]>
Commit | Line | Data |
---|---|---|
0a268dbd TH |
1 | #include "cgroup-internal.h" |
2 | ||
1592c9b2 | 3 | #include <linux/ctype.h> |
0a268dbd TH |
4 | #include <linux/kmod.h> |
5 | #include <linux/sort.h> | |
1592c9b2 | 6 | #include <linux/delay.h> |
0a268dbd TH |
7 | #include <linux/mm.h> |
8 | #include <linux/slab.h> | |
9 | #include <linux/vmalloc.h> | |
10 | #include <linux/delayacct.h> | |
11 | #include <linux/pid_namespace.h> | |
12 | #include <linux/cgroupstats.h> | |
13 | ||
14 | #include <trace/events/cgroup.h> | |
15 | ||
16 | /* | |
17 | * pidlists linger the following amount before being destroyed. The goal | |
18 | * is avoiding frequent destruction in the middle of consecutive read calls | |
19 | * Expiring in the middle is a performance problem not a correctness one. | |
20 | * 1 sec should be enough. | |
21 | */ | |
22 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
23 | ||
24 | /* Controllers blocked by the commandline in v1 */ | |
25 | static u16 cgroup_no_v1_mask; | |
26 | ||
27 | /* | |
28 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
29 | * separate workqueue as flush domain. | |
30 | */ | |
31 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
32 | ||
33 | /* | |
34 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
35 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
36 | */ | |
1592c9b2 | 37 | static DEFINE_SPINLOCK(release_agent_path_lock); |
0a268dbd | 38 | |
d62beb7f | 39 | bool cgroup1_ssid_disabled(int ssid) |
0a268dbd TH |
40 | { |
41 | return cgroup_no_v1_mask & (1 << ssid); | |
42 | } | |
43 | ||
44 | /** | |
45 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
46 | * @from: attach to all cgroups of a given task | |
47 | * @tsk: the task to be attached | |
48 | */ | |
49 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
50 | { | |
51 | struct cgroup_root *root; | |
52 | int retval = 0; | |
53 | ||
54 | mutex_lock(&cgroup_mutex); | |
55 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
56 | for_each_root(root) { | |
57 | struct cgroup *from_cgrp; | |
58 | ||
59 | if (root == &cgrp_dfl_root) | |
60 | continue; | |
61 | ||
62 | spin_lock_irq(&css_set_lock); | |
63 | from_cgrp = task_cgroup_from_root(from, root); | |
64 | spin_unlock_irq(&css_set_lock); | |
65 | ||
66 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
67 | if (retval) | |
68 | break; | |
69 | } | |
70 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
71 | mutex_unlock(&cgroup_mutex); | |
72 | ||
73 | return retval; | |
74 | } | |
75 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
76 | ||
77 | /** | |
78 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
79 | * @to: cgroup to which the tasks will be moved | |
80 | * @from: cgroup in which the tasks currently reside | |
81 | * | |
82 | * Locking rules between cgroup_post_fork() and the migration path | |
83 | * guarantee that, if a task is forking while being migrated, the new child | |
84 | * is guaranteed to be either visible in the source cgroup after the | |
85 | * parent's migration is complete or put into the target cgroup. No task | |
86 | * can slip out of migration through forking. | |
87 | */ | |
88 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
89 | { | |
e595cd70 | 90 | DEFINE_CGROUP_MGCTX(mgctx); |
0a268dbd TH |
91 | struct cgrp_cset_link *link; |
92 | struct css_task_iter it; | |
93 | struct task_struct *task; | |
94 | int ret; | |
95 | ||
96 | if (cgroup_on_dfl(to)) | |
97 | return -EINVAL; | |
98 | ||
99 | if (!cgroup_may_migrate_to(to)) | |
100 | return -EBUSY; | |
101 | ||
102 | mutex_lock(&cgroup_mutex); | |
103 | ||
104 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
105 | ||
106 | /* all tasks in @from are being moved, all csets are source */ | |
107 | spin_lock_irq(&css_set_lock); | |
108 | list_for_each_entry(link, &from->cset_links, cset_link) | |
e595cd70 | 109 | cgroup_migrate_add_src(link->cset, to, &mgctx); |
0a268dbd TH |
110 | spin_unlock_irq(&css_set_lock); |
111 | ||
e595cd70 | 112 | ret = cgroup_migrate_prepare_dst(&mgctx); |
0a268dbd TH |
113 | if (ret) |
114 | goto out_err; | |
115 | ||
116 | /* | |
117 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
118 | * ->can_attach() fails. | |
119 | */ | |
120 | do { | |
121 | css_task_iter_start(&from->self, &it); | |
122 | task = css_task_iter_next(&it); | |
123 | if (task) | |
124 | get_task_struct(task); | |
125 | css_task_iter_end(&it); | |
126 | ||
127 | if (task) { | |
bfc2cf6f | 128 | ret = cgroup_migrate(task, false, &mgctx); |
0a268dbd TH |
129 | if (!ret) |
130 | trace_cgroup_transfer_tasks(to, task, false); | |
131 | put_task_struct(task); | |
132 | } | |
133 | } while (task && !ret); | |
134 | out_err: | |
e595cd70 | 135 | cgroup_migrate_finish(&mgctx); |
0a268dbd TH |
136 | percpu_up_write(&cgroup_threadgroup_rwsem); |
137 | mutex_unlock(&cgroup_mutex); | |
138 | return ret; | |
139 | } | |
140 | ||
141 | /* | |
142 | * Stuff for reading the 'tasks'/'procs' files. | |
143 | * | |
144 | * Reading this file can return large amounts of data if a cgroup has | |
145 | * *lots* of attached tasks. So it may need several calls to read(), | |
146 | * but we cannot guarantee that the information we produce is correct | |
147 | * unless we produce it entirely atomically. | |
148 | * | |
149 | */ | |
150 | ||
151 | /* which pidlist file are we talking about? */ | |
152 | enum cgroup_filetype { | |
153 | CGROUP_FILE_PROCS, | |
154 | CGROUP_FILE_TASKS, | |
155 | }; | |
156 | ||
157 | /* | |
158 | * A pidlist is a list of pids that virtually represents the contents of one | |
159 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
160 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
161 | * to the cgroup. | |
162 | */ | |
163 | struct cgroup_pidlist { | |
164 | /* | |
165 | * used to find which pidlist is wanted. doesn't change as long as | |
166 | * this particular list stays in the list. | |
