]>
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 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
38 | #include <linux/mm.h> | |
39 | #include <linux/mutex.h> | |
40 | #include <linux/mount.h> | |
41 | #include <linux/pagemap.h> | |
a424316c | 42 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
43 | #include <linux/rcupdate.h> |
44 | #include <linux/sched.h> | |
ddbcc7e8 | 45 | #include <linux/slab.h> |
ddbcc7e8 | 46 | #include <linux/spinlock.h> |
96d365e0 | 47 | #include <linux/rwsem.h> |
ddbcc7e8 | 48 | #include <linux/string.h> |
bbcb81d0 | 49 | #include <linux/sort.h> |
81a6a5cd | 50 | #include <linux/kmod.h> |
846c7bb0 BS |
51 | #include <linux/delayacct.h> |
52 | #include <linux/cgroupstats.h> | |
0ac801fe | 53 | #include <linux/hashtable.h> |
096b7fe0 | 54 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 55 | #include <linux/idr.h> |
d1d9fd33 | 56 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 57 | #include <linux/kthread.h> |
776f02fa | 58 | #include <linux/delay.h> |
846c7bb0 | 59 | |
60063497 | 60 | #include <linux/atomic.h> |
ddbcc7e8 | 61 | |
b1a21367 TH |
62 | /* |
63 | * pidlists linger the following amount before being destroyed. The goal | |
64 | * is avoiding frequent destruction in the middle of consecutive read calls | |
65 | * Expiring in the middle is a performance problem not a correctness one. | |
66 | * 1 sec should be enough. | |
67 | */ | |
68 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
69 | ||
8d7e6fb0 TH |
70 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
71 | MAX_CFTYPE_NAME + 2) | |
72 | ||
e25e2cbb TH |
73 | /* |
74 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
75 | * hierarchy must be performed while holding it. | |
76 | * | |
0e1d768f TH |
77 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
78 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 79 | * |
0e1d768f TH |
80 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
81 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 82 | */ |
2219449a TH |
83 | #ifdef CONFIG_PROVE_RCU |
84 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
85 | DECLARE_RWSEM(css_set_rwsem); |
86 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
87 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 88 | #else |
81a6a5cd | 89 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 90 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
91 | #endif |
92 | ||
6fa4918d | 93 | /* |
15a4c835 TH |
94 | * Protects cgroup_idr and css_idr so that IDs can be released without |
95 | * grabbing cgroup_mutex. | |
6fa4918d TH |
96 | */ |
97 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
98 | ||
69e943b7 TH |
99 | /* |
100 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
101 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
102 | */ | |
103 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 104 | |
8353da1f | 105 | #define cgroup_assert_mutex_or_rcu_locked() \ |
87fb54f1 TH |
106 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
107 | lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 108 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 109 | |
e5fca243 TH |
110 | /* |
111 | * cgroup destruction makes heavy use of work items and there can be a lot | |
112 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
113 | * destruction work items don't end up filling up max_active of system_wq | |
114 | * which may lead to deadlock. | |
115 | */ | |
116 | static struct workqueue_struct *cgroup_destroy_wq; | |
117 | ||
b1a21367 TH |
118 | /* |
119 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
120 | * separate workqueue as flush domain. | |
121 | */ | |
122 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
123 | ||
3ed80a62 | 124 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 125 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 126 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
127 | #include <linux/cgroup_subsys.h> |
128 | }; | |
073219e9 TH |
129 | #undef SUBSYS |
130 | ||
131 | /* array of cgroup subsystem names */ | |
132 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
133 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
134 | #include <linux/cgroup_subsys.h> |
135 | }; | |
073219e9 | 136 | #undef SUBSYS |
ddbcc7e8 | 137 | |
ddbcc7e8 | 138 | /* |
3dd06ffa | 139 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
140 | * unattached - it never has more than a single cgroup, and all tasks are |
141 | * part of that cgroup. | |
ddbcc7e8 | 142 | */ |
a2dd4247 | 143 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 144 | |
a2dd4247 TH |
145 | /* |
146 | * The default hierarchy always exists but is hidden until mounted for the | |
147 | * first time. This is for backward compatibility. | |
148 | */ | |
149 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
150 | |
151 | /* The list of hierarchy roots */ | |
152 | ||
9871bf95 TH |
153 | static LIST_HEAD(cgroup_roots); |
154 | static int cgroup_root_count; | |
ddbcc7e8 | 155 | |
3417ae1f | 156 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 157 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 158 | |
794611a1 | 159 | /* |
0cb51d71 TH |
160 | * Assign a monotonically increasing serial number to csses. It guarantees |
161 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
162 | * Also, as csses are always appended to the parent's ->children list, it | |
163 | * guarantees that sibling csses are always sorted in the ascending serial | |
164 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 165 | */ |
0cb51d71 | 166 | static u64 css_serial_nr_next = 1; |
794611a1 | 167 | |
ddbcc7e8 | 168 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
169 | * check for fork/exit handlers to call. This avoids us having to do |
170 | * extra work in the fork/exit path if none of the subsystems need to | |
171 | * be called. | |
ddbcc7e8 | 172 | */ |
8947f9d5 | 173 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 174 | |
628f7cd4 TH |
175 | static struct cftype cgroup_base_files[]; |
176 | ||
59f5296b | 177 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 178 | static int rebind_subsystems(struct cgroup_root *dst_root, |
69dfa00c | 179 | unsigned int ss_mask); |
42809dd4 | 180 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f8f22e53 | 181 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss); |
9d755d33 | 182 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 183 | static void kill_css(struct cgroup_subsys_state *css); |
2bb566cb TH |
184 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
185 | bool is_add); | |
b1a21367 | 186 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 187 | |
6fa4918d TH |
188 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
189 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
190 | gfp_t gfp_mask) | |
191 | { | |
192 | int ret; | |
193 | ||
194 | idr_preload(gfp_mask); | |
54504e97 | 195 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 196 | ret = idr_alloc(idr, ptr, start, end, gfp_mask); |
54504e97 | 197 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
198 | idr_preload_end(); |
199 | return ret; | |
200 | } | |
201 | ||
202 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
203 | { | |
204 | void *ret; | |
205 | ||
54504e97 | 206 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 207 | ret = idr_replace(idr, ptr, id); |
54504e97 | 208 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
209 | return ret; |
210 | } | |
211 | ||
212 | static void cgroup_idr_remove(struct idr *idr, int id) | |
213 | { | |
54504e97 | 214 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 215 | idr_remove(idr, id); |
54504e97 | 216 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
217 | } |
218 | ||
d51f39b0 TH |
219 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
220 | { | |
221 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
222 | ||
223 | if (parent_css) | |
224 | return container_of(parent_css, struct cgroup, self); | |
225 | return NULL; | |
226 | } | |
227 | ||
95109b62 TH |
228 | /** |
229 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
230 | * @cgrp: the cgroup of interest | |
9d800df1 | 231 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 232 | * |
ca8bdcaf TH |
233 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
234 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
235 | * the caller is responsible for pinning the returned css if it wants to | |
236 | * keep accessing it outside the said locks. This function may return | |
237 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
238 | */ |
239 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 240 | struct cgroup_subsys *ss) |
95109b62 | 241 | { |
ca8bdcaf | 242 | if (ss) |
aec25020 | 243 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 244 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 245 | else |
9d800df1 | 246 | return &cgrp->self; |
95109b62 | 247 | } |
42809dd4 | 248 | |
aec3dfcb TH |
249 | /** |
250 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
251 | * @cgrp: the cgroup of interest | |
9d800df1 | 252 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb TH |
253 | * |
254 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
255 | * as the matching css of the nearest ancestor including self which has @ss | |
256 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
257 | * function is guaranteed to return non-NULL css. | |
258 | */ | |
259 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
260 | struct cgroup_subsys *ss) | |
261 | { | |
262 | lockdep_assert_held(&cgroup_mutex); | |
263 | ||
264 | if (!ss) | |
9d800df1 | 265 | return &cgrp->self; |
aec3dfcb TH |
266 | |
267 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
268 | return NULL; | |
269 | ||
d51f39b0 TH |
270 | while (cgroup_parent(cgrp) && |
271 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
272 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
273 | |
274 | return cgroup_css(cgrp, ss); | |
275 | } | |
276 | ||
ddbcc7e8 | 277 | /* convenient tests for these bits */ |
54766d4a | 278 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 279 | { |
184faf32 | 280 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
281 | } |
282 | ||
b4168640 | 283 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 284 | { |
2bd59d48 | 285 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 286 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
287 | |
288 | /* | |
289 | * This is open and unprotected implementation of cgroup_css(). | |
290 | * seq_css() is only called from a kernfs file operation which has | |
291 | * an active reference on the file. Because all the subsystem | |
292 | * files are drained before a css is disassociated with a cgroup, | |
293 | * the matching css from the cgroup's subsys table is guaranteed to | |
294 | * be and stay valid until the enclosing operation is complete. | |
295 | */ | |
296 | if (cft->ss) | |
297 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
298 | else | |
9d800df1 | 299 | return &cgrp->self; |
59f5296b | 300 | } |
b4168640 | 301 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 302 | |
78574cf9 LZ |
303 | /** |
304 | * cgroup_is_descendant - test ancestry | |
305 | * @cgrp: the cgroup to be tested | |
306 | * @ancestor: possible ancestor of @cgrp | |
307 | * | |
308 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
309 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
310 | * and @ancestor are accessible. | |
311 | */ | |
312 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
313 | { | |
314 | while (cgrp) { | |
315 | if (cgrp == ancestor) | |
316 | return true; | |
d51f39b0 | 317 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
318 | } |
319 | return false; | |
320 | } | |
ddbcc7e8 | 321 | |
e9685a03 | 322 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
323 | { |
324 | const int bits = | |
bd89aabc PM |
325 | (1 << CGRP_RELEASABLE) | |
326 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
327 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
328 | } |
329 | ||
e9685a03 | 330 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 331 | { |
bd89aabc | 332 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
333 | } |
334 | ||
1c6727af TH |
335 | /** |
336 | * for_each_css - iterate all css's of a cgroup | |
337 | * @css: the iteration cursor | |
338 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
339 | * @cgrp: the target cgroup to iterate css's of | |
340 | * | |
aec3dfcb | 341 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
342 | */ |
343 | #define for_each_css(css, ssid, cgrp) \ | |
344 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
345 | if (!((css) = rcu_dereference_check( \ | |
346 | (cgrp)->subsys[(ssid)], \ | |
347 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
348 | else | |
349 | ||
aec3dfcb TH |
350 | /** |
351 | * for_each_e_css - iterate all effective css's of a cgroup | |
352 | * @css: the iteration cursor | |
353 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
354 | * @cgrp: the target cgroup to iterate css's of | |
355 | * | |
356 | * Should be called under cgroup_[tree_]mutex. | |
357 | */ | |
358 | #define for_each_e_css(css, ssid, cgrp) \ | |
359 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
360 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
361 | ; \ | |
362 | else | |
363 | ||
30159ec7 | 364 | /** |
3ed80a62 | 365 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 366 | * @ss: the iteration cursor |
780cd8b3 | 367 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 368 | */ |
780cd8b3 | 369 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
370 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
371 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 372 | |
985ed670 TH |
373 | /* iterate across the hierarchies */ |
374 | #define for_each_root(root) \ | |
5549c497 | 375 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 376 | |
f8f22e53 TH |
377 | /* iterate over child cgrps, lock should be held throughout iteration */ |
378 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 379 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 380 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
381 | cgroup_is_dead(child); })) \ |
382 | ; \ | |
383 | else | |
384 | ||
81a6a5cd PM |
385 | /* the list of cgroups eligible for automatic release. Protected by |
386 | * release_list_lock */ | |
387 | static LIST_HEAD(release_list); | |
cdcc136f | 388 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
389 | static void cgroup_release_agent(struct work_struct *work); |
390 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 391 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 392 | |
69d0206c TH |
393 | /* |
394 | * A cgroup can be associated with multiple css_sets as different tasks may | |
395 | * belong to different cgroups on different hierarchies. In the other | |
396 | * direction, a css_set is naturally associated with multiple cgroups. | |
397 | * This M:N relationship is represented by the following link structure | |
398 | * which exists for each association and allows traversing the associations | |
399 | * from both sides. | |
400 | */ | |
401 | struct cgrp_cset_link { | |
402 | /* the cgroup and css_set this link associates */ | |
403 | struct cgroup *cgrp; | |
404 | struct css_set *cset; | |
405 | ||
406 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
407 | struct list_head cset_link; | |
408 | ||
409 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
410 | struct list_head cgrp_link; | |
817929ec PM |
411 | }; |
412 | ||
172a2c06 TH |
413 | /* |
414 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
415 | * hierarchies being mounted. It contains a pointer to the root state |
416 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
417 | * reference-counted, to improve performance when child cgroups | |
418 | * haven't been created. | |
419 | */ | |
5024ae29 | 420 | struct css_set init_css_set = { |
172a2c06 TH |
421 | .refcount = ATOMIC_INIT(1), |
422 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
423 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
424 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
425 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
426 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
427 | }; | |
817929ec | 428 | |
172a2c06 | 429 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 430 | |
842b597e TH |
431 | /** |
432 | * cgroup_update_populated - updated populated count of a cgroup | |
433 | * @cgrp: the target cgroup | |
434 | * @populated: inc or dec populated count | |
435 | * | |
436 | * @cgrp is either getting the first task (css_set) or losing the last. | |
437 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
438 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
439 | * cgroup and all its descendants are empty. | |
440 | * | |
441 | * @cgrp's interface file "cgroup.populated" is zero if | |
442 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
443 | * changes from or to zero, userland is notified that the content of the | |
444 | * interface file has changed. This can be used to detect when @cgrp and | |
445 | * its descendants become populated or empty. | |
446 | */ | |
447 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
448 | { | |
449 | lockdep_assert_held(&css_set_rwsem); | |
450 | ||
451 | do { | |
452 | bool trigger; | |
453 | ||
454 | if (populated) | |
455 | trigger = !cgrp->populated_cnt++; | |
456 | else | |
457 | trigger = !--cgrp->populated_cnt; | |
458 | ||
459 | if (!trigger) | |
460 | break; | |
461 | ||
462 | if (cgrp->populated_kn) | |
463 | kernfs_notify(cgrp->populated_kn); | |
d51f39b0 | 464 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
465 | } while (cgrp); |
466 | } | |
467 | ||
7717f7ba PM |
468 | /* |
469 | * hash table for cgroup groups. This improves the performance to find | |
470 | * an existing css_set. This hash doesn't (currently) take into | |
471 | * account cgroups in empty hierarchies. | |
472 | */ | |
472b1053 | 473 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 474 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 475 | |
0ac801fe | 476 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 477 | { |
0ac801fe | 478 | unsigned long key = 0UL; |
30159ec7 TH |
479 | struct cgroup_subsys *ss; |
480 | int i; | |
472b1053 | 481 | |
30159ec7 | 482 | for_each_subsys(ss, i) |
0ac801fe LZ |
483 | key += (unsigned long)css[i]; |
484 | key = (key >> 16) ^ key; | |
472b1053 | 485 | |
0ac801fe | 486 | return key; |
472b1053 LZ |
487 | } |
488 | ||
89c5509b | 489 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 490 | { |
69d0206c | 491 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
492 | struct cgroup_subsys *ss; |
493 | int ssid; | |
5abb8855 | 494 | |
89c5509b TH |
495 | lockdep_assert_held(&css_set_rwsem); |
496 | ||
497 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 498 | return; |
81a6a5cd | 499 | |
2c6ab6d2 | 500 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
501 | for_each_subsys(ss, ssid) |
502 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 503 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
504 | css_set_count--; |
505 | ||
69d0206c | 506 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 507 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 508 | |
69d0206c TH |
509 | list_del(&link->cset_link); |
510 | list_del(&link->cgrp_link); | |
71b5707e | 511 | |
96d365e0 | 512 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
513 | if (list_empty(&cgrp->cset_links)) { |
514 | cgroup_update_populated(cgrp, false); | |
515 | if (notify_on_release(cgrp)) { | |
516 | if (taskexit) | |
517 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
518 | check_for_release(cgrp); | |
519 | } | |
81a6a5cd | 520 | } |
2c6ab6d2 PM |
521 | |
522 | kfree(link); | |
81a6a5cd | 523 | } |
2c6ab6d2 | 524 | |
5abb8855 | 525 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
526 | } |
527 | ||
89c5509b TH |
528 | static void put_css_set(struct css_set *cset, bool taskexit) |
529 | { | |
530 | /* | |
531 | * Ensure that the refcount doesn't hit zero while any readers | |
532 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
533 | * rwlock | |
534 | */ | |
535 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
536 | return; | |
537 | ||
538 | down_write(&css_set_rwsem); | |
539 | put_css_set_locked(cset, taskexit); | |
540 | up_write(&css_set_rwsem); | |
541 | } | |
542 | ||
817929ec PM |
543 | /* |
544 | * refcounted get/put for css_set objects | |
545 | */ | |
5abb8855 | 546 | static inline void get_css_set(struct css_set *cset) |
817929ec | 547 | { |
5abb8855 | 548 | atomic_inc(&cset->refcount); |
817929ec PM |
549 | } |
550 | ||
b326f9d0 | 551 | /** |
7717f7ba | 552 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
553 | * @cset: candidate css_set being tested |
554 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
555 | * @new_cgrp: cgroup that's being entered by the task |
556 | * @template: desired set of css pointers in css_set (pre-calculated) | |
557 | * | |
6f4b7e63 | 558 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
559 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
560 | */ | |
5abb8855 TH |
561 | static bool compare_css_sets(struct css_set *cset, |
562 | struct css_set *old_cset, | |
7717f7ba PM |
563 | struct cgroup *new_cgrp, |
564 | struct cgroup_subsys_state *template[]) | |
565 | { | |
566 | struct list_head *l1, *l2; | |
567 | ||
aec3dfcb TH |
568 | /* |
569 | * On the default hierarchy, there can be csets which are | |
570 | * associated with the same set of cgroups but different csses. | |
571 | * Let's first ensure that csses match. | |
572 | */ | |
573 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 574 | return false; |
7717f7ba PM |
575 | |
576 | /* | |
577 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
578 | * different cgroups in hierarchies. As different cgroups may |
579 | * share the same effective css, this comparison is always | |
580 | * necessary. | |
7717f7ba | 581 | */ |
69d0206c TH |
582 | l1 = &cset->cgrp_links; |
583 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 584 | while (1) { |
69d0206c | 585 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 586 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
587 | |
588 | l1 = l1->next; | |
589 | l2 = l2->next; | |
590 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
591 | if (l1 == &cset->cgrp_links) { |
592 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
593 | break; |
594 | } else { | |
69d0206c | 595 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
596 | } |
597 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
598 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
599 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
600 | cgrp1 = link1->cgrp; | |
601 | cgrp2 = link2->cgrp; | |
7717f7ba | 602 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 603 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
604 | |
605 | /* | |
606 | * If this hierarchy is the hierarchy of the cgroup | |
607 | * that's changing, then we need to check that this | |
608 | * css_set points to the new cgroup; if it's any other | |
609 | * hierarchy, then this css_set should point to the | |
610 | * same cgroup as the old css_set. | |
611 | */ | |
5abb8855 TH |
612 | if (cgrp1->root == new_cgrp->root) { |
613 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
614 | return false; |
615 | } else { | |
5abb8855 | 616 | if (cgrp1 != cgrp2) |
7717f7ba PM |
617 | return false; |
618 | } | |
619 | } | |
620 | return true; | |
621 | } | |
622 | ||
b326f9d0 TH |
623 | /** |
624 | * find_existing_css_set - init css array and find the matching css_set | |
625 | * @old_cset: the css_set that we're using before the cgroup transition | |
626 | * @cgrp: the cgroup that we're moving into | |
627 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 628 | */ |
5abb8855 TH |
629 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
630 | struct cgroup *cgrp, | |
631 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 632 | { |
3dd06ffa | 633 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 634 | struct cgroup_subsys *ss; |
5abb8855 | 635 | struct css_set *cset; |
0ac801fe | 636 | unsigned long key; |
b326f9d0 | 637 | int i; |
817929ec | 638 | |
aae8aab4 BB |
639 | /* |
640 | * Build the set of subsystem state objects that we want to see in the | |
641 | * new css_set. while subsystems can change globally, the entries here | |
642 | * won't change, so no need for locking. | |
643 | */ | |
30159ec7 | 644 | for_each_subsys(ss, i) { |
f392e51c | 645 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
646 | /* |
647 | * @ss is in this hierarchy, so we want the | |
648 | * effective css from @cgrp. | |
649 | */ | |
650 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 651 | } else { |
aec3dfcb TH |
652 | /* |
653 | * @ss is not in this hierarchy, so we don't want | |
654 | * to change the css. | |
655 | */ | |
5abb8855 | 656 | template[i] = old_cset->subsys[i]; |
817929ec PM |
657 | } |
658 | } | |
659 | ||
0ac801fe | 660 | key = css_set_hash(template); |
5abb8855 TH |
661 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
662 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
663 | continue; |
664 | ||
665 | /* This css_set matches what we need */ | |
5abb8855 | 666 | return cset; |
472b1053 | 667 | } |
817929ec PM |
668 | |
669 | /* No existing cgroup group matched */ | |
670 | return NULL; | |
671 | } | |
672 | ||
69d0206c | 673 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 674 | { |
69d0206c | 675 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 676 | |
