]>
Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
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 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
776f02fa | 56 | #include <linux/delay.h> |
846c7bb0 | 57 | |
60063497 | 58 | #include <linux/atomic.h> |
ddbcc7e8 | 59 | |
b1a21367 TH |
60 | /* |
61 | * pidlists linger the following amount before being destroyed. The goal | |
62 | * is avoiding frequent destruction in the middle of consecutive read calls | |
63 | * Expiring in the middle is a performance problem not a correctness one. | |
64 | * 1 sec should be enough. | |
65 | */ | |
66 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
67 | ||
8d7e6fb0 TH |
68 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
69 | MAX_CFTYPE_NAME + 2) | |
70 | ||
ace2bee8 TH |
71 | /* |
72 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
73 | * creation/removal and hierarchy changing operations including cgroup | |
74 | * creation, removal, css association and controller rebinding. This outer | |
75 | * lock is needed mainly to resolve the circular dependency between kernfs | |
76 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
77 | */ | |
78 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
79 | ||
e25e2cbb TH |
80 | /* |
81 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
82 | * hierarchy must be performed while holding it. | |
83 | * | |
0e1d768f TH |
84 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
85 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 86 | * |
0e1d768f TH |
87 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
88 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 89 | */ |
2219449a TH |
90 | #ifdef CONFIG_PROVE_RCU |
91 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
92 | DECLARE_RWSEM(css_set_rwsem); |
93 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
94 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 95 | #else |
81a6a5cd | 96 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 97 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
98 | #endif |
99 | ||
69e943b7 TH |
100 | /* |
101 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
102 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
103 | */ | |
104 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 105 | |
ace2bee8 | 106 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 107 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 108 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 109 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 110 | "cgroup_[tree_]mutex or RCU read lock required"); |
780cd8b3 | 111 | |
e5fca243 TH |
112 | /* |
113 | * cgroup destruction makes heavy use of work items and there can be a lot | |
114 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
115 | * destruction work items don't end up filling up max_active of system_wq | |
116 | * which may lead to deadlock. | |
117 | */ | |
118 | static struct workqueue_struct *cgroup_destroy_wq; | |
119 | ||
b1a21367 TH |
120 | /* |
121 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
122 | * separate workqueue as flush domain. | |
123 | */ | |
124 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
125 | ||
3ed80a62 | 126 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 127 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 128 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
129 | #include <linux/cgroup_subsys.h> |
130 | }; | |
073219e9 TH |
131 | #undef SUBSYS |
132 | ||
133 | /* array of cgroup subsystem names */ | |
134 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
135 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
136 | #include <linux/cgroup_subsys.h> |
137 | }; | |
073219e9 | 138 | #undef SUBSYS |
ddbcc7e8 | 139 | |
ddbcc7e8 | 140 | /* |
3dd06ffa | 141 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
142 | * unattached - it never has more than a single cgroup, and all tasks are |
143 | * part of that cgroup. | |
ddbcc7e8 | 144 | */ |
a2dd4247 | 145 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 146 | |
a2dd4247 TH |
147 | /* |
148 | * The default hierarchy always exists but is hidden until mounted for the | |
149 | * first time. This is for backward compatibility. | |
150 | */ | |
151 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
152 | |
153 | /* The list of hierarchy roots */ | |
154 | ||
9871bf95 TH |
155 | static LIST_HEAD(cgroup_roots); |
156 | static int cgroup_root_count; | |
ddbcc7e8 | 157 | |
3417ae1f | 158 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 159 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 160 | |
794611a1 LZ |
161 | /* |
162 | * Assign a monotonically increasing serial number to cgroups. It | |
163 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
164 | * numbers. Also, as cgroups are always appended to the parent's | |
165 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
166 | * the ascending serial number order on the list. Protected by |
167 | * cgroup_mutex. | |
794611a1 | 168 | */ |
00356bd5 | 169 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 170 | |
ddbcc7e8 | 171 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
172 | * check for fork/exit handlers to call. This avoids us having to do |
173 | * extra work in the fork/exit path if none of the subsystems need to | |
174 | * be called. | |
ddbcc7e8 | 175 | */ |
8947f9d5 | 176 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 177 | |
628f7cd4 TH |
178 | static struct cftype cgroup_base_files[]; |
179 | ||
59f5296b | 180 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 181 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 182 | unsigned long ss_mask); |
f20104de | 183 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 184 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f8f22e53 TH |
185 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss); |
186 | static void kill_css(struct cgroup_subsys_state *css); | |
2bb566cb TH |
187 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
188 | bool is_add); | |
b1a21367 | 189 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 190 | |
95109b62 TH |
191 | /** |
192 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
193 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 194 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 195 | * |
ca8bdcaf TH |
196 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
197 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
198 | * the caller is responsible for pinning the returned css if it wants to | |
199 | * keep accessing it outside the said locks. This function may return | |
200 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
201 | */ |
202 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 203 | struct cgroup_subsys *ss) |
95109b62 | 204 | { |
ca8bdcaf | 205 | if (ss) |
aec25020 | 206 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
207 | lockdep_is_held(&cgroup_tree_mutex) || |
208 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
209 | else |
210 | return &cgrp->dummy_css; | |
95109b62 | 211 | } |
42809dd4 | 212 | |
aec3dfcb TH |
213 | /** |
214 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
215 | * @cgrp: the cgroup of interest | |
216 | * @ss: the subsystem of interest (%NULL returns the dummy_css) | |
217 | * | |
218 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
219 | * as the matching css of the nearest ancestor including self which has @ss | |
220 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
221 | * function is guaranteed to return non-NULL css. | |
222 | */ | |
223 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
224 | struct cgroup_subsys *ss) | |
225 | { | |
226 | lockdep_assert_held(&cgroup_mutex); | |
227 | ||
228 | if (!ss) | |
229 | return &cgrp->dummy_css; | |
230 | ||
231 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
232 | return NULL; | |
233 | ||
234 | while (cgrp->parent && | |
235 | !(cgrp->parent->child_subsys_mask & (1 << ss->id))) | |
236 | cgrp = cgrp->parent; | |
237 | ||
238 | return cgroup_css(cgrp, ss); | |
239 | } | |
240 | ||
ddbcc7e8 | 241 | /* convenient tests for these bits */ |
54766d4a | 242 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 243 | { |
54766d4a | 244 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
245 | } |
246 | ||
59f5296b TH |
247 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
248 | { | |
2bd59d48 TH |
249 | struct kernfs_open_file *of = seq->private; |
250 | struct cgroup *cgrp = of->kn->parent->priv; | |
251 | struct cftype *cft = seq_cft(seq); | |
252 | ||
253 | /* | |
254 | * This is open and unprotected implementation of cgroup_css(). | |
255 | * seq_css() is only called from a kernfs file operation which has | |
256 | * an active reference on the file. Because all the subsystem | |
257 | * files are drained before a css is disassociated with a cgroup, | |
258 | * the matching css from the cgroup's subsys table is guaranteed to | |
259 | * be and stay valid until the enclosing operation is complete. | |
260 | */ | |
261 | if (cft->ss) | |
262 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
263 | else | |
264 | return &cgrp->dummy_css; | |
59f5296b TH |
265 | } |
266 | EXPORT_SYMBOL_GPL(seq_css); | |
267 | ||
78574cf9 LZ |
268 | /** |
269 | * cgroup_is_descendant - test ancestry | |
270 | * @cgrp: the cgroup to be tested | |
271 | * @ancestor: possible ancestor of @cgrp | |
272 | * | |
273 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
274 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
275 | * and @ancestor are accessible. | |
276 | */ | |
277 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
278 | { | |
279 | while (cgrp) { | |
280 | if (cgrp == ancestor) | |
281 | return true; | |
282 | cgrp = cgrp->parent; | |
283 | } | |
284 | return false; | |
285 | } | |
ddbcc7e8 | 286 | |
e9685a03 | 287 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
288 | { |
289 | const int bits = | |
bd89aabc PM |
290 | (1 << CGRP_RELEASABLE) | |
291 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
292 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
293 | } |
294 | ||
e9685a03 | 295 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 296 | { |
bd89aabc | 297 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
298 | } |
299 | ||
1c6727af TH |
300 | /** |
301 | * for_each_css - iterate all css's of a cgroup | |
302 | * @css: the iteration cursor | |
303 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
304 | * @cgrp: the target cgroup to iterate css's of | |
305 | * | |
aec3dfcb | 306 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
307 | */ |
308 | #define for_each_css(css, ssid, cgrp) \ | |
309 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
310 | if (!((css) = rcu_dereference_check( \ | |
311 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 312 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
313 | lockdep_is_held(&cgroup_mutex)))) { } \ |
314 | else | |
315 | ||
aec3dfcb TH |
316 | /** |
317 | * for_each_e_css - iterate all effective css's of a cgroup | |
318 | * @css: the iteration cursor | |
319 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
320 | * @cgrp: the target cgroup to iterate css's of | |
321 | * | |
322 | * Should be called under cgroup_[tree_]mutex. | |
323 | */ | |
324 | #define for_each_e_css(css, ssid, cgrp) \ | |
325 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
326 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
327 | ; \ | |
328 | else | |
329 | ||
30159ec7 | 330 | /** |
3ed80a62 | 331 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 332 | * @ss: the iteration cursor |
780cd8b3 | 333 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 334 | */ |
780cd8b3 | 335 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
336 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
337 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 338 | |
985ed670 TH |
339 | /* iterate across the hierarchies */ |
340 | #define for_each_root(root) \ | |
5549c497 | 341 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 342 | |
f8f22e53 TH |
343 | /* iterate over child cgrps, lock should be held throughout iteration */ |
344 | #define cgroup_for_each_live_child(child, cgrp) \ | |
345 | list_for_each_entry((child), &(cgrp)->children, sibling) \ | |
346 | if (({ lockdep_assert_held(&cgroup_tree_mutex); \ | |
347 | cgroup_is_dead(child); })) \ | |
348 | ; \ | |
349 | else | |
350 | ||
7ae1bad9 TH |
351 | /** |
352 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
353 | * @cgrp: the cgroup to be checked for liveness | |
354 | * | |
47cfcd09 TH |
355 | * On success, returns true; the mutex should be later unlocked. On |
356 | * failure returns false with no lock held. | |
7ae1bad9 | 357 | */ |
b9777cf8 | 358 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
359 | { |
360 | mutex_lock(&cgroup_mutex); | |
54766d4a | 361 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
362 | mutex_unlock(&cgroup_mutex); |
363 | return false; | |
364 | } | |
365 | return true; | |
366 | } | |
7ae1bad9 | 367 | |
81a6a5cd PM |
368 | /* the list of cgroups eligible for automatic release. Protected by |
369 | * release_list_lock */ | |
370 | static LIST_HEAD(release_list); | |
cdcc136f | 371 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
372 | static void cgroup_release_agent(struct work_struct *work); |
373 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 374 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 375 | |
69d0206c TH |
376 | /* |
377 | * A cgroup can be associated with multiple css_sets as different tasks may | |
378 | * belong to different cgroups on different hierarchies. In the other | |
379 | * direction, a css_set is naturally associated with multiple cgroups. | |
380 | * This M:N relationship is represented by the following link structure | |
381 | * which exists for each association and allows traversing the associations | |
382 | * from both sides. | |
383 | */ | |
384 | struct cgrp_cset_link { | |
385 | /* the cgroup and css_set this link associates */ | |
386 | struct cgroup *cgrp; | |
387 | struct css_set *cset; | |
388 | ||
389 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
390 | struct list_head cset_link; | |
391 | ||
392 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
393 | struct list_head cgrp_link; | |
817929ec PM |
394 | }; |
395 | ||
172a2c06 TH |
396 | /* |
397 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
398 | * hierarchies being mounted. It contains a pointer to the root state |
399 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
400 | * reference-counted, to improve performance when child cgroups | |
401 | * haven't been created. | |
402 | */ | |
172a2c06 TH |
403 | static struct css_set init_css_set = { |
404 | .refcount = ATOMIC_INIT(1), | |
405 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
406 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
407 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
408 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
409 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
410 | }; | |
817929ec | 411 | |
172a2c06 | 412 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 413 | |
842b597e TH |
414 | /** |
415 | * cgroup_update_populated - updated populated count of a cgroup | |
416 | * @cgrp: the target cgroup | |
417 | * @populated: inc or dec populated count | |
418 | * | |
419 | * @cgrp is either getting the first task (css_set) or losing the last. | |
420 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
421 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
422 | * cgroup and all its descendants are empty. | |
423 | * | |
424 | * @cgrp's interface file "cgroup.populated" is zero if | |
425 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
426 | * changes from or to zero, userland is notified that the content of the | |
427 | * interface file has changed. This can be used to detect when @cgrp and | |
428 | * its descendants become populated or empty. | |
429 | */ | |
430 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
431 | { | |
432 | lockdep_assert_held(&css_set_rwsem); | |
433 | ||
434 | do { | |
435 | bool trigger; | |
436 | ||
437 | if (populated) | |
438 | trigger = !cgrp->populated_cnt++; | |
439 | else | |
440 | trigger = !--cgrp->populated_cnt; | |
441 | ||
442 | if (!trigger) | |
443 | break; | |
444 | ||
445 | if (cgrp->populated_kn) | |
446 | kernfs_notify(cgrp->populated_kn); | |
447 | cgrp = cgrp->parent; | |
448 | } while (cgrp); | |
449 | } | |
450 | ||
7717f7ba PM |
451 | /* |
452 | * hash table for cgroup groups. This improves the performance to find | |
453 | * an existing css_set. This hash doesn't (currently) take into | |
454 | * account cgroups in empty hierarchies. | |
455 | */ | |
472b1053 | 456 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 457 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 458 | |
0ac801fe | 459 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 460 | { |
0ac801fe | 461 | unsigned long key = 0UL; |
30159ec7 TH |
462 | struct cgroup_subsys *ss; |
463 | int i; | |
472b1053 | 464 | |
30159ec7 | 465 | for_each_subsys(ss, i) |
0ac801fe LZ |
466 | key += (unsigned long)css[i]; |
467 | key = (key >> 16) ^ key; | |
472b1053 | 468 | |
0ac801fe | 469 | return key; |
472b1053 LZ |
470 | } |
471 | ||
89c5509b | 472 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 473 | { |
69d0206c | 474 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
475 | struct cgroup_subsys *ss; |
476 | int ssid; | |
5abb8855 | 477 | |
89c5509b TH |
478 | lockdep_assert_held(&css_set_rwsem); |
479 | ||
480 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 481 | return; |
81a6a5cd | 482 | |
2c6ab6d2 | 483 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
484 | for_each_subsys(ss, ssid) |
485 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 486 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
487 | css_set_count--; |
488 | ||
69d0206c | 489 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 490 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 491 | |
69d0206c TH |
492 | list_del(&link->cset_link); |
493 | list_del(&link->cgrp_link); | |
71b5707e | 494 | |
96d365e0 | 495 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
496 | if (list_empty(&cgrp->cset_links)) { |
497 | cgroup_update_populated(cgrp, false); | |
498 | if (notify_on_release(cgrp)) { | |
499 | if (taskexit) | |
500 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
501 | check_for_release(cgrp); | |
502 | } | |
81a6a5cd | 503 | } |
2c6ab6d2 PM |
504 | |
505 | kfree(link); | |
81a6a5cd | 506 | } |
2c6ab6d2 | 507 | |
5abb8855 | 508 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
509 | } |
510 | ||
89c5509b TH |
511 | static void put_css_set(struct css_set *cset, bool taskexit) |
512 | { | |
513 | /* | |
514 | * Ensure that the refcount doesn't hit zero while any readers | |
515 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
516 | * rwlock | |
517 | */ | |
518 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
519 | return; | |
520 | ||
521 | down_write(&css_set_rwsem); | |
522 | put_css_set_locked(cset, taskexit); | |
523 | up_write(&css_set_rwsem); | |
524 | } | |
525 | ||
817929ec PM |
526 | /* |
527 | * refcounted get/put for css_set objects | |
528 | */ | |
5abb8855 | 529 | static inline void get_css_set(struct css_set *cset) |
817929ec | 530 | { |
5abb8855 | 531 | atomic_inc(&cset->refcount); |
817929ec PM |
532 | } |
533 | ||
b326f9d0 | 534 | /** |
7717f7ba | 535 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
536 | * @cset: candidate css_set being tested |
537 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
538 | * @new_cgrp: cgroup that's being entered by the task |
539 | * @template: desired set of css pointers in css_set (pre-calculated) | |
540 | * | |
6f4b7e63 | 541 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
542 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
543 | */ | |
5abb8855 TH |
544 | static bool compare_css_sets(struct css_set *cset, |
545 | struct css_set *old_cset, | |
7717f7ba PM |
546 | struct cgroup *new_cgrp, |
547 | struct cgroup_subsys_state *template[]) | |
548 | { | |
549 | struct list_head *l1, *l2; | |
550 | ||
aec3dfcb TH |
551 | /* |
552 | * On the default hierarchy, there can be csets which are | |
553 | * associated with the same set of cgroups but different csses. | |
554 | * Let's first ensure that csses match. | |
555 | */ | |
556 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 557 | return false; |
7717f7ba PM |
558 | |
559 | /* | |
560 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
561 | * different cgroups in hierarchies. As different cgroups may |
562 | * share the same effective css, this comparison is always | |
563 | * necessary. | |
7717f7ba | 564 | */ |
69d0206c TH |
565 | l1 = &cset->cgrp_links; |
566 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 567 | while (1) { |
69d0206c | 568 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 569 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
570 | |
571 | l1 = l1->next; | |
572 | l2 = l2->next; | |
573 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
574 | if (l1 == &cset->cgrp_links) { |
575 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
576 | break; |
577 | } else { | |
69d0206c | 578 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
579 | } |
580 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
581 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
582 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
583 | cgrp1 = link1->cgrp; | |
584 | cgrp2 = link2->cgrp; | |
7717f7ba | 585 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 586 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
587 | |
588 | /* | |
589 | * If this hierarchy is the hierarchy of the cgroup | |
590 | * that's changing, then we need to check that this | |
591 | * css_set points to the new cgroup; if it's any other | |
592 | * hierarchy, then this css_set should point to the | |
593 | * same cgroup as the old css_set. | |
594 | */ | |
5abb8855 TH |
595 | if (cgrp1->root == new_cgrp->root) { |
596 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
597 | return false; |
598 | } else { | |
5abb8855 | 599 | if (cgrp1 != cgrp2) |
7717f7ba PM |
600 | return false; |
601 | } | |
602 | } | |
603 | return true; | |
604 | } | |
605 | ||
b326f9d0 TH |
606 | /** |
607 | * find_existing_css_set - init css array and find the matching css_set | |
608 | * @old_cset: the css_set that we're using before the cgroup transition | |
609 | * @cgrp: the cgroup that we're moving into | |
610 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 611 | */ |
5abb8855 TH |
612 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
613 | struct cgroup *cgrp, | |
614 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 615 | { |
3dd06ffa | 616 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 617 | struct cgroup_subsys *ss; |
5abb8855 | 618 | struct css_set *cset; |
0ac801fe | 619 | unsigned long key; |
b326f9d0 | 620 | int i; |
817929ec | 621 | |
aae8aab4 BB |
622 | /* |
623 | * Build the set of subsystem state objects that we want to see in the | |
624 | * new css_set. while subsystems can change globally, the entries here | |
625 | * won't change, so no need for locking. | |
626 | */ | |
30159ec7 | 627 | for_each_subsys(ss, i) { |
f392e51c | 628 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
629 | /* |
630 | * @ss is in this hierarchy, so we want the | |
631 | * effective css from @cgrp. | |
632 | */ | |
633 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 634 | } else { |
aec3dfcb TH |
635 | /* |
636 | * @ss is not in this hierarchy, so we don't want | |
637 | * to change the css. | |
638 | */ | |
5abb8855 | 639 | template[i] = old_cset->subsys[i]; |
817929ec PM |
640 | } |
641 | } | |
642 | ||
0ac801fe | 643 | key = css_set_hash(template); |
5abb8855 TH |
644 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
645 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
646 | continue; |
647 | ||
648 | /* This css_set matches what we need */ | |
5abb8855 | 649 | return cset; |
472b1053 | 650 | } |
817929ec PM |
651 | |
652 | /* No existing cgroup group matched */ | |
653 | return NULL; | |
654 | } | |
655 | ||
69d0206c | 656 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 657 | { |
69d0206c | 658 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 659 | |
69d0206c TH |
660 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
661 | list_del(&link->cset_link); | |
36553434 LZ |
662 | kfree(link); |
663 | } | |
664 | } | |
665 | ||
69d0206c TH |
666 | /** |
667 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
668 | * @count: the number of links to allocate | |
669 | * @tmp_links: list_head the allocated links are put on | |
670 | * | |
671 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
672 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 673 | */ |
69d0206c | 674 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 675 | { |
69d0206c | 676 | struct cgrp_cset_link *link; |
817929ec | 677 | int i; |
69d0206c TH |
678 | |
679 | INIT_LIST_HEAD(tmp_links); | |
680 | ||
817929ec | 681 | for (i = 0; i < count; i++) { |
f4f4be2b | 682 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 683 | if (!link) { |
69d0206c | 684 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
685 | return -ENOMEM; |
686 | } | |
69d0206c | 687 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
