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> |
56cd6973 | 9 | #include <linux/sched/task.h> |
50ff9d13 | 10 | #include <linux/magic.h> |
0a268dbd TH |
11 | #include <linux/slab.h> |
12 | #include <linux/vmalloc.h> | |
13 | #include <linux/delayacct.h> | |
14 | #include <linux/pid_namespace.h> | |
15 | #include <linux/cgroupstats.h> | |
16 | ||
17 | #include <trace/events/cgroup.h> | |
18 | ||
19 | /* | |
20 | * pidlists linger the following amount before being destroyed. The goal | |
21 | * is avoiding frequent destruction in the middle of consecutive read calls | |
22 | * Expiring in the middle is a performance problem not a correctness one. | |
23 | * 1 sec should be enough. | |
24 | */ | |
25 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
26 | ||
27 | /* Controllers blocked by the commandline in v1 */ | |
28 | static u16 cgroup_no_v1_mask; | |
29 | ||
30 | /* | |
31 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
32 | * separate workqueue as flush domain. | |
33 | */ | |
34 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
35 | ||
36 | /* | |
37 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
38 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
39 | */ | |
1592c9b2 | 40 | static DEFINE_SPINLOCK(release_agent_path_lock); |
0a268dbd | 41 | |
d62beb7f | 42 | bool cgroup1_ssid_disabled(int ssid) |
0a268dbd TH |
43 | { |
44 | return cgroup_no_v1_mask & (1 << ssid); | |
45 | } | |
46 | ||
47 | /** | |
48 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
49 | * @from: attach to all cgroups of a given task | |
50 | * @tsk: the task to be attached | |
51 | */ | |
52 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
53 | { | |
54 | struct cgroup_root *root; | |
55 | int retval = 0; | |
56 | ||
57 | mutex_lock(&cgroup_mutex); | |
58 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
59 | for_each_root(root) { | |
60 | struct cgroup *from_cgrp; | |
61 | ||
62 | if (root == &cgrp_dfl_root) | |
63 | continue; | |
64 | ||
65 | spin_lock_irq(&css_set_lock); | |
66 | from_cgrp = task_cgroup_from_root(from, root); | |
67 | spin_unlock_irq(&css_set_lock); | |
68 | ||
69 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
70 | if (retval) | |
71 | break; | |
72 | } | |
73 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
74 | mutex_unlock(&cgroup_mutex); | |
75 | ||
76 | return retval; | |
77 | } | |
78 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
79 | ||
80 | /** | |
81 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
82 | * @to: cgroup to which the tasks will be moved | |
83 | * @from: cgroup in which the tasks currently reside | |
84 | * | |
85 | * Locking rules between cgroup_post_fork() and the migration path | |
86 | * guarantee that, if a task is forking while being migrated, the new child | |
87 | * is guaranteed to be either visible in the source cgroup after the | |
88 | * parent's migration is complete or put into the target cgroup. No task | |
89 | * can slip out of migration through forking. | |
90 | */ | |
91 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
92 | { | |
e595cd70 | 93 | DEFINE_CGROUP_MGCTX(mgctx); |
0a268dbd TH |
94 | struct cgrp_cset_link *link; |
95 | struct css_task_iter it; | |
96 | struct task_struct *task; | |
97 | int ret; | |
98 | ||
99 | if (cgroup_on_dfl(to)) | |
100 | return -EINVAL; | |
101 | ||
102 | if (!cgroup_may_migrate_to(to)) | |
103 | return -EBUSY; | |
104 | ||
105 | mutex_lock(&cgroup_mutex); | |
106 | ||
107 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
108 | ||
109 | /* all tasks in @from are being moved, all csets are source */ | |
110 | spin_lock_irq(&css_set_lock); | |
111 | list_for_each_entry(link, &from->cset_links, cset_link) | |
e595cd70 | 112 | cgroup_migrate_add_src(link->cset, to, &mgctx); |
0a268dbd TH |
113 | spin_unlock_irq(&css_set_lock); |
114 | ||
e595cd70 | 115 | ret = cgroup_migrate_prepare_dst(&mgctx); |
0a268dbd TH |
116 | if (ret) |
117 | goto out_err; | |
118 | ||
119 | /* | |
120 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
121 | * ->can_attach() fails. | |
122 | */ | |
123 | do { | |
124 | css_task_iter_start(&from->self, &it); | |
125 | task = css_task_iter_next(&it); | |
126 | if (task) | |
127 | get_task_struct(task); | |
128 | css_task_iter_end(&it); | |
129 | ||
130 | if (task) { | |
bfc2cf6f | 131 | ret = cgroup_migrate(task, false, &mgctx); |
0a268dbd TH |
132 | if (!ret) |
133 | trace_cgroup_transfer_tasks(to, task, false); | |
134 | put_task_struct(task); | |
135 | } | |
136 | } while (task && !ret); | |
137 | out_err: | |
e595cd70 | 138 | cgroup_migrate_finish(&mgctx); |
0a268dbd TH |
139 | percpu_up_write(&cgroup_threadgroup_rwsem); |
140 | mutex_unlock(&cgroup_mutex); | |
141 | return ret; | |
142 | } | |
143 | ||
144 | /* | |
145 | * Stuff for reading the 'tasks'/'procs' files. | |
146 | * | |
147 | * Reading this file can return large amounts of data if a cgroup has | |
148 | * *lots* of attached tasks. So it may need several calls to read(), | |
149 | * but we cannot guarantee that the information we produce is correct | |
150 | * unless we produce it entirely atomically. | |
151 | * | |
152 | */ | |
153 | ||
154 | /* which pidlist file are we talking about? */ | |
155 | enum cgroup_filetype { | |
156 | CGROUP_FILE_PROCS, | |
157 | CGROUP_FILE_TASKS, | |
158 | }; | |
159 | ||
160 | /* | |
161 | * A pidlist is a list of pids that virtually represents the contents of one | |
162 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
163 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
164 | * to the cgroup. | |
165 | */ | |
166 | struct cgroup_pidlist { | |
167 | /* | |
168 | * used to find which pidlist is wanted. doesn't change as long as | |
169 | * this particular list stays in the list. | |
170 | */ | |
171 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
172 | /* array of xids */ | |
173 | pid_t *list; | |
174 | /* how many elements the above list has */ | |
175 | int length; | |