69d0206c TH |
677 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
678 | list_del(&link->cset_link); | |
36553434 LZ |
679 | kfree(link); |
680 | } | |
681 | } | |
682 | ||
69d0206c TH |
683 | /** |
684 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
685 | * @count: the number of links to allocate | |
686 | * @tmp_links: list_head the allocated links are put on | |
687 | * | |
688 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
689 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 690 | */ |
69d0206c | 691 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 692 | { |
69d0206c | 693 | struct cgrp_cset_link *link; |
817929ec | 694 | int i; |
69d0206c TH |
695 | |
696 | INIT_LIST_HEAD(tmp_links); | |
697 | ||
817929ec | 698 | for (i = 0; i < count; i++) { |
f4f4be2b | 699 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 700 | if (!link) { |
69d0206c | 701 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
702 | return -ENOMEM; |
703 | } | |
69d0206c | 704 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
705 | } |
706 | return 0; | |
707 | } | |
708 | ||
c12f65d4 LZ |
709 | /** |
710 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 711 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 712 | * @cset: the css_set to be linked |
c12f65d4 LZ |
713 | * @cgrp: the destination cgroup |
714 | */ | |
69d0206c TH |
715 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
716 | struct cgroup *cgrp) | |
c12f65d4 | 717 | { |
69d0206c | 718 | struct cgrp_cset_link *link; |
c12f65d4 | 719 | |
69d0206c | 720 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
721 | |
722 | if (cgroup_on_dfl(cgrp)) | |
723 | cset->dfl_cgrp = cgrp; | |
724 | ||
69d0206c TH |
725 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
726 | link->cset = cset; | |
7717f7ba | 727 | link->cgrp = cgrp; |
842b597e TH |
728 | |
729 | if (list_empty(&cgrp->cset_links)) | |
730 | cgroup_update_populated(cgrp, true); | |
69d0206c | 731 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 732 | |
7717f7ba PM |
733 | /* |
734 | * Always add links to the tail of the list so that the list | |
735 | * is sorted by order of hierarchy creation | |
736 | */ | |
69d0206c | 737 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
738 | } |
739 | ||
b326f9d0 TH |
740 | /** |
741 | * find_css_set - return a new css_set with one cgroup updated | |
742 | * @old_cset: the baseline css_set | |
743 | * @cgrp: the cgroup to be updated | |
744 | * | |
745 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
746 | * substituted into the appropriate hierarchy. | |
817929ec | 747 | */ |
5abb8855 TH |
748 | static struct css_set *find_css_set(struct css_set *old_cset, |
749 | struct cgroup *cgrp) | |
817929ec | 750 | { |
b326f9d0 | 751 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 752 | struct css_set *cset; |
69d0206c TH |
753 | struct list_head tmp_links; |
754 | struct cgrp_cset_link *link; | |
2d8f243a | 755 | struct cgroup_subsys *ss; |
0ac801fe | 756 | unsigned long key; |
2d8f243a | 757 | int ssid; |
472b1053 | 758 | |
b326f9d0 TH |
759 | lockdep_assert_held(&cgroup_mutex); |
760 | ||
817929ec PM |
761 | /* First see if we already have a cgroup group that matches |
762 | * the desired set */ | |
96d365e0 | 763 | down_read(&css_set_rwsem); |
5abb8855 TH |
764 | cset = find_existing_css_set(old_cset, cgrp, template); |
765 | if (cset) | |
766 | get_css_set(cset); | |
96d365e0 | 767 | up_read(&css_set_rwsem); |
817929ec | 768 | |
5abb8855 TH |
769 | if (cset) |
770 | return cset; | |
817929ec | 771 | |
f4f4be2b | 772 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 773 | if (!cset) |
817929ec PM |
774 | return NULL; |
775 | ||
69d0206c | 776 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 777 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 778 | kfree(cset); |
817929ec PM |
779 | return NULL; |
780 | } | |
781 | ||
5abb8855 | 782 | atomic_set(&cset->refcount, 1); |
69d0206c | 783 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 784 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 785 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 786 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 787 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 788 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
789 | |
790 | /* Copy the set of subsystem state objects generated in | |
791 | * find_existing_css_set() */ | |
5abb8855 | 792 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 793 | |
96d365e0 | 794 | down_write(&css_set_rwsem); |
817929ec | 795 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 796 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 797 | struct cgroup *c = link->cgrp; |
69d0206c | 798 | |
7717f7ba PM |
799 | if (c->root == cgrp->root) |
800 | c = cgrp; | |
69d0206c | 801 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 802 | } |
817929ec | 803 | |
69d0206c | 804 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 805 | |
817929ec | 806 | css_set_count++; |
472b1053 | 807 | |
2d8f243a | 808 | /* Add @cset to the hash table */ |
5abb8855 TH |
809 | key = css_set_hash(cset->subsys); |
810 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 811 | |
2d8f243a TH |
812 | for_each_subsys(ss, ssid) |
813 | list_add_tail(&cset->e_cset_node[ssid], | |
814 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
815 | ||
96d365e0 | 816 | up_write(&css_set_rwsem); |
817929ec | 817 | |
5abb8855 | 818 | return cset; |
b4f48b63 PM |
819 | } |
820 | ||
3dd06ffa | 821 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 822 | { |
3dd06ffa | 823 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 824 | |
3dd06ffa | 825 | return root_cgrp->root; |
2bd59d48 TH |
826 | } |
827 | ||
3dd06ffa | 828 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
829 | { |
830 | int id; | |
831 | ||
832 | lockdep_assert_held(&cgroup_mutex); | |
833 | ||
985ed670 | 834 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
835 | if (id < 0) |
836 | return id; | |
837 | ||
838 | root->hierarchy_id = id; | |
839 | return 0; | |
840 | } | |
841 | ||
3dd06ffa | 842 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
843 | { |
844 | lockdep_assert_held(&cgroup_mutex); | |
845 | ||
846 | if (root->hierarchy_id) { | |
847 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
848 | root->hierarchy_id = 0; | |
849 | } | |
850 | } | |
851 | ||
3dd06ffa | 852 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
853 | { |
854 | if (root) { | |
855 | /* hierarhcy ID shoulid already have been released */ | |
856 | WARN_ON_ONCE(root->hierarchy_id); | |
857 | ||
858 | idr_destroy(&root->cgroup_idr); | |
859 | kfree(root); | |
860 | } | |
861 | } | |
862 | ||
3dd06ffa | 863 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 864 | { |
3dd06ffa | 865 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 866 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 867 | |
2bd59d48 | 868 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 869 | |
776f02fa | 870 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 871 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 872 | |
f2e85d57 | 873 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 874 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 875 | |
7717f7ba | 876 | /* |
f2e85d57 TH |
877 | * Release all the links from cset_links to this hierarchy's |
878 | * root cgroup | |
7717f7ba | 879 | */ |
96d365e0 | 880 | down_write(&css_set_rwsem); |
f2e85d57 TH |
881 | |
882 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
883 | list_del(&link->cset_link); | |
884 | list_del(&link->cgrp_link); | |
885 | kfree(link); | |
886 | } | |
96d365e0 | 887 | up_write(&css_set_rwsem); |
f2e85d57 TH |
888 | |
889 | if (!list_empty(&root->root_list)) { | |
890 | list_del(&root->root_list); | |
891 | cgroup_root_count--; | |
892 | } | |
893 | ||
894 | cgroup_exit_root_id(root); | |
895 | ||
896 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 897 | |
2bd59d48 | 898 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
899 | cgroup_free_root(root); |
900 | } | |
901 | ||
ceb6a081 TH |
902 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
903 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 904 | struct cgroup_root *root) |
7717f7ba | 905 | { |
7717f7ba PM |
906 | struct cgroup *res = NULL; |
907 | ||
96d365e0 TH |
908 | lockdep_assert_held(&cgroup_mutex); |
909 | lockdep_assert_held(&css_set_rwsem); | |
910 | ||
5abb8855 | 911 | if (cset == &init_css_set) { |
3dd06ffa | 912 | res = &root->cgrp; |
7717f7ba | 913 | } else { |
69d0206c TH |
914 | struct cgrp_cset_link *link; |
915 | ||
916 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 917 | struct cgroup *c = link->cgrp; |
69d0206c | 918 | |
7717f7ba PM |
919 | if (c->root == root) { |
920 | res = c; | |
921 | break; | |
922 | } | |
923 | } | |
924 | } | |
96d365e0 | 925 | |
7717f7ba PM |
926 | BUG_ON(!res); |
927 | return res; | |
928 | } | |
929 | ||
ddbcc7e8 | 930 | /* |
ceb6a081 TH |
931 | * Return the cgroup for "task" from the given hierarchy. Must be |
932 | * called with cgroup_mutex and css_set_rwsem held. | |
933 | */ | |
934 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 935 | struct cgroup_root *root) |
ceb6a081 TH |
936 | { |
937 | /* | |
938 | * No need to lock the task - since we hold cgroup_mutex the | |
939 | * task can't change groups, so the only thing that can happen | |
940 | * is that it exits and its css is set back to init_css_set. | |
941 | */ | |
942 | return cset_cgroup_from_root(task_css_set(task), root); | |
943 | } | |
944 | ||
ddbcc7e8 | 945 | /* |
ddbcc7e8 PM |
946 | * A task must hold cgroup_mutex to modify cgroups. |
947 | * | |
948 | * Any task can increment and decrement the count field without lock. | |
949 | * So in general, code holding cgroup_mutex can't rely on the count | |
950 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 951 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
952 | * means that no tasks are currently attached, therefore there is no |
953 | * way a task attached to that cgroup can fork (the other way to | |
954 | * increment the count). So code holding cgroup_mutex can safely | |
955 | * assume that if the count is zero, it will stay zero. Similarly, if | |
956 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
957 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
958 | * needs that mutex. | |
959 | * | |
ddbcc7e8 PM |
960 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
961 | * (usually) take cgroup_mutex. These are the two most performance | |
962 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
963 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
964 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
965 | * to the release agent with the name of the cgroup (path relative to |
966 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
967 | * |
968 | * A cgroup can only be deleted if both its 'count' of using tasks | |
969 | * is zero, and its list of 'children' cgroups is empty. Since all | |
970 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 971 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 972 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 973 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
974 | * |
975 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 976 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
977 | */ |
978 | ||
69dfa00c | 979 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask); |
2bd59d48 | 980 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 981 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 982 | |
8d7e6fb0 TH |
983 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
984 | char *buf) | |
ddbcc7e8 | 985 | { |
8d7e6fb0 TH |
986 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
987 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
988 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
989 | cft->ss->name, cft->name); | |
990 | else | |
991 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
992 | return buf; | |
ddbcc7e8 PM |
993 | } |
994 | ||
f2e85d57 TH |
995 | /** |
996 | * cgroup_file_mode - deduce file mode of a control file | |
997 | * @cft: the control file in question | |
998 | * | |
999 | * returns cft->mode if ->mode is not 0 | |
1000 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
1001 | * returns S_IRUGO if it has only a read handler | |
1002 | * returns S_IWUSR if it has only a write hander | |
1003 | */ | |
1004 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1005 | { |
f2e85d57 | 1006 | umode_t mode = 0; |
65dff759 | 1007 | |
f2e85d57 TH |
1008 | if (cft->mode) |
1009 | return cft->mode; | |
1010 | ||
1011 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
1012 | mode |= S_IRUGO; | |
1013 | ||
6770c64e | 1014 | if (cft->write_u64 || cft->write_s64 || cft->write) |
f2e85d57 TH |
1015 | mode |= S_IWUSR; |
1016 | ||
1017 | return mode; | |
65dff759 LZ |
1018 | } |
1019 | ||
59f5296b | 1020 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 1021 | { |
2bd59d48 | 1022 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
9d755d33 | 1023 | css_get(&cgrp->self); |
ddbcc7e8 PM |
1024 | } |
1025 | ||
59f5296b | 1026 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 1027 | { |
9d755d33 | 1028 | css_put(&cgrp->self); |
ddbcc7e8 | 1029 | } |
05ef1d7c | 1030 | |
a9746d8d TH |
1031 | /** |
1032 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1033 | * @kn: the kernfs_node being serviced | |
1034 | * | |
1035 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1036 | * the method finishes if locking succeeded. Note that once this function | |
1037 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1038 | * inaccessible any time. If the caller intends to continue to access the | |
1039 | * cgroup, it should pin it before invoking this function. | |
1040 | */ | |
1041 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
1042 | { | |
1043 | struct cgroup *cgrp; | |
1044 | ||
1045 | if (kernfs_type(kn) == KERNFS_DIR) | |
1046 | cgrp = kn->priv; | |
1047 | else | |
1048 | cgrp = kn->parent->priv; | |
1049 | ||
1050 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1051 | |
1052 | kernfs_unbreak_active_protection(kn); | |
1053 | cgroup_put(cgrp); | |
1054 | } | |
1055 | ||
1056 | /** | |
1057 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1058 | * @kn: the kernfs_node being serviced | |
1059 | * | |
1060 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1061 | * @kn. It breaks the active protection, performs cgroup locking and | |
1062 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1063 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1064 | * matching cgroup_kn_unlock() invocation. | |
1065 | * | |
1066 | * Any cgroup kernfs method implementation which requires locking the | |
1067 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1068 | * locking under kernfs active protection and allows all kernfs operations | |
1069 | * including self-removal. | |
1070 | */ | |
1071 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
1072 | { | |
1073 | struct cgroup *cgrp; | |
1074 | ||
1075 | if (kernfs_type(kn) == KERNFS_DIR) | |
1076 | cgrp = kn->priv; | |
1077 | else | |
1078 | cgrp = kn->parent->priv; | |
1079 | ||
1080 | /* | |
01f6474c | 1081 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1082 | * active_ref. cgroup liveliness check alone provides enough |
1083 | * protection against removal. Ensure @cgrp stays accessible and | |
1084 | * break the active_ref protection. | |
1085 | */ | |
1086 | cgroup_get(cgrp); | |
1087 | kernfs_break_active_protection(kn); | |
1088 | ||
a9746d8d TH |
1089 | mutex_lock(&cgroup_mutex); |
1090 | ||
1091 | if (!cgroup_is_dead(cgrp)) | |
1092 | return cgrp; | |
1093 | ||
1094 | cgroup_kn_unlock(kn); | |
1095 | return NULL; | |
1096 | } | |
1097 | ||
2739d3cc | 1098 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1099 | { |
2bd59d48 | 1100 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1101 | |
01f6474c | 1102 | lockdep_assert_held(&cgroup_mutex); |
2bd59d48 | 1103 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1104 | } |
1105 | ||
13af07df | 1106 | /** |
628f7cd4 | 1107 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1108 | * @cgrp: target cgroup |
13af07df AR |
1109 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1110 | */ | |
69dfa00c | 1111 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
05ef1d7c | 1112 | { |
13af07df | 1113 | struct cgroup_subsys *ss; |
b420ba7d | 1114 | int i; |
05ef1d7c | 1115 | |
b420ba7d | 1116 | for_each_subsys(ss, i) { |
0adb0704 | 1117 | struct cftype *cfts; |
b420ba7d | 1118 | |
69dfa00c | 1119 | if (!(subsys_mask & (1 << i))) |
13af07df | 1120 | continue; |
0adb0704 TH |
1121 | list_for_each_entry(cfts, &ss->cfts, node) |
1122 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1123 | } |
ddbcc7e8 PM |
1124 | } |
1125 | ||
69dfa00c | 1126 | static int rebind_subsystems(struct cgroup_root *dst_root, unsigned int ss_mask) |
ddbcc7e8 | 1127 | { |
30159ec7 | 1128 | struct cgroup_subsys *ss; |
2d8f243a | 1129 | int ssid, i, ret; |
ddbcc7e8 | 1130 | |
ace2bee8 | 1131 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1132 | |
5df36032 TH |
1133 | for_each_subsys(ss, ssid) { |
1134 | if (!(ss_mask & (1 << ssid))) | |
1135 | continue; | |
aae8aab4 | 1136 | |
7fd8c565 TH |
1137 | /* if @ss has non-root csses attached to it, can't move */ |
1138 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1139 | return -EBUSY; |
1d5be6b2 | 1140 | |
5df36032 | 1141 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1142 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1143 | return -EBUSY; |
ddbcc7e8 PM |
1144 | } |
1145 | ||
a2dd4247 TH |
1146 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1147 | if (ret) { | |
1148 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1149 | return ret; |
ddbcc7e8 | 1150 | |
a2dd4247 TH |
1151 | /* |
1152 | * Rebinding back to the default root is not allowed to | |
1153 | * fail. Using both default and non-default roots should | |
1154 | * be rare. Moving subsystems back and forth even more so. | |
1155 | * Just warn about it and continue. | |
1156 | */ | |
1157 | if (cgrp_dfl_root_visible) { | |
69dfa00c | 1158 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", |
a2a1f9ea | 1159 | ret, ss_mask); |
ed3d261b | 1160 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); |
a2dd4247 | 1161 | } |
5df36032 | 1162 | } |
3126121f TH |
1163 | |
1164 | /* | |
1165 | * Nothing can fail from this point on. Remove files for the | |
1166 | * removed subsystems and rebind each subsystem. | |
1167 | */ | |
5df36032 | 1168 | for_each_subsys(ss, ssid) |
a2dd4247 | 1169 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1170 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
a8a648c4 | 1171 | |
5df36032 | 1172 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1173 | struct cgroup_root *src_root; |
5df36032 | 1174 | struct cgroup_subsys_state *css; |
2d8f243a | 1175 | struct css_set *cset; |
a8a648c4 | 1176 | |
5df36032 TH |
1177 | if (!(ss_mask & (1 << ssid))) |
1178 | continue; | |
a8a648c4 | 1179 | |
5df36032 | 1180 | src_root = ss->root; |
3dd06ffa | 1181 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1182 | |
3dd06ffa | 1183 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1184 | |
3dd06ffa TH |
1185 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1186 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1187 | ss->root = dst_root; |
3dd06ffa | 1188 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1189 | |
2d8f243a TH |
1190 | down_write(&css_set_rwsem); |
1191 | hash_for_each(css_set_table, i, cset, hlist) | |
1192 | list_move_tail(&cset->e_cset_node[ss->id], | |
1193 | &dst_root->cgrp.e_csets[ss->id]); | |
1194 | up_write(&css_set_rwsem); | |
1195 | ||
f392e51c TH |
1196 | src_root->subsys_mask &= ~(1 << ssid); |
1197 | src_root->cgrp.child_subsys_mask &= ~(1 << ssid); | |
1198 | ||
bd53d617 | 1199 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1200 | dst_root->subsys_mask |= 1 << ssid; |
bd53d617 TH |
1201 | if (dst_root != &cgrp_dfl_root) |
1202 | dst_root->cgrp.child_subsys_mask |= 1 << ssid; | |
a8a648c4 | 1203 | |
5df36032 TH |
1204 | if (ss->bind) |
1205 | ss->bind(css); | |
ddbcc7e8 | 1206 | } |
ddbcc7e8 | 1207 | |
a2dd4247 | 1208 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1209 | return 0; |
1210 | } | |
1211 | ||
2bd59d48 TH |
1212 | static int cgroup_show_options(struct seq_file *seq, |
1213 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1214 | { |
3dd06ffa | 1215 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1216 | struct cgroup_subsys *ss; |
b85d2040 | 1217 | int ssid; |
ddbcc7e8 | 1218 | |
b85d2040 | 1219 | for_each_subsys(ss, ssid) |
f392e51c | 1220 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1221 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1222 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1223 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1224 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1225 | seq_puts(seq, ",noprefix"); |
93438629 | 1226 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1227 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1228 | |
1229 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1230 | if (strlen(root->release_agent_path)) |
1231 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1232 | spin_unlock(&release_agent_path_lock); |
1233 | ||
3dd06ffa | 1234 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1235 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1236 | if (strlen(root->name)) |
1237 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1238 | return 0; |
1239 | } | |
1240 | ||
1241 | struct cgroup_sb_opts { | |
69dfa00c TH |
1242 | unsigned int subsys_mask; |
1243 | unsigned int flags; | |
81a6a5cd | 1244 | char *release_agent; |
2260e7fc | 1245 | bool cpuset_clone_children; |
c6d57f33 | 1246 | char *name; |
2c6ab6d2 PM |
1247 | /* User explicitly requested empty subsystem */ |
1248 | bool none; | |
ddbcc7e8 PM |
1249 | }; |
1250 | ||
cf5d5941 | 1251 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1252 | { |
32a8cf23 DL |
1253 | char *token, *o = data; |
1254 | bool all_ss = false, one_ss = false; | |
69dfa00c | 1255 | unsigned int mask = -1U; |
30159ec7 TH |
1256 | struct cgroup_subsys *ss; |
1257 | int i; | |
f9ab5b5b LZ |
1258 | |
1259 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1260 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1261 | #endif |
ddbcc7e8 | 1262 | |
c6d57f33 | 1263 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1264 | |
1265 | while ((token = strsep(&o, ",")) != NULL) { | |
1266 | if (!*token) | |
1267 | return -EINVAL; | |
32a8cf23 | 1268 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1269 | /* Explicitly have no subsystems */ |
1270 | opts->none = true; | |
32a8cf23 DL |
1271 | continue; |
1272 | } | |
1273 | if (!strcmp(token, "all")) { | |
1274 | /* Mutually exclusive option 'all' + subsystem name */ | |
1275 | if (one_ss) | |
1276 | return -EINVAL; | |
1277 | all_ss = true; | |
1278 | continue; | |
1279 | } | |
873fe09e TH |
1280 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1281 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1282 | continue; | |
1283 | } | |
32a8cf23 | 1284 | if (!strcmp(token, "noprefix")) { |
93438629 | 1285 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1286 | continue; |
1287 | } | |
1288 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1289 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1290 | continue; |
1291 | } | |
03b1cde6 | 1292 | if (!strcmp(token, "xattr")) { |
93438629 | 1293 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1294 | continue; |
1295 | } | |
32a8cf23 | 1296 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1297 | /* Specifying two release agents is forbidden */ |
1298 | if (opts->release_agent) | |
1299 | return -EINVAL; | |
c6d57f33 | 1300 | opts->release_agent = |
e400c285 | 1301 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1302 | if (!opts->release_agent) |
1303 | return -ENOMEM; | |
32a8cf23 DL |
1304 | continue; |
1305 | } | |
1306 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1307 | const char *name = token + 5; |
1308 | /* Can't specify an empty name */ | |
1309 | if (!strlen(name)) | |
1310 | return -EINVAL; | |
1311 | /* Must match [\w.-]+ */ | |
1312 | for (i = 0; i < strlen(name); i++) { | |
1313 | char c = name[i]; | |
1314 | if (isalnum(c)) | |
1315 | continue; | |
1316 | if ((c == '.') || (c == '-') || (c == '_')) | |
1317 | continue; | |
1318 | return -EINVAL; | |
1319 | } | |
1320 | /* Specifying two names is forbidden */ | |
1321 | if (opts->name) | |
1322 | return -EINVAL; | |
1323 | opts->name = kstrndup(name, | |
e400c285 | 1324 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1325 | GFP_KERNEL); |
1326 | if (!opts->name) | |
1327 | return -ENOMEM; | |
32a8cf23 DL |
1328 | |
1329 | continue; | |
1330 | } | |
1331 | ||
30159ec7 | 1332 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1333 | if (strcmp(token, ss->name)) |
1334 | continue; | |
1335 | if (ss->disabled) | |
1336 | continue; | |
1337 | ||
1338 | /* Mutually exclusive option 'all' + subsystem name */ | |
1339 | if (all_ss) | |
1340 | return -EINVAL; | |
69dfa00c | 1341 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1342 | one_ss = true; |
1343 | ||
1344 | break; | |
1345 | } | |
1346 | if (i == CGROUP_SUBSYS_COUNT) | |
1347 | return -ENOENT; | |
1348 | } | |
1349 | ||
2c6ab6d2 PM |
1350 | /* Consistency checks */ |
1351 | ||
873fe09e | 1352 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1353 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
