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> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
60063497 | 60 | #include <linux/atomic.h> |
e93ad19d | 61 | #include <linux/cpuset.h> |
a79a908f AK |
62 | #include <linux/proc_ns.h> |
63 | #include <linux/nsproxy.h> | |
64 | #include <linux/proc_ns.h> | |
bd1060a1 | 65 | #include <net/sock.h> |
ddbcc7e8 | 66 | |
b1a21367 TH |
67 | /* |
68 | * pidlists linger the following amount before being destroyed. The goal | |
69 | * is avoiding frequent destruction in the middle of consecutive read calls | |
70 | * Expiring in the middle is a performance problem not a correctness one. | |
71 | * 1 sec should be enough. | |
72 | */ | |
73 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
74 | ||
8d7e6fb0 TH |
75 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
76 | MAX_CFTYPE_NAME + 2) | |
77 | ||
e25e2cbb TH |
78 | /* |
79 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
80 | * hierarchy must be performed while holding it. | |
81 | * | |
f0d9a5f1 | 82 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 83 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 84 | * |
0e1d768f TH |
85 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
86 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 87 | */ |
2219449a TH |
88 | #ifdef CONFIG_PROVE_RCU |
89 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 90 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 91 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 92 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 93 | #else |
81a6a5cd | 94 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 95 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
96 | #endif |
97 | ||
6fa4918d | 98 | /* |
15a4c835 TH |
99 | * Protects cgroup_idr and css_idr so that IDs can be released without |
100 | * grabbing cgroup_mutex. | |
6fa4918d TH |
101 | */ |
102 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
103 | ||
34c06254 TH |
104 | /* |
105 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
106 | * against file removal/re-creation across css hiding. | |
107 | */ | |
108 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
109 | ||
69e943b7 TH |
110 | /* |
111 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
112 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
113 | */ | |
114 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 115 | |
1ed13287 TH |
116 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
117 | ||
8353da1f | 118 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
119 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
120 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 121 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 122 | |
e5fca243 TH |
123 | /* |
124 | * cgroup destruction makes heavy use of work items and there can be a lot | |
125 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
126 | * destruction work items don't end up filling up max_active of system_wq | |
127 | * which may lead to deadlock. | |
128 | */ | |
129 | static struct workqueue_struct *cgroup_destroy_wq; | |
130 | ||
b1a21367 TH |
131 | /* |
132 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
133 | * separate workqueue as flush domain. | |
134 | */ | |
135 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
136 | ||
3ed80a62 | 137 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 138 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 139 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
140 | #include <linux/cgroup_subsys.h> |
141 | }; | |
073219e9 TH |
142 | #undef SUBSYS |
143 | ||
144 | /* array of cgroup subsystem names */ | |
145 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
146 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
147 | #include <linux/cgroup_subsys.h> |
148 | }; | |
073219e9 | 149 | #undef SUBSYS |
ddbcc7e8 | 150 | |
49d1dc4b TH |
151 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
152 | #define SUBSYS(_x) \ | |
153 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
154 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
155 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
156 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
157 | #include <linux/cgroup_subsys.h> | |
158 | #undef SUBSYS | |
159 | ||
160 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
161 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
162 | #include <linux/cgroup_subsys.h> | |
163 | }; | |
164 | #undef SUBSYS | |
165 | ||
166 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
167 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
168 | #include <linux/cgroup_subsys.h> | |
169 | }; | |
170 | #undef SUBSYS | |
171 | ||
ddbcc7e8 | 172 | /* |
3dd06ffa | 173 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
174 | * unattached - it never has more than a single cgroup, and all tasks are |
175 | * part of that cgroup. | |
ddbcc7e8 | 176 | */ |
a2dd4247 | 177 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 178 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 179 | |
a2dd4247 TH |
180 | /* |
181 | * The default hierarchy always exists but is hidden until mounted for the | |
182 | * first time. This is for backward compatibility. | |
183 | */ | |
a7165264 | 184 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 185 | |
223ffb29 | 186 | /* Controllers blocked by the commandline in v1 */ |
6e5c8307 | 187 | static u16 cgroup_no_v1_mask; |
223ffb29 | 188 | |
5533e011 | 189 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 190 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 191 | |
f6d635ad TH |
192 | /* some controllers are implicitly enabled on the default hierarchy */ |
193 | static unsigned long cgrp_dfl_implicit_ss_mask; | |
194 | ||
ddbcc7e8 PM |
195 | /* The list of hierarchy roots */ |
196 | ||
9871bf95 TH |
197 | static LIST_HEAD(cgroup_roots); |
198 | static int cgroup_root_count; | |
ddbcc7e8 | 199 | |
3417ae1f | 200 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 201 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 202 | |
794611a1 | 203 | /* |
0cb51d71 TH |
204 | * Assign a monotonically increasing serial number to csses. It guarantees |
205 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
206 | * Also, as csses are always appended to the parent's ->children list, it | |
207 | * guarantees that sibling csses are always sorted in the ascending serial | |
208 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 209 | */ |
0cb51d71 | 210 | static u64 css_serial_nr_next = 1; |
794611a1 | 211 | |
cb4a3167 AS |
212 | /* |
213 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
214 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
215 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 216 | */ |
6e5c8307 TH |
217 | static u16 have_fork_callback __read_mostly; |
218 | static u16 have_exit_callback __read_mostly; | |
219 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 220 | |
a79a908f AK |
221 | /* cgroup namespace for init task */ |
222 | struct cgroup_namespace init_cgroup_ns = { | |
223 | .count = { .counter = 2, }, | |
224 | .user_ns = &init_user_ns, | |
225 | .ns.ops = &cgroupns_operations, | |
226 | .ns.inum = PROC_CGROUP_INIT_INO, | |
227 | .root_cset = &init_css_set, | |
228 | }; | |
229 | ||
7e47682e | 230 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 231 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 232 | |
67e9c74b | 233 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
234 | static struct cftype cgroup_dfl_base_files[]; |
235 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 236 | |
6e5c8307 | 237 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
945ba199 | 238 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
334c3679 TH |
239 | static int cgroup_apply_control(struct cgroup *cgrp); |
240 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 241 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 242 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
243 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
244 | struct cgroup_subsys *ss); | |
9d755d33 | 245 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 246 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
247 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
248 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 249 | bool is_add); |
42809dd4 | 250 | |
fc5ed1e9 TH |
251 | /** |
252 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
253 | * @ssid: subsys ID of interest | |
254 | * | |
255 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
256 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
257 | * is slower static_key_enabled() based test indexed by @ssid. | |
258 | */ | |
259 | static bool cgroup_ssid_enabled(int ssid) | |
260 | { | |
cfe02a8a AB |
261 | if (CGROUP_SUBSYS_COUNT == 0) |
262 | return false; | |
263 | ||
fc5ed1e9 TH |
264 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
265 | } | |
266 | ||
223ffb29 JW |
267 | static bool cgroup_ssid_no_v1(int ssid) |
268 | { | |
269 | return cgroup_no_v1_mask & (1 << ssid); | |
270 | } | |
271 | ||
9e10a130 TH |
272 | /** |
273 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
274 | * @cgrp: the cgroup of interest | |
275 | * | |
276 | * The default hierarchy is the v2 interface of cgroup and this function | |
277 | * can be used to test whether a cgroup is on the default hierarchy for | |
278 | * cases where a subsystem should behave differnetly depending on the | |
279 | * interface version. | |
280 | * | |
281 | * The set of behaviors which change on the default hierarchy are still | |