176 | /* each of these stored in a list by its cgroup */ | |
177 | struct list_head links; | |
178 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
179 | struct cgroup *owner; | |
180 | /* for delayed destruction */ | |
181 | struct delayed_work destroy_dwork; | |
182 | }; | |
183 | ||
184 | /* | |
185 | * The following two functions "fix" the issue where there are more pids | |
186 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
187 | * TODO: replace with a kernel-wide solution to this problem | |
188 | */ | |
189 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
190 | static void *pidlist_allocate(int count) | |
191 | { | |
192 | if (PIDLIST_TOO_LARGE(count)) | |
193 | return vmalloc(count * sizeof(pid_t)); | |
194 | else | |
195 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
196 | } | |
197 | ||
198 | static void pidlist_free(void *p) | |
199 | { | |
200 | kvfree(p); | |
201 | } | |
202 | ||
203 | /* | |
204 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
205 | * should be left afterwards. | |
206 | */ | |
d62beb7f | 207 | void cgroup1_pidlist_destroy_all(struct cgroup *cgrp) |
0a268dbd TH |
208 | { |
209 | struct cgroup_pidlist *l, *tmp_l; | |
210 | ||
211 | mutex_lock(&cgrp->pidlist_mutex); | |
212 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
213 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
214 | mutex_unlock(&cgrp->pidlist_mutex); | |
215 | ||
216 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
217 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
218 | } | |
219 | ||
220 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
221 | { | |
222 | struct delayed_work *dwork = to_delayed_work(work); | |
223 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
224 | destroy_dwork); | |
225 | struct cgroup_pidlist *tofree = NULL; | |
226 | ||
227 | mutex_lock(&l->owner->pidlist_mutex); | |
228 | ||
229 | /* | |
230 | * Destroy iff we didn't get queued again. The state won't change | |
231 | * as destroy_dwork can only be queued while locked. | |
232 | */ | |
233 | if (!delayed_work_pending(dwork)) { | |
234 | list_del(&l->links); | |
235 | pidlist_free(l->list); | |
236 | put_pid_ns(l->key.ns); | |
237 | tofree = l; | |
238 | } | |
239 | ||
240 | mutex_unlock(&l->owner->pidlist_mutex); | |
241 | kfree(tofree); | |
242 | } | |
243 | ||
244 | /* | |
245 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
246 | * Returns the number of unique elements. | |
247 | */ | |
248 | static int pidlist_uniq(pid_t *list, int length) | |
249 | { | |
250 | int src, dest = 1; | |
251 | ||
252 | /* | |
253 | * we presume the 0th element is unique, so i starts at 1. trivial | |
254 | * edge cases first; no work needs to be done for either | |
255 | */ | |
256 | if (length == 0 || length == 1) | |
257 | return length; | |
258 | /* src and dest walk down the list; dest counts unique elements */ | |
259 | for (src = 1; src < length; src++) { | |
260 | /* find next unique element */ | |
261 | while (list[src] == list[src-1]) { | |
262 | src++; | |
263 | if (src == length) | |
264 | goto after; | |
265 | } | |
266 | /* dest always points to where the next unique element goes */ | |
267 | list[dest] = list[src]; | |
268 | dest++; | |
269 | } | |
270 | after: | |
271 | return dest; | |
272 | } | |
273 | ||
274 | /* | |
275 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
276 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
277 | * different per namespace, each namespace needs differently sorted list, | |
278 | * making it impossible to use, for example, single rbtree of member tasks | |
279 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
280 | * per open file is dangerous, so cgroup had to implement shared pool of | |
281 | * pidlists keyed by cgroup and namespace. | |
282 | */ | |
283 | static int cmppid(const void *a, const void *b) | |
284 | { | |
285 | return *(pid_t *)a - *(pid_t *)b; | |
286 | } | |
287 | ||
288 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
289 | enum cgroup_filetype type) | |
290 | { | |
291 | struct cgroup_pidlist *l; | |
292 | /* don't need task_nsproxy() if we're looking at ourself */ | |
293 | struct pid_namespace *ns = task_active_pid_ns(current); | |
294 | ||
295 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
296 | ||
297 | list_for_each_entry(l, &cgrp->pidlists, links) | |
298 | if (l->key.type == type && l->key.ns == ns) | |
299 | return l; | |
300 | return NULL; | |
301 | } | |
302 | ||
303 | /* | |
304 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
305 | * returns with the lock on that pidlist already held, and takes care | |
306 | * of the use count, or returns NULL with no locks held if we're out of | |
307 | * memory. | |
308 | */ | |
309 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
310 | enum cgroup_filetype type) | |
311 | { | |
312 | struct cgroup_pidlist *l; | |
313 | ||
314 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
315 | ||
316 | l = cgroup_pidlist_find(cgrp, type); | |
317 | if (l) | |
318 | return l; | |
319 | ||
320 | /* entry not found; create a new one */ | |
321 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
322 | if (!l) | |
323 | return l; | |
324 | ||
325 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
326 | l->key.type = type; | |
327 | /* don't need task_nsproxy() if we're looking at ourself */ | |
328 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
329 | l->owner = cgrp; | |
330 | list_add(&l->links, &cgrp->pidlists); | |
331 | return l; | |
332 | } | |
333 | ||
334 | /** | |
335 | * cgroup_task_count - count the number of tasks in a cgroup. | |
336 | * @cgrp: the cgroup in question | |
337 | * | |
338 | * Return the number of tasks in the cgroup. The returned number can be | |
339 | * higher than the actual number of tasks due to css_set references from | |
340 | * namespace roots and temporary usages. | |
341 | */ | |
342 | static int cgroup_task_count(const struct cgroup *cgrp) | |
