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 | 7 | #include <linux/mm.h> |
c3edc401 | 8 | #include <linux/sched/signal.h> |
0a268dbd TH |
9 | #include <linux/slab.h> |
10 | #include <linux/vmalloc.h> | |
11 | #include <linux/delayacct.h> | |
12 | #include <linux/pid_namespace.h> | |
13 | #include <linux/cgroupstats.h> | |
14 | ||
15 | #include <trace/events/cgroup.h> | |
16 | ||
17 | /* | |
18 | * pidlists linger the following amount before being destroyed. The goal | |
19 | * is avoiding frequent destruction in the middle of consecutive read calls | |
20 | * Expiring in the middle is a performance problem not a correctness one. | |
21 | * 1 sec should be enough. | |
22 | */ | |
23 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
24 | ||
25 | /* Controllers blocked by the commandline in v1 */ | |
26 | static u16 cgroup_no_v1_mask; | |
27 | ||
28 | /* | |
29 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
30 | * separate workqueue as flush domain. | |
31 | */ | |
32 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
33 | ||
34 | /* | |
35 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
36 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
37 | */ | |
1592c9b2 | 38 | static DEFINE_SPINLOCK(release_agent_path_lock); |
0a268dbd | 39 | |
d62beb7f | 40 | bool cgroup1_ssid_disabled(int ssid) |
0a268dbd TH |
41 | { |
42 | return cgroup_no_v1_mask & (1 << ssid); | |
43 | } | |
44 | ||
45 | /** | |
46 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
47 | * @from: attach to all cgroups of a given task | |
48 | * @tsk: the task to be attached | |
49 | */ | |
50 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
51 | { | |
52 | struct cgroup_root *root; | |
53 | int retval = 0; | |
54 | ||
55 | mutex_lock(&cgroup_mutex); | |
56 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
57 | for_each_root(root) { | |
58 | struct cgroup *from_cgrp; | |
59 | ||
60 | if (root == &cgrp_dfl_root) | |
61 | continue; | |
62 | ||
63 | spin_lock_irq(&css_set_lock); | |
64 | from_cgrp = task_cgroup_from_root(from, root); | |
65 | spin_unlock_irq(&css_set_lock); | |
66 | ||
67 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
68 | if (retval) | |
69 | break; | |
70 | } | |
71 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
72 | mutex_unlock(&cgroup_mutex); | |
73 | ||
74 | return retval; | |
75 | } | |
76 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
77 | ||
78 | /** | |
79 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
80 | * @to: cgroup to which the tasks will be moved | |
81 | * @from: cgroup in which the tasks currently reside | |
82 | * | |
83 | * Locking rules between cgroup_post_fork() and the migration path | |
84 | * guarantee that, if a task is forking while being migrated, the new child | |
85 | * is guaranteed to be either visible in the source cgroup after the | |
86 | * parent's migration is complete or put into the target cgroup. No task | |
87 | * can slip out of migration through forking. | |
88 | */ | |
89 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
90 | { | |
e595cd70 | 91 | DEFINE_CGROUP_MGCTX(mgctx); |
0a268dbd TH |
92 | struct cgrp_cset_link *link; |
93 | struct css_task_iter it; | |
94 | struct task_struct *task; | |
95 | int ret; | |
96 | ||
97 | if (cgroup_on_dfl(to)) | |
98 | return -EINVAL; | |
99 | ||
100 | if (!cgroup_may_migrate_to(to)) | |
101 | return -EBUSY; | |
102 | ||
103 | mutex_lock(&cgroup_mutex); | |
104 | ||
105 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
106 | ||
107 | /* all tasks in @from are being moved, all csets are source */ | |
108 | spin_lock_irq(&css_set_lock); | |
109 | list_for_each_entry(link, &from->cset_links, cset_link) | |
e595cd70 | 110 | cgroup_migrate_add_src(link->cset, to, &mgctx); |
0a268dbd TH |
111 | spin_unlock_irq(&css_set_lock); |
112 | ||
e595cd70 | 113 | ret = cgroup_migrate_prepare_dst(&mgctx); |
0a268dbd TH |
114 | if (ret) |
115 | goto out_err; | |
116 | ||
117 | /* | |
118 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
119 | * ->can_attach() fails. | |
120 | */ | |
121 | do { | |
122 | css_task_iter_start(&from->self, &it); | |
123 | task = css_task_iter_next(&it); | |
124 | if (task) | |
125 | get_task_struct(task); | |
126 | css_task_iter_end(&it); | |
127 | ||
128 | if (task) { | |
bfc2cf6f | 129 | ret = cgroup_migrate(task, false, &mgctx); |
0a268dbd TH |
130 | if (!ret) |
131 | trace_cgroup_transfer_tasks(to, task, false); | |
132 | put_task_struct(task); | |
133 | } | |
134 | } while (task && !ret); | |
135 | out_err: | |
e595cd70 | 136 | cgroup_migrate_finish(&mgctx); |
0a268dbd TH |
137 | percpu_up_write(&cgroup_threadgroup_rwsem); |
138 | mutex_unlock(&cgroup_mutex); | |
139 | return ret; | |
140 | } | |
141 | ||
142 | /* | |
143 | * Stuff for reading the 'tasks'/'procs' files. | |
144 | * | |
145 | * Reading this file can return large amounts of data if a cgroup has | |
146 | * *lots* of attached tasks. So it may need several calls to read(), | |
147 | * but we cannot guarantee that the information we produce is correct | |
148 | * unless we produce it entirely atomically. | |
149 | * | |
150 | */ | |
151 | ||
152 | /* which pidlist file are we talking about? */ | |
153 | enum cgroup_filetype { | |
154 | CGROUP_FILE_PROCS, | |
155 | CGROUP_FILE_TASKS, | |
156 | }; | |
157 | ||
158 | /* | |
159 | * A pidlist is a list of pids that virtually represents the contents of one | |
160 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
161 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
162 | * to the cgroup. | |
163 | */ | |
164 | struct cgroup_pidlist { | |
165 | /* | |
166 | * used to find which pidlist is wanted. doesn't change as long as | |
167 | * this particular list stays in the list. | |
168 | */ | |
169 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
170 | /* array of xids */ | |