873fe09e | 1354 | |
d3ba07c3 TH |
1355 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1356 | opts->cpuset_clone_children || opts->release_agent || | |
1357 | opts->name) { | |
ed3d261b | 1358 | pr_err("sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); |
873fe09e TH |
1359 | return -EINVAL; |
1360 | } | |
a2dd4247 TH |
1361 | } else { |
1362 | /* | |
1363 | * If the 'all' option was specified select all the | |
1364 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1365 | * name options were not specified, let's default to 'all' | |
1366 | */ | |
1367 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1368 | for_each_subsys(ss, i) | |
1369 | if (!ss->disabled) | |
69dfa00c | 1370 | opts->subsys_mask |= (1 << i); |
873fe09e | 1371 | |
a2dd4247 TH |
1372 | /* |
1373 | * We either have to specify by name or by subsystems. (So | |
1374 | * all empty hierarchies must have a name). | |
1375 | */ | |
1376 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1377 | return -EINVAL; |
873fe09e TH |
1378 | } |
1379 | ||
f9ab5b5b LZ |
1380 | /* |
1381 | * Option noprefix was introduced just for backward compatibility | |
1382 | * with the old cpuset, so we allow noprefix only if mounting just | |
1383 | * the cpuset subsystem. | |
1384 | */ | |
93438629 | 1385 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1386 | return -EINVAL; |
1387 | ||
2c6ab6d2 PM |
1388 | |
1389 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1390 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1391 | return -EINVAL; |
1392 | ||
ddbcc7e8 PM |
1393 | return 0; |
1394 | } | |
1395 | ||
2bd59d48 | 1396 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1397 | { |
1398 | int ret = 0; | |
3dd06ffa | 1399 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1400 | struct cgroup_sb_opts opts; |
69dfa00c | 1401 | unsigned int added_mask, removed_mask; |
ddbcc7e8 | 1402 | |
873fe09e | 1403 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1404 | pr_err("sane_behavior: remount is not allowed\n"); |
873fe09e TH |
1405 | return -EINVAL; |
1406 | } | |
1407 | ||
ddbcc7e8 PM |
1408 | mutex_lock(&cgroup_mutex); |
1409 | ||
1410 | /* See what subsystems are wanted */ | |
1411 | ret = parse_cgroupfs_options(data, &opts); | |
1412 | if (ret) | |
1413 | goto out_unlock; | |
1414 | ||
f392e51c | 1415 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1416 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1417 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1418 | |
f392e51c TH |
1419 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1420 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1421 | |
cf5d5941 | 1422 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1423 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1424 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1425 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
0ce6cba3 TH |
1426 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", |
1427 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1428 | ret = -EINVAL; |
1429 | goto out_unlock; | |
1430 | } | |
1431 | ||
f172e67c | 1432 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1433 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1434 | ret = -EBUSY; |
0670e08b | 1435 | goto out_unlock; |
cf5d5941 | 1436 | } |
ddbcc7e8 | 1437 | |
5df36032 | 1438 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1439 | if (ret) |
0670e08b | 1440 | goto out_unlock; |
ddbcc7e8 | 1441 | |
3dd06ffa | 1442 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1443 | |
69e943b7 TH |
1444 | if (opts.release_agent) { |
1445 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1446 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1447 | spin_unlock(&release_agent_path_lock); |
1448 | } | |
ddbcc7e8 | 1449 | out_unlock: |
66bdc9cf | 1450 | kfree(opts.release_agent); |
c6d57f33 | 1451 | kfree(opts.name); |
ddbcc7e8 | 1452 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1453 | return ret; |
1454 | } | |
1455 | ||
afeb0f9f TH |
1456 | /* |
1457 | * To reduce the fork() overhead for systems that are not actually using | |
1458 | * their cgroups capability, we don't maintain the lists running through | |
1459 | * each css_set to its tasks until we see the list actually used - in other | |
1460 | * words after the first mount. | |
1461 | */ | |
1462 | static bool use_task_css_set_links __read_mostly; | |
1463 | ||
1464 | static void cgroup_enable_task_cg_lists(void) | |
1465 | { | |
1466 | struct task_struct *p, *g; | |
1467 | ||
96d365e0 | 1468 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1469 | |
1470 | if (use_task_css_set_links) | |
1471 | goto out_unlock; | |
1472 | ||
1473 | use_task_css_set_links = true; | |
1474 | ||
1475 | /* | |
1476 | * We need tasklist_lock because RCU is not safe against | |
1477 | * while_each_thread(). Besides, a forking task that has passed | |
1478 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1479 | * is not guaranteed to have its child immediately visible in the | |
1480 | * tasklist if we walk through it with RCU. | |
1481 | */ | |
1482 | read_lock(&tasklist_lock); | |
1483 | do_each_thread(g, p) { | |
afeb0f9f TH |
1484 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1485 | task_css_set(p) != &init_css_set); | |
1486 | ||
1487 | /* | |
1488 | * We should check if the process is exiting, otherwise | |
1489 | * it will race with cgroup_exit() in that the list | |
1490 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1491 | * Do it while holding siglock so that we don't end up |
1492 | * racing against cgroup_exit(). | |
afeb0f9f | 1493 | */ |
f153ad11 | 1494 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1495 | if (!(p->flags & PF_EXITING)) { |
1496 | struct css_set *cset = task_css_set(p); | |
1497 | ||
1498 | list_add(&p->cg_list, &cset->tasks); | |
1499 | get_css_set(cset); | |
1500 | } | |
f153ad11 | 1501 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1502 | } while_each_thread(g, p); |
1503 | read_unlock(&tasklist_lock); | |
1504 | out_unlock: | |
96d365e0 | 1505 | up_write(&css_set_rwsem); |
afeb0f9f | 1506 | } |
ddbcc7e8 | 1507 | |
cc31edce PM |
1508 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1509 | { | |
2d8f243a TH |
1510 | struct cgroup_subsys *ss; |
1511 | int ssid; | |
1512 | ||
d5c419b6 TH |
1513 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1514 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1515 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1516 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1517 | INIT_LIST_HEAD(&cgrp->pidlists); |
1518 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1519 | cgrp->self.cgroup = cgrp; |
184faf32 | 1520 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1521 | |
1522 | for_each_subsys(ss, ssid) | |
1523 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1524 | |
1525 | init_waitqueue_head(&cgrp->offline_waitq); | |
cc31edce | 1526 | } |
c6d57f33 | 1527 | |
3dd06ffa | 1528 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1529 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1530 | { |
3dd06ffa | 1531 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1532 | |
ddbcc7e8 | 1533 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1534 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1535 | cgrp->root = root; |
cc31edce | 1536 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1537 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1538 | |
c6d57f33 PM |
1539 | root->flags = opts->flags; |
1540 | if (opts->release_agent) | |
1541 | strcpy(root->release_agent_path, opts->release_agent); | |
1542 | if (opts->name) | |
1543 | strcpy(root->name, opts->name); | |
2260e7fc | 1544 | if (opts->cpuset_clone_children) |
3dd06ffa | 1545 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1546 | } |
1547 | ||
69dfa00c | 1548 | static int cgroup_setup_root(struct cgroup_root *root, unsigned int ss_mask) |
2c6ab6d2 | 1549 | { |
d427dfeb | 1550 | LIST_HEAD(tmp_links); |
3dd06ffa | 1551 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1552 | struct css_set *cset; |
d427dfeb | 1553 | int i, ret; |
2c6ab6d2 | 1554 | |
d427dfeb | 1555 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1556 | |
6fa4918d | 1557 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_NOWAIT); |
d427dfeb | 1558 | if (ret < 0) |
2bd59d48 | 1559 | goto out; |
d427dfeb | 1560 | root_cgrp->id = ret; |
c6d57f33 | 1561 | |
9d755d33 TH |
1562 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release); |
1563 | if (ret) | |
1564 | goto out; | |
1565 | ||
d427dfeb | 1566 | /* |
96d365e0 | 1567 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1568 | * but that's OK - it can only be increased by someone holding |
1569 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1570 | * have some link structures left over | |
1571 | */ | |
1572 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1573 | if (ret) | |
9d755d33 | 1574 | goto cancel_ref; |
ddbcc7e8 | 1575 | |
985ed670 | 1576 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1577 | if (ret) |
9d755d33 | 1578 | goto cancel_ref; |
ddbcc7e8 | 1579 | |
2bd59d48 TH |
1580 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1581 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1582 | root_cgrp); | |
1583 | if (IS_ERR(root->kf_root)) { | |
1584 | ret = PTR_ERR(root->kf_root); | |
1585 | goto exit_root_id; | |
1586 | } | |
1587 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1588 | |
d427dfeb TH |
1589 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1590 | if (ret) | |
2bd59d48 | 1591 | goto destroy_root; |
ddbcc7e8 | 1592 | |
5df36032 | 1593 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1594 | if (ret) |
2bd59d48 | 1595 | goto destroy_root; |
ddbcc7e8 | 1596 | |
d427dfeb TH |
1597 | /* |
1598 | * There must be no failure case after here, since rebinding takes | |
1599 | * care of subsystems' refcounts, which are explicitly dropped in | |
1600 | * the failure exit path. | |
1601 | */ | |
1602 | list_add(&root->root_list, &cgroup_roots); | |
1603 | cgroup_root_count++; | |
0df6a63f | 1604 | |
d427dfeb | 1605 | /* |
3dd06ffa | 1606 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1607 | * objects. |
1608 | */ | |
96d365e0 | 1609 | down_write(&css_set_rwsem); |
d427dfeb TH |
1610 | hash_for_each(css_set_table, i, cset, hlist) |
1611 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1612 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1613 | |
d5c419b6 | 1614 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1615 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1616 | |
2bd59d48 | 1617 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1618 | ret = 0; |
2bd59d48 | 1619 | goto out; |
d427dfeb | 1620 | |
2bd59d48 TH |
1621 | destroy_root: |
1622 | kernfs_destroy_root(root->kf_root); | |
1623 | root->kf_root = NULL; | |
1624 | exit_root_id: | |
d427dfeb | 1625 | cgroup_exit_root_id(root); |
9d755d33 TH |
1626 | cancel_ref: |
1627 | percpu_ref_cancel_init(&root_cgrp->self.refcnt); | |
2bd59d48 | 1628 | out: |
d427dfeb TH |
1629 | free_cgrp_cset_links(&tmp_links); |
1630 | return ret; | |
ddbcc7e8 PM |
1631 | } |
1632 | ||
f7e83571 | 1633 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1634 | int flags, const char *unused_dev_name, |
f7e83571 | 1635 | void *data) |
ddbcc7e8 | 1636 | { |
3dd06ffa | 1637 | struct cgroup_root *root; |
ddbcc7e8 | 1638 | struct cgroup_sb_opts opts; |
2bd59d48 | 1639 | struct dentry *dentry; |
8e30e2b8 | 1640 | int ret; |
c6b3d5bc | 1641 | bool new_sb; |
ddbcc7e8 | 1642 | |
56fde9e0 TH |
1643 | /* |
1644 | * The first time anyone tries to mount a cgroup, enable the list | |
1645 | * linking each css_set to its tasks and fix up all existing tasks. | |
1646 | */ | |
1647 | if (!use_task_css_set_links) | |
1648 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1649 | |
aae8aab4 | 1650 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1651 | |
1652 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1653 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1654 | if (ret) |
8e30e2b8 | 1655 | goto out_unlock; |
a015edd2 | 1656 | |
2bd59d48 | 1657 | /* look for a matching existing root */ |
a2dd4247 TH |
1658 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1659 | cgrp_dfl_root_visible = true; | |
1660 | root = &cgrp_dfl_root; | |
1661 | cgroup_get(&root->cgrp); | |
1662 | ret = 0; | |
1663 | goto out_unlock; | |
ddbcc7e8 PM |
1664 | } |
1665 | ||
985ed670 | 1666 | for_each_root(root) { |
2bd59d48 | 1667 | bool name_match = false; |
3126121f | 1668 | |
3dd06ffa | 1669 | if (root == &cgrp_dfl_root) |
985ed670 | 1670 | continue; |
3126121f | 1671 | |
cf5d5941 | 1672 | /* |
2bd59d48 TH |
1673 | * If we asked for a name then it must match. Also, if |
1674 | * name matches but sybsys_mask doesn't, we should fail. | |
1675 | * Remember whether name matched. | |
cf5d5941 | 1676 | */ |
2bd59d48 TH |
1677 | if (opts.name) { |
1678 | if (strcmp(opts.name, root->name)) | |
1679 | continue; | |
1680 | name_match = true; | |
1681 | } | |
ddbcc7e8 | 1682 | |
c6d57f33 | 1683 | /* |
2bd59d48 TH |
1684 | * If we asked for subsystems (or explicitly for no |
1685 | * subsystems) then they must match. | |
c6d57f33 | 1686 | */ |
2bd59d48 | 1687 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1688 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1689 | if (!name_match) |
1690 | continue; | |
1691 | ret = -EBUSY; | |
1692 | goto out_unlock; | |
1693 | } | |
873fe09e | 1694 | |
c7ba8287 | 1695 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb | 1696 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1697 | pr_err("sane_behavior: new mount options should match the existing superblock\n"); |
2a0ff3fb | 1698 | ret = -EINVAL; |
8e30e2b8 | 1699 | goto out_unlock; |
2a0ff3fb | 1700 | } else { |
ed3d261b | 1701 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); |
2a0ff3fb | 1702 | } |
873fe09e | 1703 | } |
ddbcc7e8 | 1704 | |
776f02fa | 1705 | /* |
9d755d33 TH |
1706 | * A root's lifetime is governed by its root cgroup. |
1707 | * tryget_live failure indicate that the root is being | |
1708 | * destroyed. Wait for destruction to complete so that the | |
1709 | * subsystems are free. We can use wait_queue for the wait | |
1710 | * but this path is super cold. Let's just sleep for a bit | |
1711 | * and retry. | |
776f02fa | 1712 | */ |
9d755d33 | 1713 | if (!percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { |
776f02fa | 1714 | mutex_unlock(&cgroup_mutex); |
776f02fa | 1715 | msleep(10); |
a015edd2 TH |
1716 | ret = restart_syscall(); |
1717 | goto out_free; | |
776f02fa | 1718 | } |
ddbcc7e8 | 1719 | |
776f02fa | 1720 | ret = 0; |
2bd59d48 | 1721 | goto out_unlock; |
ddbcc7e8 | 1722 | } |
ddbcc7e8 | 1723 | |
817929ec | 1724 | /* |
172a2c06 TH |
1725 | * No such thing, create a new one. name= matching without subsys |
1726 | * specification is allowed for already existing hierarchies but we | |
1727 | * can't create new one without subsys specification. | |
817929ec | 1728 | */ |
172a2c06 TH |
1729 | if (!opts.subsys_mask && !opts.none) { |
1730 | ret = -EINVAL; | |
1731 | goto out_unlock; | |
817929ec | 1732 | } |
817929ec | 1733 | |
172a2c06 TH |
1734 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1735 | if (!root) { | |
1736 | ret = -ENOMEM; | |
2bd59d48 | 1737 | goto out_unlock; |
839ec545 | 1738 | } |
e5f6a860 | 1739 | |
172a2c06 TH |
1740 | init_cgroup_root(root, &opts); |
1741 | ||
35585573 | 1742 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1743 | if (ret) |
1744 | cgroup_free_root(root); | |
fa3ca07e | 1745 | |
8e30e2b8 | 1746 | out_unlock: |
ddbcc7e8 | 1747 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 1748 | out_free: |
c6d57f33 PM |
1749 | kfree(opts.release_agent); |
1750 | kfree(opts.name); | |
03b1cde6 | 1751 | |
2bd59d48 | 1752 | if (ret) |
8e30e2b8 | 1753 | return ERR_PTR(ret); |
2bd59d48 | 1754 | |
c6b3d5bc LZ |
1755 | dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); |
1756 | if (IS_ERR(dentry) || !new_sb) | |
3dd06ffa | 1757 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1758 | return dentry; |
1759 | } | |
1760 | ||
1761 | static void cgroup_kill_sb(struct super_block *sb) | |
1762 | { | |
1763 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1764 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1765 | |
9d755d33 TH |
1766 | /* |
1767 | * If @root doesn't have any mounts or children, start killing it. | |
1768 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
1769 | * cgroup_mount() may wait for @root's release. | |
1770 | */ | |
f3d46500 | 1771 | if (css_has_online_children(&root->cgrp.self)) |
9d755d33 TH |
1772 | cgroup_put(&root->cgrp); |
1773 | else | |
1774 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
1775 | ||
2bd59d48 | 1776 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1777 | } |
1778 | ||
1779 | static struct file_system_type cgroup_fs_type = { | |
1780 | .name = "cgroup", | |
f7e83571 | 1781 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1782 | .kill_sb = cgroup_kill_sb, |
1783 | }; | |
1784 | ||
676db4af GKH |
1785 | static struct kobject *cgroup_kobj; |
1786 | ||
857a2beb | 1787 | /** |
913ffdb5 | 1788 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1789 | * @task: target task |
857a2beb TH |
1790 | * @buf: the buffer to write the path into |
1791 | * @buflen: the length of the buffer | |
1792 | * | |
913ffdb5 TH |
1793 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1794 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1795 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1796 | * cgroup controller callbacks. | |
1797 | * | |
e61734c5 | 1798 | * Return value is the same as kernfs_path(). |
857a2beb | 1799 | */ |
e61734c5 | 1800 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1801 | { |
3dd06ffa | 1802 | struct cgroup_root *root; |
913ffdb5 | 1803 | struct cgroup *cgrp; |
e61734c5 TH |
1804 | int hierarchy_id = 1; |
1805 | char *path = NULL; | |
857a2beb TH |
1806 | |
1807 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1808 | down_read(&css_set_rwsem); |
857a2beb | 1809 | |
913ffdb5 TH |
1810 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1811 | ||
857a2beb TH |
1812 | if (root) { |
1813 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1814 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1815 | } else { |
1816 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1817 | if (strlcpy(buf, "/", buflen) < buflen) |
1818 | path = buf; | |
857a2beb TH |
1819 | } |
1820 | ||
96d365e0 | 1821 | up_read(&css_set_rwsem); |
857a2beb | 1822 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1823 | return path; |
857a2beb | 1824 | } |
913ffdb5 | 1825 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1826 | |
b3dc094e | 1827 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1828 | struct cgroup_taskset { |
b3dc094e TH |
1829 | /* the src and dst cset list running through cset->mg_node */ |
1830 | struct list_head src_csets; | |
1831 | struct list_head dst_csets; | |
1832 | ||
1833 | /* | |
1834 | * Fields for cgroup_taskset_*() iteration. | |
1835 | * | |
1836 | * Before migration is committed, the target migration tasks are on | |
1837 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1838 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1839 | * or ->dst_csets depending on whether migration is committed. | |
1840 | * | |
1841 | * ->cur_csets and ->cur_task point to the current task position | |
1842 | * during iteration. | |
1843 | */ | |
1844 | struct list_head *csets; | |
1845 | struct css_set *cur_cset; | |
1846 | struct task_struct *cur_task; | |
2f7ee569 TH |
1847 | }; |
1848 | ||
1849 | /** | |
1850 | * cgroup_taskset_first - reset taskset and return the first task | |
1851 | * @tset: taskset of interest | |
1852 | * | |
1853 | * @tset iteration is initialized and the first task is returned. | |
1854 | */ | |
1855 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1856 | { | |
b3dc094e TH |
1857 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1858 | tset->cur_task = NULL; | |
1859 | ||
1860 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1861 | } |
2f7ee569 TH |
1862 | |
1863 | /** | |
1864 | * cgroup_taskset_next - iterate to the next task in taskset | |
1865 | * @tset: taskset of interest | |
1866 | * | |
1867 | * Return the next task in @tset. Iteration must have been initialized | |
1868 | * with cgroup_taskset_first(). | |
1869 | */ | |
1870 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1871 | { | |
b3dc094e TH |
1872 | struct css_set *cset = tset->cur_cset; |
1873 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1874 | |
b3dc094e TH |
1875 | while (&cset->mg_node != tset->csets) { |
1876 | if (!task) | |
1877 | task = list_first_entry(&cset->mg_tasks, | |
1878 | struct task_struct, cg_list); | |
1879 | else | |
1880 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1881 | |
b3dc094e TH |
1882 | if (&task->cg_list != &cset->mg_tasks) { |
1883 | tset->cur_cset = cset; | |
1884 | tset->cur_task = task; | |
1885 | return task; | |
1886 | } | |
2f7ee569 | 1887 | |
b3dc094e TH |
1888 | cset = list_next_entry(cset, mg_node); |
1889 | task = NULL; | |
1890 | } | |
2f7ee569 | 1891 | |
b3dc094e | 1892 | return NULL; |
2f7ee569 | 1893 | } |
2f7ee569 | 1894 | |
cb0f1fe9 | 1895 | /** |
74a1166d | 1896 | * cgroup_task_migrate - move a task from one cgroup to another. |
60106946 | 1897 | * @old_cgrp: the cgroup @tsk is being migrated from |
cb0f1fe9 TH |
1898 | * @tsk: the task being migrated |
1899 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1900 | * |
cb0f1fe9 | 1901 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1902 | */ |
5abb8855 TH |
1903 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1904 | struct task_struct *tsk, | |
1905 | struct css_set *new_cset) | |
74a1166d | 1906 | { |
5abb8855 | 1907 | struct css_set *old_cset; |
74a1166d | 1908 | |
cb0f1fe9 TH |
1909 | lockdep_assert_held(&cgroup_mutex); |
1910 | lockdep_assert_held(&css_set_rwsem); | |
1911 | ||
74a1166d | 1912 | /* |
026085ef MSB |
1913 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1914 | * setting such that we can't race against cgroup_exit() changing the | |
1915 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1916 | */ |
c84cdf75 | 1917 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1918 | old_cset = task_css_set(tsk); |
74a1166d | 1919 | |
b3dc094e | 1920 | get_css_set(new_cset); |
5abb8855 | 1921 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1922 | |
1b9aba49 TH |
1923 | /* |
1924 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1925 | * leader after migration. This works because cgroup_migrate() | |
1926 | * ensures that the dst_cset of the leader is the first on the | |
1927 | * tset's dst_csets list. | |
1928 | */ | |
1929 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1930 | |
1931 | /* | |
5abb8855 TH |
1932 | * We just gained a reference on old_cset by taking it from the |
1933 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1934 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1935 | */ |
5abb8855 | 1936 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1937 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1938 | } |
1939 | ||
a043e3b2 | 1940 | /** |
1958d2d5 TH |
1941 | * cgroup_migrate_finish - cleanup after attach |
1942 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1943 | * |
1958d2d5 TH |
1944 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1945 | * those functions for details. | |
74a1166d | 1946 | */ |
1958d2d5 | 1947 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1948 | { |
1958d2d5 | 1949 | struct css_set *cset, *tmp_cset; |
74a1166d | 1950 | |
1958d2d5 TH |
1951 | lockdep_assert_held(&cgroup_mutex); |
1952 | ||
1953 | down_write(&css_set_rwsem); | |
1954 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1955 | cset->mg_src_cgrp = NULL; | |
1956 | cset->mg_dst_cset = NULL; | |
1957 | list_del_init(&cset->mg_preload_node); | |
1958 | put_css_set_locked(cset, false); | |
1959 | } | |
1960 | up_write(&css_set_rwsem); | |
1961 | } | |
1962 | ||
1963 | /** | |
1964 | * cgroup_migrate_add_src - add a migration source css_set | |
1965 | * @src_cset: the source css_set to add | |
1966 | * @dst_cgrp: the destination cgroup | |
1967 | * @preloaded_csets: list of preloaded css_sets | |
1968 | * | |
1969 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1970 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1971 | * up by cgroup_migrate_finish(). | |
1972 | * | |
1973 | * This function may be called without holding threadgroup_lock even if the | |
1974 | * target is a process. Threads may be created and destroyed but as long | |
1975 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1976 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1977 | */ | |
1978 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1979 | struct cgroup *dst_cgrp, | |
1980 | struct list_head *preloaded_csets) | |
1981 | { | |
1982 | struct cgroup *src_cgrp; | |
1983 | ||
1984 | lockdep_assert_held(&cgroup_mutex); | |
1985 | lockdep_assert_held(&css_set_rwsem); | |
1986 | ||
1987 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1988 | ||
1958d2d5 TH |
1989 | if (!list_empty(&src_cset->mg_preload_node)) |
1990 | return; | |
1991 | ||
1992 | WARN_ON(src_cset->mg_src_cgrp); | |
1993 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1994 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1995 | ||
1996 | src_cset->mg_src_cgrp = src_cgrp; | |
1997 | get_css_set(src_cset); | |
1998 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1999 | } | |
2000 | ||
2001 | /** | |