282 | * being determined and the mount option is prefixed with __DEVEL__. | |
283 | * | |
284 | * List of changed behaviors: | |
285 | * | |
286 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
287 | * and "name" are disallowed. | |
288 | * | |
289 | * - When mounting an existing superblock, mount options should match. | |
290 | * | |
291 | * - Remount is disallowed. | |
292 | * | |
293 | * - rename(2) is disallowed. | |
294 | * | |
295 | * - "tasks" is removed. Everything should be at process granularity. Use | |
296 | * "cgroup.procs" instead. | |
297 | * | |
298 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
299 | * recycled inbetween reads. | |
300 | * | |
301 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
302 | * notification mechanism will be implemented. | |
303 | * | |
304 | * - "cgroup.clone_children" is removed. | |
305 | * | |
306 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
307 | * and its descendants contain no task; otherwise, 1. The file also | |
308 | * generates kernfs notification which can be monitored through poll and | |
309 | * [di]notify when the value of the file changes. | |
310 | * | |
311 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
312 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
313 | * moved to an ancestor. | |
314 | * | |
315 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
316 | * masks of ancestors. | |
317 | * | |
318 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
319 | * is not created. | |
320 | * | |
321 | * - blkcg: blk-throttle becomes properly hierarchical. | |
322 | * | |
323 | * - debug: disallowed on the default hierarchy. | |
324 | */ | |
325 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
326 | { | |
327 | return cgrp->root == &cgrp_dfl_root; | |
328 | } | |
329 | ||
6fa4918d TH |
330 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
331 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
332 | gfp_t gfp_mask) | |
333 | { | |
334 | int ret; | |
335 | ||
336 | idr_preload(gfp_mask); | |
54504e97 | 337 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 338 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 339 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
340 | idr_preload_end(); |
341 | return ret; | |
342 | } | |
343 | ||
344 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
345 | { | |
346 | void *ret; | |
347 | ||
54504e97 | 348 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 349 | ret = idr_replace(idr, ptr, id); |
54504e97 | 350 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
351 | return ret; |
352 | } | |
353 | ||
354 | static void cgroup_idr_remove(struct idr *idr, int id) | |
355 | { | |
54504e97 | 356 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 357 | idr_remove(idr, id); |
54504e97 | 358 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
359 | } |
360 | ||
d51f39b0 TH |
361 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
362 | { | |
363 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
364 | ||
365 | if (parent_css) | |
366 | return container_of(parent_css, struct cgroup, self); | |
367 | return NULL; | |
368 | } | |
369 | ||
5531dc91 TH |
370 | /* subsystems visibly enabled on a cgroup */ |
371 | static u16 cgroup_control(struct cgroup *cgrp) | |
372 | { | |
373 | struct cgroup *parent = cgroup_parent(cgrp); | |
374 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
375 | ||
376 | if (parent) | |
377 | return parent->subtree_control; | |
378 | ||
379 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
380 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
381 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
382 | return root_ss_mask; |
383 | } | |
384 | ||
385 | /* subsystems enabled on a cgroup */ | |
386 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
387 | { | |
388 | struct cgroup *parent = cgroup_parent(cgrp); | |
389 | ||
390 | if (parent) | |
391 | return parent->subtree_ss_mask; | |
392 | ||
393 | return cgrp->root->subsys_mask; | |
394 | } | |
395 | ||
95109b62 TH |
396 | /** |
397 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
398 | * @cgrp: the cgroup of interest | |
9d800df1 | 399 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 400 | * |
ca8bdcaf TH |
401 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
402 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
403 | * the caller is responsible for pinning the returned css if it wants to | |
404 | * keep accessing it outside the said locks. This function may return | |
405 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
406 | */ |
407 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 408 | struct cgroup_subsys *ss) |
95109b62 | 409 | { |
ca8bdcaf | 410 | if (ss) |
aec25020 | 411 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 412 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 413 | else |
9d800df1 | 414 | return &cgrp->self; |
95109b62 | 415 | } |
42809dd4 | 416 | |
aec3dfcb TH |
417 | /** |
418 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
419 | * @cgrp: the cgroup of interest | |
9d800df1 | 420 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 421 | * |
d0f702e6 | 422 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
423 | * as the matching css of the nearest ancestor including self which has @ss |
424 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
425 | * function is guaranteed to return non-NULL css. | |
426 | */ | |
427 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
428 | struct cgroup_subsys *ss) | |
429 | { | |
430 | lockdep_assert_held(&cgroup_mutex); | |
431 | ||
432 | if (!ss) | |
9d800df1 | 433 | return &cgrp->self; |
aec3dfcb | 434 | |
eeecbd19 TH |
435 | /* |
436 | * This function is used while updating css associations and thus | |
5531dc91 | 437 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 438 | */ |
5531dc91 | 439 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 440 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
441 | if (!cgrp) |
442 | return NULL; | |
443 | } | |
aec3dfcb TH |
444 | |
445 | return cgroup_css(cgrp, ss); | |
95109b62 | 446 | } |
42809dd4 | 447 | |
eeecbd19 TH |
448 | /** |
449 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
450 | * @cgrp: the cgroup of interest | |
451 | * @ss: the subsystem of interest | |
452 | * | |
453 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
454 | * defined as the matching css of the nearest ancestor including self which | |
455 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
456 | * the root css is returned, so this function always returns a valid css. | |
457 | * The returned css must be put using css_put(). | |
458 | */ | |
459 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
460 | struct cgroup_subsys *ss) | |
461 | { | |
462 | struct cgroup_subsys_state *css; | |
463 | ||
464 | rcu_read_lock(); | |
465 | ||
466 | do { | |
467 | css = cgroup_css(cgrp, ss); | |
468 | ||
469 | if (css && css_tryget_online(css)) | |
470 | goto out_unlock; | |
471 | cgrp = cgroup_parent(cgrp); | |
472 | } while (cgrp); | |
473 | ||
474 | css = init_css_set.subsys[ss->id]; | |
475 | css_get(css); | |
476 | out_unlock: | |
477 | rcu_read_unlock(); | |
478 | return css; | |
479 | } | |
480 | ||
ddbcc7e8 | 481 | /* convenient tests for these bits */ |
54766d4a | 482 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 483 | { |
184faf32 | 484 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
485 | } |
486 | ||
052c3f3a TH |
487 | static void cgroup_get(struct cgroup *cgrp) |
488 | { | |
489 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
490 | css_get(&cgrp->self); | |
491 | } | |
492 | ||
493 | static bool cgroup_tryget(struct cgroup *cgrp) | |
494 | { | |
495 | return css_tryget(&cgrp->self); | |
496 | } | |
497 | ||
b4168640 | 498 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 499 | { |
2bd59d48 | 500 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 501 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
502 | |
503 | /* | |
504 | * This is open and unprotected implementation of cgroup_css(). | |
505 | * seq_css() is only called from a kernfs file operation which has | |
506 | * an active reference on the file. Because all the subsystem | |
507 | * files are drained before a css is disassociated with a cgroup, | |
508 | * the matching css from the cgroup's subsys table is guaranteed to | |
509 | * be and stay valid until the enclosing operation is complete. | |
510 | */ | |
511 | if (cft->ss) | |
512 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
513 | else | |
9d800df1 | 514 | return &cgrp->self; |
59f5296b | 515 | } |
b4168640 | 516 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 517 | |
e9685a03 | 518 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 519 | { |
bd89aabc | 520 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
521 | } |
522 | ||
1c6727af TH |
523 | /** |
524 | * for_each_css - iterate all css's of a cgroup | |
525 | * @css: the iteration cursor | |
526 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
527 | * @cgrp: the target cgroup to iterate css's of | |
528 | * | |
aec3dfcb | 529 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
530 | */ |
531 | #define for_each_css(css, ssid, cgrp) \ | |
532 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
533 | if (!((css) = rcu_dereference_check( \ | |
534 | (cgrp)->subsys[(ssid)], \ | |
535 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
536 | else | |
537 | ||
aec3dfcb TH |
538 | /** |
539 | * for_each_e_css - iterate all effective css's of a cgroup | |
