify an empty name */ |
979 | if (!param->size) | |
58c025f0 | 980 | return invalfc(fc, "Empty name"); |
8d2451f4 | 981 | if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1) |
58c025f0 | 982 | return invalfc(fc, "Name too long"); |
8d2451f4 AV |
983 | /* Must match [\w.-]+ */ |
984 | for (i = 0; i < param->size; i++) { | |
985 | char c = param->string[i]; | |
986 | if (isalnum(c)) | |
987 | continue; | |
988 | if ((c == '.') || (c == '-') || (c == '_')) | |
989 | continue; | |
58c025f0 | 990 | return invalfc(fc, "Invalid name"); |
8d2451f4 AV |
991 | } |
992 | /* Specifying two names is forbidden */ | |
993 | if (ctx->name) | |
58c025f0 | 994 | return invalfc(fc, "name respecified"); |
8d2451f4 AV |
995 | ctx->name = param->string; |
996 | param->string = NULL; | |
997 | break; | |
1592c9b2 | 998 | } |
f5dfb531 AV |
999 | return 0; |
1000 | } | |
1001 | ||
8d2451f4 | 1002 | static int check_cgroupfs_options(struct fs_context *fc) |
f5dfb531 | 1003 | { |
8d2451f4 | 1004 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
f5dfb531 AV |
1005 | u16 mask = U16_MAX; |
1006 | u16 enabled = 0; | |
1007 | struct cgroup_subsys *ss; | |
1008 | int i; | |
1009 | ||
1010 | #ifdef CONFIG_CPUSETS | |
1011 | mask = ~((u16)1 << cpuset_cgrp_id); | |
1012 | #endif | |
1013 | for_each_subsys(ss, i) | |
1014 | if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i)) | |
1015 | enabled |= 1 << i; | |
1016 | ||
1017 | ctx->subsys_mask &= enabled; | |
1592c9b2 TH |
1018 | |
1019 | /* | |
08b2b6fd | 1020 | * In absence of 'none', 'name=' and subsystem name options, |
f5dfb531 | 1021 | * let's default to 'all'. |
1592c9b2 | 1022 | */ |
f5dfb531 AV |
1023 | if (!ctx->subsys_mask && !ctx->none && !ctx->name) |
1024 | ctx->all_ss = true; | |
1025 | ||
1026 | if (ctx->all_ss) { | |
1027 | /* Mutually exclusive option 'all' + subsystem name */ | |
1028 | if (ctx->subsys_mask) | |
58c025f0 | 1029 | return invalfc(fc, "subsys name conflicts with all"); |
f5dfb531 AV |
1030 | /* 'all' => select all the subsystems */ |
1031 | ctx->subsys_mask = enabled; | |
1032 | } | |
1592c9b2 TH |
1033 | |
1034 | /* | |
1035 | * We either have to specify by name or by subsystems. (So all | |
1036 | * empty hierarchies must have a name). | |
1037 | */ | |
f5dfb531 | 1038 | if (!ctx->subsys_mask && !ctx->name) |
58c025f0 | 1039 | return invalfc(fc, "Need name or subsystem set"); |
1592c9b2 TH |
1040 | |
1041 | /* | |
1042 | * Option noprefix was introduced just for backward compatibility | |
1043 | * with the old cpuset, so we allow noprefix only if mounting just | |
1044 | * the cpuset subsystem. | |
1045 | */ | |
f5dfb531 | 1046 | if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask)) |
58c025f0 | 1047 | return invalfc(fc, "noprefix used incorrectly"); |
1592c9b2 TH |
1048 | |
1049 | /* Can't specify "none" and some subsystems */ | |
f5dfb531 | 1050 | if (ctx->subsys_mask && ctx->none) |
58c025f0 | 1051 | return invalfc(fc, "none used incorrectly"); |
1592c9b2 TH |
1052 | |
1053 | return 0; | |
1054 | } | |
1055 | ||
90129625 | 1056 | int cgroup1_reconfigure(struct fs_context *fc) |
1592c9b2 | 1057 | { |
90129625 AV |
1058 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1059 | struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb); | |
1592c9b2 | 1060 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
90129625 | 1061 | int ret = 0; |
1592c9b2 TH |
1062 | u16 added_mask, removed_mask; |
1063 | ||
1064 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1065 | ||
1066 | /* See what subsystems are wanted */ | |
8d2451f4 | 1067 | ret = check_cgroupfs_options(fc); |
1592c9b2 TH |
1068 | if (ret) |
1069 | goto out_unlock; | |
1070 | ||
f5dfb531 | 1071 | if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent) |
1592c9b2 TH |
1072 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
1073 | task_tgid_nr(current), current->comm); | |
1074 | ||
f5dfb531 AV |
1075 | added_mask = ctx->subsys_mask & ~root->subsys_mask; |