171 | pid_t *list; | |
172 | /* how many elements the above list has */ | |
173 | int length; | |
174 | /* each of these stored in a list by its cgroup */ | |
175 | struct list_head links; | |
176 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
177 | struct cgroup *owner; | |
178 | /* for delayed destruction */ | |
179 | struct delayed_work destroy_dwork; | |
180 | }; | |
181 | ||
182 | /* | |
183 | * The following two functions "fix" the issue where there are more pids | |
184 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
185 | * TODO: replace with a kernel-wide solution to this problem | |
186 | */ | |
187 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
188 | static void *pidlist_allocate(int count) | |
189 | { | |
190 | if (PIDLIST_TOO_LARGE(count)) | |
191 | return vmalloc(count * sizeof(pid_t)); | |
192 | else | |
193 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
194 | } | |
195 | ||
196 | static void pidlist_free(void *p) | |
197 | { | |
198 | kvfree(p); | |
199 | } | |
200 | ||
201 | /* | |
202 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
203 | * should be left afterwards. | |
204 | */ | |
d62beb7f | 205 | void cgroup1_pidlist_destroy_all(struct cgroup *cgrp) |
0a268dbd TH |
206 | { |
207 | struct cgroup_pidlist *l, *tmp_l; | |
208 | ||
209 | mutex_lock(&cgrp->pidlist_mutex); | |
210 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
211 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
212 | mutex_unlock(&cgrp->pidlist_mutex); | |
213 | ||
214 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
215 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
216 | } | |
217 | ||
218 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
219 | { | |
220 | struct delayed_work *dwork = to_delayed_work(work); | |
221 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
222 | destroy_dwork); | |
223 | struct cgroup_pidlist *tofree = NULL; | |
224 | ||
225 | mutex_lock(&l->owner->pidlist_mutex); | |
226 | ||
227 | /* | |
228 | * Destroy iff we didn't get queued again. The state won't change | |
229 | * as destroy_dwork can only be queued while locked. | |
230 | */ | |
231 | if (!delayed_work_pending(dwork)) { | |
232 | list_del(&l->links); | |
233 | pidlist_free(l->list); | |
234 | put_pid_ns(l->key.ns); | |
235 | tofree = l; | |
236 | } | |
237 | ||
238 | mutex_unlock(&l->owner->pidlist_mutex); | |
239 | kfree(tofree); | |
240 | } | |
241 | ||
242 | /* | |
243 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
244 | * Returns the number of unique elements. | |
245 | */ | |
246 | static int pidlist_uniq(pid_t *list, int length) | |
247 | { | |
248 | int src, dest = 1; | |
249 | ||
250 | /* | |
251 | * we presume the 0th element is unique, so i starts at 1. trivial | |
252 | * edge cases first; no work needs to be done for either | |
253 | */ | |
254 | if (length == 0 || length == 1) | |
255 | return length; | |
256 | /* src and dest walk down the list; dest counts unique elements */ | |
257 | for (src = 1; src < length; src++) { | |
258 | /* find next unique element */ | |
259 | while (list[src] == list[src-1]) { | |
260 | src++; | |
261 | if (src == length) | |
262 | goto after; | |
263 | } | |
264 | /* dest always points to where the next unique element goes */ | |
265 | list[dest] = list[src]; | |
266 | dest++; | |
267 | } | |
268 | after: | |
269 | return dest; | |
270 | } | |
271 | ||
272 | /* | |
273 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
274 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
275 | * different per namespace, each namespace needs differently sorted list, | |
276 | * making it impossible to use, for example, single rbtree of member tasks | |
277 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
278 | * per open file is dangerous, so cgroup had to implement shared pool of | |
279 | * pidlists keyed by cgroup and namespace. | |
280 | */ | |
281 | static int cmppid(const void *a, const void *b) | |
282 | { | |
283 | return *(pid_t *)a - *(pid_t *)b; | |
284 | } | |
285 | ||
286 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
287 | enum cgroup_filetype type) | |
288 | { | |
289 | struct cgroup_pidlist *l; | |
290 | /* don't need task_nsproxy() if we're looking at ourself */ | |
291 | struct pid_namespace *ns = task_active_pid_ns(current); | |
292 | ||
293 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
294 | ||
295 | list_for_each_entry(l, &cgrp->pidlists, links) | |
296 | if (l->key.type == type && l->key.ns == ns) | |
297 | return l; | |
298 | return NULL; | |
299 | } | |
300 | ||
301 | /* | |
302 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
303 | * returns with the lock on that pidlist already held, and takes care | |
304 | * of the use count, or returns NULL with no locks held if we're out of | |
305 | * memory. | |
306 | */ | |
307 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
308 | enum cgroup_filetype type) | |
309 | { | |
310 | struct cgroup_pidlist *l; | |
311 | ||
312 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
313 | ||
314 | l = cgroup_pidlist_find(cgrp, type); | |
315 | if (l) | |
316 | return l; | |
317 | ||
318 | /* entry not found; create a new one */ | |
319 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
320 | if (!l) | |
321 | return l; | |
322 | ||
323 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
324 | l->key.type = type; | |
325 | /* don't need task_nsproxy() if we're looking at ourself */ | |
326 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
327 | l->owner = cgrp; | |
328 | list_add(&l->links, &cgrp->pidlists); | |
329 | return l; | |
330 | } | |
331 | ||
332 | /** | |
333 | * cgroup_task_count - count the number of tasks in a cgroup. | |
334 | * @cgrp: the cgroup in question | |
335 | * | |
336 | * Return the number of tasks in the cgroup. The returned number can be | |