2002 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2003 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2004 | * @preloaded_csets: list of preloaded source css_sets |
2005 | * | |
2006 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2007 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2008 | * pins all destination css_sets, links each to its source, and append them |
2009 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2010 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2011 | * |
2012 | * This function must be called after cgroup_migrate_add_src() has been | |
2013 | * called on each migration source css_set. After migration is performed | |
2014 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2015 | * @preloaded_csets. | |
2016 | */ | |
2017 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2018 | struct list_head *preloaded_csets) | |
2019 | { | |
2020 | LIST_HEAD(csets); | |
f817de98 | 2021 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2022 | |
2023 | lockdep_assert_held(&cgroup_mutex); | |
2024 | ||
f8f22e53 TH |
2025 | /* |
2026 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2027 | * with tasks so that child cgroups don't compete against tasks. | |
2028 | */ | |
d51f39b0 | 2029 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2030 | dst_cgrp->child_subsys_mask) |
2031 | return -EBUSY; | |
2032 | ||
1958d2d5 | 2033 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2034 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2035 | struct css_set *dst_cset; |
2036 | ||
f817de98 TH |
2037 | dst_cset = find_css_set(src_cset, |
2038 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2039 | if (!dst_cset) |
2040 | goto err; | |
2041 | ||
2042 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2043 | |
2044 | /* | |
2045 | * If src cset equals dst, it's noop. Drop the src. | |
2046 | * cgroup_migrate() will skip the cset too. Note that we | |
2047 | * can't handle src == dst as some nodes are used by both. | |
2048 | */ | |
2049 | if (src_cset == dst_cset) { | |
2050 | src_cset->mg_src_cgrp = NULL; | |
2051 | list_del_init(&src_cset->mg_preload_node); | |
2052 | put_css_set(src_cset, false); | |
2053 | put_css_set(dst_cset, false); | |
2054 | continue; | |
2055 | } | |
2056 | ||
1958d2d5 TH |
2057 | src_cset->mg_dst_cset = dst_cset; |
2058 | ||
2059 | if (list_empty(&dst_cset->mg_preload_node)) | |
2060 | list_add(&dst_cset->mg_preload_node, &csets); | |
2061 | else | |
2062 | put_css_set(dst_cset, false); | |
2063 | } | |
2064 | ||
f817de98 | 2065 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2066 | return 0; |
2067 | err: | |
2068 | cgroup_migrate_finish(&csets); | |
2069 | return -ENOMEM; | |
2070 | } | |
2071 | ||
2072 | /** | |
2073 | * cgroup_migrate - migrate a process or task to a cgroup | |
2074 | * @cgrp: the destination cgroup | |
2075 | * @leader: the leader of the process or the task to migrate | |
2076 | * @threadgroup: whether @leader points to the whole process or a single task | |
2077 | * | |
2078 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2079 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2080 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2081 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2082 | * function and following up with cgroup_migrate_finish(). | |
2083 | * | |
2084 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2085 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2086 | * failure, when migrating multiple targets, the success or failure can be | |
2087 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2088 | * actually starting migrating. | |
2089 | */ | |
2090 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2091 | bool threadgroup) | |
74a1166d | 2092 | { |
b3dc094e TH |
2093 | struct cgroup_taskset tset = { |
2094 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2095 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2096 | .csets = &tset.src_csets, | |
2097 | }; | |
1c6727af | 2098 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2099 | struct css_set *cset, *tmp_cset; |
2100 | struct task_struct *task, *tmp_task; | |
2101 | int i, ret; | |
74a1166d | 2102 | |
fb5d2b4c MSB |
2103 | /* |
2104 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2105 | * already PF_EXITING could be freed from underneath us unless we | |
2106 | * take an rcu_read_lock. | |
2107 | */ | |
b3dc094e | 2108 | down_write(&css_set_rwsem); |
fb5d2b4c | 2109 | rcu_read_lock(); |
9db8de37 | 2110 | task = leader; |
74a1166d | 2111 | do { |
9db8de37 TH |
2112 | /* @task either already exited or can't exit until the end */ |
2113 | if (task->flags & PF_EXITING) | |
ea84753c | 2114 | goto next; |
134d3373 | 2115 | |
eaf797ab TH |
2116 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2117 | if (list_empty(&task->cg_list)) | |
ea84753c | 2118 | goto next; |
cd3d0952 | 2119 | |
b3dc094e | 2120 | cset = task_css_set(task); |
1958d2d5 | 2121 | if (!cset->mg_src_cgrp) |
ea84753c | 2122 | goto next; |
b3dc094e | 2123 | |
61d1d219 | 2124 | /* |
1b9aba49 TH |
2125 | * cgroup_taskset_first() must always return the leader. |
2126 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2127 | */ |
1b9aba49 TH |
2128 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2129 | if (list_empty(&cset->mg_node)) | |
2130 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2131 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2132 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2133 | &tset.dst_csets); | |
ea84753c | 2134 | next: |
081aa458 LZ |
2135 | if (!threadgroup) |
2136 | break; | |
9db8de37 | 2137 | } while_each_thread(leader, task); |
fb5d2b4c | 2138 | rcu_read_unlock(); |
b3dc094e | 2139 | up_write(&css_set_rwsem); |
74a1166d | 2140 | |
134d3373 | 2141 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2142 | if (list_empty(&tset.src_csets)) |
2143 | return 0; | |
134d3373 | 2144 | |
1958d2d5 | 2145 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2146 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2147 | if (css->ss->can_attach) { |
9db8de37 TH |
2148 | ret = css->ss->can_attach(css, &tset); |
2149 | if (ret) { | |
1c6727af | 2150 | failed_css = css; |
74a1166d BB |
2151 | goto out_cancel_attach; |
2152 | } | |
2153 | } | |
74a1166d BB |
2154 | } |
2155 | ||
2156 | /* | |
1958d2d5 TH |
2157 | * Now that we're guaranteed success, proceed to move all tasks to |
2158 | * the new cgroup. There are no failure cases after here, so this | |
2159 | * is the commit point. | |
74a1166d | 2160 | */ |
cb0f1fe9 | 2161 | down_write(&css_set_rwsem); |
b3dc094e TH |
2162 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2163 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2164 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2165 | cset->mg_dst_cset); | |
74a1166d | 2166 | } |
cb0f1fe9 | 2167 | up_write(&css_set_rwsem); |
74a1166d BB |
2168 | |
2169 | /* | |
1958d2d5 TH |
2170 | * Migration is committed, all target tasks are now on dst_csets. |
2171 | * Nothing is sensitive to fork() after this point. Notify | |
2172 | * controllers that migration is complete. | |
74a1166d | 2173 | */ |
1958d2d5 | 2174 | tset.csets = &tset.dst_csets; |
74a1166d | 2175 | |
aec3dfcb | 2176 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2177 | if (css->ss->attach) |
2178 | css->ss->attach(css, &tset); | |
74a1166d | 2179 | |
9db8de37 | 2180 | ret = 0; |
b3dc094e TH |
2181 | goto out_release_tset; |
2182 | ||
74a1166d | 2183 | out_cancel_attach: |
aec3dfcb | 2184 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2185 | if (css == failed_css) |
2186 | break; | |
2187 | if (css->ss->cancel_attach) | |
2188 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2189 | } |
b3dc094e TH |
2190 | out_release_tset: |
2191 | down_write(&css_set_rwsem); | |
2192 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2193 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2194 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2195 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2196 | } |
2197 | up_write(&css_set_rwsem); | |
9db8de37 | 2198 | return ret; |
74a1166d BB |
2199 | } |
2200 | ||
1958d2d5 TH |
2201 | /** |
2202 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2203 | * @dst_cgrp: the cgroup to attach to | |
2204 | * @leader: the task or the leader of the threadgroup to be attached | |
2205 | * @threadgroup: attach the whole threadgroup? | |
2206 | * | |
0e1d768f | 2207 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2208 | */ |
2209 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2210 | struct task_struct *leader, bool threadgroup) | |
2211 | { | |
2212 | LIST_HEAD(preloaded_csets); | |
2213 | struct task_struct *task; | |
2214 | int ret; | |
2215 | ||
2216 | /* look up all src csets */ | |
2217 | down_read(&css_set_rwsem); | |
2218 | rcu_read_lock(); | |
2219 | task = leader; | |
2220 | do { | |
2221 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2222 | &preloaded_csets); | |
2223 | if (!threadgroup) | |
2224 | break; | |
2225 | } while_each_thread(leader, task); | |
2226 | rcu_read_unlock(); | |
2227 | up_read(&css_set_rwsem); | |
2228 | ||
2229 | /* prepare dst csets and commit */ | |
2230 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2231 | if (!ret) | |
2232 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2233 | ||
2234 | cgroup_migrate_finish(&preloaded_csets); | |
2235 | return ret; | |
74a1166d BB |
2236 | } |
2237 | ||
2238 | /* | |
2239 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2240 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2241 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2242 | */ |
acbef755 TH |
2243 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2244 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2245 | { |
bbcb81d0 | 2246 | struct task_struct *tsk; |
c69e8d9c | 2247 | const struct cred *cred = current_cred(), *tcred; |
e76ecaee | 2248 | struct cgroup *cgrp; |
acbef755 | 2249 | pid_t pid; |
bbcb81d0 PM |
2250 | int ret; |
2251 | ||
acbef755 TH |
2252 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2253 | return -EINVAL; | |
2254 | ||
e76ecaee TH |
2255 | cgrp = cgroup_kn_lock_live(of->kn); |
2256 | if (!cgrp) | |
74a1166d BB |
2257 | return -ENODEV; |
2258 | ||
b78949eb MSB |
2259 | retry_find_task: |
2260 | rcu_read_lock(); | |
bbcb81d0 | 2261 | if (pid) { |
73507f33 | 2262 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2263 | if (!tsk) { |
2264 | rcu_read_unlock(); | |
dd4b0a46 | 2265 | ret = -ESRCH; |
b78949eb | 2266 | goto out_unlock_cgroup; |
bbcb81d0 | 2267 | } |
74a1166d BB |
2268 | /* |
2269 | * even if we're attaching all tasks in the thread group, we | |
2270 | * only need to check permissions on one of them. | |
2271 | */ | |
c69e8d9c | 2272 | tcred = __task_cred(tsk); |
14a590c3 EB |
2273 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2274 | !uid_eq(cred->euid, tcred->uid) && | |
2275 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2276 | rcu_read_unlock(); |
b78949eb MSB |
2277 | ret = -EACCES; |
2278 | goto out_unlock_cgroup; | |
bbcb81d0 | 2279 | } |
b78949eb MSB |
2280 | } else |
2281 | tsk = current; | |
cd3d0952 TH |
2282 | |
2283 | if (threadgroup) | |
b78949eb | 2284 | tsk = tsk->group_leader; |
c4c27fbd MG |
2285 | |
2286 | /* | |
14a40ffc | 2287 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2288 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2289 | * with no rt_runtime allocated. Just say no. | |
2290 | */ | |
14a40ffc | 2291 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2292 | ret = -EINVAL; |
2293 | rcu_read_unlock(); | |
2294 | goto out_unlock_cgroup; | |
2295 | } | |
2296 | ||
b78949eb MSB |
2297 | get_task_struct(tsk); |
2298 | rcu_read_unlock(); | |
2299 | ||
2300 | threadgroup_lock(tsk); | |
2301 | if (threadgroup) { | |
2302 | if (!thread_group_leader(tsk)) { | |
2303 | /* | |
2304 | * a race with de_thread from another thread's exec() | |
2305 | * may strip us of our leadership, if this happens, | |
2306 | * there is no choice but to throw this task away and | |
2307 | * try again; this is | |
2308 | * "double-double-toil-and-trouble-check locking". | |
2309 | */ | |
2310 | threadgroup_unlock(tsk); | |
2311 | put_task_struct(tsk); | |
2312 | goto retry_find_task; | |
2313 | } | |
081aa458 LZ |
2314 | } |
2315 | ||
2316 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2317 | ||
cd3d0952 TH |
2318 | threadgroup_unlock(tsk); |
2319 | ||
bbcb81d0 | 2320 | put_task_struct(tsk); |
b78949eb | 2321 | out_unlock_cgroup: |
e76ecaee | 2322 | cgroup_kn_unlock(of->kn); |
acbef755 | 2323 | return ret ?: nbytes; |
bbcb81d0 PM |
2324 | } |
2325 | ||
7ae1bad9 TH |
2326 | /** |
2327 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2328 | * @from: attach to all cgroups of a given task | |
2329 | * @tsk: the task to be attached | |
2330 | */ | |
2331 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2332 | { | |
3dd06ffa | 2333 | struct cgroup_root *root; |
7ae1bad9 TH |
2334 | int retval = 0; |
2335 | ||
47cfcd09 | 2336 | mutex_lock(&cgroup_mutex); |
985ed670 | 2337 | for_each_root(root) { |
96d365e0 TH |
2338 | struct cgroup *from_cgrp; |
2339 | ||
3dd06ffa | 2340 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2341 | continue; |
2342 | ||
96d365e0 TH |
2343 | down_read(&css_set_rwsem); |
2344 | from_cgrp = task_cgroup_from_root(from, root); | |
2345 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2346 | |
6f4b7e63 | 2347 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2348 | if (retval) |
2349 | break; | |
2350 | } | |
47cfcd09 | 2351 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2352 | |
2353 | return retval; | |
2354 | } | |
2355 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2356 | ||
acbef755 TH |
2357 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2358 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2359 | { |
acbef755 | 2360 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2361 | } |
2362 | ||
acbef755 TH |
2363 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2364 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2365 | { |
acbef755 | 2366 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2367 | } |
2368 | ||
451af504 TH |
2369 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2370 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2371 | { |
e76ecaee | 2372 | struct cgroup *cgrp; |
5f469907 | 2373 | |
e76ecaee TH |
2374 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
2375 | ||
2376 | cgrp = cgroup_kn_lock_live(of->kn); | |
2377 | if (!cgrp) | |
e788e066 | 2378 | return -ENODEV; |
69e943b7 | 2379 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2380 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2381 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2382 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2383 | cgroup_kn_unlock(of->kn); |
451af504 | 2384 | return nbytes; |
e788e066 PM |
2385 | } |
2386 | ||
2da8ca82 | 2387 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2388 | { |
2da8ca82 | 2389 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2390 | |
46cfeb04 | 2391 | spin_lock(&release_agent_path_lock); |
e788e066 | 2392 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2393 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2394 | seq_putc(seq, '\n'); |
e788e066 PM |
2395 | return 0; |
2396 | } | |
2397 | ||
2da8ca82 | 2398 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2399 | { |
2da8ca82 TH |
2400 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2401 | ||
2402 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2403 | return 0; |
2404 | } | |
2405 | ||
f8f22e53 TH |
2406 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
2407 | { | |
2408 | struct cgroup_subsys *ss; | |
2409 | bool printed = false; | |
2410 | int ssid; | |
2411 | ||
2412 | for_each_subsys(ss, ssid) { | |
2413 | if (ss_mask & (1 << ssid)) { | |
2414 | if (printed) | |
2415 | seq_putc(seq, ' '); | |
2416 | seq_printf(seq, "%s", ss->name); | |
2417 | printed = true; | |
2418 | } | |
2419 | } | |
2420 | if (printed) | |
2421 | seq_putc(seq, '\n'); | |
2422 | } | |
2423 | ||
2424 | /* show controllers which are currently attached to the default hierarchy */ | |
2425 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
2426 | { | |
2427 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2428 | ||
2429 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask); | |
2430 | return 0; | |
2431 | } | |
2432 | ||
2433 | /* show controllers which are enabled from the parent */ | |
2434 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
2435 | { | |
2436 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2437 | ||
d51f39b0 | 2438 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->child_subsys_mask); |
f8f22e53 TH |
2439 | return 0; |
2440 | } | |
2441 | ||
2442 | /* show controllers which are enabled for a given cgroup's children */ | |
2443 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
2444 | { | |
2445 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2446 | ||
2447 | cgroup_print_ss_mask(seq, cgrp->child_subsys_mask); | |
2448 | return 0; | |
2449 | } | |
2450 | ||
2451 | /** | |
2452 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2453 | * @cgrp: root of the subtree to update csses for | |
2454 | * | |
2455 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2456 | * css associations need to be updated accordingly. This function looks up | |
2457 | * all css_sets which are attached to the subtree, creates the matching | |
2458 | * updated css_sets and migrates the tasks to the new ones. | |
2459 | */ | |
2460 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2461 | { | |
2462 | LIST_HEAD(preloaded_csets); | |
2463 | struct cgroup_subsys_state *css; | |
2464 | struct css_set *src_cset; | |
2465 | int ret; | |
2466 | ||
f8f22e53 TH |
2467 | lockdep_assert_held(&cgroup_mutex); |
2468 | ||
2469 | /* look up all csses currently attached to @cgrp's subtree */ | |
2470 | down_read(&css_set_rwsem); | |
2471 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2472 | struct cgrp_cset_link *link; | |
2473 | ||
2474 | /* self is not affected by child_subsys_mask change */ | |
2475 | if (css->cgroup == cgrp) | |
2476 | continue; | |
2477 | ||
2478 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2479 | cgroup_migrate_add_src(link->cset, cgrp, | |
2480 | &preloaded_csets); | |
2481 | } | |
2482 | up_read(&css_set_rwsem); | |
2483 | ||
2484 | /* NULL dst indicates self on default hierarchy */ | |
2485 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2486 | if (ret) | |
2487 | goto out_finish; | |
2488 | ||
2489 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2490 | struct task_struct *last_task = NULL, *task; | |
2491 | ||
2492 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2493 | if (!src_cset->mg_src_cgrp) | |
2494 | break; | |
2495 | ||
2496 | /* | |
2497 | * All tasks in src_cset need to be migrated to the | |
2498 | * matching dst_cset. Empty it process by process. We | |
2499 | * walk tasks but migrate processes. The leader might even | |
2500 | * belong to a different cset but such src_cset would also | |
2501 | * be among the target src_csets because the default | |
2502 | * hierarchy enforces per-process membership. | |
2503 | */ | |
2504 | while (true) { | |
2505 | down_read(&css_set_rwsem); | |
2506 | task = list_first_entry_or_null(&src_cset->tasks, | |
2507 | struct task_struct, cg_list); | |
2508 | if (task) { | |
2509 | task = task->group_leader; | |
2510 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2511 | get_task_struct(task); | |
2512 | } | |
2513 | up_read(&css_set_rwsem); | |
2514 | ||
2515 | if (!task) | |
2516 | break; | |
2517 | ||
2518 | /* guard against possible infinite loop */ | |
2519 | if (WARN(last_task == task, | |
2520 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2521 | goto out_finish; | |
2522 | last_task = task; | |
2523 | ||
2524 | threadgroup_lock(task); | |
2525 | /* raced against de_thread() from another thread? */ | |
2526 | if (!thread_group_leader(task)) { | |
2527 | threadgroup_unlock(task); | |
2528 | put_task_struct(task); | |
2529 | continue; | |
2530 | } | |
2531 | ||
2532 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2533 | ||
2534 | threadgroup_unlock(task); | |
2535 | put_task_struct(task); | |
2536 | ||
2537 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2538 | goto out_finish; | |
2539 | } | |
2540 | } | |
2541 | ||
2542 | out_finish: | |
2543 | cgroup_migrate_finish(&preloaded_csets); | |
2544 | return ret; | |
2545 | } | |
2546 | ||
2547 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2548 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2549 | char *buf, size_t nbytes, | |
2550 | loff_t off) | |
f8f22e53 | 2551 | { |
7d331fa9 | 2552 | unsigned int enable = 0, disable = 0; |
a9746d8d | 2553 | struct cgroup *cgrp, *child; |
f8f22e53 | 2554 | struct cgroup_subsys *ss; |
451af504 | 2555 | char *tok; |
f8f22e53 TH |
2556 | int ssid, ret; |
2557 | ||
2558 | /* | |
d37167ab TH |
2559 | * Parse input - space separated list of subsystem names prefixed |
2560 | * with either + or -. | |
f8f22e53 | 2561 | */ |
451af504 TH |
2562 | buf = strstrip(buf); |
2563 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2564 | if (tok[0] == '\0') |
2565 | continue; | |
f8f22e53 TH |
2566 | for_each_subsys(ss, ssid) { |
2567 | if (ss->disabled || strcmp(tok + 1, ss->name)) | |
2568 | continue; | |
2569 | ||
2570 | if (*tok == '+') { | |
7d331fa9 TH |
2571 | enable |= 1 << ssid; |
2572 | disable &= ~(1 << ssid); | |
f8f22e53 | 2573 | } else if (*tok == '-') { |
7d331fa9 TH |
2574 | disable |= 1 << ssid; |
2575 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2576 | } else { |
2577 | return -EINVAL; | |
2578 | } | |
2579 | break; | |
2580 | } | |
2581 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2582 | return -EINVAL; | |
2583 | } | |
2584 | ||
a9746d8d TH |
2585 | cgrp = cgroup_kn_lock_live(of->kn); |
2586 | if (!cgrp) | |
2587 | return -ENODEV; | |
f8f22e53 TH |
2588 | |
2589 | for_each_subsys(ss, ssid) { | |
2590 | if (enable & (1 << ssid)) { | |
2591 | if (cgrp->child_subsys_mask & (1 << ssid)) { | |
2592 | enable &= ~(1 << ssid); | |
2593 | continue; | |
2594 | } | |
2595 | ||
2596 | /* | |
2597 | * Because css offlining is asynchronous, userland | |
2598 | * might try to re-enable the same controller while | |
2599 | * the previous instance is still around. In such | |
2600 | * cases, wait till it's gone using offline_waitq. | |
2601 | */ | |
2602 | cgroup_for_each_live_child(child, cgrp) { | |
0cee8b77 | 2603 | DEFINE_WAIT(wait); |
f8f22e53 TH |
2604 | |
2605 | if (!cgroup_css(child, ss)) | |
2606 | continue; | |
2607 | ||
0cee8b77 | 2608 | cgroup_get(child); |
f8f22e53 TH |
2609 | prepare_to_wait(&child->offline_waitq, &wait, |
2610 | TASK_UNINTERRUPTIBLE); | |
a9746d8d | 2611 | cgroup_kn_unlock(of->kn); |
f8f22e53 TH |
2612 | schedule(); |
2613 | finish_wait(&child->offline_waitq, &wait); | |
0cee8b77 | 2614 | cgroup_put(child); |
7d331fa9 | 2615 | |
a9746d8d | 2616 | return restart_syscall(); |
f8f22e53 TH |
2617 | } |
2618 | ||
2619 | /* unavailable or not enabled on the parent? */ | |
2620 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
d51f39b0 TH |
2621 | (cgroup_parent(cgrp) && |
2622 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ssid)))) { | |
f8f22e53 | 2623 | ret = -ENOENT; |
ddab2b6e | 2624 | goto out_unlock; |
f8f22e53 TH |
2625 | } |
2626 | } else if (disable & (1 << ssid)) { | |
2627 | if (!(cgrp->child_subsys_mask & (1 << ssid))) { | |
2628 | disable &= ~(1 << ssid); | |
2629 | continue; | |
2630 | } | |
2631 | ||
2632 | /* a child has it enabled? */ | |
2633 | cgroup_for_each_live_child(child, cgrp) { | |
2634 | if (child->child_subsys_mask & (1 << ssid)) { | |
2635 | ret = -EBUSY; | |
ddab2b6e | 2636 | goto out_unlock; |
f8f22e53 TH |
2637 | } |
2638 | } | |
2639 | } | |
2640 | } | |
2641 | ||
2642 | if (!enable && !disable) { | |
2643 | ret = 0; | |
ddab2b6e | 2644 | goto out_unlock; |
f8f22e53 TH |
2645 | } |
2646 | ||
2647 | /* | |
2648 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2649 | * with tasks so that child cgroups don't compete against tasks. | |
2650 | */ | |
d51f39b0 | 2651 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
2652 | ret = -EBUSY; |
2653 | goto out_unlock; | |
2654 | } | |
2655 | ||
2656 | /* | |
2657 | * Create csses for enables and update child_subsys_mask. This | |
2658 | * changes cgroup_e_css() results which in turn makes the | |
2659 | * subsequent cgroup_update_dfl_csses() associate all tasks in the | |
2660 | * subtree to the updated csses. | |
2661 | */ | |
2662 | for_each_subsys(ss, ssid) { | |
2663 | if (!(enable & (1 << ssid))) | |
2664 | continue; | |
2665 | ||
2666 | cgroup_for_each_live_child(child, cgrp) { | |
2667 | ret = create_css(child, ss); | |
2668 | if (ret) | |
2669 | goto err_undo_css; | |
2670 | } | |
2671 | } | |
2672 | ||
2673 | cgrp->child_subsys_mask |= enable; | |
2674 | cgrp->child_subsys_mask &= ~disable; | |
2675 | ||
2676 | ret = cgroup_update_dfl_csses(cgrp); | |
2677 | if (ret) | |
2678 | goto err_undo_css; | |
2679 | ||
2680 | /* all tasks are now migrated away from the old csses, kill them */ | |
2681 | for_each_subsys(ss, ssid) { | |
2682 | if (!(disable & (1 << ssid))) | |
2683 | continue; | |
2684 | ||
2685 | cgroup_for_each_live_child(child, cgrp) | |
2686 | kill_css(cgroup_css(child, ss)); | |
2687 | } | |
2688 | ||
2689 | kernfs_activate(cgrp->kn); | |
2690 | ret = 0; | |
2691 | out_unlock: | |
a9746d8d | 2692 | cgroup_kn_unlock(of->kn); |
451af504 | 2693 | return ret ?: nbytes; |
f8f22e53 TH |
2694 | |
2695 | err_undo_css: | |