343 | { | |
344 | int count = 0; | |
345 | struct cgrp_cset_link *link; | |
346 | ||
347 | spin_lock_irq(&css_set_lock); | |
348 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
4b9502e6 | 349 | count += refcount_read(&link->cset->refcount); |
0a268dbd TH |
350 | spin_unlock_irq(&css_set_lock); |
351 | return count; | |
352 | } | |
353 | ||
354 | /* | |
355 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
356 | */ | |
357 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
358 | struct cgroup_pidlist **lp) | |
359 | { | |
360 | pid_t *array; | |
361 | int length; | |
362 | int pid, n = 0; /* used for populating the array */ | |
363 | struct css_task_iter it; | |
364 | struct task_struct *tsk; | |
365 | struct cgroup_pidlist *l; | |
366 | ||
367 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
368 | ||
369 | /* | |
370 | * If cgroup gets more users after we read count, we won't have | |
371 | * enough space - tough. This race is indistinguishable to the | |
372 | * caller from the case that the additional cgroup users didn't | |
373 | * show up until sometime later on. | |
374 | */ | |
375 | length = cgroup_task_count(cgrp); | |
376 | array = pidlist_allocate(length); | |
377 | if (!array) | |
378 | return -ENOMEM; | |
379 | /* now, populate the array */ | |
380 | css_task_iter_start(&cgrp->self, &it); | |
381 | while ((tsk = css_task_iter_next(&it))) { | |
382 | if (unlikely(n == length)) | |
383 | break; | |
384 | /* get tgid or pid for procs or tasks file respectively */ | |
385 | if (type == CGROUP_FILE_PROCS) | |
386 | pid = task_tgid_vnr(tsk); | |
387 | else | |
388 | pid = task_pid_vnr(tsk); | |
389 | if (pid > 0) /* make sure to only use valid results */ | |
390 | array[n++] = pid; | |
391 | } | |
392 | css_task_iter_end(&it); | |
393 | length = n; | |
394 | /* now sort & (if procs) strip out duplicates */ | |
395 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
396 | if (type == CGROUP_FILE_PROCS) | |
397 | length = pidlist_uniq(array, length); | |
398 | ||
399 | l = cgroup_pidlist_find_create(cgrp, type); | |
400 | if (!l) { | |
401 | pidlist_free(array); | |
402 | return -ENOMEM; | |
403 | } | |
404 | ||
405 | /* store array, freeing old if necessary */ | |
406 | pidlist_free(l->list); | |
407 | l->list = array; | |
408 | l->length = length; | |
409 | *lp = l; | |
410 | return 0; | |
411 | } | |
412 | ||
413 | /* | |
414 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
415 | * next pid to display; the seq_file iterator is a pointer to the pid | |
416 | * in the cgroup->l->list array. | |
417 | */ | |
418 | ||
419 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
420 | { | |
421 | /* | |
422 | * Initially we receive a position value that corresponds to | |
423 | * one more than the last pid shown (or 0 on the first call or | |
424 | * after a seek to the start). Use a binary-search to find the | |
425 | * next pid to display, if any | |
426 | */ | |
427 | struct kernfs_open_file *of = s->private; | |
428 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
429 | struct cgroup_pidlist *l; | |
430 | enum cgroup_filetype type = seq_cft(s)->private; | |
431 | int index = 0, pid = *pos; | |
432 | int *iter, ret; | |
433 | ||
434 | mutex_lock(&cgrp->pidlist_mutex); | |
435 | ||
436 | /* | |
437 | * !NULL @of->priv indicates that this isn't the first start() | |
438 | * after open. If the matching pidlist is around, we can use that. | |
439 | * Look for it. Note that @of->priv can't be used directly. It | |
440 | * could already have been destroyed. | |
441 | */ | |
442 | if (of->priv) | |
443 | of->priv = cgroup_pidlist_find(cgrp, type); | |
444 | ||
445 | /* | |
446 | * Either this is the first start() after open or the matching | |
447 | * pidlist has been destroyed inbetween. Create a new one. | |
448 | */ | |
449 | if (!of->priv) { | |
450 | ret = pidlist_array_load(cgrp, type, | |
451 | (struct cgroup_pidlist **)&of->priv); | |
452 | if (ret) | |
453 | return ERR_PTR(ret); | |
454 | } | |
455 | l = of->priv; | |
456 | ||
457 | if (pid) { | |
458 | int end = l->length; | |
459 | ||
460 | while (index < end) { | |
461 | int mid = (index + end) / 2; | |
462 | if (l->list[mid] == pid) { | |
463 | index = mid; | |
464 | break; | |
465 | } else if (l->list[mid] <= pid) | |
466 | index = mid + 1; | |
467 | else | |
468 | end = mid; | |
469 | } | |
470 | } | |
471 | /* If we're off the end of the array, we're done */ | |
472 | if (index >= l->length) | |
473 | return NULL; | |
474 | /* Update the abstract position to be the actual pid that we found */ | |
475 | iter = l->list + index; | |
476 | *pos = *iter; | |
477 | return iter; | |
478 | } | |
479 | ||
480 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
481 | { | |
482 | struct kernfs_open_file *of = s->private; | |
483 | struct cgroup_pidlist *l = of->priv; | |
484 | ||
485 | if (l) | |
486 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
487 | CGROUP_PIDLIST_DESTROY_DELAY); | |
488 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); | |
489 | } | |
490 | ||
491 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
492 | { | |
493 | struct kernfs_open_file *of = s->private; | |
494 | struct cgroup_pidlist *l = of->priv; | |
495 | pid_t *p = v; | |
496 | pid_t *end = l->list + l->length; | |
497 | /* | |
498 | * Advance to the next pid in the array. If this goes off the | |
499 | * end, we're done | |
500 | */ | |
501 | p++; | |
502 | if (p >= end) { | |
503 | return NULL; | |
504 | } else { | |
505 | *pos = *p; | |
506 | return p; | |
507 | } | |
508 | } | |
509 | ||
510 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
511 | { | |
512 | seq_printf(s, "%d\n", *(int *)v); | |
513 | ||
514 | return 0; | |
515 | } | |
516 | ||
517 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, | |
518 | char *buf, size_t nbytes, loff_t off) | |
519 | { | |
520 | return __cgroup_procs_write(of, buf, nbytes, off, false); | |
521 | } | |
522 | ||
