]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Generic process-grouping system. | |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
7 | * Notifications support | |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
11 | * Copyright notices from the original cpuset code: | |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
30 | ||
31 | #include <linux/cgroup.h> | |
32 | #include <linux/cred.h> | |
33 | #include <linux/ctype.h> | |
34 | #include <linux/errno.h> | |
35 | #include <linux/init_task.h> | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/list.h> | |
38 | #include <linux/magic.h> | |
39 | #include <linux/mm.h> | |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
43 | #include <linux/proc_fs.h> | |
44 | #include <linux/rcupdate.h> | |
45 | #include <linux/sched.h> | |
46 | #include <linux/slab.h> | |
47 | #include <linux/spinlock.h> | |
48 | #include <linux/percpu-rwsem.h> | |
49 | #include <linux/string.h> | |
50 | #include <linux/sort.h> | |
51 | #include <linux/kmod.h> | |
52 | #include <linux/delayacct.h> | |
53 | #include <linux/cgroupstats.h> | |
54 | #include <linux/hashtable.h> | |
55 | #include <linux/pid_namespace.h> | |
56 | #include <linux/idr.h> | |
57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ | |
58 | #include <linux/kthread.h> | |
59 | #include <linux/delay.h> | |
60 | #include <linux/atomic.h> | |
61 | #include <linux/cpuset.h> | |
62 | #include <net/sock.h> | |
63 | ||
64 | /* | |
65 | * pidlists linger the following amount before being destroyed. The goal | |
66 | * is avoiding frequent destruction in the middle of consecutive read calls | |
67 | * Expiring in the middle is a performance problem not a correctness one. | |
68 | * 1 sec should be enough. | |
69 | */ | |
70 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
71 | ||
72 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ | |
73 | MAX_CFTYPE_NAME + 2) | |
74 | ||
75 | /* | |
76 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
77 | * hierarchy must be performed while holding it. | |
78 | * | |
79 | * css_set_lock protects task->cgroups pointer, the list of css_set | |
80 | * objects, and the chain of tasks off each css_set. | |
81 | * | |
82 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in | |
83 | * cgroup.h can use them for lockdep annotations. | |
84 | */ | |
85 | #ifdef CONFIG_PROVE_RCU | |
86 | DEFINE_MUTEX(cgroup_mutex); | |
87 | DEFINE_SPINLOCK(css_set_lock); | |
88 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
89 | EXPORT_SYMBOL_GPL(css_set_lock); | |
90 | #else | |
91 | static DEFINE_MUTEX(cgroup_mutex); | |
92 | static DEFINE_SPINLOCK(css_set_lock); | |
93 | #endif | |
94 | ||
95 | /* | |
96 | * Protects cgroup_idr and css_idr so that IDs can be released without | |
97 | * grabbing cgroup_mutex. | |
98 | */ | |
99 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
100 | ||
101 | /* | |
102 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
103 | * against file removal/re-creation across css hiding. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
106 | ||
107 | /* | |
108 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
109 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
110 | */ | |
111 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
112 | ||
113 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; | |
114 | ||
115 | #define cgroup_assert_mutex_or_rcu_locked() \ | |
116 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ | |
117 | !lockdep_is_held(&cgroup_mutex), \ | |
118 | "cgroup_mutex or RCU read lock required"); | |
119 | ||
120 | /* | |
121 | * cgroup destruction makes heavy use of work items and there can be a lot | |
122 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
123 | * destruction work items don't end up filling up max_active of system_wq | |
124 | * which may lead to deadlock. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_destroy_wq; | |
127 | ||
128 | /* | |
129 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
130 | * separate workqueue as flush domain. | |
131 | */ | |
132 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
133 | ||
134 | /* generate an array of cgroup subsystem pointers */ | |
135 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, | |
136 | static struct cgroup_subsys *cgroup_subsys[] = { | |
137 | #include <linux/cgroup_subsys.h> | |
138 | }; | |
139 | #undef SUBSYS | |
140 | ||
141 | /* array of cgroup subsystem names */ | |
142 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
143 | static const char *cgroup_subsys_name[] = { | |
144 | #include <linux/cgroup_subsys.h> | |
145 | }; | |
146 | #undef SUBSYS | |
147 | ||
148 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ | |
149 | #define SUBSYS(_x) \ | |
150 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
151 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
152 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
153 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
154 | #include <linux/cgroup_subsys.h> | |
155 | #undef SUBSYS | |
156 | ||
157 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
158 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
159 | #include <linux/cgroup_subsys.h> | |
160 | }; | |
161 | #undef SUBSYS | |
162 | ||
163 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
164 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
165 | #include <linux/cgroup_subsys.h> | |
166 | }; | |
167 | #undef SUBSYS | |
168 | ||
169 | /* | |
170 | * The default hierarchy, reserved for the subsystems that are otherwise | |
171 | * unattached - it never has more than a single cgroup, and all tasks are | |
172 | * part of that cgroup. | |
173 | */ | |
174 | struct cgroup_root cgrp_dfl_root; | |
175 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); | |
176 | ||
177 | /* | |
178 | * The default hierarchy always exists but is hidden until mounted for the | |
179 | * first time. This is for backward compatibility. | |
180 | */ | |
181 | static bool cgrp_dfl_visible; | |
182 | ||
183 | /* Controllers blocked by the commandline in v1 */ | |
184 | static u16 cgroup_no_v1_mask; | |
185 | ||
186 | /* some controllers are not supported in the default hierarchy */ | |
187 | static u16 cgrp_dfl_inhibit_ss_mask; | |
188 | ||
189 | /* The list of hierarchy roots */ | |
190 | ||
191 | static LIST_HEAD(cgroup_roots); | |
192 | static int cgroup_root_count; | |
193 | ||
194 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ | |
195 | static DEFINE_IDR(cgroup_hierarchy_idr); | |
196 | ||
197 | /* | |
198 | * Assign a monotonically increasing serial number to csses. It guarantees | |
199 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
200 | * Also, as csses are always appended to the parent's ->children list, it | |
201 | * guarantees that sibling csses are always sorted in the ascending serial | |
202 | * number order on the list. Protected by cgroup_mutex. | |
203 | */ | |
204 | static u64 css_serial_nr_next = 1; | |
205 | ||
206 | /* | |
207 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
208 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
209 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
210 | */ | |
211 | static u16 have_fork_callback __read_mostly; | |
212 | static u16 have_exit_callback __read_mostly; | |
213 | static u16 have_free_callback __read_mostly; | |
214 | ||
215 | /* Ditto for the can_fork callback. */ | |
216 | static u16 have_canfork_callback __read_mostly; | |
217 | ||
218 | static struct file_system_type cgroup2_fs_type; | |
219 | static struct cftype cgroup_dfl_base_files[]; | |
220 | static struct cftype cgroup_legacy_base_files[]; | |
221 | ||
222 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); | |
223 | static void css_task_iter_advance(struct css_task_iter *it); | |
224 | static int cgroup_destroy_locked(struct cgroup *cgrp); | |
225 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, | |
226 | struct cgroup_subsys *ss); | |
227 | static void css_release(struct percpu_ref *ref); | |
228 | static void kill_css(struct cgroup_subsys_state *css); | |
229 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, | |
230 | struct cgroup *cgrp, struct cftype cfts[], | |
231 | bool is_add); | |
232 | ||
233 | /** | |
234 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
235 | * @ssid: subsys ID of interest | |
236 | * | |
237 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
238 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
239 | * is slower static_key_enabled() based test indexed by @ssid. | |
240 | */ | |
241 | static bool cgroup_ssid_enabled(int ssid) | |
242 | { | |
243 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
244 | } | |
245 | ||
246 | static bool cgroup_ssid_no_v1(int ssid) | |
247 | { | |
248 | return cgroup_no_v1_mask & (1 << ssid); | |
249 | } | |
250 | ||
251 | /** | |
252 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
253 | * @cgrp: the cgroup of interest | |
254 | * | |
255 | * The default hierarchy is the v2 interface of cgroup and this function | |
256 | * can be used to test whether a cgroup is on the default hierarchy for | |
257 | * cases where a subsystem should behave differnetly depending on the | |
258 | * interface version. | |
259 | * | |
260 | * The set of behaviors which change on the default hierarchy are still | |
261 | * being determined and the mount option is prefixed with __DEVEL__. | |
262 | * | |
263 | * List of changed behaviors: | |
264 | * | |
265 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
266 | * and "name" are disallowed. | |
267 | * | |
268 | * - When mounting an existing superblock, mount options should match. | |
269 | * | |
270 | * - Remount is disallowed. | |
271 | * | |
272 | * - rename(2) is disallowed. | |
273 | * | |
274 | * - "tasks" is removed. Everything should be at process granularity. Use | |
275 | * "cgroup.procs" instead. | |
276 | * | |
277 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
278 | * recycled inbetween reads. | |
279 | * | |
280 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
281 | * notification mechanism will be implemented. | |
282 | * | |
283 | * - "cgroup.clone_children" is removed. | |
284 | * | |
285 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
286 | * and its descendants contain no task; otherwise, 1. The file also | |
287 | * generates kernfs notification which can be monitored through poll and | |
288 | * [di]notify when the value of the file changes. | |
289 | * | |
290 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
291 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
292 | * moved to an ancestor. | |
293 | * | |
294 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
295 | * masks of ancestors. | |
296 | * | |
297 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
298 | * is not created. | |
299 | * | |
300 | * - blkcg: blk-throttle becomes properly hierarchical. | |
301 | * | |
302 | * - debug: disallowed on the default hierarchy. | |
303 | */ | |
304 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
305 | { | |
306 | return cgrp->root == &cgrp_dfl_root; | |
307 | } | |
308 | ||
309 | /* IDR wrappers which synchronize using cgroup_idr_lock */ | |
310 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
311 | gfp_t gfp_mask) | |
312 | { | |
313 | int ret; | |
314 | ||
315 | idr_preload(gfp_mask); | |
316 | spin_lock_bh(&cgroup_idr_lock); | |
317 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); | |
318 | spin_unlock_bh(&cgroup_idr_lock); | |
319 | idr_preload_end(); | |
320 | return ret; | |
321 | } | |
322 | ||
323 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
324 | { | |
325 | void *ret; | |
326 | ||
327 | spin_lock_bh(&cgroup_idr_lock); | |
328 | ret = idr_replace(idr, ptr, id); | |
329 | spin_unlock_bh(&cgroup_idr_lock); | |
330 | return ret; | |
331 | } | |
332 | ||
333 | static void cgroup_idr_remove(struct idr *idr, int id) | |
334 | { | |
335 | spin_lock_bh(&cgroup_idr_lock); | |
336 | idr_remove(idr, id); | |
337 | spin_unlock_bh(&cgroup_idr_lock); | |
338 | } | |
339 | ||
340 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) | |
341 | { | |
342 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
343 | ||
344 | if (parent_css) | |
345 | return container_of(parent_css, struct cgroup, self); | |
346 | return NULL; | |
347 | } | |
348 | ||
349 | /** | |
350 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
351 | * @cgrp: the cgroup of interest | |
352 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) | |
353 | * | |
354 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This | |
355 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
356 | * the caller is responsible for pinning the returned css if it wants to | |
357 | * keep accessing it outside the said locks. This function may return | |
358 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
359 | */ | |
360 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
361 | struct cgroup_subsys *ss) | |
362 | { | |
363 | if (ss) | |
364 | return rcu_dereference_check(cgrp->subsys[ss->id], | |
365 | lockdep_is_held(&cgroup_mutex)); | |
366 | else | |
367 | return &cgrp->self; | |
368 | } | |
369 | ||
370 | /** | |
371 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
372 | * @cgrp: the cgroup of interest | |
373 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) | |
374 | * | |
375 | * Similar to cgroup_css() but returns the effective css, which is defined | |
376 | * as the matching css of the nearest ancestor including self which has @ss | |
377 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
378 | * function is guaranteed to return non-NULL css. | |
379 | */ | |
380 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
381 | struct cgroup_subsys *ss) | |
382 | { | |
383 | lockdep_assert_held(&cgroup_mutex); | |
384 | ||
385 | if (!ss) | |
386 | return &cgrp->self; | |
387 | ||
388 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
389 | return NULL; | |
390 | ||
391 | /* | |
392 | * This function is used while updating css associations and thus | |
393 | * can't test the csses directly. Use ->subtree_ss_mask. | |
394 | */ | |
395 | while (cgroup_parent(cgrp) && | |
396 | !(cgroup_parent(cgrp)->subtree_ss_mask & (1 << ss->id))) | |
397 | cgrp = cgroup_parent(cgrp); | |
398 | ||
399 | return cgroup_css(cgrp, ss); | |
400 | } | |
401 | ||
402 | /** | |
403 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
404 | * @cgrp: the cgroup of interest | |
405 | * @ss: the subsystem of interest | |
406 | * | |
407 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
408 | * defined as the matching css of the nearest ancestor including self which | |
409 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
410 | * the root css is returned, so this function always returns a valid css. | |
411 | * The returned css must be put using css_put(). | |
412 | */ | |
413 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
414 | struct cgroup_subsys *ss) | |
415 | { | |
416 | struct cgroup_subsys_state *css; | |
417 | ||
418 | rcu_read_lock(); | |
419 | ||
420 | do { | |
421 | css = cgroup_css(cgrp, ss); | |
422 | ||
423 | if (css && css_tryget_online(css)) | |
424 | goto out_unlock; | |
425 | cgrp = cgroup_parent(cgrp); | |
426 | } while (cgrp); | |
427 | ||
428 | css = init_css_set.subsys[ss->id]; | |
429 | css_get(css); | |
430 | out_unlock: | |
431 | rcu_read_unlock(); | |
432 | return css; | |
433 | } | |
434 | ||
435 | /* convenient tests for these bits */ | |
436 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) | |
437 | { | |
438 | return !(cgrp->self.flags & CSS_ONLINE); | |
439 | } | |
440 | ||
441 | static void cgroup_get(struct cgroup *cgrp) | |
442 | { | |
443 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
444 | css_get(&cgrp->self); | |
445 | } | |
446 | ||
447 | static bool cgroup_tryget(struct cgroup *cgrp) | |
448 | { | |
449 | return css_tryget(&cgrp->self); | |
450 | } | |
451 | ||
452 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) | |
453 | { | |
454 | struct cgroup *cgrp = of->kn->parent->priv; | |
455 | struct cftype *cft = of_cft(of); | |
456 | ||
457 | /* | |
458 | * This is open and unprotected implementation of cgroup_css(). | |
459 | * seq_css() is only called from a kernfs file operation which has | |
460 | * an active reference on the file. Because all the subsystem | |
461 | * files are drained before a css is disassociated with a cgroup, | |
462 | * the matching css from the cgroup's subsys table is guaranteed to | |
463 | * be and stay valid until the enclosing operation is complete. | |
464 | */ | |
465 | if (cft->ss) | |
466 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
467 | else | |
468 | return &cgrp->self; | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(of_css); | |
471 | ||
472 | static int notify_on_release(const struct cgroup *cgrp) | |
473 | { | |
474 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
475 | } | |
476 | ||
477 | /** | |
478 | * for_each_css - iterate all css's of a cgroup | |
479 | * @css: the iteration cursor | |
480 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
481 | * @cgrp: the target cgroup to iterate css's of | |
482 | * | |
483 | * Should be called under cgroup_[tree_]mutex. | |
484 | */ | |
485 | #define for_each_css(css, ssid, cgrp) \ | |
486 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
487 | if (!((css) = rcu_dereference_check( \ | |
488 | (cgrp)->subsys[(ssid)], \ | |
489 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
490 | else | |
491 | ||
492 | /** | |
493 | * for_each_e_css - iterate all effective css's of a cgroup | |
494 | * @css: the iteration cursor | |
495 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
496 | * @cgrp: the target cgroup to iterate css's of | |
497 | * | |
498 | * Should be called under cgroup_[tree_]mutex. | |
499 | */ | |
500 | #define for_each_e_css(css, ssid, cgrp) \ | |
501 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
502 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
503 | ; \ | |
504 | else | |
505 | ||
506 | /** | |
507 | * for_each_subsys - iterate all enabled cgroup subsystems | |
508 | * @ss: the iteration cursor | |
509 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
510 | */ | |
511 | #define for_each_subsys(ss, ssid) \ | |
512 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ | |
513 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
514 | ||
515 | /** | |
516 | * do_each_subsys_mask - filter for_each_subsys with a bitmask | |
517 | * @ss: the iteration cursor | |
518 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
519 | * @ss_mask: the bitmask | |
520 | * | |
521 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
522 | * @ss_mask is set. | |
523 | */ | |
524 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ | |
525 | unsigned long __ss_mask = (ss_mask); \ | |
526 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
527 | (ssid) = 0; \ | |
528 | break; \ | |
529 | } \ | |
530 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
531 | (ss) = cgroup_subsys[ssid]; \ | |
532 | { | |
533 | ||
534 | #define while_each_subsys_mask() \ | |
535 | } \ | |
536 | } \ | |
537 | } while (false) | |
538 | ||
539 | /* iterate across the hierarchies */ | |
540 | #define for_each_root(root) \ | |
541 | list_for_each_entry((root), &cgroup_roots, root_list) | |
542 | ||
543 | /* iterate over child cgrps, lock should be held throughout iteration */ | |
544 | #define cgroup_for_each_live_child(child, cgrp) \ | |
545 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ | |
546 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
547 | cgroup_is_dead(child); })) \ | |
548 | ; \ | |
549 | else | |
550 | ||
551 | static void cgroup_release_agent(struct work_struct *work); | |
552 | static void check_for_release(struct cgroup *cgrp); | |
553 | ||
554 | /* | |
555 | * A cgroup can be associated with multiple css_sets as different tasks may | |
556 | * belong to different cgroups on different hierarchies. In the other | |
557 | * direction, a css_set is naturally associated with multiple cgroups. | |
558 | * This M:N relationship is represented by the following link structure | |
559 | * which exists for each association and allows traversing the associations | |
560 | * from both sides. | |
561 | */ | |
562 | struct cgrp_cset_link { | |
563 | /* the cgroup and css_set this link associates */ | |
564 | struct cgroup *cgrp; | |
565 | struct css_set *cset; | |
566 | ||
567 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
568 | struct list_head cset_link; | |
569 | ||
570 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
571 | struct list_head cgrp_link; | |
572 | }; | |
573 | ||
574 | /* | |
575 | * The default css_set - used by init and its children prior to any | |
576 | * hierarchies being mounted. It contains a pointer to the root state | |
577 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
578 | * reference-counted, to improve performance when child cgroups | |
579 | * haven't been created. | |
580 | */ | |
581 | struct css_set init_css_set = { | |
582 | .refcount = ATOMIC_INIT(1), | |
583 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
584 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
585 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
586 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
587 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
588 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), | |
589 | }; | |
590 | ||
591 | static int css_set_count = 1; /* 1 for init_css_set */ | |
592 | ||
593 | /** | |
594 | * css_set_populated - does a css_set contain any tasks? | |
595 | * @cset: target css_set | |
596 | */ | |
597 | static bool css_set_populated(struct css_set *cset) | |
598 | { | |
599 | lockdep_assert_held(&css_set_lock); | |
600 | ||
601 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
602 | } | |
603 | ||
604 | /** | |
605 | * cgroup_update_populated - updated populated count of a cgroup | |
606 | * @cgrp: the target cgroup | |
607 | * @populated: inc or dec populated count | |
608 | * | |
609 | * One of the css_sets associated with @cgrp is either getting its first | |
610 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
611 | * count is propagated towards root so that a given cgroup's populated_cnt | |
612 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
613 | * | |
614 | * @cgrp's interface file "cgroup.populated" is zero if | |
615 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
616 | * changes from or to zero, userland is notified that the content of the | |
617 | * interface file has changed. This can be used to detect when @cgrp and | |
618 | * its descendants become populated or empty. | |
619 | */ | |
620 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
621 | { | |
622 | lockdep_assert_held(&css_set_lock); | |
623 | ||
624 | do { | |
625 | bool trigger; | |
626 | ||
627 | if (populated) | |
628 | trigger = !cgrp->populated_cnt++; | |
629 | else | |
630 | trigger = !--cgrp->populated_cnt; | |
631 | ||
632 | if (!trigger) | |
633 | break; | |
634 | ||
635 | check_for_release(cgrp); | |
636 | cgroup_file_notify(&cgrp->events_file); | |
637 | ||
638 | cgrp = cgroup_parent(cgrp); | |
639 | } while (cgrp); | |
640 | } | |
641 | ||
642 | /** | |
643 | * css_set_update_populated - update populated state of a css_set | |
644 | * @cset: target css_set | |
645 | * @populated: whether @cset is populated or depopulated | |
646 | * | |
647 | * @cset is either getting the first task or losing the last. Update the | |
648 | * ->populated_cnt of all associated cgroups accordingly. | |
649 | */ | |
650 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
651 | { | |
652 | struct cgrp_cset_link *link; | |
653 | ||
654 | lockdep_assert_held(&css_set_lock); | |
655 | ||
656 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
657 | cgroup_update_populated(link->cgrp, populated); | |
658 | } | |
659 | ||
660 | /** | |
661 | * css_set_move_task - move a task from one css_set to another | |
662 | * @task: task being moved | |
663 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
664 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
665 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
666 | * | |
667 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
668 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
669 | * instead of moved, @to_cset can be NULL. | |
670 | * | |
671 | * This function automatically handles populated_cnt updates and | |
672 | * css_task_iter adjustments but the caller is responsible for managing | |
673 | * @from_cset and @to_cset's reference counts. | |
674 | */ | |
675 | static void css_set_move_task(struct task_struct *task, | |
676 | struct css_set *from_cset, struct css_set *to_cset, | |
677 | bool use_mg_tasks) | |
678 | { | |
679 | lockdep_assert_held(&css_set_lock); | |
680 | ||
681 | if (to_cset && !css_set_populated(to_cset)) | |
682 | css_set_update_populated(to_cset, true); | |
683 | ||
684 | if (from_cset) { | |
685 | struct css_task_iter *it, *pos; | |
686 | ||
687 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
688 | ||
689 | /* | |
690 | * @task is leaving, advance task iterators which are | |
691 | * pointing to it so that they can resume at the next | |
692 | * position. Advancing an iterator might remove it from | |
693 | * the list, use safe walk. See css_task_iter_advance*() | |
694 | * for details. | |
695 | */ | |
696 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
697 | iters_node) | |
698 | if (it->task_pos == &task->cg_list) | |
699 | css_task_iter_advance(it); | |
700 | ||
701 | list_del_init(&task->cg_list); | |
702 | if (!css_set_populated(from_cset)) | |
703 | css_set_update_populated(from_cset, false); | |
704 | } else { | |
705 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
706 | } | |
707 | ||
708 | if (to_cset) { | |
709 | /* | |
710 | * We are synchronized through cgroup_threadgroup_rwsem | |
711 | * against PF_EXITING setting such that we can't race | |
712 | * against cgroup_exit() changing the css_set to | |
713 | * init_css_set and dropping the old one. | |
714 | */ | |
715 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
716 | ||
717 | rcu_assign_pointer(task->cgroups, to_cset); | |
718 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
719 | &to_cset->tasks); | |
720 | } | |
721 | } | |
722 | ||
723 | /* | |
724 | * hash table for cgroup groups. This improves the performance to find | |
725 | * an existing css_set. This hash doesn't (currently) take into | |
726 | * account cgroups in empty hierarchies. | |
727 | */ | |
728 | #define CSS_SET_HASH_BITS 7 | |
729 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); | |
730 | ||