167 | */ | |
168 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
169 | /* array of xids */ | |
170 | pid_t *list; | |
171 | /* how many elements the above list has */ | |
172 | int length; | |
173 | /* each of these stored in a list by its cgroup */ | |
174 | struct list_head links; | |
175 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
176 | struct cgroup *owner; | |
177 | /* for delayed destruction */ | |
178 | struct delayed_work destroy_dwork; | |
179 | }; | |
180 | ||
181 | /* | |
182 | * The following two functions "fix" the issue where there are more pids | |
183 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
184 | * TODO: replace with a kernel-wide solution to this problem | |
185 | */ | |
186 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
187 | static void *pidlist_allocate(int count) | |
188 | { | |
189 | if (PIDLIST_TOO_LARGE(count)) | |
190 | return vmalloc(count * sizeof(pid_t)); | |
191 | else | |
192 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
193 | } | |
194 | ||
195 | static void pidlist_free(void *p) | |
196 | { | |
197 | kvfree(p); | |
198 | } | |
199 | ||
200 | /* | |
201 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
202 | * should be left afterwards. | |
203 | */ | |
d62beb7f | 204 | void cgroup1_pidlist_destroy_all(struct cgroup *cgrp) |
0a268dbd TH |
205 | { |
206 | struct cgroup_pidlist *l, *tmp_l; | |
207 | ||
208 | mutex_lock(&cgrp->pidlist_mutex); | |
209 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
210 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
211 | mutex_unlock(&cgrp->pidlist_mutex); | |
212 | ||
213 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
214 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
215 | } | |
216 | ||
217 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
218 | { | |
219 | struct delayed_work *dwork = to_delayed_work(work); | |
220 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
221 | destroy_dwork); | |
222 | struct cgroup_pidlist *tofree = NULL; | |
223 | ||
224 | mutex_lock(&l->owner->pidlist_mutex); | |
225 | ||
226 | /* | |
227 | * Destroy iff we didn't get queued again. The state won't change | |
228 | * as destroy_dwork can only be queued while locked. | |
229 | */ | |
230 | if (!delayed_work_pending(dwork)) { | |
231 | list_del(&l->links); | |
232 | pidlist_free(l->list); | |
233 | put_pid_ns(l->key.ns); | |
234 | tofree = l; | |
235 | } | |
236 | ||
237 | mutex_unlock(&l->owner->pidlist_mutex); | |
238 | kfree(tofree); | |
239 | } | |
240 | ||
241 | /* | |
242 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
243 | * Returns the number of unique elements. | |
244 | */ | |
245 | static int pidlist_uniq(pid_t *list, int length) | |
246 | { | |
247 | int src, dest = 1; | |
248 | ||
249 | /* | |
250 | * we presume the 0th element is unique, so i starts at 1. trivial | |
251 | * edge cases first; no work needs to be done for either | |
252 | */ | |
253 | if (length == 0 || length == 1) | |
254 | return length; | |
255 | /* src and dest walk down the list; dest counts unique elements */ | |
256 | for (src = 1; src < length; src++) { | |
257 | /* find next unique element */ | |
258 | while (list[src] == list[src-1]) { | |
259 | src++; | |
260 | if (src == length) | |
261 | goto after; | |
262 | } | |
263 | /* dest always points to where the next unique element goes */ | |
264 | list[dest] = list[src]; | |
265 | dest++; | |
266 | } | |
267 | after: | |
268 | return dest; | |
269 | } | |
270 | ||
271 | /* | |
272 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
273 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
274 | * different per namespace, each namespace needs differently sorted list, | |
275 | * making it impossible to use, for example, single rbtree of member tasks | |
276 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
277 | * per open file is dangerous, so cgroup had to implement shared pool of | |
278 | * pidlists keyed by cgroup and namespace. | |
279 | */ | |
280 | static int cmppid(const void *a, const void *b) | |
281 | { | |
282 | return *(pid_t *)a - *(pid_t *)b; | |
283 | } | |
284 | ||
285 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
286 | enum cgroup_filetype type) | |
287 | { | |
288 | struct cgroup_pidlist *l; | |
289 | /* don't need task_nsproxy() if we're looking at ourself */ | |
290 | struct pid_namespace *ns = task_active_pid_ns(current); | |
291 | ||
292 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
293 | ||
294 | list_for_each_entry(l, &cgrp->pidlists, links) | |
295 | if (l->key.type == type && l->key.ns == ns) | |
296 | return l; | |
297 | return NULL; | |
298 | } | |
299 | ||
300 | /* | |
301 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
302 | * returns with the lock on that pidlist already held, and takes care | |
303 | * of the use count, or returns NULL with no locks held if we're out of | |
304 | * memory. | |
305 | */ | |
306 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
307 | enum cgroup_filetype type) | |
308 | { | |
309 | struct cgroup_pidlist *l; | |
310 | ||
311 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
312 | ||
313 | l = cgroup_pidlist_find(cgrp, type); | |
314 | if (l) | |
315 | return l; | |
316 | ||
317 | /* entry not found; create a new one */ | |
318 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
319 | if (!l) | |
320 | return l; | |
321 | ||
322 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
323 | l->key.type = type; | |
324 | /* don't need task_nsproxy() if we're looking at ourself */ | |
325 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
326 | l->owner = cgrp; | |
327 | list_add(&l->links, &cgrp->pidlists); | |
328 | return l; | |
329 | } | |
330 | ||
331 | /** | |
332 | * cgroup_task_count - count the number of tasks in a cgroup. | |
333 | * @cgrp: the cgroup in question | |
334 | * | |