337 | * higher than the actual number of tasks due to css_set references from | |
338 | * namespace roots and temporary usages. | |
339 | */ | |
340 | static int cgroup_task_count(const struct cgroup *cgrp) | |
341 | { | |
342 | int count = 0; | |
343 | struct cgrp_cset_link *link; | |
344 | ||
345 | spin_lock_irq(&css_set_lock); | |
346 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
347 | count += atomic_read(&link->cset->refcount); | |
348 | spin_unlock_irq(&css_set_lock); | |
349 | return count; | |
350 | } | |
351 | ||
352 | /* | |
353 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
354 | */ | |
355 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
356 | struct cgroup_pidlist **lp) | |
357 | { | |
358 | pid_t *array; | |
359 | int length; | |
360 | int pid, n = 0; /* used for populating the array */ | |
361 | struct css_task_iter it; | |
362 | struct task_struct *tsk; | |
363 | struct cgroup_pidlist *l; | |
364 | ||
365 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
366 | ||
367 | /* | |
368 | * If cgroup gets more users after we read count, we won't have | |
369 | * enough space - tough. This race is indistinguishable to the | |
370 | * caller from the case that the additional cgroup users didn't | |
371 | * show up until sometime later on. | |
372 | */ | |
373 | length = cgroup_task_count(cgrp); | |
374 | array = pidlist_allocate(length); | |
375 | if (!array) | |
376 | return -ENOMEM; | |
377 | /* now, populate the array */ | |
378 | css_task_iter_start(&cgrp->self, &it); | |
379 | while ((tsk = css_task_iter_next(&it))) { | |
380 | if (unlikely(n == length)) | |
381 | break; | |
382 | /* get tgid or pid for procs or tasks file respectively */ | |
383 | if (type == CGROUP_FILE_PROCS) | |
384 | pid = task_tgid_vnr(tsk); | |
385 | else | |
386 | pid = task_pid_vnr(tsk); | |
387 | if (pid > 0) /* make sure to only use valid results */ | |
388 | array[n++] = pid; | |
389 | } | |
390 | css_task_iter_end(&it); | |
391 | length = n; | |
392 | /* now sort & (if procs) strip out duplicates */ | |
393 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
394 | if (type == CGROUP_FILE_PROCS) | |
395 | length = pidlist_uniq(array, length); | |
396 | ||
397 | l = cgroup_pidlist_find_create(cgrp, type); | |
398 | if (!l) { | |
399 | pidlist_free(array); | |
400 | return -ENOMEM; | |
401 | } | |
402 | ||
403 | /* store array, freeing old if necessary */ | |
404 | pidlist_free(l->list); | |
405 | l->list = array; | |
406 | l->length = length; | |
407 | *lp = l; | |
408 | return 0; | |
409 | } | |
410 | ||
411 | /* | |
412 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
413 | * next pid to display; the seq_file iterator is a pointer to the pid | |
414 | * in the cgroup->l->list array. | |
415 | */ | |
416 | ||
417 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
418 | { | |
419 | /* | |
420 | * Initially we receive a position value that corresponds to | |
421 | * one more than the last pid shown (or 0 on the first call or | |
422 | * after a seek to the start). Use a binary-search to find the | |
423 | * next pid to display, if any | |
424 | */ | |
425 | struct kernfs_open_file *of = s->private; | |
426 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
427 | struct cgroup_pidlist *l; | |
428 | enum cgroup_filetype type = seq_cft(s)->private; | |
429 | int index = 0, pid = *pos; | |
430 | int *iter, ret; | |
431 | ||
432 | mutex_lock(&cgrp->pidlist_mutex); | |
433 | ||
434 | /* | |
435 | * !NULL @of->priv indicates that this isn't the first start() | |
436 | * after open. If the matching pidlist is around, we can use that. | |
437 | * Look for it. Note that @of->priv can't be used directly. It | |
438 | * could already have been destroyed. | |
439 | */ | |
440 | if (of->priv) | |
441 | of->priv = cgroup_pidlist_find(cgrp, type); | |
442 | ||
443 | /* | |
444 | * Either this is the first start() after open or the matching | |
445 | * pidlist has been destroyed inbetween. Create a new one. | |
446 | */ | |
447 | if (!of->priv) { | |
448 | ret = pidlist_array_load(cgrp, type, | |
449 | (struct cgroup_pidlist **)&of->priv); | |
450 | if (ret) | |
451 | return ERR_PTR(ret); | |
452 | } | |
453 | l = of->priv; | |
454 | ||
455 | if (pid) { | |
456 | int end = l->length; | |
457 | ||
458 | while (index < end) { | |
459 | int mid = (index + end) / 2; | |
460 | if (l->list[mid] == pid) { | |
461 | index = mid; | |
462 | break; | |
463 | } else if (l->list[mid] <= pid) | |
464 | index = mid + 1; | |
465 | else | |
466 | end = mid; | |
467 | } | |
468 | } | |
469 | /* If we're off the end of the array, we're done */ | |
470 | if (index >= l->length) | |
471 | return NULL; | |
472 | /* Update the abstract position to be the actual pid that we found */ | |
473 | iter = l->list + index; | |
474 | *pos = *iter; | |
475 | return iter; | |
476 | } | |
477 | ||
478 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
479 | { | |
480 | struct kernfs_open_file *of = s->private; | |
481 | struct cgroup_pidlist *l = of->priv; | |
482 | ||
483 | if (l) | |
484 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
485 | CGROUP_PIDLIST_DESTROY_DELAY); | |
486 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); | |
487 | } | |
488 | ||
489 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
490 | { | |
491 | struct kernfs_open_file *of = s->private; | |
492 | struct cgroup_pidlist *l = of->priv; | |
493 | pid_t *p = v; | |
494 | pid_t *end = l->list + l->length; | |
495 | /* | |
496 | * Advance to the next pid in the array. If this goes off the | |
497 | * end, we're done | |
498 | */ | |
499 | p++; | |
500 | if (p >= end) { | |
501 | return NULL; | |
502 | } else { | |
503 | *pos = *p; | |
504 | return p; | |
505 | } | |
506 | } | |
507 | ||
508 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
509 | { | |
510 | seq_printf(s, "%d\n", *(int *)v); | |
511 | ||