688 | } |
689 | return 0; | |
690 | } | |
691 | ||
c12f65d4 LZ |
692 | /** |
693 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 694 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 695 | * @cset: the css_set to be linked |
c12f65d4 LZ |
696 | * @cgrp: the destination cgroup |
697 | */ | |
69d0206c TH |
698 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
699 | struct cgroup *cgrp) | |
c12f65d4 | 700 | { |
69d0206c | 701 | struct cgrp_cset_link *link; |
c12f65d4 | 702 | |
69d0206c | 703 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
704 | |
705 | if (cgroup_on_dfl(cgrp)) | |
706 | cset->dfl_cgrp = cgrp; | |
707 | ||
69d0206c TH |
708 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
709 | link->cset = cset; | |
7717f7ba | 710 | link->cgrp = cgrp; |
842b597e TH |
711 | |
712 | if (list_empty(&cgrp->cset_links)) | |
713 | cgroup_update_populated(cgrp, true); | |
69d0206c | 714 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 715 | |
7717f7ba PM |
716 | /* |
717 | * Always add links to the tail of the list so that the list | |
718 | * is sorted by order of hierarchy creation | |
719 | */ | |
69d0206c | 720 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
721 | } |
722 | ||
b326f9d0 TH |
723 | /** |
724 | * find_css_set - return a new css_set with one cgroup updated | |
725 | * @old_cset: the baseline css_set | |
726 | * @cgrp: the cgroup to be updated | |
727 | * | |
728 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
729 | * substituted into the appropriate hierarchy. | |
817929ec | 730 | */ |
5abb8855 TH |
731 | static struct css_set *find_css_set(struct css_set *old_cset, |
732 | struct cgroup *cgrp) | |
817929ec | 733 | { |
b326f9d0 | 734 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 735 | struct css_set *cset; |
69d0206c TH |
736 | struct list_head tmp_links; |
737 | struct cgrp_cset_link *link; | |
2d8f243a | 738 | struct cgroup_subsys *ss; |
0ac801fe | 739 | unsigned long key; |
2d8f243a | 740 | int ssid; |
472b1053 | 741 | |
b326f9d0 TH |
742 | lockdep_assert_held(&cgroup_mutex); |
743 | ||
817929ec PM |
744 | /* First see if we already have a cgroup group that matches |
745 | * the desired set */ | |
96d365e0 | 746 | down_read(&css_set_rwsem); |
5abb8855 TH |
747 | cset = find_existing_css_set(old_cset, cgrp, template); |
748 | if (cset) | |
749 | get_css_set(cset); | |
96d365e0 | 750 | up_read(&css_set_rwsem); |
817929ec | 751 | |
5abb8855 TH |
752 | if (cset) |
753 | return cset; | |
817929ec | 754 | |
f4f4be2b | 755 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 756 | if (!cset) |
817929ec PM |
757 | return NULL; |
758 | ||
69d0206c | 759 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 760 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 761 | kfree(cset); |
817929ec PM |
762 | return NULL; |
763 | } | |
764 | ||
5abb8855 | 765 | atomic_set(&cset->refcount, 1); |
69d0206c | 766 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 767 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 768 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 769 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 770 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 771 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
772 | |
773 | /* Copy the set of subsystem state objects generated in | |
774 | * find_existing_css_set() */ | |
5abb8855 | 775 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 776 | |
96d365e0 | 777 | down_write(&css_set_rwsem); |
817929ec | 778 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 779 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 780 | struct cgroup *c = link->cgrp; |
69d0206c | 781 | |
7717f7ba PM |
782 | if (c->root == cgrp->root) |
783 | c = cgrp; | |
69d0206c | 784 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 785 | } |
817929ec | 786 | |
69d0206c | 787 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 788 | |
817929ec | 789 | css_set_count++; |
472b1053 | 790 | |
2d8f243a | 791 | /* Add @cset to the hash table */ |
5abb8855 TH |
792 | key = css_set_hash(cset->subsys); |
793 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 794 | |
2d8f243a TH |
795 | for_each_subsys(ss, ssid) |
796 | list_add_tail(&cset->e_cset_node[ssid], | |
797 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
798 | ||
96d365e0 | 799 | up_write(&css_set_rwsem); |
817929ec | 800 | |
5abb8855 | 801 | return cset; |
b4f48b63 PM |
802 | } |
803 | ||
3dd06ffa | 804 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 805 | { |
3dd06ffa | 806 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 807 | |
3dd06ffa | 808 | return root_cgrp->root; |
2bd59d48 TH |
809 | } |
810 | ||
3dd06ffa | 811 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
812 | { |
813 | int id; | |
814 | ||
815 | lockdep_assert_held(&cgroup_mutex); | |
816 | ||
985ed670 | 817 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
818 | if (id < 0) |
819 | return id; | |
820 | ||
821 | root->hierarchy_id = id; | |
822 | return 0; | |
823 | } | |
824 | ||
3dd06ffa | 825 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
826 | { |
827 | lockdep_assert_held(&cgroup_mutex); | |
828 | ||
829 | if (root->hierarchy_id) { | |
830 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
831 | root->hierarchy_id = 0; | |
832 | } | |
833 | } | |
834 | ||
3dd06ffa | 835 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
836 | { |
837 | if (root) { | |
838 | /* hierarhcy ID shoulid already have been released */ | |
839 | WARN_ON_ONCE(root->hierarchy_id); | |
840 | ||
841 | idr_destroy(&root->cgroup_idr); | |
842 | kfree(root); | |
843 | } | |
844 | } | |
845 | ||
3dd06ffa | 846 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 847 | { |
3dd06ffa | 848 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 849 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 850 | |
2bd59d48 | 851 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 852 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 853 | |
776f02fa | 854 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
855 | BUG_ON(!list_empty(&cgrp->children)); |
856 | ||
f2e85d57 | 857 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 858 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 859 | |
7717f7ba | 860 | /* |
f2e85d57 TH |
861 | * Release all the links from cset_links to this hierarchy's |
862 | * root cgroup | |
7717f7ba | 863 | */ |
96d365e0 | 864 | down_write(&css_set_rwsem); |
f2e85d57 TH |
865 | |
866 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
867 | list_del(&link->cset_link); | |
868 | list_del(&link->cgrp_link); | |
869 | kfree(link); | |
870 | } | |
96d365e0 | 871 | up_write(&css_set_rwsem); |
f2e85d57 TH |
872 | |
873 | if (!list_empty(&root->root_list)) { | |
874 | list_del(&root->root_list); | |
875 | cgroup_root_count--; | |
876 | } | |
877 | ||
878 | cgroup_exit_root_id(root); | |
879 | ||
880 | mutex_unlock(&cgroup_mutex); | |
881 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 882 | |
2bd59d48 | 883 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
884 | cgroup_free_root(root); |
885 | } | |
886 | ||
ceb6a081 TH |
887 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
888 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 889 | struct cgroup_root *root) |
7717f7ba | 890 | { |
7717f7ba PM |
891 | struct cgroup *res = NULL; |
892 | ||
96d365e0 TH |
893 | lockdep_assert_held(&cgroup_mutex); |
894 | lockdep_assert_held(&css_set_rwsem); | |
895 | ||
5abb8855 | 896 | if (cset == &init_css_set) { |
3dd06ffa | 897 | res = &root->cgrp; |
7717f7ba | 898 | } else { |
69d0206c TH |
899 | struct cgrp_cset_link *link; |
900 | ||
901 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 902 | struct cgroup *c = link->cgrp; |
69d0206c | 903 | |
7717f7ba PM |
904 | if (c->root == root) { |
905 | res = c; | |
906 | break; | |
907 | } | |
908 | } | |
909 | } | |
96d365e0 | 910 | |
7717f7ba PM |
911 | BUG_ON(!res); |
912 | return res; | |
913 | } | |
914 | ||
ddbcc7e8 | 915 | /* |
ceb6a081 TH |
916 | * Return the cgroup for "task" from the given hierarchy. Must be |
917 | * called with cgroup_mutex and css_set_rwsem held. | |
918 | */ | |
919 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 920 | struct cgroup_root *root) |
ceb6a081 TH |
921 | { |
922 | /* | |
923 | * No need to lock the task - since we hold cgroup_mutex the | |
924 | * task can't change groups, so the only thing that can happen | |
925 | * is that it exits and its css is set back to init_css_set. | |
926 | */ | |
927 | return cset_cgroup_from_root(task_css_set(task), root); | |
928 | } | |
929 | ||
ddbcc7e8 | 930 | /* |
ddbcc7e8 PM |
931 | * A task must hold cgroup_mutex to modify cgroups. |
932 | * | |
933 | * Any task can increment and decrement the count field without lock. | |
934 | * So in general, code holding cgroup_mutex can't rely on the count | |
935 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 936 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
937 | * means that no tasks are currently attached, therefore there is no |
938 | * way a task attached to that cgroup can fork (the other way to | |
939 | * increment the count). So code holding cgroup_mutex can safely | |
940 | * assume that if the count is zero, it will stay zero. Similarly, if | |
941 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
942 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
943 | * needs that mutex. | |
944 | * | |
ddbcc7e8 PM |
945 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
946 | * (usually) take cgroup_mutex. These are the two most performance | |
947 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
948 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
949 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
950 | * to the release agent with the name of the cgroup (path relative to |
951 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
952 | * |
953 | * A cgroup can only be deleted if both its 'count' of using tasks | |
954 | * is zero, and its list of 'children' cgroups is empty. Since all | |
955 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 956 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 957 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 958 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
959 | * |
960 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 961 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
962 | */ |
963 | ||
628f7cd4 | 964 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 965 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 966 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 967 | |
8d7e6fb0 TH |
968 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
969 | char *buf) | |
ddbcc7e8 | 970 | { |
8d7e6fb0 TH |
971 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
972 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
973 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
974 | cft->ss->name, cft->name); | |
975 | else | |
976 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
977 | return buf; | |
ddbcc7e8 PM |
978 | } |
979 | ||
f2e85d57 TH |
980 | /** |
981 | * cgroup_file_mode - deduce file mode of a control file | |
982 | * @cft: the control file in question | |
983 | * | |
984 | * returns cft->mode if ->mode is not 0 | |
985 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
986 | * returns S_IRUGO if it has only a read handler | |
987 | * returns S_IWUSR if it has only a write hander | |
988 | */ | |
989 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 990 | { |
f2e85d57 | 991 | umode_t mode = 0; |
65dff759 | 992 | |
f2e85d57 TH |
993 | if (cft->mode) |
994 | return cft->mode; | |
995 | ||
996 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
997 | mode |= S_IRUGO; | |
998 | ||
999 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
1000 | cft->trigger) | |
1001 | mode |= S_IWUSR; | |
1002 | ||
1003 | return mode; | |
65dff759 LZ |
1004 | } |
1005 | ||
be445626 LZ |
1006 | static void cgroup_free_fn(struct work_struct *work) |
1007 | { | |
ea15f8cc | 1008 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 1009 | |
3c9c825b | 1010 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 1011 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 1012 | |
776f02fa TH |
1013 | if (cgrp->parent) { |
1014 | /* | |
1015 | * We get a ref to the parent, and put the ref when this | |
1016 | * cgroup is being freed, so it's guaranteed that the | |
1017 | * parent won't be destroyed before its children. | |
1018 | */ | |
1019 | cgroup_put(cgrp->parent); | |
1020 | kernfs_put(cgrp->kn); | |
1021 | kfree(cgrp); | |
1022 | } else { | |
1023 | /* | |
3dd06ffa | 1024 | * This is root cgroup's refcnt reaching zero, which |
776f02fa TH |
1025 | * indicates that the root should be released. |
1026 | */ | |
1027 | cgroup_destroy_root(cgrp->root); | |
1028 | } | |
be445626 LZ |
1029 | } |
1030 | ||
1031 | static void cgroup_free_rcu(struct rcu_head *head) | |
1032 | { | |
1033 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
1034 | ||
ea15f8cc | 1035 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 1036 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
1037 | } |
1038 | ||
59f5296b | 1039 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 1040 | { |
2bd59d48 TH |
1041 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
1042 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
1043 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
1044 | } |
1045 | ||
59f5296b | 1046 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 1047 | { |
2bd59d48 TH |
1048 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
1049 | return; | |
776f02fa | 1050 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 1051 | return; |
05ef1d7c | 1052 | |
2739d3cc | 1053 | /* |
2bd59d48 TH |
1054 | * XXX: cgrp->id is only used to look up css's. As cgroup and |
1055 | * css's lifetimes will be decoupled, it should be made | |
1056 | * per-subsystem and moved to css->id so that lookups are | |
1057 | * successful until the target css is released. | |
2739d3cc | 1058 | */ |
2bd59d48 TH |
1059 | mutex_lock(&cgroup_mutex); |
1060 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
1061 | mutex_unlock(&cgroup_mutex); | |
1062 | cgrp->id = -1; | |
05ef1d7c | 1063 | |
2bd59d48 | 1064 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 | 1065 | } |
05ef1d7c | 1066 | |
2739d3cc | 1067 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1068 | { |
2bd59d48 | 1069 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1070 | |
ace2bee8 | 1071 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1072 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1073 | } |
1074 | ||
13af07df | 1075 | /** |
628f7cd4 | 1076 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1077 | * @cgrp: target cgroup |
13af07df AR |
1078 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1079 | */ | |
628f7cd4 | 1080 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1081 | { |
13af07df | 1082 | struct cgroup_subsys *ss; |
b420ba7d | 1083 | int i; |
05ef1d7c | 1084 | |
b420ba7d | 1085 | for_each_subsys(ss, i) { |
0adb0704 | 1086 | struct cftype *cfts; |
b420ba7d TH |
1087 | |
1088 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1089 | continue; |
0adb0704 TH |
1090 | list_for_each_entry(cfts, &ss->cfts, node) |
1091 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1092 | } |
ddbcc7e8 PM |
1093 | } |
1094 | ||
3dd06ffa | 1095 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 1096 | unsigned long ss_mask) |
ddbcc7e8 | 1097 | { |
30159ec7 | 1098 | struct cgroup_subsys *ss; |
2d8f243a | 1099 | int ssid, i, ret; |
ddbcc7e8 | 1100 | |
ace2bee8 TH |
1101 | lockdep_assert_held(&cgroup_tree_mutex); |
1102 | lockdep_assert_held(&cgroup_mutex); | |
ddbcc7e8 | 1103 | |
5df36032 TH |
1104 | for_each_subsys(ss, ssid) { |
1105 | if (!(ss_mask & (1 << ssid))) | |
1106 | continue; | |
aae8aab4 | 1107 | |
7fd8c565 TH |
1108 | /* if @ss has non-root csses attached to it, can't move */ |
1109 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1110 | return -EBUSY; |
1d5be6b2 | 1111 | |
5df36032 | 1112 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1113 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1114 | return -EBUSY; |
ddbcc7e8 PM |
1115 | } |
1116 | ||
a2dd4247 TH |
1117 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1118 | if (ret) { | |
1119 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1120 | return ret; |
ddbcc7e8 | 1121 | |
a2dd4247 TH |
1122 | /* |
1123 | * Rebinding back to the default root is not allowed to | |
1124 | * fail. Using both default and non-default roots should | |
1125 | * be rare. Moving subsystems back and forth even more so. | |
1126 | * Just warn about it and continue. | |
1127 | */ | |
1128 | if (cgrp_dfl_root_visible) { | |
a2a1f9ea JZ |
1129 | pr_warn("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n", |
1130 | ret, ss_mask); | |
1131 | pr_warn("cgroup: you may retry by moving them to a different hierarchy and unbinding\n"); | |
a2dd4247 | 1132 | } |
5df36032 | 1133 | } |
3126121f TH |
1134 | |
1135 | /* | |
1136 | * Nothing can fail from this point on. Remove files for the | |
1137 | * removed subsystems and rebind each subsystem. | |
1138 | */ | |
4ac06017 | 1139 | mutex_unlock(&cgroup_mutex); |
5df36032 | 1140 | for_each_subsys(ss, ssid) |
a2dd4247 | 1141 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1142 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
4ac06017 | 1143 | mutex_lock(&cgroup_mutex); |
a8a648c4 | 1144 | |
5df36032 | 1145 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1146 | struct cgroup_root *src_root; |
5df36032 | 1147 | struct cgroup_subsys_state *css; |
2d8f243a | 1148 | struct css_set *cset; |
a8a648c4 | 1149 | |
5df36032 TH |
1150 | if (!(ss_mask & (1 << ssid))) |
1151 | continue; | |
a8a648c4 | 1152 | |
5df36032 | 1153 | src_root = ss->root; |
3dd06ffa | 1154 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1155 | |
3dd06ffa | 1156 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1157 | |
3dd06ffa TH |
1158 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1159 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1160 | ss->root = dst_root; |
3dd06ffa | 1161 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1162 | |
2d8f243a TH |
1163 | down_write(&css_set_rwsem); |
1164 | hash_for_each(css_set_table, i, cset, hlist) | |
1165 | list_move_tail(&cset->e_cset_node[ss->id], | |
1166 | &dst_root->cgrp.e_csets[ss->id]); | |
1167 | up_write(&css_set_rwsem); | |
1168 | ||
f392e51c TH |
1169 | src_root->subsys_mask &= ~(1 << ssid); |
1170 | src_root->cgrp.child_subsys_mask &= ~(1 << ssid); | |
1171 | ||
bd53d617 | 1172 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1173 | dst_root->subsys_mask |= 1 << ssid; |
bd53d617 TH |
1174 | if (dst_root != &cgrp_dfl_root) |
1175 | dst_root->cgrp.child_subsys_mask |= 1 << ssid; | |
a8a648c4 | 1176 | |
5df36032 TH |
1177 | if (ss->bind) |
1178 | ss->bind(css); | |
ddbcc7e8 | 1179 | } |
ddbcc7e8 | 1180 | |
a2dd4247 | 1181 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1182 | return 0; |
1183 | } | |
1184 | ||
2bd59d48 TH |
1185 | static int cgroup_show_options(struct seq_file *seq, |
1186 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1187 | { |
3dd06ffa | 1188 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1189 | struct cgroup_subsys *ss; |
b85d2040 | 1190 | int ssid; |
ddbcc7e8 | 1191 | |
b85d2040 | 1192 | for_each_subsys(ss, ssid) |
f392e51c | 1193 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1194 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1195 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1196 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1197 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1198 | seq_puts(seq, ",noprefix"); |
93438629 | 1199 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1200 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1201 | |
1202 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1203 | if (strlen(root->release_agent_path)) |
1204 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1205 | spin_unlock(&release_agent_path_lock); |
1206 | ||
3dd06ffa | 1207 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1208 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1209 | if (strlen(root->name)) |
1210 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1211 | return 0; |
1212 | } | |
1213 | ||
1214 | struct cgroup_sb_opts { | |
a1a71b45 | 1215 | unsigned long subsys_mask; |
ddbcc7e8 | 1216 | unsigned long flags; |
81a6a5cd | 1217 | char *release_agent; |
2260e7fc | 1218 | bool cpuset_clone_children; |
c6d57f33 | 1219 | char *name; |
2c6ab6d2 PM |
1220 | /* User explicitly requested empty subsystem */ |
1221 | bool none; | |
ddbcc7e8 PM |
1222 | }; |
1223 | ||
cf5d5941 | 1224 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1225 | { |
32a8cf23 DL |
1226 | char *token, *o = data; |
1227 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1228 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1229 | struct cgroup_subsys *ss; |
1230 | int i; | |
f9ab5b5b LZ |
1231 | |
1232 | #ifdef CONFIG_CPUSETS | |
073219e9 | 1233 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1234 | #endif |
ddbcc7e8 | 1235 | |
c6d57f33 | 1236 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1237 | |
1238 | while ((token = strsep(&o, ",")) != NULL) { | |
1239 | if (!*token) | |
1240 | return -EINVAL; | |
32a8cf23 | 1241 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1242 | /* Explicitly have no subsystems */ |
1243 | opts->none = true; | |
32a8cf23 DL |
1244 | continue; |
1245 | } | |
1246 | if (!strcmp(token, "all")) { | |
1247 | /* Mutually exclusive option 'all' + subsystem name */ | |
1248 | if (one_ss) | |
1249 | return -EINVAL; | |
1250 | all_ss = true; | |
1251 | continue; | |
1252 | } | |
873fe09e TH |
1253 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1254 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1255 | continue; | |
1256 | } | |
32a8cf23 | 1257 | if (!strcmp(token, "noprefix")) { |
93438629 | 1258 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1259 | continue; |
1260 | } | |
1261 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1262 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1263 | continue; |
1264 | } | |
03b1cde6 | 1265 | if (!strcmp(token, "xattr")) { |
93438629 | 1266 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1267 | continue; |
1268 | } | |
32a8cf23 | 1269 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1270 | /* Specifying two release agents is forbidden */ |
1271 | if (opts->release_agent) | |
1272 | return -EINVAL; | |
c6d57f33 | 1273 | opts->release_agent = |
e400c285 | 1274 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1275 | if (!opts->release_agent) |
1276 | return -ENOMEM; | |
32a8cf23 DL |
1277 | continue; |
1278 | } | |
1279 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1280 | const char *name = token + 5; |
1281 | /* Can't specify an empty name */ | |
1282 | if (!strlen(name)) | |
1283 | return -EINVAL; | |
1284 | /* Must match [\w.-]+ */ | |
1285 | for (i = 0; i < strlen(name); i++) { | |
1286 | char c = name[i]; | |
1287 | if (isalnum(c)) | |
1288 | continue; | |
1289 | if ((c == '.') || (c == '-') || (c == '_')) | |
1290 | continue; | |
1291 | return -EINVAL; | |
1292 | } | |
1293 | /* Specifying two names is forbidden */ | |
1294 | if (opts->name) | |
1295 | return -EINVAL; | |
1296 | opts->name = kstrndup(name, | |
e400c285 | 1297 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1298 | GFP_KERNEL); |
1299 | if (!opts->name) | |
1300 | return -ENOMEM; | |
32a8cf23 DL |
1301 | |
1302 | continue; | |
1303 | } | |
1304 | ||
30159ec7 | 1305 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1306 | if (strcmp(token, ss->name)) |
1307 | continue; | |
1308 | if (ss->disabled) | |
1309 | continue; | |
1310 | ||
1311 | /* Mutually exclusive option 'all' + subsystem name */ | |
1312 | if (all_ss) | |
1313 | return -EINVAL; | |
a1a71b45 | 1314 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1315 | one_ss = true; |
1316 | ||
1317 | break; | |
1318 | } | |
1319 | if (i == CGROUP_SUBSYS_COUNT) | |
1320 | return -ENOENT; | |
1321 | } | |
1322 | ||
2c6ab6d2 PM |
1323 | /* Consistency checks */ |
1324 | ||
873fe09e | 1325 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2a1f9ea | 1326 | pr_warn("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
873fe09e | 1327 | |
d3ba07c3 TH |
1328 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1329 | opts->cpuset_clone_children || opts->release_agent || | |
1330 | opts->name) { | |
1331 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1332 | return -EINVAL; |
1333 | } | |
a2dd4247 TH |
1334 | } else { |
1335 | /* | |
1336 | * If the 'all' option was specified select all the | |
1337 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1338 | * name options were not specified, let's default to 'all' | |
1339 | */ | |
1340 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1341 | for_each_subsys(ss, i) | |
1342 | if (!ss->disabled) | |
1343 | set_bit(i, &opts->subsys_mask); | |
873fe09e | 1344 | |
a2dd4247 TH |
1345 | /* |
1346 | * We either have to specify by name or by subsystems. (So | |
1347 | * all empty hierarchies must have a name). | |
1348 | */ | |
1349 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1350 | return -EINVAL; |
873fe09e TH |
1351 | } |
1352 | ||
f9ab5b5b LZ |
1353 | /* |
1354 | * Option noprefix was introduced just for backward compatibility | |