335 | * Return the number of tasks in the cgroup. The returned number can be | |
336 | * higher than the actual number of tasks due to css_set references from | |
337 | * namespace roots and temporary usages. | |
338 | */ | |
339 | static int cgroup_task_count(const struct cgroup *cgrp) | |
340 | { | |
341 | int count = 0; | |
342 | struct cgrp_cset_link *link; | |
343 | ||
344 | spin_lock_irq(&css_set_lock); | |
345 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
346 | count += atomic_read(&link->cset->refcount); | |
347 | spin_unlock_irq(&css_set_lock); | |
348 | return count; | |
349 | } | |
350 | ||
351 | /* | |
352 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
353 | */ | |
354 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
355 | struct cgroup_pidlist **lp) | |
356 | { | |
357 | pid_t *array; | |
358 | int length; | |
359 | int pid, n = 0; /* used for populating the array */ | |
360 | struct css_task_iter it; | |
361 | struct task_struct *tsk; | |
362 | struct cgroup_pidlist *l; | |
363 | ||
364 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
365 | ||
366 | /* | |
367 | * If cgroup gets more users after we read count, we won't have | |
368 | * enough space - tough. This race is indistinguishable to the | |
369 | * caller from the case that the additional cgroup users didn't | |
370 | * show up until sometime later on. | |
371 | */ | |
372 | length = cgroup_task_count(cgrp); | |
373 | array = pidlist_allocate(length); | |
374 | if (!array) | |
375 | return -ENOMEM; | |
376 | /* now, populate the array */ | |
377 | css_task_iter_start(&cgrp->self, &it); | |
378 | while ((tsk = css_task_iter_next(&it))) { | |
379 | if (unlikely(n == length)) | |
380 | break; | |
381 | /* get tgid or pid for procs or tasks file respectively */ | |
382 | if (type == CGROUP_FILE_PROCS) | |
383 | pid = task_tgid_vnr(tsk); | |
384 | else | |
385 | pid = task_pid_vnr(tsk); | |
386 | if (pid > 0) /* make sure to only use valid results */ | |
387 | array[n++] = pid; | |
388 | } | |
389 | css_task_iter_end(&it); | |
390 | length = n; | |
391 | /* now sort & (if procs) strip out duplicates */ | |
392 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
393 | if (type == CGROUP_FILE_PROCS) | |
394 | length = pidlist_uniq(array, length); | |
395 | ||
396 | l = cgroup_pidlist_find_create(cgrp, type); | |
397 | if (!l) { | |
398 | pidlist_free(array); | |
399 | return -ENOMEM; | |
400 | } | |
401 | ||
402 | /* store array, freeing old if necessary */ | |
403 | pidlist_free(l->list); | |
404 | l->list = array; | |
405 | l->length = length; | |
406 | *lp = l; | |
407 | return 0; | |
408 | } | |
409 | ||
410 | /* | |
411 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
412 | * next pid to display; the seq_file iterator is a pointer to the pid | |
413 | * in the cgroup->l->list array. | |
414 | */ | |
415 | ||
416 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
417 | { | |
418 | /* | |
419 | * Initially we receive a position value that corresponds to | |
420 | * one more than the last pid shown (or 0 on the first call or | |
421 | * after a seek to the start). Use a binary-search to find the | |
422 | * next pid to display, if any | |
423 | */ | |
424 | struct kernfs_open_file *of = s->private; | |
425 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
426 | struct cgroup_pidlist *l; | |
427 | enum cgroup_filetype type = seq_cft(s)->private; | |
428 | int index = 0, pid = *pos; | |
429 | int *iter, ret; | |
430 | ||
431 | mutex_lock(&cgrp->pidlist_mutex); | |
432 | ||
433 | /* | |
434 | * !NULL @of->priv indicates that this isn't the first start() | |
435 | * after open. If the matching pidlist is around, we can use that. | |
436 | * Look for it. Note that @of->priv can't be used directly. It | |
437 | * could already have been destroyed. | |
438 | */ | |
439 | if (of->priv) | |
440 | of->priv = cgroup_pidlist_find(cgrp, type); | |
441 | ||
442 | /* | |
443 | * Either this is the first start() after open or the matching | |
444 | * pidlist has been destroyed inbetween. Create a new one. | |
445 | */ | |
446 | if (!of->priv) { | |
447 | ret = pidlist_array_load(cgrp, type, | |
448 | (struct cgroup_pidlist **)&of->priv); | |
449 | if (ret) | |
450 | return ERR_PTR(ret); | |
451 | } | |
452 | l = of->priv; | |
453 | ||
454 | if (pid) { | |
455 | int end = l->length; | |
456 | ||
457 | while (index < end) { | |
458 | int mid = (index + end) / 2; | |
459 | if (l->list[mid] == pid) { | |
460 | index = mid; | |
461 | break; | |
462 | } else if (l->list[mid] <= pid) | |
463 | index = mid + 1; | |
464 | else | |
465 | end = mid; | |
466 | } | |
467 | } | |
468 | /* If we're off the end of the array, we're done */ | |
469 | if (index >= l->length) | |
470 | return NULL; | |
471 | /* Update the abstract position to be the actual pid that we found */ | |
472 | iter = l->list + index; | |
473 | *pos = *iter; | |
474 | return iter; | |
475 | } | |
476 | ||
477 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
478 | { | |
479 | struct kernfs_open_file *of = s->private; | |
480 | struct cgroup_pidlist *l = of->priv; | |
481 | ||
482 | if (l) | |
483 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
484 | CGROUP_PIDLIST_DESTROY_DELAY); | |
485 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); | |
486 | } | |
487 | ||
488 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
489 | { | |
490 | struct kernfs_open_file *of = s->private; | |
491 | struct cgroup_pidlist *l = of->priv; | |
492 | pid_t *p = v; | |
493 | pid_t *end = l->list + l->length; | |
494 | /* | |
495 | * Advance to the next pid in the array. If this goes off the | |
496 | * end, we're done | |
497 | */ | |
498 | p++; | |
499 | if (p >= end) { | |
500 | return NULL; | |
501 | } else { | |
502 | *pos = *p; | |
503 | return p; | |
504 | } | |
505 | } | |
506 | ||
507 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