512 | return 0; | |
513 | } | |
514 | ||
515 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, | |
516 | char *buf, size_t nbytes, loff_t off) | |
517 | { | |
518 | return __cgroup_procs_write(of, buf, nbytes, off, false); | |
519 | } | |
520 | ||
521 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
522 | char *buf, size_t nbytes, loff_t off) | |
523 | { | |
524 | struct cgroup *cgrp; | |
525 | ||
526 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
527 | ||
528 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
529 | if (!cgrp) | |
530 | return -ENODEV; | |
531 | spin_lock(&release_agent_path_lock); | |
532 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
533 | sizeof(cgrp->root->release_agent_path)); | |
534 | spin_unlock(&release_agent_path_lock); | |
535 | cgroup_kn_unlock(of->kn); | |
536 | return nbytes; | |
537 | } | |
538 | ||
539 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
540 | { | |
541 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
542 | ||
543 | spin_lock(&release_agent_path_lock); | |
544 | seq_puts(seq, cgrp->root->release_agent_path); | |
545 | spin_unlock(&release_agent_path_lock); | |
546 | seq_putc(seq, '\n'); | |
547 | return 0; | |
548 | } | |
549 | ||
550 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
551 | { | |
552 | seq_puts(seq, "0\n"); | |
553 | return 0; | |
554 | } | |
555 | ||
556 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
557 | struct cftype *cft) | |
558 | { | |
559 | return notify_on_release(css->cgroup); | |
560 | } | |
561 | ||
562 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
563 | struct cftype *cft, u64 val) | |
564 | { | |
565 | if (val) | |
566 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
567 | else | |
568 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
569 | return 0; | |
570 | } | |
571 | ||
572 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
573 | struct cftype *cft) | |
574 | { | |
575 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
576 | } | |
577 | ||
578 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
579 | struct cftype *cft, u64 val) | |
580 | { | |
581 | if (val) | |
582 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
583 | else | |
584 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
585 | return 0; | |
586 | } | |
587 | ||
588 | /* cgroup core interface files for the legacy hierarchies */ | |
d62beb7f | 589 | struct cftype cgroup1_base_files[] = { |
0a268dbd TH |
590 | { |
591 | .name = "cgroup.procs", | |
592 | .seq_start = cgroup_pidlist_start, | |
593 | .seq_next = cgroup_pidlist_next, | |
594 | .seq_stop = cgroup_pidlist_stop, | |
595 | .seq_show = cgroup_pidlist_show, | |
596 | .private = CGROUP_FILE_PROCS, | |
597 | .write = cgroup_procs_write, | |
598 | }, | |
599 | { | |
600 | .name = "cgroup.clone_children", | |
601 | .read_u64 = cgroup_clone_children_read, | |
602 | .write_u64 = cgroup_clone_children_write, | |
603 | }, | |
604 | { | |
605 | .name = "cgroup.sane_behavior", | |
606 | .flags = CFTYPE_ONLY_ON_ROOT, | |
607 | .seq_show = cgroup_sane_behavior_show, | |
608 | }, | |
609 | { | |
610 | .name = "tasks", | |
611 | .seq_start = cgroup_pidlist_start, | |
612 | .seq_next = cgroup_pidlist_next, | |
613 | .seq_stop = cgroup_pidlist_stop, | |
614 | .seq_show = cgroup_pidlist_show, | |
615 | .private = CGROUP_FILE_TASKS, | |
616 | .write = cgroup_tasks_write, | |
617 | }, | |
618 | { | |
619 | .name = "notify_on_release", | |
620 | .read_u64 = cgroup_read_notify_on_release, | |
621 | .write_u64 = cgroup_write_notify_on_release, | |
622 | }, | |
623 | { | |
624 | .name = "release_agent", | |
625 | .flags = CFTYPE_ONLY_ON_ROOT, | |
626 | .seq_show = cgroup_release_agent_show, | |
627 | .write = cgroup_release_agent_write, | |
628 | .max_write_len = PATH_MAX - 1, | |
629 | }, | |
630 | { } /* terminate */ | |
631 | }; | |
632 | ||
633 | /* Display information about each subsystem and each hierarchy */ | |
634 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
635 | { | |
636 | struct cgroup_subsys *ss; | |
637 | int i; | |
638 | ||
639 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
640 | /* | |
641 | * ideally we don't want subsystems moving around while we do this. | |
642 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
643 | * subsys/hierarchy state. | |
644 | */ | |
645 | mutex_lock(&cgroup_mutex); | |
646 | ||
647 | for_each_subsys(ss, i) | |
648 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
649 | ss->legacy_name, ss->root->hierarchy_id, | |
650 | atomic_read(&ss->root->nr_cgrps), | |
651 | cgroup_ssid_enabled(i)); | |
652 | ||
653 | mutex_unlock(&cgroup_mutex); | |
654 | return 0; | |
655 | } | |
656 | ||
657 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
658 | { | |
659 | return single_open(file, proc_cgroupstats_show, NULL); | |
660 | } | |
661 | ||
662 | const struct file_operations proc_cgroupstats_operations = { | |
663 | .open = cgroupstats_open, | |
664 | .read = seq_read, | |
665 | .llseek = seq_lseek, | |
666 | .release = single_release, | |
667 | }; | |
668 | ||
669 | /** | |
670 | * cgroupstats_build - build and fill cgroupstats | |
671 | * @stats: cgroupstats to fill information into | |
672 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
673 | * been requested. | |
674 | * | |
675 | * Build and fill cgroupstats so that taskstats can export it to user | |
676 | * space. | |
677 | */ | |
678 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
679 | { | |
680 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
681 | struct cgroup *cgrp; | |
682 | struct css_task_iter it; | |
683 | struct task_struct *tsk; | |
684 | ||
685 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
686 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
687 | kernfs_type(kn) != KERNFS_DIR) | |
688 | return -EINVAL; | |
689 | ||
690 | mutex_lock(&cgroup_mutex); | |