731 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) | |
732 | { | |
733 | unsigned long key = 0UL; | |
734 | struct cgroup_subsys *ss; | |
735 | int i; | |
736 | ||
737 | for_each_subsys(ss, i) | |
738 | key += (unsigned long)css[i]; | |
739 | key = (key >> 16) ^ key; | |
740 | ||
741 | return key; | |
742 | } | |
743 | ||
744 | static void put_css_set_locked(struct css_set *cset) | |
745 | { | |
746 | struct cgrp_cset_link *link, *tmp_link; | |
747 | struct cgroup_subsys *ss; | |
748 | int ssid; | |
749 | ||
750 | lockdep_assert_held(&css_set_lock); | |
751 | ||
752 | if (!atomic_dec_and_test(&cset->refcount)) | |
753 | return; | |
754 | ||
755 | /* This css_set is dead. unlink it and release cgroup and css refs */ | |
756 | for_each_subsys(ss, ssid) { | |
757 | list_del(&cset->e_cset_node[ssid]); | |
758 | css_put(cset->subsys[ssid]); | |
759 | } | |
760 | hash_del(&cset->hlist); | |
761 | css_set_count--; | |
762 | ||
763 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { | |
764 | list_del(&link->cset_link); | |
765 | list_del(&link->cgrp_link); | |
766 | if (cgroup_parent(link->cgrp)) | |
767 | cgroup_put(link->cgrp); | |
768 | kfree(link); | |
769 | } | |
770 | ||
771 | kfree_rcu(cset, rcu_head); | |
772 | } | |
773 | ||
774 | static void put_css_set(struct css_set *cset) | |
775 | { | |
776 | /* | |
777 | * Ensure that the refcount doesn't hit zero while any readers | |
778 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
779 | * rwlock | |
780 | */ | |
781 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
782 | return; | |
783 | ||
784 | spin_lock_bh(&css_set_lock); | |
785 | put_css_set_locked(cset); | |
786 | spin_unlock_bh(&css_set_lock); | |
787 | } | |
788 | ||
789 | /* | |
790 | * refcounted get/put for css_set objects | |
791 | */ | |
792 | static inline void get_css_set(struct css_set *cset) | |
793 | { | |
794 | atomic_inc(&cset->refcount); | |
795 | } | |
796 | ||
797 | /** | |
798 | * compare_css_sets - helper function for find_existing_css_set(). | |
799 | * @cset: candidate css_set being tested | |
800 | * @old_cset: existing css_set for a task | |
801 | * @new_cgrp: cgroup that's being entered by the task | |
802 | * @template: desired set of css pointers in css_set (pre-calculated) | |
803 | * | |
804 | * Returns true if "cset" matches "old_cset" except for the hierarchy | |
805 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
806 | */ | |
807 | static bool compare_css_sets(struct css_set *cset, | |
808 | struct css_set *old_cset, | |
809 | struct cgroup *new_cgrp, | |
810 | struct cgroup_subsys_state *template[]) | |
811 | { | |
812 | struct list_head *l1, *l2; | |
813 | ||
814 | /* | |
815 | * On the default hierarchy, there can be csets which are | |
816 | * associated with the same set of cgroups but different csses. | |
817 | * Let's first ensure that csses match. | |
818 | */ | |
819 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
820 | return false; | |
821 | ||
822 | /* | |
823 | * Compare cgroup pointers in order to distinguish between | |
824 | * different cgroups in hierarchies. As different cgroups may | |
825 | * share the same effective css, this comparison is always | |
826 | * necessary. | |
827 | */ | |
828 | l1 = &cset->cgrp_links; | |
829 | l2 = &old_cset->cgrp_links; | |
830 | while (1) { | |
831 | struct cgrp_cset_link *link1, *link2; | |
832 | struct cgroup *cgrp1, *cgrp2; | |
833 | ||
834 | l1 = l1->next; | |
835 | l2 = l2->next; | |
836 | /* See if we reached the end - both lists are equal length. */ | |
837 | if (l1 == &cset->cgrp_links) { | |
838 | BUG_ON(l2 != &old_cset->cgrp_links); | |
839 | break; | |
840 | } else { | |
841 | BUG_ON(l2 == &old_cset->cgrp_links); | |
842 | } | |
843 | /* Locate the cgroups associated with these links. */ | |
844 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); | |
845 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
846 | cgrp1 = link1->cgrp; | |
847 | cgrp2 = link2->cgrp; | |
848 | /* Hierarchies should be linked in the same order. */ | |
849 | BUG_ON(cgrp1->root != cgrp2->root); | |
850 | ||
851 | /* | |
852 | * If this hierarchy is the hierarchy of the cgroup | |
853 | * that's changing, then we need to check that this | |
854 | * css_set points to the new cgroup; if it's any other | |
855 | * hierarchy, then this css_set should point to the | |
856 | * same cgroup as the old css_set. | |
857 | */ | |
858 | if (cgrp1->root == new_cgrp->root) { | |
859 | if (cgrp1 != new_cgrp) | |
860 | return false; | |
861 | } else { | |
862 | if (cgrp1 != cgrp2) | |
863 | return false; | |
864 | } | |
865 | } | |
866 | return true; | |
867 | } | |
868 | ||
869 | /** | |
870 | * find_existing_css_set - init css array and find the matching css_set | |
871 | * @old_cset: the css_set that we're using before the cgroup transition | |
872 | * @cgrp: the cgroup that we're moving into | |
873 | * @template: out param for the new set of csses, should be clear on entry | |
874 | */ | |
875 | static struct css_set *find_existing_css_set(struct css_set *old_cset, | |
876 | struct cgroup *cgrp, | |
877 | struct cgroup_subsys_state *template[]) | |
878 | { | |
879 | struct cgroup_root *root = cgrp->root; | |
880 | struct cgroup_subsys *ss; | |
881 | struct css_set *cset; | |
882 | unsigned long key; | |
883 | int i; | |
884 | ||
885 | /* | |
886 | * Build the set of subsystem state objects that we want to see in the | |
887 | * new css_set. while subsystems can change globally, the entries here | |
888 | * won't change, so no need for locking. | |
889 | */ | |
890 | for_each_subsys(ss, i) { | |
891 | if (root->subsys_mask & (1UL << i)) { | |
892 | /* | |
893 | * @ss is in this hierarchy, so we want the | |
894 | * effective css from @cgrp. | |
895 | */ | |
896 | template[i] = cgroup_e_css(cgrp, ss); | |
897 | } else { | |
898 | /* | |
899 | * @ss is not in this hierarchy, so we don't want | |
900 | * to change the css. | |
901 | */ | |
902 | template[i] = old_cset->subsys[i]; | |
903 | } | |
904 | } | |
905 | ||
906 | key = css_set_hash(template); | |
907 | hash_for_each_possible(css_set_table, cset, hlist, key) { | |
908 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
909 | continue; | |
910 | ||
911 | /* This css_set matches what we need */ | |
912 | return cset; | |
913 | } | |
914 | ||
915 | /* No existing cgroup group matched */ | |
916 | return NULL; | |
917 | } | |
918 | ||
919 | static void free_cgrp_cset_links(struct list_head *links_to_free) | |
920 | { | |
921 | struct cgrp_cset_link *link, *tmp_link; | |
922 | ||
923 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { | |
924 | list_del(&link->cset_link); | |
925 | kfree(link); | |
926 | } | |
927 | } | |
928 | ||
929 | /** | |
930 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
931 | * @count: the number of links to allocate | |
932 | * @tmp_links: list_head the allocated links are put on | |
933 | * | |
934 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
935 | * through ->cset_link. Returns 0 on success or -errno. | |
936 | */ | |
937 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) | |
938 | { | |
939 | struct cgrp_cset_link *link; | |
940 | int i; | |
941 | ||
942 | INIT_LIST_HEAD(tmp_links); | |
943 | ||
944 | for (i = 0; i < count; i++) { | |
945 | link = kzalloc(sizeof(*link), GFP_KERNEL); | |
946 | if (!link) { | |
947 | free_cgrp_cset_links(tmp_links); | |
948 | return -ENOMEM; | |
949 | } | |
950 | list_add(&link->cset_link, tmp_links); | |
951 | } | |
952 | return 0; | |
953 | } | |
954 | ||
955 | /** | |
956 | * link_css_set - a helper function to link a css_set to a cgroup | |
957 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() | |
958 | * @cset: the css_set to be linked | |
959 | * @cgrp: the destination cgroup | |
960 | */ | |
961 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, | |
962 | struct cgroup *cgrp) | |
963 | { | |
964 | struct cgrp_cset_link *link; | |
965 | ||
966 | BUG_ON(list_empty(tmp_links)); | |
967 | ||
968 | if (cgroup_on_dfl(cgrp)) | |
969 | cset->dfl_cgrp = cgrp; | |
970 | ||
971 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
972 | link->cset = cset; | |
973 | link->cgrp = cgrp; | |
974 | ||
975 | /* | |
976 | * Always add links to the tail of the lists so that the lists are | |
977 | * in choronological order. | |
978 | */ | |
979 | list_move_tail(&link->cset_link, &cgrp->cset_links); | |
980 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); | |
981 | ||
982 | if (cgroup_parent(cgrp)) | |
983 | cgroup_get(cgrp); | |
984 | } | |
985 | ||
986 | /** | |
987 | * find_css_set - return a new css_set with one cgroup updated | |
988 | * @old_cset: the baseline css_set | |
989 | * @cgrp: the cgroup to be updated | |
990 | * | |
991 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
992 | * substituted into the appropriate hierarchy. | |
993 | */ | |
994 | static struct css_set *find_css_set(struct css_set *old_cset, | |
995 | struct cgroup *cgrp) | |
996 | { | |
997 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; | |
998 | struct css_set *cset; | |
999 | struct list_head tmp_links; | |
1000 | struct cgrp_cset_link *link; | |
1001 | struct cgroup_subsys *ss; | |
1002 | unsigned long key; | |
1003 | int ssid; | |
1004 | ||
1005 | lockdep_assert_held(&cgroup_mutex); | |
1006 | ||
1007 | /* First see if we already have a cgroup group that matches | |
1008 | * the desired set */ | |
1009 | spin_lock_bh(&css_set_lock); | |
1010 | cset = find_existing_css_set(old_cset, cgrp, template); | |
1011 | if (cset) | |
1012 | get_css_set(cset); | |
1013 | spin_unlock_bh(&css_set_lock); | |
1014 | ||
1015 | if (cset) | |
1016 | return cset; | |
1017 | ||
1018 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); | |
1019 | if (!cset) | |
1020 | return NULL; | |
1021 | ||
1022 | /* Allocate all the cgrp_cset_link objects that we'll need */ | |
1023 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { | |
1024 | kfree(cset); | |
1025 | return NULL; | |
1026 | } | |
1027 | ||
1028 | atomic_set(&cset->refcount, 1); | |
1029 | INIT_LIST_HEAD(&cset->cgrp_links); | |
1030 | INIT_LIST_HEAD(&cset->tasks); | |
1031 | INIT_LIST_HEAD(&cset->mg_tasks); | |
1032 | INIT_LIST_HEAD(&cset->mg_preload_node); | |
1033 | INIT_LIST_HEAD(&cset->mg_node); | |
1034 | INIT_LIST_HEAD(&cset->task_iters); | |
1035 | INIT_HLIST_NODE(&cset->hlist); | |
1036 | ||
1037 | /* Copy the set of subsystem state objects generated in | |
1038 | * find_existing_css_set() */ | |
1039 | memcpy(cset->subsys, template, sizeof(cset->subsys)); | |
1040 | ||
1041 | spin_lock_bh(&css_set_lock); | |
1042 | /* Add reference counts and links from the new css_set. */ | |
1043 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { | |
1044 | struct cgroup *c = link->cgrp; | |
1045 | ||
1046 | if (c->root == cgrp->root) | |
1047 | c = cgrp; | |
1048 | link_css_set(&tmp_links, cset, c); | |
1049 | } | |
1050 | ||
1051 | BUG_ON(!list_empty(&tmp_links)); | |
1052 | ||
1053 | css_set_count++; | |
1054 | ||
1055 | /* Add @cset to the hash table */ | |
1056 | key = css_set_hash(cset->subsys); | |
1057 | hash_add(css_set_table, &cset->hlist, key); | |
1058 | ||
1059 | for_each_subsys(ss, ssid) { | |
1060 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1061 | ||
1062 | list_add_tail(&cset->e_cset_node[ssid], | |
1063 | &css->cgroup->e_csets[ssid]); | |
1064 | css_get(css); | |
1065 | } | |
1066 | ||
1067 | spin_unlock_bh(&css_set_lock); | |
1068 | ||
1069 | return cset; | |
1070 | } | |
1071 | ||
1072 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) | |
1073 | { | |
1074 | struct cgroup *root_cgrp = kf_root->kn->priv; | |
1075 | ||
1076 | return root_cgrp->root; | |
1077 | } | |
1078 | ||
1079 | static int cgroup_init_root_id(struct cgroup_root *root) | |
1080 | { | |
1081 | int id; | |
1082 | ||
1083 | lockdep_assert_held(&cgroup_mutex); | |
1084 | ||
1085 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); | |
1086 | if (id < 0) | |
1087 | return id; | |
1088 | ||
1089 | root->hierarchy_id = id; | |
1090 | return 0; | |
1091 | } | |
1092 | ||
1093 | static void cgroup_exit_root_id(struct cgroup_root *root) | |
1094 | { | |
1095 | lockdep_assert_held(&cgroup_mutex); | |
1096 | ||
1097 | if (root->hierarchy_id) { | |
1098 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1099 | root->hierarchy_id = 0; | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | static void cgroup_free_root(struct cgroup_root *root) | |
1104 | { | |
1105 | if (root) { | |
1106 | /* hierarchy ID should already have been released */ | |
1107 | WARN_ON_ONCE(root->hierarchy_id); | |
1108 | ||
1109 | idr_destroy(&root->cgroup_idr); | |
1110 | kfree(root); | |
1111 | } | |
1112 | } | |
1113 | ||
1114 | static void cgroup_destroy_root(struct cgroup_root *root) | |
1115 | { | |
1116 | struct cgroup *cgrp = &root->cgrp; | |
1117 | struct cgrp_cset_link *link, *tmp_link; | |
1118 | ||
1119 | mutex_lock(&cgroup_mutex); | |
1120 | ||
1121 | BUG_ON(atomic_read(&root->nr_cgrps)); | |
1122 | BUG_ON(!list_empty(&cgrp->self.children)); | |
1123 | ||
1124 | /* Rebind all subsystems back to the default hierarchy */ | |
1125 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); | |
1126 | ||
1127 | /* | |
1128 | * Release all the links from cset_links to this hierarchy's | |
1129 | * root cgroup | |
1130 | */ | |
1131 | spin_lock_bh(&css_set_lock); | |
1132 | ||
1133 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1134 | list_del(&link->cset_link); | |
1135 | list_del(&link->cgrp_link); | |
1136 | kfree(link); | |
1137 | } | |
1138 | ||
1139 | spin_unlock_bh(&css_set_lock); | |
1140 | ||
1141 | if (!list_empty(&root->root_list)) { | |
1142 | list_del(&root->root_list); | |
1143 | cgroup_root_count--; | |
1144 | } | |
1145 | ||
1146 | cgroup_exit_root_id(root); | |
1147 | ||
1148 | mutex_unlock(&cgroup_mutex); | |
1149 | ||
1150 | kernfs_destroy_root(root->kf_root); | |
1151 | cgroup_free_root(root); | |
1152 | } | |
1153 | ||
1154 | /* look up cgroup associated with given css_set on the specified hierarchy */ | |
1155 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
1156 | struct cgroup_root *root) | |
1157 | { | |
1158 | struct cgroup *res = NULL; | |
1159 | ||
1160 | lockdep_assert_held(&cgroup_mutex); | |
1161 | lockdep_assert_held(&css_set_lock); | |
1162 | ||
1163 | if (cset == &init_css_set) { | |
1164 | res = &root->cgrp; | |
1165 | } else { | |
1166 | struct cgrp_cset_link *link; | |
1167 | ||
1168 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1169 | struct cgroup *c = link->cgrp; | |
1170 | ||
1171 | if (c->root == root) { | |
1172 | res = c; | |
1173 | break; | |
1174 | } | |
1175 | } | |
1176 | } | |
1177 | ||
1178 | BUG_ON(!res); | |
1179 | return res; | |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * Return the cgroup for "task" from the given hierarchy. Must be | |
1184 | * called with cgroup_mutex and css_set_lock held. | |
1185 | */ | |
1186 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
1187 | struct cgroup_root *root) | |
1188 | { | |
1189 | /* | |
1190 | * No need to lock the task - since we hold cgroup_mutex the | |
1191 | * task can't change groups, so the only thing that can happen | |
1192 | * is that it exits and its css is set back to init_css_set. | |
1193 | */ | |
1194 | return cset_cgroup_from_root(task_css_set(task), root); | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * A task must hold cgroup_mutex to modify cgroups. | |
1199 | * | |
1200 | * Any task can increment and decrement the count field without lock. | |
1201 | * So in general, code holding cgroup_mutex can't rely on the count | |
1202 | * field not changing. However, if the count goes to zero, then only | |
1203 | * cgroup_attach_task() can increment it again. Because a count of zero | |
1204 | * means that no tasks are currently attached, therefore there is no | |
1205 | * way a task attached to that cgroup can fork (the other way to | |
1206 | * increment the count). So code holding cgroup_mutex can safely | |
1207 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1208 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1209 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1210 | * needs that mutex. | |
1211 | * | |
1212 | * A cgroup can only be deleted if both its 'count' of using tasks | |
1213 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1214 | * tasks in the system use _some_ cgroup, and since there is always at | |
1215 | * least one task in the system (init, pid == 1), therefore, root cgroup | |
1216 | * always has either children cgroups and/or using tasks. So we don't | |
1217 | * need a special hack to ensure that root cgroup cannot be deleted. | |
1218 | * | |
1219 | * P.S. One more locking exception. RCU is used to guard the | |
1220 | * update of a tasks cgroup pointer by cgroup_attach_task() | |
1221 | */ | |
1222 | ||
1223 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; | |
1224 | static const struct file_operations proc_cgroupstats_operations; | |
1225 | ||
1226 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, | |
1227 | char *buf) | |
1228 | { | |
1229 | struct cgroup_subsys *ss = cft->ss; | |
1230 | ||
1231 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
1232 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1233 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
1234 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, | |
1235 | cft->name); | |
1236 | else | |
1237 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1238 | return buf; | |
1239 | } | |
1240 | ||
1241 | /** | |
1242 | * cgroup_file_mode - deduce file mode of a control file | |
1243 | * @cft: the control file in question | |
1244 | * | |
1245 | * S_IRUGO for read, S_IWUSR for write. | |
1246 | */ | |
1247 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
1248 | { | |
1249 | umode_t mode = 0; | |
1250 | ||
1251 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
1252 | mode |= S_IRUGO; | |
1253 | ||
1254 | if (cft->write_u64 || cft->write_s64 || cft->write) { | |
1255 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1256 | mode |= S_IWUGO; | |
1257 | else | |
1258 | mode |= S_IWUSR; | |
1259 | } | |
1260 | ||
1261 | return mode; | |
1262 | } | |
1263 | ||
1264 | /** | |
1265 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask | |
1266 | * @cgrp: the target cgroup | |
1267 | * @subtree_control: the new subtree_control mask to consider | |
1268 | * | |
1269 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1270 | * enabled together through its ->depends_on mask. In such cases, more | |
1271 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1272 | * | |
1273 | * This function calculates which subsystems need to be enabled if | |
1274 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1275 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
1276 | */ | |
1277 | static u16 cgroup_calc_subtree_ss_mask(struct cgroup *cgrp, u16 subtree_control) | |
1278 | { | |
1279 | struct cgroup *parent = cgroup_parent(cgrp); | |
1280 | u16 cur_ss_mask = subtree_control; | |
1281 | struct cgroup_subsys *ss; | |
1282 | int ssid; | |
1283 | ||
1284 | lockdep_assert_held(&cgroup_mutex); | |
1285 | ||
1286 | if (!cgroup_on_dfl(cgrp)) | |
1287 | return cur_ss_mask; | |
1288 | ||
1289 | while (true) { | |
1290 | u16 new_ss_mask = cur_ss_mask; | |
1291 | ||
1292 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { | |
1293 | new_ss_mask |= ss->depends_on; | |
1294 | } while_each_subsys_mask(); | |
1295 | ||
1296 | /* | |
1297 | * Mask out subsystems which aren't available. This can | |
1298 | * happen only if some depended-upon subsystems were bound | |
1299 | * to non-default hierarchies. | |
1300 | */ | |
1301 | if (parent) | |
1302 | new_ss_mask &= parent->subtree_ss_mask; | |
1303 | else | |
1304 | new_ss_mask &= cgrp->root->subsys_mask; | |
1305 | ||
1306 | if (new_ss_mask == cur_ss_mask) | |
1307 | break; | |
1308 | cur_ss_mask = new_ss_mask; | |
1309 | } | |
1310 | ||
1311 | return cur_ss_mask; | |
1312 | } | |
1313 | ||
1314 | /** | |
1315 | * cgroup_refresh_subtree_ss_mask - update subtree_ss_mask | |
1316 | * @cgrp: the target cgroup | |
1317 | * | |
1318 | * Update @cgrp->subtree_ss_mask according to the current | |
1319 | * @cgrp->subtree_control using cgroup_calc_subtree_ss_mask(). | |
1320 | */ | |
1321 | static void cgroup_refresh_subtree_ss_mask(struct cgroup *cgrp) | |
1322 | { | |
1323 | cgrp->subtree_ss_mask = | |
1324 | cgroup_calc_subtree_ss_mask(cgrp, cgrp->subtree_control); | |
1325 | } | |
1326 | ||
1327 | /** | |
1328 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1329 | * @kn: the kernfs_node being serviced | |
1330 | * | |
1331 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1332 | * the method finishes if locking succeeded. Note that once this function | |
1333 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1334 | * inaccessible any time. If the caller intends to continue to access the | |
1335 | * cgroup, it should pin it before invoking this function. | |
1336 | */ | |
1337 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
1338 | { | |
1339 | struct cgroup *cgrp; | |
1340 | ||
1341 | if (kernfs_type(kn) == KERNFS_DIR) | |
1342 | cgrp = kn->priv; | |
1343 | else | |
1344 | cgrp = kn->parent->priv; | |
1345 | ||
1346 | mutex_unlock(&cgroup_mutex); | |
1347 | ||
1348 | kernfs_unbreak_active_protection(kn); | |
1349 | cgroup_put(cgrp); | |
1350 | } | |
1351 | ||
1352 | /** | |
1353 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1354 | * @kn: the kernfs_node being serviced | |
1355 | * | |
1356 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1357 | * @kn. It breaks the active protection, performs cgroup locking and | |
1358 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1359 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1360 | * matching cgroup_kn_unlock() invocation. | |
1361 | * | |
1362 | * Any cgroup kernfs method implementation which requires locking the | |
1363 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1364 | * locking under kernfs active protection and allows all kernfs operations | |
1365 | * including self-removal. | |
1366 | */ | |
1367 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
1368 | { | |
1369 | struct cgroup *cgrp; | |
1370 | ||
1371 | if (kernfs_type(kn) == KERNFS_DIR) | |
1372 | cgrp = kn->priv; | |
1373 | else | |
1374 | cgrp = kn->parent->priv; | |
1375 | ||
1376 | /* | |
1377 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
1378 | * active_ref. cgroup liveliness check alone provides enough | |
1379 | * protection against removal. Ensure @cgrp stays accessible and | |
1380 | * break the active_ref protection. | |
1381 | */ | |
1382 | if (!cgroup_tryget(cgrp)) | |
1383 | return NULL; | |
1384 | kernfs_break_active_protection(kn); | |
1385 | ||
1386 | mutex_lock(&cgroup_mutex); | |
1387 | ||
1388 | if (!cgroup_is_dead(cgrp)) | |
1389 | return cgrp; | |
1390 | ||
1391 | cgroup_kn_unlock(kn); | |
1392 | return NULL; | |
1393 | } | |
1394 | ||
1395 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) | |
1396 | { | |
1397 | char name[CGROUP_FILE_NAME_MAX]; | |
1398 | ||
1399 | lockdep_assert_held(&cgroup_mutex); | |
1400 | ||
1401 | if (cft->file_offset) { | |
1402 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1403 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1404 | ||
1405 | spin_lock_irq(&cgroup_file_kn_lock); | |
1406 | cfile->kn = NULL; | |
1407 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1408 | } | |
1409 | ||
1410 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); | |
1411 | } | |
1412 | ||
1413 | /** | |
1414 | * css_clear_dir - remove subsys files in a cgroup directory | |
1415 | * @css: taget css | |
1416 | * @cgrp_override: specify if target cgroup is different from css->cgroup | |
1417 | */ | |
1418 | static void css_clear_dir(struct cgroup_subsys_state *css, | |
1419 | struct cgroup *cgrp_override) | |
1420 | { | |
1421 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; | |
1422 | struct cftype *cfts; | |
1423 | ||
1424 | if (!(css->flags & CSS_VISIBLE)) | |
1425 | return; | |
1426 | ||
1427 | css->flags &= ~CSS_VISIBLE; | |
1428 | ||
1429 | list_for_each_entry(cfts, &css->ss->cfts, node) | |
1430 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1431 | } | |
1432 | ||
1433 | /** | |
1434 | * css_populate_dir - create subsys files in a cgroup directory | |
1435 | * @css: target css | |
1436 | * @cgrp_overried: specify if target cgroup is different from css->cgroup | |
1437 | * | |
1438 | * On failure, no file is added. | |
1439 | */ | |
1440 | static int css_populate_dir(struct cgroup_subsys_state *css, | |
1441 | struct cgroup *cgrp_override) | |
1442 | { | |
1443 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; | |
1444 | struct cftype *cfts, *failed_cfts; | |
1445 | int ret; | |
1446 | ||
1447 | if (css->flags & CSS_VISIBLE) | |
1448 | return 0; | |
1449 | ||
1450 | if (!css->ss) { | |
1451 | if (cgroup_on_dfl(cgrp)) | |
1452 | cfts = cgroup_dfl_base_files; | |
1453 | else | |
1454 | cfts = cgroup_legacy_base_files; | |
1455 | ||
1456 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); | |
1457 | } | |
1458 | ||
1459 | list_for_each_entry(cfts, &css->ss->cfts, node) { | |
1460 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1461 | if (ret < 0) { | |
1462 | failed_cfts = cfts; | |
1463 | goto err; | |
1464 | } | |
1465 | } | |
1466 | ||
1467 | css->flags |= CSS_VISIBLE; | |
1468 | ||
1469 | return 0; | |
1470 | err: | |
1471 | list_for_each_entry(cfts, &css->ss->cfts, node) { | |
1472 | if (cfts == failed_cfts) | |
1473 | break; | |
1474 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1475 | } | |
1476 | return ret; | |
1477 | } | |
1478 | ||
1479 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) | |
1480 | { | |
1481 | struct cgroup *dcgrp = &dst_root->cgrp; | |
1482 | struct cgroup_subsys *ss; | |
1483 | u16 tmp_ss_mask; | |
1484 | int ssid, i, ret; | |
1485 | ||
1486 | lockdep_assert_held(&cgroup_mutex); | |
1487 | ||
1488 | do_each_subsys_mask(ss, ssid, ss_mask) { | |
1489 | /* if @ss has non-root csses attached to it, can't move */ | |
1490 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
1491 | return -EBUSY; | |
1492 | ||
1493 | /* can't move between two non-dummy roots either */ | |
1494 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) | |
1495 | return -EBUSY; | |
1496 | } while_each_subsys_mask(); | |
1497 | ||
1498 | /* skip creating root files on dfl_root for inhibited subsystems */ | |
1499 | tmp_ss_mask = ss_mask; | |
1500 | if (dst_root == &cgrp_dfl_root) | |
1501 | tmp_ss_mask &= ~cgrp_dfl_inhibit_ss_mask; | |
1502 | ||
1503 | do_each_subsys_mask(ss, ssid, tmp_ss_mask) { | |
1504 | struct cgroup *scgrp = &ss->root->cgrp; | |
1505 | int tssid; | |