508 | { | |
509 | seq_printf(s, "%d\n", *(int *)v); | |
510 | ||
511 | return 0; | |
512 | } | |
513 | ||
514 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, | |
515 | char *buf, size_t nbytes, loff_t off) | |
516 | { | |
517 | return __cgroup_procs_write(of, buf, nbytes, off, false); | |
518 | } | |
519 | ||
520 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
521 | char *buf, size_t nbytes, loff_t off) | |
522 | { | |
523 | struct cgroup *cgrp; | |
524 | ||
525 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
526 | ||
527 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
528 | if (!cgrp) | |
529 | return -ENODEV; | |
530 | spin_lock(&release_agent_path_lock); | |
531 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
532 | sizeof(cgrp->root->release_agent_path)); | |
533 | spin_unlock(&release_agent_path_lock); | |
534 | cgroup_kn_unlock(of->kn); | |
535 | return nbytes; | |
536 | } | |
537 | ||
538 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
539 | { | |
540 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
541 | ||
542 | spin_lock(&release_agent_path_lock); | |
543 | seq_puts(seq, cgrp->root->release_agent_path); | |
544 | spin_unlock(&release_agent_path_lock); | |
545 | seq_putc(seq, '\n'); | |
546 | return 0; | |
547 | } | |
548 | ||
549 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
550 | { | |
551 | seq_puts(seq, "0\n"); | |
552 | return 0; | |
553 | } | |
554 | ||
555 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
556 | struct cftype *cft) | |
557 | { | |
558 | return notify_on_release(css->cgroup); | |
559 | } | |
560 | ||
561 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
562 | struct cftype *cft, u64 val) | |
563 | { | |
564 | if (val) | |
565 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
566 | else | |
567 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
568 | return 0; | |
569 | } | |
570 | ||
571 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
572 | struct cftype *cft) | |
573 | { | |
574 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
575 | } | |
576 | ||
577 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
578 | struct cftype *cft, u64 val) | |
579 | { | |
580 | if (val) | |
581 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
582 | else | |
583 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
584 | return 0; | |
585 | } | |
586 | ||
587 | /* cgroup core interface files for the legacy hierarchies */ | |
d62beb7f | 588 | struct cftype cgroup1_base_files[] = { |
0a268dbd TH |
589 | { |
590 | .name = "cgroup.procs", | |
591 | .seq_start = cgroup_pidlist_start, | |
592 | .seq_next = cgroup_pidlist_next, | |
593 | .seq_stop = cgroup_pidlist_stop, | |
594 | .seq_show = cgroup_pidlist_show, | |
595 | .private = CGROUP_FILE_PROCS, | |
596 | .write = cgroup_procs_write, | |
597 | }, | |
598 | { | |
599 | .name = "cgroup.clone_children", | |
600 | .read_u64 = cgroup_clone_children_read, | |
601 | .write_u64 = cgroup_clone_children_write, | |
602 | }, | |
603 | { | |
604 | .name = "cgroup.sane_behavior", | |
605 | .flags = CFTYPE_ONLY_ON_ROOT, | |
606 | .seq_show = cgroup_sane_behavior_show, | |
607 | }, | |
608 | { | |
609 | .name = "tasks", | |
610 | .seq_start = cgroup_pidlist_start, | |
611 | .seq_next = cgroup_pidlist_next, | |
612 | .seq_stop = cgroup_pidlist_stop, | |
613 | .seq_show = cgroup_pidlist_show, | |
614 | .private = CGROUP_FILE_TASKS, | |
615 | .write = cgroup_tasks_write, | |
616 | }, | |
617 | { | |
618 | .name = "notify_on_release", | |
619 | .read_u64 = cgroup_read_notify_on_release, | |
620 | .write_u64 = cgroup_write_notify_on_release, | |
621 | }, | |
622 | { | |
623 | .name = "release_agent", | |
624 | .flags = CFTYPE_ONLY_ON_ROOT, | |
625 | .seq_show = cgroup_release_agent_show, | |
626 | .write = cgroup_release_agent_write, | |
627 | .max_write_len = PATH_MAX - 1, | |
628 | }, | |
629 | { } /* terminate */ | |
630 | }; | |
631 | ||
632 | /* Display information about each subsystem and each hierarchy */ | |
633 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
634 | { | |
635 | struct cgroup_subsys *ss; | |
636 | int i; | |
637 | ||
638 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
639 | /* | |
640 | * ideally we don't want subsystems moving around while we do this. | |
641 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
642 | * subsys/hierarchy state. | |
643 | */ | |
644 | mutex_lock(&cgroup_mutex); | |
645 | ||
646 | for_each_subsys(ss, i) | |
647 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
648 | ss->legacy_name, ss->root->hierarchy_id, | |
649 | atomic_read(&ss->root->nr_cgrps), | |
650 | cgroup_ssid_enabled(i)); | |
651 | ||
652 | mutex_unlock(&cgroup_mutex); | |
653 | return 0; | |
654 | } | |
655 | ||
656 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
657 | { | |
658 | return single_open(file, proc_cgroupstats_show, NULL); | |
659 | } | |
660 | ||
661 | const struct file_operations proc_cgroupstats_operations = { | |
662 | .open = cgroupstats_open, | |
663 | .read = seq_read, | |
664 | .llseek = seq_lseek, | |
665 | .release = single_release, | |
666 | }; | |
667 | ||
668 | /** | |
669 | * cgroupstats_build - build and fill cgroupstats | |
670 | * @stats: cgroupstats to fill information into | |
671 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
672 | * been requested. | |
673 | * | |
674 | * Build and fill cgroupstats so that taskstats can export it to user | |
675 | * space. | |
676 | */ | |
677 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
678 | { | |
679 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
680 | struct cgroup *cgrp; | |
681 | struct css_task_iter it; | |
682 | struct task_struct *tsk; | |
683 | ||
684 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
685 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
686 | kernfs_type(kn) != KERNFS_DIR) | |