540 | * @css: the iteration cursor | |
541 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
542 | * @cgrp: the target cgroup to iterate css's of | |
543 | * | |
544 | * Should be called under cgroup_[tree_]mutex. | |
545 | */ | |
546 | #define for_each_e_css(css, ssid, cgrp) \ | |
547 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
548 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
549 | ; \ | |
550 | else | |
551 | ||
30159ec7 | 552 | /** |
3ed80a62 | 553 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 554 | * @ss: the iteration cursor |
780cd8b3 | 555 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 556 | */ |
780cd8b3 | 557 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
558 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
559 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 560 | |
cb4a3167 | 561 | /** |
b4e0eeaf | 562 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
563 | * @ss: the iteration cursor |
564 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 565 | * @ss_mask: the bitmask |
cb4a3167 AS |
566 | * |
567 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 568 | * @ss_mask is set. |
cb4a3167 | 569 | */ |
b4e0eeaf TH |
570 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
571 | unsigned long __ss_mask = (ss_mask); \ | |
572 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 573 | (ssid) = 0; \ |
b4e0eeaf TH |
574 | break; \ |
575 | } \ | |
576 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
577 | (ss) = cgroup_subsys[ssid]; \ | |
578 | { | |
579 | ||
580 | #define while_each_subsys_mask() \ | |
581 | } \ | |
582 | } \ | |
583 | } while (false) | |
cb4a3167 | 584 | |
985ed670 TH |
585 | /* iterate across the hierarchies */ |
586 | #define for_each_root(root) \ | |
5549c497 | 587 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 588 | |
f8f22e53 TH |
589 | /* iterate over child cgrps, lock should be held throughout iteration */ |
590 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 591 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 592 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
593 | cgroup_is_dead(child); })) \ |
594 | ; \ | |
595 | else | |
7ae1bad9 | 596 | |
ce3f1d9d TH |
597 | /* walk live descendants in preorder */ |
598 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
599 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
600 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
601 | (dsct) = (d_css)->cgroup; \ | |
602 | cgroup_is_dead(dsct); })) \ | |
603 | ; \ | |
604 | else | |
605 | ||
606 | /* walk live descendants in postorder */ | |
607 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
608 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
609 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
610 | (dsct) = (d_css)->cgroup; \ | |
611 | cgroup_is_dead(dsct); })) \ | |
612 | ; \ | |
613 | else | |
614 | ||
81a6a5cd | 615 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 616 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 617 | |
69d0206c TH |
618 | /* |
619 | * A cgroup can be associated with multiple css_sets as different tasks may | |
620 | * belong to different cgroups on different hierarchies. In the other | |
621 | * direction, a css_set is naturally associated with multiple cgroups. | |
622 | * This M:N relationship is represented by the following link structure | |
623 | * which exists for each association and allows traversing the associations | |
624 | * from both sides. | |
625 | */ | |
626 | struct cgrp_cset_link { | |
627 | /* the cgroup and css_set this link associates */ | |
628 | struct cgroup *cgrp; | |
629 | struct css_set *cset; | |
630 | ||
631 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
632 | struct list_head cset_link; | |
633 | ||
634 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
635 | struct list_head cgrp_link; | |
817929ec PM |
636 | }; |
637 | ||
172a2c06 TH |
638 | /* |
639 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
640 | * hierarchies being mounted. It contains a pointer to the root state |
641 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
642 | * reference-counted, to improve performance when child cgroups | |
643 | * haven't been created. | |
644 | */ | |
5024ae29 | 645 | struct css_set init_css_set = { |
172a2c06 TH |
646 | .refcount = ATOMIC_INIT(1), |
647 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
648 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
649 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
650 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
651 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 652 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 653 | }; |
817929ec | 654 | |
172a2c06 | 655 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 656 | |
0de0942d TH |
657 | /** |
658 | * css_set_populated - does a css_set contain any tasks? | |
659 | * @cset: target css_set | |
660 | */ | |
661 | static bool css_set_populated(struct css_set *cset) | |
662 | { | |
f0d9a5f1 | 663 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
664 | |
665 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
666 | } | |
667 | ||
842b597e TH |
668 | /** |
669 | * cgroup_update_populated - updated populated count of a cgroup | |
670 | * @cgrp: the target cgroup | |
671 | * @populated: inc or dec populated count | |
672 | * | |
0de0942d TH |
673 | * One of the css_sets associated with @cgrp is either getting its first |
674 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
675 | * count is propagated towards root so that a given cgroup's populated_cnt | |
676 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
677 | * |
678 | * @cgrp's interface file "cgroup.populated" is zero if | |
679 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
680 | * changes from or to zero, userland is notified that the content of the | |
681 | * interface file has changed. This can be used to detect when @cgrp and | |
682 | * its descendants become populated or empty. | |
683 | */ | |
684 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
685 | { | |
f0d9a5f1 | 686 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
687 | |
688 | do { | |
689 | bool trigger; | |
690 | ||
691 | if (populated) | |
692 | trigger = !cgrp->populated_cnt++; | |
693 | else | |
694 | trigger = !--cgrp->populated_cnt; | |
695 | ||
696 | if (!trigger) | |
697 | break; | |
698 | ||
ad2ed2b3 | 699 | check_for_release(cgrp); |
6f60eade TH |
700 | cgroup_file_notify(&cgrp->events_file); |
701 | ||
d51f39b0 | 702 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
703 | } while (cgrp); |
704 | } | |
705 | ||
0de0942d TH |
706 | /** |
707 | * css_set_update_populated - update populated state of a css_set | |
708 | * @cset: target css_set | |
709 | * @populated: whether @cset is populated or depopulated | |
710 | * | |
711 | * @cset is either getting the first task or losing the last. Update the | |
712 | * ->populated_cnt of all associated cgroups accordingly. | |
713 | */ | |
714 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
715 | { | |
716 | struct cgrp_cset_link *link; | |
717 | ||
f0d9a5f1 | 718 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
719 | |
720 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
721 | cgroup_update_populated(link->cgrp, populated); | |
722 | } | |
723 | ||
f6d7d049 TH |
724 | /** |
725 | * css_set_move_task - move a task from one css_set to another | |
726 | * @task: task being moved | |
727 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
728 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
729 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
730 | * | |
731 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
732 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
733 | * instead of moved, @to_cset can be NULL. | |
734 | * | |
ed27b9f7 TH |
735 | * This function automatically handles populated_cnt updates and |
736 | * css_task_iter adjustments but the caller is responsible for managing | |
737 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
738 | */ |
739 | static void css_set_move_task(struct task_struct *task, | |
740 | struct css_set *from_cset, struct css_set *to_cset, | |
741 | bool use_mg_tasks) | |
742 | { | |
f0d9a5f1 | 743 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 744 | |
20b454a6 TH |
745 | if (to_cset && !css_set_populated(to_cset)) |
746 | css_set_update_populated(to_cset, true); | |
747 | ||
f6d7d049 | 748 | if (from_cset) { |
ed27b9f7 TH |
749 | struct css_task_iter *it, *pos; |
750 | ||
f6d7d049 | 751 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
752 | |
753 | /* | |
754 | * @task is leaving, advance task iterators which are | |
755 | * pointing to it so that they can resume at the next | |
756 | * position. Advancing an iterator might remove it from | |
757 | * the list, use safe walk. See css_task_iter_advance*() | |
758 | * for details. | |
759 | */ | |
760 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
761 | iters_node) | |
762 | if (it->task_pos == &task->cg_list) | |
763 | css_task_iter_advance(it); | |
764 | ||
f6d7d049 TH |
765 | list_del_init(&task->cg_list); |
766 | if (!css_set_populated(from_cset)) | |
767 | css_set_update_populated(from_cset, false); | |
768 | } else { | |
769 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
770 | } | |
771 | ||
772 | if (to_cset) { | |
773 | /* | |
774 | * We are synchronized through cgroup_threadgroup_rwsem | |
775 | * against PF_EXITING setting such that we can't race | |
776 | * against cgroup_exit() changing the css_set to | |