523 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
524 | char *buf, size_t nbytes, loff_t off) | |
525 | { | |
526 | struct cgroup *cgrp; | |
527 | ||
528 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
529 | ||
530 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
531 | if (!cgrp) | |
532 | return -ENODEV; | |
533 | spin_lock(&release_agent_path_lock); | |
534 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
535 | sizeof(cgrp->root->release_agent_path)); | |
536 | spin_unlock(&release_agent_path_lock); | |
537 | cgroup_kn_unlock(of->kn); | |
538 | return nbytes; | |
539 | } | |
540 | ||
541 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
542 | { | |
543 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
544 | ||
545 | spin_lock(&release_agent_path_lock); | |
546 | seq_puts(seq, cgrp->root->release_agent_path); | |
547 | spin_unlock(&release_agent_path_lock); | |
548 | seq_putc(seq, '\n'); | |
549 | return 0; | |
550 | } | |
551 | ||
552 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
553 | { | |
554 | seq_puts(seq, "0\n"); | |
555 | return 0; | |
556 | } | |
557 | ||
558 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
559 | struct cftype *cft) | |
560 | { | |
561 | return notify_on_release(css->cgroup); | |
562 | } | |
563 | ||
564 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
565 | struct cftype *cft, u64 val) | |
566 | { | |
567 | if (val) | |
568 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
569 | else | |
570 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
571 | return 0; | |
572 | } | |
573 | ||
574 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
575 | struct cftype *cft) | |
576 | { | |
577 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
578 | } | |
579 | ||
580 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
581 | struct cftype *cft, u64 val) | |
582 | { | |
583 | if (val) | |
584 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
585 | else | |
586 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
587 | return 0; | |
588 | } | |
589 | ||
590 | /* cgroup core interface files for the legacy hierarchies */ | |
d62beb7f | 591 | struct cftype cgroup1_base_files[] = { |
0a268dbd TH |
592 | { |
593 | .name = "cgroup.procs", | |
594 | .seq_start = cgroup_pidlist_start, | |
595 | .seq_next = cgroup_pidlist_next, | |
596 | .seq_stop = cgroup_pidlist_stop, | |
597 | .seq_show = cgroup_pidlist_show, | |
598 | .private = CGROUP_FILE_PROCS, | |
599 | .write = cgroup_procs_write, | |
600 | }, | |
601 | { | |
602 | .name = "cgroup.clone_children", | |
603 | .read_u64 = cgroup_clone_children_read, | |
604 | .write_u64 = cgroup_clone_children_write, | |
605 | }, | |
606 | { | |
607 | .name = "cgroup.sane_behavior", | |
608 | .flags = CFTYPE_ONLY_ON_ROOT, | |
609 | .seq_show = cgroup_sane_behavior_show, | |
610 | }, | |
611 | { | |
612 | .name = "tasks", | |
613 | .seq_start = cgroup_pidlist_start, | |
614 | .seq_next = cgroup_pidlist_next, | |
615 | .seq_stop = cgroup_pidlist_stop, | |
616 | .seq_show = cgroup_pidlist_show, | |
617 | .private = CGROUP_FILE_TASKS, | |
618 | .write = cgroup_tasks_write, | |
619 | }, | |
620 | { | |
621 | .name = "notify_on_release", | |
622 | .read_u64 = cgroup_read_notify_on_release, | |
623 | .write_u64 = cgroup_write_notify_on_release, | |
624 | }, | |
625 | { | |
626 | .name = "release_agent", | |
627 | .flags = CFTYPE_ONLY_ON_ROOT, | |
628 | .seq_show = cgroup_release_agent_show, | |
629 | .write = cgroup_release_agent_write, | |
630 | .max_write_len = PATH_MAX - 1, | |
631 | }, | |
632 | { } /* terminate */ | |
633 | }; | |
634 | ||
635 | /* Display information about each subsystem and each hierarchy */ | |
636 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
637 | { | |
638 | struct cgroup_subsys *ss; | |
639 | int i; | |
640 | ||
641 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
642 | /* | |
643 | * ideally we don't want subsystems moving around while we do this. | |
644 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
645 | * subsys/hierarchy state. | |
646 | */ | |
647 | mutex_lock(&cgroup_mutex); | |
648 | ||
649 | for_each_subsys(ss, i) | |
650 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
651 | ss->legacy_name, ss->root->hierarchy_id, | |
652 | atomic_read(&ss->root->nr_cgrps), | |
653 | cgroup_ssid_enabled(i)); | |
654 | ||
655 | mutex_unlock(&cgroup_mutex); | |
656 | return 0; | |
657 | } | |
658 | ||
659 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
660 | { | |
661 | return single_open(file, proc_cgroupstats_show, NULL); | |
662 | } | |
663 | ||
664 | const struct file_operations proc_cgroupstats_operations = { | |
665 | .open = cgroupstats_open, | |
666 | .read = seq_read, | |
667 | .llseek = seq_lseek, | |
668 | .release = single_release, | |
669 | }; | |
670 | ||
671 | /** | |
672 | * cgroupstats_build - build and fill cgroupstats | |
673 | * @stats: cgroupstats to fill information into | |
674 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
675 | * been requested. | |
676 | * | |
677 | * Build and fill cgroupstats so that taskstats can export it to user | |
678 | * space. | |
679 | */ | |
680 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
681 | { | |
682 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
683 | struct cgroup *cgrp; | |
684 | struct css_task_iter it; | |
685 | struct task_struct *tsk; | |
686 | ||
687 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
688 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
689 | kernfs_type(kn) != KERNFS_DIR) | |
690 | return -EINVAL; | |
691 | ||
692 | mutex_lock(&cgroup_mutex); | |
693 | ||
694 | /* | |
695 | * We aren't being called from kernfs and there's no guarantee on | |
696 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
697 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
698 | */ | |
699 | rcu_read_lock(); | |
e0aed7c7 | 700 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
0a268dbd TH |