2696 | cgrp->child_subsys_mask &= ~enable; | |
2697 | cgrp->child_subsys_mask |= disable; | |
2698 | ||
2699 | for_each_subsys(ss, ssid) { | |
2700 | if (!(enable & (1 << ssid))) | |
2701 | continue; | |
2702 | ||
2703 | cgroup_for_each_live_child(child, cgrp) { | |
2704 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
2705 | if (css) | |
2706 | kill_css(css); | |
2707 | } | |
2708 | } | |
2709 | goto out_unlock; | |
2710 | } | |
2711 | ||
842b597e TH |
2712 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2713 | { | |
2714 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2715 | return 0; | |
2716 | } | |
2717 | ||
2bd59d48 TH |
2718 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2719 | size_t nbytes, loff_t off) | |
355e0c48 | 2720 | { |
2bd59d48 TH |
2721 | struct cgroup *cgrp = of->kn->parent->priv; |
2722 | struct cftype *cft = of->kn->priv; | |
2723 | struct cgroup_subsys_state *css; | |
a742c59d | 2724 | int ret; |
355e0c48 | 2725 | |
b4168640 TH |
2726 | if (cft->write) |
2727 | return cft->write(of, buf, nbytes, off); | |
2728 | ||
2bd59d48 TH |
2729 | /* |
2730 | * kernfs guarantees that a file isn't deleted with operations in | |
2731 | * flight, which means that the matching css is and stays alive and | |
2732 | * doesn't need to be pinned. The RCU locking is not necessary | |
2733 | * either. It's just for the convenience of using cgroup_css(). | |
2734 | */ | |
2735 | rcu_read_lock(); | |
2736 | css = cgroup_css(cgrp, cft->ss); | |
2737 | rcu_read_unlock(); | |
a742c59d | 2738 | |
451af504 | 2739 | if (cft->write_u64) { |
a742c59d TH |
2740 | unsigned long long v; |
2741 | ret = kstrtoull(buf, 0, &v); | |
2742 | if (!ret) | |
2743 | ret = cft->write_u64(css, cft, v); | |
2744 | } else if (cft->write_s64) { | |
2745 | long long v; | |
2746 | ret = kstrtoll(buf, 0, &v); | |
2747 | if (!ret) | |
2748 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2749 | } else { |
a742c59d | 2750 | ret = -EINVAL; |
e73d2c61 | 2751 | } |
2bd59d48 | 2752 | |
a742c59d | 2753 | return ret ?: nbytes; |
355e0c48 PM |
2754 | } |
2755 | ||
6612f05b | 2756 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2757 | { |
2bd59d48 | 2758 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2759 | } |
2760 | ||
6612f05b | 2761 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2762 | { |
2bd59d48 | 2763 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2764 | } |
2765 | ||
6612f05b | 2766 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2767 | { |
2bd59d48 | 2768 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2769 | } |
2770 | ||
91796569 | 2771 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2772 | { |
7da11279 TH |
2773 | struct cftype *cft = seq_cft(m); |
2774 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2775 | |
2da8ca82 TH |
2776 | if (cft->seq_show) |
2777 | return cft->seq_show(m, arg); | |
e73d2c61 | 2778 | |
f4c753b7 | 2779 | if (cft->read_u64) |
896f5199 TH |
2780 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2781 | else if (cft->read_s64) | |
2782 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2783 | else | |
2784 | return -EINVAL; | |
2785 | return 0; | |
91796569 PM |
2786 | } |
2787 | ||
2bd59d48 TH |
2788 | static struct kernfs_ops cgroup_kf_single_ops = { |
2789 | .atomic_write_len = PAGE_SIZE, | |
2790 | .write = cgroup_file_write, | |
2791 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2792 | }; |
2793 | ||
2bd59d48 TH |
2794 | static struct kernfs_ops cgroup_kf_ops = { |
2795 | .atomic_write_len = PAGE_SIZE, | |
2796 | .write = cgroup_file_write, | |
2797 | .seq_start = cgroup_seqfile_start, | |
2798 | .seq_next = cgroup_seqfile_next, | |
2799 | .seq_stop = cgroup_seqfile_stop, | |
2800 | .seq_show = cgroup_seqfile_show, | |
2801 | }; | |
ddbcc7e8 PM |
2802 | |
2803 | /* | |
2804 | * cgroup_rename - Only allow simple rename of directories in place. | |
2805 | */ | |
2bd59d48 TH |
2806 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2807 | const char *new_name_str) | |
ddbcc7e8 | 2808 | { |
2bd59d48 | 2809 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2810 | int ret; |
65dff759 | 2811 | |
2bd59d48 | 2812 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2813 | return -ENOTDIR; |
2bd59d48 | 2814 | if (kn->parent != new_parent) |
ddbcc7e8 | 2815 | return -EIO; |
65dff759 | 2816 | |
6db8e85c TH |
2817 | /* |
2818 | * This isn't a proper migration and its usefulness is very | |
2819 | * limited. Disallow if sane_behavior. | |
2820 | */ | |
2821 | if (cgroup_sane_behavior(cgrp)) | |
2822 | return -EPERM; | |
099fca32 | 2823 | |
e1b2dc17 | 2824 | /* |
8353da1f | 2825 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 2826 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 2827 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
2828 | */ |
2829 | kernfs_break_active_protection(new_parent); | |
2830 | kernfs_break_active_protection(kn); | |
099fca32 | 2831 | |
2bd59d48 | 2832 | mutex_lock(&cgroup_mutex); |
099fca32 | 2833 | |
2bd59d48 | 2834 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2835 | |
2bd59d48 | 2836 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
2837 | |
2838 | kernfs_unbreak_active_protection(kn); | |
2839 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2840 | return ret; |
099fca32 LZ |
2841 | } |
2842 | ||
49957f8e TH |
2843 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2844 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2845 | { | |
2846 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2847 | .ia_uid = current_fsuid(), | |
2848 | .ia_gid = current_fsgid(), }; | |
2849 | ||
2850 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2851 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2852 | return 0; | |
2853 | ||
2854 | return kernfs_setattr(kn, &iattr); | |
2855 | } | |
2856 | ||
2bb566cb | 2857 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2858 | { |
8d7e6fb0 | 2859 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2860 | struct kernfs_node *kn; |
2861 | struct lock_class_key *key = NULL; | |
49957f8e | 2862 | int ret; |
05ef1d7c | 2863 | |
2bd59d48 TH |
2864 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2865 | key = &cft->lockdep_key; | |
2866 | #endif | |
2867 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2868 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2869 | NULL, false, key); | |
49957f8e TH |
2870 | if (IS_ERR(kn)) |
2871 | return PTR_ERR(kn); | |
2872 | ||
2873 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 2874 | if (ret) { |
49957f8e | 2875 | kernfs_remove(kn); |
f8f22e53 TH |
2876 | return ret; |
2877 | } | |
2878 | ||
b7fc5ad2 | 2879 | if (cft->seq_show == cgroup_populated_show) |
842b597e | 2880 | cgrp->populated_kn = kn; |
f8f22e53 | 2881 | return 0; |
ddbcc7e8 PM |
2882 | } |
2883 | ||
b1f28d31 TH |
2884 | /** |
2885 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2886 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2887 | * @cfts: array of cftypes to be added |
2888 | * @is_add: whether to add or remove | |
2889 | * | |
2890 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2891 | * For removals, this function never fails. If addition fails, this |
2892 | * function doesn't remove files already added. The caller is responsible | |
2893 | * for cleaning up. | |
b1f28d31 | 2894 | */ |
2bb566cb TH |
2895 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2896 | bool is_add) | |
ddbcc7e8 | 2897 | { |
03b1cde6 | 2898 | struct cftype *cft; |
b1f28d31 TH |
2899 | int ret; |
2900 | ||
01f6474c | 2901 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 TH |
2902 | |
2903 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2904 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2905 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2906 | continue; | |
873fe09e TH |
2907 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2908 | continue; | |
d51f39b0 | 2909 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 2910 | continue; |
d51f39b0 | 2911 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
2912 | continue; |
2913 | ||
2739d3cc | 2914 | if (is_add) { |
2bb566cb | 2915 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2916 | if (ret) { |
ed3d261b JP |
2917 | pr_warn("%s: failed to add %s, err=%d\n", |
2918 | __func__, cft->name, ret); | |
b1f28d31 TH |
2919 | return ret; |
2920 | } | |
2739d3cc LZ |
2921 | } else { |
2922 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2923 | } |
ddbcc7e8 | 2924 | } |
b1f28d31 | 2925 | return 0; |
ddbcc7e8 PM |
2926 | } |
2927 | ||
21a2d343 | 2928 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2929 | { |
2930 | LIST_HEAD(pending); | |
2bb566cb | 2931 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2932 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2933 | struct cgroup_subsys_state *css; |
9ccece80 | 2934 | int ret = 0; |
8e3f6541 | 2935 | |
01f6474c | 2936 | lockdep_assert_held(&cgroup_mutex); |
8e3f6541 | 2937 | |
e8c82d20 | 2938 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2939 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2940 | struct cgroup *cgrp = css->cgroup; |
2941 | ||
e8c82d20 LZ |
2942 | if (cgroup_is_dead(cgrp)) |
2943 | continue; | |
2944 | ||
21a2d343 | 2945 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2946 | if (ret) |
2947 | break; | |
8e3f6541 | 2948 | } |
21a2d343 TH |
2949 | |
2950 | if (is_add && !ret) | |
2951 | kernfs_activate(root->kn); | |
9ccece80 | 2952 | return ret; |
8e3f6541 TH |
2953 | } |
2954 | ||
2da440a2 | 2955 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2956 | { |
2bb566cb | 2957 | struct cftype *cft; |
8e3f6541 | 2958 | |
2bd59d48 TH |
2959 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2960 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2961 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2962 | kfree(cft->kf_ops); | |
2963 | cft->kf_ops = NULL; | |
2da440a2 | 2964 | cft->ss = NULL; |
2bd59d48 | 2965 | } |
2da440a2 TH |
2966 | } |
2967 | ||
2bd59d48 | 2968 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2969 | { |
2970 | struct cftype *cft; | |
2971 | ||
2bd59d48 TH |
2972 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2973 | struct kernfs_ops *kf_ops; | |
2974 | ||
0adb0704 TH |
2975 | WARN_ON(cft->ss || cft->kf_ops); |
2976 | ||
2bd59d48 TH |
2977 | if (cft->seq_start) |
2978 | kf_ops = &cgroup_kf_ops; | |
2979 | else | |
2980 | kf_ops = &cgroup_kf_single_ops; | |
2981 | ||
2982 | /* | |
2983 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2984 | * make a copy of kf_ops to set its atomic_write_len. | |
2985 | */ | |
2986 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2987 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2988 | if (!kf_ops) { | |
2989 | cgroup_exit_cftypes(cfts); | |
2990 | return -ENOMEM; | |
2991 | } | |
2992 | kf_ops->atomic_write_len = cft->max_write_len; | |
2993 | } | |
8e3f6541 | 2994 | |
2bd59d48 | 2995 | cft->kf_ops = kf_ops; |
2bb566cb | 2996 | cft->ss = ss; |
2bd59d48 | 2997 | } |
2bb566cb | 2998 | |
2bd59d48 | 2999 | return 0; |
2da440a2 TH |
3000 | } |
3001 | ||
21a2d343 TH |
3002 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3003 | { | |
01f6474c | 3004 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3005 | |
3006 | if (!cfts || !cfts[0].ss) | |
3007 | return -ENOENT; | |
3008 | ||
3009 | list_del(&cfts->node); | |
3010 | cgroup_apply_cftypes(cfts, false); | |
3011 | cgroup_exit_cftypes(cfts); | |
3012 | return 0; | |
8e3f6541 | 3013 | } |
8e3f6541 | 3014 | |
79578621 TH |
3015 | /** |
3016 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3017 | * @cfts: zero-length name terminated array of cftypes |
3018 | * | |
2bb566cb TH |
3019 | * Unregister @cfts. Files described by @cfts are removed from all |
3020 | * existing cgroups and all future cgroups won't have them either. This | |
3021 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3022 | * |
3023 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3024 | * registered. |
79578621 | 3025 | */ |
2bb566cb | 3026 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3027 | { |
21a2d343 | 3028 | int ret; |
79578621 | 3029 | |
01f6474c | 3030 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3031 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3032 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3033 | return ret; |
80b13586 TH |
3034 | } |
3035 | ||
8e3f6541 TH |
3036 | /** |
3037 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3038 | * @ss: target cgroup subsystem | |
3039 | * @cfts: zero-length name terminated array of cftypes | |
3040 | * | |
3041 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3042 | * existing cgroups to which @ss is attached and all future cgroups will | |
3043 | * have them too. This function can be called anytime whether @ss is | |
3044 | * attached or not. | |
3045 | * | |
3046 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3047 | * function currently returns 0 as long as @cfts registration is successful | |
3048 | * even if some file creation attempts on existing cgroups fail. | |
3049 | */ | |
03b1cde6 | 3050 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3051 | { |
9ccece80 | 3052 | int ret; |
8e3f6541 | 3053 | |
dc5736ed LZ |
3054 | if (!cfts || cfts[0].name[0] == '\0') |
3055 | return 0; | |
2bb566cb | 3056 | |
2bd59d48 TH |
3057 | ret = cgroup_init_cftypes(ss, cfts); |
3058 | if (ret) | |
3059 | return ret; | |
79578621 | 3060 | |
01f6474c | 3061 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3062 | |
0adb0704 | 3063 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3064 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3065 | if (ret) |
21a2d343 | 3066 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3067 | |
01f6474c | 3068 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3069 | return ret; |
79578621 TH |
3070 | } |
3071 | ||
a043e3b2 LZ |
3072 | /** |
3073 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3074 | * @cgrp: the cgroup in question | |
3075 | * | |
3076 | * Return the number of tasks in the cgroup. | |
3077 | */ | |
07bc356e | 3078 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3079 | { |
3080 | int count = 0; | |
69d0206c | 3081 | struct cgrp_cset_link *link; |
817929ec | 3082 | |
96d365e0 | 3083 | down_read(&css_set_rwsem); |
69d0206c TH |
3084 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3085 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3086 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3087 | return count; |
3088 | } | |
3089 | ||
53fa5261 | 3090 | /** |
492eb21b | 3091 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3092 | * @pos: the current position (%NULL to initiate traversal) |
3093 | * @parent: css whose children to walk | |
53fa5261 | 3094 | * |
c2931b70 | 3095 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3096 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3097 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3098 | * be returned regardless of their states. | |
3099 | * | |
3100 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3101 | * css which finished ->css_online() is guaranteed to be visible in the | |
3102 | * future iterations and will stay visible until the last reference is put. | |
3103 | * A css which hasn't finished ->css_online() or already finished | |
3104 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3105 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3106 | */ |
c2931b70 TH |
3107 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3108 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3109 | { |
c2931b70 | 3110 | struct cgroup_subsys_state *next; |
53fa5261 | 3111 | |
8353da1f | 3112 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3113 | |
3114 | /* | |
de3f0341 TH |
3115 | * @pos could already have been unlinked from the sibling list. |
3116 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3117 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3118 | * @pos is taken off list, at which time its next pointer is valid, | |
3119 | * and, as releases are serialized, the one pointed to by the next | |
3120 | * pointer is guaranteed to not have started release yet. This | |
3121 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3122 | * critical section, the one pointed to by its next pointer is | |
3123 | * guaranteed to not have finished its RCU grace period even if we | |
3124 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3125 | * |
de3f0341 TH |
3126 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3127 | * dereferenced; however, as each css is given a monotonically | |
3128 | * increasing unique serial number and always appended to the | |
3129 | * sibling list, the next one can be found by walking the parent's | |
3130 | * children until the first css with higher serial number than | |
3131 | * @pos's. While this path can be slower, it happens iff iteration | |
3132 | * races against release and the race window is very small. | |
53fa5261 | 3133 | */ |
3b287a50 | 3134 | if (!pos) { |
c2931b70 TH |
3135 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3136 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3137 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3138 | } else { |
c2931b70 TH |
3139 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3140 | if (next->serial_nr > pos->serial_nr) | |
3b287a50 | 3141 | break; |
53fa5261 TH |
3142 | } |
3143 | ||
3b281afb TH |
3144 | /* |
3145 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3146 | * the next sibling. |
3b281afb | 3147 | */ |
c2931b70 TH |
3148 | if (&next->sibling != &parent->children) |
3149 | return next; | |
3b281afb | 3150 | return NULL; |
53fa5261 | 3151 | } |
53fa5261 | 3152 | |
574bd9f7 | 3153 | /** |
492eb21b | 3154 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3155 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3156 | * @root: css whose descendants to walk |
574bd9f7 | 3157 | * |
492eb21b | 3158 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3159 | * to visit for pre-order traversal of @root's descendants. @root is |
3160 | * included in the iteration and the first node to be visited. | |
75501a6d | 3161 | * |
87fb54f1 TH |
3162 | * While this function requires cgroup_mutex or RCU read locking, it |
3163 | * doesn't require the whole traversal to be contained in a single critical | |
3164 | * section. This function will return the correct next descendant as long | |
3165 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3166 | * |
3167 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3168 | * css which finished ->css_online() is guaranteed to be visible in the | |
3169 | * future iterations and will stay visible until the last reference is put. | |
3170 | * A css which hasn't finished ->css_online() or already finished | |
3171 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3172 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3173 | */ |
492eb21b TH |
3174 | struct cgroup_subsys_state * |
3175 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3176 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3177 | { |
492eb21b | 3178 | struct cgroup_subsys_state *next; |
574bd9f7 | 3179 | |
8353da1f | 3180 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3181 | |
bd8815a6 | 3182 | /* if first iteration, visit @root */ |
7805d000 | 3183 | if (!pos) |
bd8815a6 | 3184 | return root; |
574bd9f7 TH |
3185 | |
3186 | /* visit the first child if exists */ | |
492eb21b | 3187 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3188 | if (next) |
3189 | return next; | |
3190 | ||
3191 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3192 | while (pos != root) { |
5c9d535b | 3193 | next = css_next_child(pos, pos->parent); |
75501a6d | 3194 | if (next) |
574bd9f7 | 3195 | return next; |
5c9d535b | 3196 | pos = pos->parent; |
7805d000 | 3197 | } |
574bd9f7 TH |
3198 | |
3199 | return NULL; | |
3200 | } | |
574bd9f7 | 3201 | |
12a9d2fe | 3202 | /** |
492eb21b TH |
3203 | * css_rightmost_descendant - return the rightmost descendant of a css |
3204 | * @pos: css of interest | |
12a9d2fe | 3205 | * |
492eb21b TH |
3206 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3207 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3208 | * subtree of @pos. |
75501a6d | 3209 | * |
87fb54f1 TH |
3210 | * While this function requires cgroup_mutex or RCU read locking, it |
3211 | * doesn't require the whole traversal to be contained in a single critical | |
3212 | * section. This function will return the correct rightmost descendant as | |
3213 | * long as @pos is accessible. | |
12a9d2fe | 3214 | */ |
492eb21b TH |
3215 | struct cgroup_subsys_state * |
3216 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3217 | { |
492eb21b | 3218 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3219 | |
8353da1f | 3220 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3221 | |
3222 | do { | |
3223 | last = pos; | |
3224 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3225 | pos = NULL; | |
492eb21b | 3226 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3227 | pos = tmp; |
3228 | } while (pos); | |
3229 | ||
3230 | return last; | |
3231 | } | |
12a9d2fe | 3232 | |
492eb21b TH |
3233 | static struct cgroup_subsys_state * |
3234 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3235 | { |
492eb21b | 3236 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3237 | |
3238 | do { | |
3239 | last = pos; | |
492eb21b | 3240 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3241 | } while (pos); |
3242 | ||
3243 | return last; | |
3244 | } | |
3245 | ||
3246 | /** | |
492eb21b | 3247 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3248 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3249 | * @root: css whose descendants to walk |
574bd9f7 | 3250 | * |
492eb21b | 3251 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3252 | * to visit for post-order traversal of @root's descendants. @root is |
3253 | * included in the iteration and the last node to be visited. | |
75501a6d | 3254 | * |
87fb54f1 TH |
3255 | * While this function requires cgroup_mutex or RCU read locking, it |
3256 | * doesn't require the whole traversal to be contained in a single critical | |
3257 | * section. This function will return the correct next descendant as long | |
3258 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3259 | * @cgroup. | |
c2931b70 TH |
3260 | * |
3261 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3262 | * css which finished ->css_online() is guaranteed to be visible in the | |
3263 | * future iterations and will stay visible until the last reference is put. | |
3264 | * A css which hasn't finished ->css_online() or already finished | |
3265 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3266 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3267 | */ |
492eb21b TH |
3268 | struct cgroup_subsys_state * |
3269 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3270 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3271 | { |
492eb21b | 3272 | struct cgroup_subsys_state *next; |
574bd9f7 | 3273 | |
8353da1f | 3274 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3275 | |
58b79a91 TH |
3276 | /* if first iteration, visit leftmost descendant which may be @root */ |
3277 | if (!pos) | |
3278 | return css_leftmost_descendant(root); | |
574bd9f7 | 3279 | |
bd8815a6 TH |
3280 | /* if we visited @root, we're done */ |
3281 | if (pos == root) | |
3282 | return NULL; | |
3283 | ||
574bd9f7 | 3284 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3285 | next = css_next_child(pos, pos->parent); |
75501a6d | 3286 | if (next) |
492eb21b | 3287 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3288 | |
3289 | /* no sibling left, visit parent */ | |
5c9d535b | 3290 | return pos->parent; |
574bd9f7 | 3291 | } |
574bd9f7 | 3292 | |
f3d46500 TH |
3293 | /** |
3294 | * css_has_online_children - does a css have online children | |
3295 | * @css: the target css | |
3296 | * | |
3297 | * Returns %true if @css has any online children; otherwise, %false. This | |
3298 | * function can be called from any context but the caller is responsible | |
3299 | * for synchronizing against on/offlining as necessary. | |
3300 | */ | |
3301 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3302 | { |
f3d46500 TH |
3303 | struct cgroup_subsys_state *child; |
3304 | bool ret = false; | |
cbc125ef TH |
3305 | |
3306 | rcu_read_lock(); | |
f3d46500 TH |
3307 | css_for_each_child(child, css) { |
3308 | if (css->flags & CSS_ONLINE) { | |
3309 | ret = true; | |
3310 | break; | |
cbc125ef TH |
3311 | } |
3312 | } | |
3313 | rcu_read_unlock(); | |
f3d46500 | 3314 | return ret; |
cbc125ef TH |
3315 | } |
3316 | ||
0942eeee | 3317 | /** |
72ec7029 | 3318 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3319 | * @it: the iterator to advance |
3320 | * | |
3321 | * Advance @it to the next css_set to walk. | |
d515876e | 3322 | */ |
72ec7029 | 3323 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3324 | { |
0f0a2b4f | 3325 | struct list_head *l = it->cset_pos; |
d515876e TH |
3326 | struct cgrp_cset_link *link; |
3327 | struct css_set *cset; | |
3328 | ||
3329 | /* Advance to the next non-empty css_set */ | |
3330 | do { | |
3331 | l = l->next; | |
0f0a2b4f TH |
3332 | if (l == it->cset_head) { |
3333 | it->cset_pos = NULL; | |
d515876e TH |
3334 | return; |
3335 | } | |
3ebb2b6e TH |
3336 | |
3337 | if (it->ss) { | |
3338 | cset = container_of(l, struct css_set, | |
3339 | e_cset_node[it->ss->id]); | |
3340 | } else { | |
3341 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3342 | cset = link->cset; | |
3343 | } | |
c7561128 TH |
3344 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3345 | ||