1355 | * with the old cpuset, so we allow noprefix only if mounting just | |
1356 | * the cpuset subsystem. | |
1357 | */ | |
93438629 | 1358 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1359 | return -EINVAL; |
1360 | ||
2c6ab6d2 PM |
1361 | |
1362 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1363 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1364 | return -EINVAL; |
1365 | ||
ddbcc7e8 PM |
1366 | return 0; |
1367 | } | |
1368 | ||
2bd59d48 | 1369 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1370 | { |
1371 | int ret = 0; | |
3dd06ffa | 1372 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1373 | struct cgroup_sb_opts opts; |
a1a71b45 | 1374 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1375 | |
873fe09e TH |
1376 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1377 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1378 | return -EINVAL; | |
1379 | } | |
1380 | ||
ace2bee8 | 1381 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1382 | mutex_lock(&cgroup_mutex); |
1383 | ||
1384 | /* See what subsystems are wanted */ | |
1385 | ret = parse_cgroupfs_options(data, &opts); | |
1386 | if (ret) | |
1387 | goto out_unlock; | |
1388 | ||
f392e51c | 1389 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
a2a1f9ea JZ |
1390 | pr_warn("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1391 | task_tgid_nr(current), current->comm); | |
8b5a5a9d | 1392 | |
f392e51c TH |
1393 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1394 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1395 | |
cf5d5941 | 1396 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1397 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1398 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1399 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1400 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1401 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1402 | ret = -EINVAL; |
1403 | goto out_unlock; | |
1404 | } | |
1405 | ||
f172e67c | 1406 | /* remounting is not allowed for populated hierarchies */ |
3dd06ffa | 1407 | if (!list_empty(&root->cgrp.children)) { |
f172e67c | 1408 | ret = -EBUSY; |
0670e08b | 1409 | goto out_unlock; |
cf5d5941 | 1410 | } |
ddbcc7e8 | 1411 | |
5df36032 | 1412 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1413 | if (ret) |
0670e08b | 1414 | goto out_unlock; |
ddbcc7e8 | 1415 | |
3dd06ffa | 1416 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1417 | |
69e943b7 TH |
1418 | if (opts.release_agent) { |
1419 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1420 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1421 | spin_unlock(&release_agent_path_lock); |
1422 | } | |
ddbcc7e8 | 1423 | out_unlock: |
66bdc9cf | 1424 | kfree(opts.release_agent); |
c6d57f33 | 1425 | kfree(opts.name); |
ddbcc7e8 | 1426 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1427 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1428 | return ret; |
1429 | } | |
1430 | ||
afeb0f9f TH |
1431 | /* |
1432 | * To reduce the fork() overhead for systems that are not actually using | |
1433 | * their cgroups capability, we don't maintain the lists running through | |
1434 | * each css_set to its tasks until we see the list actually used - in other | |
1435 | * words after the first mount. | |
1436 | */ | |
1437 | static bool use_task_css_set_links __read_mostly; | |
1438 | ||
1439 | static void cgroup_enable_task_cg_lists(void) | |
1440 | { | |
1441 | struct task_struct *p, *g; | |
1442 | ||
96d365e0 | 1443 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1444 | |
1445 | if (use_task_css_set_links) | |
1446 | goto out_unlock; | |
1447 | ||
1448 | use_task_css_set_links = true; | |
1449 | ||
1450 | /* | |
1451 | * We need tasklist_lock because RCU is not safe against | |
1452 | * while_each_thread(). Besides, a forking task that has passed | |
1453 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1454 | * is not guaranteed to have its child immediately visible in the | |
1455 | * tasklist if we walk through it with RCU. | |
1456 | */ | |
1457 | read_lock(&tasklist_lock); | |
1458 | do_each_thread(g, p) { | |
afeb0f9f TH |
1459 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1460 | task_css_set(p) != &init_css_set); | |
1461 | ||
1462 | /* | |
1463 | * We should check if the process is exiting, otherwise | |
1464 | * it will race with cgroup_exit() in that the list | |
1465 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1466 | * Do it while holding siglock so that we don't end up |
1467 | * racing against cgroup_exit(). | |
afeb0f9f | 1468 | */ |
f153ad11 | 1469 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1470 | if (!(p->flags & PF_EXITING)) { |
1471 | struct css_set *cset = task_css_set(p); | |
1472 | ||
1473 | list_add(&p->cg_list, &cset->tasks); | |
1474 | get_css_set(cset); | |
1475 | } | |
f153ad11 | 1476 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1477 | } while_each_thread(g, p); |
1478 | read_unlock(&tasklist_lock); | |
1479 | out_unlock: | |
96d365e0 | 1480 | up_write(&css_set_rwsem); |
afeb0f9f | 1481 | } |
ddbcc7e8 | 1482 | |
cc31edce PM |
1483 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1484 | { | |
2d8f243a TH |
1485 | struct cgroup_subsys *ss; |
1486 | int ssid; | |
1487 | ||
2bd59d48 | 1488 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1489 | INIT_LIST_HEAD(&cgrp->sibling); |
1490 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1491 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1492 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1493 | INIT_LIST_HEAD(&cgrp->pidlists); |
1494 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1495 | cgrp->dummy_css.cgroup = cgrp; |
2d8f243a TH |
1496 | |
1497 | for_each_subsys(ss, ssid) | |
1498 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1499 | |
1500 | init_waitqueue_head(&cgrp->offline_waitq); | |
cc31edce | 1501 | } |
c6d57f33 | 1502 | |
3dd06ffa | 1503 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1504 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1505 | { |
3dd06ffa | 1506 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1507 | |
ddbcc7e8 | 1508 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1509 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1510 | cgrp->root = root; |
cc31edce | 1511 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1512 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1513 | |
c6d57f33 PM |
1514 | root->flags = opts->flags; |
1515 | if (opts->release_agent) | |
1516 | strcpy(root->release_agent_path, opts->release_agent); | |
1517 | if (opts->name) | |
1518 | strcpy(root->name, opts->name); | |
2260e7fc | 1519 | if (opts->cpuset_clone_children) |
3dd06ffa | 1520 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1521 | } |
1522 | ||
3dd06ffa | 1523 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1524 | { |
d427dfeb | 1525 | LIST_HEAD(tmp_links); |
3dd06ffa | 1526 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1527 | struct css_set *cset; |
d427dfeb | 1528 | int i, ret; |
2c6ab6d2 | 1529 | |
d427dfeb TH |
1530 | lockdep_assert_held(&cgroup_tree_mutex); |
1531 | lockdep_assert_held(&cgroup_mutex); | |
c6d57f33 | 1532 | |
d427dfeb TH |
1533 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); |
1534 | if (ret < 0) | |
2bd59d48 | 1535 | goto out; |
d427dfeb | 1536 | root_cgrp->id = ret; |
c6d57f33 | 1537 | |
d427dfeb | 1538 | /* |
96d365e0 | 1539 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1540 | * but that's OK - it can only be increased by someone holding |
1541 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1542 | * have some link structures left over | |
1543 | */ | |
1544 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1545 | if (ret) | |
2bd59d48 | 1546 | goto out; |
ddbcc7e8 | 1547 | |
985ed670 | 1548 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1549 | if (ret) |
2bd59d48 | 1550 | goto out; |
ddbcc7e8 | 1551 | |
2bd59d48 TH |
1552 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1553 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1554 | root_cgrp); | |
1555 | if (IS_ERR(root->kf_root)) { | |
1556 | ret = PTR_ERR(root->kf_root); | |
1557 | goto exit_root_id; | |
1558 | } | |
1559 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1560 | |
d427dfeb TH |
1561 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1562 | if (ret) | |
2bd59d48 | 1563 | goto destroy_root; |
ddbcc7e8 | 1564 | |
5df36032 | 1565 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1566 | if (ret) |
2bd59d48 | 1567 | goto destroy_root; |
ddbcc7e8 | 1568 | |
d427dfeb TH |
1569 | /* |
1570 | * There must be no failure case after here, since rebinding takes | |
1571 | * care of subsystems' refcounts, which are explicitly dropped in | |
1572 | * the failure exit path. | |
1573 | */ | |
1574 | list_add(&root->root_list, &cgroup_roots); | |
1575 | cgroup_root_count++; | |
0df6a63f | 1576 | |
d427dfeb | 1577 | /* |
3dd06ffa | 1578 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1579 | * objects. |
1580 | */ | |
96d365e0 | 1581 | down_write(&css_set_rwsem); |
d427dfeb TH |
1582 | hash_for_each(css_set_table, i, cset, hlist) |
1583 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1584 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1585 | |
d427dfeb | 1586 | BUG_ON(!list_empty(&root_cgrp->children)); |
3c9c825b | 1587 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1588 | |
2bd59d48 | 1589 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1590 | ret = 0; |
2bd59d48 | 1591 | goto out; |
d427dfeb | 1592 | |
2bd59d48 TH |
1593 | destroy_root: |
1594 | kernfs_destroy_root(root->kf_root); | |
1595 | root->kf_root = NULL; | |
1596 | exit_root_id: | |
d427dfeb | 1597 | cgroup_exit_root_id(root); |
2bd59d48 | 1598 | out: |
d427dfeb TH |
1599 | free_cgrp_cset_links(&tmp_links); |
1600 | return ret; | |
ddbcc7e8 PM |
1601 | } |
1602 | ||
f7e83571 | 1603 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1604 | int flags, const char *unused_dev_name, |
f7e83571 | 1605 | void *data) |
ddbcc7e8 | 1606 | { |
3dd06ffa | 1607 | struct cgroup_root *root; |
ddbcc7e8 | 1608 | struct cgroup_sb_opts opts; |
2bd59d48 | 1609 | struct dentry *dentry; |
8e30e2b8 | 1610 | int ret; |
c6b3d5bc | 1611 | bool new_sb; |
ddbcc7e8 | 1612 | |
56fde9e0 TH |
1613 | /* |
1614 | * The first time anyone tries to mount a cgroup, enable the list | |
1615 | * linking each css_set to its tasks and fix up all existing tasks. | |
1616 | */ | |
1617 | if (!use_task_css_set_links) | |
1618 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1619 | |
8e30e2b8 | 1620 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1621 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1622 | |
1623 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1624 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1625 | if (ret) |
8e30e2b8 | 1626 | goto out_unlock; |
e37a06f1 | 1627 | retry: |
2bd59d48 | 1628 | /* look for a matching existing root */ |
a2dd4247 TH |
1629 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1630 | cgrp_dfl_root_visible = true; | |
1631 | root = &cgrp_dfl_root; | |
1632 | cgroup_get(&root->cgrp); | |
1633 | ret = 0; | |
1634 | goto out_unlock; | |
ddbcc7e8 PM |
1635 | } |
1636 | ||
985ed670 | 1637 | for_each_root(root) { |
2bd59d48 | 1638 | bool name_match = false; |
3126121f | 1639 | |
3dd06ffa | 1640 | if (root == &cgrp_dfl_root) |
985ed670 | 1641 | continue; |
3126121f | 1642 | |
cf5d5941 | 1643 | /* |
2bd59d48 TH |
1644 | * If we asked for a name then it must match. Also, if |
1645 | * name matches but sybsys_mask doesn't, we should fail. | |
1646 | * Remember whether name matched. | |
cf5d5941 | 1647 | */ |
2bd59d48 TH |
1648 | if (opts.name) { |
1649 | if (strcmp(opts.name, root->name)) | |
1650 | continue; | |
1651 | name_match = true; | |
1652 | } | |
ddbcc7e8 | 1653 | |
c6d57f33 | 1654 | /* |
2bd59d48 TH |
1655 | * If we asked for subsystems (or explicitly for no |
1656 | * subsystems) then they must match. | |
c6d57f33 | 1657 | */ |
2bd59d48 | 1658 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1659 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1660 | if (!name_match) |
1661 | continue; | |
1662 | ret = -EBUSY; | |
1663 | goto out_unlock; | |
1664 | } | |
873fe09e | 1665 | |
c7ba8287 | 1666 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1667 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1668 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1669 | ret = -EINVAL; | |
8e30e2b8 | 1670 | goto out_unlock; |
2a0ff3fb | 1671 | } else { |
a2a1f9ea | 1672 | pr_warn("cgroup: new mount options do not match the existing superblock, will be ignored\n"); |
2a0ff3fb | 1673 | } |
873fe09e | 1674 | } |
ddbcc7e8 | 1675 | |
776f02fa | 1676 | /* |
3dd06ffa | 1677 | * A root's lifetime is governed by its root cgroup. Zero |
776f02fa TH |
1678 | * ref indicate that the root is being destroyed. Wait for |
1679 | * destruction to complete so that the subsystems are free. | |
1680 | * We can use wait_queue for the wait but this path is | |
1681 | * super cold. Let's just sleep for a bit and retry. | |
1682 | */ | |
3dd06ffa | 1683 | if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { |
776f02fa TH |
1684 | mutex_unlock(&cgroup_mutex); |
1685 | mutex_unlock(&cgroup_tree_mutex); | |
1686 | msleep(10); | |
e37a06f1 LZ |
1687 | mutex_lock(&cgroup_tree_mutex); |
1688 | mutex_lock(&cgroup_mutex); | |
776f02fa TH |
1689 | goto retry; |
1690 | } | |
ddbcc7e8 | 1691 | |
776f02fa | 1692 | ret = 0; |
2bd59d48 | 1693 | goto out_unlock; |
ddbcc7e8 | 1694 | } |
ddbcc7e8 | 1695 | |
817929ec | 1696 | /* |
172a2c06 TH |
1697 | * No such thing, create a new one. name= matching without subsys |
1698 | * specification is allowed for already existing hierarchies but we | |
1699 | * can't create new one without subsys specification. | |
817929ec | 1700 | */ |
172a2c06 TH |
1701 | if (!opts.subsys_mask && !opts.none) { |
1702 | ret = -EINVAL; | |
1703 | goto out_unlock; | |
817929ec | 1704 | } |
817929ec | 1705 | |
172a2c06 TH |
1706 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1707 | if (!root) { | |
1708 | ret = -ENOMEM; | |
2bd59d48 | 1709 | goto out_unlock; |
839ec545 | 1710 | } |
e5f6a860 | 1711 | |
172a2c06 TH |
1712 | init_cgroup_root(root, &opts); |
1713 | ||
35585573 | 1714 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1715 | if (ret) |
1716 | cgroup_free_root(root); | |
fa3ca07e | 1717 | |
8e30e2b8 | 1718 | out_unlock: |
ddbcc7e8 | 1719 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1720 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 | 1721 | |
c6d57f33 PM |
1722 | kfree(opts.release_agent); |
1723 | kfree(opts.name); | |
03b1cde6 | 1724 | |
2bd59d48 | 1725 | if (ret) |
8e30e2b8 | 1726 | return ERR_PTR(ret); |
2bd59d48 | 1727 | |
c6b3d5bc LZ |
1728 | dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); |
1729 | if (IS_ERR(dentry) || !new_sb) | |
3dd06ffa | 1730 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1731 | return dentry; |
1732 | } | |
1733 | ||
1734 | static void cgroup_kill_sb(struct super_block *sb) | |
1735 | { | |
1736 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1737 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1738 | |
3dd06ffa | 1739 | cgroup_put(&root->cgrp); |
2bd59d48 | 1740 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1741 | } |
1742 | ||
1743 | static struct file_system_type cgroup_fs_type = { | |
1744 | .name = "cgroup", | |
f7e83571 | 1745 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1746 | .kill_sb = cgroup_kill_sb, |
1747 | }; | |
1748 | ||
676db4af GKH |
1749 | static struct kobject *cgroup_kobj; |
1750 | ||
857a2beb | 1751 | /** |
913ffdb5 | 1752 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1753 | * @task: target task |
857a2beb TH |
1754 | * @buf: the buffer to write the path into |
1755 | * @buflen: the length of the buffer | |
1756 | * | |
913ffdb5 TH |
1757 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1758 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1759 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1760 | * cgroup controller callbacks. | |
1761 | * | |
e61734c5 | 1762 | * Return value is the same as kernfs_path(). |
857a2beb | 1763 | */ |
e61734c5 | 1764 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1765 | { |
3dd06ffa | 1766 | struct cgroup_root *root; |
913ffdb5 | 1767 | struct cgroup *cgrp; |
e61734c5 TH |
1768 | int hierarchy_id = 1; |
1769 | char *path = NULL; | |
857a2beb TH |
1770 | |
1771 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1772 | down_read(&css_set_rwsem); |
857a2beb | 1773 | |
913ffdb5 TH |
1774 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1775 | ||
857a2beb TH |
1776 | if (root) { |
1777 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1778 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1779 | } else { |
1780 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1781 | if (strlcpy(buf, "/", buflen) < buflen) |
1782 | path = buf; | |
857a2beb TH |
1783 | } |
1784 | ||
96d365e0 | 1785 | up_read(&css_set_rwsem); |
857a2beb | 1786 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1787 | return path; |
857a2beb | 1788 | } |
913ffdb5 | 1789 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1790 | |
b3dc094e | 1791 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1792 | struct cgroup_taskset { |
b3dc094e TH |
1793 | /* the src and dst cset list running through cset->mg_node */ |
1794 | struct list_head src_csets; | |
1795 | struct list_head dst_csets; | |
1796 | ||
1797 | /* | |
1798 | * Fields for cgroup_taskset_*() iteration. | |
1799 | * | |
1800 | * Before migration is committed, the target migration tasks are on | |
1801 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1802 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1803 | * or ->dst_csets depending on whether migration is committed. | |
1804 | * | |
1805 | * ->cur_csets and ->cur_task point to the current task position | |
1806 | * during iteration. | |
1807 | */ | |
1808 | struct list_head *csets; | |
1809 | struct css_set *cur_cset; | |
1810 | struct task_struct *cur_task; | |
2f7ee569 TH |
1811 | }; |
1812 | ||
1813 | /** | |
1814 | * cgroup_taskset_first - reset taskset and return the first task | |
1815 | * @tset: taskset of interest | |
1816 | * | |
1817 | * @tset iteration is initialized and the first task is returned. | |
1818 | */ | |
1819 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1820 | { | |
b3dc094e TH |
1821 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1822 | tset->cur_task = NULL; | |
1823 | ||
1824 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1825 | } |
2f7ee569 TH |
1826 | |
1827 | /** | |
1828 | * cgroup_taskset_next - iterate to the next task in taskset | |
1829 | * @tset: taskset of interest | |
1830 | * | |
1831 | * Return the next task in @tset. Iteration must have been initialized | |
1832 | * with cgroup_taskset_first(). | |
1833 | */ | |
1834 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1835 | { | |
b3dc094e TH |
1836 | struct css_set *cset = tset->cur_cset; |
1837 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1838 | |
b3dc094e TH |
1839 | while (&cset->mg_node != tset->csets) { |
1840 | if (!task) | |
1841 | task = list_first_entry(&cset->mg_tasks, | |
1842 | struct task_struct, cg_list); | |
1843 | else | |
1844 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1845 | |
b3dc094e TH |
1846 | if (&task->cg_list != &cset->mg_tasks) { |
1847 | tset->cur_cset = cset; | |
1848 | tset->cur_task = task; | |
1849 | return task; | |
1850 | } | |
2f7ee569 | 1851 | |
b3dc094e TH |
1852 | cset = list_next_entry(cset, mg_node); |
1853 | task = NULL; | |
1854 | } | |
2f7ee569 | 1855 | |
b3dc094e | 1856 | return NULL; |
2f7ee569 | 1857 | } |
2f7ee569 | 1858 | |
cb0f1fe9 | 1859 | /** |
74a1166d | 1860 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1861 | * @old_cgrp; the cgroup @tsk is being migrated from |
1862 | * @tsk: the task being migrated | |
1863 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1864 | * |
cb0f1fe9 | 1865 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1866 | */ |
5abb8855 TH |
1867 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1868 | struct task_struct *tsk, | |
1869 | struct css_set *new_cset) | |
74a1166d | 1870 | { |
5abb8855 | 1871 | struct css_set *old_cset; |
74a1166d | 1872 | |
cb0f1fe9 TH |
1873 | lockdep_assert_held(&cgroup_mutex); |
1874 | lockdep_assert_held(&css_set_rwsem); | |
1875 | ||
74a1166d | 1876 | /* |
026085ef MSB |
1877 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1878 | * setting such that we can't race against cgroup_exit() changing the | |
1879 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1880 | */ |
c84cdf75 | 1881 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1882 | old_cset = task_css_set(tsk); |
74a1166d | 1883 | |
b3dc094e | 1884 | get_css_set(new_cset); |
5abb8855 | 1885 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1886 | |
1b9aba49 TH |
1887 | /* |
1888 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1889 | * leader after migration. This works because cgroup_migrate() | |
1890 | * ensures that the dst_cset of the leader is the first on the | |
1891 | * tset's dst_csets list. | |
1892 | */ | |
1893 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1894 | |
1895 | /* | |
5abb8855 TH |
1896 | * We just gained a reference on old_cset by taking it from the |
1897 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1898 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1899 | */ |
5abb8855 | 1900 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1901 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1902 | } |
1903 | ||
a043e3b2 | 1904 | /** |
1958d2d5 TH |
1905 | * cgroup_migrate_finish - cleanup after attach |
1906 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1907 | * |
1958d2d5 TH |
1908 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1909 | * those functions for details. | |
74a1166d | 1910 | */ |
1958d2d5 | 1911 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1912 | { |
1958d2d5 | 1913 | struct css_set *cset, *tmp_cset; |
74a1166d | 1914 | |
1958d2d5 TH |
1915 | lockdep_assert_held(&cgroup_mutex); |
1916 | ||
1917 | down_write(&css_set_rwsem); | |
1918 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1919 | cset->mg_src_cgrp = NULL; | |
1920 | cset->mg_dst_cset = NULL; | |
1921 | list_del_init(&cset->mg_preload_node); | |
1922 | put_css_set_locked(cset, false); | |
1923 | } | |
1924 | up_write(&css_set_rwsem); | |
1925 | } | |
1926 | ||
1927 | /** | |
1928 | * cgroup_migrate_add_src - add a migration source css_set | |
1929 | * @src_cset: the source css_set to add | |
1930 | * @dst_cgrp: the destination cgroup | |
1931 | * @preloaded_csets: list of preloaded css_sets | |
1932 | * | |
1933 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1934 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1935 | * up by cgroup_migrate_finish(). | |
1936 | * | |
1937 | * This function may be called without holding threadgroup_lock even if the | |
1938 | * target is a process. Threads may be created and destroyed but as long | |
1939 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1940 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1941 | */ | |
1942 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1943 | struct cgroup *dst_cgrp, | |
1944 | struct list_head *preloaded_csets) | |
1945 | { | |
1946 | struct cgroup *src_cgrp; | |
1947 | ||
1948 | lockdep_assert_held(&cgroup_mutex); | |
1949 | lockdep_assert_held(&css_set_rwsem); | |
1950 | ||
1951 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1952 | ||
1958d2d5 TH |
1953 | if (!list_empty(&src_cset->mg_preload_node)) |
1954 | return; | |
1955 | ||
1956 | WARN_ON(src_cset->mg_src_cgrp); | |
1957 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1958 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1959 | ||
1960 | src_cset->mg_src_cgrp = src_cgrp; | |
1961 | get_css_set(src_cset); | |
1962 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1963 | } | |
1964 | ||
1965 | /** | |
1966 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 1967 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
1968 | * @preloaded_csets: list of preloaded source css_sets |
1969 | * | |
1970 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1971 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
1972 | * pins all destination css_sets, links each to its source, and append them |
1973 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
1974 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
1975 | * |
1976 | * This function must be called after cgroup_migrate_add_src() has been | |
1977 | * called on each migration source css_set. After migration is performed | |
1978 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
1979 | * @preloaded_csets. | |
1980 | */ | |
1981 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
1982 | struct list_head *preloaded_csets) | |
1983 | { | |
1984 | LIST_HEAD(csets); | |
f817de98 | 1985 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
1986 | |
1987 | lockdep_assert_held(&cgroup_mutex); | |
1988 | ||
f8f22e53 TH |
1989 | /* |
1990 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
1991 | * with tasks so that child cgroups don't compete against tasks. | |
1992 | */ | |
1993 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && dst_cgrp->parent && | |
1994 | dst_cgrp->child_subsys_mask) | |