777 | * init_css_set and dropping the old one. | |
778 | */ | |
779 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
780 | ||
f6d7d049 TH |
781 | rcu_assign_pointer(task->cgroups, to_cset); |
782 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
783 | &to_cset->tasks); | |
784 | } | |
785 | } | |
786 | ||
7717f7ba PM |
787 | /* |
788 | * hash table for cgroup groups. This improves the performance to find | |
789 | * an existing css_set. This hash doesn't (currently) take into | |
790 | * account cgroups in empty hierarchies. | |
791 | */ | |
472b1053 | 792 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 793 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 794 | |
0ac801fe | 795 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 796 | { |
0ac801fe | 797 | unsigned long key = 0UL; |
30159ec7 TH |
798 | struct cgroup_subsys *ss; |
799 | int i; | |
472b1053 | 800 | |
30159ec7 | 801 | for_each_subsys(ss, i) |
0ac801fe LZ |
802 | key += (unsigned long)css[i]; |
803 | key = (key >> 16) ^ key; | |
472b1053 | 804 | |
0ac801fe | 805 | return key; |
472b1053 LZ |
806 | } |
807 | ||
a25eb52e | 808 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 809 | { |
69d0206c | 810 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
811 | struct cgroup_subsys *ss; |
812 | int ssid; | |
5abb8855 | 813 | |
f0d9a5f1 | 814 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
815 | |
816 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 817 | return; |
81a6a5cd | 818 | |
53254f90 TH |
819 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
820 | for_each_subsys(ss, ssid) { | |
2d8f243a | 821 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
822 | css_put(cset->subsys[ssid]); |
823 | } | |
5abb8855 | 824 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
825 | css_set_count--; |
826 | ||
69d0206c | 827 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
828 | list_del(&link->cset_link); |
829 | list_del(&link->cgrp_link); | |
2ceb231b TH |
830 | if (cgroup_parent(link->cgrp)) |
831 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 832 | kfree(link); |
81a6a5cd | 833 | } |
2c6ab6d2 | 834 | |
5abb8855 | 835 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
836 | } |
837 | ||
a25eb52e | 838 | static void put_css_set(struct css_set *cset) |
89c5509b | 839 | { |
82d6489d DBO |
840 | unsigned long flags; |
841 | ||
89c5509b TH |
842 | /* |
843 | * Ensure that the refcount doesn't hit zero while any readers | |
844 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
845 | * rwlock | |
846 | */ | |
847 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
848 | return; | |
849 | ||
82d6489d | 850 | spin_lock_irqsave(&css_set_lock, flags); |
a25eb52e | 851 | put_css_set_locked(cset); |
82d6489d | 852 | spin_unlock_irqrestore(&css_set_lock, flags); |
89c5509b TH |
853 | } |
854 | ||
817929ec PM |
855 | /* |
856 | * refcounted get/put for css_set objects | |
857 | */ | |
5abb8855 | 858 | static inline void get_css_set(struct css_set *cset) |
817929ec | 859 | { |
5abb8855 | 860 | atomic_inc(&cset->refcount); |
817929ec PM |
861 | } |
862 | ||
b326f9d0 | 863 | /** |
7717f7ba | 864 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
865 | * @cset: candidate css_set being tested |
866 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
867 | * @new_cgrp: cgroup that's being entered by the task |
868 | * @template: desired set of css pointers in css_set (pre-calculated) | |
869 | * | |
6f4b7e63 | 870 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
871 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
872 | */ | |
5abb8855 TH |
873 | static bool compare_css_sets(struct css_set *cset, |
874 | struct css_set *old_cset, | |
7717f7ba PM |
875 | struct cgroup *new_cgrp, |
876 | struct cgroup_subsys_state *template[]) | |
877 | { | |
878 | struct list_head *l1, *l2; | |
879 | ||
aec3dfcb TH |
880 | /* |
881 | * On the default hierarchy, there can be csets which are | |
882 | * associated with the same set of cgroups but different csses. | |
883 | * Let's first ensure that csses match. | |
884 | */ | |
885 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 886 | return false; |
7717f7ba PM |
887 | |
888 | /* | |
889 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
890 | * different cgroups in hierarchies. As different cgroups may |
891 | * share the same effective css, this comparison is always | |
892 | * necessary. | |
7717f7ba | 893 | */ |
69d0206c TH |
894 | l1 = &cset->cgrp_links; |
895 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 896 | while (1) { |
69d0206c | 897 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 898 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
899 | |
900 | l1 = l1->next; | |
901 | l2 = l2->next; | |
902 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
903 | if (l1 == &cset->cgrp_links) { |
904 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
905 | break; |
906 | } else { | |
69d0206c | 907 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
908 | } |
909 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
910 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
911 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
912 | cgrp1 = link1->cgrp; | |
913 | cgrp2 = link2->cgrp; | |
7717f7ba | 914 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 915 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
916 | |
917 | /* | |
918 | * If this hierarchy is the hierarchy of the cgroup | |
919 | * that's changing, then we need to check that this | |
920 | * css_set points to the new cgroup; if it's any other | |
921 | * hierarchy, then this css_set should point to the | |
922 | * same cgroup as the old css_set. | |
923 | */ | |
5abb8855 TH |
924 | if (cgrp1->root == new_cgrp->root) { |
925 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
926 | return false; |
927 | } else { | |
5abb8855 | 928 | if (cgrp1 != cgrp2) |
7717f7ba PM |
929 | return false; |
930 | } | |
931 | } | |
932 | return true; | |
933 | } | |
934 | ||
b326f9d0 TH |
935 | /** |
936 | * find_existing_css_set - init css array and find the matching css_set | |
937 | * @old_cset: the css_set that we're using before the cgroup transition | |
938 | * @cgrp: the cgroup that we're moving into | |
939 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 940 | */ |
5abb8855 TH |
941 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
942 | struct cgroup *cgrp, | |
943 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 944 | { |
3dd06ffa | 945 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 946 | struct cgroup_subsys *ss; |
5abb8855 | 947 | struct css_set *cset; |
0ac801fe | 948 | unsigned long key; |
b326f9d0 | 949 | int i; |
817929ec | 950 | |
aae8aab4 BB |
951 | /* |
952 | * Build the set of subsystem state objects that we want to see in the | |
953 | * new css_set. while subsystems can change globally, the entries here | |
954 | * won't change, so no need for locking. | |
955 | */ | |
30159ec7 | 956 | for_each_subsys(ss, i) { |
f392e51c | 957 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
958 | /* |
959 | * @ss is in this hierarchy, so we want the | |
960 | * effective css from @cgrp. | |
961 | */ | |
962 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 963 | } else { |
aec3dfcb TH |
964 | /* |
965 | * @ss is not in this hierarchy, so we don't want | |
966 | * to change the css. | |
967 | */ | |
5abb8855 | 968 | template[i] = old_cset->subsys[i]; |
817929ec PM |
969 | } |
970 | } | |
971 | ||
0ac801fe | 972 | key = css_set_hash(template); |
5abb8855 TH |
973 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
974 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
975 | continue; |
976 | ||
977 | /* This css_set matches what we need */ | |
5abb8855 | 978 | return cset; |
472b1053 | 979 | } |
817929ec PM |
980 | |
981 | /* No existing cgroup group matched */ | |
982 | return NULL; | |
983 | } | |
984 | ||
69d0206c | 985 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 986 | { |
69d0206c | 987 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 988 | |
69d0206c TH |
989 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
990 | list_del(&link->cset_link); | |
36553434 LZ |
991 | kfree(link); |
992 | } | |
993 | } | |
994 | ||
69d0206c TH |
995 | /** |
996 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
997 | * @count: the number of links to allocate | |
998 | * @tmp_links: list_head the allocated links are put on | |
999 | * | |
1000 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1001 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1002 | */ |
69d0206c | 1003 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1004 | { |
69d0206c | 1005 | struct cgrp_cset_link *link; |
817929ec | 1006 | int i; |
69d0206c TH |
1007 | |
1008 | INIT_LIST_HEAD(tmp_links); | |
1009 | ||
817929ec | 1010 | for (i = 0; i < count; i++) { |
f4f4be2b | 1011 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1012 | if (!link) { |
69d0206c | 1013 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1014 | return -ENOMEM; |
1015 | } | |
69d0206c | 1016 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1017 | } |
1018 | return 0; | |
1019 | } | |
1020 | ||
c12f65d4 LZ |
1021 | /** |
1022 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1023 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1024 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1025 | * @cgrp: the destination cgroup |
1026 | */ | |
69d0206c TH |
1027 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1028 | struct cgroup *cgrp) | |
c12f65d4 | 1029 | { |