1506 | ||
1507 | ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp); | |
1508 | if (!ret) | |
1509 | continue; | |
1510 | ||
1511 | /* | |
1512 | * Rebinding back to the default root is not allowed to | |
1513 | * fail. Using both default and non-default roots should | |
1514 | * be rare. Moving subsystems back and forth even more so. | |
1515 | * Just warn about it and continue. | |
1516 | */ | |
1517 | if (dst_root == &cgrp_dfl_root) { | |
1518 | if (cgrp_dfl_visible) { | |
1519 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", | |
1520 | ret, ss_mask); | |
1521 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); | |
1522 | } | |
1523 | continue; | |
1524 | } | |
1525 | ||
1526 | do_each_subsys_mask(ss, tssid, tmp_ss_mask) { | |
1527 | if (tssid == ssid) | |
1528 | break; | |
1529 | css_clear_dir(cgroup_css(scgrp, ss), dcgrp); | |
1530 | } while_each_subsys_mask(); | |
1531 | return ret; | |
1532 | } while_each_subsys_mask(); | |
1533 | ||
1534 | /* | |
1535 | * Nothing can fail from this point on. Remove files for the | |
1536 | * removed subsystems and rebind each subsystem. | |
1537 | */ | |
1538 | do_each_subsys_mask(ss, ssid, ss_mask) { | |
1539 | struct cgroup_root *src_root = ss->root; | |
1540 | struct cgroup *scgrp = &src_root->cgrp; | |
1541 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
1542 | struct css_set *cset; | |
1543 | ||
1544 | WARN_ON(!css || cgroup_css(dcgrp, ss)); | |
1545 | ||
1546 | css_clear_dir(css, NULL); | |
1547 | ||
1548 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); | |
1549 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
1550 | ss->root = dst_root; | |
1551 | css->cgroup = dcgrp; | |
1552 | ||
1553 | spin_lock_bh(&css_set_lock); | |
1554 | hash_for_each(css_set_table, i, cset, hlist) | |
1555 | list_move_tail(&cset->e_cset_node[ss->id], | |
1556 | &dcgrp->e_csets[ss->id]); | |
1557 | spin_unlock_bh(&css_set_lock); | |
1558 | ||
1559 | src_root->subsys_mask &= ~(1 << ssid); | |
1560 | scgrp->subtree_control &= ~(1 << ssid); | |
1561 | cgroup_refresh_subtree_ss_mask(scgrp); | |
1562 | ||
1563 | /* default hierarchy doesn't enable controllers by default */ | |
1564 | dst_root->subsys_mask |= 1 << ssid; | |
1565 | if (dst_root == &cgrp_dfl_root) { | |
1566 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1567 | } else { | |
1568 | dcgrp->subtree_control |= 1 << ssid; | |
1569 | cgroup_refresh_subtree_ss_mask(dcgrp); | |
1570 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); | |
1571 | } | |
1572 | ||
1573 | if (ss->bind) | |
1574 | ss->bind(css); | |
1575 | } while_each_subsys_mask(); | |
1576 | ||
1577 | kernfs_activate(dcgrp->kn); | |
1578 | return 0; | |
1579 | } | |
1580 | ||
1581 | static int cgroup_show_options(struct seq_file *seq, | |
1582 | struct kernfs_root *kf_root) | |
1583 | { | |
1584 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
1585 | struct cgroup_subsys *ss; | |
1586 | int ssid; | |
1587 | ||
1588 | if (root != &cgrp_dfl_root) | |
1589 | for_each_subsys(ss, ssid) | |
1590 | if (root->subsys_mask & (1 << ssid)) | |
1591 | seq_show_option(seq, ss->legacy_name, NULL); | |
1592 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
1593 | seq_puts(seq, ",noprefix"); | |
1594 | if (root->flags & CGRP_ROOT_XATTR) | |
1595 | seq_puts(seq, ",xattr"); | |
1596 | ||
1597 | spin_lock(&release_agent_path_lock); | |
1598 | if (strlen(root->release_agent_path)) | |
1599 | seq_show_option(seq, "release_agent", | |
1600 | root->release_agent_path); | |
1601 | spin_unlock(&release_agent_path_lock); | |
1602 | ||
1603 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
1604 | seq_puts(seq, ",clone_children"); | |
1605 | if (strlen(root->name)) | |
1606 | seq_show_option(seq, "name", root->name); | |
1607 | return 0; | |
1608 | } | |
1609 | ||
1610 | struct cgroup_sb_opts { | |
1611 | u16 subsys_mask; | |
1612 | unsigned int flags; | |
1613 | char *release_agent; | |
1614 | bool cpuset_clone_children; | |
1615 | char *name; | |
1616 | /* User explicitly requested empty subsystem */ | |
1617 | bool none; | |
1618 | }; | |
1619 | ||
1620 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) | |
1621 | { | |
1622 | char *token, *o = data; | |
1623 | bool all_ss = false, one_ss = false; | |
1624 | u16 mask = U16_MAX; | |
1625 | struct cgroup_subsys *ss; | |
1626 | int nr_opts = 0; | |
1627 | int i; | |
1628 | ||
1629 | #ifdef CONFIG_CPUSETS | |
1630 | mask = ~((u16)1 << cpuset_cgrp_id); | |
1631 | #endif | |
1632 | ||
1633 | memset(opts, 0, sizeof(*opts)); | |
1634 | ||
1635 | while ((token = strsep(&o, ",")) != NULL) { | |
1636 | nr_opts++; | |
1637 | ||
1638 | if (!*token) | |
1639 | return -EINVAL; | |
1640 | if (!strcmp(token, "none")) { | |
1641 | /* Explicitly have no subsystems */ | |
1642 | opts->none = true; | |
1643 | continue; | |
1644 | } | |
1645 | if (!strcmp(token, "all")) { | |
1646 | /* Mutually exclusive option 'all' + subsystem name */ | |
1647 | if (one_ss) | |
1648 | return -EINVAL; | |
1649 | all_ss = true; | |
1650 | continue; | |
1651 | } | |
1652 | if (!strcmp(token, "noprefix")) { | |
1653 | opts->flags |= CGRP_ROOT_NOPREFIX; | |
1654 | continue; | |
1655 | } | |
1656 | if (!strcmp(token, "clone_children")) { | |
1657 | opts->cpuset_clone_children = true; | |
1658 | continue; | |
1659 | } | |
1660 | if (!strcmp(token, "xattr")) { | |
1661 | opts->flags |= CGRP_ROOT_XATTR; | |
1662 | continue; | |
1663 | } | |
1664 | if (!strncmp(token, "release_agent=", 14)) { | |
1665 | /* Specifying two release agents is forbidden */ | |
1666 | if (opts->release_agent) | |
1667 | return -EINVAL; | |
1668 | opts->release_agent = | |
1669 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); | |
1670 | if (!opts->release_agent) | |
1671 | return -ENOMEM; | |
1672 | continue; | |
1673 | } | |
1674 | if (!strncmp(token, "name=", 5)) { | |
1675 | const char *name = token + 5; | |
1676 | /* Can't specify an empty name */ | |
1677 | if (!strlen(name)) | |
1678 | return -EINVAL; | |
1679 | /* Must match [\w.-]+ */ | |
1680 | for (i = 0; i < strlen(name); i++) { | |
1681 | char c = name[i]; | |
1682 | if (isalnum(c)) | |
1683 | continue; | |
1684 | if ((c == '.') || (c == '-') || (c == '_')) | |
1685 | continue; | |
1686 | return -EINVAL; | |
1687 | } | |
1688 | /* Specifying two names is forbidden */ | |
1689 | if (opts->name) | |
1690 | return -EINVAL; | |
1691 | opts->name = kstrndup(name, | |
1692 | MAX_CGROUP_ROOT_NAMELEN - 1, | |
1693 | GFP_KERNEL); | |
1694 | if (!opts->name) | |
1695 | return -ENOMEM; | |
1696 | ||
1697 | continue; | |
1698 | } | |
1699 | ||
1700 | for_each_subsys(ss, i) { | |
1701 | if (strcmp(token, ss->legacy_name)) | |
1702 | continue; | |
1703 | if (!cgroup_ssid_enabled(i)) | |
1704 | continue; | |
1705 | if (cgroup_ssid_no_v1(i)) | |
1706 | continue; | |
1707 | ||
1708 | /* Mutually exclusive option 'all' + subsystem name */ | |
1709 | if (all_ss) | |
1710 | return -EINVAL; | |
1711 | opts->subsys_mask |= (1 << i); | |
1712 | one_ss = true; | |
1713 | ||
1714 | break; | |
1715 | } | |
1716 | if (i == CGROUP_SUBSYS_COUNT) | |
1717 | return -ENOENT; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * If the 'all' option was specified select all the subsystems, | |
1722 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1723 | * not specified, let's default to 'all' | |
1724 | */ | |
1725 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1726 | for_each_subsys(ss, i) | |
1727 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) | |
1728 | opts->subsys_mask |= (1 << i); | |
1729 | ||
1730 | /* | |
1731 | * We either have to specify by name or by subsystems. (So all | |
1732 | * empty hierarchies must have a name). | |
1733 | */ | |
1734 | if (!opts->subsys_mask && !opts->name) | |
1735 | return -EINVAL; | |
1736 | ||
1737 | /* | |
1738 | * Option noprefix was introduced just for backward compatibility | |
1739 | * with the old cpuset, so we allow noprefix only if mounting just | |
1740 | * the cpuset subsystem. | |
1741 | */ | |
1742 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) | |
1743 | return -EINVAL; | |
1744 | ||
1745 | /* Can't specify "none" and some subsystems */ | |
1746 | if (opts->subsys_mask && opts->none) | |
1747 | return -EINVAL; | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
1752 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) | |
1753 | { | |
1754 | int ret = 0; | |
1755 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
1756 | struct cgroup_sb_opts opts; | |
1757 | u16 added_mask, removed_mask; | |
1758 | ||
1759 | if (root == &cgrp_dfl_root) { | |
1760 | pr_err("remount is not allowed\n"); | |
1761 | return -EINVAL; | |
1762 | } | |
1763 | ||
1764 | mutex_lock(&cgroup_mutex); | |
1765 | ||
1766 | /* See what subsystems are wanted */ | |
1767 | ret = parse_cgroupfs_options(data, &opts); | |
1768 | if (ret) | |
1769 | goto out_unlock; | |
1770 | ||
1771 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) | |
1772 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", | |
1773 | task_tgid_nr(current), current->comm); | |
1774 | ||
1775 | added_mask = opts.subsys_mask & ~root->subsys_mask; | |
1776 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
1777 | ||
1778 | /* Don't allow flags or name to change at remount */ | |
1779 | if ((opts.flags ^ root->flags) || | |
1780 | (opts.name && strcmp(opts.name, root->name))) { | |
1781 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", | |
1782 | opts.flags, opts.name ?: "", root->flags, root->name); | |
1783 | ret = -EINVAL; | |
1784 | goto out_unlock; | |
1785 | } | |
1786 | ||
1787 | /* remounting is not allowed for populated hierarchies */ | |
1788 | if (!list_empty(&root->cgrp.self.children)) { | |
1789 | ret = -EBUSY; | |
1790 | goto out_unlock; | |
1791 | } | |
1792 | ||
1793 | ret = rebind_subsystems(root, added_mask); | |
1794 | if (ret) | |
1795 | goto out_unlock; | |
1796 | ||
1797 | rebind_subsystems(&cgrp_dfl_root, removed_mask); | |
1798 | ||
1799 | if (opts.release_agent) { | |
1800 | spin_lock(&release_agent_path_lock); | |
1801 | strcpy(root->release_agent_path, opts.release_agent); | |
1802 | spin_unlock(&release_agent_path_lock); | |
1803 | } | |
1804 | out_unlock: | |
1805 | kfree(opts.release_agent); | |
1806 | kfree(opts.name); | |
1807 | mutex_unlock(&cgroup_mutex); | |
1808 | return ret; | |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * To reduce the fork() overhead for systems that are not actually using | |
1813 | * their cgroups capability, we don't maintain the lists running through | |
1814 | * each css_set to its tasks until we see the list actually used - in other | |
1815 | * words after the first mount. | |
1816 | */ | |
1817 | static bool use_task_css_set_links __read_mostly; | |
1818 | ||
1819 | static void cgroup_enable_task_cg_lists(void) | |
1820 | { | |
1821 | struct task_struct *p, *g; | |
1822 | ||
1823 | spin_lock_bh(&css_set_lock); | |
1824 | ||
1825 | if (use_task_css_set_links) | |
1826 | goto out_unlock; | |
1827 | ||
1828 | use_task_css_set_links = true; | |
1829 | ||
1830 | /* | |
1831 | * We need tasklist_lock because RCU is not safe against | |
1832 | * while_each_thread(). Besides, a forking task that has passed | |
1833 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1834 | * is not guaranteed to have its child immediately visible in the | |
1835 | * tasklist if we walk through it with RCU. | |
1836 | */ | |
1837 | read_lock(&tasklist_lock); | |
1838 | do_each_thread(g, p) { | |
1839 | WARN_ON_ONCE(!list_empty(&p->cg_list) || | |
1840 | task_css_set(p) != &init_css_set); | |
1841 | ||
1842 | /* | |
1843 | * We should check if the process is exiting, otherwise | |
1844 | * it will race with cgroup_exit() in that the list | |
1845 | * entry won't be deleted though the process has exited. | |
1846 | * Do it while holding siglock so that we don't end up | |
1847 | * racing against cgroup_exit(). | |
1848 | */ | |
1849 | spin_lock_irq(&p->sighand->siglock); | |
1850 | if (!(p->flags & PF_EXITING)) { | |
1851 | struct css_set *cset = task_css_set(p); | |
1852 | ||
1853 | if (!css_set_populated(cset)) | |
1854 | css_set_update_populated(cset, true); | |
1855 | list_add_tail(&p->cg_list, &cset->tasks); | |
1856 | get_css_set(cset); | |
1857 | } | |
1858 | spin_unlock_irq(&p->sighand->siglock); | |
1859 | } while_each_thread(g, p); | |
1860 | read_unlock(&tasklist_lock); | |
1861 | out_unlock: | |
1862 | spin_unlock_bh(&css_set_lock); | |
1863 | } | |
1864 | ||
1865 | static void init_cgroup_housekeeping(struct cgroup *cgrp) | |
1866 | { | |
1867 | struct cgroup_subsys *ss; | |
1868 | int ssid; | |
1869 | ||
1870 | INIT_LIST_HEAD(&cgrp->self.sibling); | |
1871 | INIT_LIST_HEAD(&cgrp->self.children); | |
1872 | INIT_LIST_HEAD(&cgrp->cset_links); | |
1873 | INIT_LIST_HEAD(&cgrp->pidlists); | |
1874 | mutex_init(&cgrp->pidlist_mutex); | |
1875 | cgrp->self.cgroup = cgrp; | |
1876 | cgrp->self.flags |= CSS_ONLINE; | |
1877 | ||
1878 | for_each_subsys(ss, ssid) | |
1879 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
1880 | ||
1881 | init_waitqueue_head(&cgrp->offline_waitq); | |
1882 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); | |
1883 | } | |
1884 | ||
1885 | static void init_cgroup_root(struct cgroup_root *root, | |
1886 | struct cgroup_sb_opts *opts) | |
1887 | { | |
1888 | struct cgroup *cgrp = &root->cgrp; | |
1889 | ||
1890 | INIT_LIST_HEAD(&root->root_list); | |
1891 | atomic_set(&root->nr_cgrps, 1); | |
1892 | cgrp->root = root; | |
1893 | init_cgroup_housekeeping(cgrp); | |
1894 | idr_init(&root->cgroup_idr); | |
1895 | ||
1896 | root->flags = opts->flags; | |
1897 | if (opts->release_agent) | |
1898 | strcpy(root->release_agent_path, opts->release_agent); | |
1899 | if (opts->name) | |
1900 | strcpy(root->name, opts->name); | |
1901 | if (opts->cpuset_clone_children) | |
1902 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); | |
1903 | } | |
1904 | ||
1905 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) | |
1906 | { | |
1907 | LIST_HEAD(tmp_links); | |
1908 | struct cgroup *root_cgrp = &root->cgrp; | |
1909 | struct css_set *cset; | |
1910 | int i, ret; | |
1911 | ||
1912 | lockdep_assert_held(&cgroup_mutex); | |
1913 | ||
1914 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); | |
1915 | if (ret < 0) | |
1916 | goto out; | |
1917 | root_cgrp->id = ret; | |
1918 | root_cgrp->ancestor_ids[0] = ret; | |
1919 | ||
1920 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, | |
1921 | GFP_KERNEL); | |
1922 | if (ret) | |
1923 | goto out; | |
1924 | ||
1925 | /* | |
1926 | * We're accessing css_set_count without locking css_set_lock here, | |
1927 | * but that's OK - it can only be increased by someone holding | |
1928 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1929 | * have some link structures left over | |
1930 | */ | |
1931 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1932 | if (ret) | |
1933 | goto cancel_ref; | |
1934 | ||
1935 | ret = cgroup_init_root_id(root); | |
1936 | if (ret) | |
1937 | goto cancel_ref; | |
1938 | ||
1939 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, | |
1940 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1941 | root_cgrp); | |
1942 | if (IS_ERR(root->kf_root)) { | |
1943 | ret = PTR_ERR(root->kf_root); | |
1944 | goto exit_root_id; | |
1945 | } | |
1946 | root_cgrp->kn = root->kf_root->kn; | |
1947 | ||
1948 | ret = css_populate_dir(&root_cgrp->self, NULL); | |
1949 | if (ret) | |
1950 | goto destroy_root; | |
1951 | ||
1952 | ret = rebind_subsystems(root, ss_mask); | |
1953 | if (ret) | |
1954 | goto destroy_root; | |
1955 | ||
1956 | /* | |
1957 | * There must be no failure case after here, since rebinding takes | |
1958 | * care of subsystems' refcounts, which are explicitly dropped in | |
1959 | * the failure exit path. | |
1960 | */ | |
1961 | list_add(&root->root_list, &cgroup_roots); | |
1962 | cgroup_root_count++; | |
1963 | ||
1964 | /* | |
1965 | * Link the root cgroup in this hierarchy into all the css_set | |
1966 | * objects. | |
1967 | */ | |
1968 | spin_lock_bh(&css_set_lock); | |
1969 | hash_for_each(css_set_table, i, cset, hlist) { | |
1970 | link_css_set(&tmp_links, cset, root_cgrp); | |
1971 | if (css_set_populated(cset)) | |
1972 | cgroup_update_populated(root_cgrp, true); | |
1973 | } | |
1974 | spin_unlock_bh(&css_set_lock); | |
1975 | ||
1976 | BUG_ON(!list_empty(&root_cgrp->self.children)); | |
1977 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); | |
1978 | ||
1979 | kernfs_activate(root_cgrp->kn); | |
1980 | ret = 0; | |
1981 | goto out; | |
1982 | ||
1983 | destroy_root: | |
1984 | kernfs_destroy_root(root->kf_root); | |
1985 | root->kf_root = NULL; | |
1986 | exit_root_id: | |
1987 | cgroup_exit_root_id(root); | |
1988 | cancel_ref: | |
1989 | percpu_ref_exit(&root_cgrp->self.refcnt); | |
1990 | out: | |
1991 | free_cgrp_cset_links(&tmp_links); | |
1992 | return ret; | |
1993 | } | |
1994 | ||
1995 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, | |
1996 | int flags, const char *unused_dev_name, | |
1997 | void *data) | |
1998 | { | |
1999 | bool is_v2 = fs_type == &cgroup2_fs_type; | |
2000 | struct super_block *pinned_sb = NULL; | |
2001 | struct cgroup_subsys *ss; | |
2002 | struct cgroup_root *root; | |
2003 | struct cgroup_sb_opts opts; | |
2004 | struct dentry *dentry; | |
2005 | int ret; | |
2006 | int i; | |
2007 | bool new_sb; | |
2008 | ||
2009 | /* | |
2010 | * The first time anyone tries to mount a cgroup, enable the list | |
2011 | * linking each css_set to its tasks and fix up all existing tasks. | |
2012 | */ | |
2013 | if (!use_task_css_set_links) | |
2014 | cgroup_enable_task_cg_lists(); | |
2015 | ||
2016 | if (is_v2) { | |
2017 | if (data) { | |
2018 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
2019 | return ERR_PTR(-EINVAL); | |
2020 | } | |
2021 | cgrp_dfl_visible = true; | |
2022 | root = &cgrp_dfl_root; | |
2023 | cgroup_get(&root->cgrp); | |
2024 | goto out_mount; | |
2025 | } | |
2026 | ||
2027 | mutex_lock(&cgroup_mutex); | |
2028 | ||
2029 | /* First find the desired set of subsystems */ | |
2030 | ret = parse_cgroupfs_options(data, &opts); | |
2031 | if (ret) | |
2032 | goto out_unlock; | |
2033 | ||
2034 | /* | |
2035 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2036 | * still be dying after the previous unmount. Let's drain the | |
2037 | * dying subsystems. We just need to ensure that the ones | |
2038 | * unmounted previously finish dying and don't care about new ones | |
2039 | * starting. Testing ref liveliness is good enough. | |
2040 | */ | |
2041 | for_each_subsys(ss, i) { | |
2042 | if (!(opts.subsys_mask & (1 << i)) || | |
2043 | ss->root == &cgrp_dfl_root) | |
2044 | continue; | |
2045 | ||
2046 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2047 | mutex_unlock(&cgroup_mutex); | |
2048 | msleep(10); | |
2049 | ret = restart_syscall(); | |
2050 | goto out_free; | |
2051 | } | |
2052 | cgroup_put(&ss->root->cgrp); | |
2053 | } | |
2054 | ||
2055 | for_each_root(root) { | |
2056 | bool name_match = false; | |
2057 | ||
2058 | if (root == &cgrp_dfl_root) | |
2059 | continue; | |
2060 | ||
2061 | /* | |
2062 | * If we asked for a name then it must match. Also, if | |
2063 | * name matches but sybsys_mask doesn't, we should fail. | |
2064 | * Remember whether name matched. | |
2065 | */ | |
2066 | if (opts.name) { | |
2067 | if (strcmp(opts.name, root->name)) | |
2068 | continue; | |
2069 | name_match = true; | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * If we asked for subsystems (or explicitly for no | |
2074 | * subsystems) then they must match. | |
2075 | */ | |
2076 | if ((opts.subsys_mask || opts.none) && | |
2077 | (opts.subsys_mask != root->subsys_mask)) { | |
2078 | if (!name_match) | |
2079 | continue; | |
2080 | ret = -EBUSY; | |
2081 | goto out_unlock; | |
2082 | } | |
2083 | ||
2084 | if (root->flags ^ opts.flags) | |
2085 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
2086 | ||
2087 | /* | |
2088 | * We want to reuse @root whose lifetime is governed by its | |
2089 | * ->cgrp. Let's check whether @root is alive and keep it | |
2090 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2091 | * want to block it by pinning the sb so that @root doesn't | |
2092 | * get killed before mount is complete. | |
2093 | * | |
2094 | * With the sb pinned, tryget_live can reliably indicate | |
2095 | * whether @root can be reused. If it's being killed, | |
2096 | * drain it. We can use wait_queue for the wait but this | |
2097 | * path is super cold. Let's just sleep a bit and retry. | |
2098 | */ | |
2099 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); | |
2100 | if (IS_ERR(pinned_sb) || | |
2101 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
2102 | mutex_unlock(&cgroup_mutex); | |
2103 | if (!IS_ERR_OR_NULL(pinned_sb)) | |
2104 | deactivate_super(pinned_sb); | |
2105 | msleep(10); | |
2106 | ret = restart_syscall(); | |
2107 | goto out_free; | |
2108 | } | |
2109 | ||
2110 | ret = 0; | |
2111 | goto out_unlock; | |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * No such thing, create a new one. name= matching without subsys | |
2116 | * specification is allowed for already existing hierarchies but we | |
2117 | * can't create new one without subsys specification. | |
2118 | */ | |
2119 | if (!opts.subsys_mask && !opts.none) { | |
2120 | ret = -EINVAL; | |
2121 | goto out_unlock; | |
2122 | } | |
2123 | ||
2124 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
2125 | if (!root) { | |
2126 | ret = -ENOMEM; | |
2127 | goto out_unlock; | |
2128 | } | |
2129 | ||
2130 | init_cgroup_root(root, &opts); | |
2131 | ||
2132 | ret = cgroup_setup_root(root, opts.subsys_mask); | |
2133 | if (ret) | |
2134 | cgroup_free_root(root); | |
2135 | ||
2136 | out_unlock: | |
2137 | mutex_unlock(&cgroup_mutex); | |
2138 | out_free: | |
2139 | kfree(opts.release_agent); | |
2140 | kfree(opts.name); | |
2141 | ||
2142 | if (ret) | |
2143 | return ERR_PTR(ret); | |
2144 | out_mount: | |
2145 | dentry = kernfs_mount(fs_type, flags, root->kf_root, | |
2146 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, | |
2147 | &new_sb); | |
2148 | if (IS_ERR(dentry) || !new_sb) | |
2149 | cgroup_put(&root->cgrp); | |
2150 | ||
2151 | /* | |
2152 | * If @pinned_sb, we're reusing an existing root and holding an | |
2153 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2154 | */ | |
2155 | if (pinned_sb) { | |
2156 | WARN_ON(new_sb); | |
2157 | deactivate_super(pinned_sb); | |
2158 | } | |
2159 | ||
2160 | return dentry; | |
2161 | } | |
2162 | ||
2163 | static void cgroup_kill_sb(struct super_block *sb) | |
2164 | { | |
2165 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
2166 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
2167 | ||
2168 | /* | |
2169 | * If @root doesn't have any mounts or children, start killing it. | |
2170 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2171 | * cgroup_mount() may wait for @root's release. | |
2172 | * | |
2173 | * And don't kill the default root. | |
2174 | */ | |
2175 | if (!list_empty(&root->cgrp.self.children) || | |
2176 | root == &cgrp_dfl_root) | |
2177 | cgroup_put(&root->cgrp); | |
2178 | else | |
2179 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2180 | ||
2181 | kernfs_kill_sb(sb); | |
2182 | } | |
2183 | ||
2184 | static struct file_system_type cgroup_fs_type = { | |
2185 | .name = "cgroup", | |
2186 | .mount = cgroup_mount, | |
2187 | .kill_sb = cgroup_kill_sb, | |
2188 | }; | |
2189 | ||
2190 | static struct file_system_type cgroup2_fs_type = { | |
2191 | .name = "cgroup2", | |
2192 | .mount = cgroup_mount, | |
2193 | .kill_sb = cgroup_kill_sb, | |
2194 | }; | |
2195 | ||
2196 | /** | |
2197 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy | |
2198 | * @task: target task | |
2199 | * @buf: the buffer to write the path into | |
2200 | * @buflen: the length of the buffer | |
2201 | * | |
2202 | * Determine @task's cgroup on the first (the one with the lowest non-zero | |
2203 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2204 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2205 | * cgroup controller callbacks. | |
2206 | * | |
2207 | * Return value is the same as kernfs_path(). | |
2208 | */ | |
2209 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) | |
2210 | { | |
2211 | struct cgroup_root *root; | |
2212 | struct cgroup *cgrp; | |
2213 | int hierarchy_id = 1; | |
2214 | char *path = NULL; | |
2215 | ||
2216 | mutex_lock(&cgroup_mutex); | |
2217 | spin_lock_bh(&css_set_lock); | |
2218 | ||
2219 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); | |
2220 | ||
2221 | if (root) { | |
2222 | cgrp = task_cgroup_from_root(task, root); | |
2223 | path = cgroup_path(cgrp, buf, buflen); | |
2224 | } else { | |
2225 | /* if no hierarchy exists, everyone is in "/" */ | |
2226 | if (strlcpy(buf, "/", buflen) < buflen) | |
2227 | path = buf; | |
2228 | } | |
2229 | ||
2230 | spin_unlock_bh(&css_set_lock); | |
2231 | mutex_unlock(&cgroup_mutex); | |
2232 | return path; | |
2233 | } | |
2234 | EXPORT_SYMBOL_GPL(task_cgroup_path); | |
2235 | ||
2236 | /* used to track tasks and other necessary states during migration */ | |
2237 | struct cgroup_taskset { | |
2238 | /* the src and dst cset list running through cset->mg_node */ | |
2239 | struct list_head src_csets; | |
2240 | struct list_head dst_csets; | |
2241 | ||
2242 | /* the subsys currently being processed */ | |
2243 | int ssid; | |
2244 | ||
2245 | /* | |
2246 | * Fields for cgroup_taskset_*() iteration. | |
2247 | * | |
2248 | * Before migration is committed, the target migration tasks are on | |
2249 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2250 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2251 | * or ->dst_csets depending on whether migration is committed. | |
2252 | * | |
2253 | * ->cur_csets and ->cur_task point to the current task position | |
2254 | * during iteration. | |
2255 | */ | |
2256 | struct list_head *csets; | |
2257 | struct css_set *cur_cset; | |
2258 | struct task_struct *cur_task; | |
2259 | }; | |
2260 | ||
2261 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ | |
2262 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2263 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2264 | .csets = &tset.src_csets, \ | |
2265 | } | |
2266 | ||
2267 | /** | |
2268 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2269 | * @task: target task | |
2270 | * @tset: target taskset | |
2271 | * | |
2272 | * Add @task, which is a migration target, to @tset. This function becomes | |
2273 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2274 | * been added as a migration source and @task->cg_list will be moved from | |
2275 | * the css_set's tasks list to mg_tasks one. | |