0f0a2b4f | 3346 | it->cset_pos = l; |
c7561128 TH |
3347 | |
3348 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3349 | it->task_pos = cset->tasks.next; |
c7561128 | 3350 | else |
0f0a2b4f TH |
3351 | it->task_pos = cset->mg_tasks.next; |
3352 | ||
3353 | it->tasks_head = &cset->tasks; | |
3354 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3355 | } |
3356 | ||
0942eeee | 3357 | /** |
72ec7029 TH |
3358 | * css_task_iter_start - initiate task iteration |
3359 | * @css: the css to walk tasks of | |
0942eeee TH |
3360 | * @it: the task iterator to use |
3361 | * | |
72ec7029 TH |
3362 | * Initiate iteration through the tasks of @css. The caller can call |
3363 | * css_task_iter_next() to walk through the tasks until the function | |
3364 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3365 | * called. | |
0942eeee TH |
3366 | * |
3367 | * Note that this function acquires a lock which is released when the | |
3368 | * iteration finishes. The caller can't sleep while iteration is in | |
3369 | * progress. | |
3370 | */ | |
72ec7029 TH |
3371 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3372 | struct css_task_iter *it) | |
96d365e0 | 3373 | __acquires(css_set_rwsem) |
817929ec | 3374 | { |
56fde9e0 TH |
3375 | /* no one should try to iterate before mounting cgroups */ |
3376 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3377 | |
96d365e0 | 3378 | down_read(&css_set_rwsem); |
c59cd3d8 | 3379 | |
3ebb2b6e TH |
3380 | it->ss = css->ss; |
3381 | ||
3382 | if (it->ss) | |
3383 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3384 | else | |
3385 | it->cset_pos = &css->cgroup->cset_links; | |
3386 | ||
0f0a2b4f | 3387 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3388 | |
72ec7029 | 3389 | css_advance_task_iter(it); |
817929ec PM |
3390 | } |
3391 | ||
0942eeee | 3392 | /** |
72ec7029 | 3393 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3394 | * @it: the task iterator being iterated |
3395 | * | |
3396 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3397 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3398 | * reaches the end. | |
0942eeee | 3399 | */ |
72ec7029 | 3400 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3401 | { |
3402 | struct task_struct *res; | |
0f0a2b4f | 3403 | struct list_head *l = it->task_pos; |
817929ec PM |
3404 | |
3405 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3406 | if (!it->cset_pos) |
817929ec PM |
3407 | return NULL; |
3408 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3409 | |
3410 | /* | |
3411 | * Advance iterator to find next entry. cset->tasks is consumed | |
3412 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3413 | * next cset. | |
3414 | */ | |
817929ec | 3415 | l = l->next; |
c7561128 | 3416 | |
0f0a2b4f TH |
3417 | if (l == it->tasks_head) |
3418 | l = it->mg_tasks_head->next; | |
c7561128 | 3419 | |
0f0a2b4f | 3420 | if (l == it->mg_tasks_head) |
72ec7029 | 3421 | css_advance_task_iter(it); |
c7561128 | 3422 | else |
0f0a2b4f | 3423 | it->task_pos = l; |
c7561128 | 3424 | |
817929ec PM |
3425 | return res; |
3426 | } | |
3427 | ||
0942eeee | 3428 | /** |
72ec7029 | 3429 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3430 | * @it: the task iterator to finish |
3431 | * | |
72ec7029 | 3432 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3433 | */ |
72ec7029 | 3434 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3435 | __releases(css_set_rwsem) |
31a7df01 | 3436 | { |
96d365e0 | 3437 | up_read(&css_set_rwsem); |
31a7df01 CW |
3438 | } |
3439 | ||
3440 | /** | |
8cc99345 TH |
3441 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3442 | * @to: cgroup to which the tasks will be moved | |
3443 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3444 | * |
eaf797ab TH |
3445 | * Locking rules between cgroup_post_fork() and the migration path |
3446 | * guarantee that, if a task is forking while being migrated, the new child | |
3447 | * is guaranteed to be either visible in the source cgroup after the | |
3448 | * parent's migration is complete or put into the target cgroup. No task | |
3449 | * can slip out of migration through forking. | |
31a7df01 | 3450 | */ |
8cc99345 | 3451 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3452 | { |
952aaa12 TH |
3453 | LIST_HEAD(preloaded_csets); |
3454 | struct cgrp_cset_link *link; | |
72ec7029 | 3455 | struct css_task_iter it; |
e406d1cf | 3456 | struct task_struct *task; |
952aaa12 | 3457 | int ret; |
31a7df01 | 3458 | |
952aaa12 | 3459 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3460 | |
952aaa12 TH |
3461 | /* all tasks in @from are being moved, all csets are source */ |
3462 | down_read(&css_set_rwsem); | |
3463 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3464 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3465 | up_read(&css_set_rwsem); | |
31a7df01 | 3466 | |
952aaa12 TH |
3467 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3468 | if (ret) | |
3469 | goto out_err; | |
8cc99345 | 3470 | |
952aaa12 TH |
3471 | /* |
3472 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3473 | * ->can_attach() fails. | |
3474 | */ | |
e406d1cf | 3475 | do { |
9d800df1 | 3476 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
3477 | task = css_task_iter_next(&it); |
3478 | if (task) | |
3479 | get_task_struct(task); | |
3480 | css_task_iter_end(&it); | |
3481 | ||
3482 | if (task) { | |
952aaa12 | 3483 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3484 | put_task_struct(task); |
3485 | } | |
3486 | } while (task && !ret); | |
952aaa12 TH |
3487 | out_err: |
3488 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3489 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3490 | return ret; |
8cc99345 TH |
3491 | } |
3492 | ||
bbcb81d0 | 3493 | /* |
102a775e | 3494 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3495 | * |
3496 | * Reading this file can return large amounts of data if a cgroup has | |
3497 | * *lots* of attached tasks. So it may need several calls to read(), | |
3498 | * but we cannot guarantee that the information we produce is correct | |
3499 | * unless we produce it entirely atomically. | |
3500 | * | |
bbcb81d0 | 3501 | */ |
bbcb81d0 | 3502 | |
24528255 LZ |
3503 | /* which pidlist file are we talking about? */ |
3504 | enum cgroup_filetype { | |
3505 | CGROUP_FILE_PROCS, | |
3506 | CGROUP_FILE_TASKS, | |
3507 | }; | |
3508 | ||
3509 | /* | |
3510 | * A pidlist is a list of pids that virtually represents the contents of one | |
3511 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3512 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3513 | * to the cgroup. | |
3514 | */ | |
3515 | struct cgroup_pidlist { | |
3516 | /* | |
3517 | * used to find which pidlist is wanted. doesn't change as long as | |
3518 | * this particular list stays in the list. | |
3519 | */ | |
3520 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3521 | /* array of xids */ | |
3522 | pid_t *list; | |
3523 | /* how many elements the above list has */ | |
3524 | int length; | |
24528255 LZ |
3525 | /* each of these stored in a list by its cgroup */ |
3526 | struct list_head links; | |
3527 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3528 | struct cgroup *owner; | |
b1a21367 TH |
3529 | /* for delayed destruction */ |
3530 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3531 | }; |
3532 | ||
d1d9fd33 BB |
3533 | /* |
3534 | * The following two functions "fix" the issue where there are more pids | |
3535 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3536 | * TODO: replace with a kernel-wide solution to this problem | |
3537 | */ | |
3538 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3539 | static void *pidlist_allocate(int count) | |
3540 | { | |
3541 | if (PIDLIST_TOO_LARGE(count)) | |
3542 | return vmalloc(count * sizeof(pid_t)); | |
3543 | else | |
3544 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3545 | } | |
b1a21367 | 3546 | |
d1d9fd33 BB |
3547 | static void pidlist_free(void *p) |
3548 | { | |
3549 | if (is_vmalloc_addr(p)) | |
3550 | vfree(p); | |
3551 | else | |
3552 | kfree(p); | |
3553 | } | |
d1d9fd33 | 3554 | |
b1a21367 TH |
3555 | /* |
3556 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3557 | * should be left afterwards. | |
3558 | */ | |
3559 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3560 | { | |
3561 | struct cgroup_pidlist *l, *tmp_l; | |
3562 | ||
3563 | mutex_lock(&cgrp->pidlist_mutex); | |
3564 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3565 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3566 | mutex_unlock(&cgrp->pidlist_mutex); | |
3567 | ||
3568 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3569 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3570 | } | |
3571 | ||
3572 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3573 | { | |
3574 | struct delayed_work *dwork = to_delayed_work(work); | |
3575 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3576 | destroy_dwork); | |
3577 | struct cgroup_pidlist *tofree = NULL; | |
3578 | ||
3579 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3580 | |
3581 | /* | |
04502365 TH |
3582 | * Destroy iff we didn't get queued again. The state won't change |
3583 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3584 | */ |
04502365 | 3585 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3586 | list_del(&l->links); |
3587 | pidlist_free(l->list); | |
3588 | put_pid_ns(l->key.ns); | |
3589 | tofree = l; | |
3590 | } | |
3591 | ||
b1a21367 TH |
3592 | mutex_unlock(&l->owner->pidlist_mutex); |
3593 | kfree(tofree); | |
3594 | } | |
3595 | ||
bbcb81d0 | 3596 | /* |
102a775e | 3597 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3598 | * Returns the number of unique elements. |
bbcb81d0 | 3599 | */ |
6ee211ad | 3600 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3601 | { |
102a775e | 3602 | int src, dest = 1; |
102a775e BB |
3603 | |
3604 | /* | |
3605 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3606 | * edge cases first; no work needs to be done for either | |
3607 | */ | |
3608 | if (length == 0 || length == 1) | |
3609 | return length; | |
3610 | /* src and dest walk down the list; dest counts unique elements */ | |
3611 | for (src = 1; src < length; src++) { | |
3612 | /* find next unique element */ | |
3613 | while (list[src] == list[src-1]) { | |
3614 | src++; | |
3615 | if (src == length) | |
3616 | goto after; | |
3617 | } | |
3618 | /* dest always points to where the next unique element goes */ | |
3619 | list[dest] = list[src]; | |
3620 | dest++; | |
3621 | } | |
3622 | after: | |
102a775e BB |
3623 | return dest; |
3624 | } | |
3625 | ||
afb2bc14 TH |
3626 | /* |
3627 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3628 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3629 | * different per namespace, each namespace needs differently sorted list, | |
3630 | * making it impossible to use, for example, single rbtree of member tasks | |
3631 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3632 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3633 | * pidlists keyed by cgroup and namespace. | |
3634 | * | |
3635 | * All this extra complexity was caused by the original implementation | |
3636 | * committing to an entirely unnecessary property. In the long term, we | |
3637 | * want to do away with it. Explicitly scramble sort order if | |
3638 | * sane_behavior so that no such expectation exists in the new interface. | |
3639 | * | |
3640 | * Scrambling is done by swapping every two consecutive bits, which is | |
3641 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3642 | */ | |
3643 | static pid_t pid_fry(pid_t pid) | |
3644 | { | |
3645 | unsigned a = pid & 0x55555555; | |
3646 | unsigned b = pid & 0xAAAAAAAA; | |
3647 | ||
3648 | return (a << 1) | (b >> 1); | |
3649 | } | |
3650 | ||
3651 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3652 | { | |
3653 | if (cgroup_sane_behavior(cgrp)) | |
3654 | return pid_fry(pid); | |
3655 | else | |
3656 | return pid; | |
3657 | } | |
3658 | ||
102a775e BB |
3659 | static int cmppid(const void *a, const void *b) |
3660 | { | |
3661 | return *(pid_t *)a - *(pid_t *)b; | |
3662 | } | |
3663 | ||
afb2bc14 TH |
3664 | static int fried_cmppid(const void *a, const void *b) |
3665 | { | |
3666 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3667 | } | |
3668 | ||
e6b81710 TH |
3669 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3670 | enum cgroup_filetype type) | |
3671 | { | |
3672 | struct cgroup_pidlist *l; | |
3673 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3674 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3675 | ||
3676 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3677 | ||
3678 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3679 | if (l->key.type == type && l->key.ns == ns) | |
3680 | return l; | |
3681 | return NULL; | |
3682 | } | |
3683 | ||
72a8cb30 BB |
3684 | /* |
3685 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3686 | * returns with the lock on that pidlist already held, and takes care | |
3687 | * of the use count, or returns NULL with no locks held if we're out of | |
3688 | * memory. | |
3689 | */ | |
e6b81710 TH |
3690 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3691 | enum cgroup_filetype type) | |
72a8cb30 BB |
3692 | { |
3693 | struct cgroup_pidlist *l; | |
b70cc5fd | 3694 | |
e6b81710 TH |
3695 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3696 | ||
3697 | l = cgroup_pidlist_find(cgrp, type); | |
3698 | if (l) | |
3699 | return l; | |
3700 | ||
72a8cb30 | 3701 | /* entry not found; create a new one */ |
f4f4be2b | 3702 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3703 | if (!l) |
72a8cb30 | 3704 | return l; |
e6b81710 | 3705 | |
b1a21367 | 3706 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3707 | l->key.type = type; |
e6b81710 TH |
3708 | /* don't need task_nsproxy() if we're looking at ourself */ |
3709 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3710 | l->owner = cgrp; |
3711 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3712 | return l; |
3713 | } | |
3714 | ||
102a775e BB |
3715 | /* |
3716 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3717 | */ | |
72a8cb30 BB |
3718 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3719 | struct cgroup_pidlist **lp) | |
102a775e BB |
3720 | { |
3721 | pid_t *array; | |
3722 | int length; | |
3723 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3724 | struct css_task_iter it; |
817929ec | 3725 | struct task_struct *tsk; |
102a775e BB |
3726 | struct cgroup_pidlist *l; |
3727 | ||
4bac00d1 TH |
3728 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3729 | ||
102a775e BB |
3730 | /* |
3731 | * If cgroup gets more users after we read count, we won't have | |
3732 | * enough space - tough. This race is indistinguishable to the | |
3733 | * caller from the case that the additional cgroup users didn't | |
3734 | * show up until sometime later on. | |
3735 | */ | |
3736 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3737 | array = pidlist_allocate(length); |
102a775e BB |
3738 | if (!array) |
3739 | return -ENOMEM; | |
3740 | /* now, populate the array */ | |
9d800df1 | 3741 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 3742 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 3743 | if (unlikely(n == length)) |
817929ec | 3744 | break; |
102a775e | 3745 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3746 | if (type == CGROUP_FILE_PROCS) |
3747 | pid = task_tgid_vnr(tsk); | |
3748 | else | |
3749 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3750 | if (pid > 0) /* make sure to only use valid results */ |
3751 | array[n++] = pid; | |
817929ec | 3752 | } |
72ec7029 | 3753 | css_task_iter_end(&it); |
102a775e BB |
3754 | length = n; |
3755 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3756 | if (cgroup_sane_behavior(cgrp)) |
3757 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3758 | else | |
3759 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3760 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3761 | length = pidlist_uniq(array, length); |
e6b81710 | 3762 | |
e6b81710 | 3763 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3764 | if (!l) { |
e6b81710 | 3765 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3766 | pidlist_free(array); |
72a8cb30 | 3767 | return -ENOMEM; |
102a775e | 3768 | } |
e6b81710 TH |
3769 | |
3770 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3771 | pidlist_free(l->list); |
102a775e BB |
3772 | l->list = array; |
3773 | l->length = length; | |
72a8cb30 | 3774 | *lp = l; |
102a775e | 3775 | return 0; |
bbcb81d0 PM |
3776 | } |
3777 | ||
846c7bb0 | 3778 | /** |
a043e3b2 | 3779 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3780 | * @stats: cgroupstats to fill information into |
3781 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3782 | * been requested. | |
a043e3b2 LZ |
3783 | * |
3784 | * Build and fill cgroupstats so that taskstats can export it to user | |
3785 | * space. | |
846c7bb0 BS |
3786 | */ |
3787 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3788 | { | |
2bd59d48 | 3789 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3790 | struct cgroup *cgrp; |
72ec7029 | 3791 | struct css_task_iter it; |
846c7bb0 | 3792 | struct task_struct *tsk; |
33d283be | 3793 | |
2bd59d48 TH |
3794 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3795 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3796 | kernfs_type(kn) != KERNFS_DIR) | |
3797 | return -EINVAL; | |
3798 | ||
bad34660 LZ |
3799 | mutex_lock(&cgroup_mutex); |
3800 | ||
846c7bb0 | 3801 | /* |
2bd59d48 | 3802 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 3803 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 3804 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 3805 | */ |
2bd59d48 TH |
3806 | rcu_read_lock(); |
3807 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3808 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3809 | rcu_read_unlock(); |
bad34660 | 3810 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3811 | return -ENOENT; |
3812 | } | |
bad34660 | 3813 | rcu_read_unlock(); |
846c7bb0 | 3814 | |
9d800df1 | 3815 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 3816 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
3817 | switch (tsk->state) { |
3818 | case TASK_RUNNING: | |
3819 | stats->nr_running++; | |
3820 | break; | |
3821 | case TASK_INTERRUPTIBLE: | |
3822 | stats->nr_sleeping++; | |
3823 | break; | |
3824 | case TASK_UNINTERRUPTIBLE: | |
3825 | stats->nr_uninterruptible++; | |
3826 | break; | |
3827 | case TASK_STOPPED: | |
3828 | stats->nr_stopped++; | |
3829 | break; | |
3830 | default: | |
3831 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3832 | stats->nr_io_wait++; | |
3833 | break; | |
3834 | } | |
3835 | } | |
72ec7029 | 3836 | css_task_iter_end(&it); |
846c7bb0 | 3837 | |
bad34660 | 3838 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3839 | return 0; |
846c7bb0 BS |
3840 | } |
3841 | ||
8f3ff208 | 3842 | |
bbcb81d0 | 3843 | /* |
102a775e | 3844 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3845 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3846 | * in the cgroup->l->list array. |
bbcb81d0 | 3847 | */ |
cc31edce | 3848 | |
102a775e | 3849 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3850 | { |
cc31edce PM |
3851 | /* |
3852 | * Initially we receive a position value that corresponds to | |
3853 | * one more than the last pid shown (or 0 on the first call or | |
3854 | * after a seek to the start). Use a binary-search to find the | |
3855 | * next pid to display, if any | |
3856 | */ | |
2bd59d48 | 3857 | struct kernfs_open_file *of = s->private; |
7da11279 | 3858 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3859 | struct cgroup_pidlist *l; |
7da11279 | 3860 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3861 | int index = 0, pid = *pos; |
4bac00d1 TH |
3862 | int *iter, ret; |
3863 | ||
3864 | mutex_lock(&cgrp->pidlist_mutex); | |
3865 | ||
3866 | /* | |
5d22444f | 3867 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3868 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3869 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3870 | * could already have been destroyed. |
3871 | */ | |
5d22444f TH |
3872 | if (of->priv) |
3873 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3874 | |
3875 | /* | |
3876 | * Either this is the first start() after open or the matching | |
3877 | * pidlist has been destroyed inbetween. Create a new one. | |
3878 | */ | |
5d22444f TH |
3879 | if (!of->priv) { |
3880 | ret = pidlist_array_load(cgrp, type, | |
3881 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3882 | if (ret) |
3883 | return ERR_PTR(ret); | |
3884 | } | |
5d22444f | 3885 | l = of->priv; |
cc31edce | 3886 | |
cc31edce | 3887 | if (pid) { |
102a775e | 3888 | int end = l->length; |
20777766 | 3889 | |
cc31edce PM |
3890 | while (index < end) { |
3891 | int mid = (index + end) / 2; | |
afb2bc14 | 3892 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3893 | index = mid; |
3894 | break; | |
afb2bc14 | 3895 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3896 | index = mid + 1; |
3897 | else | |
3898 | end = mid; | |
3899 | } | |
3900 | } | |
3901 | /* If we're off the end of the array, we're done */ | |
102a775e | 3902 | if (index >= l->length) |
cc31edce PM |
3903 | return NULL; |
3904 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3905 | iter = l->list + index; |
afb2bc14 | 3906 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3907 | return iter; |
3908 | } | |
3909 | ||
102a775e | 3910 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3911 | { |
2bd59d48 | 3912 | struct kernfs_open_file *of = s->private; |
5d22444f | 3913 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3914 | |
5d22444f TH |
3915 | if (l) |
3916 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3917 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3918 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3919 | } |
3920 | ||
102a775e | 3921 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3922 | { |
2bd59d48 | 3923 | struct kernfs_open_file *of = s->private; |
5d22444f | 3924 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3925 | pid_t *p = v; |
3926 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3927 | /* |
3928 | * Advance to the next pid in the array. If this goes off the | |
3929 | * end, we're done | |
3930 | */ | |
3931 | p++; | |
3932 | if (p >= end) { | |
3933 | return NULL; | |
3934 | } else { | |
7da11279 | 3935 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3936 | return p; |
3937 | } | |
3938 | } | |
3939 | ||
102a775e | 3940 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3941 | { |
3942 | return seq_printf(s, "%d\n", *(int *)v); | |
3943 | } | |
bbcb81d0 | 3944 | |
182446d0 TH |
3945 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3946 | struct cftype *cft) | |
81a6a5cd | 3947 | { |
182446d0 | 3948 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3949 | } |
3950 | ||
182446d0 TH |
3951 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3952 | struct cftype *cft, u64 val) | |
6379c106 | 3953 | { |
182446d0 | 3954 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3955 | if (val) |
182446d0 | 3956 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3957 | else |
182446d0 | 3958 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3959 | return 0; |
3960 | } | |
3961 | ||
182446d0 TH |
3962 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3963 | struct cftype *cft) | |
97978e6d | 3964 | { |
182446d0 | 3965 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3966 | } |
3967 | ||
182446d0 TH |
3968 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3969 | struct cftype *cft, u64 val) | |
97978e6d DL |
3970 | { |
3971 | if (val) | |
182446d0 | 3972 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3973 | else |
182446d0 | 3974 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3975 | return 0; |
3976 | } | |
3977 | ||
d5c56ced | 3978 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3979 | { |
d5c56ced | 3980 | .name = "cgroup.procs", |
6612f05b TH |
3981 | .seq_start = cgroup_pidlist_start, |
3982 | .seq_next = cgroup_pidlist_next, | |
3983 | .seq_stop = cgroup_pidlist_stop, | |
3984 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3985 | .private = CGROUP_FILE_PROCS, |
acbef755 | 3986 | .write = cgroup_procs_write, |
74a1166d | 3987 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3988 | }, |
97978e6d DL |
3989 | { |
3990 | .name = "cgroup.clone_children", | |
873fe09e | 3991 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3992 | .read_u64 = cgroup_clone_children_read, |
3993 | .write_u64 = cgroup_clone_children_write, | |
3994 | }, | |
873fe09e TH |
3995 | { |
3996 | .name = "cgroup.sane_behavior", | |
3997 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3998 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3999 | }, |
f8f22e53 TH |
4000 | { |
4001 | .name = "cgroup.controllers", | |
4002 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_ONLY_ON_ROOT, | |
4003 | .seq_show = cgroup_root_controllers_show, | |
4004 | }, | |
4005 | { | |
4006 | .name = "cgroup.controllers", | |
4007 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
4008 | .seq_show = cgroup_controllers_show, | |
4009 | }, | |
4010 | { | |
4011 | .name = "cgroup.subtree_control", | |
4012 | .flags = CFTYPE_ONLY_ON_DFL, | |
4013 | .seq_show = cgroup_subtree_control_show, | |
451af504 | 4014 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4015 | }, |
842b597e TH |
4016 | { |
4017 | .name = "cgroup.populated", | |
4018 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