691 | ||
692 | /* | |
693 | * We aren't being called from kernfs and there's no guarantee on | |
694 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
695 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
696 | */ | |
697 | rcu_read_lock(); | |
e0aed7c7 | 698 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
0a268dbd TH |
699 | if (!cgrp || cgroup_is_dead(cgrp)) { |
700 | rcu_read_unlock(); | |
701 | mutex_unlock(&cgroup_mutex); | |
702 | return -ENOENT; | |
703 | } | |
704 | rcu_read_unlock(); | |
705 | ||
706 | css_task_iter_start(&cgrp->self, &it); | |
707 | while ((tsk = css_task_iter_next(&it))) { | |
708 | switch (tsk->state) { | |
709 | case TASK_RUNNING: | |
710 | stats->nr_running++; | |
711 | break; | |
712 | case TASK_INTERRUPTIBLE: | |
713 | stats->nr_sleeping++; | |
714 | break; | |
715 | case TASK_UNINTERRUPTIBLE: | |
716 | stats->nr_uninterruptible++; | |
717 | break; | |
718 | case TASK_STOPPED: | |
719 | stats->nr_stopped++; | |
720 | break; | |
721 | default: | |
722 | if (delayacct_is_task_waiting_on_io(tsk)) | |
723 | stats->nr_io_wait++; | |
724 | break; | |
725 | } | |
726 | } | |
727 | css_task_iter_end(&it); | |
728 | ||
729 | mutex_unlock(&cgroup_mutex); | |
730 | return 0; | |
731 | } | |
732 | ||
d62beb7f | 733 | void cgroup1_check_for_release(struct cgroup *cgrp) |
0a268dbd TH |
734 | { |
735 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
736 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
737 | schedule_work(&cgrp->release_agent_work); | |
738 | } | |
739 | ||
740 | /* | |
741 | * Notify userspace when a cgroup is released, by running the | |
742 | * configured release agent with the name of the cgroup (path | |
743 | * relative to the root of cgroup file system) as the argument. | |
744 | * | |
745 | * Most likely, this user command will try to rmdir this cgroup. | |
746 | * | |
747 | * This races with the possibility that some other task will be | |
748 | * attached to this cgroup before it is removed, or that some other | |
749 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
750 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
751 | * unused, and this cgroup will be reprieved from its death sentence, | |
752 | * to continue to serve a useful existence. Next time it's released, | |
753 | * we will get notified again, if it still has 'notify_on_release' set. | |
754 | * | |
755 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
756 | * means only wait until the task is successfully execve()'d. The | |
757 | * separate release agent task is forked by call_usermodehelper(), | |
758 | * then control in this thread returns here, without waiting for the | |
759 | * release agent task. We don't bother to wait because the caller of | |
760 | * this routine has no use for the exit status of the release agent | |
761 | * task, so no sense holding our caller up for that. | |
762 | */ | |
d62beb7f | 763 | void cgroup1_release_agent(struct work_struct *work) |
0a268dbd TH |
764 | { |
765 | struct cgroup *cgrp = | |
766 | container_of(work, struct cgroup, release_agent_work); | |
767 | char *pathbuf = NULL, *agentbuf = NULL; | |
768 | char *argv[3], *envp[3]; | |
769 | int ret; | |
770 | ||
771 | mutex_lock(&cgroup_mutex); | |
772 | ||
773 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
774 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
775 | if (!pathbuf || !agentbuf) | |
776 | goto out; | |
777 | ||
778 | spin_lock_irq(&css_set_lock); | |
779 | ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
780 | spin_unlock_irq(&css_set_lock); | |
781 | if (ret < 0 || ret >= PATH_MAX) | |
782 | goto out; | |
783 | ||
784 | argv[0] = agentbuf; | |
785 | argv[1] = pathbuf; | |
786 | argv[2] = NULL; | |
787 | ||
788 | /* minimal command environment */ | |
789 | envp[0] = "HOME=/"; | |
790 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
791 | envp[2] = NULL; | |
792 | ||
793 | mutex_unlock(&cgroup_mutex); | |
794 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
795 | goto out_free; | |
796 | out: | |
797 | mutex_unlock(&cgroup_mutex); | |
798 | out_free: | |
799 | kfree(agentbuf); | |
800 | kfree(pathbuf); | |
801 | } | |
802 | ||
803 | /* | |
804 | * cgroup_rename - Only allow simple rename of directories in place. | |
805 | */ | |
1592c9b2 TH |
806 | static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
807 | const char *new_name_str) | |
0a268dbd TH |
808 | { |
809 | struct cgroup *cgrp = kn->priv; | |
810 | int ret; | |
811 | ||
812 | if (kernfs_type(kn) != KERNFS_DIR) | |
813 | return -ENOTDIR; | |
814 | if (kn->parent != new_parent) | |
815 | return -EIO; | |
816 | ||
0a268dbd TH |
817 | /* |
818 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
819 | * active_ref. kernfs_rename() doesn't require active_ref | |
820 | * protection. Break them before grabbing cgroup_mutex. | |
821 | */ | |
822 | kernfs_break_active_protection(new_parent); | |
823 | kernfs_break_active_protection(kn); | |
824 | ||
825 | mutex_lock(&cgroup_mutex); | |
826 | ||
827 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
828 | if (!ret) | |
829 | trace_cgroup_rename(cgrp); | |
830 | ||
831 | mutex_unlock(&cgroup_mutex); | |
832 | ||
833 | kernfs_unbreak_active_protection(kn); | |
834 | kernfs_unbreak_active_protection(new_parent); | |
835 | return ret; | |
836 | } | |
837 | ||
1592c9b2 TH |
838 | static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root) |
839 | { | |
840 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
841 | struct cgroup_subsys *ss; | |
842 | int ssid; | |
843 | ||
844 | for_each_subsys(ss, ssid) | |
845 | if (root->subsys_mask & (1 << ssid)) | |
846 | seq_show_option(seq, ss->legacy_name, NULL); | |
847 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
848 | seq_puts(seq, ",noprefix"); | |
849 | if (root->flags & CGRP_ROOT_XATTR) | |
850 | seq_puts(seq, ",xattr"); | |