687 | return -EINVAL; | |
688 | ||
689 | mutex_lock(&cgroup_mutex); | |
690 | ||
691 | /* | |
692 | * We aren't being called from kernfs and there's no guarantee on | |
693 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
694 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
695 | */ | |
696 | rcu_read_lock(); | |
e0aed7c7 | 697 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
0a268dbd TH |
698 | if (!cgrp || cgroup_is_dead(cgrp)) { |
699 | rcu_read_unlock(); | |
700 | mutex_unlock(&cgroup_mutex); | |
701 | return -ENOENT; | |
702 | } | |
703 | rcu_read_unlock(); | |
704 | ||
705 | css_task_iter_start(&cgrp->self, &it); | |
706 | while ((tsk = css_task_iter_next(&it))) { | |
707 | switch (tsk->state) { | |
708 | case TASK_RUNNING: | |
709 | stats->nr_running++; | |
710 | break; | |
711 | case TASK_INTERRUPTIBLE: | |
712 | stats->nr_sleeping++; | |
713 | break; | |
714 | case TASK_UNINTERRUPTIBLE: | |
715 | stats->nr_uninterruptible++; | |
716 | break; | |
717 | case TASK_STOPPED: | |
718 | stats->nr_stopped++; | |
719 | break; | |
720 | default: | |
721 | if (delayacct_is_task_waiting_on_io(tsk)) | |
722 | stats->nr_io_wait++; | |
723 | break; | |
724 | } | |
725 | } | |
726 | css_task_iter_end(&it); | |
727 | ||
728 | mutex_unlock(&cgroup_mutex); | |
729 | return 0; | |
730 | } | |
731 | ||
d62beb7f | 732 | void cgroup1_check_for_release(struct cgroup *cgrp) |
0a268dbd TH |
733 | { |
734 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
735 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
736 | schedule_work(&cgrp->release_agent_work); | |
737 | } | |
738 | ||
739 | /* | |
740 | * Notify userspace when a cgroup is released, by running the | |
741 | * configured release agent with the name of the cgroup (path | |
742 | * relative to the root of cgroup file system) as the argument. | |
743 | * | |
744 | * Most likely, this user command will try to rmdir this cgroup. | |
745 | * | |
746 | * This races with the possibility that some other task will be | |
747 | * attached to this cgroup before it is removed, or that some other | |
748 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
749 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
750 | * unused, and this cgroup will be reprieved from its death sentence, | |
751 | * to continue to serve a useful existence. Next time it's released, | |
752 | * we will get notified again, if it still has 'notify_on_release' set. | |
753 | * | |
754 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
755 | * means only wait until the task is successfully execve()'d. The | |
756 | * separate release agent task is forked by call_usermodehelper(), | |
757 | * then control in this thread returns here, without waiting for the | |
758 | * release agent task. We don't bother to wait because the caller of | |
759 | * this routine has no use for the exit status of the release agent | |
760 | * task, so no sense holding our caller up for that. | |
761 | */ | |
d62beb7f | 762 | void cgroup1_release_agent(struct work_struct *work) |
0a268dbd TH |
763 | { |
764 | struct cgroup *cgrp = | |
765 | container_of(work, struct cgroup, release_agent_work); | |
766 | char *pathbuf = NULL, *agentbuf = NULL; | |
767 | char *argv[3], *envp[3]; | |
768 | int ret; | |
769 | ||
770 | mutex_lock(&cgroup_mutex); | |
771 | ||
772 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
773 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
774 | if (!pathbuf || !agentbuf) | |
775 | goto out; | |
776 | ||
777 | spin_lock_irq(&css_set_lock); | |
778 | ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
779 | spin_unlock_irq(&css_set_lock); | |
780 | if (ret < 0 || ret >= PATH_MAX) | |
781 | goto out; | |
782 | ||
783 | argv[0] = agentbuf; | |
784 | argv[1] = pathbuf; | |
785 | argv[2] = NULL; | |
786 | ||
787 | /* minimal command environment */ | |
788 | envp[0] = "HOME=/"; | |
789 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
790 | envp[2] = NULL; | |
791 | ||
792 | mutex_unlock(&cgroup_mutex); | |
793 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
794 | goto out_free; | |
795 | out: | |
796 | mutex_unlock(&cgroup_mutex); | |
797 | out_free: | |
798 | kfree(agentbuf); | |
799 | kfree(pathbuf); | |
800 | } | |
801 | ||
802 | /* | |
803 | * cgroup_rename - Only allow simple rename of directories in place. | |
804 | */ | |
1592c9b2 TH |
805 | static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
806 | const char *new_name_str) | |
0a268dbd TH |
807 | { |
808 | struct cgroup *cgrp = kn->priv; | |
809 | int ret; | |
810 | ||
811 | if (kernfs_type(kn) != KERNFS_DIR) | |
812 | return -ENOTDIR; | |
813 | if (kn->parent != new_parent) | |
814 | return -EIO; | |
815 | ||
0a268dbd TH |
816 | /* |
817 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
818 | * active_ref. kernfs_rename() doesn't require active_ref | |
819 | * protection. Break them before grabbing cgroup_mutex. | |
820 | */ | |
821 | kernfs_break_active_protection(new_parent); | |
822 | kernfs_break_active_protection(kn); | |
823 | ||
824 | mutex_lock(&cgroup_mutex); | |
825 | ||
826 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
827 | if (!ret) | |
828 | trace_cgroup_rename(cgrp); | |
829 | ||
830 | mutex_unlock(&cgroup_mutex); | |
831 | ||
832 | kernfs_unbreak_active_protection(kn); | |
833 | kernfs_unbreak_active_protection(new_parent); | |
834 | return ret; | |
835 | } | |
836 | ||
1592c9b2 TH |
837 | static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root) |
838 | { | |
839 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
840 | struct cgroup_subsys *ss; | |
841 | int ssid; | |
842 | ||
843 | for_each_subsys(ss, ssid) | |
844 | if (root->subsys_mask & (1 << ssid)) | |
845 | seq_show_option(seq, ss->legacy_name, NULL); | |
846 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
847 | seq_puts(seq, ",noprefix"); | |