4019 | .seq_show = cgroup_populated_show, | |
4020 | }, | |
d5c56ced TH |
4021 | |
4022 | /* | |
4023 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
4024 | * don't exist if sane_behavior. If you're depending on these, be | |
4025 | * prepared to be burned. | |
4026 | */ | |
4027 | { | |
4028 | .name = "tasks", | |
4029 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
4030 | .seq_start = cgroup_pidlist_start, |
4031 | .seq_next = cgroup_pidlist_next, | |
4032 | .seq_stop = cgroup_pidlist_stop, | |
4033 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4034 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4035 | .write = cgroup_tasks_write, |
d5c56ced TH |
4036 | .mode = S_IRUGO | S_IWUSR, |
4037 | }, | |
4038 | { | |
4039 | .name = "notify_on_release", | |
4040 | .flags = CFTYPE_INSANE, | |
4041 | .read_u64 = cgroup_read_notify_on_release, | |
4042 | .write_u64 = cgroup_write_notify_on_release, | |
4043 | }, | |
6e6ff25b TH |
4044 | { |
4045 | .name = "release_agent", | |
cc5943a7 | 4046 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4047 | .seq_show = cgroup_release_agent_show, |
451af504 | 4048 | .write = cgroup_release_agent_write, |
5f469907 | 4049 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4050 | }, |
db0416b6 | 4051 | { } /* terminate */ |
bbcb81d0 PM |
4052 | }; |
4053 | ||
13af07df | 4054 | /** |
628f7cd4 | 4055 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4056 | * @cgrp: target cgroup |
13af07df | 4057 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4058 | * |
4059 | * On failure, no file is added. | |
13af07df | 4060 | */ |
69dfa00c | 4061 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
ddbcc7e8 | 4062 | { |
ddbcc7e8 | 4063 | struct cgroup_subsys *ss; |
b420ba7d | 4064 | int i, ret = 0; |
bbcb81d0 | 4065 | |
8e3f6541 | 4066 | /* process cftsets of each subsystem */ |
b420ba7d | 4067 | for_each_subsys(ss, i) { |
0adb0704 | 4068 | struct cftype *cfts; |
b420ba7d | 4069 | |
69dfa00c | 4070 | if (!(subsys_mask & (1 << i))) |
13af07df | 4071 | continue; |
8e3f6541 | 4072 | |
0adb0704 TH |
4073 | list_for_each_entry(cfts, &ss->cfts, node) { |
4074 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4075 | if (ret < 0) |
4076 | goto err; | |
4077 | } | |
ddbcc7e8 | 4078 | } |
ddbcc7e8 | 4079 | return 0; |
bee55099 TH |
4080 | err: |
4081 | cgroup_clear_dir(cgrp, subsys_mask); | |
4082 | return ret; | |
ddbcc7e8 PM |
4083 | } |
4084 | ||
0c21ead1 TH |
4085 | /* |
4086 | * css destruction is four-stage process. | |
4087 | * | |
4088 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4089 | * Implemented in kill_css(). | |
4090 | * | |
4091 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4092 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4093 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4094 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4095 | * |
4096 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4097 | * accessors are inside RCU read sections. css_release() schedules the | |
4098 | * RCU callback. | |
4099 | * | |
4100 | * 4. After the grace period, the css can be freed. Implemented in | |
4101 | * css_free_work_fn(). | |
4102 | * | |
4103 | * It is actually hairier because both step 2 and 4 require process context | |
4104 | * and thus involve punting to css->destroy_work adding two additional | |
4105 | * steps to the already complex sequence. | |
4106 | */ | |
35ef10da | 4107 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4108 | { |
4109 | struct cgroup_subsys_state *css = | |
35ef10da | 4110 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4111 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4112 | |
9d755d33 TH |
4113 | if (css->ss) { |
4114 | /* css free path */ | |
4115 | if (css->parent) | |
4116 | css_put(css->parent); | |
0ae78e0b | 4117 | |
9d755d33 TH |
4118 | css->ss->css_free(css); |
4119 | cgroup_put(cgrp); | |
4120 | } else { | |
4121 | /* cgroup free path */ | |
4122 | atomic_dec(&cgrp->root->nr_cgrps); | |
4123 | cgroup_pidlist_destroy_all(cgrp); | |
4124 | ||
d51f39b0 | 4125 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4126 | /* |
4127 | * We get a ref to the parent, and put the ref when | |
4128 | * this cgroup is being freed, so it's guaranteed | |
4129 | * that the parent won't be destroyed before its | |
4130 | * children. | |
4131 | */ | |
d51f39b0 | 4132 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4133 | kernfs_put(cgrp->kn); |
4134 | kfree(cgrp); | |
4135 | } else { | |
4136 | /* | |
4137 | * This is root cgroup's refcnt reaching zero, | |
4138 | * which indicates that the root should be | |
4139 | * released. | |
4140 | */ | |
4141 | cgroup_destroy_root(cgrp->root); | |
4142 | } | |
4143 | } | |
48ddbe19 TH |
4144 | } |
4145 | ||
0c21ead1 | 4146 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4147 | { |
4148 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4149 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4150 | |
35ef10da | 4151 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4152 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4153 | } |
4154 | ||
25e15d83 | 4155 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4156 | { |
4157 | struct cgroup_subsys_state *css = | |
25e15d83 | 4158 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4159 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4160 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4161 | |
1fed1b2e TH |
4162 | mutex_lock(&cgroup_mutex); |
4163 | ||
de3f0341 | 4164 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4165 | list_del_rcu(&css->sibling); |
4166 | ||
9d755d33 TH |
4167 | if (ss) { |
4168 | /* css release path */ | |
4169 | cgroup_idr_remove(&ss->css_idr, css->id); | |
4170 | } else { | |
4171 | /* cgroup release path */ | |
9d755d33 TH |
4172 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4173 | cgrp->id = -1; | |
4174 | } | |
d3daf28d | 4175 | |
1fed1b2e TH |
4176 | mutex_unlock(&cgroup_mutex); |
4177 | ||
0c21ead1 | 4178 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4179 | } |
4180 | ||
25e15d83 TH |
4181 | static void css_release(struct percpu_ref *ref) |
4182 | { | |
4183 | struct cgroup_subsys_state *css = | |
4184 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4185 | ||
4186 | INIT_WORK(&css->destroy_work, css_release_work_fn); | |
4187 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
4188 | } | |
4189 | ||
ddfcadab TH |
4190 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4191 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4192 | { |
0cb51d71 TH |
4193 | lockdep_assert_held(&cgroup_mutex); |
4194 | ||
ddfcadab TH |
4195 | cgroup_get(cgrp); |
4196 | ||
d5c419b6 | 4197 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4198 | css->cgroup = cgrp; |
72c97e54 | 4199 | css->ss = ss; |
d5c419b6 TH |
4200 | INIT_LIST_HEAD(&css->sibling); |
4201 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4202 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4203 | |
d51f39b0 TH |
4204 | if (cgroup_parent(cgrp)) { |
4205 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4206 | css_get(css->parent); |
ddfcadab | 4207 | } |
48ddbe19 | 4208 | |
ca8bdcaf | 4209 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4210 | } |
4211 | ||
2a4ac633 | 4212 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4213 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4214 | { |
623f926b | 4215 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4216 | int ret = 0; |
4217 | ||
a31f2d3f TH |
4218 | lockdep_assert_held(&cgroup_mutex); |
4219 | ||
92fb9748 | 4220 | if (ss->css_online) |
eb95419b | 4221 | ret = ss->css_online(css); |
ae7f164a | 4222 | if (!ret) { |
eb95419b | 4223 | css->flags |= CSS_ONLINE; |
aec25020 | 4224 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4225 | } |
b1929db4 | 4226 | return ret; |
a31f2d3f TH |
4227 | } |
4228 | ||
2a4ac633 | 4229 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4230 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4231 | { |
623f926b | 4232 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4233 | |
4234 | lockdep_assert_held(&cgroup_mutex); | |
4235 | ||
4236 | if (!(css->flags & CSS_ONLINE)) | |
4237 | return; | |
4238 | ||
d7eeac19 | 4239 | if (ss->css_offline) |
eb95419b | 4240 | ss->css_offline(css); |
a31f2d3f | 4241 | |
eb95419b | 4242 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4243 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4244 | |
4245 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4246 | } |
4247 | ||
c81c925a TH |
4248 | /** |
4249 | * create_css - create a cgroup_subsys_state | |
4250 | * @cgrp: the cgroup new css will be associated with | |
4251 | * @ss: the subsys of new css | |
4252 | * | |
4253 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4254 | * css is online and installed in @cgrp with all interface files created. | |
4255 | * Returns 0 on success, -errno on failure. | |
4256 | */ | |
4257 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4258 | { | |
d51f39b0 | 4259 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4260 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4261 | struct cgroup_subsys_state *css; |
4262 | int err; | |
4263 | ||
c81c925a TH |
4264 | lockdep_assert_held(&cgroup_mutex); |
4265 | ||
1fed1b2e | 4266 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4267 | if (IS_ERR(css)) |
4268 | return PTR_ERR(css); | |
4269 | ||
ddfcadab | 4270 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4271 | |
c81c925a TH |
4272 | err = percpu_ref_init(&css->refcnt, css_release); |
4273 | if (err) | |
3eb59ec6 | 4274 | goto err_free_css; |
c81c925a | 4275 | |
15a4c835 TH |
4276 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_NOWAIT); |
4277 | if (err < 0) | |
4278 | goto err_free_percpu_ref; | |
4279 | css->id = err; | |
4280 | ||
aec25020 | 4281 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 4282 | if (err) |
15a4c835 TH |
4283 | goto err_free_id; |
4284 | ||
4285 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4286 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4287 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4288 | |
4289 | err = online_css(css); | |
4290 | if (err) | |
1fed1b2e | 4291 | goto err_list_del; |
c81c925a | 4292 | |
c81c925a | 4293 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4294 | cgroup_parent(parent)) { |
ed3d261b | 4295 | pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", |
a2a1f9ea | 4296 | current->comm, current->pid, ss->name); |
c81c925a | 4297 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4298 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4299 | ss->warned_broken_hierarchy = true; |
4300 | } | |
4301 | ||
4302 | return 0; | |
4303 | ||
1fed1b2e TH |
4304 | err_list_del: |
4305 | list_del_rcu(&css->sibling); | |
32d01dc7 | 4306 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
15a4c835 TH |
4307 | err_free_id: |
4308 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4309 | err_free_percpu_ref: |
c81c925a | 4310 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 4311 | err_free_css: |
a2bed820 | 4312 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4313 | return err; |
4314 | } | |
4315 | ||
b3bfd983 TH |
4316 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4317 | umode_t mode) | |
ddbcc7e8 | 4318 | { |
a9746d8d TH |
4319 | struct cgroup *parent, *cgrp; |
4320 | struct cgroup_root *root; | |
ddbcc7e8 | 4321 | struct cgroup_subsys *ss; |
2bd59d48 | 4322 | struct kernfs_node *kn; |
b3bfd983 | 4323 | int ssid, ret; |
65dff759 | 4324 | |
a9746d8d TH |
4325 | parent = cgroup_kn_lock_live(parent_kn); |
4326 | if (!parent) | |
4327 | return -ENODEV; | |
4328 | root = parent->root; | |
ba0f4d76 TH |
4329 | |
4330 | /* allocate the cgroup and its ID, 0 is reserved for the root */ | |
4331 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); | |
4332 | if (!cgrp) { | |
4333 | ret = -ENOMEM; | |
4334 | goto out_unlock; | |
0ab02ca8 LZ |
4335 | } |
4336 | ||
9d755d33 TH |
4337 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release); |
4338 | if (ret) | |
4339 | goto out_free_cgrp; | |
4340 | ||
0ab02ca8 LZ |
4341 | /* |
4342 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4343 | * a half-baked cgroup. | |
4344 | */ | |
6fa4918d | 4345 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_NOWAIT); |
0ab02ca8 | 4346 | if (cgrp->id < 0) { |
ba0f4d76 | 4347 | ret = -ENOMEM; |
9d755d33 | 4348 | goto out_cancel_ref; |
976c06bc TH |
4349 | } |
4350 | ||
cc31edce | 4351 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4352 | |
9d800df1 | 4353 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4354 | cgrp->root = root; |
ddbcc7e8 | 4355 | |
b6abdb0e LZ |
4356 | if (notify_on_release(parent)) |
4357 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4358 | ||
2260e7fc TH |
4359 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4360 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4361 | |
2bd59d48 | 4362 | /* create the directory */ |
e61734c5 | 4363 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4364 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4365 | ret = PTR_ERR(kn); |
4366 | goto out_free_id; | |
2bd59d48 TH |
4367 | } |
4368 | cgrp->kn = kn; | |
ddbcc7e8 | 4369 | |
4e139afc | 4370 | /* |
6f30558f TH |
4371 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4372 | * that @cgrp->kn is always accessible. | |
4e139afc | 4373 | */ |
6f30558f | 4374 | kernfs_get(kn); |
ddbcc7e8 | 4375 | |
0cb51d71 | 4376 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4377 | |
4e139afc | 4378 | /* allocation complete, commit to creation */ |
d5c419b6 | 4379 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4380 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4381 | cgroup_get(parent); |
415cf07a | 4382 | |
0d80255e TH |
4383 | /* |
4384 | * @cgrp is now fully operational. If something fails after this | |
4385 | * point, it'll be released via the normal destruction path. | |
4386 | */ | |
6fa4918d | 4387 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4388 | |
ba0f4d76 TH |
4389 | ret = cgroup_kn_set_ugid(kn); |
4390 | if (ret) | |
4391 | goto out_destroy; | |
49957f8e | 4392 | |
ba0f4d76 TH |
4393 | ret = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
4394 | if (ret) | |
4395 | goto out_destroy; | |
628f7cd4 | 4396 | |
9d403e99 | 4397 | /* let's create and online css's */ |
b85d2040 | 4398 | for_each_subsys(ss, ssid) { |
f392e51c | 4399 | if (parent->child_subsys_mask & (1 << ssid)) { |
ba0f4d76 TH |
4400 | ret = create_css(cgrp, ss); |
4401 | if (ret) | |
4402 | goto out_destroy; | |
b85d2040 | 4403 | } |
a8638030 | 4404 | } |
ddbcc7e8 | 4405 | |
bd53d617 TH |
4406 | /* |
4407 | * On the default hierarchy, a child doesn't automatically inherit | |
4408 | * child_subsys_mask from the parent. Each is configured manually. | |
4409 | */ | |
4410 | if (!cgroup_on_dfl(cgrp)) | |
4411 | cgrp->child_subsys_mask = parent->child_subsys_mask; | |
f392e51c | 4412 | |
2bd59d48 TH |
4413 | kernfs_activate(kn); |
4414 | ||
ba0f4d76 TH |
4415 | ret = 0; |
4416 | goto out_unlock; | |
ddbcc7e8 | 4417 | |
ba0f4d76 | 4418 | out_free_id: |
6fa4918d | 4419 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 TH |
4420 | out_cancel_ref: |
4421 | percpu_ref_cancel_init(&cgrp->self.refcnt); | |
ba0f4d76 TH |
4422 | out_free_cgrp: |
4423 | kfree(cgrp); | |
4424 | out_unlock: | |
a9746d8d | 4425 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4426 | return ret; |
4b8b47eb | 4427 | |
ba0f4d76 | 4428 | out_destroy: |
4b8b47eb | 4429 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4430 | goto out_unlock; |
ddbcc7e8 PM |
4431 | } |
4432 | ||
223dbc38 TH |
4433 | /* |
4434 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4435 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4436 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4437 | */ |
4438 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4439 | { |
223dbc38 TH |
4440 | struct cgroup_subsys_state *css = |
4441 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4442 | |
f20104de | 4443 | mutex_lock(&cgroup_mutex); |
09a503ea | 4444 | offline_css(css); |
f20104de | 4445 | mutex_unlock(&cgroup_mutex); |
09a503ea | 4446 | |
09a503ea | 4447 | css_put(css); |
d3daf28d TH |
4448 | } |
4449 | ||
223dbc38 TH |
4450 | /* css kill confirmation processing requires process context, bounce */ |
4451 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4452 | { |
4453 | struct cgroup_subsys_state *css = | |
4454 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4455 | ||
223dbc38 | 4456 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4457 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4458 | } |
4459 | ||
f392e51c TH |
4460 | /** |
4461 | * kill_css - destroy a css | |
4462 | * @css: css to destroy | |
4463 | * | |
4464 | * This function initiates destruction of @css by removing cgroup interface | |
4465 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4466 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4467 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4468 | */ |
4469 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4470 | { |
01f6474c | 4471 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4472 | |
2bd59d48 TH |
4473 | /* |
4474 | * This must happen before css is disassociated with its cgroup. | |
4475 | * See seq_css() for details. | |
4476 | */ | |
aec25020 | 4477 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4478 | |
edae0c33 TH |
4479 | /* |
4480 | * Killing would put the base ref, but we need to keep it alive | |
4481 | * until after ->css_offline(). | |
4482 | */ | |
4483 | css_get(css); | |
4484 | ||
4485 | /* | |
4486 | * cgroup core guarantees that, by the time ->css_offline() is | |
4487 | * invoked, no new css reference will be given out via | |
ec903c0c | 4488 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4489 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4490 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4491 | * | |
4492 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4493 | * css is confirmed to be seen as killed on all CPUs. | |
4494 | */ | |
4495 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4496 | } |
4497 | ||
4498 | /** | |
4499 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4500 | * @cgrp: cgroup to be destroyed | |
4501 | * | |
4502 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4503 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4504 | * guarantee that css_tryget_online() won't succeed by the time |
4505 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4506 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4507 | * |
4508 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4509 | * userland visible parts and start killing the percpu refcnts of | |
4510 | * css's. Set up so that the next stage will be kicked off once all | |
4511 | * the percpu refcnts are confirmed to be killed. | |
4512 | * | |
4513 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4514 | * rest of destruction. Once all cgroup references are gone, the | |
4515 | * cgroup is RCU-freed. | |
4516 | * | |
4517 | * This function implements s1. After this step, @cgrp is gone as far as | |
4518 | * the userland is concerned and a new cgroup with the same name may be | |
4519 | * created. As cgroup doesn't care about the names internally, this | |
4520 | * doesn't cause any problem. | |
4521 | */ | |
42809dd4 TH |
4522 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4523 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4524 | { |
2bd59d48 | 4525 | struct cgroup_subsys_state *css; |
ddd69148 | 4526 | bool empty; |
1c6727af | 4527 | int ssid; |
ddbcc7e8 | 4528 | |
42809dd4 TH |
4529 | lockdep_assert_held(&cgroup_mutex); |
4530 | ||
ddd69148 | 4531 | /* |
96d365e0 | 4532 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4533 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4534 | */ |
96d365e0 | 4535 | down_read(&css_set_rwsem); |
bb78a92f | 4536 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4537 | up_read(&css_set_rwsem); |
ddd69148 | 4538 | if (!empty) |
ddbcc7e8 | 4539 | return -EBUSY; |
a043e3b2 | 4540 | |
bb78a92f | 4541 | /* |
d5c419b6 TH |
4542 | * Make sure there's no live children. We can't test emptiness of |
4543 | * ->self.children as dead children linger on it while being | |
4544 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 4545 | */ |
f3d46500 | 4546 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
4547 | return -EBUSY; |
4548 | ||
455050d2 TH |
4549 | /* |
4550 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 4551 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 4552 | */ |
184faf32 | 4553 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 4554 | |
249f3468 | 4555 | /* initiate massacre of all css's */ |
1c6727af TH |
4556 | for_each_css(css, ssid, cgrp) |
4557 | kill_css(css); | |
455050d2 | 4558 | |
184faf32 | 4559 | /* CSS_ONLINE is clear, remove from ->release_list for the last time */ |
455050d2 TH |
4560 | raw_spin_lock(&release_list_lock); |
4561 | if (!list_empty(&cgrp->release_list)) | |
4562 | list_del_init(&cgrp->release_list); | |
4563 | raw_spin_unlock(&release_list_lock); | |
4564 | ||
01f6474c TH |
4565 | /* |
4566 | * Remove @cgrp directory along with the base files. @cgrp has an | |
4567 | * extra ref on its kn. | |
4568 | */ | |
4569 | kernfs_remove(cgrp->kn); | |
455050d2 | 4570 | |
d51f39b0 TH |
4571 | set_bit(CGRP_RELEASABLE, &cgroup_parent(cgrp)->flags); |
4572 | check_for_release(cgroup_parent(cgrp)); | |
9e4173e1 | 4573 | |
249f3468 | 4574 | /* put the base reference */ |
9d755d33 | 4575 | percpu_ref_kill(&cgrp->self.refcnt); |
249f3468 | 4576 | |
ea15f8cc TH |
4577 | return 0; |
4578 | }; | |
4579 | ||
2bd59d48 | 4580 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4581 | { |
a9746d8d | 4582 | struct cgroup *cgrp; |
2bd59d48 | 4583 | int ret = 0; |
42809dd4 | 4584 | |
a9746d8d TH |
4585 | cgrp = cgroup_kn_lock_live(kn); |
4586 | if (!cgrp) | |
4587 | return 0; | |
4588 | cgroup_get(cgrp); /* for @kn->priv clearing */ | |
2bb566cb | 4589 | |
a9746d8d | 4590 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 4591 | |
a9746d8d | 4592 | cgroup_kn_unlock(kn); |
cfc79d5b TH |
4593 | |
4594 | /* | |
4595 | * There are two control paths which try to determine cgroup from | |
4596 | * dentry without going through kernfs - cgroupstats_build() and | |
4597 | * css_tryget_online_from_dir(). Those are supported by RCU | |
4598 | * protecting clearing of cgrp->kn->priv backpointer, which should | |
4599 | * happen after all files under it have been removed. | |
4600 | */ | |
4601 | if (!ret) | |
4602 | RCU_INIT_POINTER(*(void __rcu __force **)&kn->priv, NULL); | |
4603 | ||
2bd59d48 | 4604 | cgroup_put(cgrp); |
42809dd4 | 4605 | return ret; |
8e3f6541 TH |
4606 | } |
4607 | ||
2bd59d48 TH |
4608 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4609 | .remount_fs = cgroup_remount, | |
4610 | .show_options = cgroup_show_options, | |
4611 | .mkdir = cgroup_mkdir, | |
4612 | .rmdir = cgroup_rmdir, | |
4613 | .rename = cgroup_rename, | |
4614 | }; | |
4615 | ||
15a4c835 | 4616 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4617 | { |
ddbcc7e8 | 4618 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4619 | |
4620 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4621 | |
648bb56d TH |
4622 | mutex_lock(&cgroup_mutex); |
4623 | ||
15a4c835 | 4624 | idr_init(&ss->css_idr); |
0adb0704 | 4625 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4626 | |
3dd06ffa TH |
4627 | /* Create the root cgroup state for this subsystem */ |
4628 | ss->root = &cgrp_dfl_root; | |
4629 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4630 | /* We don't handle early failures gracefully */ |
4631 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4632 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
4633 | |
4634 | /* | |
4635 | * Root csses are never destroyed and we can't initialize | |
4636 | * percpu_ref during early init. Disable refcnting. | |
4637 | */ | |
4638 | css->flags |= CSS_NO_REF; | |
4639 | ||
15a4c835 | 4640 | if (early) { |
9395a450 | 4641 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
4642 | css->id = 1; |
4643 | } else { | |
4644 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4645 | BUG_ON(css->id < 0); | |
4646 | } | |
ddbcc7e8 | 4647 | |
e8d55fde | 4648 | /* Update the init_css_set to contain a subsys |
817929ec | 4649 | * pointer to this state - since the subsystem is |
e8d55fde | 4650 | * newly registered, all tasks and hence the |
3dd06ffa | 4651 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4652 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4653 | |
4654 | need_forkexit_callback |= ss->fork || ss->exit; | |
4655 | ||
e8d55fde LZ |
4656 | /* At system boot, before all subsystems have been |
4657 | * registered, no tasks have been forked, so we don't | |
4658 | * need to invoke fork callbacks here. */ | |
4659 | BUG_ON(!list_empty(&init_task.tasks)); | |
4660 | ||
ae7f164a | 4661 | BUG_ON(online_css(css)); |
a8638030 | 4662 | |
f392e51c | 4663 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4664 | |
4665 | mutex_unlock(&cgroup_mutex); | |
4666 | } | |
cf5d5941 | 4667 | |
ddbcc7e8 | 4668 | /** |
a043e3b2 LZ |
4669 | * cgroup_init_early - cgroup initialization at system boot |
4670 | * | |
4671 | * Initialize cgroups at system boot, and initialize any | |
4672 | * subsystems that request early init. | |
ddbcc7e8 PM |
4673 | */ |
4674 | int __init cgroup_init_early(void) | |
4675 | { | |
a2dd4247 TH |
4676 | static struct cgroup_sb_opts __initdata opts = |