851 | ||
852 | spin_lock(&release_agent_path_lock); | |
853 | if (strlen(root->release_agent_path)) | |
854 | seq_show_option(seq, "release_agent", | |
855 | root->release_agent_path); | |
856 | spin_unlock(&release_agent_path_lock); | |
857 | ||
858 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
859 | seq_puts(seq, ",clone_children"); | |
860 | if (strlen(root->name)) | |
861 | seq_show_option(seq, "name", root->name); | |
862 | return 0; | |
863 | } | |
864 | ||
865 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) | |
866 | { | |
867 | char *token, *o = data; | |
868 | bool all_ss = false, one_ss = false; | |
869 | u16 mask = U16_MAX; | |
870 | struct cgroup_subsys *ss; | |
871 | int nr_opts = 0; | |
872 | int i; | |
873 | ||
874 | #ifdef CONFIG_CPUSETS | |
875 | mask = ~((u16)1 << cpuset_cgrp_id); | |
876 | #endif | |
877 | ||
878 | memset(opts, 0, sizeof(*opts)); | |
879 | ||
880 | while ((token = strsep(&o, ",")) != NULL) { | |
881 | nr_opts++; | |
882 | ||
883 | if (!*token) | |
884 | return -EINVAL; | |
885 | if (!strcmp(token, "none")) { | |
886 | /* Explicitly have no subsystems */ | |
887 | opts->none = true; | |
888 | continue; | |
889 | } | |
890 | if (!strcmp(token, "all")) { | |
891 | /* Mutually exclusive option 'all' + subsystem name */ | |
892 | if (one_ss) | |
893 | return -EINVAL; | |
894 | all_ss = true; | |
895 | continue; | |
896 | } | |
897 | if (!strcmp(token, "noprefix")) { | |
898 | opts->flags |= CGRP_ROOT_NOPREFIX; | |
899 | continue; | |
900 | } | |
901 | if (!strcmp(token, "clone_children")) { | |
902 | opts->cpuset_clone_children = true; | |
903 | continue; | |
904 | } | |
905 | if (!strcmp(token, "xattr")) { | |
906 | opts->flags |= CGRP_ROOT_XATTR; | |
907 | continue; | |
908 | } | |
909 | if (!strncmp(token, "release_agent=", 14)) { | |
910 | /* Specifying two release agents is forbidden */ | |
911 | if (opts->release_agent) | |
912 | return -EINVAL; | |
913 | opts->release_agent = | |
914 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); | |
915 | if (!opts->release_agent) | |
916 | return -ENOMEM; | |
917 | continue; | |
918 | } | |
919 | if (!strncmp(token, "name=", 5)) { | |
920 | const char *name = token + 5; | |
921 | /* Can't specify an empty name */ | |
922 | if (!strlen(name)) | |
923 | return -EINVAL; | |
924 | /* Must match [\w.-]+ */ | |
925 | for (i = 0; i < strlen(name); i++) { | |
926 | char c = name[i]; | |
927 | if (isalnum(c)) | |
928 | continue; | |
929 | if ((c == '.') || (c == '-') || (c == '_')) | |
930 | continue; | |
931 | return -EINVAL; | |
932 | } | |
933 | /* Specifying two names is forbidden */ | |
934 | if (opts->name) | |
935 | return -EINVAL; | |
936 | opts->name = kstrndup(name, | |
937 | MAX_CGROUP_ROOT_NAMELEN - 1, | |
938 | GFP_KERNEL); | |
939 | if (!opts->name) | |
940 | return -ENOMEM; | |
941 | ||
942 | continue; | |
943 | } | |
944 | ||
945 | for_each_subsys(ss, i) { | |
946 | if (strcmp(token, ss->legacy_name)) | |
947 | continue; | |
948 | if (!cgroup_ssid_enabled(i)) | |
949 | continue; | |
d62beb7f | 950 | if (cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
951 | continue; |
952 | ||
953 | /* Mutually exclusive option 'all' + subsystem name */ | |
954 | if (all_ss) | |
955 | return -EINVAL; | |
956 | opts->subsys_mask |= (1 << i); | |
957 | one_ss = true; | |
958 | ||
959 | break; | |
960 | } | |
961 | if (i == CGROUP_SUBSYS_COUNT) | |
962 | return -ENOENT; | |
963 | } | |
964 | ||
965 | /* | |
966 | * If the 'all' option was specified select all the subsystems, | |
967 | * otherwise if 'none', 'name=' and a subsystem name options were | |
968 | * not specified, let's default to 'all' | |
969 | */ | |
970 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
971 | for_each_subsys(ss, i) | |
d62beb7f | 972 | if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
973 | opts->subsys_mask |= (1 << i); |
974 | ||
975 | /* | |
976 | * We either have to specify by name or by subsystems. (So all | |
977 | * empty hierarchies must have a name). | |
978 | */ | |
979 | if (!opts->subsys_mask && !opts->name) | |
980 | return -EINVAL; | |
981 | ||
982 | /* | |
983 | * Option noprefix was introduced just for backward compatibility | |
984 | * with the old cpuset, so we allow noprefix only if mounting just | |
985 | * the cpuset subsystem. | |
986 | */ | |
987 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) | |
988 | return -EINVAL; | |
989 | ||
990 | /* Can't specify "none" and some subsystems */ | |
991 | if (opts->subsys_mask && opts->none) | |
992 | return -EINVAL; | |
993 | ||
994 | return 0; | |
995 | } | |
996 | ||
997 | static int cgroup1_remount(struct kernfs_root *kf_root, int *flags, char *data) | |
998 | { | |
999 | int ret = 0; | |
1000 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
1001 | struct cgroup_sb_opts opts; | |
1002 | u16 added_mask, removed_mask; | |
1003 | ||
1004 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1005 | ||
1006 | /* See what subsystems are wanted */ | |
1007 | ret = parse_cgroupfs_options(data, &opts); | |
1008 | if (ret) | |
1009 | goto out_unlock; | |
1010 | ||
1011 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) | |
1012 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", | |
1013 | task_tgid_nr(current), current->comm); | |
1014 | ||
1015 | added_mask = opts.subsys_mask & ~root->subsys_mask; | |
1016 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
1017 | ||
1018 | /* Don't allow flags or name to change at remount */ | |
1019 | if ((opts.flags ^ root->flags) || | |
1020 | (opts.name && strcmp(opts.name, root->name))) { | |
1021 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", | |
1022 | opts.flags, opts.name ?: "", root->flags, root->name); | |
1023 | ret = -EINVAL; | |
1024 | goto out_unlock; | |
1025 | } | |
1026 | ||
1027 | /* remounting is not allowed for populated hierarchies */ | |