69d0206c | 1030 | struct cgrp_cset_link *link; |
c12f65d4 | 1031 | |
69d0206c | 1032 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1033 | |
1034 | if (cgroup_on_dfl(cgrp)) | |
1035 | cset->dfl_cgrp = cgrp; | |
1036 | ||
69d0206c TH |
1037 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1038 | link->cset = cset; | |
7717f7ba | 1039 | link->cgrp = cgrp; |
842b597e | 1040 | |
7717f7ba | 1041 | /* |
389b9c1b TH |
1042 | * Always add links to the tail of the lists so that the lists are |
1043 | * in choronological order. | |
7717f7ba | 1044 | */ |
389b9c1b | 1045 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1046 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1047 | |
1048 | if (cgroup_parent(cgrp)) | |
1049 | cgroup_get(cgrp); | |
c12f65d4 LZ |
1050 | } |
1051 | ||
b326f9d0 TH |
1052 | /** |
1053 | * find_css_set - return a new css_set with one cgroup updated | |
1054 | * @old_cset: the baseline css_set | |
1055 | * @cgrp: the cgroup to be updated | |
1056 | * | |
1057 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1058 | * substituted into the appropriate hierarchy. | |
817929ec | 1059 | */ |
5abb8855 TH |
1060 | static struct css_set *find_css_set(struct css_set *old_cset, |
1061 | struct cgroup *cgrp) | |
817929ec | 1062 | { |
b326f9d0 | 1063 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1064 | struct css_set *cset; |
69d0206c TH |
1065 | struct list_head tmp_links; |
1066 | struct cgrp_cset_link *link; | |
2d8f243a | 1067 | struct cgroup_subsys *ss; |
0ac801fe | 1068 | unsigned long key; |
2d8f243a | 1069 | int ssid; |
472b1053 | 1070 | |
b326f9d0 TH |
1071 | lockdep_assert_held(&cgroup_mutex); |
1072 | ||
817929ec PM |
1073 | /* First see if we already have a cgroup group that matches |
1074 | * the desired set */ | |
82d6489d | 1075 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1076 | cset = find_existing_css_set(old_cset, cgrp, template); |
1077 | if (cset) | |
1078 | get_css_set(cset); | |
82d6489d | 1079 | spin_unlock_irq(&css_set_lock); |
817929ec | 1080 | |
5abb8855 TH |
1081 | if (cset) |
1082 | return cset; | |
817929ec | 1083 | |
f4f4be2b | 1084 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1085 | if (!cset) |
817929ec PM |
1086 | return NULL; |
1087 | ||
69d0206c | 1088 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1089 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1090 | kfree(cset); |
817929ec PM |
1091 | return NULL; |
1092 | } | |
1093 | ||
5abb8855 | 1094 | atomic_set(&cset->refcount, 1); |
69d0206c | 1095 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1096 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1097 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1098 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1099 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1100 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1101 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1102 | |
1103 | /* Copy the set of subsystem state objects generated in | |
1104 | * find_existing_css_set() */ | |
5abb8855 | 1105 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1106 | |
82d6489d | 1107 | spin_lock_irq(&css_set_lock); |
817929ec | 1108 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1109 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1110 | struct cgroup *c = link->cgrp; |
69d0206c | 1111 | |
7717f7ba PM |
1112 | if (c->root == cgrp->root) |
1113 | c = cgrp; | |
69d0206c | 1114 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1115 | } |
817929ec | 1116 | |
69d0206c | 1117 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1118 | |
817929ec | 1119 | css_set_count++; |
472b1053 | 1120 | |
2d8f243a | 1121 | /* Add @cset to the hash table */ |
5abb8855 TH |
1122 | key = css_set_hash(cset->subsys); |
1123 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1124 | |
53254f90 TH |
1125 | for_each_subsys(ss, ssid) { |
1126 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1127 | ||
2d8f243a | 1128 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1129 | &css->cgroup->e_csets[ssid]); |
1130 | css_get(css); | |
1131 | } | |
2d8f243a | 1132 | |
82d6489d | 1133 | spin_unlock_irq(&css_set_lock); |
817929ec | 1134 | |
5abb8855 | 1135 | return cset; |
b4f48b63 PM |
1136 | } |
1137 | ||
3dd06ffa | 1138 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1139 | { |
3dd06ffa | 1140 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1141 | |
3dd06ffa | 1142 | return root_cgrp->root; |
2bd59d48 TH |
1143 | } |
1144 | ||
3dd06ffa | 1145 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1146 | { |
1147 | int id; | |
1148 | ||
1149 | lockdep_assert_held(&cgroup_mutex); | |
1150 | ||
985ed670 | 1151 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1152 | if (id < 0) |
1153 | return id; | |
1154 | ||
1155 | root->hierarchy_id = id; | |
1156 | return 0; | |
1157 | } | |
1158 | ||
3dd06ffa | 1159 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1160 | { |
1161 | lockdep_assert_held(&cgroup_mutex); | |
1162 | ||
1163 | if (root->hierarchy_id) { | |
1164 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1165 | root->hierarchy_id = 0; | |
1166 | } | |
1167 | } | |
1168 | ||
3dd06ffa | 1169 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1170 | { |
1171 | if (root) { | |
d0f702e6 | 1172 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1173 | WARN_ON_ONCE(root->hierarchy_id); |
1174 | ||
1175 | idr_destroy(&root->cgroup_idr); | |
1176 | kfree(root); | |
1177 | } | |
1178 | } | |
1179 | ||
3dd06ffa | 1180 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1181 | { |
3dd06ffa | 1182 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1183 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1184 | |
334c3679 | 1185 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1186 | |
776f02fa | 1187 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1188 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1189 | |
f2e85d57 | 1190 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1191 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1192 | |
7717f7ba | 1193 | /* |
f2e85d57 TH |
1194 | * Release all the links from cset_links to this hierarchy's |
1195 | * root cgroup | |
7717f7ba | 1196 | */ |
82d6489d | 1197 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1198 | |
1199 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1200 | list_del(&link->cset_link); | |
1201 | list_del(&link->cgrp_link); | |
1202 | kfree(link); | |
1203 | } | |
f0d9a5f1 | 1204 | |
82d6489d | 1205 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1206 | |
1207 | if (!list_empty(&root->root_list)) { | |
1208 | list_del(&root->root_list); | |
1209 | cgroup_root_count--; | |
1210 | } | |
1211 | ||
1212 | cgroup_exit_root_id(root); | |
1213 | ||
1214 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1215 | |
2bd59d48 | 1216 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1217 | cgroup_free_root(root); |
1218 | } | |
1219 | ||
4f41fc59 SH |
1220 | /* |
1221 | * look up cgroup associated with current task's cgroup namespace on the | |
1222 | * specified hierarchy | |
1223 | */ | |
1224 | static struct cgroup * | |
1225 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1226 | { | |
1227 | struct cgroup *res = NULL; | |
1228 | struct css_set *cset; | |
1229 | ||
1230 | lockdep_assert_held(&css_set_lock); | |
1231 | ||
1232 | rcu_read_lock(); | |
1233 | ||
1234 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1235 | if (cset == &init_css_set) { | |
1236 | res = &root->cgrp; | |
1237 | } else { | |
1238 | struct cgrp_cset_link *link; | |
1239 | ||
1240 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1241 | struct cgroup *c = link->cgrp; | |
1242 | ||
1243 | if (c->root == root) { | |
1244 | res = c; | |
1245 | break; | |
1246 | } | |
1247 | } | |
1248 | } | |
1249 | rcu_read_unlock(); | |
1250 | ||
1251 | BUG_ON(!res); | |
1252 | return res; | |
1253 | } | |
1254 | ||
ceb6a081 TH |
1255 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1256 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1257 | struct cgroup_root *root) |
7717f7ba | 1258 | { |
7717f7ba PM |
1259 | struct cgroup *res = NULL; |
1260 | ||
96d365e0 | 1261 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1262 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1263 | |
5abb8855 | 1264 | if (cset == &init_css_set) { |
3dd06ffa | 1265 | res = &root->cgrp; |
7717f7ba | 1266 | } else { |
69d0206c TH |
1267 | struct cgrp_cset_link *link; |
1268 | ||
1269 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1270 | struct cgroup *c = link->cgrp; |
69d0206c | 1271 | |
7717f7ba PM |
1272 | if (c->root == root) { |
1273 | res = c; | |
1274 | break; | |
1275 | } | |
1276 | } | |
1277 | } | |
96d365e0 | 1278 | |
7717f7ba PM |
1279 | BUG_ON(!res); |
1280 | return res; | |
1281 | } | |
1282 | ||
ddbcc7e8 | 1283 | /* |
ceb6a081 | 1284 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1285 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1286 | */ |
1287 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1288 | struct cgroup_root *root) |
ceb6a081 TH |
1289 | { |
1290 | /* | |
1291 | * No need to lock the task - since we hold cgroup_mutex the | |
1292 | * task can't change groups, so the only thing that can happen | |
1293 | * is that it exits and its css is set back to init_css_set. | |
1294 | */ | |
1295 | return cset_cgroup_from_root(task_css_set(task), root); | |
1296 | } | |
1297 | ||
ddbcc7e8 | 1298 | /* |
ddbcc7e8 PM |
1299 | * A task must hold cgroup_mutex to modify cgroups. |
1300 | * | |