1076 | removed_mask = root->subsys_mask & ~ctx->subsys_mask; | |
1592c9b2 TH |
1077 | |
1078 | /* Don't allow flags or name to change at remount */ | |
f5dfb531 AV |
1079 | if ((ctx->flags ^ root->flags) || |
1080 | (ctx->name && strcmp(ctx->name, root->name))) { | |
58c025f0 | 1081 | errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"", |
f5dfb531 | 1082 | ctx->flags, ctx->name ?: "", root->flags, root->name); |
1592c9b2 TH |
1083 | ret = -EINVAL; |
1084 | goto out_unlock; | |
1085 | } | |
1086 | ||
1087 | /* remounting is not allowed for populated hierarchies */ | |
1088 | if (!list_empty(&root->cgrp.self.children)) { | |
1089 | ret = -EBUSY; | |
1090 | goto out_unlock; | |
1091 | } | |
1092 | ||
1093 | ret = rebind_subsystems(root, added_mask); | |
1094 | if (ret) | |
1095 | goto out_unlock; | |
1096 | ||
1097 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); | |
1098 | ||
f5dfb531 | 1099 | if (ctx->release_agent) { |
1592c9b2 | 1100 | spin_lock(&release_agent_path_lock); |
f5dfb531 | 1101 | strcpy(root->release_agent_path, ctx->release_agent); |
1592c9b2 TH |
1102 | spin_unlock(&release_agent_path_lock); |
1103 | } | |
1104 | ||
1105 | trace_cgroup_remount(root); | |
1106 | ||
1107 | out_unlock: | |
1592c9b2 TH |
1108 | mutex_unlock(&cgroup_mutex); |
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | struct kernfs_syscall_ops cgroup1_kf_syscall_ops = { | |
1113 | .rename = cgroup1_rename, | |
1114 | .show_options = cgroup1_show_options, | |
1592c9b2 TH |
1115 | .mkdir = cgroup_mkdir, |
1116 | .rmdir = cgroup_rmdir, | |
1117 | .show_path = cgroup_show_path, | |
1118 | }; | |
1119 | ||
6678889f AV |
1120 | /* |
1121 | * The guts of cgroup1 mount - find or create cgroup_root to use. | |
1122 | * Called with cgroup_mutex held; returns 0 on success, -E... on | |
1123 | * error and positive - in case when the candidate is busy dying. | |
1124 | * On success it stashes a reference to cgroup_root into given | |
1125 | * cgroup_fs_context; that reference is *NOT* counting towards the | |
1126 | * cgroup_root refcount. | |
1127 | */ | |
1128 | static int cgroup1_root_to_use(struct fs_context *fc) | |
1592c9b2 | 1129 | { |
7feeef58 | 1130 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1592c9b2 TH |
1131 | struct cgroup_root *root; |
1132 | struct cgroup_subsys *ss; | |
1592c9b2 TH |
1133 | int i, ret; |
1134 | ||
1592c9b2 | 1135 | /* First find the desired set of subsystems */ |
8d2451f4 | 1136 | ret = check_cgroupfs_options(fc); |
1592c9b2 | 1137 | if (ret) |
6678889f | 1138 | return ret; |
1592c9b2 TH |
1139 | |
1140 | /* | |
1141 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1142 | * still be dying after the previous unmount. Let's drain the | |
1143 | * dying subsystems. We just need to ensure that the ones | |
1144 | * unmounted previously finish dying and don't care about new ones | |
1145 | * starting. Testing ref liveliness is good enough. | |
1146 | */ | |
1147 | for_each_subsys(ss, i) { | |
f5dfb531 | 1148 | if (!(ctx->subsys_mask & (1 << i)) || |
1592c9b2 TH |
1149 | ss->root == &cgrp_dfl_root) |
1150 | continue; | |
1151 | ||
6678889f AV |
1152 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) |
1153 | return 1; /* restart */ | |
1592c9b2 TH |
1154 | cgroup_put(&ss->root->cgrp); |
1155 | } | |
1156 | ||
1157 | for_each_root(root) { | |
1158 | bool name_match = false; | |
1159 | ||
1160 | if (root == &cgrp_dfl_root) | |
1161 | continue; | |
1162 | ||
1163 | /* | |
1164 | * If we asked for a name then it must match. Also, if | |
1165 | * name matches but sybsys_mask doesn't, we should fail. | |
1166 | * Remember whether name matched. | |
1167 | */ | |
f5dfb531 AV |
1168 | if (ctx->name) { |
1169 | if (strcmp(ctx->name, root->name)) | |
1592c9b2 TH |
1170 | continue; |
1171 | name_match = true; | |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * If we asked for subsystems (or explicitly for no | |
1176 | * subsystems) then they must match. | |