1028 | if (!list_empty(&root->cgrp.self.children)) { | |
1029 | ret = -EBUSY; | |
1030 | goto out_unlock; | |
1031 | } | |
1032 | ||
1033 | ret = rebind_subsystems(root, added_mask); | |
1034 | if (ret) | |
1035 | goto out_unlock; | |
1036 | ||
1037 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); | |
1038 | ||
1039 | if (opts.release_agent) { | |
1040 | spin_lock(&release_agent_path_lock); | |
1041 | strcpy(root->release_agent_path, opts.release_agent); | |
1042 | spin_unlock(&release_agent_path_lock); | |
1043 | } | |
1044 | ||
1045 | trace_cgroup_remount(root); | |
1046 | ||
1047 | out_unlock: | |
1048 | kfree(opts.release_agent); | |
1049 | kfree(opts.name); | |
1050 | mutex_unlock(&cgroup_mutex); | |
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | struct kernfs_syscall_ops cgroup1_kf_syscall_ops = { | |
1055 | .rename = cgroup1_rename, | |
1056 | .show_options = cgroup1_show_options, | |
1057 | .remount_fs = cgroup1_remount, | |
1058 | .mkdir = cgroup_mkdir, | |
1059 | .rmdir = cgroup_rmdir, | |
1060 | .show_path = cgroup_show_path, | |
1061 | }; | |
1062 | ||
1063 | struct dentry *cgroup1_mount(struct file_system_type *fs_type, int flags, | |
1064 | void *data, unsigned long magic, | |
1065 | struct cgroup_namespace *ns) | |
1066 | { | |
1067 | struct super_block *pinned_sb = NULL; | |
1068 | struct cgroup_sb_opts opts; | |
1069 | struct cgroup_root *root; | |
1070 | struct cgroup_subsys *ss; | |
1071 | struct dentry *dentry; | |
1072 | int i, ret; | |
1073 | ||
1074 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1075 | ||
1076 | /* First find the desired set of subsystems */ | |
1077 | ret = parse_cgroupfs_options(data, &opts); | |
1078 | if (ret) | |
1079 | goto out_unlock; | |
1080 | ||
1081 | /* | |
1082 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1083 | * still be dying after the previous unmount. Let's drain the | |
1084 | * dying subsystems. We just need to ensure that the ones | |
1085 | * unmounted previously finish dying and don't care about new ones | |
1086 | * starting. Testing ref liveliness is good enough. | |
1087 | */ | |
1088 | for_each_subsys(ss, i) { | |
1089 | if (!(opts.subsys_mask & (1 << i)) || | |
1090 | ss->root == &cgrp_dfl_root) | |
1091 | continue; | |
1092 | ||
1093 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1094 | mutex_unlock(&cgroup_mutex); | |
1095 | msleep(10); | |
1096 | ret = restart_syscall(); | |
1097 | goto out_free; | |
1098 | } | |
1099 | cgroup_put(&ss->root->cgrp); | |
1100 | } | |
1101 | ||
1102 | for_each_root(root) { | |
1103 | bool name_match = false; | |
1104 | ||
1105 | if (root == &cgrp_dfl_root) | |
1106 | continue; | |
1107 | ||
1108 | /* | |
1109 | * If we asked for a name then it must match. Also, if | |
1110 | * name matches but sybsys_mask doesn't, we should fail. | |
1111 | * Remember whether name matched. | |
1112 | */ | |
1113 | if (opts.name) { | |
1114 | if (strcmp(opts.name, root->name)) | |
1115 | continue; | |
1116 | name_match = true; | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * If we asked for subsystems (or explicitly for no | |
1121 | * subsystems) then they must match. | |
1122 | */ | |
1123 | if ((opts.subsys_mask || opts.none) && | |
1124 | (opts.subsys_mask != root->subsys_mask)) { | |
1125 | if (!name_match) | |
1126 | continue; | |
1127 | ret = -EBUSY; | |
1128 | goto out_unlock; | |
1129 | } | |
1130 | ||
1131 | if (root->flags ^ opts.flags) | |
1132 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
1133 | ||
1134 | /* | |
1135 | * We want to reuse @root whose lifetime is governed by its | |
1136 | * ->cgrp. Let's check whether @root is alive and keep it | |
1137 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1138 | * want to block it by pinning the sb so that @root doesn't | |
1139 | * get killed before mount is complete. | |
1140 | * | |
1141 | * With the sb pinned, tryget_live can reliably indicate | |
1142 | * whether @root can be reused. If it's being killed, | |
1143 | * drain it. We can use wait_queue for the wait but this | |
1144 | * path is super cold. Let's just sleep a bit and retry. | |
1145 | */ | |
1146 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); | |
1147 | if (IS_ERR(pinned_sb) || | |
1148 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
1149 | mutex_unlock(&cgroup_mutex); | |
1150 | if (!IS_ERR_OR_NULL(pinned_sb)) | |
1151 | deactivate_super(pinned_sb); | |
1152 | msleep(10); | |
1153 | ret = restart_syscall(); | |
1154 | goto out_free; | |
1155 | } | |
1156 | ||
1157 | ret = 0; | |
1158 | goto out_unlock; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * No such thing, create a new one. name= matching without subsys | |
1163 | * specification is allowed for already existing hierarchies but we | |
1164 | * can't create new one without subsys specification. | |
1165 | */ | |
1166 | if (!opts.subsys_mask && !opts.none) { | |
1167 | ret = -EINVAL; | |
1168 | goto out_unlock; | |
1169 | } | |
1170 | ||
1171 | /* Hierarchies may only be created in the initial cgroup namespace. */ | |
1172 | if (ns != &init_cgroup_ns) { | |
1173 | ret = -EPERM; | |
1174 | goto out_unlock; | |
1175 | } | |
1176 | ||
1177 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1178 | if (!root) { | |
1179 | ret = -ENOMEM; | |
1180 | goto out_unlock; | |
1181 | } | |
1182 | ||
1183 | init_cgroup_root(root, &opts); | |
1184 | ||
1185 | ret = cgroup_setup_root(root, opts.subsys_mask); | |
1186 | if (ret) | |
1187 | cgroup_free_root(root); | |
1188 | ||
1189 | out_unlock: | |
1190 | mutex_unlock(&cgroup_mutex); | |
1191 | out_free: | |
1192 | kfree(opts.release_agent); | |
1193 | kfree(opts.name); | |
1194 | ||
1195 | if (ret) | |
1196 | return ERR_PTR(ret); | |
1197 | ||
1198 | dentry = cgroup_do_mount(&cgroup_fs_type, flags, root, | |
1199 | CGROUP_SUPER_MAGIC, ns); | |
1200 | ||