1301 | * Any task can increment and decrement the count field without lock. | |
1302 | * So in general, code holding cgroup_mutex can't rely on the count | |
1303 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1304 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1305 | * means that no tasks are currently attached, therefore there is no |
1306 | * way a task attached to that cgroup can fork (the other way to | |
1307 | * increment the count). So code holding cgroup_mutex can safely | |
1308 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1309 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1310 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1311 | * needs that mutex. | |
1312 | * | |
ddbcc7e8 PM |
1313 | * A cgroup can only be deleted if both its 'count' of using tasks |
1314 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1315 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1316 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1317 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1318 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1319 | * |
1320 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1321 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1322 | */ |
1323 | ||
2bd59d48 | 1324 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1325 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1326 | |
8d7e6fb0 TH |
1327 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1328 | char *buf) | |
ddbcc7e8 | 1329 | { |
3e1d2eed TH |
1330 | struct cgroup_subsys *ss = cft->ss; |
1331 | ||
8d7e6fb0 TH |
1332 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1333 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1334 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1335 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1336 | cft->name); | |
8d7e6fb0 TH |
1337 | else |
1338 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1339 | return buf; | |
ddbcc7e8 PM |
1340 | } |
1341 | ||
f2e85d57 TH |
1342 | /** |
1343 | * cgroup_file_mode - deduce file mode of a control file | |
1344 | * @cft: the control file in question | |
1345 | * | |
7dbdb199 | 1346 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1347 | */ |
1348 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1349 | { |
f2e85d57 | 1350 | umode_t mode = 0; |
65dff759 | 1351 | |
f2e85d57 TH |
1352 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1353 | mode |= S_IRUGO; | |
1354 | ||
7dbdb199 TH |
1355 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1356 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1357 | mode |= S_IWUGO; | |
1358 | else | |
1359 | mode |= S_IWUSR; | |
1360 | } | |
f2e85d57 TH |
1361 | |
1362 | return mode; | |
65dff759 LZ |
1363 | } |
1364 | ||
af0ba678 | 1365 | /** |
8699b776 | 1366 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1367 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1368 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1369 | * |
1370 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1371 | * enabled together through its ->depends_on mask. In such cases, more | |
1372 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1373 | * | |
0f060deb | 1374 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1375 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1376 | */ |
5ced2518 | 1377 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1378 | { |
6e5c8307 | 1379 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1380 | struct cgroup_subsys *ss; |
1381 | int ssid; | |
1382 | ||
1383 | lockdep_assert_held(&cgroup_mutex); | |
1384 | ||
f6d635ad TH |
1385 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1386 | ||
af0ba678 | 1387 | while (true) { |
6e5c8307 | 1388 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1389 | |
b4e0eeaf | 1390 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1391 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1392 | } while_each_subsys_mask(); |
af0ba678 TH |
1393 | |
1394 | /* | |
1395 | * Mask out subsystems which aren't available. This can | |
1396 | * happen only if some depended-upon subsystems were bound | |
1397 | * to non-default hierarchies. | |
1398 | */ | |
5ced2518 | 1399 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1400 | |
1401 | if (new_ss_mask == cur_ss_mask) | |
1402 | break; | |
1403 | cur_ss_mask = new_ss_mask; | |
1404 | } | |
1405 | ||
0f060deb TH |
1406 | return cur_ss_mask; |
1407 | } | |
1408 | ||
a9746d8d TH |
1409 | /** |
1410 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1411 | * @kn: the kernfs_node being serviced | |
1412 | * | |
1413 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1414 | * the method finishes if locking succeeded. Note that once this function | |
1415 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1416 | * inaccessible any time. If the caller intends to continue to access the | |
1417 | * cgroup, it should pin it before invoking this function. | |
1418 | */ | |
1419 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1420 | { |
a9746d8d TH |
1421 | struct cgroup *cgrp; |
1422 | ||
1423 | if (kernfs_type(kn) == KERNFS_DIR) | |
1424 | cgrp = kn->priv; | |
1425 | else | |
1426 | cgrp = kn->parent->priv; | |
1427 | ||
1428 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1429 | |
1430 | kernfs_unbreak_active_protection(kn); | |
1431 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1432 | } |
1433 | ||
a9746d8d TH |
1434 | /** |
1435 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1436 | * @kn: the kernfs_node being serviced | |
945ba199 | 1437 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1438 | * |
1439 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1440 | * @kn. It breaks the active protection, performs cgroup locking and | |
1441 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1442 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1443 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1444 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1445 | * |
1446 | * Any cgroup kernfs method implementation which requires locking the | |
1447 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1448 | * locking under kernfs active protection and allows all kernfs operations | |
1449 | * including self-removal. | |
1450 | */ | |
945ba199 TH |
1451 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1452 | bool drain_offline) | |
05ef1d7c | 1453 | { |
a9746d8d TH |
1454 | struct cgroup *cgrp; |
1455 | ||
1456 | if (kernfs_type(kn) == KERNFS_DIR) | |
1457 | cgrp = kn->priv; | |
1458 | else | |
1459 | cgrp = kn->parent->priv; | |
05ef1d7c | 1460 | |
2739d3cc | 1461 | /* |
01f6474c | 1462 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1463 | * active_ref. cgroup liveliness check alone provides enough |
1464 | * protection against removal. Ensure @cgrp stays accessible and | |
1465 | * break the active_ref protection. | |
2739d3cc | 1466 | */ |
aa32362f LZ |
1467 | if (!cgroup_tryget(cgrp)) |
1468 | return NULL; | |
a9746d8d TH |
1469 | kernfs_break_active_protection(kn); |
1470 | ||
945ba199 TH |
1471 | if (drain_offline) |
1472 | cgroup_lock_and_drain_offline(cgrp); | |
1473 | else | |
1474 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1475 | |
a9746d8d TH |
1476 | if (!cgroup_is_dead(cgrp)) |
1477 | return cgrp; | |
1478 | ||
1479 | cgroup_kn_unlock(kn); | |
1480 | return NULL; | |
ddbcc7e8 | 1481 | } |
05ef1d7c | 1482 | |
2739d3cc | 1483 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1484 | { |
2bd59d48 | 1485 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1486 | |
01f6474c | 1487 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1488 | |
1489 | if (cft->file_offset) { | |
1490 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1491 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1492 | ||
1493 | spin_lock_irq(&cgroup_file_kn_lock); | |
1494 | cfile->kn = NULL; | |
1495 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1496 | } | |
1497 | ||
2bd59d48 | 1498 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1499 | } |
1500 | ||
13af07df | 1501 | /** |
4df8dc90 TH |
1502 | * css_clear_dir - remove subsys files in a cgroup directory |
1503 | * @css: taget css | |
13af07df | 1504 | */ |
334c3679 | 1505 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1506 | { |
334c3679 | 1507 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1508 | struct cftype *cfts; |
05ef1d7c | 1509 | |
88cb04b9 TH |
1510 | if (!(css->flags & CSS_VISIBLE)) |
1511 | return; | |
1512 | ||
1513 | css->flags &= ~CSS_VISIBLE; | |
1514 | ||
4df8dc90 TH |
1515 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1516 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1517 | } |
1518 | ||
ccdca218 | 1519 | /** |
4df8dc90 TH |
1520 | * css_populate_dir - create subsys files in a cgroup directory |
1521 | * @css: target css | |
ccdca218 TH |
1522 | * |
1523 | * On failure, no file is added. | |
1524 | */ | |
334c3679 | 1525 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1526 | { |
334c3679 | 1527 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1528 | struct cftype *cfts, *failed_cfts; |
1529 | int ret; | |
ccdca218 | 1530 | |
03970d3c | 1531 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1532 | return 0; |
1533 | ||
4df8dc90 TH |
1534 | if (!css->ss) { |
1535 | if (cgroup_on_dfl(cgrp)) | |
1536 | cfts = cgroup_dfl_base_files; | |
1537 | else | |
1538 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1539 | |
4df8dc90 TH |
1540 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1541 | } | |
ccdca218 | 1542 | |
4df8dc90 TH |