701 | if (!cgrp || cgroup_is_dead(cgrp)) { |
702 | rcu_read_unlock(); | |
703 | mutex_unlock(&cgroup_mutex); | |
704 | return -ENOENT; | |
705 | } | |
706 | rcu_read_unlock(); | |
707 | ||
708 | css_task_iter_start(&cgrp->self, &it); | |
709 | while ((tsk = css_task_iter_next(&it))) { | |
710 | switch (tsk->state) { | |
711 | case TASK_RUNNING: | |
712 | stats->nr_running++; | |
713 | break; | |
714 | case TASK_INTERRUPTIBLE: | |
715 | stats->nr_sleeping++; | |
716 | break; | |
717 | case TASK_UNINTERRUPTIBLE: | |
718 | stats->nr_uninterruptible++; | |
719 | break; | |
720 | case TASK_STOPPED: | |
721 | stats->nr_stopped++; | |
722 | break; | |
723 | default: | |
724 | if (delayacct_is_task_waiting_on_io(tsk)) | |
725 | stats->nr_io_wait++; | |
726 | break; | |
727 | } | |
728 | } | |
729 | css_task_iter_end(&it); | |
730 | ||
731 | mutex_unlock(&cgroup_mutex); | |
732 | return 0; | |
733 | } | |
734 | ||
d62beb7f | 735 | void cgroup1_check_for_release(struct cgroup *cgrp) |
0a268dbd TH |
736 | { |
737 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
738 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
739 | schedule_work(&cgrp->release_agent_work); | |
740 | } | |
741 | ||
742 | /* | |
743 | * Notify userspace when a cgroup is released, by running the | |
744 | * configured release agent with the name of the cgroup (path | |
745 | * relative to the root of cgroup file system) as the argument. | |
746 | * | |
747 | * Most likely, this user command will try to rmdir this cgroup. | |
748 | * | |
749 | * This races with the possibility that some other task will be | |
750 | * attached to this cgroup before it is removed, or that some other | |
751 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
752 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
753 | * unused, and this cgroup will be reprieved from its death sentence, | |
754 | * to continue to serve a useful existence. Next time it's released, | |
755 | * we will get notified again, if it still has 'notify_on_release' set. | |
756 | * | |
757 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
758 | * means only wait until the task is successfully execve()'d. The | |
759 | * separate release agent task is forked by call_usermodehelper(), | |
760 | * then control in this thread returns here, without waiting for the | |
761 | * release agent task. We don't bother to wait because the caller of | |
762 | * this routine has no use for the exit status of the release agent | |
763 | * task, so no sense holding our caller up for that. | |
764 | */ | |
d62beb7f | 765 | void cgroup1_release_agent(struct work_struct *work) |
0a268dbd TH |
766 | { |
767 | struct cgroup *cgrp = | |
768 | container_of(work, struct cgroup, release_agent_work); | |
769 | char *pathbuf = NULL, *agentbuf = NULL; | |
770 | char *argv[3], *envp[3]; | |
771 | int ret; | |
772 | ||
773 | mutex_lock(&cgroup_mutex); | |
774 | ||
775 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
776 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
777 | if (!pathbuf || !agentbuf) | |
778 | goto out; | |
779 | ||
780 | spin_lock_irq(&css_set_lock); | |
781 | ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
782 | spin_unlock_irq(&css_set_lock); | |
783 | if (ret < 0 || ret >= PATH_MAX) | |
784 | goto out; | |
785 | ||
786 | argv[0] = agentbuf; | |
787 | argv[1] = pathbuf; | |
788 | argv[2] = NULL; | |
789 | ||
790 | /* minimal command environment */ | |
791 | envp[0] = "HOME=/"; | |
792 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
793 | envp[2] = NULL; | |
794 | ||
795 | mutex_unlock(&cgroup_mutex); | |
796 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
797 | goto out_free; | |
798 | out: | |
799 | mutex_unlock(&cgroup_mutex); | |
800 | out_free: | |
801 | kfree(agentbuf); | |
802 | kfree(pathbuf); | |
803 | } | |
804 | ||
805 | /* | |
806 | * cgroup_rename - Only allow simple rename of directories in place. | |
807 | */ | |
1592c9b2 TH |
808 | static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
809 | const char *new_name_str) | |
0a268dbd TH |
810 | { |
811 | struct cgroup *cgrp = kn->priv; | |
812 | int ret; | |
813 | ||
814 | if (kernfs_type(kn) != KERNFS_DIR) | |
815 | return -ENOTDIR; | |
816 | if (kn->parent != new_parent) | |
817 | return -EIO; | |
818 | ||
0a268dbd TH |
819 | /* |
820 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
821 | * active_ref. kernfs_rename() doesn't require active_ref | |
822 | * protection. Break them before grabbing cgroup_mutex. | |
823 | */ | |
824 | kernfs_break_active_protection(new_parent); | |
825 | kernfs_break_active_protection(kn); | |
826 | ||
827 | mutex_lock(&cgroup_mutex); | |
828 | ||
829 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
830 | if (!ret) | |
831 | trace_cgroup_rename(cgrp); | |
832 | ||
833 | mutex_unlock(&cgroup_mutex); | |
834 | ||
835 | kernfs_unbreak_active_protection(kn); | |
836 | kernfs_unbreak_active_protection(new_parent); | |
837 | return ret; | |
838 | } | |
839 | ||
1592c9b2 TH |
840 | static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root) |
841 | { | |
842 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
843 | struct cgroup_subsys *ss; | |
844 | int ssid; | |
845 | ||
846 | for_each_subsys(ss, ssid) | |
847 | if (root->subsys_mask & (1 << ssid)) | |
848 | seq_show_option(seq, ss->legacy_name, NULL); | |
849 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
850 | seq_puts(seq, ",noprefix"); | |
851 | if (root->flags & CGRP_ROOT_XATTR) | |
852 | seq_puts(seq, ",xattr"); | |
853 | ||
854 | spin_lock(&release_agent_path_lock); | |
855 | if (strlen(root->release_agent_path)) | |
856 | seq_show_option(seq, "release_agent", | |
857 | root->release_agent_path); | |
858 | spin_unlock(&release_agent_path_lock); | |
859 | ||
860 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
861 | seq_puts(seq, ",clone_children"); | |
862 | if (strlen(root->name)) | |
863 | seq_show_option(seq, "name", root->name); | |