1177 | */ | |
f5dfb531 AV |
1178 | if ((ctx->subsys_mask || ctx->none) && |
1179 | (ctx->subsys_mask != root->subsys_mask)) { | |
1592c9b2 TH |
1180 | if (!name_match) |
1181 | continue; | |
6678889f | 1182 | return -EBUSY; |
1592c9b2 TH |
1183 | } |
1184 | ||
f5dfb531 | 1185 | if (root->flags ^ ctx->flags) |
1592c9b2 TH |
1186 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); |
1187 | ||
cf6299b1 | 1188 | ctx->root = root; |
6678889f | 1189 | return 0; |
1592c9b2 TH |
1190 | } |
1191 | ||
1192 | /* | |
1193 | * No such thing, create a new one. name= matching without subsys | |
1194 | * specification is allowed for already existing hierarchies but we | |
1195 | * can't create new one without subsys specification. | |
1196 | */ | |
6678889f | 1197 | if (!ctx->subsys_mask && !ctx->none) |
58c025f0 | 1198 | return invalfc(fc, "No subsys list or none specified"); |
1592c9b2 TH |
1199 | |
1200 | /* Hierarchies may only be created in the initial cgroup namespace. */ | |
cca8f327 | 1201 | if (ctx->ns != &init_cgroup_ns) |
6678889f | 1202 | return -EPERM; |
1592c9b2 TH |
1203 | |
1204 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
6678889f AV |
1205 | if (!root) |
1206 | return -ENOMEM; | |
1592c9b2 | 1207 | |
cf6299b1 AV |
1208 | ctx->root = root; |
1209 | init_cgroup_root(ctx); | |
1592c9b2 | 1210 | |
f5dfb531 | 1211 | ret = cgroup_setup_root(root, ctx->subsys_mask); |
1592c9b2 TH |
1212 | if (ret) |
1213 | cgroup_free_root(root); | |
6678889f AV |
1214 | return ret; |
1215 | } | |
1216 | ||
1217 | int cgroup1_get_tree(struct fs_context *fc) | |
1218 | { | |
6678889f AV |
1219 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1220 | int ret; | |
1221 | ||
1222 | /* Check if the caller has permission to mount. */ | |
cca8f327 | 1223 | if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN)) |
6678889f AV |
1224 | return -EPERM; |
1225 | ||
1226 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1227 | ||
1228 | ret = cgroup1_root_to_use(fc); | |
1229 | if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt)) | |
1230 | ret = 1; /* restart */ | |
1592c9b2 | 1231 | |
1592c9b2 | 1232 | mutex_unlock(&cgroup_mutex); |
1592c9b2 | 1233 | |
6678889f | 1234 | if (!ret) |
cca8f327 | 1235 | ret = cgroup_do_get_tree(fc); |
6678889f AV |
1236 | |
1237 | if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) { | |
1e7107c5 | 1238 | fc_drop_locked(fc); |
6678889f AV |
1239 | ret = 1; |
1240 | } | |
1241 | ||
1242 | if (unlikely(ret > 0)) { | |
35ac1184 | 1243 | msleep(10); |
7feeef58 | 1244 | return restart_syscall(); |
9732adc5 | 1245 | } |
71d883c3 | 1246 | return ret; |
1592c9b2 TH |
1247 | } |
1248 | ||
0a268dbd TH |
1249 | static int __init cgroup1_wq_init(void) |
1250 | { | |
1251 | /* | |
1252 | * Used to destroy pidlists and separate to serve as flush domain. | |
1253 | * Cap @max_active to 1 too. | |
1254 | */ | |
1255 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
1256 | 0, 1); | |
1257 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
1258 | return 0; | |
1259 | } | |
1260 | core_initcall(cgroup1_wq_init); | |
1261 | ||
1262 | static int __init cgroup_no_v1(char *str) | |
1263 | { | |
1264 | struct cgroup_subsys *ss; | |
1265 | char *token; | |
1266 | int i; | |
1267 | ||
1268 | while ((token = strsep(&str, ",")) != NULL) { | |
1269 | if (!*token) | |
1270 | continue; | |
1271 | ||
1272 | if (!strcmp(token, "all")) { | |
1273 | cgroup_no_v1_mask = U16_MAX; | |
3fc9c12d TH |
1274 | continue; |
1275 | } | |
1276 | ||
1277 | if (!strcmp(token, "named")) { | |
1278 | cgroup_no_v1_named = true; | |
1279 | continue; | |
0a268dbd TH |
1280 | } |
1281 | ||
1282 | for_each_subsys(ss, i) { | |
1283 | if (strcmp(token, ss->name) && | |
1284 | strcmp(token, ss->legacy_name)) | |
1285 | continue; | |
1286 | ||
1287 | cgroup_no_v1_mask |= 1 << i; | |
1288 | } | |
1289 | } | |
1290 | return 1; | |
1291 | } | |
1292 | __setup("cgroup_no_v1=", cgroup_no_v1); |