4677 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4678 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4679 | int i; |
30159ec7 | 4680 | |
3dd06ffa | 4681 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
4682 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
4683 | ||
a4ea1cc9 | 4684 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4685 | |
3ed80a62 | 4686 | for_each_subsys(ss, i) { |
aec25020 | 4687 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4688 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4689 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4690 | ss->id, ss->name); |
073219e9 TH |
4691 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4692 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4693 | ||
aec25020 | 4694 | ss->id = i; |
073219e9 | 4695 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4696 | |
4697 | if (ss->early_init) | |
15a4c835 | 4698 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
4699 | } |
4700 | return 0; | |
4701 | } | |
4702 | ||
4703 | /** | |
a043e3b2 LZ |
4704 | * cgroup_init - cgroup initialization |
4705 | * | |
4706 | * Register cgroup filesystem and /proc file, and initialize | |
4707 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4708 | */ |
4709 | int __init cgroup_init(void) | |
4710 | { | |
30159ec7 | 4711 | struct cgroup_subsys *ss; |
0ac801fe | 4712 | unsigned long key; |
172a2c06 | 4713 | int ssid, err; |
ddbcc7e8 | 4714 | |
2bd59d48 | 4715 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4716 | |
54e7b4eb | 4717 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4718 | |
82fe9b0d TH |
4719 | /* Add init_css_set to the hash table */ |
4720 | key = css_set_hash(init_css_set.subsys); | |
4721 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4722 | ||
3dd06ffa | 4723 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4724 | |
54e7b4eb TH |
4725 | mutex_unlock(&cgroup_mutex); |
4726 | ||
172a2c06 | 4727 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
4728 | if (ss->early_init) { |
4729 | struct cgroup_subsys_state *css = | |
4730 | init_css_set.subsys[ss->id]; | |
4731 | ||
4732 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
4733 | GFP_KERNEL); | |
4734 | BUG_ON(css->id < 0); | |
4735 | } else { | |
4736 | cgroup_init_subsys(ss, false); | |
4737 | } | |
172a2c06 | 4738 | |
2d8f243a TH |
4739 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4740 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
4741 | ||
172a2c06 TH |
4742 | /* |
4743 | * cftype registration needs kmalloc and can't be done | |
4744 | * during early_init. Register base cftypes separately. | |
4745 | */ | |
4746 | if (ss->base_cftypes) | |
4747 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GKH |
4748 | } |
4749 | ||
676db4af | 4750 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4751 | if (!cgroup_kobj) |
4752 | return -ENOMEM; | |
676db4af | 4753 | |
ddbcc7e8 | 4754 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4755 | if (err < 0) { |
4756 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4757 | return err; |
676db4af | 4758 | } |
ddbcc7e8 | 4759 | |
46ae220b | 4760 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4761 | return 0; |
ddbcc7e8 | 4762 | } |
b4f48b63 | 4763 | |
e5fca243 TH |
4764 | static int __init cgroup_wq_init(void) |
4765 | { | |
4766 | /* | |
4767 | * There isn't much point in executing destruction path in | |
4768 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4769 | * Use 1 for @max_active. |
e5fca243 TH |
4770 | * |
4771 | * We would prefer to do this in cgroup_init() above, but that | |
4772 | * is called before init_workqueues(): so leave this until after. | |
4773 | */ | |
1a11533f | 4774 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4775 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4776 | |
4777 | /* | |
4778 | * Used to destroy pidlists and separate to serve as flush domain. | |
4779 | * Cap @max_active to 1 too. | |
4780 | */ | |
4781 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4782 | 0, 1); | |
4783 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4784 | ||
e5fca243 TH |
4785 | return 0; |
4786 | } | |
4787 | core_initcall(cgroup_wq_init); | |
4788 | ||
a424316c PM |
4789 | /* |
4790 | * proc_cgroup_show() | |
4791 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4792 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4793 | */ |
4794 | ||
4795 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4796 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4797 | { |
4798 | struct pid *pid; | |
4799 | struct task_struct *tsk; | |
e61734c5 | 4800 | char *buf, *path; |
a424316c | 4801 | int retval; |
3dd06ffa | 4802 | struct cgroup_root *root; |
a424316c PM |
4803 | |
4804 | retval = -ENOMEM; | |
e61734c5 | 4805 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4806 | if (!buf) |
4807 | goto out; | |
4808 | ||
4809 | retval = -ESRCH; | |
4810 | pid = m->private; | |
4811 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4812 | if (!tsk) | |
4813 | goto out_free; | |
4814 | ||
4815 | retval = 0; | |
4816 | ||
4817 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4818 | down_read(&css_set_rwsem); |
a424316c | 4819 | |
985ed670 | 4820 | for_each_root(root) { |
a424316c | 4821 | struct cgroup_subsys *ss; |
bd89aabc | 4822 | struct cgroup *cgrp; |
b85d2040 | 4823 | int ssid, count = 0; |
a424316c | 4824 | |
a2dd4247 | 4825 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4826 | continue; |
4827 | ||
2c6ab6d2 | 4828 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4829 | for_each_subsys(ss, ssid) |
f392e51c | 4830 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 4831 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4832 | if (strlen(root->name)) |
4833 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4834 | root->name); | |
a424316c | 4835 | seq_putc(m, ':'); |
7717f7ba | 4836 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4837 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4838 | if (!path) { | |
4839 | retval = -ENAMETOOLONG; | |
a424316c | 4840 | goto out_unlock; |
e61734c5 TH |
4841 | } |
4842 | seq_puts(m, path); | |
a424316c PM |
4843 | seq_putc(m, '\n'); |
4844 | } | |
4845 | ||
4846 | out_unlock: | |
96d365e0 | 4847 | up_read(&css_set_rwsem); |
a424316c PM |
4848 | mutex_unlock(&cgroup_mutex); |
4849 | put_task_struct(tsk); | |
4850 | out_free: | |
4851 | kfree(buf); | |
4852 | out: | |
4853 | return retval; | |
4854 | } | |
4855 | ||
a424316c PM |
4856 | /* Display information about each subsystem and each hierarchy */ |
4857 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4858 | { | |
30159ec7 | 4859 | struct cgroup_subsys *ss; |
a424316c | 4860 | int i; |
a424316c | 4861 | |
8bab8dde | 4862 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4863 | /* |
4864 | * ideally we don't want subsystems moving around while we do this. | |
4865 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4866 | * subsys/hierarchy state. | |
4867 | */ | |
a424316c | 4868 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4869 | |
4870 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4871 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4872 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4873 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4874 | |
a424316c PM |
4875 | mutex_unlock(&cgroup_mutex); |
4876 | return 0; | |
4877 | } | |
4878 | ||
4879 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4880 | { | |
9dce07f1 | 4881 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4882 | } |
4883 | ||
828c0950 | 4884 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4885 | .open = cgroupstats_open, |
4886 | .read = seq_read, | |
4887 | .llseek = seq_lseek, | |
4888 | .release = single_release, | |
4889 | }; | |
4890 | ||
b4f48b63 | 4891 | /** |
eaf797ab | 4892 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4893 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4894 | * |
eaf797ab TH |
4895 | * A task is associated with the init_css_set until cgroup_post_fork() |
4896 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4897 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4898 | */ |
4899 | void cgroup_fork(struct task_struct *child) | |
4900 | { | |
eaf797ab | 4901 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4902 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4903 | } |
4904 | ||
817929ec | 4905 | /** |
a043e3b2 LZ |
4906 | * cgroup_post_fork - called on a new task after adding it to the task list |
4907 | * @child: the task in question | |
4908 | * | |
5edee61e TH |
4909 | * Adds the task to the list running through its css_set if necessary and |
4910 | * call the subsystem fork() callbacks. Has to be after the task is | |
4911 | * visible on the task list in case we race with the first call to | |
0942eeee | 4912 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4913 | * list. |
a043e3b2 | 4914 | */ |
817929ec PM |
4915 | void cgroup_post_fork(struct task_struct *child) |
4916 | { | |
30159ec7 | 4917 | struct cgroup_subsys *ss; |
5edee61e TH |
4918 | int i; |
4919 | ||
3ce3230a | 4920 | /* |
eaf797ab TH |
4921 | * This may race against cgroup_enable_task_cg_links(). As that |
4922 | * function sets use_task_css_set_links before grabbing | |
4923 | * tasklist_lock and we just went through tasklist_lock to add | |
4924 | * @child, it's guaranteed that either we see the set | |
4925 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4926 | * @child during its iteration. | |
4927 | * | |
4928 | * If we won the race, @child is associated with %current's | |
4929 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4930 | * association is stable, and, on completion of the parent's | |
4931 | * migration, @child is visible in the source of migration or | |
4932 | * already in the destination cgroup. This guarantee is necessary | |
4933 | * when implementing operations which need to migrate all tasks of | |
4934 | * a cgroup to another. | |
4935 | * | |
4936 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4937 | * will remain in init_css_set. This is safe because all tasks are | |
4938 | * in the init_css_set before cg_links is enabled and there's no | |
4939 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4940 | */ |
817929ec | 4941 | if (use_task_css_set_links) { |
eaf797ab TH |
4942 | struct css_set *cset; |
4943 | ||
96d365e0 | 4944 | down_write(&css_set_rwsem); |
0e1d768f | 4945 | cset = task_css_set(current); |
eaf797ab TH |
4946 | if (list_empty(&child->cg_list)) { |
4947 | rcu_assign_pointer(child->cgroups, cset); | |
4948 | list_add(&child->cg_list, &cset->tasks); | |
4949 | get_css_set(cset); | |
4950 | } | |
96d365e0 | 4951 | up_write(&css_set_rwsem); |
817929ec | 4952 | } |
5edee61e TH |
4953 | |
4954 | /* | |
4955 | * Call ss->fork(). This must happen after @child is linked on | |
4956 | * css_set; otherwise, @child might change state between ->fork() | |
4957 | * and addition to css_set. | |
4958 | */ | |
4959 | if (need_forkexit_callback) { | |
3ed80a62 | 4960 | for_each_subsys(ss, i) |
5edee61e TH |
4961 | if (ss->fork) |
4962 | ss->fork(child); | |
5edee61e | 4963 | } |
817929ec | 4964 | } |
5edee61e | 4965 | |
b4f48b63 PM |
4966 | /** |
4967 | * cgroup_exit - detach cgroup from exiting task | |
4968 | * @tsk: pointer to task_struct of exiting process | |
4969 | * | |
4970 | * Description: Detach cgroup from @tsk and release it. | |
4971 | * | |
4972 | * Note that cgroups marked notify_on_release force every task in | |
4973 | * them to take the global cgroup_mutex mutex when exiting. | |
4974 | * This could impact scaling on very large systems. Be reluctant to | |
4975 | * use notify_on_release cgroups where very high task exit scaling | |
4976 | * is required on large systems. | |
4977 | * | |
0e1d768f TH |
4978 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4979 | * call cgroup_exit() while the task is still competent to handle | |
4980 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4981 | * each hierarchy for the remainder of its exit. No need to bother with | |
4982 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4983 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4984 | */ |
1ec41830 | 4985 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4986 | { |
30159ec7 | 4987 | struct cgroup_subsys *ss; |
5abb8855 | 4988 | struct css_set *cset; |
eaf797ab | 4989 | bool put_cset = false; |
d41d5a01 | 4990 | int i; |
817929ec PM |
4991 | |
4992 | /* | |
0e1d768f TH |
4993 | * Unlink from @tsk from its css_set. As migration path can't race |
4994 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
4995 | */ |
4996 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4997 | down_write(&css_set_rwsem); |
0e1d768f | 4998 | list_del_init(&tsk->cg_list); |
96d365e0 | 4999 | up_write(&css_set_rwsem); |
0e1d768f | 5000 | put_cset = true; |
817929ec PM |
5001 | } |
5002 | ||
b4f48b63 | 5003 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
5004 | cset = task_css_set(tsk); |
5005 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 5006 | |
1ec41830 | 5007 | if (need_forkexit_callback) { |
3ed80a62 TH |
5008 | /* see cgroup_post_fork() for details */ |
5009 | for_each_subsys(ss, i) { | |
d41d5a01 | 5010 | if (ss->exit) { |
eb95419b TH |
5011 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
5012 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5013 | |
eb95419b | 5014 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
5015 | } |
5016 | } | |
5017 | } | |
d41d5a01 | 5018 | |
eaf797ab TH |
5019 | if (put_cset) |
5020 | put_css_set(cset, true); | |
b4f48b63 | 5021 | } |
697f4161 | 5022 | |
bd89aabc | 5023 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5024 | { |
f3d46500 TH |
5025 | if (cgroup_is_releasable(cgrp) && list_empty(&cgrp->cset_links) && |
5026 | !css_has_online_children(&cgrp->self)) { | |
f50daa70 LZ |
5027 | /* |
5028 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 5029 | * already queued for a userspace notification, queue |
f50daa70 LZ |
5030 | * it now |
5031 | */ | |
81a6a5cd | 5032 | int need_schedule_work = 0; |
f50daa70 | 5033 | |
cdcc136f | 5034 | raw_spin_lock(&release_list_lock); |
54766d4a | 5035 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
5036 | list_empty(&cgrp->release_list)) { |
5037 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5038 | need_schedule_work = 1; |
5039 | } | |
cdcc136f | 5040 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5041 | if (need_schedule_work) |
5042 | schedule_work(&release_agent_work); | |
5043 | } | |
5044 | } | |
5045 | ||
81a6a5cd PM |
5046 | /* |
5047 | * Notify userspace when a cgroup is released, by running the | |
5048 | * configured release agent with the name of the cgroup (path | |
5049 | * relative to the root of cgroup file system) as the argument. | |
5050 | * | |
5051 | * Most likely, this user command will try to rmdir this cgroup. | |
5052 | * | |
5053 | * This races with the possibility that some other task will be | |
5054 | * attached to this cgroup before it is removed, or that some other | |
5055 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5056 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5057 | * unused, and this cgroup will be reprieved from its death sentence, | |
5058 | * to continue to serve a useful existence. Next time it's released, | |
5059 | * we will get notified again, if it still has 'notify_on_release' set. | |
5060 | * | |
5061 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5062 | * means only wait until the task is successfully execve()'d. The | |
5063 | * separate release agent task is forked by call_usermodehelper(), | |
5064 | * then control in this thread returns here, without waiting for the | |
5065 | * release agent task. We don't bother to wait because the caller of | |
5066 | * this routine has no use for the exit status of the release agent | |
5067 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5068 | */ |
81a6a5cd PM |
5069 | static void cgroup_release_agent(struct work_struct *work) |
5070 | { | |
5071 | BUG_ON(work != &release_agent_work); | |
5072 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5073 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5074 | while (!list_empty(&release_list)) { |
5075 | char *argv[3], *envp[3]; | |
5076 | int i; | |
e61734c5 | 5077 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 5078 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5079 | struct cgroup, |
5080 | release_list); | |
bd89aabc | 5081 | list_del_init(&cgrp->release_list); |
cdcc136f | 5082 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 5083 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
5084 | if (!pathbuf) |
5085 | goto continue_free; | |
e61734c5 TH |
5086 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
5087 | if (!path) | |
e788e066 PM |
5088 | goto continue_free; |
5089 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5090 | if (!agentbuf) | |
5091 | goto continue_free; | |
81a6a5cd PM |
5092 | |
5093 | i = 0; | |
e788e066 | 5094 | argv[i++] = agentbuf; |
e61734c5 | 5095 | argv[i++] = path; |
81a6a5cd PM |
5096 | argv[i] = NULL; |
5097 | ||
5098 | i = 0; | |
5099 | /* minimal command environment */ | |
5100 | envp[i++] = "HOME=/"; | |
5101 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5102 | envp[i] = NULL; | |
5103 | ||
5104 | /* Drop the lock while we invoke the usermode helper, | |
5105 | * since the exec could involve hitting disk and hence | |
5106 | * be a slow process */ | |
5107 | mutex_unlock(&cgroup_mutex); | |
5108 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5109 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5110 | continue_free: |
5111 | kfree(pathbuf); | |
5112 | kfree(agentbuf); | |
cdcc136f | 5113 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5114 | } |
cdcc136f | 5115 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5116 | mutex_unlock(&cgroup_mutex); |
5117 | } | |
8bab8dde PM |
5118 | |
5119 | static int __init cgroup_disable(char *str) | |
5120 | { | |
30159ec7 | 5121 | struct cgroup_subsys *ss; |
8bab8dde | 5122 | char *token; |
30159ec7 | 5123 | int i; |
8bab8dde PM |
5124 | |
5125 | while ((token = strsep(&str, ",")) != NULL) { | |
5126 | if (!*token) | |
5127 | continue; | |
be45c900 | 5128 | |
3ed80a62 | 5129 | for_each_subsys(ss, i) { |
8bab8dde PM |
5130 | if (!strcmp(token, ss->name)) { |
5131 | ss->disabled = 1; | |
5132 | printk(KERN_INFO "Disabling %s control group" | |
5133 | " subsystem\n", ss->name); | |
5134 | break; | |
5135 | } | |
5136 | } | |
5137 | } | |
5138 | return 1; | |
5139 | } | |
5140 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5141 | |
b77d7b60 | 5142 | /** |
ec903c0c | 5143 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5144 | * @dentry: directory dentry of interest |
5145 | * @ss: subsystem of interest | |
b77d7b60 | 5146 | * |
5a17f543 TH |
5147 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5148 | * to get the corresponding css and return it. If such css doesn't exist | |
5149 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5150 | */ |
ec903c0c TH |
5151 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5152 | struct cgroup_subsys *ss) | |
e5d1367f | 5153 | { |
2bd59d48 TH |
5154 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5155 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5156 | struct cgroup *cgrp; |
e5d1367f | 5157 | |
35cf0836 | 5158 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5159 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5160 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5161 | return ERR_PTR(-EBADF); |
5162 | ||
5a17f543 TH |
5163 | rcu_read_lock(); |
5164 | ||
2bd59d48 TH |
5165 | /* |
5166 | * This path doesn't originate from kernfs and @kn could already | |
5167 | * have been or be removed at any point. @kn->priv is RCU | |
cfc79d5b | 5168 | * protected for this access. See cgroup_rmdir() for details. |
2bd59d48 TH |
5169 | */ |
5170 | cgrp = rcu_dereference(kn->priv); | |
5171 | if (cgrp) | |
5172 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5173 | |
ec903c0c | 5174 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5175 | css = ERR_PTR(-ENOENT); |
5176 | ||
5177 | rcu_read_unlock(); | |
5178 | return css; | |
e5d1367f | 5179 | } |
e5d1367f | 5180 | |
1cb650b9 LZ |
5181 | /** |
5182 | * css_from_id - lookup css by id | |
5183 | * @id: the cgroup id | |
5184 | * @ss: cgroup subsys to be looked into | |
5185 | * | |
5186 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5187 | * Should be called under rcu_read_lock(). | |
5188 | */ | |
5189 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5190 | { | |
6fa4918d | 5191 | WARN_ON_ONCE(!rcu_read_lock_held()); |
15a4c835 | 5192 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5193 | } |
5194 | ||
fe693435 | 5195 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5196 | static struct cgroup_subsys_state * |
5197 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5198 | { |
5199 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5200 | ||
5201 | if (!css) | |
5202 | return ERR_PTR(-ENOMEM); | |
5203 | ||
5204 | return css; | |
5205 | } | |
5206 | ||
eb95419b | 5207 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5208 | { |
eb95419b | 5209 | kfree(css); |
fe693435 PM |
5210 | } |
5211 | ||
182446d0 TH |
5212 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5213 | struct cftype *cft) | |
fe693435 | 5214 | { |
182446d0 | 5215 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5216 | } |
5217 | ||
182446d0 TH |
5218 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5219 | struct cftype *cft) | |
fe693435 PM |
5220 | { |
5221 | return (u64)(unsigned long)current->cgroups; | |
5222 | } | |
5223 | ||
182446d0 | 5224 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5225 | struct cftype *cft) |
fe693435 PM |
5226 | { |
5227 | u64 count; | |
5228 | ||
5229 | rcu_read_lock(); | |
a8ad805c | 5230 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5231 | rcu_read_unlock(); |
5232 | return count; | |
5233 | } | |
5234 | ||
2da8ca82 | 5235 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5236 | { |
69d0206c | 5237 | struct cgrp_cset_link *link; |
5abb8855 | 5238 | struct css_set *cset; |
e61734c5 TH |
5239 | char *name_buf; |
5240 | ||
5241 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5242 | if (!name_buf) | |
5243 | return -ENOMEM; | |
7717f7ba | 5244 | |
96d365e0 | 5245 | down_read(&css_set_rwsem); |
7717f7ba | 5246 | rcu_read_lock(); |
5abb8855 | 5247 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5248 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5249 | struct cgroup *c = link->cgrp; |
7717f7ba | 5250 | |
a2dd4247 | 5251 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5252 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5253 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5254 | } |
5255 | rcu_read_unlock(); | |
96d365e0 | 5256 | up_read(&css_set_rwsem); |
e61734c5 | 5257 | kfree(name_buf); |
7717f7ba PM |
5258 | return 0; |
5259 | } | |
5260 | ||
5261 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5262 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5263 | { |
2da8ca82 | 5264 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5265 | struct cgrp_cset_link *link; |
7717f7ba | 5266 | |
96d365e0 | 5267 | down_read(&css_set_rwsem); |
182446d0 | 5268 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5269 | struct css_set *cset = link->cset; |
7717f7ba PM |
5270 | struct task_struct *task; |
5271 | int count = 0; | |
c7561128 | 5272 | |
5abb8855 | 5273 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5274 | |
5abb8855 | 5275 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5276 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5277 | goto overflow; | |
5278 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5279 | } | |
5280 | ||
5281 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5282 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5283 | goto overflow; | |
5284 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5285 | } |
c7561128 TH |
5286 | continue; |
5287 | overflow: | |
5288 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5289 | } |
96d365e0 | 5290 | up_read(&css_set_rwsem); |
7717f7ba PM |
5291 | return 0; |
5292 | } | |
5293 | ||
182446d0 | 5294 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5295 | { |
182446d0 | 5296 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5297 | } |
5298 | ||
5299 | static struct cftype debug_files[] = { | |
fe693435 PM |
5300 | { |
5301 | .name = "taskcount", | |
5302 | .read_u64 = debug_taskcount_read, | |
5303 | }, | |
5304 | ||
5305 | { | |
5306 | .name = "current_css_set", | |
5307 | .read_u64 = current_css_set_read, | |
5308 | }, | |
5309 | ||
5310 | { | |
5311 | .name = "current_css_set_refcount", | |
5312 | .read_u64 = current_css_set_refcount_read, | |
5313 | }, | |
5314 | ||
7717f7ba PM |
5315 | { |
5316 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5317 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5318 | }, |
5319 | ||
5320 | { | |
5321 | .name = "cgroup_css_links", | |
2da8ca82 | 5322 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5323 | }, |
5324 | ||
fe693435 PM |
5325 | { |
5326 | .name = "releasable", | |
5327 | .read_u64 = releasable_read, | |
5328 | }, | |
fe693435 | 5329 | |
4baf6e33 TH |
5330 | { } /* terminate */ |
5331 | }; | |
fe693435 | 5332 | |
073219e9 | 5333 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5334 | .css_alloc = debug_css_alloc, |
5335 | .css_free = debug_css_free, | |
4baf6e33 | 5336 | .base_cftypes = debug_files, |
fe693435 PM |
5337 | }; |
5338 | #endif /* CONFIG_CGROUP_DEBUG */ |