1543 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1544 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1545 | if (ret < 0) { | |
1546 | failed_cfts = cfts; | |
1547 | goto err; | |
ccdca218 TH |
1548 | } |
1549 | } | |
88cb04b9 TH |
1550 | |
1551 | css->flags |= CSS_VISIBLE; | |
1552 | ||
ccdca218 TH |
1553 | return 0; |
1554 | err: | |
4df8dc90 TH |
1555 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1556 | if (cfts == failed_cfts) | |
1557 | break; | |
1558 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1559 | } | |
ccdca218 TH |
1560 | return ret; |
1561 | } | |
1562 | ||
6e5c8307 | 1563 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1564 | { |
1ada4838 | 1565 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1566 | struct cgroup_subsys *ss; |
2d8f243a | 1567 | int ssid, i, ret; |
ddbcc7e8 | 1568 | |
ace2bee8 | 1569 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1570 | |
b4e0eeaf | 1571 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1572 | /* |
1573 | * If @ss has non-root csses attached to it, can't move. | |
1574 | * If @ss is an implicit controller, it is exempt from this | |
1575 | * rule and can be stolen. | |
1576 | */ | |
1577 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1578 | !ss->implicit_on_dfl) | |
3ed80a62 | 1579 | return -EBUSY; |
1d5be6b2 | 1580 | |
5df36032 | 1581 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1582 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1583 | return -EBUSY; |
b4e0eeaf | 1584 | } while_each_subsys_mask(); |
ddbcc7e8 | 1585 | |
b4e0eeaf | 1586 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1587 | struct cgroup_root *src_root = ss->root; |
1588 | struct cgroup *scgrp = &src_root->cgrp; | |
1589 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1590 | struct css_set *cset; |
a8a648c4 | 1591 | |
1ada4838 | 1592 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1593 | |
334c3679 TH |
1594 | /* disable from the source */ |
1595 | src_root->subsys_mask &= ~(1 << ssid); | |
1596 | WARN_ON(cgroup_apply_control(scgrp)); | |
1597 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1598 | |
334c3679 | 1599 | /* rebind */ |
1ada4838 TH |
1600 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1601 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1602 | ss->root = dst_root; |
1ada4838 | 1603 | css->cgroup = dcgrp; |
73e80ed8 | 1604 | |
82d6489d | 1605 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1606 | hash_for_each(css_set_table, i, cset, hlist) |
1607 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1608 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1609 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1610 | |
bd53d617 | 1611 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1612 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1613 | if (dst_root == &cgrp_dfl_root) { |
1614 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1615 | } else { | |
1ada4838 | 1616 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1617 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1618 | } |
a8a648c4 | 1619 | |
334c3679 TH |
1620 | ret = cgroup_apply_control(dcgrp); |
1621 | if (ret) | |
1622 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1623 | ss->name, ret); | |
1624 | ||
5df36032 TH |
1625 | if (ss->bind) |
1626 | ss->bind(css); | |
b4e0eeaf | 1627 | } while_each_subsys_mask(); |
ddbcc7e8 | 1628 | |
1ada4838 | 1629 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1630 | return 0; |
1631 | } | |
1632 | ||
4f41fc59 SH |
1633 | static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1634 | struct kernfs_root *kf_root) | |
1635 | { | |
09be4c82 | 1636 | int len = 0; |
4f41fc59 SH |
1637 | char *buf = NULL; |
1638 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1639 | struct cgroup *ns_cgroup; | |
1640 | ||
1641 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1642 | if (!buf) | |
1643 | return -ENOMEM; | |
1644 | ||
82d6489d | 1645 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1646 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1647 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1648 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1649 | |
1650 | if (len >= PATH_MAX) | |
1651 | len = -ERANGE; | |
1652 | else if (len > 0) { | |
1653 | seq_escape(sf, buf, " \t\n\\"); | |
1654 | len = 0; | |
1655 | } | |
1656 | kfree(buf); | |
1657 | return len; | |
1658 | } | |
1659 | ||
2bd59d48 TH |
1660 | static int cgroup_show_options(struct seq_file *seq, |
1661 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1662 | { |
3dd06ffa | 1663 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1664 | struct cgroup_subsys *ss; |
b85d2040 | 1665 | int ssid; |
ddbcc7e8 | 1666 | |
d98817d4 TH |
1667 | if (root != &cgrp_dfl_root) |
1668 | for_each_subsys(ss, ssid) | |
1669 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1670 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1671 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1672 | seq_puts(seq, ",noprefix"); |
93438629 | 1673 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1674 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1675 | |
1676 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1677 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1678 | seq_show_option(seq, "release_agent", |
1679 | root->release_agent_path); | |
69e943b7 TH |
1680 | spin_unlock(&release_agent_path_lock); |
1681 | ||
3dd06ffa | 1682 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1683 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1684 | if (strlen(root->name)) |
a068acf2 | 1685 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1686 | return 0; |
1687 | } | |
1688 | ||
1689 | struct cgroup_sb_opts { | |
6e5c8307 | 1690 | u16 subsys_mask; |
69dfa00c | 1691 | unsigned int flags; |
81a6a5cd | 1692 | char *release_agent; |
2260e7fc | 1693 | bool cpuset_clone_children; |
c6d57f33 | 1694 | char *name; |
2c6ab6d2 PM |
1695 | /* User explicitly requested empty subsystem */ |
1696 | bool none; | |
ddbcc7e8 PM |
1697 | }; |
1698 | ||
cf5d5941 | 1699 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1700 | { |
32a8cf23 DL |
1701 | char *token, *o = data; |
1702 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1703 | u16 mask = U16_MAX; |
30159ec7 | 1704 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1705 | int nr_opts = 0; |
30159ec7 | 1706 | int i; |
f9ab5b5b LZ |
1707 | |
1708 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1709 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1710 | #endif |
ddbcc7e8 | 1711 | |
c6d57f33 | 1712 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1713 | |
1714 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1715 | nr_opts++; |
1716 | ||
ddbcc7e8 PM |
1717 | if (!*token) |
1718 | return -EINVAL; | |
32a8cf23 | 1719 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1720 | /* Explicitly have no subsystems */ |
1721 | opts->none = true; | |
32a8cf23 DL |
1722 | continue; |
1723 | } | |
1724 | if (!strcmp(token, "all")) { | |
1725 | /* Mutually exclusive option 'all' + subsystem name */ | |
1726 | if (one_ss) | |
1727 | return -EINVAL; | |
1728 | all_ss = true; | |
1729 | continue; | |
1730 | } | |
1731 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1732 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1733 | continue; |
1734 | } | |
1735 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1736 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1737 | continue; |
1738 | } | |
03b1cde6 | 1739 | if (!strcmp(token, "xattr")) { |
93438629 | 1740 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1741 | continue; |
1742 | } | |
32a8cf23 | 1743 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1744 | /* Specifying two release agents is forbidden */ |
1745 | if (opts->release_agent) | |
1746 | return -EINVAL; | |
c6d57f33 | 1747 | opts->release_agent = |
e400c285 | 1748 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1749 | if (!opts->release_agent) |
1750 | return -ENOMEM; | |
32a8cf23 DL |
1751 | continue; |
1752 | } | |
1753 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1754 | const char *name = token + 5; |
1755 | /* Can't specify an empty name */ | |
1756 | if (!strlen(name)) | |
1757 | return -EINVAL; | |
1758 | /* Must match [\w.-]+ */ | |
1759 | for (i = 0; i < strlen(name); i++) { | |
1760 | char c = name[i]; | |
1761 | if (isalnum(c)) | |
1762 | continue; | |
1763 | if ((c == '.') || (c == '-') || (c == '_')) | |
1764 | continue; | |
1765 | return -EINVAL; | |
1766 | } | |
1767 | /* Specifying two names is forbidden */ | |
1768 | if (opts->name) | |
1769 | return -EINVAL; | |
1770 | opts->name = kstrndup(name, | |
e400c285 | 1771 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1772 | GFP_KERNEL); |
1773 | if (!opts->name) | |
1774 | return -ENOMEM; | |
32a8cf23 DL |
1775 | |
1776 | continue; | |
1777 | } | |
1778 | ||
30159ec7 | 1779 | for_each_subsys(ss, i) { |
3e1d2eed | 1780 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1781 | continue; |
fc5ed1e9 | 1782 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1783 | continue; |
223ffb29 JW |
1784 | if (cgroup_ssid_no_v1(i)) |
1785 | continue; | |
32a8cf23 DL |
1786 | |
1787 | /* Mutually exclusive option 'all' + subsystem name */ | |
1788 | if (all_ss) | |
1789 | return -EINVAL; | |
69dfa00c | 1790 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1791 | one_ss = true; |
1792 | ||
1793 | break; | |
1794 | } | |
1795 | if (i == CGROUP_SUBSYS_COUNT) | |
1796 | return -ENOENT; | |
1797 | } | |
1798 | ||
7b9a6ba5 TH |
1799 | /* |
1800 | * If the 'all' option was specified select all the subsystems, | |