848 | if (root->flags & CGRP_ROOT_XATTR) | |
849 | seq_puts(seq, ",xattr"); | |
850 | ||
851 | spin_lock(&release_agent_path_lock); | |
852 | if (strlen(root->release_agent_path)) | |
853 | seq_show_option(seq, "release_agent", | |
854 | root->release_agent_path); | |
855 | spin_unlock(&release_agent_path_lock); | |
856 | ||
857 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
858 | seq_puts(seq, ",clone_children"); | |
859 | if (strlen(root->name)) | |
860 | seq_show_option(seq, "name", root->name); | |
861 | return 0; | |
862 | } | |
863 | ||
864 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) | |
865 | { | |
866 | char *token, *o = data; | |
867 | bool all_ss = false, one_ss = false; | |
868 | u16 mask = U16_MAX; | |
869 | struct cgroup_subsys *ss; | |
870 | int nr_opts = 0; | |
871 | int i; | |
872 | ||
873 | #ifdef CONFIG_CPUSETS | |
874 | mask = ~((u16)1 << cpuset_cgrp_id); | |
875 | #endif | |
876 | ||
877 | memset(opts, 0, sizeof(*opts)); | |
878 | ||
879 | while ((token = strsep(&o, ",")) != NULL) { | |
880 | nr_opts++; | |
881 | ||
882 | if (!*token) | |
883 | return -EINVAL; | |
884 | if (!strcmp(token, "none")) { | |
885 | /* Explicitly have no subsystems */ | |
886 | opts->none = true; | |
887 | continue; | |
888 | } | |
889 | if (!strcmp(token, "all")) { | |
890 | /* Mutually exclusive option 'all' + subsystem name */ | |
891 | if (one_ss) | |
892 | return -EINVAL; | |
893 | all_ss = true; | |
894 | continue; | |
895 | } | |
896 | if (!strcmp(token, "noprefix")) { | |
897 | opts->flags |= CGRP_ROOT_NOPREFIX; | |
898 | continue; | |
899 | } | |
900 | if (!strcmp(token, "clone_children")) { | |
901 | opts->cpuset_clone_children = true; | |
902 | continue; | |
903 | } | |
904 | if (!strcmp(token, "xattr")) { | |
905 | opts->flags |= CGRP_ROOT_XATTR; | |
906 | continue; | |
907 | } | |
908 | if (!strncmp(token, "release_agent=", 14)) { | |
909 | /* Specifying two release agents is forbidden */ | |
910 | if (opts->release_agent) | |
911 | return -EINVAL; | |
912 | opts->release_agent = | |
913 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); | |
914 | if (!opts->release_agent) | |
915 | return -ENOMEM; | |
916 | continue; | |
917 | } | |
918 | if (!strncmp(token, "name=", 5)) { | |
919 | const char *name = token + 5; | |
920 | /* Can't specify an empty name */ | |
921 | if (!strlen(name)) | |
922 | return -EINVAL; | |
923 | /* Must match [\w.-]+ */ | |
924 | for (i = 0; i < strlen(name); i++) { | |
925 | char c = name[i]; | |
926 | if (isalnum(c)) | |
927 | continue; | |
928 | if ((c == '.') || (c == '-') || (c == '_')) | |
929 | continue; | |
930 | return -EINVAL; | |
931 | } | |
932 | /* Specifying two names is forbidden */ | |
933 | if (opts->name) | |
934 | return -EINVAL; | |
935 | opts->name = kstrndup(name, | |
936 | MAX_CGROUP_ROOT_NAMELEN - 1, | |
937 | GFP_KERNEL); | |
938 | if (!opts->name) | |
939 | return -ENOMEM; | |
940 | ||
941 | continue; | |
942 | } | |
943 | ||
944 | for_each_subsys(ss, i) { | |
945 | if (strcmp(token, ss->legacy_name)) | |
946 | continue; | |
947 | if (!cgroup_ssid_enabled(i)) | |
948 | continue; | |
d62beb7f | 949 | if (cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
950 | continue; |
951 | ||
952 | /* Mutually exclusive option 'all' + subsystem name */ | |
953 | if (all_ss) | |
954 | return -EINVAL; | |
955 | opts->subsys_mask |= (1 << i); | |
956 | one_ss = true; | |
957 | ||
958 | break; | |
959 | } | |
960 | if (i == CGROUP_SUBSYS_COUNT) | |
961 | return -ENOENT; | |
962 | } | |
963 | ||
964 | /* | |
965 | * If the 'all' option was specified select all the subsystems, | |
966 | * otherwise if 'none', 'name=' and a subsystem name options were | |
967 | * not specified, let's default to 'all' | |
968 | */ | |
969 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
970 | for_each_subsys(ss, i) | |
d62beb7f | 971 | if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
972 | opts->subsys_mask |= (1 << i); |
973 | ||
974 | /* | |
975 | * We either have to specify by name or by subsystems. (So all | |
976 | * empty hierarchies must have a name). | |
977 | */ | |
978 | if (!opts->subsys_mask && !opts->name) | |
979 | return -EINVAL; | |
980 | ||
981 | /* | |
982 | * Option noprefix was introduced just for backward compatibility | |
983 | * with the old cpuset, so we allow noprefix only if mounting just | |
984 | * the cpuset subsystem. | |
985 | */ | |
986 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) | |
987 | return -EINVAL; | |
988 | ||
989 | /* Can't specify "none" and some subsystems */ | |
990 | if (opts->subsys_mask && opts->none) | |
991 | return -EINVAL; | |
992 | ||
993 | return 0; | |
994 | } | |
995 | ||
996 | static int cgroup1_remount(struct kernfs_root *kf_root, int *flags, char *data) | |
997 | { | |
998 | int ret = 0; | |
999 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
1000 | struct cgroup_sb_opts opts; | |
1001 | u16 added_mask, removed_mask; | |
1002 | ||
1003 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1004 | ||
1005 | /* See what subsystems are wanted */ | |
1006 | ret = parse_cgroupfs_options(data, &opts); | |
1007 | if (ret) | |
1008 | goto out_unlock; | |
1009 | ||
1010 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) | |
1011 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", | |
1012 | task_tgid_nr(current), current->comm); | |
1013 | ||
1014 | added_mask = opts.subsys_mask & ~root->subsys_mask; | |
1015 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
1016 | ||
1017 | /* Don't allow flags or name to change at remount */ | |
1018 | if ((opts.flags ^ root->flags) || | |
1019 | (opts.name && strcmp(opts.name, root->name))) { | |
1020 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", | |
1021 | opts.flags, opts.name ?: "", root->flags, root->name); | |
1022 | ret = -EINVAL; | |
1023 | goto out_unlock; | |
1024 | } | |
1025 | ||