864 | return 0; | |
865 | } | |
866 | ||
867 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) | |
868 | { | |
869 | char *token, *o = data; | |
870 | bool all_ss = false, one_ss = false; | |
871 | u16 mask = U16_MAX; | |
872 | struct cgroup_subsys *ss; | |
873 | int nr_opts = 0; | |
874 | int i; | |
875 | ||
876 | #ifdef CONFIG_CPUSETS | |
877 | mask = ~((u16)1 << cpuset_cgrp_id); | |
878 | #endif | |
879 | ||
880 | memset(opts, 0, sizeof(*opts)); | |
881 | ||
882 | while ((token = strsep(&o, ",")) != NULL) { | |
883 | nr_opts++; | |
884 | ||
885 | if (!*token) | |
886 | return -EINVAL; | |
887 | if (!strcmp(token, "none")) { | |
888 | /* Explicitly have no subsystems */ | |
889 | opts->none = true; | |
890 | continue; | |
891 | } | |
892 | if (!strcmp(token, "all")) { | |
893 | /* Mutually exclusive option 'all' + subsystem name */ | |
894 | if (one_ss) | |
895 | return -EINVAL; | |
896 | all_ss = true; | |
897 | continue; | |
898 | } | |
899 | if (!strcmp(token, "noprefix")) { | |
900 | opts->flags |= CGRP_ROOT_NOPREFIX; | |
901 | continue; | |
902 | } | |
903 | if (!strcmp(token, "clone_children")) { | |
904 | opts->cpuset_clone_children = true; | |
905 | continue; | |
906 | } | |
907 | if (!strcmp(token, "xattr")) { | |
908 | opts->flags |= CGRP_ROOT_XATTR; | |
909 | continue; | |
910 | } | |
911 | if (!strncmp(token, "release_agent=", 14)) { | |
912 | /* Specifying two release agents is forbidden */ | |
913 | if (opts->release_agent) | |
914 | return -EINVAL; | |
915 | opts->release_agent = | |
916 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); | |
917 | if (!opts->release_agent) | |
918 | return -ENOMEM; | |
919 | continue; | |
920 | } | |
921 | if (!strncmp(token, "name=", 5)) { | |
922 | const char *name = token + 5; | |
923 | /* Can't specify an empty name */ | |
924 | if (!strlen(name)) | |
925 | return -EINVAL; | |
926 | /* Must match [\w.-]+ */ | |
927 | for (i = 0; i < strlen(name); i++) { | |
928 | char c = name[i]; | |
929 | if (isalnum(c)) | |
930 | continue; | |
931 | if ((c == '.') || (c == '-') || (c == '_')) | |
932 | continue; | |
933 | return -EINVAL; | |
934 | } | |
935 | /* Specifying two names is forbidden */ | |
936 | if (opts->name) | |
937 | return -EINVAL; | |
938 | opts->name = kstrndup(name, | |
939 | MAX_CGROUP_ROOT_NAMELEN - 1, | |
940 | GFP_KERNEL); | |
941 | if (!opts->name) | |
942 | return -ENOMEM; | |
943 | ||
944 | continue; | |
945 | } | |
946 | ||
947 | for_each_subsys(ss, i) { | |
948 | if (strcmp(token, ss->legacy_name)) | |
949 | continue; | |
950 | if (!cgroup_ssid_enabled(i)) | |
951 | continue; | |
d62beb7f | 952 | if (cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
953 | continue; |
954 | ||
955 | /* Mutually exclusive option 'all' + subsystem name */ | |
956 | if (all_ss) | |
957 | return -EINVAL; | |
958 | opts->subsys_mask |= (1 << i); | |
959 | one_ss = true; | |
960 | ||
961 | break; | |
962 | } | |
963 | if (i == CGROUP_SUBSYS_COUNT) | |
964 | return -ENOENT; | |
965 | } | |
966 | ||
967 | /* | |
968 | * If the 'all' option was specified select all the subsystems, | |
969 | * otherwise if 'none', 'name=' and a subsystem name options were | |
970 | * not specified, let's default to 'all' | |
971 | */ | |
972 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
973 | for_each_subsys(ss, i) | |
d62beb7f | 974 | if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i)) |
1592c9b2 TH |
975 | opts->subsys_mask |= (1 << i); |
976 | ||
977 | /* | |
978 | * We either have to specify by name or by subsystems. (So all | |
979 | * empty hierarchies must have a name). | |
980 | */ | |
981 | if (!opts->subsys_mask && !opts->name) | |
982 | return -EINVAL; | |
983 | ||
984 | /* | |
985 | * Option noprefix was introduced just for backward compatibility | |
986 | * with the old cpuset, so we allow noprefix only if mounting just | |
987 | * the cpuset subsystem. | |
988 | */ | |
989 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) | |
990 | return -EINVAL; | |
991 | ||
992 | /* Can't specify "none" and some subsystems */ | |
993 | if (opts->subsys_mask && opts->none) | |
994 | return -EINVAL; | |
995 | ||
996 | return 0; | |
997 | } | |
998 | ||
999 | static int cgroup1_remount(struct kernfs_root *kf_root, int *flags, char *data) | |
1000 | { | |
1001 | int ret = 0; | |
1002 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
1003 | struct cgroup_sb_opts opts; | |
1004 | u16 added_mask, removed_mask; | |
1005 | ||
1006 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1007 | ||
1008 | /* See what subsystems are wanted */ | |
1009 | ret = parse_cgroupfs_options(data, &opts); | |
1010 | if (ret) | |
1011 | goto out_unlock; | |
1012 | ||
1013 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) | |
1014 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", | |
1015 | task_tgid_nr(current), current->comm); | |
1016 | ||
1017 | added_mask = opts.subsys_mask & ~root->subsys_mask; | |
1018 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
1019 | ||
1020 | /* Don't allow flags or name to change at remount */ | |
1021 | if ((opts.flags ^ root->flags) || | |
1022 | (opts.name && strcmp(opts.name, root->name))) { | |
1023 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", | |
1024 | opts.flags, opts.name ?: "", root->flags, root->name); | |
1025 | ret = -EINVAL; | |
1026 | goto out_unlock; | |
1027 | } | |
1028 | ||
1029 | /* remounting is not allowed for populated hierarchies */ | |
1030 | if (!list_empty(&root->cgrp.self.children)) { | |
1031 | ret = -EBUSY; | |
1032 | goto out_unlock; | |
1033 | } | |
1034 | ||
1035 | ret = rebind_subsystems(root, added_mask); | |
1036 | if (ret) | |
1037 | goto out_unlock; | |
1038 | ||
1039 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); | |
1040 | ||
1041 | if (opts.release_agent) { | |
1042 | spin_lock(&release_agent_path_lock); | |
1043 | strcpy(root->release_agent_path, opts.release_agent); | |
1044 | spin_unlock(&release_agent_path_lock); | |