2276 | */ | |
2277 | static void cgroup_taskset_add(struct task_struct *task, | |
2278 | struct cgroup_taskset *tset) | |
2279 | { | |
2280 | struct css_set *cset; | |
2281 | ||
2282 | lockdep_assert_held(&css_set_lock); | |
2283 | ||
2284 | /* @task either already exited or can't exit until the end */ | |
2285 | if (task->flags & PF_EXITING) | |
2286 | return; | |
2287 | ||
2288 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2289 | if (list_empty(&task->cg_list)) | |
2290 | return; | |
2291 | ||
2292 | cset = task_css_set(task); | |
2293 | if (!cset->mg_src_cgrp) | |
2294 | return; | |
2295 | ||
2296 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2297 | if (list_empty(&cset->mg_node)) | |
2298 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2299 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2300 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2301 | &tset->dst_csets); | |
2302 | } | |
2303 | ||
2304 | /** | |
2305 | * cgroup_taskset_first - reset taskset and return the first task | |
2306 | * @tset: taskset of interest | |
2307 | * @dst_cssp: output variable for the destination css | |
2308 | * | |
2309 | * @tset iteration is initialized and the first task is returned. | |
2310 | */ | |
2311 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, | |
2312 | struct cgroup_subsys_state **dst_cssp) | |
2313 | { | |
2314 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); | |
2315 | tset->cur_task = NULL; | |
2316 | ||
2317 | return cgroup_taskset_next(tset, dst_cssp); | |
2318 | } | |
2319 | ||
2320 | /** | |
2321 | * cgroup_taskset_next - iterate to the next task in taskset | |
2322 | * @tset: taskset of interest | |
2323 | * @dst_cssp: output variable for the destination css | |
2324 | * | |
2325 | * Return the next task in @tset. Iteration must have been initialized | |
2326 | * with cgroup_taskset_first(). | |
2327 | */ | |
2328 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, | |
2329 | struct cgroup_subsys_state **dst_cssp) | |
2330 | { | |
2331 | struct css_set *cset = tset->cur_cset; | |
2332 | struct task_struct *task = tset->cur_task; | |
2333 | ||
2334 | while (&cset->mg_node != tset->csets) { | |
2335 | if (!task) | |
2336 | task = list_first_entry(&cset->mg_tasks, | |
2337 | struct task_struct, cg_list); | |
2338 | else | |
2339 | task = list_next_entry(task, cg_list); | |
2340 | ||
2341 | if (&task->cg_list != &cset->mg_tasks) { | |
2342 | tset->cur_cset = cset; | |
2343 | tset->cur_task = task; | |
2344 | ||
2345 | /* | |
2346 | * This function may be called both before and | |
2347 | * after cgroup_taskset_migrate(). The two cases | |
2348 | * can be distinguished by looking at whether @cset | |
2349 | * has its ->mg_dst_cset set. | |
2350 | */ | |
2351 | if (cset->mg_dst_cset) | |
2352 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2353 | else | |
2354 | *dst_cssp = cset->subsys[tset->ssid]; | |
2355 | ||
2356 | return task; | |
2357 | } | |
2358 | ||
2359 | cset = list_next_entry(cset, mg_node); | |
2360 | task = NULL; | |
2361 | } | |
2362 | ||
2363 | return NULL; | |
2364 | } | |
2365 | ||
2366 | /** | |
2367 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2368 | * @tset: taget taskset | |
2369 | * @dst_cgrp: destination cgroup | |
2370 | * | |
2371 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2372 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2373 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2374 | * success. | |
2375 | */ | |
2376 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2377 | struct cgroup *dst_cgrp) | |
2378 | { | |
2379 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2380 | struct task_struct *task, *tmp_task; | |
2381 | struct css_set *cset, *tmp_cset; | |
2382 | int i, ret; | |
2383 | ||
2384 | /* methods shouldn't be called if no task is actually migrating */ | |
2385 | if (list_empty(&tset->src_csets)) | |
2386 | return 0; | |
2387 | ||
2388 | /* check that we can legitimately attach to the cgroup */ | |
2389 | for_each_e_css(css, i, dst_cgrp) { | |
2390 | if (css->ss->can_attach) { | |
2391 | tset->ssid = i; | |
2392 | ret = css->ss->can_attach(tset); | |
2393 | if (ret) { | |
2394 | failed_css = css; | |
2395 | goto out_cancel_attach; | |
2396 | } | |
2397 | } | |
2398 | } | |
2399 | ||
2400 | /* | |
2401 | * Now that we're guaranteed success, proceed to move all tasks to | |
2402 | * the new cgroup. There are no failure cases after here, so this | |
2403 | * is the commit point. | |
2404 | */ | |
2405 | spin_lock_bh(&css_set_lock); | |
2406 | list_for_each_entry(cset, &tset->src_csets, mg_node) { | |
2407 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { | |
2408 | struct css_set *from_cset = task_css_set(task); | |
2409 | struct css_set *to_cset = cset->mg_dst_cset; | |
2410 | ||
2411 | get_css_set(to_cset); | |
2412 | css_set_move_task(task, from_cset, to_cset, true); | |
2413 | put_css_set_locked(from_cset); | |
2414 | } | |
2415 | } | |
2416 | spin_unlock_bh(&css_set_lock); | |
2417 | ||
2418 | /* | |
2419 | * Migration is committed, all target tasks are now on dst_csets. | |
2420 | * Nothing is sensitive to fork() after this point. Notify | |
2421 | * controllers that migration is complete. | |
2422 | */ | |
2423 | tset->csets = &tset->dst_csets; | |
2424 | ||
2425 | for_each_e_css(css, i, dst_cgrp) { | |
2426 | if (css->ss->attach) { | |
2427 | tset->ssid = i; | |
2428 | css->ss->attach(tset); | |
2429 | } | |
2430 | } | |
2431 | ||
2432 | ret = 0; | |
2433 | goto out_release_tset; | |
2434 | ||
2435 | out_cancel_attach: | |
2436 | for_each_e_css(css, i, dst_cgrp) { | |
2437 | if (css == failed_css) | |
2438 | break; | |
2439 | if (css->ss->cancel_attach) { | |
2440 | tset->ssid = i; | |
2441 | css->ss->cancel_attach(tset); | |
2442 | } | |
2443 | } | |
2444 | out_release_tset: | |
2445 | spin_lock_bh(&css_set_lock); | |
2446 | list_splice_init(&tset->dst_csets, &tset->src_csets); | |
2447 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2448 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2449 | list_del_init(&cset->mg_node); | |
2450 | } | |
2451 | spin_unlock_bh(&css_set_lock); | |
2452 | return ret; | |
2453 | } | |
2454 | ||
2455 | /** | |
2456 | * cgroup_migrate_finish - cleanup after attach | |
2457 | * @preloaded_csets: list of preloaded css_sets | |
2458 | * | |
2459 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See | |
2460 | * those functions for details. | |
2461 | */ | |
2462 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) | |
2463 | { | |
2464 | struct css_set *cset, *tmp_cset; | |
2465 | ||
2466 | lockdep_assert_held(&cgroup_mutex); | |
2467 | ||
2468 | spin_lock_bh(&css_set_lock); | |
2469 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2470 | cset->mg_src_cgrp = NULL; | |
2471 | cset->mg_dst_cset = NULL; | |
2472 | list_del_init(&cset->mg_preload_node); | |
2473 | put_css_set_locked(cset); | |
2474 | } | |
2475 | spin_unlock_bh(&css_set_lock); | |
2476 | } | |
2477 | ||
2478 | /** | |
2479 | * cgroup_migrate_add_src - add a migration source css_set | |
2480 | * @src_cset: the source css_set to add | |
2481 | * @dst_cgrp: the destination cgroup | |
2482 | * @preloaded_csets: list of preloaded css_sets | |
2483 | * | |
2484 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2485 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2486 | * up by cgroup_migrate_finish(). | |
2487 | * | |
2488 | * This function may be called without holding cgroup_threadgroup_rwsem | |
2489 | * even if the target is a process. Threads may be created and destroyed | |
2490 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2491 | * into play and the preloaded css_sets are guaranteed to cover all | |
2492 | * migrations. | |
2493 | */ | |
2494 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2495 | struct cgroup *dst_cgrp, | |
2496 | struct list_head *preloaded_csets) | |
2497 | { | |
2498 | struct cgroup *src_cgrp; | |
2499 | ||
2500 | lockdep_assert_held(&cgroup_mutex); | |
2501 | lockdep_assert_held(&css_set_lock); | |
2502 | ||
2503 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2504 | ||
2505 | if (!list_empty(&src_cset->mg_preload_node)) | |
2506 | return; | |
2507 | ||
2508 | WARN_ON(src_cset->mg_src_cgrp); | |
2509 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2510 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2511 | ||
2512 | src_cset->mg_src_cgrp = src_cgrp; | |
2513 | get_css_set(src_cset); | |
2514 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2515 | } | |
2516 | ||
2517 | /** | |
2518 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
2519 | * @dst_cgrp: the destination cgroup (may be %NULL) | |
2520 | * @preloaded_csets: list of preloaded source css_sets | |
2521 | * | |
2522 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2523 | * have been preloaded to @preloaded_csets. This function looks up and | |
2524 | * pins all destination css_sets, links each to its source, and append them | |
2525 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2526 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
2527 | * | |
2528 | * This function must be called after cgroup_migrate_add_src() has been | |
2529 | * called on each migration source css_set. After migration is performed | |
2530 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2531 | * @preloaded_csets. | |
2532 | */ | |
2533 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2534 | struct list_head *preloaded_csets) | |
2535 | { | |
2536 | LIST_HEAD(csets); | |
2537 | struct css_set *src_cset, *tmp_cset; | |
2538 | ||
2539 | lockdep_assert_held(&cgroup_mutex); | |
2540 | ||
2541 | /* | |
2542 | * Except for the root, subtree_control must be zero for a cgroup | |
2543 | * with tasks so that child cgroups don't compete against tasks. | |
2544 | */ | |
2545 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && | |
2546 | dst_cgrp->subtree_control) | |
2547 | return -EBUSY; | |
2548 | ||
2549 | /* look up the dst cset for each src cset and link it to src */ | |
2550 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2551 | struct css_set *dst_cset; | |
2552 | ||
2553 | dst_cset = find_css_set(src_cset, | |
2554 | dst_cgrp ?: src_cset->dfl_cgrp); | |
2555 | if (!dst_cset) | |
2556 | goto err; | |
2557 | ||
2558 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
2559 | ||
2560 | /* | |
2561 | * If src cset equals dst, it's noop. Drop the src. | |
2562 | * cgroup_migrate() will skip the cset too. Note that we | |
2563 | * can't handle src == dst as some nodes are used by both. | |
2564 | */ | |
2565 | if (src_cset == dst_cset) { | |
2566 | src_cset->mg_src_cgrp = NULL; | |
2567 | list_del_init(&src_cset->mg_preload_node); | |
2568 | put_css_set(src_cset); | |
2569 | put_css_set(dst_cset); | |
2570 | continue; | |
2571 | } | |
2572 | ||
2573 | src_cset->mg_dst_cset = dst_cset; | |
2574 | ||
2575 | if (list_empty(&dst_cset->mg_preload_node)) | |
2576 | list_add(&dst_cset->mg_preload_node, &csets); | |
2577 | else | |
2578 | put_css_set(dst_cset); | |
2579 | } | |
2580 | ||
2581 | list_splice_tail(&csets, preloaded_csets); | |
2582 | return 0; | |
2583 | err: | |
2584 | cgroup_migrate_finish(&csets); | |
2585 | return -ENOMEM; | |
2586 | } | |
2587 | ||
2588 | /** | |
2589 | * cgroup_migrate - migrate a process or task to a cgroup | |
2590 | * @leader: the leader of the process or the task to migrate | |
2591 | * @threadgroup: whether @leader points to the whole process or a single task | |
2592 | * @cgrp: the destination cgroup | |
2593 | * | |
2594 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2595 | * process, the caller must be holding cgroup_threadgroup_rwsem. The | |
2596 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2597 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2598 | * function and following up with cgroup_migrate_finish(). | |
2599 | * | |
2600 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2601 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2602 | * failure, when migrating multiple targets, the success or failure can be | |
2603 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2604 | * actually starting migrating. | |
2605 | */ | |
2606 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, | |
2607 | struct cgroup *cgrp) | |
2608 | { | |
2609 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); | |
2610 | struct task_struct *task; | |
2611 | ||
2612 | /* | |
2613 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2614 | * already PF_EXITING could be freed from underneath us unless we | |
2615 | * take an rcu_read_lock. | |
2616 | */ | |
2617 | spin_lock_bh(&css_set_lock); | |
2618 | rcu_read_lock(); | |
2619 | task = leader; | |
2620 | do { | |
2621 | cgroup_taskset_add(task, &tset); | |
2622 | if (!threadgroup) | |
2623 | break; | |
2624 | } while_each_thread(leader, task); | |
2625 | rcu_read_unlock(); | |
2626 | spin_unlock_bh(&css_set_lock); | |
2627 | ||
2628 | return cgroup_taskset_migrate(&tset, cgrp); | |
2629 | } | |
2630 | ||
2631 | /** | |
2632 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2633 | * @dst_cgrp: the cgroup to attach to | |
2634 | * @leader: the task or the leader of the threadgroup to be attached | |
2635 | * @threadgroup: attach the whole threadgroup? | |
2636 | * | |
2637 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. | |
2638 | */ | |
2639 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2640 | struct task_struct *leader, bool threadgroup) | |
2641 | { | |
2642 | LIST_HEAD(preloaded_csets); | |
2643 | struct task_struct *task; | |
2644 | int ret; | |
2645 | ||
2646 | /* look up all src csets */ | |
2647 | spin_lock_bh(&css_set_lock); | |
2648 | rcu_read_lock(); | |
2649 | task = leader; | |
2650 | do { | |
2651 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2652 | &preloaded_csets); | |
2653 | if (!threadgroup) | |
2654 | break; | |
2655 | } while_each_thread(leader, task); | |
2656 | rcu_read_unlock(); | |
2657 | spin_unlock_bh(&css_set_lock); | |
2658 | ||
2659 | /* prepare dst csets and commit */ | |
2660 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2661 | if (!ret) | |
2662 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); | |
2663 | ||
2664 | cgroup_migrate_finish(&preloaded_csets); | |
2665 | return ret; | |
2666 | } | |
2667 | ||
2668 | static int cgroup_procs_write_permission(struct task_struct *task, | |
2669 | struct cgroup *dst_cgrp, | |
2670 | struct kernfs_open_file *of) | |
2671 | { | |
2672 | const struct cred *cred = current_cred(); | |
2673 | const struct cred *tcred = get_task_cred(task); | |
2674 | int ret = 0; | |
2675 | ||
2676 | /* | |
2677 | * even if we're attaching all tasks in the thread group, we only | |
2678 | * need to check permissions on one of them. | |
2679 | */ | |
2680 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2681 | !uid_eq(cred->euid, tcred->uid) && | |
2682 | !uid_eq(cred->euid, tcred->suid)) | |
2683 | ret = -EACCES; | |
2684 | ||
2685 | if (!ret && cgroup_on_dfl(dst_cgrp)) { | |
2686 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2687 | struct cgroup *cgrp; | |
2688 | struct inode *inode; | |
2689 | ||
2690 | spin_lock_bh(&css_set_lock); | |
2691 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
2692 | spin_unlock_bh(&css_set_lock); | |
2693 | ||
2694 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2695 | cgrp = cgroup_parent(cgrp); | |
2696 | ||
2697 | ret = -ENOMEM; | |
2698 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); | |
2699 | if (inode) { | |
2700 | ret = inode_permission(inode, MAY_WRITE); | |
2701 | iput(inode); | |
2702 | } | |
2703 | } | |
2704 | ||
2705 | put_cred(tcred); | |
2706 | return ret; | |
2707 | } | |
2708 | ||
2709 | /* | |
2710 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
2711 | * function to attach either it or all tasks in its threadgroup. Will lock | |
2712 | * cgroup_mutex and threadgroup. | |
2713 | */ | |
2714 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, | |
2715 | size_t nbytes, loff_t off, bool threadgroup) | |
2716 | { | |
2717 | struct task_struct *tsk; | |
2718 | struct cgroup *cgrp; | |
2719 | pid_t pid; | |
2720 | int ret; | |
2721 | ||
2722 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) | |
2723 | return -EINVAL; | |
2724 | ||
2725 | cgrp = cgroup_kn_lock_live(of->kn); | |
2726 | if (!cgrp) | |
2727 | return -ENODEV; | |
2728 | ||
2729 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
2730 | rcu_read_lock(); | |
2731 | if (pid) { | |
2732 | tsk = find_task_by_vpid(pid); | |
2733 | if (!tsk) { | |
2734 | ret = -ESRCH; | |
2735 | goto out_unlock_rcu; | |
2736 | } | |
2737 | } else { | |
2738 | tsk = current; | |
2739 | } | |
2740 | ||
2741 | if (threadgroup) | |
2742 | tsk = tsk->group_leader; | |
2743 | ||
2744 | /* | |
2745 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become | |
2746 | * trapped in a cpuset, or RT worker may be born in a cgroup | |
2747 | * with no rt_runtime allocated. Just say no. | |
2748 | */ | |
2749 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { | |
2750 | ret = -EINVAL; | |
2751 | goto out_unlock_rcu; | |
2752 | } | |
2753 | ||
2754 | get_task_struct(tsk); | |
2755 | rcu_read_unlock(); | |
2756 | ||
2757 | ret = cgroup_procs_write_permission(tsk, cgrp, of); | |
2758 | if (!ret) | |
2759 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2760 | ||
2761 | put_task_struct(tsk); | |
2762 | goto out_unlock_threadgroup; | |
2763 | ||
2764 | out_unlock_rcu: | |
2765 | rcu_read_unlock(); | |
2766 | out_unlock_threadgroup: | |
2767 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2768 | cgroup_kn_unlock(of->kn); | |
2769 | cpuset_post_attach_flush(); | |
2770 | return ret ?: nbytes; | |
2771 | } | |
2772 | ||
2773 | /** | |
2774 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2775 | * @from: attach to all cgroups of a given task | |
2776 | * @tsk: the task to be attached | |
2777 | */ | |
2778 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2779 | { | |
2780 | struct cgroup_root *root; | |
2781 | int retval = 0; | |
2782 | ||
2783 | mutex_lock(&cgroup_mutex); | |
2784 | for_each_root(root) { | |
2785 | struct cgroup *from_cgrp; | |
2786 | ||
2787 | if (root == &cgrp_dfl_root) | |
2788 | continue; | |
2789 | ||
2790 | spin_lock_bh(&css_set_lock); | |
2791 | from_cgrp = task_cgroup_from_root(from, root); | |
2792 | spin_unlock_bh(&css_set_lock); | |
2793 | ||
2794 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
2795 | if (retval) | |
2796 | break; | |
2797 | } | |
2798 | mutex_unlock(&cgroup_mutex); | |
2799 | ||
2800 | return retval; | |
2801 | } | |
2802 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2803 | ||
2804 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, | |
2805 | char *buf, size_t nbytes, loff_t off) | |
2806 | { | |
2807 | return __cgroup_procs_write(of, buf, nbytes, off, false); | |
2808 | } | |
2809 | ||
2810 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, | |
2811 | char *buf, size_t nbytes, loff_t off) | |
2812 | { | |
2813 | return __cgroup_procs_write(of, buf, nbytes, off, true); | |
2814 | } | |
2815 | ||
2816 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
2817 | char *buf, size_t nbytes, loff_t off) | |
2818 | { | |
2819 | struct cgroup *cgrp; | |
2820 | ||
2821 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
2822 | ||
2823 | cgrp = cgroup_kn_lock_live(of->kn); | |
2824 | if (!cgrp) | |
2825 | return -ENODEV; | |
2826 | spin_lock(&release_agent_path_lock); | |
2827 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
2828 | sizeof(cgrp->root->release_agent_path)); | |
2829 | spin_unlock(&release_agent_path_lock); | |
2830 | cgroup_kn_unlock(of->kn); | |
2831 | return nbytes; | |
2832 | } | |
2833 | ||
2834 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
2835 | { | |
2836 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2837 | ||
2838 | spin_lock(&release_agent_path_lock); | |
2839 | seq_puts(seq, cgrp->root->release_agent_path); | |
2840 | spin_unlock(&release_agent_path_lock); | |
2841 | seq_putc(seq, '\n'); | |
2842 | return 0; | |
2843 | } | |
2844 | ||
2845 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
2846 | { | |
2847 | seq_puts(seq, "0\n"); | |
2848 | return 0; | |
2849 | } | |
2850 | ||
2851 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) | |
2852 | { | |
2853 | struct cgroup_subsys *ss; | |
2854 | bool printed = false; | |
2855 | int ssid; | |
2856 | ||
2857 | do_each_subsys_mask(ss, ssid, ss_mask) { | |
2858 | if (printed) | |
2859 | seq_putc(seq, ' '); | |
2860 | seq_printf(seq, "%s", ss->name); | |
2861 | printed = true; | |
2862 | } while_each_subsys_mask(); | |
2863 | if (printed) | |
2864 | seq_putc(seq, '\n'); | |
2865 | } | |
2866 | ||
2867 | /* show controllers which are currently attached to the default hierarchy */ | |
2868 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
2869 | { | |
2870 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2871 | ||
2872 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & | |
2873 | ~cgrp_dfl_inhibit_ss_mask); | |
2874 | return 0; | |
2875 | } | |
2876 | ||
2877 | /* show controllers which are enabled from the parent */ | |
2878 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
2879 | { | |
2880 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2881 | ||
2882 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); | |
2883 | return 0; | |
2884 | } | |
2885 | ||
2886 | /* show controllers which are enabled for a given cgroup's children */ | |
2887 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
2888 | { | |
2889 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2890 | ||
2891 | cgroup_print_ss_mask(seq, cgrp->subtree_control); | |
2892 | return 0; | |
2893 | } | |
2894 | ||
2895 | /** | |
2896 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2897 | * @cgrp: root of the subtree to update csses for | |
2898 | * | |
2899 | * @cgrp's subtree_ss_mask has changed and its subtree's (self excluded) | |
2900 | * css associations need to be updated accordingly. This function looks up | |
2901 | * all css_sets which are attached to the subtree, creates the matching | |
2902 | * updated css_sets and migrates the tasks to the new ones. | |
2903 | */ | |
2904 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2905 | { | |
2906 | LIST_HEAD(preloaded_csets); | |
2907 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); | |
2908 | struct cgroup_subsys_state *css; | |
2909 | struct css_set *src_cset; | |
2910 | int ret; | |
2911 | ||
2912 | lockdep_assert_held(&cgroup_mutex); | |
2913 | ||
2914 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
2915 | ||
2916 | /* look up all csses currently attached to @cgrp's subtree */ | |
2917 | spin_lock_bh(&css_set_lock); | |
2918 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2919 | struct cgrp_cset_link *link; | |
2920 | ||
2921 | /* self is not affected by subtree_ss_mask change */ | |
2922 | if (css->cgroup == cgrp) | |
2923 | continue; | |
2924 | ||
2925 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2926 | cgroup_migrate_add_src(link->cset, cgrp, | |
2927 | &preloaded_csets); | |
2928 | } | |
2929 | spin_unlock_bh(&css_set_lock); | |
2930 | ||
2931 | /* NULL dst indicates self on default hierarchy */ | |
2932 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2933 | if (ret) | |
2934 | goto out_finish; | |
2935 | ||
2936 | spin_lock_bh(&css_set_lock); | |
2937 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2938 | struct task_struct *task, *ntask; | |
2939 | ||
2940 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2941 | if (!src_cset->mg_src_cgrp) | |
2942 | break; | |
2943 | ||
2944 | /* all tasks in src_csets need to be migrated */ | |
2945 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2946 | cgroup_taskset_add(task, &tset); | |
2947 | } | |
2948 | spin_unlock_bh(&css_set_lock); | |
2949 | ||
2950 | ret = cgroup_taskset_migrate(&tset, cgrp); | |
2951 | out_finish: | |
2952 | cgroup_migrate_finish(&preloaded_csets); | |
2953 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2954 | return ret; | |
2955 | } | |
2956 | ||
2957 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
2958 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, | |
2959 | char *buf, size_t nbytes, | |
2960 | loff_t off) | |
2961 | { | |
2962 | u16 enable = 0, disable = 0; | |
2963 | u16 css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
2964 | struct cgroup *cgrp, *child; | |
2965 | struct cgroup_subsys *ss; | |
2966 | char *tok; | |
2967 | int ssid, ret; | |
2968 | ||
2969 | /* | |
2970 | * Parse input - space separated list of subsystem names prefixed | |
2971 | * with either + or -. | |
2972 | */ | |
2973 | buf = strstrip(buf); | |
2974 | while ((tok = strsep(&buf, " "))) { | |
2975 | if (tok[0] == '\0') | |
2976 | continue; | |
2977 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { | |
2978 | if (!cgroup_ssid_enabled(ssid) || | |
2979 | strcmp(tok + 1, ss->name)) | |
2980 | continue; | |
2981 | ||
2982 | if (*tok == '+') { | |
2983 | enable |= 1 << ssid; | |
2984 | disable &= ~(1 << ssid); | |
2985 | } else if (*tok == '-') { | |
2986 | disable |= 1 << ssid; | |
2987 | enable &= ~(1 << ssid); | |
2988 | } else { | |
2989 | return -EINVAL; | |
2990 | } | |
2991 | break; | |
2992 | } while_each_subsys_mask(); | |
2993 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2994 | return -EINVAL; | |
2995 | } | |
2996 | ||
2997 | cgrp = cgroup_kn_lock_live(of->kn); | |
2998 | if (!cgrp) | |
2999 | return -ENODEV; | |
3000 | ||
3001 | for_each_subsys(ss, ssid) { | |
3002 | if (enable & (1 << ssid)) { | |
3003 | if (cgrp->subtree_control & (1 << ssid)) { | |
3004 | enable &= ~(1 << ssid); | |
3005 | continue; | |
3006 | } | |
3007 | ||
3008 | /* unavailable or not enabled on the parent? */ | |
3009 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
3010 | (cgroup_parent(cgrp) && | |
3011 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { | |
3012 | ret = -ENOENT; | |
3013 | goto out_unlock; | |
3014 | } | |
3015 | } else if (disable & (1 << ssid)) { | |
3016 | if (!(cgrp->subtree_control & (1 << ssid))) { | |
3017 | disable &= ~(1 << ssid); | |
3018 | continue; | |
3019 | } | |
3020 | ||
3021 | /* a child has it enabled? */ | |
3022 | cgroup_for_each_live_child(child, cgrp) { | |
3023 | if (child->subtree_control & (1 << ssid)) { | |
3024 | ret = -EBUSY; | |
3025 | goto out_unlock; | |
3026 | } | |
3027 | } | |
3028 | } | |
3029 | } | |
3030 | ||
3031 | if (!enable && !disable) { | |
3032 | ret = 0; | |
3033 | goto out_unlock; | |
3034 | } | |
3035 | ||
3036 | /* | |
3037 | * Except for the root, subtree_control must be zero for a cgroup | |
3038 | * with tasks so that child cgroups don't compete against tasks. | |
3039 | */ | |