1995 | return -EBUSY; | |
1996 | ||
1958d2d5 | 1997 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 1998 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
1999 | struct css_set *dst_cset; |
2000 | ||
f817de98 TH |
2001 | dst_cset = find_css_set(src_cset, |
2002 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2003 | if (!dst_cset) |
2004 | goto err; | |
2005 | ||
2006 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2007 | |
2008 | /* | |
2009 | * If src cset equals dst, it's noop. Drop the src. | |
2010 | * cgroup_migrate() will skip the cset too. Note that we | |
2011 | * can't handle src == dst as some nodes are used by both. | |
2012 | */ | |
2013 | if (src_cset == dst_cset) { | |
2014 | src_cset->mg_src_cgrp = NULL; | |
2015 | list_del_init(&src_cset->mg_preload_node); | |
2016 | put_css_set(src_cset, false); | |
2017 | put_css_set(dst_cset, false); | |
2018 | continue; | |
2019 | } | |
2020 | ||
1958d2d5 TH |
2021 | src_cset->mg_dst_cset = dst_cset; |
2022 | ||
2023 | if (list_empty(&dst_cset->mg_preload_node)) | |
2024 | list_add(&dst_cset->mg_preload_node, &csets); | |
2025 | else | |
2026 | put_css_set(dst_cset, false); | |
2027 | } | |
2028 | ||
f817de98 | 2029 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2030 | return 0; |
2031 | err: | |
2032 | cgroup_migrate_finish(&csets); | |
2033 | return -ENOMEM; | |
2034 | } | |
2035 | ||
2036 | /** | |
2037 | * cgroup_migrate - migrate a process or task to a cgroup | |
2038 | * @cgrp: the destination cgroup | |
2039 | * @leader: the leader of the process or the task to migrate | |
2040 | * @threadgroup: whether @leader points to the whole process or a single task | |
2041 | * | |
2042 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2043 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2044 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2045 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2046 | * function and following up with cgroup_migrate_finish(). | |
2047 | * | |
2048 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2049 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2050 | * failure, when migrating multiple targets, the success or failure can be | |
2051 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2052 | * actually starting migrating. | |
2053 | */ | |
2054 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2055 | bool threadgroup) | |
74a1166d | 2056 | { |
b3dc094e TH |
2057 | struct cgroup_taskset tset = { |
2058 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2059 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2060 | .csets = &tset.src_csets, | |
2061 | }; | |
1c6727af | 2062 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2063 | struct css_set *cset, *tmp_cset; |
2064 | struct task_struct *task, *tmp_task; | |
2065 | int i, ret; | |
74a1166d | 2066 | |
fb5d2b4c MSB |
2067 | /* |
2068 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2069 | * already PF_EXITING could be freed from underneath us unless we | |
2070 | * take an rcu_read_lock. | |
2071 | */ | |
b3dc094e | 2072 | down_write(&css_set_rwsem); |
fb5d2b4c | 2073 | rcu_read_lock(); |
9db8de37 | 2074 | task = leader; |
74a1166d | 2075 | do { |
9db8de37 TH |
2076 | /* @task either already exited or can't exit until the end */ |
2077 | if (task->flags & PF_EXITING) | |
ea84753c | 2078 | goto next; |
134d3373 | 2079 | |
eaf797ab TH |
2080 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2081 | if (list_empty(&task->cg_list)) | |
ea84753c | 2082 | goto next; |
cd3d0952 | 2083 | |
b3dc094e | 2084 | cset = task_css_set(task); |
1958d2d5 | 2085 | if (!cset->mg_src_cgrp) |
ea84753c | 2086 | goto next; |
b3dc094e | 2087 | |
61d1d219 | 2088 | /* |
1b9aba49 TH |
2089 | * cgroup_taskset_first() must always return the leader. |
2090 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2091 | */ |
1b9aba49 TH |
2092 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2093 | if (list_empty(&cset->mg_node)) | |
2094 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2095 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2096 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2097 | &tset.dst_csets); | |
ea84753c | 2098 | next: |
081aa458 LZ |
2099 | if (!threadgroup) |
2100 | break; | |
9db8de37 | 2101 | } while_each_thread(leader, task); |
fb5d2b4c | 2102 | rcu_read_unlock(); |
b3dc094e | 2103 | up_write(&css_set_rwsem); |
74a1166d | 2104 | |
134d3373 | 2105 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2106 | if (list_empty(&tset.src_csets)) |
2107 | return 0; | |
134d3373 | 2108 | |
1958d2d5 | 2109 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2110 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2111 | if (css->ss->can_attach) { |
9db8de37 TH |
2112 | ret = css->ss->can_attach(css, &tset); |
2113 | if (ret) { | |
1c6727af | 2114 | failed_css = css; |
74a1166d BB |
2115 | goto out_cancel_attach; |
2116 | } | |
2117 | } | |
74a1166d BB |
2118 | } |
2119 | ||
2120 | /* | |
1958d2d5 TH |
2121 | * Now that we're guaranteed success, proceed to move all tasks to |
2122 | * the new cgroup. There are no failure cases after here, so this | |
2123 | * is the commit point. | |
74a1166d | 2124 | */ |
cb0f1fe9 | 2125 | down_write(&css_set_rwsem); |
b3dc094e TH |
2126 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2127 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2128 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2129 | cset->mg_dst_cset); | |
74a1166d | 2130 | } |
cb0f1fe9 | 2131 | up_write(&css_set_rwsem); |
74a1166d BB |
2132 | |
2133 | /* | |
1958d2d5 TH |
2134 | * Migration is committed, all target tasks are now on dst_csets. |
2135 | * Nothing is sensitive to fork() after this point. Notify | |
2136 | * controllers that migration is complete. | |
74a1166d | 2137 | */ |
1958d2d5 | 2138 | tset.csets = &tset.dst_csets; |
74a1166d | 2139 | |
aec3dfcb | 2140 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2141 | if (css->ss->attach) |
2142 | css->ss->attach(css, &tset); | |
74a1166d | 2143 | |
9db8de37 | 2144 | ret = 0; |
b3dc094e TH |
2145 | goto out_release_tset; |
2146 | ||
74a1166d | 2147 | out_cancel_attach: |
aec3dfcb | 2148 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2149 | if (css == failed_css) |
2150 | break; | |
2151 | if (css->ss->cancel_attach) | |
2152 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2153 | } |
b3dc094e TH |
2154 | out_release_tset: |
2155 | down_write(&css_set_rwsem); | |
2156 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2157 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2158 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2159 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2160 | } |
2161 | up_write(&css_set_rwsem); | |
9db8de37 | 2162 | return ret; |
74a1166d BB |
2163 | } |
2164 | ||
1958d2d5 TH |
2165 | /** |
2166 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2167 | * @dst_cgrp: the cgroup to attach to | |
2168 | * @leader: the task or the leader of the threadgroup to be attached | |
2169 | * @threadgroup: attach the whole threadgroup? | |
2170 | * | |
0e1d768f | 2171 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2172 | */ |
2173 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2174 | struct task_struct *leader, bool threadgroup) | |
2175 | { | |
2176 | LIST_HEAD(preloaded_csets); | |
2177 | struct task_struct *task; | |
2178 | int ret; | |
2179 | ||
2180 | /* look up all src csets */ | |
2181 | down_read(&css_set_rwsem); | |
2182 | rcu_read_lock(); | |
2183 | task = leader; | |
2184 | do { | |
2185 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2186 | &preloaded_csets); | |
2187 | if (!threadgroup) | |
2188 | break; | |
2189 | } while_each_thread(leader, task); | |
2190 | rcu_read_unlock(); | |
2191 | up_read(&css_set_rwsem); | |
2192 | ||
2193 | /* prepare dst csets and commit */ | |
2194 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2195 | if (!ret) | |
2196 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2197 | ||
2198 | cgroup_migrate_finish(&preloaded_csets); | |
2199 | return ret; | |
74a1166d BB |
2200 | } |
2201 | ||
2202 | /* | |
2203 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2204 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2205 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2206 | */ |
74a1166d | 2207 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2208 | { |
bbcb81d0 | 2209 | struct task_struct *tsk; |
c69e8d9c | 2210 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2211 | int ret; |
2212 | ||
74a1166d BB |
2213 | if (!cgroup_lock_live_group(cgrp)) |
2214 | return -ENODEV; | |
2215 | ||
b78949eb MSB |
2216 | retry_find_task: |
2217 | rcu_read_lock(); | |
bbcb81d0 | 2218 | if (pid) { |
73507f33 | 2219 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2220 | if (!tsk) { |
2221 | rcu_read_unlock(); | |
dd4b0a46 | 2222 | ret = -ESRCH; |
b78949eb | 2223 | goto out_unlock_cgroup; |
bbcb81d0 | 2224 | } |
74a1166d BB |
2225 | /* |
2226 | * even if we're attaching all tasks in the thread group, we | |
2227 | * only need to check permissions on one of them. | |
2228 | */ | |
c69e8d9c | 2229 | tcred = __task_cred(tsk); |
14a590c3 EB |
2230 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2231 | !uid_eq(cred->euid, tcred->uid) && | |
2232 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2233 | rcu_read_unlock(); |
b78949eb MSB |
2234 | ret = -EACCES; |
2235 | goto out_unlock_cgroup; | |
bbcb81d0 | 2236 | } |
b78949eb MSB |
2237 | } else |
2238 | tsk = current; | |
cd3d0952 TH |
2239 | |
2240 | if (threadgroup) | |
b78949eb | 2241 | tsk = tsk->group_leader; |
c4c27fbd MG |
2242 | |
2243 | /* | |
14a40ffc | 2244 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2245 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2246 | * with no rt_runtime allocated. Just say no. | |
2247 | */ | |
14a40ffc | 2248 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2249 | ret = -EINVAL; |
2250 | rcu_read_unlock(); | |
2251 | goto out_unlock_cgroup; | |
2252 | } | |
2253 | ||
b78949eb MSB |
2254 | get_task_struct(tsk); |
2255 | rcu_read_unlock(); | |
2256 | ||
2257 | threadgroup_lock(tsk); | |
2258 | if (threadgroup) { | |
2259 | if (!thread_group_leader(tsk)) { | |
2260 | /* | |
2261 | * a race with de_thread from another thread's exec() | |
2262 | * may strip us of our leadership, if this happens, | |
2263 | * there is no choice but to throw this task away and | |
2264 | * try again; this is | |
2265 | * "double-double-toil-and-trouble-check locking". | |
2266 | */ | |
2267 | threadgroup_unlock(tsk); | |
2268 | put_task_struct(tsk); | |
2269 | goto retry_find_task; | |
2270 | } | |
081aa458 LZ |
2271 | } |
2272 | ||
2273 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2274 | ||
cd3d0952 TH |
2275 | threadgroup_unlock(tsk); |
2276 | ||
bbcb81d0 | 2277 | put_task_struct(tsk); |
b78949eb | 2278 | out_unlock_cgroup: |
47cfcd09 | 2279 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2280 | return ret; |
2281 | } | |
2282 | ||
7ae1bad9 TH |
2283 | /** |
2284 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2285 | * @from: attach to all cgroups of a given task | |
2286 | * @tsk: the task to be attached | |
2287 | */ | |
2288 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2289 | { | |
3dd06ffa | 2290 | struct cgroup_root *root; |
7ae1bad9 TH |
2291 | int retval = 0; |
2292 | ||
47cfcd09 | 2293 | mutex_lock(&cgroup_mutex); |
985ed670 | 2294 | for_each_root(root) { |
96d365e0 TH |
2295 | struct cgroup *from_cgrp; |
2296 | ||
3dd06ffa | 2297 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2298 | continue; |
2299 | ||
96d365e0 TH |
2300 | down_read(&css_set_rwsem); |
2301 | from_cgrp = task_cgroup_from_root(from, root); | |
2302 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2303 | |
6f4b7e63 | 2304 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2305 | if (retval) |
2306 | break; | |
2307 | } | |
47cfcd09 | 2308 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2309 | |
2310 | return retval; | |
2311 | } | |
2312 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2313 | ||
182446d0 TH |
2314 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2315 | struct cftype *cft, u64 pid) | |
74a1166d | 2316 | { |
182446d0 | 2317 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2318 | } |
2319 | ||
182446d0 TH |
2320 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2321 | struct cftype *cft, u64 tgid) | |
af351026 | 2322 | { |
182446d0 | 2323 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2324 | } |
2325 | ||
182446d0 | 2326 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
4d3bb511 | 2327 | struct cftype *cft, char *buffer) |
e788e066 | 2328 | { |
3dd06ffa | 2329 | struct cgroup_root *root = css->cgroup->root; |
5f469907 TH |
2330 | |
2331 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2332 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2333 | return -ENODEV; |
69e943b7 | 2334 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2335 | strlcpy(root->release_agent_path, buffer, |
2336 | sizeof(root->release_agent_path)); | |
69e943b7 | 2337 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2338 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2339 | return 0; |
2340 | } | |
2341 | ||
2da8ca82 | 2342 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2343 | { |
2da8ca82 | 2344 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2345 | |
e788e066 PM |
2346 | if (!cgroup_lock_live_group(cgrp)) |
2347 | return -ENODEV; | |
2348 | seq_puts(seq, cgrp->root->release_agent_path); | |
2349 | seq_putc(seq, '\n'); | |
47cfcd09 | 2350 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2351 | return 0; |
2352 | } | |
2353 | ||
2da8ca82 | 2354 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2355 | { |
2da8ca82 TH |
2356 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2357 | ||
2358 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2359 | return 0; |
2360 | } | |
2361 | ||
f8f22e53 TH |
2362 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
2363 | { | |
2364 | struct cgroup_subsys *ss; | |
2365 | bool printed = false; | |
2366 | int ssid; | |
2367 | ||
2368 | for_each_subsys(ss, ssid) { | |
2369 | if (ss_mask & (1 << ssid)) { | |
2370 | if (printed) | |
2371 | seq_putc(seq, ' '); | |
2372 | seq_printf(seq, "%s", ss->name); | |
2373 | printed = true; | |
2374 | } | |
2375 | } | |
2376 | if (printed) | |
2377 | seq_putc(seq, '\n'); | |
2378 | } | |
2379 | ||
2380 | /* show controllers which are currently attached to the default hierarchy */ | |
2381 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
2382 | { | |
2383 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2384 | ||
2385 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask); | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | /* show controllers which are enabled from the parent */ | |
2390 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
2391 | { | |
2392 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2393 | ||
2394 | cgroup_print_ss_mask(seq, cgrp->parent->child_subsys_mask); | |
2395 | return 0; | |
2396 | } | |
2397 | ||
2398 | /* show controllers which are enabled for a given cgroup's children */ | |
2399 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
2400 | { | |
2401 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2402 | ||
2403 | cgroup_print_ss_mask(seq, cgrp->child_subsys_mask); | |
2404 | return 0; | |
2405 | } | |
2406 | ||
2407 | /** | |
2408 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2409 | * @cgrp: root of the subtree to update csses for | |
2410 | * | |
2411 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2412 | * css associations need to be updated accordingly. This function looks up | |
2413 | * all css_sets which are attached to the subtree, creates the matching | |
2414 | * updated css_sets and migrates the tasks to the new ones. | |
2415 | */ | |
2416 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2417 | { | |
2418 | LIST_HEAD(preloaded_csets); | |
2419 | struct cgroup_subsys_state *css; | |
2420 | struct css_set *src_cset; | |
2421 | int ret; | |
2422 | ||
2423 | lockdep_assert_held(&cgroup_tree_mutex); | |
2424 | lockdep_assert_held(&cgroup_mutex); | |
2425 | ||
2426 | /* look up all csses currently attached to @cgrp's subtree */ | |
2427 | down_read(&css_set_rwsem); | |
2428 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2429 | struct cgrp_cset_link *link; | |
2430 | ||
2431 | /* self is not affected by child_subsys_mask change */ | |
2432 | if (css->cgroup == cgrp) | |
2433 | continue; | |
2434 | ||
2435 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2436 | cgroup_migrate_add_src(link->cset, cgrp, | |
2437 | &preloaded_csets); | |
2438 | } | |
2439 | up_read(&css_set_rwsem); | |
2440 | ||
2441 | /* NULL dst indicates self on default hierarchy */ | |
2442 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2443 | if (ret) | |
2444 | goto out_finish; | |
2445 | ||
2446 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2447 | struct task_struct *last_task = NULL, *task; | |
2448 | ||
2449 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2450 | if (!src_cset->mg_src_cgrp) | |
2451 | break; | |
2452 | ||
2453 | /* | |
2454 | * All tasks in src_cset need to be migrated to the | |
2455 | * matching dst_cset. Empty it process by process. We | |
2456 | * walk tasks but migrate processes. The leader might even | |
2457 | * belong to a different cset but such src_cset would also | |
2458 | * be among the target src_csets because the default | |
2459 | * hierarchy enforces per-process membership. | |
2460 | */ | |
2461 | while (true) { | |
2462 | down_read(&css_set_rwsem); | |
2463 | task = list_first_entry_or_null(&src_cset->tasks, | |
2464 | struct task_struct, cg_list); | |
2465 | if (task) { | |
2466 | task = task->group_leader; | |
2467 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2468 | get_task_struct(task); | |
2469 | } | |
2470 | up_read(&css_set_rwsem); | |
2471 | ||
2472 | if (!task) | |
2473 | break; | |
2474 | ||
2475 | /* guard against possible infinite loop */ | |
2476 | if (WARN(last_task == task, | |
2477 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2478 | goto out_finish; | |
2479 | last_task = task; | |
2480 | ||
2481 | threadgroup_lock(task); | |
2482 | /* raced against de_thread() from another thread? */ | |
2483 | if (!thread_group_leader(task)) { | |
2484 | threadgroup_unlock(task); | |
2485 | put_task_struct(task); | |
2486 | continue; | |
2487 | } | |
2488 | ||
2489 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2490 | ||
2491 | threadgroup_unlock(task); | |
2492 | put_task_struct(task); | |
2493 | ||
2494 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2495 | goto out_finish; | |
2496 | } | |
2497 | } | |
2498 | ||
2499 | out_finish: | |
2500 | cgroup_migrate_finish(&preloaded_csets); | |
2501 | return ret; | |
2502 | } | |
2503 | ||
2504 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
2505 | static int cgroup_subtree_control_write(struct cgroup_subsys_state *dummy_css, | |
2506 | struct cftype *cft, char *buffer) | |
2507 | { | |
2508 | unsigned long enable_req = 0, disable_req = 0, enable, disable; | |
2509 | struct cgroup *cgrp = dummy_css->cgroup, *child; | |
2510 | struct cgroup_subsys *ss; | |
2511 | char *tok, *p; | |
2512 | int ssid, ret; | |
2513 | ||
2514 | /* | |
2515 | * Parse input - white space separated list of subsystem names | |
2516 | * prefixed with either + or -. | |
2517 | */ | |
2518 | p = buffer; | |
2519 | while ((tok = strsep(&p, " \t\n"))) { | |
2520 | for_each_subsys(ss, ssid) { | |
2521 | if (ss->disabled || strcmp(tok + 1, ss->name)) | |
2522 | continue; | |
2523 | ||
2524 | if (*tok == '+') { | |
2525 | enable_req |= 1 << ssid; | |
2526 | disable_req &= ~(1 << ssid); | |
2527 | } else if (*tok == '-') { | |
2528 | disable_req |= 1 << ssid; | |
2529 | enable_req &= ~(1 << ssid); | |
2530 | } else { | |
2531 | return -EINVAL; | |
2532 | } | |
2533 | break; | |
2534 | } | |
2535 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2536 | return -EINVAL; | |
2537 | } | |
2538 | ||
2539 | /* | |
2540 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2541 | * active_ref. cgroup_lock_live_group() already provides enough | |
2542 | * protection. Ensure @cgrp stays accessible and break the | |
2543 | * active_ref protection. | |
2544 | */ | |
2545 | cgroup_get(cgrp); | |
2546 | kernfs_break_active_protection(cgrp->control_kn); | |
2547 | retry: | |
2548 | enable = enable_req; | |
2549 | disable = disable_req; | |
2550 | ||
2551 | mutex_lock(&cgroup_tree_mutex); | |
2552 | ||
2553 | for_each_subsys(ss, ssid) { | |
2554 | if (enable & (1 << ssid)) { | |
2555 | if (cgrp->child_subsys_mask & (1 << ssid)) { | |
2556 | enable &= ~(1 << ssid); | |
2557 | continue; | |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * Because css offlining is asynchronous, userland | |
2562 | * might try to re-enable the same controller while | |
2563 | * the previous instance is still around. In such | |
2564 | * cases, wait till it's gone using offline_waitq. | |
2565 | */ | |
2566 | cgroup_for_each_live_child(child, cgrp) { | |
2567 | wait_queue_t wait; | |
2568 | ||
2569 | if (!cgroup_css(child, ss)) | |
2570 | continue; | |
2571 | ||
2572 | prepare_to_wait(&child->offline_waitq, &wait, | |
2573 | TASK_UNINTERRUPTIBLE); | |
2574 | mutex_unlock(&cgroup_tree_mutex); | |
2575 | schedule(); | |
2576 | finish_wait(&child->offline_waitq, &wait); | |
2577 | goto retry; | |
2578 | } | |
2579 | ||
2580 | /* unavailable or not enabled on the parent? */ | |
2581 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2582 | (cgrp->parent && | |
2583 | !(cgrp->parent->child_subsys_mask & (1 << ssid)))) { | |
2584 | ret = -ENOENT; | |
2585 | goto out_unlock_tree; | |
2586 | } | |
2587 | } else if (disable & (1 << ssid)) { | |
2588 | if (!(cgrp->child_subsys_mask & (1 << ssid))) { | |
2589 | disable &= ~(1 << ssid); | |
2590 | continue; | |
2591 | } | |
2592 | ||
2593 | /* a child has it enabled? */ | |
2594 | cgroup_for_each_live_child(child, cgrp) { | |
2595 | if (child->child_subsys_mask & (1 << ssid)) { | |
2596 | ret = -EBUSY; | |
2597 | goto out_unlock_tree; | |
2598 | } | |
2599 | } | |
2600 | } | |
2601 | } | |
2602 | ||
2603 | if (!enable && !disable) { | |
2604 | ret = 0; | |
2605 | goto out_unlock_tree; | |
2606 | } | |
2607 | ||
2608 | if (!cgroup_lock_live_group(cgrp)) { | |
2609 | ret = -ENODEV; | |
2610 | goto out_unlock_tree; | |
2611 | } | |
2612 | ||
2613 | /* | |
2614 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2615 | * with tasks so that child cgroups don't compete against tasks. | |
2616 | */ | |
2617 | if (enable && cgrp->parent && !list_empty(&cgrp->cset_links)) { | |
2618 | ret = -EBUSY; | |
2619 | goto out_unlock; | |
2620 | } | |
2621 | ||
2622 | /* | |
2623 | * Create csses for enables and update child_subsys_mask. This | |
2624 | * changes cgroup_e_css() results which in turn makes the | |
2625 | * subsequent cgroup_update_dfl_csses() associate all tasks in the | |
2626 | * subtree to the updated csses. | |
2627 | */ | |
2628 | for_each_subsys(ss, ssid) { | |
2629 | if (!(enable & (1 << ssid))) | |
2630 | continue; | |
2631 | ||
2632 | cgroup_for_each_live_child(child, cgrp) { | |
2633 | ret = create_css(child, ss); | |
2634 | if (ret) | |
2635 | goto err_undo_css; | |
2636 | } | |
2637 | } | |
2638 | ||
2639 | cgrp->child_subsys_mask |= enable; | |
2640 | cgrp->child_subsys_mask &= ~disable; | |
2641 | ||
2642 | ret = cgroup_update_dfl_csses(cgrp); | |
2643 | if (ret) | |
2644 | goto err_undo_css; | |
2645 | ||
2646 | /* all tasks are now migrated away from the old csses, kill them */ | |
2647 | for_each_subsys(ss, ssid) { | |
2648 | if (!(disable & (1 << ssid))) | |
2649 | continue; | |
2650 | ||
2651 | cgroup_for_each_live_child(child, cgrp) | |
2652 | kill_css(cgroup_css(child, ss)); | |
2653 | } | |
2654 | ||
2655 | kernfs_activate(cgrp->kn); | |
2656 | ret = 0; | |
2657 | out_unlock: | |
2658 | mutex_unlock(&cgroup_mutex); | |
2659 | out_unlock_tree: | |
2660 | mutex_unlock(&cgroup_tree_mutex); | |
2661 | kernfs_unbreak_active_protection(cgrp->control_kn); | |
2662 | cgroup_put(cgrp); | |
2663 | return ret; | |
2664 | ||
2665 | err_undo_css: | |
2666 | cgrp->child_subsys_mask &= ~enable; | |
2667 | cgrp->child_subsys_mask |= disable; | |
2668 | ||
2669 | for_each_subsys(ss, ssid) { | |
2670 | if (!(enable & (1 << ssid))) | |
2671 | continue; | |
2672 | ||
2673 | cgroup_for_each_live_child(child, cgrp) { | |
2674 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
2675 | if (css) | |
2676 | kill_css(css); | |
2677 | } | |
2678 | } | |
2679 | goto out_unlock; | |
2680 | } | |
2681 | ||
842b597e TH |
2682 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2683 | { | |
2684 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2685 | return 0; | |
2686 | } | |
2687 | ||
2bd59d48 TH |
2688 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2689 | size_t nbytes, loff_t off) | |
355e0c48 | 2690 | { |
2bd59d48 TH |
2691 | struct cgroup *cgrp = of->kn->parent->priv; |
2692 | struct cftype *cft = of->kn->priv; | |
2693 | struct cgroup_subsys_state *css; | |
a742c59d | 2694 | int ret; |
355e0c48 | 2695 | |
2bd59d48 TH |
2696 | /* |
2697 | * kernfs guarantees that a file isn't deleted with operations in | |
2698 | * flight, which means that the matching css is and stays alive and | |
2699 | * doesn't need to be pinned. The RCU locking is not necessary | |