1045 | } | |
1046 | ||
1047 | trace_cgroup_remount(root); | |
1048 | ||
1049 | out_unlock: | |
1050 | kfree(opts.release_agent); | |
1051 | kfree(opts.name); | |
1052 | mutex_unlock(&cgroup_mutex); | |
1053 | return ret; | |
1054 | } | |
1055 | ||
1056 | struct kernfs_syscall_ops cgroup1_kf_syscall_ops = { | |
1057 | .rename = cgroup1_rename, | |
1058 | .show_options = cgroup1_show_options, | |
1059 | .remount_fs = cgroup1_remount, | |
1060 | .mkdir = cgroup_mkdir, | |
1061 | .rmdir = cgroup_rmdir, | |
1062 | .show_path = cgroup_show_path, | |
1063 | }; | |
1064 | ||
1065 | struct dentry *cgroup1_mount(struct file_system_type *fs_type, int flags, | |
1066 | void *data, unsigned long magic, | |
1067 | struct cgroup_namespace *ns) | |
1068 | { | |
1069 | struct super_block *pinned_sb = NULL; | |
1070 | struct cgroup_sb_opts opts; | |
1071 | struct cgroup_root *root; | |
1072 | struct cgroup_subsys *ss; | |
1073 | struct dentry *dentry; | |
1074 | int i, ret; | |
9732adc5 | 1075 | bool new_root = false; |
1592c9b2 TH |
1076 | |
1077 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1078 | ||
1079 | /* First find the desired set of subsystems */ | |
1080 | ret = parse_cgroupfs_options(data, &opts); | |
1081 | if (ret) | |
1082 | goto out_unlock; | |
1083 | ||
1084 | /* | |
1085 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1086 | * still be dying after the previous unmount. Let's drain the | |
1087 | * dying subsystems. We just need to ensure that the ones | |
1088 | * unmounted previously finish dying and don't care about new ones | |
1089 | * starting. Testing ref liveliness is good enough. | |
1090 | */ | |
1091 | for_each_subsys(ss, i) { | |
1092 | if (!(opts.subsys_mask & (1 << i)) || | |
1093 | ss->root == &cgrp_dfl_root) | |
1094 | continue; | |
1095 | ||
1096 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1097 | mutex_unlock(&cgroup_mutex); | |
1098 | msleep(10); | |
1099 | ret = restart_syscall(); | |
1100 | goto out_free; | |
1101 | } | |
1102 | cgroup_put(&ss->root->cgrp); | |
1103 | } | |
1104 | ||
1105 | for_each_root(root) { | |
1106 | bool name_match = false; | |
1107 | ||
1108 | if (root == &cgrp_dfl_root) | |
1109 | continue; | |
1110 | ||
1111 | /* | |
1112 | * If we asked for a name then it must match. Also, if | |
1113 | * name matches but sybsys_mask doesn't, we should fail. | |
1114 | * Remember whether name matched. | |
1115 | */ | |
1116 | if (opts.name) { | |
1117 | if (strcmp(opts.name, root->name)) | |
1118 | continue; | |
1119 | name_match = true; | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * If we asked for subsystems (or explicitly for no | |
1124 | * subsystems) then they must match. | |
1125 | */ | |
1126 | if ((opts.subsys_mask || opts.none) && | |
1127 | (opts.subsys_mask != root->subsys_mask)) { | |
1128 | if (!name_match) | |
1129 | continue; | |
1130 | ret = -EBUSY; | |
1131 | goto out_unlock; | |
1132 | } | |
1133 | ||
1134 | if (root->flags ^ opts.flags) | |
1135 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
1136 | ||
1137 | /* | |
1138 | * We want to reuse @root whose lifetime is governed by its | |
1139 | * ->cgrp. Let's check whether @root is alive and keep it | |
1140 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1141 | * want to block it by pinning the sb so that @root doesn't | |
1142 | * get killed before mount is complete. | |
1143 | * | |
1144 | * With the sb pinned, tryget_live can reliably indicate | |
1145 | * whether @root can be reused. If it's being killed, | |
1146 | * drain it. We can use wait_queue for the wait but this | |
1147 | * path is super cold. Let's just sleep a bit and retry. | |
1148 | */ | |
1149 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); | |
330c4186 | 1150 | if (IS_ERR(pinned_sb) || |
1592c9b2 TH |
1151 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { |
1152 | mutex_unlock(&cgroup_mutex); | |
1153 | if (!IS_ERR_OR_NULL(pinned_sb)) | |
1154 | deactivate_super(pinned_sb); | |
1155 | msleep(10); | |
1156 | ret = restart_syscall(); | |
1157 | goto out_free; | |
1158 | } | |
1159 | ||
1160 | ret = 0; | |
1161 | goto out_unlock; | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * No such thing, create a new one. name= matching without subsys | |
1166 | * specification is allowed for already existing hierarchies but we | |
1167 | * can't create new one without subsys specification. | |
1168 | */ | |
1169 | if (!opts.subsys_mask && !opts.none) { | |
1170 | ret = -EINVAL; | |
1171 | goto out_unlock; | |
1172 | } | |
1173 | ||
1174 | /* Hierarchies may only be created in the initial cgroup namespace. */ | |
1175 | if (ns != &init_cgroup_ns) { | |
1176 | ret = -EPERM; | |
1177 | goto out_unlock; | |
1178 | } | |
1179 | ||
1180 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1181 | if (!root) { | |
1182 | ret = -ENOMEM; | |
1183 | goto out_unlock; | |
1184 | } | |
9732adc5 | 1185 | new_root = true; |
1592c9b2 TH |
1186 | |
1187 | init_cgroup_root(root, &opts); | |
1188 | ||
9732adc5 | 1189 | ret = cgroup_setup_root(root, opts.subsys_mask, PERCPU_REF_INIT_DEAD); |
1592c9b2 TH |
1190 | if (ret) |
1191 | cgroup_free_root(root); | |
1192 | ||
1193 | out_unlock: | |
1194 | mutex_unlock(&cgroup_mutex); | |
1195 | out_free: | |
1196 | kfree(opts.release_agent); | |
1197 | kfree(opts.name); | |
1198 | ||
1199 | if (ret) | |
1200 | return ERR_PTR(ret); | |
1201 | ||
1202 | dentry = cgroup_do_mount(&cgroup_fs_type, flags, root, | |
1203 | CGROUP_SUPER_MAGIC, ns); | |
1204 | ||
9732adc5 ZL |
1205 | /* |
1206 | * There's a race window after we release cgroup_mutex and before | |
1207 | * allocating a superblock. Make sure a concurrent process won't | |
1208 | * be able to re-use the root during this window by delaying the | |
1209 | * initialization of root refcnt. | |
1210 | */ | |
1211 | if (new_root) { | |
1212 | mutex_lock(&cgroup_mutex); | |
1213 | percpu_ref_reinit(&root->cgrp.self.refcnt); | |