3040 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { | |
3041 | ret = -EBUSY; | |
3042 | goto out_unlock; | |
3043 | } | |
3044 | ||
3045 | /* | |
3046 | * Update subsys masks and calculate what needs to be done. More | |
3047 | * subsystems than specified may need to be enabled or disabled | |
3048 | * depending on subsystem dependencies. | |
3049 | */ | |
3050 | old_sc = cgrp->subtree_control; | |
3051 | old_ss = cgrp->subtree_ss_mask; | |
3052 | new_sc = (old_sc | enable) & ~disable; | |
3053 | new_ss = cgroup_calc_subtree_ss_mask(cgrp, new_sc); | |
3054 | ||
3055 | css_enable = ~old_ss & new_ss; | |
3056 | css_disable = old_ss & ~new_ss; | |
3057 | enable |= css_enable; | |
3058 | disable |= css_disable; | |
3059 | ||
3060 | /* | |
3061 | * Because css offlining is asynchronous, userland might try to | |
3062 | * re-enable the same controller while the previous instance is | |
3063 | * still around. In such cases, wait till it's gone using | |
3064 | * offline_waitq. | |
3065 | */ | |
3066 | do_each_subsys_mask(ss, ssid, css_enable) { | |
3067 | cgroup_for_each_live_child(child, cgrp) { | |
3068 | DEFINE_WAIT(wait); | |
3069 | ||
3070 | if (!cgroup_css(child, ss)) | |
3071 | continue; | |
3072 | ||
3073 | cgroup_get(child); | |
3074 | prepare_to_wait(&child->offline_waitq, &wait, | |
3075 | TASK_UNINTERRUPTIBLE); | |
3076 | cgroup_kn_unlock(of->kn); | |
3077 | schedule(); | |
3078 | finish_wait(&child->offline_waitq, &wait); | |
3079 | cgroup_put(child); | |
3080 | ||
3081 | return restart_syscall(); | |
3082 | } | |
3083 | } while_each_subsys_mask(); | |
3084 | ||
3085 | cgrp->subtree_control = new_sc; | |
3086 | cgrp->subtree_ss_mask = new_ss; | |
3087 | ||
3088 | /* | |
3089 | * Create new csses or make the existing ones visible. A css is | |
3090 | * created invisible if it's being implicitly enabled through | |
3091 | * dependency. An invisible css is made visible when the userland | |
3092 | * explicitly enables it. | |
3093 | */ | |
3094 | do_each_subsys_mask(ss, ssid, enable) { | |
3095 | cgroup_for_each_live_child(child, cgrp) { | |
3096 | if (css_enable & (1 << ssid)) { | |
3097 | struct cgroup_subsys_state *css; | |
3098 | ||
3099 | css = css_create(child, ss); | |
3100 | if (IS_ERR(css)) { | |
3101 | ret = PTR_ERR(css); | |
3102 | goto err_undo_css; | |
3103 | } | |
3104 | ||
3105 | if (cgrp->subtree_control & (1 << ssid)) { | |
3106 | ret = css_populate_dir(css, NULL); | |
3107 | if (ret) | |
3108 | goto err_undo_css; | |
3109 | } | |
3110 | } else { | |
3111 | ret = css_populate_dir(cgroup_css(child, ss), | |
3112 | NULL); | |
3113 | if (ret) | |
3114 | goto err_undo_css; | |
3115 | } | |
3116 | } | |
3117 | } while_each_subsys_mask(); | |
3118 | ||
3119 | /* | |
3120 | * At this point, cgroup_e_css() results reflect the new csses | |
3121 | * making the following cgroup_update_dfl_csses() properly update | |
3122 | * css associations of all tasks in the subtree. | |
3123 | */ | |
3124 | ret = cgroup_update_dfl_csses(cgrp); | |
3125 | if (ret) | |
3126 | goto err_undo_css; | |
3127 | ||
3128 | /* | |
3129 | * All tasks are migrated out of disabled csses. Kill or hide | |
3130 | * them. A css is hidden when the userland requests it to be | |
3131 | * disabled while other subsystems are still depending on it. The | |
3132 | * css must not actively control resources and be in the vanilla | |
3133 | * state if it's made visible again later. Controllers which may | |
3134 | * be depended upon should provide ->css_reset() for this purpose. | |
3135 | */ | |
3136 | do_each_subsys_mask(ss, ssid, disable) { | |
3137 | cgroup_for_each_live_child(child, cgrp) { | |
3138 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
3139 | ||
3140 | if (css_disable & (1 << ssid)) { | |
3141 | kill_css(css); | |
3142 | } else { | |
3143 | css_clear_dir(css, NULL); | |
3144 | if (ss->css_reset) | |
3145 | ss->css_reset(css); | |
3146 | } | |
3147 | } | |
3148 | } while_each_subsys_mask(); | |
3149 | ||
3150 | kernfs_activate(cgrp->kn); | |
3151 | ret = 0; | |
3152 | out_unlock: | |
3153 | cgroup_kn_unlock(of->kn); | |
3154 | return ret ?: nbytes; | |
3155 | ||
3156 | err_undo_css: | |
3157 | cgrp->subtree_control = old_sc; | |
3158 | cgrp->subtree_ss_mask = old_ss; | |
3159 | ||
3160 | do_each_subsys_mask(ss, ssid, enable) { | |
3161 | cgroup_for_each_live_child(child, cgrp) { | |
3162 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
3163 | ||
3164 | if (!css) | |
3165 | continue; | |
3166 | ||
3167 | if (css_enable & (1 << ssid)) | |
3168 | kill_css(css); | |
3169 | else | |
3170 | css_clear_dir(css, NULL); | |
3171 | } | |
3172 | } while_each_subsys_mask(); | |
3173 | goto out_unlock; | |
3174 | } | |
3175 | ||
3176 | static int cgroup_events_show(struct seq_file *seq, void *v) | |
3177 | { | |
3178 | seq_printf(seq, "populated %d\n", | |
3179 | cgroup_is_populated(seq_css(seq)->cgroup)); | |
3180 | return 0; | |
3181 | } | |
3182 | ||
3183 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, | |
3184 | size_t nbytes, loff_t off) | |
3185 | { | |
3186 | struct cgroup *cgrp = of->kn->parent->priv; | |
3187 | struct cftype *cft = of->kn->priv; | |
3188 | struct cgroup_subsys_state *css; | |
3189 | int ret; | |
3190 | ||
3191 | if (cft->write) | |
3192 | return cft->write(of, buf, nbytes, off); | |
3193 | ||
3194 | /* | |
3195 | * kernfs guarantees that a file isn't deleted with operations in | |
3196 | * flight, which means that the matching css is and stays alive and | |
3197 | * doesn't need to be pinned. The RCU locking is not necessary | |
3198 | * either. It's just for the convenience of using cgroup_css(). | |
3199 | */ | |
3200 | rcu_read_lock(); | |
3201 | css = cgroup_css(cgrp, cft->ss); | |
3202 | rcu_read_unlock(); | |
3203 | ||
3204 | if (cft->write_u64) { | |
3205 | unsigned long long v; | |
3206 | ret = kstrtoull(buf, 0, &v); | |
3207 | if (!ret) | |
3208 | ret = cft->write_u64(css, cft, v); | |
3209 | } else if (cft->write_s64) { | |
3210 | long long v; | |
3211 | ret = kstrtoll(buf, 0, &v); | |
3212 | if (!ret) | |
3213 | ret = cft->write_s64(css, cft, v); | |
3214 | } else { | |
3215 | ret = -EINVAL; | |
3216 | } | |
3217 | ||
3218 | return ret ?: nbytes; | |
3219 | } | |
3220 | ||
3221 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) | |
3222 | { | |
3223 | return seq_cft(seq)->seq_start(seq, ppos); | |
3224 | } | |
3225 | ||
3226 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) | |
3227 | { | |
3228 | return seq_cft(seq)->seq_next(seq, v, ppos); | |
3229 | } | |
3230 | ||
3231 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) | |
3232 | { | |
3233 | seq_cft(seq)->seq_stop(seq, v); | |
3234 | } | |
3235 | ||
3236 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
3237 | { | |
3238 | struct cftype *cft = seq_cft(m); | |
3239 | struct cgroup_subsys_state *css = seq_css(m); | |
3240 | ||
3241 | if (cft->seq_show) | |
3242 | return cft->seq_show(m, arg); | |
3243 | ||
3244 | if (cft->read_u64) | |
3245 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); | |
3246 | else if (cft->read_s64) | |
3247 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3248 | else | |
3249 | return -EINVAL; | |
3250 | return 0; | |
3251 | } | |
3252 | ||
3253 | static struct kernfs_ops cgroup_kf_single_ops = { | |
3254 | .atomic_write_len = PAGE_SIZE, | |
3255 | .write = cgroup_file_write, | |
3256 | .seq_show = cgroup_seqfile_show, | |
3257 | }; | |
3258 | ||
3259 | static struct kernfs_ops cgroup_kf_ops = { | |
3260 | .atomic_write_len = PAGE_SIZE, | |
3261 | .write = cgroup_file_write, | |
3262 | .seq_start = cgroup_seqfile_start, | |
3263 | .seq_next = cgroup_seqfile_next, | |
3264 | .seq_stop = cgroup_seqfile_stop, | |
3265 | .seq_show = cgroup_seqfile_show, | |
3266 | }; | |
3267 | ||
3268 | /* | |
3269 | * cgroup_rename - Only allow simple rename of directories in place. | |
3270 | */ | |
3271 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, | |
3272 | const char *new_name_str) | |
3273 | { | |
3274 | struct cgroup *cgrp = kn->priv; | |
3275 | int ret; | |
3276 | ||
3277 | if (kernfs_type(kn) != KERNFS_DIR) | |
3278 | return -ENOTDIR; | |
3279 | if (kn->parent != new_parent) | |
3280 | return -EIO; | |
3281 | ||
3282 | /* | |
3283 | * This isn't a proper migration and its usefulness is very | |
3284 | * limited. Disallow on the default hierarchy. | |
3285 | */ | |
3286 | if (cgroup_on_dfl(cgrp)) | |
3287 | return -EPERM; | |
3288 | ||
3289 | /* | |
3290 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
3291 | * active_ref. kernfs_rename() doesn't require active_ref | |
3292 | * protection. Break them before grabbing cgroup_mutex. | |
3293 | */ | |
3294 | kernfs_break_active_protection(new_parent); | |
3295 | kernfs_break_active_protection(kn); | |
3296 | ||
3297 | mutex_lock(&cgroup_mutex); | |
3298 | ||
3299 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
3300 | ||
3301 | mutex_unlock(&cgroup_mutex); | |
3302 | ||
3303 | kernfs_unbreak_active_protection(kn); | |
3304 | kernfs_unbreak_active_protection(new_parent); | |
3305 | return ret; | |
3306 | } | |
3307 | ||
3308 | /* set uid and gid of cgroup dirs and files to that of the creator */ | |
3309 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3310 | { | |
3311 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3312 | .ia_uid = current_fsuid(), | |
3313 | .ia_gid = current_fsgid(), }; | |
3314 | ||
3315 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3316 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3317 | return 0; | |
3318 | ||
3319 | return kernfs_setattr(kn, &iattr); | |
3320 | } | |
3321 | ||
3322 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, | |
3323 | struct cftype *cft) | |
3324 | { | |
3325 | char name[CGROUP_FILE_NAME_MAX]; | |
3326 | struct kernfs_node *kn; | |
3327 | struct lock_class_key *key = NULL; | |
3328 | int ret; | |
3329 | ||
3330 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
3331 | key = &cft->lockdep_key; | |
3332 | #endif | |
3333 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3334 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
3335 | NULL, key); | |
3336 | if (IS_ERR(kn)) | |
3337 | return PTR_ERR(kn); | |
3338 | ||
3339 | ret = cgroup_kn_set_ugid(kn); | |
3340 | if (ret) { | |
3341 | kernfs_remove(kn); | |
3342 | return ret; | |
3343 | } | |
3344 | ||
3345 | if (cft->file_offset) { | |
3346 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3347 | ||
3348 | spin_lock_irq(&cgroup_file_kn_lock); | |
3349 | cfile->kn = kn; | |
3350 | spin_unlock_irq(&cgroup_file_kn_lock); | |
3351 | } | |
3352 | ||
3353 | return 0; | |
3354 | } | |
3355 | ||
3356 | /** | |
3357 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
3358 | * @css: the target css | |
3359 | * @cgrp: the target cgroup (usually css->cgroup) | |
3360 | * @cfts: array of cftypes to be added | |
3361 | * @is_add: whether to add or remove | |
3362 | * | |
3363 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
3364 | * For removals, this function never fails. | |
3365 | */ | |
3366 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, | |
3367 | struct cgroup *cgrp, struct cftype cfts[], | |
3368 | bool is_add) | |
3369 | { | |
3370 | struct cftype *cft, *cft_end = NULL; | |
3371 | int ret = 0; | |
3372 | ||
3373 | lockdep_assert_held(&cgroup_mutex); | |
3374 | ||
3375 | restart: | |
3376 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
3377 | /* does cft->flags tell us to skip this file on @cgrp? */ | |
3378 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) | |
3379 | continue; | |
3380 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) | |
3381 | continue; | |
3382 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) | |
3383 | continue; | |
3384 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) | |
3385 | continue; | |
3386 | ||
3387 | if (is_add) { | |
3388 | ret = cgroup_add_file(css, cgrp, cft); | |
3389 | if (ret) { | |
3390 | pr_warn("%s: failed to add %s, err=%d\n", | |
3391 | __func__, cft->name, ret); | |
3392 | cft_end = cft; | |
3393 | is_add = false; | |
3394 | goto restart; | |
3395 | } | |
3396 | } else { | |
3397 | cgroup_rm_file(cgrp, cft); | |
3398 | } | |
3399 | } | |
3400 | return ret; | |
3401 | } | |
3402 | ||
3403 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) | |
3404 | { | |
3405 | LIST_HEAD(pending); | |
3406 | struct cgroup_subsys *ss = cfts[0].ss; | |
3407 | struct cgroup *root = &ss->root->cgrp; | |
3408 | struct cgroup_subsys_state *css; | |
3409 | int ret = 0; | |
3410 | ||
3411 | lockdep_assert_held(&cgroup_mutex); | |
3412 | ||
3413 | /* add/rm files for all cgroups created before */ | |
3414 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { | |
3415 | struct cgroup *cgrp = css->cgroup; | |
3416 | ||
3417 | if (!(css->flags & CSS_VISIBLE)) | |
3418 | continue; | |
3419 | ||
3420 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); | |
3421 | if (ret) | |
3422 | break; | |
3423 | } | |
3424 | ||
3425 | if (is_add && !ret) | |
3426 | kernfs_activate(root->kn); | |
3427 | return ret; | |
3428 | } | |
3429 | ||
3430 | static void cgroup_exit_cftypes(struct cftype *cfts) | |
3431 | { | |
3432 | struct cftype *cft; | |
3433 | ||
3434 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
3435 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3436 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3437 | kfree(cft->kf_ops); | |
3438 | cft->kf_ops = NULL; | |
3439 | cft->ss = NULL; | |
3440 | ||
3441 | /* revert flags set by cgroup core while adding @cfts */ | |
3442 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); | |
3443 | } | |
3444 | } | |
3445 | ||
3446 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3447 | { | |
3448 | struct cftype *cft; | |
3449 | ||
3450 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
3451 | struct kernfs_ops *kf_ops; | |
3452 | ||
3453 | WARN_ON(cft->ss || cft->kf_ops); | |
3454 | ||
3455 | if (cft->seq_start) | |
3456 | kf_ops = &cgroup_kf_ops; | |
3457 | else | |
3458 | kf_ops = &cgroup_kf_single_ops; | |
3459 | ||
3460 | /* | |
3461 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3462 | * make a copy of kf_ops to set its atomic_write_len. | |
3463 | */ | |
3464 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3465 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3466 | if (!kf_ops) { | |
3467 | cgroup_exit_cftypes(cfts); | |
3468 | return -ENOMEM; | |
3469 | } | |
3470 | kf_ops->atomic_write_len = cft->max_write_len; | |
3471 | } | |
3472 | ||
3473 | cft->kf_ops = kf_ops; | |
3474 | cft->ss = ss; | |
3475 | } | |
3476 | ||
3477 | return 0; | |
3478 | } | |
3479 | ||
3480 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) | |
3481 | { | |
3482 | lockdep_assert_held(&cgroup_mutex); | |
3483 | ||
3484 | if (!cfts || !cfts[0].ss) | |
3485 | return -ENOENT; | |
3486 | ||
3487 | list_del(&cfts->node); | |
3488 | cgroup_apply_cftypes(cfts, false); | |
3489 | cgroup_exit_cftypes(cfts); | |
3490 | return 0; | |
3491 | } | |
3492 | ||
3493 | /** | |
3494 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
3495 | * @cfts: zero-length name terminated array of cftypes | |
3496 | * | |
3497 | * Unregister @cfts. Files described by @cfts are removed from all | |
3498 | * existing cgroups and all future cgroups won't have them either. This | |
3499 | * function can be called anytime whether @cfts' subsys is attached or not. | |
3500 | * | |
3501 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
3502 | * registered. | |
3503 | */ | |
3504 | int cgroup_rm_cftypes(struct cftype *cfts) | |
3505 | { | |
3506 | int ret; | |
3507 | ||
3508 | mutex_lock(&cgroup_mutex); | |
3509 | ret = cgroup_rm_cftypes_locked(cfts); | |
3510 | mutex_unlock(&cgroup_mutex); | |
3511 | return ret; | |
3512 | } | |
3513 | ||
3514 | /** | |
3515 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3516 | * @ss: target cgroup subsystem | |
3517 | * @cfts: zero-length name terminated array of cftypes | |
3518 | * | |
3519 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3520 | * existing cgroups to which @ss is attached and all future cgroups will | |
3521 | * have them too. This function can be called anytime whether @ss is | |
3522 | * attached or not. | |
3523 | * | |
3524 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3525 | * function currently returns 0 as long as @cfts registration is successful | |
3526 | * even if some file creation attempts on existing cgroups fail. | |
3527 | */ | |
3528 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3529 | { | |
3530 | int ret; | |
3531 | ||
3532 | if (!cgroup_ssid_enabled(ss->id)) | |
3533 | return 0; | |
3534 | ||
3535 | if (!cfts || cfts[0].name[0] == '\0') | |
3536 | return 0; | |
3537 | ||
3538 | ret = cgroup_init_cftypes(ss, cfts); | |
3539 | if (ret) | |
3540 | return ret; | |
3541 | ||
3542 | mutex_lock(&cgroup_mutex); | |
3543 | ||
3544 | list_add_tail(&cfts->node, &ss->cfts); | |
3545 | ret = cgroup_apply_cftypes(cfts, true); | |
3546 | if (ret) | |
3547 | cgroup_rm_cftypes_locked(cfts); | |
3548 | ||
3549 | mutex_unlock(&cgroup_mutex); | |
3550 | return ret; | |
3551 | } | |
3552 | ||
3553 | /** | |
3554 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3555 | * @ss: target cgroup subsystem | |
3556 | * @cfts: zero-length name terminated array of cftypes | |
3557 | * | |
3558 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3559 | * the default hierarchy. | |
3560 | */ | |
3561 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3562 | { | |
3563 | struct cftype *cft; | |
3564 | ||
3565 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
3566 | cft->flags |= __CFTYPE_ONLY_ON_DFL; | |
3567 | return cgroup_add_cftypes(ss, cfts); | |
3568 | } | |
3569 | ||
3570 | /** | |
3571 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3572 | * @ss: target cgroup subsystem | |
3573 | * @cfts: zero-length name terminated array of cftypes | |
3574 | * | |
3575 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3576 | * the legacy hierarchies. | |
3577 | */ | |
3578 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3579 | { | |
3580 | struct cftype *cft; | |
3581 | ||
3582 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
3583 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
3584 | return cgroup_add_cftypes(ss, cfts); | |
3585 | } | |
3586 | ||
3587 | /** | |
3588 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3589 | * @cfile: target cgroup_file | |
3590 | * | |
3591 | * @cfile must have been obtained by setting cftype->file_offset. | |
3592 | */ | |
3593 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3594 | { | |
3595 | unsigned long flags; | |
3596 | ||
3597 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3598 | if (cfile->kn) | |
3599 | kernfs_notify(cfile->kn); | |
3600 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3601 | } | |
3602 | ||
3603 | /** | |
3604 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3605 | * @cgrp: the cgroup in question | |
3606 | * | |
3607 | * Return the number of tasks in the cgroup. | |
3608 | */ | |
3609 | static int cgroup_task_count(const struct cgroup *cgrp) | |
3610 | { | |
3611 | int count = 0; | |
3612 | struct cgrp_cset_link *link; | |
3613 | ||
3614 | spin_lock_bh(&css_set_lock); | |
3615 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
3616 | count += atomic_read(&link->cset->refcount); | |
3617 | spin_unlock_bh(&css_set_lock); | |
3618 | return count; | |
3619 | } | |
3620 | ||
3621 | /** | |
3622 | * css_next_child - find the next child of a given css | |
3623 | * @pos: the current position (%NULL to initiate traversal) | |
3624 | * @parent: css whose children to walk | |
3625 | * | |
3626 | * This function returns the next child of @parent and should be called | |
3627 | * under either cgroup_mutex or RCU read lock. The only requirement is | |
3628 | * that @parent and @pos are accessible. The next sibling is guaranteed to | |
3629 | * be returned regardless of their states. | |
3630 | * | |
3631 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3632 | * css which finished ->css_online() is guaranteed to be visible in the | |
3633 | * future iterations and will stay visible until the last reference is put. | |
3634 | * A css which hasn't finished ->css_online() or already finished | |
3635 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3636 | * responsibility to synchronize against on/offlining. | |
3637 | */ | |
3638 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, | |
3639 | struct cgroup_subsys_state *parent) | |
3640 | { | |
3641 | struct cgroup_subsys_state *next; | |
3642 | ||
3643 | cgroup_assert_mutex_or_rcu_locked(); | |
3644 | ||
3645 | /* | |
3646 | * @pos could already have been unlinked from the sibling list. | |
3647 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3648 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3649 | * @pos is taken off list, at which time its next pointer is valid, | |
3650 | * and, as releases are serialized, the one pointed to by the next | |
3651 | * pointer is guaranteed to not have started release yet. This | |
3652 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3653 | * critical section, the one pointed to by its next pointer is | |
3654 | * guaranteed to not have finished its RCU grace period even if we | |
3655 | * have dropped rcu_read_lock() inbetween iterations. | |
3656 | * | |
3657 | * If @pos has CSS_RELEASED set, its next pointer can't be | |
3658 | * dereferenced; however, as each css is given a monotonically | |
3659 | * increasing unique serial number and always appended to the | |
3660 | * sibling list, the next one can be found by walking the parent's | |
3661 | * children until the first css with higher serial number than | |
3662 | * @pos's. While this path can be slower, it happens iff iteration | |
3663 | * races against release and the race window is very small. | |
3664 | */ | |
3665 | if (!pos) { | |
3666 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); | |
3667 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3668 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3669 | } else { | |
3670 | list_for_each_entry_rcu(next, &parent->children, sibling) | |
3671 | if (next->serial_nr > pos->serial_nr) | |
3672 | break; | |
3673 | } | |
3674 | ||
3675 | /* | |
3676 | * @next, if not pointing to the head, can be dereferenced and is | |
3677 | * the next sibling. | |
3678 | */ | |
3679 | if (&next->sibling != &parent->children) | |
3680 | return next; | |
3681 | return NULL; | |
3682 | } | |
3683 | ||
3684 | /** | |
3685 | * css_next_descendant_pre - find the next descendant for pre-order walk | |
3686 | * @pos: the current position (%NULL to initiate traversal) | |
3687 | * @root: css whose descendants to walk | |
3688 | * | |
3689 | * To be used by css_for_each_descendant_pre(). Find the next descendant | |
3690 | * to visit for pre-order traversal of @root's descendants. @root is | |
3691 | * included in the iteration and the first node to be visited. | |
3692 | * | |
3693 | * While this function requires cgroup_mutex or RCU read locking, it | |
3694 | * doesn't require the whole traversal to be contained in a single critical | |
3695 | * section. This function will return the correct next descendant as long | |
3696 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
3697 | * | |
3698 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3699 | * css which finished ->css_online() is guaranteed to be visible in the | |
3700 | * future iterations and will stay visible until the last reference is put. | |
3701 | * A css which hasn't finished ->css_online() or already finished | |
3702 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3703 | * responsibility to synchronize against on/offlining. | |
3704 | */ | |
3705 | struct cgroup_subsys_state * | |
3706 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3707 | struct cgroup_subsys_state *root) | |
3708 | { | |
3709 | struct cgroup_subsys_state *next; | |
3710 | ||
3711 | cgroup_assert_mutex_or_rcu_locked(); | |
3712 | ||
3713 | /* if first iteration, visit @root */ | |
3714 | if (!pos) | |
3715 | return root; | |
3716 | ||
3717 | /* visit the first child if exists */ | |
3718 | next = css_next_child(NULL, pos); | |
3719 | if (next) | |
3720 | return next; | |
3721 | ||
3722 | /* no child, visit my or the closest ancestor's next sibling */ | |
3723 | while (pos != root) { | |
3724 | next = css_next_child(pos, pos->parent); | |
3725 | if (next) | |
3726 | return next; | |
3727 | pos = pos->parent; | |
3728 | } | |
3729 | ||
3730 | return NULL; | |
3731 | } | |
3732 | ||
3733 | /** | |
3734 | * css_rightmost_descendant - return the rightmost descendant of a css | |
3735 | * @pos: css of interest | |
3736 | * | |
3737 | * Return the rightmost descendant of @pos. If there's no descendant, @pos | |
3738 | * is returned. This can be used during pre-order traversal to skip | |
3739 | * subtree of @pos. | |
3740 | * | |
3741 | * While this function requires cgroup_mutex or RCU read locking, it | |
3742 | * doesn't require the whole traversal to be contained in a single critical | |
3743 | * section. This function will return the correct rightmost descendant as | |
3744 | * long as @pos is accessible. | |
3745 | */ | |
3746 | struct cgroup_subsys_state * | |
3747 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
3748 | { | |
3749 | struct cgroup_subsys_state *last, *tmp; | |
3750 | ||
3751 | cgroup_assert_mutex_or_rcu_locked(); | |
3752 | ||
3753 | do { | |
3754 | last = pos; | |
3755 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3756 | pos = NULL; | |
3757 | css_for_each_child(tmp, last) | |
3758 | pos = tmp; | |
3759 | } while (pos); | |
3760 | ||
3761 | return last; | |
3762 | } | |
3763 | ||
3764 | static struct cgroup_subsys_state * | |
3765 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
3766 | { | |
3767 | struct cgroup_subsys_state *last; | |
3768 | ||
3769 | do { | |
3770 | last = pos; | |
3771 | pos = css_next_child(NULL, pos); | |
3772 | } while (pos); | |
3773 | ||
3774 | return last; | |
3775 | } | |
3776 | ||
3777 | /** | |
3778 | * css_next_descendant_post - find the next descendant for post-order walk | |
3779 | * @pos: the current position (%NULL to initiate traversal) | |
3780 | * @root: css whose descendants to walk | |
3781 | * | |
3782 | * To be used by css_for_each_descendant_post(). Find the next descendant | |
3783 | * to visit for post-order traversal of @root's descendants. @root is | |
3784 | * included in the iteration and the last node to be visited. | |
3785 | * | |
3786 | * While this function requires cgroup_mutex or RCU read locking, it | |
3787 | * doesn't require the whole traversal to be contained in a single critical | |
3788 | * section. This function will return the correct next descendant as long | |