2700 | * either. It's just for the convenience of using cgroup_css(). | |
2701 | */ | |
2702 | rcu_read_lock(); | |
2703 | css = cgroup_css(cgrp, cft->ss); | |
2704 | rcu_read_unlock(); | |
a742c59d TH |
2705 | |
2706 | if (cft->write_string) { | |
2707 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2708 | } else if (cft->write_u64) { | |
2709 | unsigned long long v; | |
2710 | ret = kstrtoull(buf, 0, &v); | |
2711 | if (!ret) | |
2712 | ret = cft->write_u64(css, cft, v); | |
2713 | } else if (cft->write_s64) { | |
2714 | long long v; | |
2715 | ret = kstrtoll(buf, 0, &v); | |
2716 | if (!ret) | |
2717 | ret = cft->write_s64(css, cft, v); | |
2718 | } else if (cft->trigger) { | |
2719 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2720 | } else { |
a742c59d | 2721 | ret = -EINVAL; |
e73d2c61 | 2722 | } |
2bd59d48 | 2723 | |
a742c59d | 2724 | return ret ?: nbytes; |
355e0c48 PM |
2725 | } |
2726 | ||
6612f05b | 2727 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2728 | { |
2bd59d48 | 2729 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2730 | } |
2731 | ||
6612f05b | 2732 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2733 | { |
2bd59d48 | 2734 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2735 | } |
2736 | ||
6612f05b | 2737 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2738 | { |
2bd59d48 | 2739 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2740 | } |
2741 | ||
91796569 | 2742 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2743 | { |
7da11279 TH |
2744 | struct cftype *cft = seq_cft(m); |
2745 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2746 | |
2da8ca82 TH |
2747 | if (cft->seq_show) |
2748 | return cft->seq_show(m, arg); | |
e73d2c61 | 2749 | |
f4c753b7 | 2750 | if (cft->read_u64) |
896f5199 TH |
2751 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2752 | else if (cft->read_s64) | |
2753 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2754 | else | |
2755 | return -EINVAL; | |
2756 | return 0; | |
91796569 PM |
2757 | } |
2758 | ||
2bd59d48 TH |
2759 | static struct kernfs_ops cgroup_kf_single_ops = { |
2760 | .atomic_write_len = PAGE_SIZE, | |
2761 | .write = cgroup_file_write, | |
2762 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2763 | }; |
2764 | ||
2bd59d48 TH |
2765 | static struct kernfs_ops cgroup_kf_ops = { |
2766 | .atomic_write_len = PAGE_SIZE, | |
2767 | .write = cgroup_file_write, | |
2768 | .seq_start = cgroup_seqfile_start, | |
2769 | .seq_next = cgroup_seqfile_next, | |
2770 | .seq_stop = cgroup_seqfile_stop, | |
2771 | .seq_show = cgroup_seqfile_show, | |
2772 | }; | |
ddbcc7e8 PM |
2773 | |
2774 | /* | |
2775 | * cgroup_rename - Only allow simple rename of directories in place. | |
2776 | */ | |
2bd59d48 TH |
2777 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2778 | const char *new_name_str) | |
ddbcc7e8 | 2779 | { |
2bd59d48 | 2780 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2781 | int ret; |
65dff759 | 2782 | |
2bd59d48 | 2783 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2784 | return -ENOTDIR; |
2bd59d48 | 2785 | if (kn->parent != new_parent) |
ddbcc7e8 | 2786 | return -EIO; |
65dff759 | 2787 | |
6db8e85c TH |
2788 | /* |
2789 | * This isn't a proper migration and its usefulness is very | |
2790 | * limited. Disallow if sane_behavior. | |
2791 | */ | |
2792 | if (cgroup_sane_behavior(cgrp)) | |
2793 | return -EPERM; | |
099fca32 | 2794 | |
e1b2dc17 TH |
2795 | /* |
2796 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2797 | * active_ref. kernfs_rename() doesn't require active_ref | |
2798 | * protection. Break them before grabbing cgroup_tree_mutex. | |
2799 | */ | |
2800 | kernfs_break_active_protection(new_parent); | |
2801 | kernfs_break_active_protection(kn); | |
099fca32 | 2802 | |
2bd59d48 TH |
2803 | mutex_lock(&cgroup_tree_mutex); |
2804 | mutex_lock(&cgroup_mutex); | |
099fca32 | 2805 | |
2bd59d48 | 2806 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2807 | |
2bd59d48 TH |
2808 | mutex_unlock(&cgroup_mutex); |
2809 | mutex_unlock(&cgroup_tree_mutex); | |
e1b2dc17 TH |
2810 | |
2811 | kernfs_unbreak_active_protection(kn); | |
2812 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2813 | return ret; |
099fca32 LZ |
2814 | } |
2815 | ||
49957f8e TH |
2816 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2817 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2818 | { | |
2819 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2820 | .ia_uid = current_fsuid(), | |
2821 | .ia_gid = current_fsgid(), }; | |
2822 | ||
2823 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2824 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2825 | return 0; | |
2826 | ||
2827 | return kernfs_setattr(kn, &iattr); | |
2828 | } | |
2829 | ||
2bb566cb | 2830 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2831 | { |
8d7e6fb0 | 2832 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2833 | struct kernfs_node *kn; |
2834 | struct lock_class_key *key = NULL; | |
49957f8e | 2835 | int ret; |
05ef1d7c | 2836 | |
2bd59d48 TH |
2837 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2838 | key = &cft->lockdep_key; | |
2839 | #endif | |
2840 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2841 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2842 | NULL, false, key); | |
49957f8e TH |
2843 | if (IS_ERR(kn)) |
2844 | return PTR_ERR(kn); | |
2845 | ||
2846 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 2847 | if (ret) { |
49957f8e | 2848 | kernfs_remove(kn); |
f8f22e53 TH |
2849 | return ret; |
2850 | } | |
2851 | ||
2852 | if (cft->seq_show == cgroup_subtree_control_show) | |
2853 | cgrp->control_kn = kn; | |
842b597e TH |
2854 | else if (cft->seq_show == cgroup_populated_show) |
2855 | cgrp->populated_kn = kn; | |
f8f22e53 | 2856 | return 0; |
ddbcc7e8 PM |
2857 | } |
2858 | ||
b1f28d31 TH |
2859 | /** |
2860 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2861 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2862 | * @cfts: array of cftypes to be added |
2863 | * @is_add: whether to add or remove | |
2864 | * | |
2865 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2866 | * For removals, this function never fails. If addition fails, this |
2867 | * function doesn't remove files already added. The caller is responsible | |
2868 | * for cleaning up. | |
b1f28d31 | 2869 | */ |
2bb566cb TH |
2870 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2871 | bool is_add) | |
ddbcc7e8 | 2872 | { |
03b1cde6 | 2873 | struct cftype *cft; |
b1f28d31 TH |
2874 | int ret; |
2875 | ||
ace2bee8 | 2876 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2877 | |
2878 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2879 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2880 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2881 | continue; | |
873fe09e TH |
2882 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2883 | continue; | |
f33fddc2 G |
2884 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2885 | continue; | |
2886 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2887 | continue; | |
2888 | ||
2739d3cc | 2889 | if (is_add) { |
2bb566cb | 2890 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2891 | if (ret) { |
2739d3cc | 2892 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2893 | cft->name, ret); |
2894 | return ret; | |
2895 | } | |
2739d3cc LZ |
2896 | } else { |
2897 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2898 | } |
ddbcc7e8 | 2899 | } |
b1f28d31 | 2900 | return 0; |
ddbcc7e8 PM |
2901 | } |
2902 | ||
21a2d343 | 2903 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2904 | { |
2905 | LIST_HEAD(pending); | |
2bb566cb | 2906 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2907 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2908 | struct cgroup_subsys_state *css; |
9ccece80 | 2909 | int ret = 0; |
8e3f6541 | 2910 | |
21a2d343 | 2911 | lockdep_assert_held(&cgroup_tree_mutex); |
8e3f6541 | 2912 | |
e8c82d20 | 2913 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2914 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2915 | struct cgroup *cgrp = css->cgroup; |
2916 | ||
e8c82d20 LZ |
2917 | if (cgroup_is_dead(cgrp)) |
2918 | continue; | |
2919 | ||
21a2d343 | 2920 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2921 | if (ret) |
2922 | break; | |
8e3f6541 | 2923 | } |
21a2d343 TH |
2924 | |
2925 | if (is_add && !ret) | |
2926 | kernfs_activate(root->kn); | |
9ccece80 | 2927 | return ret; |
8e3f6541 TH |
2928 | } |
2929 | ||
2da440a2 | 2930 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2931 | { |
2bb566cb | 2932 | struct cftype *cft; |
8e3f6541 | 2933 | |
2bd59d48 TH |
2934 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2935 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2936 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2937 | kfree(cft->kf_ops); | |
2938 | cft->kf_ops = NULL; | |
2da440a2 | 2939 | cft->ss = NULL; |
2bd59d48 | 2940 | } |
2da440a2 TH |
2941 | } |
2942 | ||
2bd59d48 | 2943 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2944 | { |
2945 | struct cftype *cft; | |
2946 | ||
2bd59d48 TH |
2947 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2948 | struct kernfs_ops *kf_ops; | |
2949 | ||
0adb0704 TH |
2950 | WARN_ON(cft->ss || cft->kf_ops); |
2951 | ||
2bd59d48 TH |
2952 | if (cft->seq_start) |
2953 | kf_ops = &cgroup_kf_ops; | |
2954 | else | |
2955 | kf_ops = &cgroup_kf_single_ops; | |
2956 | ||
2957 | /* | |
2958 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2959 | * make a copy of kf_ops to set its atomic_write_len. | |
2960 | */ | |
2961 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2962 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2963 | if (!kf_ops) { | |
2964 | cgroup_exit_cftypes(cfts); | |
2965 | return -ENOMEM; | |
2966 | } | |
2967 | kf_ops->atomic_write_len = cft->max_write_len; | |
2968 | } | |
8e3f6541 | 2969 | |
2bd59d48 | 2970 | cft->kf_ops = kf_ops; |
2bb566cb | 2971 | cft->ss = ss; |
2bd59d48 | 2972 | } |
2bb566cb | 2973 | |
2bd59d48 | 2974 | return 0; |
2da440a2 TH |
2975 | } |
2976 | ||
21a2d343 TH |
2977 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2978 | { | |
2979 | lockdep_assert_held(&cgroup_tree_mutex); | |
2980 | ||
2981 | if (!cfts || !cfts[0].ss) | |
2982 | return -ENOENT; | |
2983 | ||
2984 | list_del(&cfts->node); | |
2985 | cgroup_apply_cftypes(cfts, false); | |
2986 | cgroup_exit_cftypes(cfts); | |
2987 | return 0; | |
8e3f6541 | 2988 | } |
8e3f6541 | 2989 | |
79578621 TH |
2990 | /** |
2991 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2992 | * @cfts: zero-length name terminated array of cftypes |
2993 | * | |
2bb566cb TH |
2994 | * Unregister @cfts. Files described by @cfts are removed from all |
2995 | * existing cgroups and all future cgroups won't have them either. This | |
2996 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2997 | * |
2998 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2999 | * registered. |
79578621 | 3000 | */ |
2bb566cb | 3001 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3002 | { |
21a2d343 | 3003 | int ret; |
79578621 | 3004 | |
21a2d343 TH |
3005 | mutex_lock(&cgroup_tree_mutex); |
3006 | ret = cgroup_rm_cftypes_locked(cfts); | |
3007 | mutex_unlock(&cgroup_tree_mutex); | |
3008 | return ret; | |
80b13586 TH |
3009 | } |
3010 | ||
8e3f6541 TH |
3011 | /** |
3012 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3013 | * @ss: target cgroup subsystem | |
3014 | * @cfts: zero-length name terminated array of cftypes | |
3015 | * | |
3016 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3017 | * existing cgroups to which @ss is attached and all future cgroups will | |
3018 | * have them too. This function can be called anytime whether @ss is | |
3019 | * attached or not. | |
3020 | * | |
3021 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3022 | * function currently returns 0 as long as @cfts registration is successful | |
3023 | * even if some file creation attempts on existing cgroups fail. | |
3024 | */ | |
03b1cde6 | 3025 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3026 | { |
9ccece80 | 3027 | int ret; |
8e3f6541 | 3028 | |
dc5736ed LZ |
3029 | if (!cfts || cfts[0].name[0] == '\0') |
3030 | return 0; | |
2bb566cb | 3031 | |
2bd59d48 TH |
3032 | ret = cgroup_init_cftypes(ss, cfts); |
3033 | if (ret) | |
3034 | return ret; | |
79578621 | 3035 | |
21a2d343 TH |
3036 | mutex_lock(&cgroup_tree_mutex); |
3037 | ||
0adb0704 | 3038 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3039 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3040 | if (ret) |
21a2d343 | 3041 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3042 | |
21a2d343 | 3043 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 3044 | return ret; |
79578621 TH |
3045 | } |
3046 | ||
a043e3b2 LZ |
3047 | /** |
3048 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3049 | * @cgrp: the cgroup in question | |
3050 | * | |
3051 | * Return the number of tasks in the cgroup. | |
3052 | */ | |
07bc356e | 3053 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3054 | { |
3055 | int count = 0; | |
69d0206c | 3056 | struct cgrp_cset_link *link; |
817929ec | 3057 | |
96d365e0 | 3058 | down_read(&css_set_rwsem); |
69d0206c TH |
3059 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3060 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3061 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3062 | return count; |
3063 | } | |
3064 | ||
53fa5261 | 3065 | /** |
492eb21b TH |
3066 | * css_next_child - find the next child of a given css |
3067 | * @pos_css: the current position (%NULL to initiate traversal) | |
3068 | * @parent_css: css whose children to walk | |
53fa5261 | 3069 | * |
492eb21b | 3070 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
3071 | * under either cgroup_mutex or RCU read lock. The only requirement is |
3072 | * that @parent_css and @pos_css are accessible. The next sibling is | |
3073 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 3074 | */ |
492eb21b TH |
3075 | struct cgroup_subsys_state * |
3076 | css_next_child(struct cgroup_subsys_state *pos_css, | |
3077 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 3078 | { |
492eb21b TH |
3079 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
3080 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
3081 | struct cgroup *next; |
3082 | ||
ace2bee8 | 3083 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
3084 | |
3085 | /* | |
3086 | * @pos could already have been removed. Once a cgroup is removed, | |
3087 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
3088 | * changes. As CGRP_DEAD assertion is serialized and happens |
3089 | * before the cgroup is taken off the ->sibling list, if we see it | |
3090 | * unasserted, it's guaranteed that the next sibling hasn't | |
3091 | * finished its grace period even if it's already removed, and thus | |
3092 | * safe to dereference from this RCU critical section. If | |
3093 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
3094 | * to be visible as %true here. | |
3b287a50 TH |
3095 | * |
3096 | * If @pos is dead, its next pointer can't be dereferenced; | |
3097 | * however, as each cgroup is given a monotonically increasing | |
3098 | * unique serial number and always appended to the sibling list, | |
3099 | * the next one can be found by walking the parent's children until | |
3100 | * we see a cgroup with higher serial number than @pos's. While | |
3101 | * this path can be slower, it's taken only when either the current | |
3102 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 3103 | */ |
3b287a50 TH |
3104 | if (!pos) { |
3105 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
3106 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 3107 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
3108 | } else { |
3109 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
3110 | if (next->serial_nr > pos->serial_nr) | |
3111 | break; | |
53fa5261 TH |
3112 | } |
3113 | ||
3b281afb TH |
3114 | /* |
3115 | * @next, if not pointing to the head, can be dereferenced and is | |
3116 | * the next sibling; however, it might have @ss disabled. If so, | |
3117 | * fast-forward to the next enabled one. | |
3118 | */ | |
3119 | while (&next->sibling != &cgrp->children) { | |
3120 | struct cgroup_subsys_state *next_css = cgroup_css(next, parent_css->ss); | |
492eb21b | 3121 | |
3b281afb TH |
3122 | if (next_css) |
3123 | return next_css; | |
3124 | next = list_entry_rcu(next->sibling.next, struct cgroup, sibling); | |
3125 | } | |
3126 | return NULL; | |
53fa5261 | 3127 | } |
53fa5261 | 3128 | |
574bd9f7 | 3129 | /** |
492eb21b | 3130 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3131 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3132 | * @root: css whose descendants to walk |
574bd9f7 | 3133 | * |
492eb21b | 3134 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3135 | * to visit for pre-order traversal of @root's descendants. @root is |
3136 | * included in the iteration and the first node to be visited. | |
75501a6d | 3137 | * |
87fb54f1 TH |
3138 | * While this function requires cgroup_mutex or RCU read locking, it |
3139 | * doesn't require the whole traversal to be contained in a single critical | |
3140 | * section. This function will return the correct next descendant as long | |
3141 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 3142 | */ |
492eb21b TH |
3143 | struct cgroup_subsys_state * |
3144 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3145 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3146 | { |
492eb21b | 3147 | struct cgroup_subsys_state *next; |
574bd9f7 | 3148 | |
ace2bee8 | 3149 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3150 | |
bd8815a6 | 3151 | /* if first iteration, visit @root */ |
7805d000 | 3152 | if (!pos) |
bd8815a6 | 3153 | return root; |
574bd9f7 TH |
3154 | |
3155 | /* visit the first child if exists */ | |
492eb21b | 3156 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3157 | if (next) |
3158 | return next; | |
3159 | ||
3160 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
3161 | while (pos != root) { |
3162 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 3163 | if (next) |
574bd9f7 | 3164 | return next; |
492eb21b | 3165 | pos = css_parent(pos); |
7805d000 | 3166 | } |
574bd9f7 TH |
3167 | |
3168 | return NULL; | |
3169 | } | |
574bd9f7 | 3170 | |
12a9d2fe | 3171 | /** |
492eb21b TH |
3172 | * css_rightmost_descendant - return the rightmost descendant of a css |
3173 | * @pos: css of interest | |
12a9d2fe | 3174 | * |
492eb21b TH |
3175 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3176 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3177 | * subtree of @pos. |
75501a6d | 3178 | * |
87fb54f1 TH |
3179 | * While this function requires cgroup_mutex or RCU read locking, it |
3180 | * doesn't require the whole traversal to be contained in a single critical | |
3181 | * section. This function will return the correct rightmost descendant as | |
3182 | * long as @pos is accessible. | |
12a9d2fe | 3183 | */ |
492eb21b TH |
3184 | struct cgroup_subsys_state * |
3185 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3186 | { |
492eb21b | 3187 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3188 | |
ace2bee8 | 3189 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
3190 | |
3191 | do { | |
3192 | last = pos; | |
3193 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3194 | pos = NULL; | |
492eb21b | 3195 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3196 | pos = tmp; |
3197 | } while (pos); | |
3198 | ||
3199 | return last; | |
3200 | } | |
12a9d2fe | 3201 | |
492eb21b TH |
3202 | static struct cgroup_subsys_state * |
3203 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3204 | { |
492eb21b | 3205 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3206 | |
3207 | do { | |
3208 | last = pos; | |
492eb21b | 3209 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3210 | } while (pos); |
3211 | ||
3212 | return last; | |
3213 | } | |
3214 | ||
3215 | /** | |
492eb21b | 3216 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3217 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3218 | * @root: css whose descendants to walk |
574bd9f7 | 3219 | * |
492eb21b | 3220 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3221 | * to visit for post-order traversal of @root's descendants. @root is |
3222 | * included in the iteration and the last node to be visited. | |
75501a6d | 3223 | * |
87fb54f1 TH |
3224 | * While this function requires cgroup_mutex or RCU read locking, it |
3225 | * doesn't require the whole traversal to be contained in a single critical | |
3226 | * section. This function will return the correct next descendant as long | |
3227 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3228 | * @cgroup. | |
574bd9f7 | 3229 | */ |
492eb21b TH |
3230 | struct cgroup_subsys_state * |
3231 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3232 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3233 | { |
492eb21b | 3234 | struct cgroup_subsys_state *next; |
574bd9f7 | 3235 | |
ace2bee8 | 3236 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3237 | |
58b79a91 TH |
3238 | /* if first iteration, visit leftmost descendant which may be @root */ |
3239 | if (!pos) | |
3240 | return css_leftmost_descendant(root); | |
574bd9f7 | 3241 | |
bd8815a6 TH |
3242 | /* if we visited @root, we're done */ |
3243 | if (pos == root) | |
3244 | return NULL; | |
3245 | ||
574bd9f7 | 3246 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3247 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3248 | if (next) |
492eb21b | 3249 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3250 | |
3251 | /* no sibling left, visit parent */ | |
bd8815a6 | 3252 | return css_parent(pos); |
574bd9f7 | 3253 | } |
574bd9f7 | 3254 | |
0942eeee | 3255 | /** |
72ec7029 | 3256 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3257 | * @it: the iterator to advance |
3258 | * | |
3259 | * Advance @it to the next css_set to walk. | |
d515876e | 3260 | */ |
72ec7029 | 3261 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3262 | { |
0f0a2b4f | 3263 | struct list_head *l = it->cset_pos; |
d515876e TH |
3264 | struct cgrp_cset_link *link; |
3265 | struct css_set *cset; | |
3266 | ||
3267 | /* Advance to the next non-empty css_set */ | |
3268 | do { | |
3269 | l = l->next; | |
0f0a2b4f TH |
3270 | if (l == it->cset_head) { |
3271 | it->cset_pos = NULL; | |
d515876e TH |
3272 | return; |
3273 | } | |
3ebb2b6e TH |
3274 | |
3275 | if (it->ss) { | |
3276 | cset = container_of(l, struct css_set, | |
3277 | e_cset_node[it->ss->id]); | |
3278 | } else { | |
3279 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3280 | cset = link->cset; | |
3281 | } | |
c7561128 TH |
3282 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3283 | ||
0f0a2b4f | 3284 | it->cset_pos = l; |
c7561128 TH |
3285 | |
3286 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3287 | it->task_pos = cset->tasks.next; |
c7561128 | 3288 | else |
0f0a2b4f TH |
3289 | it->task_pos = cset->mg_tasks.next; |
3290 | ||
3291 | it->tasks_head = &cset->tasks; | |
3292 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3293 | } |
3294 | ||
0942eeee | 3295 | /** |
72ec7029 TH |
3296 | * css_task_iter_start - initiate task iteration |
3297 | * @css: the css to walk tasks of | |
0942eeee TH |
3298 | * @it: the task iterator to use |
3299 | * | |
72ec7029 TH |
3300 | * Initiate iteration through the tasks of @css. The caller can call |
3301 | * css_task_iter_next() to walk through the tasks until the function | |
3302 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3303 | * called. | |
0942eeee TH |
3304 | * |
3305 | * Note that this function acquires a lock which is released when the | |
3306 | * iteration finishes. The caller can't sleep while iteration is in | |
3307 | * progress. | |
3308 | */ | |
72ec7029 TH |
3309 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3310 | struct css_task_iter *it) | |
96d365e0 | 3311 | __acquires(css_set_rwsem) |
817929ec | 3312 | { |
56fde9e0 TH |
3313 | /* no one should try to iterate before mounting cgroups */ |
3314 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3315 | |
96d365e0 | 3316 | down_read(&css_set_rwsem); |
c59cd3d8 | 3317 | |
3ebb2b6e TH |
3318 | it->ss = css->ss; |
3319 | ||
3320 | if (it->ss) | |
3321 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3322 | else | |
3323 | it->cset_pos = &css->cgroup->cset_links; | |
3324 | ||
0f0a2b4f | 3325 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3326 | |
72ec7029 | 3327 | css_advance_task_iter(it); |
817929ec PM |
3328 | } |
3329 | ||
0942eeee | 3330 | /** |
72ec7029 | 3331 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3332 | * @it: the task iterator being iterated |
3333 | * | |
3334 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3335 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3336 | * reaches the end. | |
0942eeee | 3337 | */ |
72ec7029 | 3338 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3339 | { |
3340 | struct task_struct *res; | |
0f0a2b4f | 3341 | struct list_head *l = it->task_pos; |
817929ec PM |
3342 | |
3343 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3344 | if (!it->cset_pos) |
817929ec PM |
3345 | return NULL; |
3346 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3347 | |
3348 | /* | |
3349 | * Advance iterator to find next entry. cset->tasks is consumed | |