1214 | mutex_unlock(&cgroup_mutex); | |
1215 | } | |
1216 | ||
1592c9b2 TH |
1217 | /* |
1218 | * If @pinned_sb, we're reusing an existing root and holding an | |
1219 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1220 | */ | |
1221 | if (pinned_sb) | |
1222 | deactivate_super(pinned_sb); | |
1223 | ||
1224 | return dentry; | |
1225 | } | |
1226 | ||
0a268dbd TH |
1227 | static int __init cgroup1_wq_init(void) |
1228 | { | |
1229 | /* | |
1230 | * Used to destroy pidlists and separate to serve as flush domain. | |
1231 | * Cap @max_active to 1 too. | |
1232 | */ | |
1233 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
1234 | 0, 1); | |
1235 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
1236 | return 0; | |
1237 | } | |
1238 | core_initcall(cgroup1_wq_init); | |
1239 | ||
1240 | static int __init cgroup_no_v1(char *str) | |
1241 | { | |
1242 | struct cgroup_subsys *ss; | |
1243 | char *token; | |
1244 | int i; | |
1245 | ||
1246 | while ((token = strsep(&str, ",")) != NULL) { | |
1247 | if (!*token) | |
1248 | continue; | |
1249 | ||
1250 | if (!strcmp(token, "all")) { | |
1251 | cgroup_no_v1_mask = U16_MAX; | |
1252 | break; | |
1253 | } | |
1254 | ||
1255 | for_each_subsys(ss, i) { | |
1256 | if (strcmp(token, ss->name) && | |
1257 | strcmp(token, ss->legacy_name)) | |
1258 | continue; | |
1259 | ||
1260 | cgroup_no_v1_mask |= 1 << i; | |
1261 | } | |
1262 | } | |
1263 | return 1; | |
1264 | } | |
1265 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
1266 | ||
1267 | ||
1268 | #ifdef CONFIG_CGROUP_DEBUG | |
1269 | static struct cgroup_subsys_state * | |
1270 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
1271 | { | |
1272 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
1273 | ||
1274 | if (!css) | |
1275 | return ERR_PTR(-ENOMEM); | |
1276 | ||
1277 | return css; | |
1278 | } | |
1279 | ||
1280 | static void debug_css_free(struct cgroup_subsys_state *css) | |
1281 | { | |
1282 | kfree(css); | |
1283 | } | |
1284 | ||
1285 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, | |
1286 | struct cftype *cft) | |
1287 | { | |
1288 | return cgroup_task_count(css->cgroup); | |
1289 | } | |
1290 | ||
1291 | static u64 current_css_set_read(struct cgroup_subsys_state *css, | |
1292 | struct cftype *cft) | |
1293 | { | |
1294 | return (u64)(unsigned long)current->cgroups; | |
1295 | } | |
1296 | ||
1297 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, | |
1298 | struct cftype *cft) | |
1299 | { | |
1300 | u64 count; | |
1301 | ||
1302 | rcu_read_lock(); | |
4b9502e6 | 1303 | count = refcount_read(&task_css_set(current)->refcount); |
0a268dbd TH |
1304 | rcu_read_unlock(); |
1305 | return count; | |
1306 | } | |
1307 | ||
1308 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) | |
1309 | { | |
1310 | struct cgrp_cset_link *link; | |
1311 | struct css_set *cset; | |
1312 | char *name_buf; | |
1313 | ||
1314 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
1315 | if (!name_buf) | |
1316 | return -ENOMEM; | |
1317 | ||
1318 | spin_lock_irq(&css_set_lock); | |
1319 | rcu_read_lock(); | |
1320 | cset = rcu_dereference(current->cgroups); | |
1321 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1322 | struct cgroup *c = link->cgrp; | |
1323 | ||
1324 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
1325 | seq_printf(seq, "Root %d group %s\n", | |
1326 | c->root->hierarchy_id, name_buf); | |
1327 | } | |
1328 | rcu_read_unlock(); | |
1329 | spin_unlock_irq(&css_set_lock); | |
1330 | kfree(name_buf); | |
1331 | return 0; | |
1332 | } | |
1333 | ||
1334 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
1335 | static int cgroup_css_links_read(struct seq_file *seq, void *v) | |
1336 | { | |
1337 | struct cgroup_subsys_state *css = seq_css(seq); | |
1338 | struct cgrp_cset_link *link; | |
1339 | ||
1340 | spin_lock_irq(&css_set_lock); | |
1341 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { | |
1342 | struct css_set *cset = link->cset; | |
1343 | struct task_struct *task; | |
1344 | int count = 0; | |
1345 | ||
b6a6759d | 1346 | seq_printf(seq, "css_set %pK\n", cset); |
0a268dbd TH |
1347 | |
1348 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
1349 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1350 | goto overflow; | |
1351 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1352 | } | |
1353 | ||
1354 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
1355 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
1356 | goto overflow; | |
1357 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
1358 | } | |
1359 | continue; | |
1360 | overflow: | |
1361 | seq_puts(seq, " ...\n"); | |
1362 | } | |
1363 | spin_unlock_irq(&css_set_lock); | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) | |
1368 | { | |
1369 | return (!cgroup_is_populated(css->cgroup) && | |
1370 | !css_has_online_children(&css->cgroup->self)); | |
1371 | } | |
1372 | ||
1373 | static struct cftype debug_files[] = { | |
1374 | { | |
1375 | .name = "taskcount", | |
1376 | .read_u64 = debug_taskcount_read, | |
1377 | }, | |
1378 | ||
1379 | { | |
1380 | .name = "current_css_set", | |
1381 | .read_u64 = current_css_set_read, | |
1382 | }, | |
1383 | ||
1384 | { | |
1385 | .name = "current_css_set_refcount", | |
1386 | .read_u64 = current_css_set_refcount_read, | |
1387 | }, | |
1388 | ||
1389 | { | |
1390 | .name = "current_css_set_cg_links", | |
1391 | .seq_show = current_css_set_cg_links_read, | |
1392 | }, | |
1393 | ||
1394 | { | |
1395 | .name = "cgroup_css_links", | |
1396 | .seq_show = cgroup_css_links_read, | |
1397 | }, | |
1398 | ||
1399 | { | |
1400 | .name = "releasable", | |
1401 | .read_u64 = releasable_read, | |
1402 | }, | |
1403 | ||
1404 | { } /* terminate */ | |
1405 | }; | |
1406 | ||
1407 | struct cgroup_subsys debug_cgrp_subsys = { | |
1408 | .css_alloc = debug_css_alloc, | |
1409 | .css_free = debug_css_free, | |
1410 | .legacy_cftypes = debug_files, | |
1411 | }; | |
1412 | #endif /* CONFIG_CGROUP_DEBUG */ |