1201 | /* | |
1202 | * If @pinned_sb, we're reusing an existing root and holding an | |
1203 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1204 | */ | |
1205 | if (pinned_sb) | |
1206 | deactivate_super(pinned_sb); | |
1207 | ||
1208 | return dentry; | |
1209 | } | |
1210 | ||
0a268dbd TH |
1211 | static int __init cgroup1_wq_init(void) |
1212 | { | |
1213 | /* | |
1214 | * Used to destroy pidlists and separate to serve as flush domain. | |
1215 | * Cap @max_active to 1 too. | |
1216 | */ | |
1217 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
1218 | 0, 1); | |
1219 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
1220 | return 0; | |
1221 | } | |
1222 | core_initcall(cgroup1_wq_init); | |
1223 | ||
1224 | static int __init cgroup_no_v1(char *str) | |
1225 | { | |
1226 | struct cgroup_subsys *ss; | |
1227 | char *token; | |
1228 | int i; | |
1229 | ||
1230 | while ((token = strsep(&str, ",")) != NULL) { | |
1231 | if (!*token) | |
1232 | continue; | |
1233 | ||
1234 | if (!strcmp(token, "all")) { | |
1235 | cgroup_no_v1_mask = U16_MAX; | |
1236 | break; | |
1237 | } | |
1238 | ||
1239 | for_each_subsys(ss, i) { | |
1240 | if (strcmp(token, ss->name) && | |
1241 | strcmp(token, ss->legacy_name)) | |
1242 | continue; | |
1243 | ||
1244 | cgroup_no_v1_mask |= 1 << i; | |
1245 | } | |
1246 | } | |
1247 | return 1; | |
1248 | } | |
1249 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
1250 | ||
1251 | ||
1252 | #ifdef CONFIG_CGROUP_DEBUG | |
1253 | static struct cgroup_subsys_state * | |
1254 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
1255 | { | |
1256 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
1257 | ||
1258 | if (!css) | |
1259 | return ERR_PTR(-ENOMEM); | |
1260 | ||
1261 | return css; | |
1262 | } | |
1263 | ||
1264 | static void debug_css_free(struct cgroup_subsys_state *css) | |
1265 | { | |
1266 | kfree(css); | |
1267 | } | |
1268 | ||
1269 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, | |
1270 | struct cftype *cft) | |
1271 | { | |
1272 | return cgroup_task_count(css->cgroup); | |
1273 | } | |
1274 | ||
1275 | static u64 current_css_set_read(struct cgroup_subsys_state *css, | |
1276 | struct cftype *cft) | |
1277 | { | |
1278 | return (u64)(unsigned long)current->cgroups; | |
1279 | } | |
1280 | ||
1281 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, | |
1282 | struct cftype *cft) | |
1283 | { | |
1284 | u64 count; | |
1285 | ||
1286 | rcu_read_lock(); | |
1287 | count = atomic_read(&task_css_set(current)->refcount); | |
1288 | rcu_read_unlock(); | |
1289 | return count; | |
1290 | } | |
1291 | ||
1292 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) | |
1293 | { | |
1294 | struct cgrp_cset_link *link; | |
1295 | struct css_set *cset; | |
1296 | char *name_buf; | |
1297 | ||
1298 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
1299 | if (!name_buf) | |
1300 | return -ENOMEM; | |
1301 | ||
1302 | spin_lock_irq(&css_set_lock); | |
1303 | rcu_read_lock(); | |
1304 | cset = rcu_dereference(current->cgroups); | |
1305 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1306 | struct cgroup *c = link->cgrp; | |
1307 | ||
1308 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
1309 | seq_printf(seq, "Root %d group %s\n", | |
1310 | c->root->hierarchy_id, name_buf); | |
1311 | } | |
1312 | rcu_read_unlock(); | |
1313 | spin_unlock_irq(&css_set_lock); | |
1314 | kfree(name_buf); | |
1315 | return 0; | |
1316 | } | |
1317 | ||
1318 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
1319 | static int cgroup_css_links_read(struct seq_file *seq, void *v) | |
1320 | { | |
1321 | struct cgroup_subsys_state *css = seq_css(seq); | |
1322 | struct cgrp_cset_link *link; | |
1323 | ||
1324 | spin_lock_irq(&css_set_lock); | |
1325 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { | |
1326 | struct css_set *cset = link->cset; | |
1327 | struct task_struct *task; | |
1328 | int count = 0; | |
1329 | ||
1330 | seq_printf(seq, "css_set %p\n", cset); | |
1331 | ||
1332 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
1333 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1334 | goto overflow; | |
1335 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1336 | } | |
1337 | ||
1338 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
1339 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1340 | goto overflow; | |
1341 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1342 | } | |
1343 | continue; | |
1344 | overflow: | |
1345 | seq_puts(seq, " ...\n"); | |
1346 | } | |
1347 | spin_unlock_irq(&css_set_lock); | |
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) | |
1352 | { | |
1353 | return (!cgroup_is_populated(css->cgroup) && | |
1354 | !css_has_online_children(&css->cgroup->self)); | |
1355 | } | |
1356 | ||
1357 | static struct cftype debug_files[] = { | |
1358 | { | |
1359 | .name = "taskcount", | |
1360 | .read_u64 = debug_taskcount_read, | |
1361 | }, | |
1362 | ||
1363 | { | |
1364 | .name = "current_css_set", | |
1365 | .read_u64 = current_css_set_read, | |
1366 | }, | |
1367 | ||
1368 | { | |
1369 | .name = "current_css_set_refcount", | |
1370 | .read_u64 = current_css_set_refcount_read, | |
1371 | }, | |
1372 | ||
1373 | { | |
1374 | .name = "current_css_set_cg_links", | |
1375 | .seq_show = current_css_set_cg_links_read, | |
1376 | }, | |
1377 | ||
1378 | { | |
1379 | .name = "cgroup_css_links", | |
1380 | .seq_show = cgroup_css_links_read, | |
1381 | }, | |
1382 | ||
1383 | { | |
1384 | .name = "releasable", | |
1385 | .read_u64 = releasable_read, | |
1386 | }, | |
1387 | ||
1388 | { } /* terminate */ | |
1389 | }; | |
1390 | ||
1391 | struct cgroup_subsys debug_cgrp_subsys = { | |
1392 | .css_alloc = debug_css_alloc, | |
1393 | .css_free = debug_css_free, | |
1394 | .legacy_cftypes = debug_files, | |
1395 | }; | |
1396 | #endif /* CONFIG_CGROUP_DEBUG */ |