3350 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3351 | * next cset. | |
3352 | */ | |
817929ec | 3353 | l = l->next; |
c7561128 | 3354 | |
0f0a2b4f TH |
3355 | if (l == it->tasks_head) |
3356 | l = it->mg_tasks_head->next; | |
c7561128 | 3357 | |
0f0a2b4f | 3358 | if (l == it->mg_tasks_head) |
72ec7029 | 3359 | css_advance_task_iter(it); |
c7561128 | 3360 | else |
0f0a2b4f | 3361 | it->task_pos = l; |
c7561128 | 3362 | |
817929ec PM |
3363 | return res; |
3364 | } | |
3365 | ||
0942eeee | 3366 | /** |
72ec7029 | 3367 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3368 | * @it: the task iterator to finish |
3369 | * | |
72ec7029 | 3370 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3371 | */ |
72ec7029 | 3372 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3373 | __releases(css_set_rwsem) |
31a7df01 | 3374 | { |
96d365e0 | 3375 | up_read(&css_set_rwsem); |
31a7df01 CW |
3376 | } |
3377 | ||
3378 | /** | |
8cc99345 TH |
3379 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3380 | * @to: cgroup to which the tasks will be moved | |
3381 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3382 | * |
eaf797ab TH |
3383 | * Locking rules between cgroup_post_fork() and the migration path |
3384 | * guarantee that, if a task is forking while being migrated, the new child | |
3385 | * is guaranteed to be either visible in the source cgroup after the | |
3386 | * parent's migration is complete or put into the target cgroup. No task | |
3387 | * can slip out of migration through forking. | |
31a7df01 | 3388 | */ |
8cc99345 | 3389 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3390 | { |
952aaa12 TH |
3391 | LIST_HEAD(preloaded_csets); |
3392 | struct cgrp_cset_link *link; | |
72ec7029 | 3393 | struct css_task_iter it; |
e406d1cf | 3394 | struct task_struct *task; |
952aaa12 | 3395 | int ret; |
31a7df01 | 3396 | |
952aaa12 | 3397 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3398 | |
952aaa12 TH |
3399 | /* all tasks in @from are being moved, all csets are source */ |
3400 | down_read(&css_set_rwsem); | |
3401 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3402 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3403 | up_read(&css_set_rwsem); | |
31a7df01 | 3404 | |
952aaa12 TH |
3405 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3406 | if (ret) | |
3407 | goto out_err; | |
8cc99345 | 3408 | |
952aaa12 TH |
3409 | /* |
3410 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3411 | * ->can_attach() fails. | |
3412 | */ | |
e406d1cf TH |
3413 | do { |
3414 | css_task_iter_start(&from->dummy_css, &it); | |
3415 | task = css_task_iter_next(&it); | |
3416 | if (task) | |
3417 | get_task_struct(task); | |
3418 | css_task_iter_end(&it); | |
3419 | ||
3420 | if (task) { | |
952aaa12 | 3421 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3422 | put_task_struct(task); |
3423 | } | |
3424 | } while (task && !ret); | |
952aaa12 TH |
3425 | out_err: |
3426 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3427 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3428 | return ret; |
8cc99345 TH |
3429 | } |
3430 | ||
bbcb81d0 | 3431 | /* |
102a775e | 3432 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3433 | * |
3434 | * Reading this file can return large amounts of data if a cgroup has | |
3435 | * *lots* of attached tasks. So it may need several calls to read(), | |
3436 | * but we cannot guarantee that the information we produce is correct | |
3437 | * unless we produce it entirely atomically. | |
3438 | * | |
bbcb81d0 | 3439 | */ |
bbcb81d0 | 3440 | |
24528255 LZ |
3441 | /* which pidlist file are we talking about? */ |
3442 | enum cgroup_filetype { | |
3443 | CGROUP_FILE_PROCS, | |
3444 | CGROUP_FILE_TASKS, | |
3445 | }; | |
3446 | ||
3447 | /* | |
3448 | * A pidlist is a list of pids that virtually represents the contents of one | |
3449 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3450 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3451 | * to the cgroup. | |
3452 | */ | |
3453 | struct cgroup_pidlist { | |
3454 | /* | |
3455 | * used to find which pidlist is wanted. doesn't change as long as | |
3456 | * this particular list stays in the list. | |
3457 | */ | |
3458 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3459 | /* array of xids */ | |
3460 | pid_t *list; | |
3461 | /* how many elements the above list has */ | |
3462 | int length; | |
24528255 LZ |
3463 | /* each of these stored in a list by its cgroup */ |
3464 | struct list_head links; | |
3465 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3466 | struct cgroup *owner; | |
b1a21367 TH |
3467 | /* for delayed destruction */ |
3468 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3469 | }; |
3470 | ||
d1d9fd33 BB |
3471 | /* |
3472 | * The following two functions "fix" the issue where there are more pids | |
3473 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3474 | * TODO: replace with a kernel-wide solution to this problem | |
3475 | */ | |
3476 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3477 | static void *pidlist_allocate(int count) | |
3478 | { | |
3479 | if (PIDLIST_TOO_LARGE(count)) | |
3480 | return vmalloc(count * sizeof(pid_t)); | |
3481 | else | |
3482 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3483 | } | |
b1a21367 | 3484 | |
d1d9fd33 BB |
3485 | static void pidlist_free(void *p) |
3486 | { | |
3487 | if (is_vmalloc_addr(p)) | |
3488 | vfree(p); | |
3489 | else | |
3490 | kfree(p); | |
3491 | } | |
d1d9fd33 | 3492 | |
b1a21367 TH |
3493 | /* |
3494 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3495 | * should be left afterwards. | |
3496 | */ | |
3497 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3498 | { | |
3499 | struct cgroup_pidlist *l, *tmp_l; | |
3500 | ||
3501 | mutex_lock(&cgrp->pidlist_mutex); | |
3502 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3503 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3504 | mutex_unlock(&cgrp->pidlist_mutex); | |
3505 | ||
3506 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3507 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3508 | } | |
3509 | ||
3510 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3511 | { | |
3512 | struct delayed_work *dwork = to_delayed_work(work); | |
3513 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3514 | destroy_dwork); | |
3515 | struct cgroup_pidlist *tofree = NULL; | |
3516 | ||
3517 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3518 | |
3519 | /* | |
04502365 TH |
3520 | * Destroy iff we didn't get queued again. The state won't change |
3521 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3522 | */ |
04502365 | 3523 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3524 | list_del(&l->links); |
3525 | pidlist_free(l->list); | |
3526 | put_pid_ns(l->key.ns); | |
3527 | tofree = l; | |
3528 | } | |
3529 | ||
b1a21367 TH |
3530 | mutex_unlock(&l->owner->pidlist_mutex); |
3531 | kfree(tofree); | |
3532 | } | |
3533 | ||
bbcb81d0 | 3534 | /* |
102a775e | 3535 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3536 | * Returns the number of unique elements. |
bbcb81d0 | 3537 | */ |
6ee211ad | 3538 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3539 | { |
102a775e | 3540 | int src, dest = 1; |
102a775e BB |
3541 | |
3542 | /* | |
3543 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3544 | * edge cases first; no work needs to be done for either | |
3545 | */ | |
3546 | if (length == 0 || length == 1) | |
3547 | return length; | |
3548 | /* src and dest walk down the list; dest counts unique elements */ | |
3549 | for (src = 1; src < length; src++) { | |
3550 | /* find next unique element */ | |
3551 | while (list[src] == list[src-1]) { | |
3552 | src++; | |
3553 | if (src == length) | |
3554 | goto after; | |
3555 | } | |
3556 | /* dest always points to where the next unique element goes */ | |
3557 | list[dest] = list[src]; | |
3558 | dest++; | |
3559 | } | |
3560 | after: | |
102a775e BB |
3561 | return dest; |
3562 | } | |
3563 | ||
afb2bc14 TH |
3564 | /* |
3565 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3566 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3567 | * different per namespace, each namespace needs differently sorted list, | |
3568 | * making it impossible to use, for example, single rbtree of member tasks | |
3569 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3570 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3571 | * pidlists keyed by cgroup and namespace. | |
3572 | * | |
3573 | * All this extra complexity was caused by the original implementation | |
3574 | * committing to an entirely unnecessary property. In the long term, we | |
3575 | * want to do away with it. Explicitly scramble sort order if | |
3576 | * sane_behavior so that no such expectation exists in the new interface. | |
3577 | * | |
3578 | * Scrambling is done by swapping every two consecutive bits, which is | |
3579 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3580 | */ | |
3581 | static pid_t pid_fry(pid_t pid) | |
3582 | { | |
3583 | unsigned a = pid & 0x55555555; | |
3584 | unsigned b = pid & 0xAAAAAAAA; | |
3585 | ||
3586 | return (a << 1) | (b >> 1); | |
3587 | } | |
3588 | ||
3589 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3590 | { | |
3591 | if (cgroup_sane_behavior(cgrp)) | |
3592 | return pid_fry(pid); | |
3593 | else | |
3594 | return pid; | |
3595 | } | |
3596 | ||
102a775e BB |
3597 | static int cmppid(const void *a, const void *b) |
3598 | { | |
3599 | return *(pid_t *)a - *(pid_t *)b; | |
3600 | } | |
3601 | ||
afb2bc14 TH |
3602 | static int fried_cmppid(const void *a, const void *b) |
3603 | { | |
3604 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3605 | } | |
3606 | ||
e6b81710 TH |
3607 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3608 | enum cgroup_filetype type) | |
3609 | { | |
3610 | struct cgroup_pidlist *l; | |
3611 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3612 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3613 | ||
3614 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3615 | ||
3616 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3617 | if (l->key.type == type && l->key.ns == ns) | |
3618 | return l; | |
3619 | return NULL; | |
3620 | } | |
3621 | ||
72a8cb30 BB |
3622 | /* |
3623 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3624 | * returns with the lock on that pidlist already held, and takes care | |
3625 | * of the use count, or returns NULL with no locks held if we're out of | |
3626 | * memory. | |
3627 | */ | |
e6b81710 TH |
3628 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3629 | enum cgroup_filetype type) | |
72a8cb30 BB |
3630 | { |
3631 | struct cgroup_pidlist *l; | |
b70cc5fd | 3632 | |
e6b81710 TH |
3633 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3634 | ||
3635 | l = cgroup_pidlist_find(cgrp, type); | |
3636 | if (l) | |
3637 | return l; | |
3638 | ||
72a8cb30 | 3639 | /* entry not found; create a new one */ |
f4f4be2b | 3640 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3641 | if (!l) |
72a8cb30 | 3642 | return l; |
e6b81710 | 3643 | |
b1a21367 | 3644 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3645 | l->key.type = type; |
e6b81710 TH |
3646 | /* don't need task_nsproxy() if we're looking at ourself */ |
3647 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3648 | l->owner = cgrp; |
3649 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3650 | return l; |
3651 | } | |
3652 | ||
102a775e BB |
3653 | /* |
3654 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3655 | */ | |
72a8cb30 BB |
3656 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3657 | struct cgroup_pidlist **lp) | |
102a775e BB |
3658 | { |
3659 | pid_t *array; | |
3660 | int length; | |
3661 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3662 | struct css_task_iter it; |
817929ec | 3663 | struct task_struct *tsk; |
102a775e BB |
3664 | struct cgroup_pidlist *l; |
3665 | ||
4bac00d1 TH |
3666 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3667 | ||
102a775e BB |
3668 | /* |
3669 | * If cgroup gets more users after we read count, we won't have | |
3670 | * enough space - tough. This race is indistinguishable to the | |
3671 | * caller from the case that the additional cgroup users didn't | |
3672 | * show up until sometime later on. | |
3673 | */ | |
3674 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3675 | array = pidlist_allocate(length); |
102a775e BB |
3676 | if (!array) |
3677 | return -ENOMEM; | |
3678 | /* now, populate the array */ | |
72ec7029 TH |
3679 | css_task_iter_start(&cgrp->dummy_css, &it); |
3680 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3681 | if (unlikely(n == length)) |
817929ec | 3682 | break; |
102a775e | 3683 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3684 | if (type == CGROUP_FILE_PROCS) |
3685 | pid = task_tgid_vnr(tsk); | |
3686 | else | |
3687 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3688 | if (pid > 0) /* make sure to only use valid results */ |
3689 | array[n++] = pid; | |
817929ec | 3690 | } |
72ec7029 | 3691 | css_task_iter_end(&it); |
102a775e BB |
3692 | length = n; |
3693 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3694 | if (cgroup_sane_behavior(cgrp)) |
3695 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3696 | else | |
3697 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3698 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3699 | length = pidlist_uniq(array, length); |
e6b81710 | 3700 | |
e6b81710 | 3701 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3702 | if (!l) { |
e6b81710 | 3703 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3704 | pidlist_free(array); |
72a8cb30 | 3705 | return -ENOMEM; |
102a775e | 3706 | } |
e6b81710 TH |
3707 | |
3708 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3709 | pidlist_free(l->list); |
102a775e BB |
3710 | l->list = array; |
3711 | l->length = length; | |
72a8cb30 | 3712 | *lp = l; |
102a775e | 3713 | return 0; |
bbcb81d0 PM |
3714 | } |
3715 | ||
846c7bb0 | 3716 | /** |
a043e3b2 | 3717 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3718 | * @stats: cgroupstats to fill information into |
3719 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3720 | * been requested. | |
a043e3b2 LZ |
3721 | * |
3722 | * Build and fill cgroupstats so that taskstats can export it to user | |
3723 | * space. | |
846c7bb0 BS |
3724 | */ |
3725 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3726 | { | |
2bd59d48 | 3727 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3728 | struct cgroup *cgrp; |
72ec7029 | 3729 | struct css_task_iter it; |
846c7bb0 | 3730 | struct task_struct *tsk; |
33d283be | 3731 | |
2bd59d48 TH |
3732 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3733 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3734 | kernfs_type(kn) != KERNFS_DIR) | |
3735 | return -EINVAL; | |
3736 | ||
bad34660 LZ |
3737 | mutex_lock(&cgroup_mutex); |
3738 | ||
846c7bb0 | 3739 | /* |
2bd59d48 TH |
3740 | * We aren't being called from kernfs and there's no guarantee on |
3741 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3742 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3743 | */ |
2bd59d48 TH |
3744 | rcu_read_lock(); |
3745 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3746 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3747 | rcu_read_unlock(); |
bad34660 | 3748 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3749 | return -ENOENT; |
3750 | } | |
bad34660 | 3751 | rcu_read_unlock(); |
846c7bb0 | 3752 | |
72ec7029 TH |
3753 | css_task_iter_start(&cgrp->dummy_css, &it); |
3754 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3755 | switch (tsk->state) { |
3756 | case TASK_RUNNING: | |
3757 | stats->nr_running++; | |
3758 | break; | |
3759 | case TASK_INTERRUPTIBLE: | |
3760 | stats->nr_sleeping++; | |
3761 | break; | |
3762 | case TASK_UNINTERRUPTIBLE: | |
3763 | stats->nr_uninterruptible++; | |
3764 | break; | |
3765 | case TASK_STOPPED: | |
3766 | stats->nr_stopped++; | |
3767 | break; | |
3768 | default: | |
3769 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3770 | stats->nr_io_wait++; | |
3771 | break; | |
3772 | } | |
3773 | } | |
72ec7029 | 3774 | css_task_iter_end(&it); |
846c7bb0 | 3775 | |
bad34660 | 3776 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3777 | return 0; |
846c7bb0 BS |
3778 | } |
3779 | ||
8f3ff208 | 3780 | |
bbcb81d0 | 3781 | /* |
102a775e | 3782 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3783 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3784 | * in the cgroup->l->list array. |
bbcb81d0 | 3785 | */ |
cc31edce | 3786 | |
102a775e | 3787 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3788 | { |
cc31edce PM |
3789 | /* |
3790 | * Initially we receive a position value that corresponds to | |
3791 | * one more than the last pid shown (or 0 on the first call or | |
3792 | * after a seek to the start). Use a binary-search to find the | |
3793 | * next pid to display, if any | |
3794 | */ | |
2bd59d48 | 3795 | struct kernfs_open_file *of = s->private; |
7da11279 | 3796 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3797 | struct cgroup_pidlist *l; |
7da11279 | 3798 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3799 | int index = 0, pid = *pos; |
4bac00d1 TH |
3800 | int *iter, ret; |
3801 | ||
3802 | mutex_lock(&cgrp->pidlist_mutex); | |
3803 | ||
3804 | /* | |
5d22444f | 3805 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3806 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3807 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3808 | * could already have been destroyed. |
3809 | */ | |
5d22444f TH |
3810 | if (of->priv) |
3811 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3812 | |
3813 | /* | |
3814 | * Either this is the first start() after open or the matching | |
3815 | * pidlist has been destroyed inbetween. Create a new one. | |
3816 | */ | |
5d22444f TH |
3817 | if (!of->priv) { |
3818 | ret = pidlist_array_load(cgrp, type, | |
3819 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3820 | if (ret) |
3821 | return ERR_PTR(ret); | |
3822 | } | |
5d22444f | 3823 | l = of->priv; |
cc31edce | 3824 | |
cc31edce | 3825 | if (pid) { |
102a775e | 3826 | int end = l->length; |
20777766 | 3827 | |
cc31edce PM |
3828 | while (index < end) { |
3829 | int mid = (index + end) / 2; | |
afb2bc14 | 3830 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3831 | index = mid; |
3832 | break; | |
afb2bc14 | 3833 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3834 | index = mid + 1; |
3835 | else | |
3836 | end = mid; | |
3837 | } | |
3838 | } | |
3839 | /* If we're off the end of the array, we're done */ | |
102a775e | 3840 | if (index >= l->length) |
cc31edce PM |
3841 | return NULL; |
3842 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3843 | iter = l->list + index; |
afb2bc14 | 3844 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3845 | return iter; |
3846 | } | |
3847 | ||
102a775e | 3848 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3849 | { |
2bd59d48 | 3850 | struct kernfs_open_file *of = s->private; |
5d22444f | 3851 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3852 | |
5d22444f TH |
3853 | if (l) |
3854 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3855 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3856 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3857 | } |
3858 | ||
102a775e | 3859 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3860 | { |
2bd59d48 | 3861 | struct kernfs_open_file *of = s->private; |
5d22444f | 3862 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3863 | pid_t *p = v; |
3864 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3865 | /* |
3866 | * Advance to the next pid in the array. If this goes off the | |
3867 | * end, we're done | |
3868 | */ | |
3869 | p++; | |
3870 | if (p >= end) { | |
3871 | return NULL; | |
3872 | } else { | |
7da11279 | 3873 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3874 | return p; |
3875 | } | |
3876 | } | |
3877 | ||
102a775e | 3878 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3879 | { |
3880 | return seq_printf(s, "%d\n", *(int *)v); | |
3881 | } | |
bbcb81d0 | 3882 | |
182446d0 TH |
3883 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3884 | struct cftype *cft) | |
81a6a5cd | 3885 | { |
182446d0 | 3886 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3887 | } |
3888 | ||
182446d0 TH |
3889 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3890 | struct cftype *cft, u64 val) | |
6379c106 | 3891 | { |
182446d0 | 3892 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3893 | if (val) |
182446d0 | 3894 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3895 | else |
182446d0 | 3896 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3897 | return 0; |
3898 | } | |
3899 | ||
182446d0 TH |
3900 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3901 | struct cftype *cft) | |
97978e6d | 3902 | { |
182446d0 | 3903 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3904 | } |
3905 | ||
182446d0 TH |
3906 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3907 | struct cftype *cft, u64 val) | |
97978e6d DL |
3908 | { |
3909 | if (val) | |
182446d0 | 3910 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3911 | else |
182446d0 | 3912 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3913 | return 0; |
3914 | } | |
3915 | ||
d5c56ced | 3916 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3917 | { |
d5c56ced | 3918 | .name = "cgroup.procs", |
6612f05b TH |
3919 | .seq_start = cgroup_pidlist_start, |
3920 | .seq_next = cgroup_pidlist_next, | |
3921 | .seq_stop = cgroup_pidlist_stop, | |
3922 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3923 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3924 | .write_u64 = cgroup_procs_write, |
74a1166d | 3925 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3926 | }, |
97978e6d DL |
3927 | { |
3928 | .name = "cgroup.clone_children", | |
873fe09e | 3929 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3930 | .read_u64 = cgroup_clone_children_read, |
3931 | .write_u64 = cgroup_clone_children_write, | |
3932 | }, | |
873fe09e TH |
3933 | { |
3934 | .name = "cgroup.sane_behavior", | |
3935 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3936 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3937 | }, |
f8f22e53 TH |
3938 | { |
3939 | .name = "cgroup.controllers", | |
3940 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_ONLY_ON_ROOT, | |
3941 | .seq_show = cgroup_root_controllers_show, | |
3942 | }, | |
3943 | { | |
3944 | .name = "cgroup.controllers", | |
3945 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
3946 | .seq_show = cgroup_controllers_show, | |
3947 | }, | |
3948 | { | |
3949 | .name = "cgroup.subtree_control", | |
3950 | .flags = CFTYPE_ONLY_ON_DFL, | |
3951 | .seq_show = cgroup_subtree_control_show, | |
3952 | .write_string = cgroup_subtree_control_write, | |
3953 | }, | |
842b597e TH |
3954 | { |
3955 | .name = "cgroup.populated", | |
3956 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
3957 | .seq_show = cgroup_populated_show, | |
3958 | }, | |
d5c56ced TH |
3959 | |
3960 | /* | |
3961 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3962 | * don't exist if sane_behavior. If you're depending on these, be | |
3963 | * prepared to be burned. | |
3964 | */ | |
3965 | { | |
3966 | .name = "tasks", | |
3967 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3968 | .seq_start = cgroup_pidlist_start, |
3969 | .seq_next = cgroup_pidlist_next, | |
3970 | .seq_stop = cgroup_pidlist_stop, | |
3971 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3972 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3973 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3974 | .mode = S_IRUGO | S_IWUSR, |
3975 | }, | |
3976 | { | |
3977 | .name = "notify_on_release", | |
3978 | .flags = CFTYPE_INSANE, | |
3979 | .read_u64 = cgroup_read_notify_on_release, | |
3980 | .write_u64 = cgroup_write_notify_on_release, | |
3981 | }, | |
6e6ff25b TH |
3982 | { |
3983 | .name = "release_agent", | |
cc5943a7 | 3984 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3985 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3986 | .write_string = cgroup_release_agent_write, |
5f469907 | 3987 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3988 | }, |
db0416b6 | 3989 | { } /* terminate */ |
bbcb81d0 PM |
3990 | }; |
3991 | ||
13af07df | 3992 | /** |
628f7cd4 | 3993 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3994 | * @cgrp: target cgroup |
13af07df | 3995 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3996 | * |
3997 | * On failure, no file is added. | |
13af07df | 3998 | */ |
628f7cd4 | 3999 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 4000 | { |
ddbcc7e8 | 4001 | struct cgroup_subsys *ss; |
b420ba7d | 4002 | int i, ret = 0; |
bbcb81d0 | 4003 | |
8e3f6541 | 4004 | /* process cftsets of each subsystem */ |
b420ba7d | 4005 | for_each_subsys(ss, i) { |
0adb0704 | 4006 | struct cftype *cfts; |
b420ba7d TH |
4007 | |
4008 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 4009 | continue; |
8e3f6541 | 4010 | |
0adb0704 TH |
4011 | list_for_each_entry(cfts, &ss->cfts, node) { |
4012 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4013 | if (ret < 0) |
4014 | goto err; | |
4015 | } | |
ddbcc7e8 | 4016 | } |
ddbcc7e8 | 4017 | return 0; |
bee55099 TH |
4018 | err: |
4019 | cgroup_clear_dir(cgrp, subsys_mask); | |
4020 | return ret; | |