1026 | /* remounting is not allowed for populated hierarchies */ | |
1027 | if (!list_empty(&root->cgrp.self.children)) { | |
1028 | ret = -EBUSY; | |
1029 | goto out_unlock; | |
1030 | } | |
1031 | ||
1032 | ret = rebind_subsystems(root, added_mask); | |
1033 | if (ret) | |
1034 | goto out_unlock; | |
1035 | ||
1036 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); | |
1037 | ||
1038 | if (opts.release_agent) { | |
1039 | spin_lock(&release_agent_path_lock); | |
1040 | strcpy(root->release_agent_path, opts.release_agent); | |
1041 | spin_unlock(&release_agent_path_lock); | |
1042 | } | |
1043 | ||
1044 | trace_cgroup_remount(root); | |
1045 | ||
1046 | out_unlock: | |
1047 | kfree(opts.release_agent); | |
1048 | kfree(opts.name); | |
1049 | mutex_unlock(&cgroup_mutex); | |
1050 | return ret; | |
1051 | } | |
1052 | ||
1053 | struct kernfs_syscall_ops cgroup1_kf_syscall_ops = { | |
1054 | .rename = cgroup1_rename, | |
1055 | .show_options = cgroup1_show_options, | |
1056 | .remount_fs = cgroup1_remount, | |
1057 | .mkdir = cgroup_mkdir, | |
1058 | .rmdir = cgroup_rmdir, | |
1059 | .show_path = cgroup_show_path, | |
1060 | }; | |
1061 | ||
1062 | struct dentry *cgroup1_mount(struct file_system_type *fs_type, int flags, | |
1063 | void *data, unsigned long magic, | |
1064 | struct cgroup_namespace *ns) | |
1065 | { | |
1066 | struct super_block *pinned_sb = NULL; | |
1067 | struct cgroup_sb_opts opts; | |
1068 | struct cgroup_root *root; | |
1069 | struct cgroup_subsys *ss; | |
1070 | struct dentry *dentry; | |
1071 | int i, ret; | |
1072 | ||
1073 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1074 | ||
1075 | /* First find the desired set of subsystems */ | |
1076 | ret = parse_cgroupfs_options(data, &opts); | |
1077 | if (ret) | |
1078 | goto out_unlock; | |
1079 | ||
1080 | /* | |
1081 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1082 | * still be dying after the previous unmount. Let's drain the | |
1083 | * dying subsystems. We just need to ensure that the ones | |
1084 | * unmounted previously finish dying and don't care about new ones | |
1085 | * starting. Testing ref liveliness is good enough. | |
1086 | */ | |
1087 | for_each_subsys(ss, i) { | |
1088 | if (!(opts.subsys_mask & (1 << i)) || | |
1089 | ss->root == &cgrp_dfl_root) | |
1090 | continue; | |
1091 | ||
1092 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1093 | mutex_unlock(&cgroup_mutex); | |
1094 | msleep(10); | |
1095 | ret = restart_syscall(); | |
1096 | goto out_free; | |
1097 | } | |
1098 | cgroup_put(&ss->root->cgrp); | |
1099 | } | |
1100 | ||
1101 | for_each_root(root) { | |
1102 | bool name_match = false; | |
1103 | ||
1104 | if (root == &cgrp_dfl_root) | |
1105 | continue; | |
1106 | ||
1107 | /* | |
1108 | * If we asked for a name then it must match. Also, if | |
1109 | * name matches but sybsys_mask doesn't, we should fail. | |
1110 | * Remember whether name matched. | |
1111 | */ | |
1112 | if (opts.name) { | |
1113 | if (strcmp(opts.name, root->name)) | |
1114 | continue; | |
1115 | name_match = true; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * If we asked for subsystems (or explicitly for no | |
1120 | * subsystems) then they must match. | |
1121 | */ | |
1122 | if ((opts.subsys_mask || opts.none) && | |
1123 | (opts.subsys_mask != root->subsys_mask)) { | |
1124 | if (!name_match) | |
1125 | continue; | |
1126 | ret = -EBUSY; | |
1127 | goto out_unlock; | |
1128 | } | |
1129 | ||
1130 | if (root->flags ^ opts.flags) | |
1131 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
1132 | ||
1133 | /* | |
1134 | * We want to reuse @root whose lifetime is governed by its | |
1135 | * ->cgrp. Let's check whether @root is alive and keep it | |
1136 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1137 | * want to block it by pinning the sb so that @root doesn't | |
1138 | * get killed before mount is complete. | |
1139 | * | |
1140 | * With the sb pinned, tryget_live can reliably indicate | |
1141 | * whether @root can be reused. If it's being killed, | |
1142 | * drain it. We can use wait_queue for the wait but this | |
1143 | * path is super cold. Let's just sleep a bit and retry. | |
1144 | */ | |
1145 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); | |
1146 | if (IS_ERR(pinned_sb) || | |
1147 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
1148 | mutex_unlock(&cgroup_mutex); | |
1149 | if (!IS_ERR_OR_NULL(pinned_sb)) | |
1150 | deactivate_super(pinned_sb); | |
1151 | msleep(10); | |
1152 | ret = restart_syscall(); | |
1153 | goto out_free; | |
1154 | } | |
1155 | ||
1156 | ret = 0; | |
1157 | goto out_unlock; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * No such thing, create a new one. name= matching without subsys | |
1162 | * specification is allowed for already existing hierarchies but we | |
1163 | * can't create new one without subsys specification. | |
1164 | */ | |
1165 | if (!opts.subsys_mask && !opts.none) { | |
1166 | ret = -EINVAL; | |
1167 | goto out_unlock; | |
1168 | } | |
1169 | ||
1170 | /* Hierarchies may only be created in the initial cgroup namespace. */ | |
1171 | if (ns != &init_cgroup_ns) { | |
1172 | ret = -EPERM; | |
1173 | goto out_unlock; | |
1174 | } | |
1175 | ||
1176 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1177 | if (!root) { | |
1178 | ret = -ENOMEM; | |
1179 | goto out_unlock; | |
1180 | } | |
1181 | ||
1182 | init_cgroup_root(root, &opts); | |
1183 | ||
1184 | ret = cgroup_setup_root(root, opts.subsys_mask); | |
1185 | if (ret) | |
1186 | cgroup_free_root(root); | |
1187 | ||
1188 | out_unlock: | |
1189 | mutex_unlock(&cgroup_mutex); | |
1190 | out_free: | |
1191 | kfree(opts.release_agent); | |
1192 | kfree(opts.name); | |
1193 | ||
1194 | if (ret) | |
1195 | return ERR_PTR(ret); | |
1196 | ||
1197 | dentry = cgroup_do_mount(&cgroup_fs_type, flags, root, | |
1198 | CGROUP_SUPER_MAGIC, ns); | |