3789 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3790 | * @cgroup. | |
3791 | * | |
3792 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3793 | * css which finished ->css_online() is guaranteed to be visible in the | |
3794 | * future iterations and will stay visible until the last reference is put. | |
3795 | * A css which hasn't finished ->css_online() or already finished | |
3796 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3797 | * responsibility to synchronize against on/offlining. | |
3798 | */ | |
3799 | struct cgroup_subsys_state * | |
3800 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3801 | struct cgroup_subsys_state *root) | |
3802 | { | |
3803 | struct cgroup_subsys_state *next; | |
3804 | ||
3805 | cgroup_assert_mutex_or_rcu_locked(); | |
3806 | ||
3807 | /* if first iteration, visit leftmost descendant which may be @root */ | |
3808 | if (!pos) | |
3809 | return css_leftmost_descendant(root); | |
3810 | ||
3811 | /* if we visited @root, we're done */ | |
3812 | if (pos == root) | |
3813 | return NULL; | |
3814 | ||
3815 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
3816 | next = css_next_child(pos, pos->parent); | |
3817 | if (next) | |
3818 | return css_leftmost_descendant(next); | |
3819 | ||
3820 | /* no sibling left, visit parent */ | |
3821 | return pos->parent; | |
3822 | } | |
3823 | ||
3824 | /** | |
3825 | * css_has_online_children - does a css have online children | |
3826 | * @css: the target css | |
3827 | * | |
3828 | * Returns %true if @css has any online children; otherwise, %false. This | |
3829 | * function can be called from any context but the caller is responsible | |
3830 | * for synchronizing against on/offlining as necessary. | |
3831 | */ | |
3832 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
3833 | { | |
3834 | struct cgroup_subsys_state *child; | |
3835 | bool ret = false; | |
3836 | ||
3837 | rcu_read_lock(); | |
3838 | css_for_each_child(child, css) { | |
3839 | if (child->flags & CSS_ONLINE) { | |
3840 | ret = true; | |
3841 | break; | |
3842 | } | |
3843 | } | |
3844 | rcu_read_unlock(); | |
3845 | return ret; | |
3846 | } | |
3847 | ||
3848 | /** | |
3849 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set | |
3850 | * @it: the iterator to advance | |
3851 | * | |
3852 | * Advance @it to the next css_set to walk. | |
3853 | */ | |
3854 | static void css_task_iter_advance_css_set(struct css_task_iter *it) | |
3855 | { | |
3856 | struct list_head *l = it->cset_pos; | |
3857 | struct cgrp_cset_link *link; | |
3858 | struct css_set *cset; | |
3859 | ||
3860 | lockdep_assert_held(&css_set_lock); | |
3861 | ||
3862 | /* Advance to the next non-empty css_set */ | |
3863 | do { | |
3864 | l = l->next; | |
3865 | if (l == it->cset_head) { | |
3866 | it->cset_pos = NULL; | |
3867 | it->task_pos = NULL; | |
3868 | return; | |
3869 | } | |
3870 | ||
3871 | if (it->ss) { | |
3872 | cset = container_of(l, struct css_set, | |
3873 | e_cset_node[it->ss->id]); | |
3874 | } else { | |
3875 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3876 | cset = link->cset; | |
3877 | } | |
3878 | } while (!css_set_populated(cset)); | |
3879 | ||
3880 | it->cset_pos = l; | |
3881 | ||
3882 | if (!list_empty(&cset->tasks)) | |
3883 | it->task_pos = cset->tasks.next; | |
3884 | else | |
3885 | it->task_pos = cset->mg_tasks.next; | |
3886 | ||
3887 | it->tasks_head = &cset->tasks; | |
3888 | it->mg_tasks_head = &cset->mg_tasks; | |
3889 | ||
3890 | /* | |
3891 | * We don't keep css_sets locked across iteration steps and thus | |
3892 | * need to take steps to ensure that iteration can be resumed after | |
3893 | * the lock is re-acquired. Iteration is performed at two levels - | |
3894 | * css_sets and tasks in them. | |
3895 | * | |
3896 | * Once created, a css_set never leaves its cgroup lists, so a | |
3897 | * pinned css_set is guaranteed to stay put and we can resume | |
3898 | * iteration afterwards. | |
3899 | * | |
3900 | * Tasks may leave @cset across iteration steps. This is resolved | |
3901 | * by registering each iterator with the css_set currently being | |
3902 | * walked and making css_set_move_task() advance iterators whose | |
3903 | * next task is leaving. | |
3904 | */ | |
3905 | if (it->cur_cset) { | |
3906 | list_del(&it->iters_node); | |
3907 | put_css_set_locked(it->cur_cset); | |
3908 | } | |
3909 | get_css_set(cset); | |
3910 | it->cur_cset = cset; | |
3911 | list_add(&it->iters_node, &cset->task_iters); | |
3912 | } | |
3913 | ||
3914 | static void css_task_iter_advance(struct css_task_iter *it) | |
3915 | { | |
3916 | struct list_head *l = it->task_pos; | |
3917 | ||
3918 | lockdep_assert_held(&css_set_lock); | |
3919 | WARN_ON_ONCE(!l); | |
3920 | ||
3921 | /* | |
3922 | * Advance iterator to find next entry. cset->tasks is consumed | |
3923 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3924 | * next cset. | |
3925 | */ | |
3926 | l = l->next; | |
3927 | ||
3928 | if (l == it->tasks_head) | |
3929 | l = it->mg_tasks_head->next; | |
3930 | ||
3931 | if (l == it->mg_tasks_head) | |
3932 | css_task_iter_advance_css_set(it); | |
3933 | else | |
3934 | it->task_pos = l; | |
3935 | } | |
3936 | ||
3937 | /** | |
3938 | * css_task_iter_start - initiate task iteration | |
3939 | * @css: the css to walk tasks of | |
3940 | * @it: the task iterator to use | |
3941 | * | |
3942 | * Initiate iteration through the tasks of @css. The caller can call | |
3943 | * css_task_iter_next() to walk through the tasks until the function | |
3944 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3945 | * called. | |
3946 | */ | |
3947 | void css_task_iter_start(struct cgroup_subsys_state *css, | |
3948 | struct css_task_iter *it) | |
3949 | { | |
3950 | /* no one should try to iterate before mounting cgroups */ | |
3951 | WARN_ON_ONCE(!use_task_css_set_links); | |
3952 | ||
3953 | memset(it, 0, sizeof(*it)); | |
3954 | ||
3955 | spin_lock_bh(&css_set_lock); | |
3956 | ||
3957 | it->ss = css->ss; | |
3958 | ||
3959 | if (it->ss) | |
3960 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3961 | else | |
3962 | it->cset_pos = &css->cgroup->cset_links; | |
3963 | ||
3964 | it->cset_head = it->cset_pos; | |
3965 | ||
3966 | css_task_iter_advance_css_set(it); | |
3967 | ||
3968 | spin_unlock_bh(&css_set_lock); | |
3969 | } | |
3970 | ||
3971 | /** | |
3972 | * css_task_iter_next - return the next task for the iterator | |
3973 | * @it: the task iterator being iterated | |
3974 | * | |
3975 | * The "next" function for task iteration. @it should have been | |
3976 | * initialized via css_task_iter_start(). Returns NULL when the iteration | |
3977 | * reaches the end. | |
3978 | */ | |
3979 | struct task_struct *css_task_iter_next(struct css_task_iter *it) | |
3980 | { | |
3981 | if (it->cur_task) { | |
3982 | put_task_struct(it->cur_task); | |
3983 | it->cur_task = NULL; | |
3984 | } | |
3985 | ||
3986 | spin_lock_bh(&css_set_lock); | |
3987 | ||
3988 | if (it->task_pos) { | |
3989 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
3990 | cg_list); | |
3991 | get_task_struct(it->cur_task); | |
3992 | css_task_iter_advance(it); | |
3993 | } | |
3994 | ||
3995 | spin_unlock_bh(&css_set_lock); | |
3996 | ||
3997 | return it->cur_task; | |
3998 | } | |
3999 | ||
4000 | /** | |
4001 | * css_task_iter_end - finish task iteration | |
4002 | * @it: the task iterator to finish | |
4003 | * | |
4004 | * Finish task iteration started by css_task_iter_start(). | |
4005 | */ | |
4006 | void css_task_iter_end(struct css_task_iter *it) | |
4007 | { | |
4008 | if (it->cur_cset) { | |
4009 | spin_lock_bh(&css_set_lock); | |
4010 | list_del(&it->iters_node); | |
4011 | put_css_set_locked(it->cur_cset); | |
4012 | spin_unlock_bh(&css_set_lock); | |
4013 | } | |
4014 | ||
4015 | if (it->cur_task) | |
4016 | put_task_struct(it->cur_task); | |
4017 | } | |
4018 | ||
4019 | /** | |
4020 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
4021 | * @to: cgroup to which the tasks will be moved | |
4022 | * @from: cgroup in which the tasks currently reside | |
4023 | * | |
4024 | * Locking rules between cgroup_post_fork() and the migration path | |
4025 | * guarantee that, if a task is forking while being migrated, the new child | |
4026 | * is guaranteed to be either visible in the source cgroup after the | |
4027 | * parent's migration is complete or put into the target cgroup. No task | |
4028 | * can slip out of migration through forking. | |
4029 | */ | |
4030 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
4031 | { | |
4032 | LIST_HEAD(preloaded_csets); | |
4033 | struct cgrp_cset_link *link; | |
4034 | struct css_task_iter it; | |
4035 | struct task_struct *task; | |
4036 | int ret; | |
4037 | ||
4038 | mutex_lock(&cgroup_mutex); | |
4039 | ||
4040 | /* all tasks in @from are being moved, all csets are source */ | |
4041 | spin_lock_bh(&css_set_lock); | |
4042 | list_for_each_entry(link, &from->cset_links, cset_link) | |
4043 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
4044 | spin_unlock_bh(&css_set_lock); | |
4045 | ||
4046 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); | |
4047 | if (ret) | |
4048 | goto out_err; | |
4049 | ||
4050 | /* | |
4051 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
4052 | * ->can_attach() fails. | |
4053 | */ | |
4054 | do { | |
4055 | css_task_iter_start(&from->self, &it); | |
4056 | task = css_task_iter_next(&it); | |
4057 | if (task) | |
4058 | get_task_struct(task); | |
4059 | css_task_iter_end(&it); | |
4060 | ||
4061 | if (task) { | |
4062 | ret = cgroup_migrate(task, false, to); | |
4063 | put_task_struct(task); | |
4064 | } | |
4065 | } while (task && !ret); | |
4066 | out_err: | |
4067 | cgroup_migrate_finish(&preloaded_csets); | |
4068 | mutex_unlock(&cgroup_mutex); | |
4069 | return ret; | |
4070 | } | |
4071 | ||
4072 | /* | |
4073 | * Stuff for reading the 'tasks'/'procs' files. | |
4074 | * | |
4075 | * Reading this file can return large amounts of data if a cgroup has | |
4076 | * *lots* of attached tasks. So it may need several calls to read(), | |
4077 | * but we cannot guarantee that the information we produce is correct | |
4078 | * unless we produce it entirely atomically. | |
4079 | * | |
4080 | */ | |
4081 | ||
4082 | /* which pidlist file are we talking about? */ | |
4083 | enum cgroup_filetype { | |
4084 | CGROUP_FILE_PROCS, | |
4085 | CGROUP_FILE_TASKS, | |
4086 | }; | |
4087 | ||
4088 | /* | |
4089 | * A pidlist is a list of pids that virtually represents the contents of one | |
4090 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4091 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4092 | * to the cgroup. | |
4093 | */ | |
4094 | struct cgroup_pidlist { | |
4095 | /* | |
4096 | * used to find which pidlist is wanted. doesn't change as long as | |
4097 | * this particular list stays in the list. | |
4098 | */ | |
4099 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4100 | /* array of xids */ | |
4101 | pid_t *list; | |
4102 | /* how many elements the above list has */ | |
4103 | int length; | |
4104 | /* each of these stored in a list by its cgroup */ | |
4105 | struct list_head links; | |
4106 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4107 | struct cgroup *owner; | |
4108 | /* for delayed destruction */ | |
4109 | struct delayed_work destroy_dwork; | |
4110 | }; | |
4111 | ||
4112 | /* | |
4113 | * The following two functions "fix" the issue where there are more pids | |
4114 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4115 | * TODO: replace with a kernel-wide solution to this problem | |
4116 | */ | |
4117 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4118 | static void *pidlist_allocate(int count) | |
4119 | { | |
4120 | if (PIDLIST_TOO_LARGE(count)) | |
4121 | return vmalloc(count * sizeof(pid_t)); | |
4122 | else | |
4123 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4124 | } | |
4125 | ||
4126 | static void pidlist_free(void *p) | |
4127 | { | |
4128 | kvfree(p); | |
4129 | } | |
4130 | ||
4131 | /* | |
4132 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4133 | * should be left afterwards. | |
4134 | */ | |
4135 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4136 | { | |
4137 | struct cgroup_pidlist *l, *tmp_l; | |
4138 | ||
4139 | mutex_lock(&cgrp->pidlist_mutex); | |
4140 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4141 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4142 | mutex_unlock(&cgrp->pidlist_mutex); | |
4143 | ||
4144 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4145 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4146 | } | |
4147 | ||
4148 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4149 | { | |
4150 | struct delayed_work *dwork = to_delayed_work(work); | |
4151 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4152 | destroy_dwork); | |
4153 | struct cgroup_pidlist *tofree = NULL; | |
4154 | ||
4155 | mutex_lock(&l->owner->pidlist_mutex); | |
4156 | ||
4157 | /* | |
4158 | * Destroy iff we didn't get queued again. The state won't change | |
4159 | * as destroy_dwork can only be queued while locked. | |
4160 | */ | |
4161 | if (!delayed_work_pending(dwork)) { | |
4162 | list_del(&l->links); | |
4163 | pidlist_free(l->list); | |
4164 | put_pid_ns(l->key.ns); | |
4165 | tofree = l; | |
4166 | } | |
4167 | ||
4168 | mutex_unlock(&l->owner->pidlist_mutex); | |
4169 | kfree(tofree); | |
4170 | } | |
4171 | ||
4172 | /* | |
4173 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
4174 | * Returns the number of unique elements. | |
4175 | */ | |
4176 | static int pidlist_uniq(pid_t *list, int length) | |
4177 | { | |
4178 | int src, dest = 1; | |
4179 | ||
4180 | /* | |
4181 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4182 | * edge cases first; no work needs to be done for either | |
4183 | */ | |
4184 | if (length == 0 || length == 1) | |
4185 | return length; | |
4186 | /* src and dest walk down the list; dest counts unique elements */ | |
4187 | for (src = 1; src < length; src++) { | |
4188 | /* find next unique element */ | |
4189 | while (list[src] == list[src-1]) { | |
4190 | src++; | |
4191 | if (src == length) | |
4192 | goto after; | |
4193 | } | |
4194 | /* dest always points to where the next unique element goes */ | |
4195 | list[dest] = list[src]; | |
4196 | dest++; | |
4197 | } | |
4198 | after: | |
4199 | return dest; | |
4200 | } | |
4201 | ||
4202 | /* | |
4203 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4204 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4205 | * different per namespace, each namespace needs differently sorted list, | |
4206 | * making it impossible to use, for example, single rbtree of member tasks | |
4207 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4208 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4209 | * pidlists keyed by cgroup and namespace. | |
4210 | * | |
4211 | * All this extra complexity was caused by the original implementation | |
4212 | * committing to an entirely unnecessary property. In the long term, we | |
4213 | * want to do away with it. Explicitly scramble sort order if on the | |
4214 | * default hierarchy so that no such expectation exists in the new | |
4215 | * interface. | |
4216 | * | |
4217 | * Scrambling is done by swapping every two consecutive bits, which is | |
4218 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4219 | */ | |
4220 | static pid_t pid_fry(pid_t pid) | |
4221 | { | |
4222 | unsigned a = pid & 0x55555555; | |
4223 | unsigned b = pid & 0xAAAAAAAA; | |
4224 | ||
4225 | return (a << 1) | (b >> 1); | |
4226 | } | |
4227 | ||
4228 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4229 | { | |
4230 | if (cgroup_on_dfl(cgrp)) | |
4231 | return pid_fry(pid); | |
4232 | else | |
4233 | return pid; | |
4234 | } | |
4235 | ||
4236 | static int cmppid(const void *a, const void *b) | |
4237 | { | |
4238 | return *(pid_t *)a - *(pid_t *)b; | |
4239 | } | |
4240 | ||
4241 | static int fried_cmppid(const void *a, const void *b) | |
4242 | { | |
4243 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4244 | } | |
4245 | ||
4246 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
4247 | enum cgroup_filetype type) | |
4248 | { | |
4249 | struct cgroup_pidlist *l; | |
4250 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4251 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4252 | ||
4253 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4254 | ||
4255 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4256 | if (l->key.type == type && l->key.ns == ns) | |
4257 | return l; | |
4258 | return NULL; | |
4259 | } | |
4260 | ||
4261 | /* | |
4262 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4263 | * returns with the lock on that pidlist already held, and takes care | |
4264 | * of the use count, or returns NULL with no locks held if we're out of | |
4265 | * memory. | |
4266 | */ | |
4267 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
4268 | enum cgroup_filetype type) | |
4269 | { | |
4270 | struct cgroup_pidlist *l; | |
4271 | ||
4272 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4273 | ||
4274 | l = cgroup_pidlist_find(cgrp, type); | |
4275 | if (l) | |
4276 | return l; | |
4277 | ||
4278 | /* entry not found; create a new one */ | |
4279 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
4280 | if (!l) | |
4281 | return l; | |
4282 | ||
4283 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
4284 | l->key.type = type; | |
4285 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4286 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
4287 | l->owner = cgrp; | |
4288 | list_add(&l->links, &cgrp->pidlists); | |
4289 | return l; | |
4290 | } | |
4291 | ||
4292 | /* | |
4293 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4294 | */ | |
4295 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
4296 | struct cgroup_pidlist **lp) | |
4297 | { | |
4298 | pid_t *array; | |
4299 | int length; | |
4300 | int pid, n = 0; /* used for populating the array */ | |
4301 | struct css_task_iter it; | |
4302 | struct task_struct *tsk; | |
4303 | struct cgroup_pidlist *l; | |
4304 | ||
4305 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4306 | ||
4307 | /* | |
4308 | * If cgroup gets more users after we read count, we won't have | |
4309 | * enough space - tough. This race is indistinguishable to the | |
4310 | * caller from the case that the additional cgroup users didn't | |
4311 | * show up until sometime later on. | |
4312 | */ | |
4313 | length = cgroup_task_count(cgrp); | |
4314 | array = pidlist_allocate(length); | |
4315 | if (!array) | |
4316 | return -ENOMEM; | |
4317 | /* now, populate the array */ | |
4318 | css_task_iter_start(&cgrp->self, &it); | |
4319 | while ((tsk = css_task_iter_next(&it))) { | |
4320 | if (unlikely(n == length)) | |
4321 | break; | |
4322 | /* get tgid or pid for procs or tasks file respectively */ | |
4323 | if (type == CGROUP_FILE_PROCS) | |
4324 | pid = task_tgid_vnr(tsk); | |
4325 | else | |
4326 | pid = task_pid_vnr(tsk); | |
4327 | if (pid > 0) /* make sure to only use valid results */ | |
4328 | array[n++] = pid; | |
4329 | } | |
4330 | css_task_iter_end(&it); | |
4331 | length = n; | |
4332 | /* now sort & (if procs) strip out duplicates */ | |
4333 | if (cgroup_on_dfl(cgrp)) | |
4334 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
4335 | else | |
4336 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
4337 | if (type == CGROUP_FILE_PROCS) | |
4338 | length = pidlist_uniq(array, length); | |
4339 | ||
4340 | l = cgroup_pidlist_find_create(cgrp, type); | |
4341 | if (!l) { | |
4342 | pidlist_free(array); | |
4343 | return -ENOMEM; | |
4344 | } | |
4345 | ||
4346 | /* store array, freeing old if necessary */ | |
4347 | pidlist_free(l->list); | |
4348 | l->list = array; | |
4349 | l->length = length; | |
4350 | *lp = l; | |
4351 | return 0; | |
4352 | } | |
4353 | ||
4354 | /** | |
4355 | * cgroupstats_build - build and fill cgroupstats | |
4356 | * @stats: cgroupstats to fill information into | |
4357 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4358 | * been requested. | |
4359 | * | |
4360 | * Build and fill cgroupstats so that taskstats can export it to user | |
4361 | * space. | |
4362 | */ | |
4363 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4364 | { | |
4365 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
4366 | struct cgroup *cgrp; | |
4367 | struct css_task_iter it; | |
4368 | struct task_struct *tsk; | |
4369 | ||
4370 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
4371 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4372 | kernfs_type(kn) != KERNFS_DIR) | |
4373 | return -EINVAL; | |
4374 | ||
4375 | mutex_lock(&cgroup_mutex); | |
4376 | ||
4377 | /* | |
4378 | * We aren't being called from kernfs and there's no guarantee on | |
4379 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
4380 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
4381 | */ | |
4382 | rcu_read_lock(); | |
4383 | cgrp = rcu_dereference(kn->priv); | |
4384 | if (!cgrp || cgroup_is_dead(cgrp)) { | |
4385 | rcu_read_unlock(); | |
4386 | mutex_unlock(&cgroup_mutex); | |
4387 | return -ENOENT; | |
4388 | } | |
4389 | rcu_read_unlock(); | |
4390 | ||
4391 | css_task_iter_start(&cgrp->self, &it); | |
4392 | while ((tsk = css_task_iter_next(&it))) { | |
4393 | switch (tsk->state) { | |
4394 | case TASK_RUNNING: | |
4395 | stats->nr_running++; | |
4396 | break; | |
4397 | case TASK_INTERRUPTIBLE: | |
4398 | stats->nr_sleeping++; | |
4399 | break; | |
4400 | case TASK_UNINTERRUPTIBLE: | |
4401 | stats->nr_uninterruptible++; | |
4402 | break; | |
4403 | case TASK_STOPPED: | |
4404 | stats->nr_stopped++; | |
4405 | break; | |
4406 | default: | |
4407 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4408 | stats->nr_io_wait++; | |
4409 | break; | |
4410 | } | |
4411 | } | |
4412 | css_task_iter_end(&it); | |
4413 | ||
4414 | mutex_unlock(&cgroup_mutex); | |
4415 | return 0; | |
4416 | } | |
4417 | ||
4418 | ||
4419 | /* | |
4420 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
4421 | * next pid to display; the seq_file iterator is a pointer to the pid | |
4422 | * in the cgroup->l->list array. | |
4423 | */ | |
4424 | ||
4425 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
4426 | { | |
4427 | /* | |
4428 | * Initially we receive a position value that corresponds to | |
4429 | * one more than the last pid shown (or 0 on the first call or | |
4430 | * after a seek to the start). Use a binary-search to find the | |
4431 | * next pid to display, if any | |
4432 | */ | |
4433 | struct kernfs_open_file *of = s->private; | |
4434 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4435 | struct cgroup_pidlist *l; | |
4436 | enum cgroup_filetype type = seq_cft(s)->private; | |
4437 | int index = 0, pid = *pos; | |
4438 | int *iter, ret; | |
4439 | ||
4440 | mutex_lock(&cgrp->pidlist_mutex); | |
4441 | ||
4442 | /* | |
4443 | * !NULL @of->priv indicates that this isn't the first start() | |
4444 | * after open. If the matching pidlist is around, we can use that. | |
4445 | * Look for it. Note that @of->priv can't be used directly. It | |
4446 | * could already have been destroyed. | |
4447 | */ | |
4448 | if (of->priv) | |
4449 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4450 | ||
4451 | /* | |
4452 | * Either this is the first start() after open or the matching | |
4453 | * pidlist has been destroyed inbetween. Create a new one. | |
4454 | */ | |
4455 | if (!of->priv) { | |
4456 | ret = pidlist_array_load(cgrp, type, | |
4457 | (struct cgroup_pidlist **)&of->priv); | |
4458 | if (ret) | |
4459 | return ERR_PTR(ret); | |
4460 | } | |
4461 | l = of->priv; | |
4462 | ||
4463 | if (pid) { | |
4464 | int end = l->length; | |
4465 | ||
4466 | while (index < end) { | |
4467 | int mid = (index + end) / 2; | |
4468 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { | |
4469 | index = mid; | |
4470 | break; | |
4471 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) | |
4472 | index = mid + 1; | |
4473 | else | |
4474 | end = mid; | |
4475 | } | |
4476 | } | |
4477 | /* If we're off the end of the array, we're done */ | |
4478 | if (index >= l->length) | |
4479 | return NULL; | |
4480 | /* Update the abstract position to be the actual pid that we found */ | |
4481 | iter = l->list + index; | |
4482 | *pos = cgroup_pid_fry(cgrp, *iter); | |
4483 | return iter; | |
4484 | } | |
4485 | ||
4486 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
4487 | { | |
4488 | struct kernfs_open_file *of = s->private; | |
4489 | struct cgroup_pidlist *l = of->priv; | |
4490 | ||
4491 | if (l) | |
4492 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
4493 | CGROUP_PIDLIST_DESTROY_DELAY); | |
4494 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); | |
4495 | } | |
4496 | ||
4497 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
4498 | { | |
4499 | struct kernfs_open_file *of = s->private; | |
4500 | struct cgroup_pidlist *l = of->priv; | |
4501 | pid_t *p = v; | |
4502 | pid_t *end = l->list + l->length; | |
4503 | /* | |
4504 | * Advance to the next pid in the array. If this goes off the | |
4505 | * end, we're done | |
4506 | */ | |
4507 | p++; | |
4508 | if (p >= end) { | |
4509 | return NULL; | |
4510 | } else { | |
4511 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); | |
4512 | return p; | |
4513 | } | |
4514 | } | |
4515 | ||
4516 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
4517 | { | |
4518 | seq_printf(s, "%d\n", *(int *)v); | |
4519 | ||
4520 | return 0; | |
4521 | } | |
4522 | ||
4523 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
4524 | struct cftype *cft) | |
4525 | { | |
4526 | return notify_on_release(css->cgroup); | |
4527 | } | |
4528 | ||
4529 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
4530 | struct cftype *cft, u64 val) | |
4531 | { | |
4532 | if (val) | |
4533 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
4534 | else | |
4535 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
4536 | return 0; | |
4537 | } | |
4538 | ||
4539 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
4540 | struct cftype *cft) | |
4541 | { | |
4542 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
4543 | } | |
4544 | ||
4545 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
4546 | struct cftype *cft, u64 val) | |
4547 | { | |
4548 | if (val) | |
4549 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
4550 | else | |
4551 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
4552 | return 0; | |
4553 | } | |
4554 | ||
4555 | /* cgroup core interface files for the default hierarchy */ | |
4556 | static struct cftype cgroup_dfl_base_files[] = { | |
4557 | { | |
4558 | .name = "cgroup.procs", | |