ddbcc7e8 PM |
4021 | } |
4022 | ||
0c21ead1 TH |
4023 | /* |
4024 | * css destruction is four-stage process. | |
4025 | * | |
4026 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4027 | * Implemented in kill_css(). | |
4028 | * | |
4029 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
4030 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
4031 | * by invoking offline_css(). After offlining, the base ref is put. | |
4032 | * Implemented in css_killed_work_fn(). | |
4033 | * | |
4034 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4035 | * accessors are inside RCU read sections. css_release() schedules the | |
4036 | * RCU callback. | |
4037 | * | |
4038 | * 4. After the grace period, the css can be freed. Implemented in | |
4039 | * css_free_work_fn(). | |
4040 | * | |
4041 | * It is actually hairier because both step 2 and 4 require process context | |
4042 | * and thus involve punting to css->destroy_work adding two additional | |
4043 | * steps to the already complex sequence. | |
4044 | */ | |
35ef10da | 4045 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4046 | { |
4047 | struct cgroup_subsys_state *css = | |
35ef10da | 4048 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4049 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4050 | |
0ae78e0b TH |
4051 | if (css->parent) |
4052 | css_put(css->parent); | |
4053 | ||
0c21ead1 | 4054 | css->ss->css_free(css); |
2bd59d48 | 4055 | cgroup_put(cgrp); |
48ddbe19 TH |
4056 | } |
4057 | ||
0c21ead1 | 4058 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4059 | { |
4060 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4061 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4062 | |
35ef10da | 4063 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4064 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4065 | } |
4066 | ||
d3daf28d TH |
4067 | static void css_release(struct percpu_ref *ref) |
4068 | { | |
4069 | struct cgroup_subsys_state *css = | |
4070 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4071 | ||
01a97140 | 4072 | RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 4073 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4074 | } |
4075 | ||
623f926b TH |
4076 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
4077 | struct cgroup *cgrp) | |
ddbcc7e8 | 4078 | { |
bd89aabc | 4079 | css->cgroup = cgrp; |
72c97e54 | 4080 | css->ss = ss; |
ddbcc7e8 | 4081 | css->flags = 0; |
0ae78e0b TH |
4082 | |
4083 | if (cgrp->parent) | |
ca8bdcaf | 4084 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 4085 | else |
38b53aba | 4086 | css->flags |= CSS_ROOT; |
48ddbe19 | 4087 | |
ca8bdcaf | 4088 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4089 | } |
4090 | ||
2a4ac633 | 4091 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4092 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4093 | { |
623f926b | 4094 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4095 | int ret = 0; |
4096 | ||
ace2bee8 | 4097 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4098 | lockdep_assert_held(&cgroup_mutex); |
4099 | ||
92fb9748 | 4100 | if (ss->css_online) |
eb95419b | 4101 | ret = ss->css_online(css); |
ae7f164a | 4102 | if (!ret) { |
eb95419b | 4103 | css->flags |= CSS_ONLINE; |
f20104de | 4104 | css->cgroup->nr_css++; |
aec25020 | 4105 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4106 | } |
b1929db4 | 4107 | return ret; |
a31f2d3f TH |
4108 | } |
4109 | ||
2a4ac633 | 4110 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4111 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4112 | { |
623f926b | 4113 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 4114 | |
ace2bee8 | 4115 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4116 | lockdep_assert_held(&cgroup_mutex); |
4117 | ||
4118 | if (!(css->flags & CSS_ONLINE)) | |
4119 | return; | |
4120 | ||
d7eeac19 | 4121 | if (ss->css_offline) |
eb95419b | 4122 | ss->css_offline(css); |
a31f2d3f | 4123 | |
eb95419b | 4124 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4125 | css->cgroup->nr_css--; |
e3297803 | 4126 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4127 | |
4128 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4129 | } |
4130 | ||
c81c925a TH |
4131 | /** |
4132 | * create_css - create a cgroup_subsys_state | |
4133 | * @cgrp: the cgroup new css will be associated with | |
4134 | * @ss: the subsys of new css | |
4135 | * | |
4136 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4137 | * css is online and installed in @cgrp with all interface files created. | |
4138 | * Returns 0 on success, -errno on failure. | |
4139 | */ | |
4140 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4141 | { | |
4142 | struct cgroup *parent = cgrp->parent; | |
4143 | struct cgroup_subsys_state *css; | |
4144 | int err; | |
4145 | ||
c81c925a TH |
4146 | lockdep_assert_held(&cgroup_mutex); |
4147 | ||
4148 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
4149 | if (IS_ERR(css)) | |
4150 | return PTR_ERR(css); | |
4151 | ||
4152 | err = percpu_ref_init(&css->refcnt, css_release); | |
4153 | if (err) | |
3eb59ec6 | 4154 | goto err_free_css; |
c81c925a TH |
4155 | |
4156 | init_css(css, ss, cgrp); | |
4157 | ||
aec25020 | 4158 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 4159 | if (err) |
3eb59ec6 | 4160 | goto err_free_percpu_ref; |
c81c925a TH |
4161 | |
4162 | err = online_css(css); | |
4163 | if (err) | |
3eb59ec6 | 4164 | goto err_clear_dir; |
c81c925a | 4165 | |
59f5296b | 4166 | cgroup_get(cgrp); |
c81c925a TH |
4167 | css_get(css->parent); |
4168 | ||
4169 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4170 | parent->parent) { | |
a2a1f9ea JZ |
4171 | pr_warn("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", |
4172 | current->comm, current->pid, ss->name); | |
c81c925a | 4173 | if (!strcmp(ss->name, "memory")) |
a2a1f9ea | 4174 | pr_warn("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); |
c81c925a TH |
4175 | ss->warned_broken_hierarchy = true; |
4176 | } | |
4177 | ||
4178 | return 0; | |
4179 | ||
3eb59ec6 | 4180 | err_clear_dir: |
32d01dc7 | 4181 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3eb59ec6 | 4182 | err_free_percpu_ref: |
c81c925a | 4183 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 4184 | err_free_css: |
c81c925a TH |
4185 | ss->css_free(css); |
4186 | return err; | |
4187 | } | |
4188 | ||
2bd59d48 | 4189 | /** |
a043e3b2 LZ |
4190 | * cgroup_create - create a cgroup |
4191 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 4192 | * @name: name of the new cgroup |
2bd59d48 | 4193 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 4194 | */ |
e61734c5 | 4195 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 4196 | umode_t mode) |
ddbcc7e8 | 4197 | { |
bd89aabc | 4198 | struct cgroup *cgrp; |
3dd06ffa | 4199 | struct cgroup_root *root = parent->root; |
b58c8998 | 4200 | int ssid, err; |
ddbcc7e8 | 4201 | struct cgroup_subsys *ss; |
2bd59d48 | 4202 | struct kernfs_node *kn; |
ddbcc7e8 | 4203 | |
0a950f65 | 4204 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4205 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4206 | if (!cgrp) | |
ddbcc7e8 PM |
4207 | return -ENOMEM; |
4208 | ||
ace2bee8 | 4209 | mutex_lock(&cgroup_tree_mutex); |
65dff759 | 4210 | |
976c06bc TH |
4211 | /* |
4212 | * Only live parents can have children. Note that the liveliness | |
4213 | * check isn't strictly necessary because cgroup_mkdir() and | |
4214 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4215 | * anyway so that locking is contained inside cgroup proper and we | |
4216 | * don't get nasty surprises if we ever grow another caller. | |
4217 | */ | |
4218 | if (!cgroup_lock_live_group(parent)) { | |
4219 | err = -ENODEV; | |
ace2bee8 | 4220 | goto err_unlock_tree; |
0ab02ca8 LZ |
4221 | } |
4222 | ||
4223 | /* | |
4224 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4225 | * a half-baked cgroup. | |
4226 | */ | |
4227 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
4228 | if (cgrp->id < 0) { | |
4229 | err = -ENOMEM; | |
4230 | goto err_unlock; | |
976c06bc TH |
4231 | } |
4232 | ||
cc31edce | 4233 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4234 | |
bd89aabc | 4235 | cgrp->parent = parent; |
0ae78e0b | 4236 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4237 | cgrp->root = parent->root; |
ddbcc7e8 | 4238 | |
b6abdb0e LZ |
4239 | if (notify_on_release(parent)) |
4240 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4241 | ||
2260e7fc TH |
4242 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4243 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4244 | |
2bd59d48 | 4245 | /* create the directory */ |
e61734c5 | 4246 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
4247 | if (IS_ERR(kn)) { |
4248 | err = PTR_ERR(kn); | |
0ab02ca8 | 4249 | goto err_free_id; |
2bd59d48 TH |
4250 | } |
4251 | cgrp->kn = kn; | |
ddbcc7e8 | 4252 | |
4e139afc | 4253 | /* |
6f30558f TH |
4254 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4255 | * that @cgrp->kn is always accessible. | |
4e139afc | 4256 | */ |
6f30558f | 4257 | kernfs_get(kn); |
ddbcc7e8 | 4258 | |
00356bd5 | 4259 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4260 | |
4e139afc | 4261 | /* allocation complete, commit to creation */ |
4e139afc | 4262 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 4263 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4264 | cgroup_get(parent); |
415cf07a | 4265 | |
0d80255e TH |
4266 | /* |
4267 | * @cgrp is now fully operational. If something fails after this | |
4268 | * point, it'll be released via the normal destruction path. | |
4269 | */ | |
4e96ee8e LZ |
4270 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4271 | ||
49957f8e TH |
4272 | err = cgroup_kn_set_ugid(kn); |
4273 | if (err) | |
4274 | goto err_destroy; | |
4275 | ||
2bb566cb | 4276 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4277 | if (err) |
4278 | goto err_destroy; | |
4279 | ||
9d403e99 | 4280 | /* let's create and online css's */ |
b85d2040 | 4281 | for_each_subsys(ss, ssid) { |
f392e51c | 4282 | if (parent->child_subsys_mask & (1 << ssid)) { |
b85d2040 TH |
4283 | err = create_css(cgrp, ss); |
4284 | if (err) | |
4285 | goto err_destroy; | |
4286 | } | |
a8638030 | 4287 | } |
ddbcc7e8 | 4288 | |
bd53d617 TH |
4289 | /* |
4290 | * On the default hierarchy, a child doesn't automatically inherit | |
4291 | * child_subsys_mask from the parent. Each is configured manually. | |
4292 | */ | |
4293 | if (!cgroup_on_dfl(cgrp)) | |
4294 | cgrp->child_subsys_mask = parent->child_subsys_mask; | |
f392e51c | 4295 | |
2bd59d48 TH |
4296 | kernfs_activate(kn); |
4297 | ||
ddbcc7e8 | 4298 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4299 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
4300 | |
4301 | return 0; | |
4302 | ||
0a950f65 | 4303 | err_free_id: |
4e96ee8e | 4304 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
4305 | err_unlock: |
4306 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
4307 | err_unlock_tree: |
4308 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 4309 | kfree(cgrp); |
ddbcc7e8 | 4310 | return err; |
4b8b47eb TH |
4311 | |
4312 | err_destroy: | |
4313 | cgroup_destroy_locked(cgrp); | |
4314 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4315 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 4316 | return err; |
ddbcc7e8 PM |
4317 | } |
4318 | ||
2bd59d48 TH |
4319 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4320 | umode_t mode) | |
ddbcc7e8 | 4321 | { |
2bd59d48 | 4322 | struct cgroup *parent = parent_kn->priv; |
e1b2dc17 | 4323 | int ret; |
ddbcc7e8 | 4324 | |
e1b2dc17 TH |
4325 | /* |
4326 | * cgroup_create() grabs cgroup_tree_mutex which nests outside | |
4327 | * kernfs active_ref and cgroup_create() already synchronizes | |
4328 | * properly against removal through cgroup_lock_live_group(). | |
4329 | * Break it before calling cgroup_create(). | |
4330 | */ | |
4331 | cgroup_get(parent); | |
4332 | kernfs_break_active_protection(parent_kn); | |
ddbcc7e8 | 4333 | |
e1b2dc17 TH |
4334 | ret = cgroup_create(parent, name, mode); |
4335 | ||
4336 | kernfs_unbreak_active_protection(parent_kn); | |
4337 | cgroup_put(parent); | |
4338 | return ret; | |
ddbcc7e8 PM |
4339 | } |
4340 | ||
223dbc38 TH |
4341 | /* |
4342 | * This is called when the refcnt of a css is confirmed to be killed. | |
4343 | * css_tryget() is now guaranteed to fail. | |
4344 | */ | |
4345 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4346 | { |
223dbc38 TH |
4347 | struct cgroup_subsys_state *css = |
4348 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4349 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4350 | |
ace2bee8 | 4351 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
4352 | mutex_lock(&cgroup_mutex); |
4353 | ||
09a503ea TH |
4354 | /* |
4355 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
4356 | * initate destruction. | |
4357 | */ | |
4358 | offline_css(css); | |
4359 | ||
f20104de TH |
4360 | /* |
4361 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4362 | * be disabled before proceeding to the second phase of cgroup | |
4363 | * destruction. If we are the last one, kick it off. | |
4364 | */ | |
09a503ea | 4365 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4366 | cgroup_destroy_css_killed(cgrp); |
4367 | ||
4368 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4369 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
4370 | |
4371 | /* | |
4372 | * Put the css refs from kill_css(). Each css holds an extra | |
4373 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4374 | * regardless of css refs. On the last put of each css, whenever | |
4375 | * that may be, the extra dentry ref is put so that dentry | |
4376 | * destruction happens only after all css's are released. | |
4377 | */ | |
4378 | css_put(css); | |
d3daf28d TH |
4379 | } |
4380 | ||
223dbc38 TH |
4381 | /* css kill confirmation processing requires process context, bounce */ |
4382 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4383 | { |
4384 | struct cgroup_subsys_state *css = | |
4385 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4386 | ||
223dbc38 | 4387 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4388 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4389 | } |
4390 | ||
f392e51c TH |
4391 | /** |
4392 | * kill_css - destroy a css | |
4393 | * @css: css to destroy | |
4394 | * | |
4395 | * This function initiates destruction of @css by removing cgroup interface | |
4396 | * files and putting its base reference. ->css_offline() will be invoked | |
4397 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
4398 | * reference count reaches zero, @css will be released. | |
4399 | */ | |
4400 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4401 | { |
94419627 TH |
4402 | lockdep_assert_held(&cgroup_tree_mutex); |
4403 | ||
2bd59d48 TH |
4404 | /* |
4405 | * This must happen before css is disassociated with its cgroup. | |
4406 | * See seq_css() for details. | |
4407 | */ | |
aec25020 | 4408 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4409 | |
edae0c33 TH |
4410 | /* |
4411 | * Killing would put the base ref, but we need to keep it alive | |
4412 | * until after ->css_offline(). | |
4413 | */ | |
4414 | css_get(css); | |
4415 | ||
4416 | /* | |
4417 | * cgroup core guarantees that, by the time ->css_offline() is | |
4418 | * invoked, no new css reference will be given out via | |
4419 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
4420 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
4421 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4422 | * | |
4423 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4424 | * css is confirmed to be seen as killed on all CPUs. | |
4425 | */ | |
4426 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4427 | } |
4428 | ||
4429 | /** | |
4430 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4431 | * @cgrp: cgroup to be destroyed | |
4432 | * | |
4433 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4434 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
4435 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
4436 | * invoked. To satisfy all the requirements, destruction is implemented in | |
4437 | * the following two steps. | |
4438 | * | |
4439 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4440 | * userland visible parts and start killing the percpu refcnts of | |
4441 | * css's. Set up so that the next stage will be kicked off once all | |
4442 | * the percpu refcnts are confirmed to be killed. | |
4443 | * | |
4444 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4445 | * rest of destruction. Once all cgroup references are gone, the | |
4446 | * cgroup is RCU-freed. | |
4447 | * | |
4448 | * This function implements s1. After this step, @cgrp is gone as far as | |
4449 | * the userland is concerned and a new cgroup with the same name may be | |
4450 | * created. As cgroup doesn't care about the names internally, this | |
4451 | * doesn't cause any problem. | |
4452 | */ | |
42809dd4 TH |
4453 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4454 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4455 | { |
bb78a92f | 4456 | struct cgroup *child; |
2bd59d48 | 4457 | struct cgroup_subsys_state *css; |
ddd69148 | 4458 | bool empty; |
1c6727af | 4459 | int ssid; |
ddbcc7e8 | 4460 | |
ace2bee8 | 4461 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4462 | lockdep_assert_held(&cgroup_mutex); |
4463 | ||
ddd69148 | 4464 | /* |
96d365e0 | 4465 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4466 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4467 | */ |
96d365e0 | 4468 | down_read(&css_set_rwsem); |
bb78a92f | 4469 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4470 | up_read(&css_set_rwsem); |
ddd69148 | 4471 | if (!empty) |
ddbcc7e8 | 4472 | return -EBUSY; |
a043e3b2 | 4473 | |
bb78a92f HD |
4474 | /* |
4475 | * Make sure there's no live children. We can't test ->children | |
4476 | * emptiness as dead children linger on it while being destroyed; | |
4477 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4478 | */ | |
4479 | empty = true; | |
4480 | rcu_read_lock(); | |
4481 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4482 | empty = cgroup_is_dead(child); | |
4483 | if (!empty) | |
4484 | break; | |
4485 | } | |
4486 | rcu_read_unlock(); | |
4487 | if (!empty) | |
4488 | return -EBUSY; | |
4489 | ||
455050d2 TH |
4490 | /* |
4491 | * Mark @cgrp dead. This prevents further task migration and child | |
4492 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4493 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4494 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4495 | * css_next_child() for details. |
455050d2 | 4496 | */ |
54766d4a | 4497 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4498 | |
88703267 | 4499 | /* |
edae0c33 TH |
4500 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4501 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4502 | * percpu refs of all css's are confirmed to be killed. This |
4503 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4504 | */ |
4ac06017 | 4505 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4506 | for_each_css(css, ssid, cgrp) |
4507 | kill_css(css); | |
4ac06017 | 4508 | mutex_lock(&cgroup_mutex); |
455050d2 | 4509 | |
455050d2 TH |
4510 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4511 | raw_spin_lock(&release_list_lock); | |
4512 | if (!list_empty(&cgrp->release_list)) | |
4513 | list_del_init(&cgrp->release_list); | |
4514 | raw_spin_unlock(&release_list_lock); | |
4515 | ||
4516 | /* | |
f20104de TH |
4517 | * If @cgrp has css's attached, the second stage of cgroup |
4518 | * destruction is kicked off from css_killed_work_fn() after the | |
4519 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4520 | * any css, we kick it off here. | |
4521 | */ | |
4522 | if (!cgrp->nr_css) | |
4523 | cgroup_destroy_css_killed(cgrp); | |
4524 | ||
2bd59d48 TH |
4525 | /* remove @cgrp directory along with the base files */ |
4526 | mutex_unlock(&cgroup_mutex); | |
4527 | ||
455050d2 | 4528 | /* |
2bd59d48 TH |
4529 | * There are two control paths which try to determine cgroup from |
4530 | * dentry without going through kernfs - cgroupstats_build() and | |
4531 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4532 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4533 | * after all files under it have been removed. | |
455050d2 | 4534 | */ |
6f30558f | 4535 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4536 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4537 | |
4ac06017 | 4538 | mutex_lock(&cgroup_mutex); |
455050d2 | 4539 | |
ea15f8cc TH |
4540 | return 0; |
4541 | }; | |
4542 | ||
d3daf28d | 4543 | /** |
f20104de | 4544 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4545 | * @work: cgroup->destroy_free_work |
4546 | * | |
4547 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4548 | * destroyed after all css's are offlined and performs the rest of |
4549 | * destruction. This is the second step of destruction described in the | |
4550 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4551 | */ |
f20104de | 4552 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4553 | { |
ea15f8cc | 4554 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4555 | |
ace2bee8 | 4556 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4557 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4558 | |
999cd8a4 | 4559 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4560 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4561 | |
59f5296b | 4562 | cgroup_put(cgrp); |
ddbcc7e8 | 4563 | |
bd89aabc | 4564 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4565 | check_for_release(parent); |
ddbcc7e8 PM |
4566 | } |
4567 | ||
2bd59d48 | 4568 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4569 | { |
2bd59d48 TH |
4570 | struct cgroup *cgrp = kn->priv; |
4571 | int ret = 0; | |
42809dd4 | 4572 | |
2bd59d48 TH |
4573 | /* |
4574 | * This is self-destruction but @kn can't be removed while this | |
4575 | * callback is in progress. Let's break active protection. Once | |
4576 | * the protection is broken, @cgrp can be destroyed at any point. | |
4577 | * Pin it so that it stays accessible. | |
4578 | */ | |
4579 | cgroup_get(cgrp); | |
4580 | kernfs_break_active_protection(kn); | |
42809dd4 | 4581 | |
ace2bee8 | 4582 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4583 | mutex_lock(&cgroup_mutex); |
8e3f6541 TH |
4584 | |
4585 | /* | |
2bd59d48 TH |
4586 | * @cgrp might already have been destroyed while we're trying to |
4587 | * grab the mutexes. | |
8e3f6541 | 4588 | */ |
2bd59d48 TH |
4589 | if (!cgroup_is_dead(cgrp)) |
4590 | ret = cgroup_destroy_locked(cgrp); | |
2bb566cb | 4591 | |
42809dd4 | 4592 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4593 | mutex_unlock(&cgroup_tree_mutex); |
2bb566cb | 4594 | |
2bd59d48 TH |
4595 | kernfs_unbreak_active_protection(kn); |
4596 | cgroup_put(cgrp); | |
42809dd4 | 4597 | return ret; |
8e3f6541 TH |
4598 | } |
4599 | ||
2bd59d48 TH |
4600 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4601 | .remount_fs = cgroup_remount, | |
4602 | .show_options = cgroup_show_options, | |
4603 | .mkdir = cgroup_mkdir, | |
4604 | .rmdir = cgroup_rmdir, | |
4605 | .rename = cgroup_rename, | |
4606 | }; | |
4607 | ||
06a11920 | 4608 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4609 | { |
ddbcc7e8 | 4610 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4611 | |
4612 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4613 | |
ace2bee8 | 4614 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4615 | mutex_lock(&cgroup_mutex); |
4616 | ||
0adb0704 | 4617 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4618 | |
3dd06ffa TH |
4619 | /* Create the root cgroup state for this subsystem */ |
4620 | ss->root = &cgrp_dfl_root; | |
4621 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4622 | /* We don't handle early failures gracefully */ |
4623 | BUG_ON(IS_ERR(css)); | |
3dd06ffa | 4624 | init_css(css, ss, &cgrp_dfl_root.cgrp); |
ddbcc7e8 | 4625 | |
e8d55fde | 4626 | /* Update the init_css_set to contain a subsys |
817929ec | 4627 | * pointer to this state - since the subsystem is |
e8d55fde | 4628 | * newly registered, all tasks and hence the |
3dd06ffa | 4629 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4630 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4631 | |
4632 | need_forkexit_callback |= ss->fork || ss->exit; | |
4633 | ||
e8d55fde LZ |
4634 | /* At system boot, before all subsystems have been |
4635 | * registered, no tasks have been forked, so we don't | |
4636 | * need to invoke fork callbacks here. */ | |
4637 | BUG_ON(!list_empty(&init_task.tasks)); | |
4638 | ||
ae7f164a | 4639 | BUG_ON(online_css(css)); |
a8638030 | 4640 | |
f392e51c | 4641 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4642 | |
4643 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4644 | mutex_unlock(&cgroup_tree_mutex); |
cf5d5941 | 4645 | } |
cf5d5941 | 4646 | |
ddbcc7e8 | 4647 | /** |
a043e3b2 LZ |
4648 | * cgroup_init_early - cgroup initialization at system boot |
4649 | * | |
4650 | * Initialize cgroups at system boot, and initialize any | |
4651 | * subsystems that request early init. | |
ddbcc7e8 PM |
4652 | */ |
4653 | int __init cgroup_init_early(void) | |
4654 | { | |
a2dd4247 TH |
4655 | static struct cgroup_sb_opts __initdata opts = |