1199 | ||
1200 | /* | |
1201 | * If @pinned_sb, we're reusing an existing root and holding an | |
1202 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1203 | */ | |
1204 | if (pinned_sb) | |
1205 | deactivate_super(pinned_sb); | |
1206 | ||
1207 | return dentry; | |
1208 | } | |
1209 | ||
0a268dbd TH |
1210 | static int __init cgroup1_wq_init(void) |
1211 | { | |
1212 | /* | |
1213 | * Used to destroy pidlists and separate to serve as flush domain. | |
1214 | * Cap @max_active to 1 too. | |
1215 | */ | |
1216 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
1217 | 0, 1); | |
1218 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
1219 | return 0; | |
1220 | } | |
1221 | core_initcall(cgroup1_wq_init); | |
1222 | ||
1223 | static int __init cgroup_no_v1(char *str) | |
1224 | { | |
1225 | struct cgroup_subsys *ss; | |
1226 | char *token; | |
1227 | int i; | |
1228 | ||
1229 | while ((token = strsep(&str, ",")) != NULL) { | |
1230 | if (!*token) | |
1231 | continue; | |
1232 | ||
1233 | if (!strcmp(token, "all")) { | |
1234 | cgroup_no_v1_mask = U16_MAX; | |
1235 | break; | |
1236 | } | |
1237 | ||
1238 | for_each_subsys(ss, i) { | |
1239 | if (strcmp(token, ss->name) && | |
1240 | strcmp(token, ss->legacy_name)) | |
1241 | continue; | |
1242 | ||
1243 | cgroup_no_v1_mask |= 1 << i; | |
1244 | } | |
1245 | } | |
1246 | return 1; | |
1247 | } | |
1248 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
1249 | ||
1250 | ||
1251 | #ifdef CONFIG_CGROUP_DEBUG | |
1252 | static struct cgroup_subsys_state * | |
1253 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
1254 | { | |
1255 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
1256 | ||
1257 | if (!css) | |
1258 | return ERR_PTR(-ENOMEM); | |
1259 | ||
1260 | return css; | |
1261 | } | |
1262 | ||
1263 | static void debug_css_free(struct cgroup_subsys_state *css) | |
1264 | { | |
1265 | kfree(css); | |
1266 | } | |
1267 | ||
1268 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, | |
1269 | struct cftype *cft) | |
1270 | { | |
1271 | return cgroup_task_count(css->cgroup); | |
1272 | } | |
1273 | ||
1274 | static u64 current_css_set_read(struct cgroup_subsys_state *css, | |
1275 | struct cftype *cft) | |
1276 | { | |
1277 | return (u64)(unsigned long)current->cgroups; | |
1278 | } | |
1279 | ||
1280 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, | |
1281 | struct cftype *cft) | |
1282 | { | |
1283 | u64 count; | |
1284 | ||
1285 | rcu_read_lock(); | |
1286 | count = atomic_read(&task_css_set(current)->refcount); | |
1287 | rcu_read_unlock(); | |
1288 | return count; | |
1289 | } | |
1290 | ||
1291 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) | |
1292 | { | |
1293 | struct cgrp_cset_link *link; | |
1294 | struct css_set *cset; | |
1295 | char *name_buf; | |
1296 | ||
1297 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
1298 | if (!name_buf) | |
1299 | return -ENOMEM; | |
1300 | ||
1301 | spin_lock_irq(&css_set_lock); | |
1302 | rcu_read_lock(); | |
1303 | cset = rcu_dereference(current->cgroups); | |
1304 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1305 | struct cgroup *c = link->cgrp; | |
1306 | ||
1307 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
1308 | seq_printf(seq, "Root %d group %s\n", | |
1309 | c->root->hierarchy_id, name_buf); | |
1310 | } | |
1311 | rcu_read_unlock(); | |
1312 | spin_unlock_irq(&css_set_lock); | |
1313 | kfree(name_buf); | |
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
1318 | static int cgroup_css_links_read(struct seq_file *seq, void *v) | |
1319 | { | |
1320 | struct cgroup_subsys_state *css = seq_css(seq); | |
1321 | struct cgrp_cset_link *link; | |
1322 | ||
1323 | spin_lock_irq(&css_set_lock); | |
1324 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { | |
1325 | struct css_set *cset = link->cset; | |
1326 | struct task_struct *task; | |
1327 | int count = 0; | |
1328 | ||
1329 | seq_printf(seq, "css_set %p\n", cset); | |
1330 | ||
1331 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
1332 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1333 | goto overflow; | |
1334 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1335 | } | |
1336 | ||
1337 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
1338 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1339 | goto overflow; | |
1340 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1341 | } | |
1342 | continue; | |
1343 | overflow: | |
1344 | seq_puts(seq, " ...\n"); | |
1345 | } | |
1346 | spin_unlock_irq(&css_set_lock); | |
1347 | return 0; | |
1348 | } | |
1349 | ||
1350 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) | |
1351 | { | |
1352 | return (!cgroup_is_populated(css->cgroup) && | |
1353 | !css_has_online_children(&css->cgroup->self)); | |
1354 | } | |
1355 | ||
1356 | static struct cftype debug_files[] = { | |
1357 | { | |
1358 | .name = "taskcount", | |
1359 | .read_u64 = debug_taskcount_read, | |
1360 | }, | |
1361 | ||
1362 | { | |
1363 | .name = "current_css_set", | |
1364 | .read_u64 = current_css_set_read, | |
1365 | }, | |
1366 | ||
1367 | { | |
1368 | .name = "current_css_set_refcount", | |
1369 | .read_u64 = current_css_set_refcount_read, | |
1370 | }, | |
1371 | ||
1372 | { | |
1373 | .name = "current_css_set_cg_links", | |
1374 | .seq_show = current_css_set_cg_links_read, | |
1375 | }, | |
1376 | ||
1377 | { | |
1378 | .name = "cgroup_css_links", | |
1379 | .seq_show = cgroup_css_links_read, | |
1380 | }, | |
1381 | ||
1382 | { | |
1383 | .name = "releasable", | |
1384 | .read_u64 = releasable_read, | |
1385 | }, | |
1386 | ||
1387 | { } /* terminate */ | |
1388 | }; | |
1389 | ||
1390 | struct cgroup_subsys debug_cgrp_subsys = { | |
1391 | .css_alloc = debug_css_alloc, | |
1392 | .css_free = debug_css_free, | |
1393 | .legacy_cftypes = debug_files, | |
1394 | }; | |
1395 | #endif /* CONFIG_CGROUP_DEBUG */ |