4559 | .file_offset = offsetof(struct cgroup, procs_file), | |
4560 | .seq_start = cgroup_pidlist_start, | |
4561 | .seq_next = cgroup_pidlist_next, | |
4562 | .seq_stop = cgroup_pidlist_stop, | |
4563 | .seq_show = cgroup_pidlist_show, | |
4564 | .private = CGROUP_FILE_PROCS, | |
4565 | .write = cgroup_procs_write, | |
4566 | }, | |
4567 | { | |
4568 | .name = "cgroup.controllers", | |
4569 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4570 | .seq_show = cgroup_root_controllers_show, | |
4571 | }, | |
4572 | { | |
4573 | .name = "cgroup.controllers", | |
4574 | .flags = CFTYPE_NOT_ON_ROOT, | |
4575 | .seq_show = cgroup_controllers_show, | |
4576 | }, | |
4577 | { | |
4578 | .name = "cgroup.subtree_control", | |
4579 | .seq_show = cgroup_subtree_control_show, | |
4580 | .write = cgroup_subtree_control_write, | |
4581 | }, | |
4582 | { | |
4583 | .name = "cgroup.events", | |
4584 | .flags = CFTYPE_NOT_ON_ROOT, | |
4585 | .file_offset = offsetof(struct cgroup, events_file), | |
4586 | .seq_show = cgroup_events_show, | |
4587 | }, | |
4588 | { } /* terminate */ | |
4589 | }; | |
4590 | ||
4591 | /* cgroup core interface files for the legacy hierarchies */ | |
4592 | static struct cftype cgroup_legacy_base_files[] = { | |
4593 | { | |
4594 | .name = "cgroup.procs", | |
4595 | .seq_start = cgroup_pidlist_start, | |
4596 | .seq_next = cgroup_pidlist_next, | |
4597 | .seq_stop = cgroup_pidlist_stop, | |
4598 | .seq_show = cgroup_pidlist_show, | |
4599 | .private = CGROUP_FILE_PROCS, | |
4600 | .write = cgroup_procs_write, | |
4601 | }, | |
4602 | { | |
4603 | .name = "cgroup.clone_children", | |
4604 | .read_u64 = cgroup_clone_children_read, | |
4605 | .write_u64 = cgroup_clone_children_write, | |
4606 | }, | |
4607 | { | |
4608 | .name = "cgroup.sane_behavior", | |
4609 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4610 | .seq_show = cgroup_sane_behavior_show, | |
4611 | }, | |
4612 | { | |
4613 | .name = "tasks", | |
4614 | .seq_start = cgroup_pidlist_start, | |
4615 | .seq_next = cgroup_pidlist_next, | |
4616 | .seq_stop = cgroup_pidlist_stop, | |
4617 | .seq_show = cgroup_pidlist_show, | |
4618 | .private = CGROUP_FILE_TASKS, | |
4619 | .write = cgroup_tasks_write, | |
4620 | }, | |
4621 | { | |
4622 | .name = "notify_on_release", | |
4623 | .read_u64 = cgroup_read_notify_on_release, | |
4624 | .write_u64 = cgroup_write_notify_on_release, | |
4625 | }, | |
4626 | { | |
4627 | .name = "release_agent", | |
4628 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4629 | .seq_show = cgroup_release_agent_show, | |
4630 | .write = cgroup_release_agent_write, | |
4631 | .max_write_len = PATH_MAX - 1, | |
4632 | }, | |
4633 | { } /* terminate */ | |
4634 | }; | |
4635 | ||
4636 | /* | |
4637 | * css destruction is four-stage process. | |
4638 | * | |
4639 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4640 | * Implemented in kill_css(). | |
4641 | * | |
4642 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
4643 | * and thus css_tryget_online() is guaranteed to fail, the css can be | |
4644 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4645 | * put. Implemented in css_killed_work_fn(). | |
4646 | * | |
4647 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4648 | * accessors are inside RCU read sections. css_release() schedules the | |
4649 | * RCU callback. | |
4650 | * | |
4651 | * 4. After the grace period, the css can be freed. Implemented in | |
4652 | * css_free_work_fn(). | |
4653 | * | |
4654 | * It is actually hairier because both step 2 and 4 require process context | |
4655 | * and thus involve punting to css->destroy_work adding two additional | |
4656 | * steps to the already complex sequence. | |
4657 | */ | |
4658 | static void css_free_work_fn(struct work_struct *work) | |
4659 | { | |
4660 | struct cgroup_subsys_state *css = | |
4661 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4662 | struct cgroup_subsys *ss = css->ss; | |
4663 | struct cgroup *cgrp = css->cgroup; | |
4664 | ||
4665 | percpu_ref_exit(&css->refcnt); | |
4666 | ||
4667 | if (ss) { | |
4668 | /* css free path */ | |
4669 | struct cgroup_subsys_state *parent = css->parent; | |
4670 | int id = css->id; | |
4671 | ||
4672 | ss->css_free(css); | |
4673 | cgroup_idr_remove(&ss->css_idr, id); | |
4674 | cgroup_put(cgrp); | |
4675 | ||
4676 | if (parent) | |
4677 | css_put(parent); | |
4678 | } else { | |
4679 | /* cgroup free path */ | |
4680 | atomic_dec(&cgrp->root->nr_cgrps); | |
4681 | cgroup_pidlist_destroy_all(cgrp); | |
4682 | cancel_work_sync(&cgrp->release_agent_work); | |
4683 | ||
4684 | if (cgroup_parent(cgrp)) { | |
4685 | /* | |
4686 | * We get a ref to the parent, and put the ref when | |
4687 | * this cgroup is being freed, so it's guaranteed | |
4688 | * that the parent won't be destroyed before its | |
4689 | * children. | |
4690 | */ | |
4691 | cgroup_put(cgroup_parent(cgrp)); | |
4692 | kernfs_put(cgrp->kn); | |
4693 | kfree(cgrp); | |
4694 | } else { | |
4695 | /* | |
4696 | * This is root cgroup's refcnt reaching zero, | |
4697 | * which indicates that the root should be | |
4698 | * released. | |
4699 | */ | |
4700 | cgroup_destroy_root(cgrp->root); | |
4701 | } | |
4702 | } | |
4703 | } | |
4704 | ||
4705 | static void css_free_rcu_fn(struct rcu_head *rcu_head) | |
4706 | { | |
4707 | struct cgroup_subsys_state *css = | |
4708 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); | |
4709 | ||
4710 | INIT_WORK(&css->destroy_work, css_free_work_fn); | |
4711 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
4712 | } | |
4713 | ||
4714 | static void css_release_work_fn(struct work_struct *work) | |
4715 | { | |
4716 | struct cgroup_subsys_state *css = | |
4717 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4718 | struct cgroup_subsys *ss = css->ss; | |
4719 | struct cgroup *cgrp = css->cgroup; | |
4720 | ||
4721 | mutex_lock(&cgroup_mutex); | |
4722 | ||
4723 | css->flags |= CSS_RELEASED; | |
4724 | list_del_rcu(&css->sibling); | |
4725 | ||
4726 | if (ss) { | |
4727 | /* css release path */ | |
4728 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); | |
4729 | if (ss->css_released) | |
4730 | ss->css_released(css); | |
4731 | } else { | |
4732 | /* cgroup release path */ | |
4733 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
4734 | cgrp->id = -1; | |
4735 | ||
4736 | /* | |
4737 | * There are two control paths which try to determine | |
4738 | * cgroup from dentry without going through kernfs - | |
4739 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4740 | * Those are supported by RCU protecting clearing of | |
4741 | * cgrp->kn->priv backpointer. | |
4742 | */ | |
4743 | if (cgrp->kn) | |
4744 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4745 | NULL); | |
4746 | } | |
4747 | ||
4748 | mutex_unlock(&cgroup_mutex); | |
4749 | ||
4750 | call_rcu(&css->rcu_head, css_free_rcu_fn); | |
4751 | } | |
4752 | ||
4753 | static void css_release(struct percpu_ref *ref) | |
4754 | { | |
4755 | struct cgroup_subsys_state *css = | |
4756 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4757 | ||
4758 | INIT_WORK(&css->destroy_work, css_release_work_fn); | |
4759 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
4760 | } | |
4761 | ||
4762 | static void init_and_link_css(struct cgroup_subsys_state *css, | |
4763 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
4764 | { | |
4765 | lockdep_assert_held(&cgroup_mutex); | |
4766 | ||
4767 | cgroup_get(cgrp); | |
4768 | ||
4769 | memset(css, 0, sizeof(*css)); | |
4770 | css->cgroup = cgrp; | |
4771 | css->ss = ss; | |
4772 | INIT_LIST_HEAD(&css->sibling); | |
4773 | INIT_LIST_HEAD(&css->children); | |
4774 | css->serial_nr = css_serial_nr_next++; | |
4775 | atomic_set(&css->online_cnt, 0); | |
4776 | ||
4777 | if (cgroup_parent(cgrp)) { | |
4778 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
4779 | css_get(css->parent); | |
4780 | } | |
4781 | ||
4782 | BUG_ON(cgroup_css(cgrp, ss)); | |
4783 | } | |
4784 | ||
4785 | /* invoke ->css_online() on a new CSS and mark it online if successful */ | |
4786 | static int online_css(struct cgroup_subsys_state *css) | |
4787 | { | |
4788 | struct cgroup_subsys *ss = css->ss; | |
4789 | int ret = 0; | |
4790 | ||
4791 | lockdep_assert_held(&cgroup_mutex); | |
4792 | ||
4793 | if (ss->css_online) | |
4794 | ret = ss->css_online(css); | |
4795 | if (!ret) { | |
4796 | css->flags |= CSS_ONLINE; | |
4797 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); | |
4798 | ||
4799 | atomic_inc(&css->online_cnt); | |
4800 | if (css->parent) | |
4801 | atomic_inc(&css->parent->online_cnt); | |
4802 | } | |
4803 | return ret; | |
4804 | } | |
4805 | ||
4806 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ | |
4807 | static void offline_css(struct cgroup_subsys_state *css) | |
4808 | { | |
4809 | struct cgroup_subsys *ss = css->ss; | |
4810 | ||
4811 | lockdep_assert_held(&cgroup_mutex); | |
4812 | ||
4813 | if (!(css->flags & CSS_ONLINE)) | |
4814 | return; | |
4815 | ||
4816 | if (ss->css_reset) | |
4817 | ss->css_reset(css); | |
4818 | ||
4819 | if (ss->css_offline) | |
4820 | ss->css_offline(css); | |
4821 | ||
4822 | css->flags &= ~CSS_ONLINE; | |
4823 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); | |
4824 | ||
4825 | wake_up_all(&css->cgroup->offline_waitq); | |
4826 | } | |
4827 | ||
4828 | /** | |
4829 | * css_create - create a cgroup_subsys_state | |
4830 | * @cgrp: the cgroup new css will be associated with | |
4831 | * @ss: the subsys of new css | |
4832 | * | |
4833 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4834 | * css is online and installed in @cgrp. This function doesn't create the | |
4835 | * interface files. Returns 0 on success, -errno on failure. | |
4836 | */ | |
4837 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, | |
4838 | struct cgroup_subsys *ss) | |
4839 | { | |
4840 | struct cgroup *parent = cgroup_parent(cgrp); | |
4841 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); | |
4842 | struct cgroup_subsys_state *css; | |
4843 | int err; | |
4844 | ||
4845 | lockdep_assert_held(&cgroup_mutex); | |
4846 | ||
4847 | css = ss->css_alloc(parent_css); | |
4848 | if (IS_ERR(css)) | |
4849 | return css; | |
4850 | ||
4851 | init_and_link_css(css, ss, cgrp); | |
4852 | ||
4853 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); | |
4854 | if (err) | |
4855 | goto err_free_css; | |
4856 | ||
4857 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); | |
4858 | if (err < 0) | |
4859 | goto err_free_percpu_ref; | |
4860 | css->id = err; | |
4861 | ||
4862 | /* @css is ready to be brought online now, make it visible */ | |
4863 | list_add_tail_rcu(&css->sibling, &parent_css->children); | |
4864 | cgroup_idr_replace(&ss->css_idr, css, css->id); | |
4865 | ||
4866 | err = online_css(css); | |
4867 | if (err) | |
4868 | goto err_list_del; | |
4869 | ||
4870 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4871 | cgroup_parent(parent)) { | |
4872 | pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", | |
4873 | current->comm, current->pid, ss->name); | |
4874 | if (!strcmp(ss->name, "memory")) | |
4875 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); | |
4876 | ss->warned_broken_hierarchy = true; | |
4877 | } | |
4878 | ||
4879 | return css; | |
4880 | ||
4881 | err_list_del: | |
4882 | list_del_rcu(&css->sibling); | |
4883 | cgroup_idr_remove(&ss->css_idr, css->id); | |
4884 | err_free_percpu_ref: | |
4885 | percpu_ref_exit(&css->refcnt); | |
4886 | err_free_css: | |
4887 | call_rcu(&css->rcu_head, css_free_rcu_fn); | |
4888 | return ERR_PTR(err); | |
4889 | } | |
4890 | ||
4891 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
4892 | umode_t mode) | |
4893 | { | |
4894 | struct cgroup *parent, *cgrp, *tcgrp; | |
4895 | struct cgroup_root *root; | |
4896 | struct cgroup_subsys *ss; | |
4897 | struct kernfs_node *kn; | |
4898 | int level, ssid, ret; | |
4899 | ||
4900 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. | |
4901 | */ | |
4902 | if (strchr(name, '\n')) | |
4903 | return -EINVAL; | |
4904 | ||
4905 | parent = cgroup_kn_lock_live(parent_kn); | |
4906 | if (!parent) | |
4907 | return -ENODEV; | |
4908 | root = parent->root; | |
4909 | level = parent->level + 1; | |
4910 | ||
4911 | /* allocate the cgroup and its ID, 0 is reserved for the root */ | |
4912 | cgrp = kzalloc(sizeof(*cgrp) + | |
4913 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
4914 | if (!cgrp) { | |
4915 | ret = -ENOMEM; | |
4916 | goto out_unlock; | |
4917 | } | |
4918 | ||
4919 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); | |
4920 | if (ret) | |
4921 | goto out_free_cgrp; | |
4922 | ||
4923 | /* | |
4924 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4925 | * a half-baked cgroup. | |
4926 | */ | |
4927 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); | |
4928 | if (cgrp->id < 0) { | |
4929 | ret = -ENOMEM; | |
4930 | goto out_cancel_ref; | |
4931 | } | |
4932 | ||
4933 | init_cgroup_housekeeping(cgrp); | |
4934 | ||
4935 | cgrp->self.parent = &parent->self; | |
4936 | cgrp->root = root; | |
4937 | cgrp->level = level; | |
4938 | ||
4939 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
4940 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
4941 | ||
4942 | if (notify_on_release(parent)) | |
4943 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4944 | ||
4945 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) | |
4946 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
4947 | ||
4948 | /* create the directory */ | |
4949 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
4950 | if (IS_ERR(kn)) { | |
4951 | ret = PTR_ERR(kn); | |
4952 | goto out_free_id; | |
4953 | } | |
4954 | cgrp->kn = kn; | |
4955 | ||
4956 | /* | |
4957 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
4958 | * that @cgrp->kn is always accessible. | |
4959 | */ | |
4960 | kernfs_get(kn); | |
4961 | ||
4962 | cgrp->self.serial_nr = css_serial_nr_next++; | |
4963 | ||
4964 | /* allocation complete, commit to creation */ | |
4965 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); | |
4966 | atomic_inc(&root->nr_cgrps); | |
4967 | cgroup_get(parent); | |
4968 | ||
4969 | /* | |
4970 | * @cgrp is now fully operational. If something fails after this | |
4971 | * point, it'll be released via the normal destruction path. | |
4972 | */ | |
4973 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); | |
4974 | ||
4975 | ret = cgroup_kn_set_ugid(kn); | |
4976 | if (ret) | |
4977 | goto out_destroy; | |
4978 | ||
4979 | ret = css_populate_dir(&cgrp->self, NULL); | |
4980 | if (ret) | |
4981 | goto out_destroy; | |
4982 | ||
4983 | /* let's create and online css's */ | |
4984 | do_each_subsys_mask(ss, ssid, parent->subtree_ss_mask) { | |
4985 | struct cgroup_subsys_state *css; | |
4986 | ||
4987 | css = css_create(cgrp, ss); | |
4988 | if (IS_ERR(css)) { | |
4989 | ret = PTR_ERR(css); | |
4990 | goto out_destroy; | |
4991 | } | |
4992 | ||
4993 | if (parent->subtree_control & (1 << ssid)) { | |
4994 | ret = css_populate_dir(css, NULL); | |
4995 | if (ret) | |
4996 | goto out_destroy; | |
4997 | } | |
4998 | } while_each_subsys_mask(); | |
4999 | ||
5000 | /* | |
5001 | * On the default hierarchy, a child doesn't automatically inherit | |
5002 | * subtree_control from the parent. Each is configured manually. | |
5003 | */ | |
5004 | if (!cgroup_on_dfl(cgrp)) { | |
5005 | cgrp->subtree_control = parent->subtree_control; | |
5006 | cgroup_refresh_subtree_ss_mask(cgrp); | |
5007 | } | |
5008 | ||
5009 | kernfs_activate(kn); | |
5010 | ||
5011 | ret = 0; | |
5012 | goto out_unlock; | |
5013 | ||
5014 | out_free_id: | |
5015 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); | |
5016 | out_cancel_ref: | |
5017 | percpu_ref_exit(&cgrp->self.refcnt); | |
5018 | out_free_cgrp: | |
5019 | kfree(cgrp); | |
5020 | out_unlock: | |
5021 | cgroup_kn_unlock(parent_kn); | |
5022 | return ret; | |
5023 | ||
5024 | out_destroy: | |
5025 | cgroup_destroy_locked(cgrp); | |
5026 | goto out_unlock; | |
5027 | } | |
5028 | ||
5029 | /* | |
5030 | * This is called when the refcnt of a css is confirmed to be killed. | |
5031 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to | |
5032 | * initate destruction and put the css ref from kill_css(). | |
5033 | */ | |
5034 | static void css_killed_work_fn(struct work_struct *work) | |
5035 | { | |
5036 | struct cgroup_subsys_state *css = | |
5037 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
5038 | ||
5039 | mutex_lock(&cgroup_mutex); | |
5040 | ||
5041 | do { | |
5042 | offline_css(css); | |
5043 | css_put(css); | |
5044 | /* @css can't go away while we're holding cgroup_mutex */ | |
5045 | css = css->parent; | |
5046 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5047 | ||
5048 | mutex_unlock(&cgroup_mutex); | |
5049 | } | |
5050 | ||
5051 | /* css kill confirmation processing requires process context, bounce */ | |
5052 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
5053 | { | |
5054 | struct cgroup_subsys_state *css = | |
5055 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5056 | ||
5057 | if (atomic_dec_and_test(&css->online_cnt)) { | |
5058 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5059 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5060 | } | |
5061 | } | |
5062 | ||
5063 | /** | |
5064 | * kill_css - destroy a css | |
5065 | * @css: css to destroy | |
5066 | * | |
5067 | * This function initiates destruction of @css by removing cgroup interface | |
5068 | * files and putting its base reference. ->css_offline() will be invoked | |
5069 | * asynchronously once css_tryget_online() is guaranteed to fail and when | |
5070 | * the reference count reaches zero, @css will be released. | |
5071 | */ | |
5072 | static void kill_css(struct cgroup_subsys_state *css) | |
5073 | { | |
5074 | lockdep_assert_held(&cgroup_mutex); | |
5075 | ||
5076 | /* | |
5077 | * This must happen before css is disassociated with its cgroup. | |
5078 | * See seq_css() for details. | |
5079 | */ | |
5080 | css_clear_dir(css, NULL); | |
5081 | ||
5082 | /* | |
5083 | * Killing would put the base ref, but we need to keep it alive | |
5084 | * until after ->css_offline(). | |
5085 | */ | |
5086 | css_get(css); | |
5087 | ||
5088 | /* | |
5089 | * cgroup core guarantees that, by the time ->css_offline() is | |
5090 | * invoked, no new css reference will be given out via | |
5091 | * css_tryget_online(). We can't simply call percpu_ref_kill() and | |
5092 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
5093 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5094 | * | |
5095 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5096 | * css is confirmed to be seen as killed on all CPUs. | |
5097 | */ | |
5098 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
5099 | } | |
5100 | ||
5101 | /** | |
5102 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5103 | * @cgrp: cgroup to be destroyed | |
5104 | * | |
5105 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5106 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
5107 | * guarantee that css_tryget_online() won't succeed by the time | |
5108 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5109 | * destruction is implemented in the following two steps. | |
5110 | * | |
5111 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5112 | * userland visible parts and start killing the percpu refcnts of | |
5113 | * css's. Set up so that the next stage will be kicked off once all | |
5114 | * the percpu refcnts are confirmed to be killed. | |
5115 | * | |
5116 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5117 | * rest of destruction. Once all cgroup references are gone, the | |
5118 | * cgroup is RCU-freed. | |
5119 | * | |
5120 | * This function implements s1. After this step, @cgrp is gone as far as | |
5121 | * the userland is concerned and a new cgroup with the same name may be | |
5122 | * created. As cgroup doesn't care about the names internally, this | |
5123 | * doesn't cause any problem. | |
5124 | */ | |
5125 | static int cgroup_destroy_locked(struct cgroup *cgrp) | |
5126 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
5127 | { | |
5128 | struct cgroup_subsys_state *css; | |
5129 | int ssid; | |
5130 | ||
5131 | lockdep_assert_held(&cgroup_mutex); | |
5132 | ||
5133 | /* | |
5134 | * Only migration can raise populated from zero and we're already | |
5135 | * holding cgroup_mutex. | |
5136 | */ | |
5137 | if (cgroup_is_populated(cgrp)) | |
5138 | return -EBUSY; | |
5139 | ||
5140 | /* | |
5141 | * Make sure there's no live children. We can't test emptiness of | |
5142 | * ->self.children as dead children linger on it while being | |
5143 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
5144 | */ | |
5145 | if (css_has_online_children(&cgrp->self)) | |
5146 | return -EBUSY; | |
5147 | ||
5148 | /* | |
5149 | * Mark @cgrp dead. This prevents further task migration and child | |
5150 | * creation by disabling cgroup_lock_live_group(). | |
5151 | */ | |
5152 | cgrp->self.flags &= ~CSS_ONLINE; | |
5153 | ||
5154 | /* initiate massacre of all css's */ | |
5155 | for_each_css(css, ssid, cgrp) | |
5156 | kill_css(css); | |
5157 | ||
5158 | /* | |
5159 | * Remove @cgrp directory along with the base files. @cgrp has an | |
5160 | * extra ref on its kn. | |
5161 | */ | |
5162 | kernfs_remove(cgrp->kn); | |
5163 | ||
5164 | check_for_release(cgroup_parent(cgrp)); | |
5165 | ||
5166 | /* put the base reference */ | |
5167 | percpu_ref_kill(&cgrp->self.refcnt); | |
5168 | ||
5169 | return 0; | |
5170 | }; | |
5171 | ||
5172 | static int cgroup_rmdir(struct kernfs_node *kn) | |
5173 | { | |
5174 | struct cgroup *cgrp; | |
5175 | int ret = 0; | |
5176 | ||
5177 | cgrp = cgroup_kn_lock_live(kn); | |
5178 | if (!cgrp) | |
5179 | return 0; | |
5180 | ||
5181 | ret = cgroup_destroy_locked(cgrp); | |
5182 | ||
5183 | cgroup_kn_unlock(kn); | |
5184 | return ret; | |
5185 | } | |
5186 | ||
5187 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { | |
5188 | .remount_fs = cgroup_remount, | |
5189 | .show_options = cgroup_show_options, | |
5190 | .mkdir = cgroup_mkdir, | |
5191 | .rmdir = cgroup_rmdir, | |
5192 | .rename = cgroup_rename, | |
5193 | }; | |
5194 | ||
5195 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) | |
5196 | { | |
5197 | struct cgroup_subsys_state *css; | |
5198 | ||
5199 | pr_debug("Initializing cgroup subsys %s\n", ss->name); | |
5200 | ||
5201 | mutex_lock(&cgroup_mutex); | |
5202 | ||
5203 | idr_init(&ss->css_idr); | |
5204 | INIT_LIST_HEAD(&ss->cfts); | |
5205 | ||
5206 | /* Create the root cgroup state for this subsystem */ | |
5207 | ss->root = &cgrp_dfl_root; | |
5208 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
5209 | /* We don't handle early failures gracefully */ | |
5210 | BUG_ON(IS_ERR(css)); | |
5211 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); | |
5212 | ||
5213 | /* | |
5214 | * Root csses are never destroyed and we can't initialize | |
5215 | * percpu_ref during early init. Disable refcnting. | |
5216 | */ | |
5217 | css->flags |= CSS_NO_REF; | |
5218 | ||
5219 | if (early) { | |
5220 | /* allocation can't be done safely during early init */ | |
5221 | css->id = 1; | |
5222 | } else { | |
5223 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5224 | BUG_ON(css->id < 0); | |
5225 | } | |
5226 | ||
5227 | /* Update the init_css_set to contain a subsys | |
5228 | * pointer to this state - since the subsystem is | |
5229 | * newly registered, all tasks and hence the | |
5230 | * init_css_set is in the subsystem's root cgroup. */ | |
5231 | init_css_set.subsys[ss->id] = css; | |
5232 | ||
5233 | have_fork_callback |= (bool)ss->fork << ss->id; | |
5234 | have_exit_callback |= (bool)ss->exit << ss->id; | |
5235 | have_free_callback |= (bool)ss->free << ss->id; | |
5236 | have_canfork_callback |= (bool)ss->can_fork << ss->id; | |
5237 | ||
5238 | /* At system boot, before all subsystems have been | |
5239 | * registered, no tasks have been forked, so we don't | |
5240 | * need to invoke fork callbacks here. */ | |
5241 | BUG_ON(!list_empty(&init_task.tasks)); | |
5242 | ||
5243 | BUG_ON(online_css(css)); | |
5244 | ||
5245 | mutex_unlock(&cgroup_mutex); | |
5246 | } | |
5247 | ||
5248 | /** | |
5249 | * cgroup_init_early - cgroup initialization at system boot | |
5250 | * | |
5251 | * Initialize cgroups at system boot, and initialize any | |
5252 | * subsystems that request early init. | |
5253 | */ | |
5254 | int __init cgroup_init_early(void) | |
5255 | { | |
5256 | static struct cgroup_sb_opts __initdata opts; | |
5257 | struct cgroup_subsys *ss; | |
5258 | int i; | |
5259 | ||
5260 | init_cgroup_root(&cgrp_dfl_root, &opts); | |
5261 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; | |
5262 | ||
5263 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); | |
5264 | ||
5265 | for_each_subsys(ss, i) { | |
5266 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, | |
5267 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", | |
5268 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
5269 | ss->id, ss->name); | |
5270 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, | |
5271 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5272 | ||
5273 | ss->id = i; | |
5274 | ss->name = cgroup_subsys_name[i]; | |
5275 | if (!ss->legacy_name) | |
5276 | ss->legacy_name = cgroup_subsys_name[i]; | |
5277 | ||
5278 | if (ss->early_init) | |
5279 | cgroup_init_subsys(ss, true); | |
5280 | } | |
5281 | return 0; | |
5282 | } | |
5283 | ||
5284 | static u16 cgroup_disable_mask __initdata; | |
5285 | ||
5286 | /** | |
5287 | * cgroup_init - cgroup initialization | |
5288 | * | |
5289 | * Register cgroup filesystem and /proc file, and initialize | |
5290 | * any subsystems that didn't request early init. | |
5291 | */ | |
5292 | int __init cgroup_init(void) | |
5293 | { | |
5294 | struct cgroup_subsys *ss; | |
5295 | int ssid; | |
5296 | ||
5297 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); | |
5298 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); | |
5299 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); | |
5300 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
5301 | ||
5302 | mutex_lock(&cgroup_mutex); | |
5303 | ||
5304 | /* | |
5305 | * Add init_css_set to the hash table so that dfl_root can link to | |
5306 | * it during init. | |
5307 | */ | |
5308 | hash_add(css_set_table, &init_css_set.hlist, | |
5309 | css_set_hash(init_css_set.subsys)); | |
5310 | ||
5311 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); | |
5312 | ||
5313 | mutex_unlock(&cgroup_mutex); | |
5314 | ||
5315 | for_each_subsys(ss, ssid) { | |
5316 | if (ss->early_init) { | |
5317 | struct cgroup_subsys_state *css = | |
5318 | init_css_set.subsys[ss->id]; | |