4656 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4657 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4658 | int i; |
30159ec7 | 4659 | |
3dd06ffa | 4660 | init_cgroup_root(&cgrp_dfl_root, &opts); |
a4ea1cc9 | 4661 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4662 | |
3ed80a62 | 4663 | for_each_subsys(ss, i) { |
aec25020 | 4664 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4665 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4666 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4667 | ss->id, ss->name); |
073219e9 TH |
4668 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4669 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4670 | ||
aec25020 | 4671 | ss->id = i; |
073219e9 | 4672 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4673 | |
4674 | if (ss->early_init) | |
4675 | cgroup_init_subsys(ss); | |
4676 | } | |
4677 | return 0; | |
4678 | } | |
4679 | ||
4680 | /** | |
a043e3b2 LZ |
4681 | * cgroup_init - cgroup initialization |
4682 | * | |
4683 | * Register cgroup filesystem and /proc file, and initialize | |
4684 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4685 | */ |
4686 | int __init cgroup_init(void) | |
4687 | { | |
30159ec7 | 4688 | struct cgroup_subsys *ss; |
0ac801fe | 4689 | unsigned long key; |
172a2c06 | 4690 | int ssid, err; |
ddbcc7e8 | 4691 | |
2bd59d48 | 4692 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4693 | |
985ed670 | 4694 | mutex_lock(&cgroup_tree_mutex); |
54e7b4eb | 4695 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4696 | |
82fe9b0d TH |
4697 | /* Add init_css_set to the hash table */ |
4698 | key = css_set_hash(init_css_set.subsys); | |
4699 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4700 | ||
3dd06ffa | 4701 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4702 | |
54e7b4eb | 4703 | mutex_unlock(&cgroup_mutex); |
985ed670 | 4704 | mutex_unlock(&cgroup_tree_mutex); |
54e7b4eb | 4705 | |
172a2c06 TH |
4706 | for_each_subsys(ss, ssid) { |
4707 | if (!ss->early_init) | |
4708 | cgroup_init_subsys(ss); | |
4709 | ||
2d8f243a TH |
4710 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4711 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
4712 | ||
172a2c06 TH |
4713 | /* |
4714 | * cftype registration needs kmalloc and can't be done | |
4715 | * during early_init. Register base cftypes separately. | |
4716 | */ | |
4717 | if (ss->base_cftypes) | |
4718 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GKH |
4719 | } |
4720 | ||
676db4af | 4721 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4722 | if (!cgroup_kobj) |
4723 | return -ENOMEM; | |
676db4af | 4724 | |
ddbcc7e8 | 4725 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4726 | if (err < 0) { |
4727 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4728 | return err; |
676db4af | 4729 | } |
ddbcc7e8 | 4730 | |
46ae220b | 4731 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4732 | return 0; |
ddbcc7e8 | 4733 | } |
b4f48b63 | 4734 | |
e5fca243 TH |
4735 | static int __init cgroup_wq_init(void) |
4736 | { | |
4737 | /* | |
4738 | * There isn't much point in executing destruction path in | |
4739 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4740 | * Use 1 for @max_active. |
e5fca243 TH |
4741 | * |
4742 | * We would prefer to do this in cgroup_init() above, but that | |
4743 | * is called before init_workqueues(): so leave this until after. | |
4744 | */ | |
1a11533f | 4745 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4746 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4747 | |
4748 | /* | |
4749 | * Used to destroy pidlists and separate to serve as flush domain. | |
4750 | * Cap @max_active to 1 too. | |
4751 | */ | |
4752 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4753 | 0, 1); | |
4754 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4755 | ||
e5fca243 TH |
4756 | return 0; |
4757 | } | |
4758 | core_initcall(cgroup_wq_init); | |
4759 | ||
a424316c PM |
4760 | /* |
4761 | * proc_cgroup_show() | |
4762 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4763 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4764 | */ |
4765 | ||
4766 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4767 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4768 | { |
4769 | struct pid *pid; | |
4770 | struct task_struct *tsk; | |
e61734c5 | 4771 | char *buf, *path; |
a424316c | 4772 | int retval; |
3dd06ffa | 4773 | struct cgroup_root *root; |
a424316c PM |
4774 | |
4775 | retval = -ENOMEM; | |
e61734c5 | 4776 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4777 | if (!buf) |
4778 | goto out; | |
4779 | ||
4780 | retval = -ESRCH; | |
4781 | pid = m->private; | |
4782 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4783 | if (!tsk) | |
4784 | goto out_free; | |
4785 | ||
4786 | retval = 0; | |
4787 | ||
4788 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4789 | down_read(&css_set_rwsem); |
a424316c | 4790 | |
985ed670 | 4791 | for_each_root(root) { |
a424316c | 4792 | struct cgroup_subsys *ss; |
bd89aabc | 4793 | struct cgroup *cgrp; |
b85d2040 | 4794 | int ssid, count = 0; |
a424316c | 4795 | |
a2dd4247 | 4796 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4797 | continue; |
4798 | ||
2c6ab6d2 | 4799 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4800 | for_each_subsys(ss, ssid) |
f392e51c | 4801 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 4802 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4803 | if (strlen(root->name)) |
4804 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4805 | root->name); | |
a424316c | 4806 | seq_putc(m, ':'); |
7717f7ba | 4807 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4808 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4809 | if (!path) { | |
4810 | retval = -ENAMETOOLONG; | |
a424316c | 4811 | goto out_unlock; |
e61734c5 TH |
4812 | } |
4813 | seq_puts(m, path); | |
a424316c PM |
4814 | seq_putc(m, '\n'); |
4815 | } | |
4816 | ||
4817 | out_unlock: | |
96d365e0 | 4818 | up_read(&css_set_rwsem); |
a424316c PM |
4819 | mutex_unlock(&cgroup_mutex); |
4820 | put_task_struct(tsk); | |
4821 | out_free: | |
4822 | kfree(buf); | |
4823 | out: | |
4824 | return retval; | |
4825 | } | |
4826 | ||
a424316c PM |
4827 | /* Display information about each subsystem and each hierarchy */ |
4828 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4829 | { | |
30159ec7 | 4830 | struct cgroup_subsys *ss; |
a424316c | 4831 | int i; |
a424316c | 4832 | |
8bab8dde | 4833 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4834 | /* |
4835 | * ideally we don't want subsystems moving around while we do this. | |
4836 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4837 | * subsys/hierarchy state. | |
4838 | */ | |
a424316c | 4839 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4840 | |
4841 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4842 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4843 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4844 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4845 | |
a424316c PM |
4846 | mutex_unlock(&cgroup_mutex); |
4847 | return 0; | |
4848 | } | |
4849 | ||
4850 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4851 | { | |
9dce07f1 | 4852 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4853 | } |
4854 | ||
828c0950 | 4855 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4856 | .open = cgroupstats_open, |
4857 | .read = seq_read, | |
4858 | .llseek = seq_lseek, | |
4859 | .release = single_release, | |
4860 | }; | |
4861 | ||
b4f48b63 | 4862 | /** |
eaf797ab | 4863 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4864 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4865 | * |
eaf797ab TH |
4866 | * A task is associated with the init_css_set until cgroup_post_fork() |
4867 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4868 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4869 | */ |
4870 | void cgroup_fork(struct task_struct *child) | |
4871 | { | |
eaf797ab | 4872 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4873 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4874 | } |
4875 | ||
817929ec | 4876 | /** |
a043e3b2 LZ |
4877 | * cgroup_post_fork - called on a new task after adding it to the task list |
4878 | * @child: the task in question | |
4879 | * | |
5edee61e TH |
4880 | * Adds the task to the list running through its css_set if necessary and |
4881 | * call the subsystem fork() callbacks. Has to be after the task is | |
4882 | * visible on the task list in case we race with the first call to | |
0942eeee | 4883 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4884 | * list. |
a043e3b2 | 4885 | */ |
817929ec PM |
4886 | void cgroup_post_fork(struct task_struct *child) |
4887 | { | |
30159ec7 | 4888 | struct cgroup_subsys *ss; |
5edee61e TH |
4889 | int i; |
4890 | ||
3ce3230a | 4891 | /* |
eaf797ab TH |
4892 | * This may race against cgroup_enable_task_cg_links(). As that |
4893 | * function sets use_task_css_set_links before grabbing | |
4894 | * tasklist_lock and we just went through tasklist_lock to add | |
4895 | * @child, it's guaranteed that either we see the set | |
4896 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4897 | * @child during its iteration. | |
4898 | * | |
4899 | * If we won the race, @child is associated with %current's | |
4900 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4901 | * association is stable, and, on completion of the parent's | |
4902 | * migration, @child is visible in the source of migration or | |
4903 | * already in the destination cgroup. This guarantee is necessary | |
4904 | * when implementing operations which need to migrate all tasks of | |
4905 | * a cgroup to another. | |
4906 | * | |
4907 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4908 | * will remain in init_css_set. This is safe because all tasks are | |
4909 | * in the init_css_set before cg_links is enabled and there's no | |
4910 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4911 | */ |
817929ec | 4912 | if (use_task_css_set_links) { |
eaf797ab TH |
4913 | struct css_set *cset; |
4914 | ||
96d365e0 | 4915 | down_write(&css_set_rwsem); |
0e1d768f | 4916 | cset = task_css_set(current); |
eaf797ab TH |
4917 | if (list_empty(&child->cg_list)) { |
4918 | rcu_assign_pointer(child->cgroups, cset); | |
4919 | list_add(&child->cg_list, &cset->tasks); | |
4920 | get_css_set(cset); | |
4921 | } | |
96d365e0 | 4922 | up_write(&css_set_rwsem); |
817929ec | 4923 | } |
5edee61e TH |
4924 | |
4925 | /* | |
4926 | * Call ss->fork(). This must happen after @child is linked on | |
4927 | * css_set; otherwise, @child might change state between ->fork() | |
4928 | * and addition to css_set. | |
4929 | */ | |
4930 | if (need_forkexit_callback) { | |
3ed80a62 | 4931 | for_each_subsys(ss, i) |
5edee61e TH |
4932 | if (ss->fork) |
4933 | ss->fork(child); | |
5edee61e | 4934 | } |
817929ec | 4935 | } |
5edee61e | 4936 | |
b4f48b63 PM |
4937 | /** |
4938 | * cgroup_exit - detach cgroup from exiting task | |
4939 | * @tsk: pointer to task_struct of exiting process | |
4940 | * | |
4941 | * Description: Detach cgroup from @tsk and release it. | |
4942 | * | |
4943 | * Note that cgroups marked notify_on_release force every task in | |
4944 | * them to take the global cgroup_mutex mutex when exiting. | |
4945 | * This could impact scaling on very large systems. Be reluctant to | |
4946 | * use notify_on_release cgroups where very high task exit scaling | |
4947 | * is required on large systems. | |
4948 | * | |
0e1d768f TH |
4949 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4950 | * call cgroup_exit() while the task is still competent to handle | |
4951 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4952 | * each hierarchy for the remainder of its exit. No need to bother with | |
4953 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4954 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4955 | */ |
1ec41830 | 4956 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4957 | { |
30159ec7 | 4958 | struct cgroup_subsys *ss; |
5abb8855 | 4959 | struct css_set *cset; |
eaf797ab | 4960 | bool put_cset = false; |
d41d5a01 | 4961 | int i; |
817929ec PM |
4962 | |
4963 | /* | |
0e1d768f TH |
4964 | * Unlink from @tsk from its css_set. As migration path can't race |
4965 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
4966 | */ |
4967 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4968 | down_write(&css_set_rwsem); |
0e1d768f | 4969 | list_del_init(&tsk->cg_list); |
96d365e0 | 4970 | up_write(&css_set_rwsem); |
0e1d768f | 4971 | put_cset = true; |
817929ec PM |
4972 | } |
4973 | ||
b4f48b63 | 4974 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
4975 | cset = task_css_set(tsk); |
4976 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 4977 | |
1ec41830 | 4978 | if (need_forkexit_callback) { |
3ed80a62 TH |
4979 | /* see cgroup_post_fork() for details */ |
4980 | for_each_subsys(ss, i) { | |
d41d5a01 | 4981 | if (ss->exit) { |
eb95419b TH |
4982 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4983 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4984 | |
eb95419b | 4985 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4986 | } |
4987 | } | |
4988 | } | |
d41d5a01 | 4989 | |
eaf797ab TH |
4990 | if (put_cset) |
4991 | put_css_set(cset, true); | |
b4f48b63 | 4992 | } |
697f4161 | 4993 | |
bd89aabc | 4994 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4995 | { |
f50daa70 | 4996 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4997 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4998 | /* |
4999 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 5000 | * already queued for a userspace notification, queue |
f50daa70 LZ |
5001 | * it now |
5002 | */ | |
81a6a5cd | 5003 | int need_schedule_work = 0; |
f50daa70 | 5004 | |
cdcc136f | 5005 | raw_spin_lock(&release_list_lock); |
54766d4a | 5006 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
5007 | list_empty(&cgrp->release_list)) { |
5008 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5009 | need_schedule_work = 1; |
5010 | } | |
cdcc136f | 5011 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5012 | if (need_schedule_work) |
5013 | schedule_work(&release_agent_work); | |
5014 | } | |
5015 | } | |
5016 | ||
81a6a5cd PM |
5017 | /* |
5018 | * Notify userspace when a cgroup is released, by running the | |
5019 | * configured release agent with the name of the cgroup (path | |
5020 | * relative to the root of cgroup file system) as the argument. | |
5021 | * | |
5022 | * Most likely, this user command will try to rmdir this cgroup. | |
5023 | * | |
5024 | * This races with the possibility that some other task will be | |
5025 | * attached to this cgroup before it is removed, or that some other | |
5026 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5027 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5028 | * unused, and this cgroup will be reprieved from its death sentence, | |
5029 | * to continue to serve a useful existence. Next time it's released, | |
5030 | * we will get notified again, if it still has 'notify_on_release' set. | |
5031 | * | |
5032 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5033 | * means only wait until the task is successfully execve()'d. The | |
5034 | * separate release agent task is forked by call_usermodehelper(), | |
5035 | * then control in this thread returns here, without waiting for the | |
5036 | * release agent task. We don't bother to wait because the caller of | |
5037 | * this routine has no use for the exit status of the release agent | |
5038 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5039 | */ |
81a6a5cd PM |
5040 | static void cgroup_release_agent(struct work_struct *work) |
5041 | { | |
5042 | BUG_ON(work != &release_agent_work); | |
5043 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5044 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5045 | while (!list_empty(&release_list)) { |
5046 | char *argv[3], *envp[3]; | |
5047 | int i; | |
e61734c5 | 5048 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 5049 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5050 | struct cgroup, |
5051 | release_list); | |
bd89aabc | 5052 | list_del_init(&cgrp->release_list); |
cdcc136f | 5053 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 5054 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
5055 | if (!pathbuf) |
5056 | goto continue_free; | |
e61734c5 TH |
5057 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
5058 | if (!path) | |
e788e066 PM |
5059 | goto continue_free; |
5060 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5061 | if (!agentbuf) | |
5062 | goto continue_free; | |
81a6a5cd PM |
5063 | |
5064 | i = 0; | |
e788e066 | 5065 | argv[i++] = agentbuf; |
e61734c5 | 5066 | argv[i++] = path; |
81a6a5cd PM |
5067 | argv[i] = NULL; |
5068 | ||
5069 | i = 0; | |
5070 | /* minimal command environment */ | |
5071 | envp[i++] = "HOME=/"; | |
5072 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5073 | envp[i] = NULL; | |
5074 | ||
5075 | /* Drop the lock while we invoke the usermode helper, | |
5076 | * since the exec could involve hitting disk and hence | |
5077 | * be a slow process */ | |
5078 | mutex_unlock(&cgroup_mutex); | |
5079 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5080 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5081 | continue_free: |
5082 | kfree(pathbuf); | |
5083 | kfree(agentbuf); | |
cdcc136f | 5084 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5085 | } |
cdcc136f | 5086 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5087 | mutex_unlock(&cgroup_mutex); |
5088 | } | |
8bab8dde PM |
5089 | |
5090 | static int __init cgroup_disable(char *str) | |
5091 | { | |
30159ec7 | 5092 | struct cgroup_subsys *ss; |
8bab8dde | 5093 | char *token; |
30159ec7 | 5094 | int i; |
8bab8dde PM |
5095 | |
5096 | while ((token = strsep(&str, ",")) != NULL) { | |
5097 | if (!*token) | |
5098 | continue; | |
be45c900 | 5099 | |
3ed80a62 | 5100 | for_each_subsys(ss, i) { |
8bab8dde PM |
5101 | if (!strcmp(token, ss->name)) { |
5102 | ss->disabled = 1; | |
5103 | printk(KERN_INFO "Disabling %s control group" | |
5104 | " subsystem\n", ss->name); | |
5105 | break; | |
5106 | } | |
5107 | } | |
5108 | } | |
5109 | return 1; | |
5110 | } | |
5111 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5112 | |
b77d7b60 | 5113 | /** |
5a17f543 | 5114 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
5115 | * @dentry: directory dentry of interest |
5116 | * @ss: subsystem of interest | |
b77d7b60 | 5117 | * |
5a17f543 TH |
5118 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5119 | * to get the corresponding css and return it. If such css doesn't exist | |
5120 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5121 | */ |
5a17f543 TH |
5122 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
5123 | struct cgroup_subsys *ss) | |
e5d1367f | 5124 | { |
2bd59d48 TH |
5125 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5126 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5127 | struct cgroup *cgrp; |
e5d1367f | 5128 | |
35cf0836 | 5129 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5130 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5131 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5132 | return ERR_PTR(-EBADF); |
5133 | ||
5a17f543 TH |
5134 | rcu_read_lock(); |
5135 | ||
2bd59d48 TH |
5136 | /* |
5137 | * This path doesn't originate from kernfs and @kn could already | |
5138 | * have been or be removed at any point. @kn->priv is RCU | |
5139 | * protected for this access. See destroy_locked() for details. | |
5140 | */ | |
5141 | cgrp = rcu_dereference(kn->priv); | |
5142 | if (cgrp) | |
5143 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
5144 | |
5145 | if (!css || !css_tryget(css)) | |
5146 | css = ERR_PTR(-ENOENT); | |
5147 | ||
5148 | rcu_read_unlock(); | |
5149 | return css; | |
e5d1367f | 5150 | } |
e5d1367f | 5151 | |
1cb650b9 LZ |
5152 | /** |
5153 | * css_from_id - lookup css by id | |
5154 | * @id: the cgroup id | |
5155 | * @ss: cgroup subsys to be looked into | |
5156 | * | |
5157 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5158 | * Should be called under rcu_read_lock(). | |
5159 | */ | |
5160 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5161 | { | |
5162 | struct cgroup *cgrp; | |
5163 | ||
ace2bee8 | 5164 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
5165 | |
5166 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
5167 | if (cgrp) | |
d1625964 | 5168 | return cgroup_css(cgrp, ss); |
1cb650b9 | 5169 | return NULL; |
e5d1367f SE |
5170 | } |
5171 | ||
fe693435 | 5172 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5173 | static struct cgroup_subsys_state * |
5174 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5175 | { |
5176 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5177 | ||
5178 | if (!css) | |
5179 | return ERR_PTR(-ENOMEM); | |
5180 | ||
5181 | return css; | |
5182 | } | |
5183 | ||
eb95419b | 5184 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5185 | { |
eb95419b | 5186 | kfree(css); |
fe693435 PM |
5187 | } |
5188 | ||
182446d0 TH |
5189 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5190 | struct cftype *cft) | |
fe693435 | 5191 | { |
182446d0 | 5192 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5193 | } |
5194 | ||
182446d0 TH |
5195 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5196 | struct cftype *cft) | |
fe693435 PM |
5197 | { |
5198 | return (u64)(unsigned long)current->cgroups; | |
5199 | } | |
5200 | ||
182446d0 | 5201 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5202 | struct cftype *cft) |
fe693435 PM |
5203 | { |
5204 | u64 count; | |
5205 | ||
5206 | rcu_read_lock(); | |
a8ad805c | 5207 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5208 | rcu_read_unlock(); |
5209 | return count; | |
5210 | } | |
5211 | ||
2da8ca82 | 5212 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5213 | { |
69d0206c | 5214 | struct cgrp_cset_link *link; |
5abb8855 | 5215 | struct css_set *cset; |
e61734c5 TH |
5216 | char *name_buf; |
5217 | ||
5218 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5219 | if (!name_buf) | |
5220 | return -ENOMEM; | |
7717f7ba | 5221 | |
96d365e0 | 5222 | down_read(&css_set_rwsem); |
7717f7ba | 5223 | rcu_read_lock(); |
5abb8855 | 5224 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5225 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5226 | struct cgroup *c = link->cgrp; |
7717f7ba | 5227 | |
a2dd4247 | 5228 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5229 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5230 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5231 | } |
5232 | rcu_read_unlock(); | |
96d365e0 | 5233 | up_read(&css_set_rwsem); |
e61734c5 | 5234 | kfree(name_buf); |
7717f7ba PM |
5235 | return 0; |
5236 | } | |
5237 | ||
5238 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5239 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5240 | { |
2da8ca82 | 5241 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5242 | struct cgrp_cset_link *link; |
7717f7ba | 5243 | |
96d365e0 | 5244 | down_read(&css_set_rwsem); |
182446d0 | 5245 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5246 | struct css_set *cset = link->cset; |
7717f7ba PM |
5247 | struct task_struct *task; |
5248 | int count = 0; | |
c7561128 | 5249 | |
5abb8855 | 5250 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5251 | |
5abb8855 | 5252 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5253 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5254 | goto overflow; | |
5255 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5256 | } | |
5257 | ||
5258 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5259 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5260 | goto overflow; | |
5261 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5262 | } |
c7561128 TH |
5263 | continue; |
5264 | overflow: | |
5265 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5266 | } |
96d365e0 | 5267 | up_read(&css_set_rwsem); |
7717f7ba PM |
5268 | return 0; |
5269 | } | |
5270 | ||
182446d0 | 5271 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5272 | { |
182446d0 | 5273 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5274 | } |
5275 | ||
5276 | static struct cftype debug_files[] = { | |
fe693435 PM |
5277 | { |
5278 | .name = "taskcount", | |
5279 | .read_u64 = debug_taskcount_read, | |
5280 | }, | |
5281 | ||
5282 | { | |
5283 | .name = "current_css_set", | |
5284 | .read_u64 = current_css_set_read, | |
5285 | }, | |
5286 | ||
5287 | { | |
5288 | .name = "current_css_set_refcount", | |
5289 | .read_u64 = current_css_set_refcount_read, | |
5290 | }, | |
5291 | ||
7717f7ba PM |
5292 | { |
5293 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5294 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5295 | }, |
5296 | ||
5297 | { | |
5298 | .name = "cgroup_css_links", | |
2da8ca82 | 5299 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5300 | }, |
5301 | ||
fe693435 PM |
5302 | { |
5303 | .name = "releasable", | |
5304 | .read_u64 = releasable_read, | |
5305 | }, | |
fe693435 | 5306 | |
4baf6e33 TH |
5307 | { } /* terminate */ |
5308 | }; | |
fe693435 | 5309 | |
073219e9 | 5310 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5311 | .css_alloc = debug_css_alloc, |
5312 | .css_free = debug_css_free, | |
4baf6e33 | 5313 | .base_cftypes = debug_files, |
fe693435 PM |
5314 | }; |
5315 | #endif /* CONFIG_CGROUP_DEBUG */ |