5319 | ||
5320 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5321 | GFP_KERNEL); | |
5322 | BUG_ON(css->id < 0); | |
5323 | } else { | |
5324 | cgroup_init_subsys(ss, false); | |
5325 | } | |
5326 | ||
5327 | list_add_tail(&init_css_set.e_cset_node[ssid], | |
5328 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
5329 | ||
5330 | /* | |
5331 | * Setting dfl_root subsys_mask needs to consider the | |
5332 | * disabled flag and cftype registration needs kmalloc, | |
5333 | * both of which aren't available during early_init. | |
5334 | */ | |
5335 | if (cgroup_disable_mask & (1 << ssid)) { | |
5336 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5337 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5338 | ss->name); | |
5339 | continue; | |
5340 | } | |
5341 | ||
5342 | if (cgroup_ssid_no_v1(ssid)) | |
5343 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5344 | ss->name); | |
5345 | ||
5346 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
5347 | ||
5348 | if (!ss->dfl_cftypes) | |
5349 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; | |
5350 | ||
5351 | if (ss->dfl_cftypes == ss->legacy_cftypes) { | |
5352 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5353 | } else { | |
5354 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5355 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
5356 | } | |
5357 | ||
5358 | if (ss->bind) | |
5359 | ss->bind(init_css_set.subsys[ssid]); | |
5360 | } | |
5361 | ||
5362 | /* init_css_set.subsys[] has been updated, re-hash */ | |
5363 | hash_del(&init_css_set.hlist); | |
5364 | hash_add(css_set_table, &init_css_set.hlist, | |
5365 | css_set_hash(init_css_set.subsys)); | |
5366 | ||
5367 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); | |
5368 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
5369 | WARN_ON(register_filesystem(&cgroup2_fs_type)); | |
5370 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); | |
5371 | ||
5372 | return 0; | |
5373 | } | |
5374 | ||
5375 | static int __init cgroup_wq_init(void) | |
5376 | { | |
5377 | /* | |
5378 | * There isn't much point in executing destruction path in | |
5379 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
5380 | * Use 1 for @max_active. | |
5381 | * | |
5382 | * We would prefer to do this in cgroup_init() above, but that | |
5383 | * is called before init_workqueues(): so leave this until after. | |
5384 | */ | |
5385 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); | |
5386 | BUG_ON(!cgroup_destroy_wq); | |
5387 | ||
5388 | /* | |
5389 | * Used to destroy pidlists and separate to serve as flush domain. | |
5390 | * Cap @max_active to 1 too. | |
5391 | */ | |
5392 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5393 | 0, 1); | |
5394 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5395 | ||
5396 | return 0; | |
5397 | } | |
5398 | core_initcall(cgroup_wq_init); | |
5399 | ||
5400 | /* | |
5401 | * proc_cgroup_show() | |
5402 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5403 | * - Used for /proc/<pid>/cgroup. | |
5404 | */ | |
5405 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, | |
5406 | struct pid *pid, struct task_struct *tsk) | |
5407 | { | |
5408 | char *buf, *path; | |
5409 | int retval; | |
5410 | struct cgroup_root *root; | |
5411 | ||
5412 | retval = -ENOMEM; | |
5413 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5414 | if (!buf) | |
5415 | goto out; | |
5416 | ||
5417 | mutex_lock(&cgroup_mutex); | |
5418 | spin_lock_bh(&css_set_lock); | |
5419 | ||
5420 | for_each_root(root) { | |
5421 | struct cgroup_subsys *ss; | |
5422 | struct cgroup *cgrp; | |
5423 | int ssid, count = 0; | |
5424 | ||
5425 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) | |
5426 | continue; | |
5427 | ||
5428 | seq_printf(m, "%d:", root->hierarchy_id); | |
5429 | if (root != &cgrp_dfl_root) | |
5430 | for_each_subsys(ss, ssid) | |
5431 | if (root->subsys_mask & (1 << ssid)) | |
5432 | seq_printf(m, "%s%s", count++ ? "," : "", | |
5433 | ss->legacy_name); | |
5434 | if (strlen(root->name)) | |
5435 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5436 | root->name); | |
5437 | seq_putc(m, ':'); | |
5438 | ||
5439 | cgrp = task_cgroup_from_root(tsk, root); | |
5440 | ||
5441 | /* | |
5442 | * On traditional hierarchies, all zombie tasks show up as | |
5443 | * belonging to the root cgroup. On the default hierarchy, | |
5444 | * while a zombie doesn't show up in "cgroup.procs" and | |
5445 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5446 | * reporting the cgroup it belonged to before exiting. If | |
5447 | * the cgroup is removed before the zombie is reaped, | |
5448 | * " (deleted)" is appended to the cgroup path. | |
5449 | */ | |
5450 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
5451 | path = cgroup_path(cgrp, buf, PATH_MAX); | |
5452 | if (!path) { | |
5453 | retval = -ENAMETOOLONG; | |
5454 | goto out_unlock; | |
5455 | } | |
5456 | } else { | |
5457 | path = "/"; | |
5458 | } | |
5459 | ||
5460 | seq_puts(m, path); | |
5461 | ||
5462 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5463 | seq_puts(m, " (deleted)\n"); | |
5464 | else | |
5465 | seq_putc(m, '\n'); | |
5466 | } | |
5467 | ||
5468 | retval = 0; | |
5469 | out_unlock: | |
5470 | spin_unlock_bh(&css_set_lock); | |
5471 | mutex_unlock(&cgroup_mutex); | |
5472 | kfree(buf); | |
5473 | out: | |
5474 | return retval; | |
5475 | } | |
5476 | ||
5477 | /* Display information about each subsystem and each hierarchy */ | |
5478 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5479 | { | |
5480 | struct cgroup_subsys *ss; | |
5481 | int i; | |
5482 | ||
5483 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
5484 | /* | |
5485 | * ideally we don't want subsystems moving around while we do this. | |
5486 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5487 | * subsys/hierarchy state. | |
5488 | */ | |
5489 | mutex_lock(&cgroup_mutex); | |
5490 | ||
5491 | for_each_subsys(ss, i) | |
5492 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
5493 | ss->legacy_name, ss->root->hierarchy_id, | |
5494 | atomic_read(&ss->root->nr_cgrps), | |
5495 | cgroup_ssid_enabled(i)); | |
5496 | ||
5497 | mutex_unlock(&cgroup_mutex); | |
5498 | return 0; | |
5499 | } | |
5500 | ||
5501 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5502 | { | |
5503 | return single_open(file, proc_cgroupstats_show, NULL); | |
5504 | } | |
5505 | ||
5506 | static const struct file_operations proc_cgroupstats_operations = { | |
5507 | .open = cgroupstats_open, | |
5508 | .read = seq_read, | |
5509 | .llseek = seq_lseek, | |
5510 | .release = single_release, | |
5511 | }; | |
5512 | ||
5513 | /** | |
5514 | * cgroup_fork - initialize cgroup related fields during copy_process() | |
5515 | * @child: pointer to task_struct of forking parent process. | |
5516 | * | |
5517 | * A task is associated with the init_css_set until cgroup_post_fork() | |
5518 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5519 | * @child isn't holding reference to its css_set. | |
5520 | */ | |
5521 | void cgroup_fork(struct task_struct *child) | |
5522 | { | |
5523 | RCU_INIT_POINTER(child->cgroups, &init_css_set); | |
5524 | INIT_LIST_HEAD(&child->cg_list); | |
5525 | } | |
5526 | ||
5527 | /** | |
5528 | * cgroup_can_fork - called on a new task before the process is exposed | |
5529 | * @child: the task in question. | |
5530 | * | |
5531 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5532 | * returns an error, the fork aborts with that error code. This allows for | |
5533 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5534 | */ | |
5535 | int cgroup_can_fork(struct task_struct *child) | |
5536 | { | |
5537 | struct cgroup_subsys *ss; | |
5538 | int i, j, ret; | |
5539 | ||
5540 | do_each_subsys_mask(ss, i, have_canfork_callback) { | |
5541 | ret = ss->can_fork(child); | |
5542 | if (ret) | |
5543 | goto out_revert; | |
5544 | } while_each_subsys_mask(); | |
5545 | ||
5546 | return 0; | |
5547 | ||
5548 | out_revert: | |
5549 | for_each_subsys(ss, j) { | |
5550 | if (j >= i) | |
5551 | break; | |
5552 | if (ss->cancel_fork) | |
5553 | ss->cancel_fork(child); | |
5554 | } | |
5555 | ||
5556 | return ret; | |
5557 | } | |
5558 | ||
5559 | /** | |
5560 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5561 | * @child: the task in question | |
5562 | * | |
5563 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5564 | * cgroup_can_fork() succeded. | |
5565 | */ | |
5566 | void cgroup_cancel_fork(struct task_struct *child) | |
5567 | { | |
5568 | struct cgroup_subsys *ss; | |
5569 | int i; | |
5570 | ||
5571 | for_each_subsys(ss, i) | |
5572 | if (ss->cancel_fork) | |
5573 | ss->cancel_fork(child); | |
5574 | } | |
5575 | ||
5576 | /** | |
5577 | * cgroup_post_fork - called on a new task after adding it to the task list | |
5578 | * @child: the task in question | |
5579 | * | |
5580 | * Adds the task to the list running through its css_set if necessary and | |
5581 | * call the subsystem fork() callbacks. Has to be after the task is | |
5582 | * visible on the task list in case we race with the first call to | |
5583 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its | |
5584 | * list. | |
5585 | */ | |
5586 | void cgroup_post_fork(struct task_struct *child) | |
5587 | { | |
5588 | struct cgroup_subsys *ss; | |
5589 | int i; | |
5590 | ||
5591 | /* | |
5592 | * This may race against cgroup_enable_task_cg_lists(). As that | |
5593 | * function sets use_task_css_set_links before grabbing | |
5594 | * tasklist_lock and we just went through tasklist_lock to add | |
5595 | * @child, it's guaranteed that either we see the set | |
5596 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5597 | * @child during its iteration. | |
5598 | * | |
5599 | * If we won the race, @child is associated with %current's | |
5600 | * css_set. Grabbing css_set_lock guarantees both that the | |
5601 | * association is stable, and, on completion of the parent's | |
5602 | * migration, @child is visible in the source of migration or | |
5603 | * already in the destination cgroup. This guarantee is necessary | |
5604 | * when implementing operations which need to migrate all tasks of | |
5605 | * a cgroup to another. | |
5606 | * | |
5607 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child | |
5608 | * will remain in init_css_set. This is safe because all tasks are | |
5609 | * in the init_css_set before cg_links is enabled and there's no | |
5610 | * operation which transfers all tasks out of init_css_set. | |
5611 | */ | |
5612 | if (use_task_css_set_links) { | |
5613 | struct css_set *cset; | |
5614 | ||
5615 | spin_lock_bh(&css_set_lock); | |
5616 | cset = task_css_set(current); | |
5617 | if (list_empty(&child->cg_list)) { | |
5618 | get_css_set(cset); | |
5619 | css_set_move_task(child, NULL, cset, false); | |
5620 | } | |
5621 | spin_unlock_bh(&css_set_lock); | |
5622 | } | |
5623 | ||
5624 | /* | |
5625 | * Call ss->fork(). This must happen after @child is linked on | |
5626 | * css_set; otherwise, @child might change state between ->fork() | |
5627 | * and addition to css_set. | |
5628 | */ | |
5629 | do_each_subsys_mask(ss, i, have_fork_callback) { | |
5630 | ss->fork(child); | |
5631 | } while_each_subsys_mask(); | |
5632 | } | |
5633 | ||
5634 | /** | |
5635 | * cgroup_exit - detach cgroup from exiting task | |
5636 | * @tsk: pointer to task_struct of exiting process | |
5637 | * | |
5638 | * Description: Detach cgroup from @tsk and release it. | |
5639 | * | |
5640 | * Note that cgroups marked notify_on_release force every task in | |
5641 | * them to take the global cgroup_mutex mutex when exiting. | |
5642 | * This could impact scaling on very large systems. Be reluctant to | |
5643 | * use notify_on_release cgroups where very high task exit scaling | |
5644 | * is required on large systems. | |
5645 | * | |
5646 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We | |
5647 | * call cgroup_exit() while the task is still competent to handle | |
5648 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5649 | * each hierarchy for the remainder of its exit. No need to bother with | |
5650 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
5651 | * with migration path - PF_EXITING is visible to migration path. | |
5652 | */ | |
5653 | void cgroup_exit(struct task_struct *tsk) | |
5654 | { | |
5655 | struct cgroup_subsys *ss; | |
5656 | struct css_set *cset; | |
5657 | int i; | |
5658 | ||
5659 | /* | |
5660 | * Unlink from @tsk from its css_set. As migration path can't race | |
5661 | * with us, we can check css_set and cg_list without synchronization. | |
5662 | */ | |
5663 | cset = task_css_set(tsk); | |
5664 | ||
5665 | if (!list_empty(&tsk->cg_list)) { | |
5666 | spin_lock_bh(&css_set_lock); | |
5667 | css_set_move_task(tsk, cset, NULL, false); | |
5668 | spin_unlock_bh(&css_set_lock); | |
5669 | } else { | |
5670 | get_css_set(cset); | |
5671 | } | |
5672 | ||
5673 | /* see cgroup_post_fork() for details */ | |
5674 | do_each_subsys_mask(ss, i, have_exit_callback) { | |
5675 | ss->exit(tsk); | |
5676 | } while_each_subsys_mask(); | |
5677 | } | |
5678 | ||
5679 | void cgroup_free(struct task_struct *task) | |
5680 | { | |
5681 | struct css_set *cset = task_css_set(task); | |
5682 | struct cgroup_subsys *ss; | |
5683 | int ssid; | |
5684 | ||
5685 | do_each_subsys_mask(ss, ssid, have_free_callback) { | |
5686 | ss->free(task); | |
5687 | } while_each_subsys_mask(); | |
5688 | ||
5689 | put_css_set(cset); | |
5690 | } | |
5691 | ||
5692 | static void check_for_release(struct cgroup *cgrp) | |
5693 | { | |
5694 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
5695 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
5696 | schedule_work(&cgrp->release_agent_work); | |
5697 | } | |
5698 | ||
5699 | /* | |
5700 | * Notify userspace when a cgroup is released, by running the | |
5701 | * configured release agent with the name of the cgroup (path | |
5702 | * relative to the root of cgroup file system) as the argument. | |
5703 | * | |
5704 | * Most likely, this user command will try to rmdir this cgroup. | |
5705 | * | |
5706 | * This races with the possibility that some other task will be | |
5707 | * attached to this cgroup before it is removed, or that some other | |
5708 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5709 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5710 | * unused, and this cgroup will be reprieved from its death sentence, | |
5711 | * to continue to serve a useful existence. Next time it's released, | |
5712 | * we will get notified again, if it still has 'notify_on_release' set. | |
5713 | * | |
5714 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5715 | * means only wait until the task is successfully execve()'d. The | |
5716 | * separate release agent task is forked by call_usermodehelper(), | |
5717 | * then control in this thread returns here, without waiting for the | |
5718 | * release agent task. We don't bother to wait because the caller of | |
5719 | * this routine has no use for the exit status of the release agent | |
5720 | * task, so no sense holding our caller up for that. | |
5721 | */ | |
5722 | static void cgroup_release_agent(struct work_struct *work) | |
5723 | { | |
5724 | struct cgroup *cgrp = | |
5725 | container_of(work, struct cgroup, release_agent_work); | |
5726 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5727 | char *argv[3], *envp[3]; | |
5728 | ||
5729 | mutex_lock(&cgroup_mutex); | |
5730 | ||
5731 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5732 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5733 | if (!pathbuf || !agentbuf) | |
5734 | goto out; | |
5735 | ||
5736 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5737 | if (!path) | |
5738 | goto out; | |
5739 | ||
5740 | argv[0] = agentbuf; | |
5741 | argv[1] = path; | |
5742 | argv[2] = NULL; | |
5743 | ||
5744 | /* minimal command environment */ | |
5745 | envp[0] = "HOME=/"; | |
5746 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5747 | envp[2] = NULL; | |
5748 | ||
5749 | mutex_unlock(&cgroup_mutex); | |
5750 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
5751 | goto out_free; | |
5752 | out: | |
5753 | mutex_unlock(&cgroup_mutex); | |
5754 | out_free: | |
5755 | kfree(agentbuf); | |
5756 | kfree(pathbuf); | |
5757 | } | |
5758 | ||
5759 | static int __init cgroup_disable(char *str) | |
5760 | { | |
5761 | struct cgroup_subsys *ss; | |
5762 | char *token; | |
5763 | int i; | |
5764 | ||
5765 | while ((token = strsep(&str, ",")) != NULL) { | |
5766 | if (!*token) | |
5767 | continue; | |
5768 | ||
5769 | for_each_subsys(ss, i) { | |
5770 | if (strcmp(token, ss->name) && | |
5771 | strcmp(token, ss->legacy_name)) | |
5772 | continue; | |
5773 | cgroup_disable_mask |= 1 << i; | |
5774 | } | |
5775 | } | |
5776 | return 1; | |
5777 | } | |
5778 | __setup("cgroup_disable=", cgroup_disable); | |
5779 | ||
5780 | static int __init cgroup_no_v1(char *str) | |
5781 | { | |
5782 | struct cgroup_subsys *ss; | |
5783 | char *token; | |
5784 | int i; | |
5785 | ||
5786 | while ((token = strsep(&str, ",")) != NULL) { | |
5787 | if (!*token) | |
5788 | continue; | |
5789 | ||
5790 | if (!strcmp(token, "all")) { | |
5791 | cgroup_no_v1_mask = U16_MAX; | |
5792 | break; | |
5793 | } | |
5794 | ||
5795 | for_each_subsys(ss, i) { | |
5796 | if (strcmp(token, ss->name) && | |
5797 | strcmp(token, ss->legacy_name)) | |
5798 | continue; | |
5799 | ||
5800 | cgroup_no_v1_mask |= 1 << i; | |
5801 | } | |
5802 | } | |
5803 | return 1; | |
5804 | } | |
5805 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
5806 | ||
5807 | /** | |
5808 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry | |
5809 | * @dentry: directory dentry of interest | |
5810 | * @ss: subsystem of interest | |
5811 | * | |
5812 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try | |
5813 | * to get the corresponding css and return it. If such css doesn't exist | |
5814 | * or can't be pinned, an ERR_PTR value is returned. | |
5815 | */ | |
5816 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, | |
5817 | struct cgroup_subsys *ss) | |
5818 | { | |
5819 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
5820 | struct file_system_type *s_type = dentry->d_sb->s_type; | |
5821 | struct cgroup_subsys_state *css = NULL; | |
5822 | struct cgroup *cgrp; | |
5823 | ||
5824 | /* is @dentry a cgroup dir? */ | |
5825 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || | |
5826 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
5827 | return ERR_PTR(-EBADF); | |
5828 | ||
5829 | rcu_read_lock(); | |
5830 | ||
5831 | /* | |
5832 | * This path doesn't originate from kernfs and @kn could already | |
5833 | * have been or be removed at any point. @kn->priv is RCU | |
5834 | * protected for this access. See css_release_work_fn() for details. | |
5835 | */ | |
5836 | cgrp = rcu_dereference(kn->priv); | |
5837 | if (cgrp) | |
5838 | css = cgroup_css(cgrp, ss); | |
5839 | ||
5840 | if (!css || !css_tryget_online(css)) | |
5841 | css = ERR_PTR(-ENOENT); | |
5842 | ||
5843 | rcu_read_unlock(); | |
5844 | return css; | |
5845 | } | |
5846 | ||
5847 | /** | |
5848 | * css_from_id - lookup css by id | |
5849 | * @id: the cgroup id | |
5850 | * @ss: cgroup subsys to be looked into | |
5851 | * | |
5852 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5853 | * Should be called under rcu_read_lock(). | |
5854 | */ | |
5855 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5856 | { | |
5857 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5858 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; | |
5859 | } | |
5860 | ||
5861 | /** | |
5862 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5863 | * @path: path on the default hierarchy | |
5864 | * | |
5865 | * Find the cgroup at @path on the default hierarchy, increment its | |
5866 | * reference count and return it. Returns pointer to the found cgroup on | |
5867 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5868 | * if @path points to a non-directory. | |
5869 | */ | |
5870 | struct cgroup *cgroup_get_from_path(const char *path) | |
5871 | { | |
5872 | struct kernfs_node *kn; | |
5873 | struct cgroup *cgrp; | |
5874 | ||
5875 | mutex_lock(&cgroup_mutex); | |
5876 | ||
5877 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5878 | if (kn) { | |
5879 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5880 | cgrp = kn->priv; | |
5881 | cgroup_get(cgrp); | |
5882 | } else { | |
5883 | cgrp = ERR_PTR(-ENOTDIR); | |
5884 | } | |
5885 | kernfs_put(kn); | |
5886 | } else { | |
5887 | cgrp = ERR_PTR(-ENOENT); | |
5888 | } | |
5889 | ||
5890 | mutex_unlock(&cgroup_mutex); | |
5891 | return cgrp; | |
5892 | } | |
5893 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5894 | ||
5895 | /* | |
5896 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5897 | * definition in cgroup-defs.h. | |
5898 | */ | |
5899 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5900 | ||
5901 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5902 | ||
5903 | DEFINE_SPINLOCK(cgroup_sk_update_lock); | |
5904 | static bool cgroup_sk_alloc_disabled __read_mostly; | |
5905 | ||
5906 | void cgroup_sk_alloc_disable(void) | |
5907 | { | |
5908 | if (cgroup_sk_alloc_disabled) | |
5909 | return; | |
5910 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5911 | cgroup_sk_alloc_disabled = true; | |
5912 | } | |
5913 | ||
5914 | #else | |
5915 | ||
5916 | #define cgroup_sk_alloc_disabled false | |
5917 | ||
5918 | #endif | |
5919 | ||
5920 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5921 | { | |
5922 | if (cgroup_sk_alloc_disabled) | |
5923 | return; | |
5924 | ||
5925 | rcu_read_lock(); | |
5926 | ||
5927 | while (true) { | |
5928 | struct css_set *cset; | |
5929 | ||
5930 | cset = task_css_set(current); | |
5931 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5932 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5933 | break; | |
5934 | } | |
5935 | cpu_relax(); | |
5936 | } | |
5937 | ||
5938 | rcu_read_unlock(); | |
5939 | } | |
5940 | ||
5941 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5942 | { | |
5943 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5944 | } | |
5945 | ||
5946 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5947 | ||
5948 | #ifdef CONFIG_CGROUP_DEBUG | |
5949 | static struct cgroup_subsys_state * | |
5950 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
5951 | { | |
5952 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5953 | ||
5954 | if (!css) | |
5955 | return ERR_PTR(-ENOMEM); | |
5956 | ||
5957 | return css; | |
5958 | } | |
5959 | ||
5960 | static void debug_css_free(struct cgroup_subsys_state *css) | |
5961 | { | |
5962 | kfree(css); | |
5963 | } | |
5964 | ||
5965 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, | |
5966 | struct cftype *cft) | |
5967 | { | |
5968 | return cgroup_task_count(css->cgroup); | |
5969 | } | |
5970 | ||
5971 | static u64 current_css_set_read(struct cgroup_subsys_state *css, | |
5972 | struct cftype *cft) | |
5973 | { | |
5974 | return (u64)(unsigned long)current->cgroups; | |
5975 | } | |
5976 | ||
5977 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, | |
5978 | struct cftype *cft) | |
5979 | { | |
5980 | u64 count; | |
5981 | ||
5982 | rcu_read_lock(); | |
5983 | count = atomic_read(&task_css_set(current)->refcount); | |
5984 | rcu_read_unlock(); | |
5985 | return count; | |
5986 | } | |
5987 | ||
5988 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) | |
5989 | { | |
5990 | struct cgrp_cset_link *link; | |
5991 | struct css_set *cset; | |
5992 | char *name_buf; | |
5993 | ||
5994 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5995 | if (!name_buf) | |
5996 | return -ENOMEM; | |
5997 | ||
5998 | spin_lock_bh(&css_set_lock); | |
5999 | rcu_read_lock(); | |
6000 | cset = rcu_dereference(current->cgroups); | |
6001 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
6002 | struct cgroup *c = link->cgrp; | |
6003 | ||
6004 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
6005 | seq_printf(seq, "Root %d group %s\n", | |
6006 | c->root->hierarchy_id, name_buf); | |
6007 | } | |
6008 | rcu_read_unlock(); | |
6009 | spin_unlock_bh(&css_set_lock); | |
6010 | kfree(name_buf); | |
6011 | return 0; | |
6012 | } | |
6013 | ||
6014 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
6015 | static int cgroup_css_links_read(struct seq_file *seq, void *v) | |
6016 | { | |
6017 | struct cgroup_subsys_state *css = seq_css(seq); | |
6018 | struct cgrp_cset_link *link; | |
6019 | ||
6020 | spin_lock_bh(&css_set_lock); | |
6021 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { | |
6022 | struct css_set *cset = link->cset; | |
6023 | struct task_struct *task; | |
6024 | int count = 0; | |
6025 | ||
6026 | seq_printf(seq, "css_set %p\n", cset); | |
6027 | ||
6028 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
6029 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6030 | goto overflow; | |
6031 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6032 | } | |
6033 | ||
6034 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6035 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6036 | goto overflow; | |
6037 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6038 | } | |
6039 | continue; | |
6040 | overflow: | |
6041 | seq_puts(seq, " ...\n"); | |
6042 | } | |
6043 | spin_unlock_bh(&css_set_lock); | |
6044 | return 0; | |
6045 | } | |
6046 | ||
6047 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) | |
6048 | { | |
6049 | return (!cgroup_is_populated(css->cgroup) && | |
6050 | !css_has_online_children(&css->cgroup->self)); | |
6051 | } | |
6052 | ||
6053 | static struct cftype debug_files[] = { | |
6054 | { | |
6055 | .name = "taskcount", | |
6056 | .read_u64 = debug_taskcount_read, | |
6057 | }, | |
6058 | ||
6059 | { | |
6060 | .name = "current_css_set", | |
6061 | .read_u64 = current_css_set_read, | |
6062 | }, | |
6063 | ||
6064 | { | |
6065 | .name = "current_css_set_refcount", | |
6066 | .read_u64 = current_css_set_refcount_read, | |
6067 | }, | |
6068 | ||
6069 | { | |
6070 | .name = "current_css_set_cg_links", | |
6071 | .seq_show = current_css_set_cg_links_read, | |
6072 | }, | |
6073 | ||
6074 | { | |
6075 | .name = "cgroup_css_links", | |
6076 | .seq_show = cgroup_css_links_read, | |
6077 | }, | |
6078 | ||
6079 | { | |
6080 | .name = "releasable", | |
6081 | .read_u64 = releasable_read, | |
6082 | }, | |
6083 | ||
6084 | { } /* terminate */ | |
6085 | }; | |
6086 | ||
6087 | struct cgroup_subsys debug_cgrp_subsys = { | |
6088 | .css_alloc = debug_css_alloc, | |
6089 | .css_free = debug_css_free, | |
6090 | .legacy_cftypes = debug_files, | |
6091 | }; | |
6092 | #endif /* CONFIG_CGROUP_DEBUG */ |