1801 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1802 | * not specified, let's default to 'all' | |
1803 | */ | |
1804 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1805 | for_each_subsys(ss, i) | |
223ffb29 | 1806 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1807 | opts->subsys_mask |= (1 << i); |
1808 | ||
1809 | /* | |
1810 | * We either have to specify by name or by subsystems. (So all | |
1811 | * empty hierarchies must have a name). | |
1812 | */ | |
1813 | if (!opts->subsys_mask && !opts->name) | |
1814 | return -EINVAL; | |
1815 | ||
f9ab5b5b LZ |
1816 | /* |
1817 | * Option noprefix was introduced just for backward compatibility | |
1818 | * with the old cpuset, so we allow noprefix only if mounting just | |
1819 | * the cpuset subsystem. | |
1820 | */ | |
93438629 | 1821 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1822 | return -EINVAL; |
1823 | ||
2c6ab6d2 | 1824 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1825 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1826 | return -EINVAL; |
1827 | ||
ddbcc7e8 PM |
1828 | return 0; |
1829 | } | |
1830 | ||
2bd59d48 | 1831 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1832 | { |
1833 | int ret = 0; | |
3dd06ffa | 1834 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1835 | struct cgroup_sb_opts opts; |
6e5c8307 | 1836 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1837 | |
aa6ec29b TH |
1838 | if (root == &cgrp_dfl_root) { |
1839 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1840 | return -EINVAL; |
1841 | } | |
1842 | ||
334c3679 | 1843 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1844 | |
1845 | /* See what subsystems are wanted */ | |
1846 | ret = parse_cgroupfs_options(data, &opts); | |
1847 | if (ret) | |
1848 | goto out_unlock; | |
1849 | ||
f392e51c | 1850 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1851 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1852 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1853 | |
f392e51c TH |
1854 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1855 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1856 | |
cf5d5941 | 1857 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1858 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1859 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1860 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1861 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1862 | ret = -EINVAL; |
1863 | goto out_unlock; | |
1864 | } | |
1865 | ||
f172e67c | 1866 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1867 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1868 | ret = -EBUSY; |
0670e08b | 1869 | goto out_unlock; |
cf5d5941 | 1870 | } |
ddbcc7e8 | 1871 | |
5df36032 | 1872 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1873 | if (ret) |
0670e08b | 1874 | goto out_unlock; |
ddbcc7e8 | 1875 | |
334c3679 | 1876 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1877 | |
69e943b7 TH |
1878 | if (opts.release_agent) { |
1879 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1880 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1881 | spin_unlock(&release_agent_path_lock); |
1882 | } | |
ddbcc7e8 | 1883 | out_unlock: |
66bdc9cf | 1884 | kfree(opts.release_agent); |
c6d57f33 | 1885 | kfree(opts.name); |
ddbcc7e8 | 1886 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1887 | return ret; |
1888 | } | |
1889 | ||
afeb0f9f TH |
1890 | /* |
1891 | * To reduce the fork() overhead for systems that are not actually using | |
1892 | * their cgroups capability, we don't maintain the lists running through | |
1893 | * each css_set to its tasks until we see the list actually used - in other | |
1894 | * words after the first mount. | |
1895 | */ | |
1896 | static bool use_task_css_set_links __read_mostly; | |
1897 | ||
1898 | static void cgroup_enable_task_cg_lists(void) | |
1899 | { | |
1900 | struct task_struct *p, *g; | |
1901 | ||
82d6489d | 1902 | spin_lock_irq(&css_set_lock); |
afeb0f9f TH |
1903 | |
1904 | if (use_task_css_set_links) | |
1905 | goto out_unlock; | |
1906 | ||
1907 | use_task_css_set_links = true; | |
1908 | ||
1909 | /* | |
1910 | * We need tasklist_lock because RCU is not safe against | |
1911 | * while_each_thread(). Besides, a forking task that has passed | |
1912 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1913 | * is not guaranteed to have its child immediately visible in the | |
1914 | * tasklist if we walk through it with RCU. | |
1915 | */ | |
1916 | read_lock(&tasklist_lock); | |
1917 | do_each_thread(g, p) { | |
afeb0f9f TH |
1918 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1919 | task_css_set(p) != &init_css_set); | |
1920 | ||
1921 | /* | |
1922 | * We should check if the process is exiting, otherwise | |
1923 | * it will race with cgroup_exit() in that the list | |
1924 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1925 | * Do it while holding siglock so that we don't end up |
1926 | * racing against cgroup_exit(). | |
82d6489d DBO |
1927 | * |
1928 | * Interrupts were already disabled while acquiring | |
1929 | * the css_set_lock, so we do not need to disable it | |
1930 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1931 | */ |
82d6489d | 1932 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1933 | if (!(p->flags & PF_EXITING)) { |
1934 | struct css_set *cset = task_css_set(p); | |
1935 | ||
0de0942d TH |
1936 | if (!css_set_populated(cset)) |
1937 | css_set_update_populated(cset, true); | |
389b9c1b | 1938 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1939 | get_css_set(cset); |
1940 | } | |
82d6489d | 1941 | spin_unlock(&p->sighand->siglock); |
afeb0f9f TH |
1942 | } while_each_thread(g, p); |
1943 | read_unlock(&tasklist_lock); | |
1944 | out_unlock: | |
82d6489d | 1945 | spin_unlock_irq(&css_set_lock); |
afeb0f9f | 1946 | } |
ddbcc7e8 | 1947 | |
cc31edce PM |
1948 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1949 | { | |
2d8f243a TH |
1950 | struct cgroup_subsys *ss; |
1951 | int ssid; | |
1952 | ||
d5c419b6 TH |
1953 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1954 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1955 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1956 | INIT_LIST_HEAD(&cgrp->pidlists); |
1957 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1958 | cgrp->self.cgroup = cgrp; |
184faf32 | 1959 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1960 | |
1961 | for_each_subsys(ss, ssid) | |
1962 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1963 | |
1964 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1965 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1966 | } |
c6d57f33 | 1967 | |
3dd06ffa | 1968 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1969 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1970 | { |
3dd06ffa | 1971 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1972 | |
ddbcc7e8 | 1973 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1974 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1975 | cgrp->root = root; |
cc31edce | 1976 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1977 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1978 | |
c6d57f33 PM |
1979 | root->flags = opts->flags; |
1980 | if (opts->release_agent) | |
1981 | strcpy(root->release_agent_path, opts->release_agent); | |
1982 | if (opts->name) | |
1983 | strcpy(root->name, opts->name); | |
2260e7fc | 1984 | if (opts->cpuset_clone_children) |
3dd06ffa | 1985 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1986 | } |
1987 | ||
6e5c8307 | 1988 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1989 | { |
d427dfeb | 1990 | LIST_HEAD(tmp_links); |
3dd06ffa | 1991 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1992 | struct css_set *cset; |
d427dfeb | 1993 | int i, ret; |
2c6ab6d2 | 1994 | |
d427dfeb | 1995 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1996 | |
cf780b7d | 1997 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1998 | if (ret < 0) |
2bd59d48 | 1999 | goto out; |
d427dfeb | 2000 | root_cgrp->id = ret; |
b11cfb58 | 2001 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 2002 | |
2aad2a86 TH |
2003 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
2004 | GFP_KERNEL); | |
9d755d33 TH |
2005 | if (ret) |
2006 | goto out; | |
2007 | ||
d427dfeb | 2008 | /* |
f0d9a5f1 | 2009 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 2010 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
2011 | * cgroup_lock, and that's us. Later rebinding may disable |
2012 | * controllers on the default hierarchy and thus create new csets, | |
2013 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 2014 | */ |
04313591 | 2015 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 2016 | if (ret) |
9d755d33 | 2017 | goto cancel_ref; |
ddbcc7e8 | 2018 | |
985ed670 | 2019 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 2020 | if (ret) |
9d755d33 | 2021 | goto cancel_ref; |
ddbcc7e8 | 2022 | |
2bd59d48 TH |
2023 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
2024 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
2025 | root_cgrp); | |
2026 | if (IS_ERR(root->kf_root)) { | |
2027 | ret = PTR_ERR(root->kf_root); | |
2028 | goto exit_root_id; | |
2029 | } | |
2030 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 2031 | |
334c3679 | 2032 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 2033 | if (ret) |
2bd59d48 | 2034 | goto destroy_root; |
ddbcc7e8 | 2035 | |
5df36032 | 2036 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 2037 | if (ret) |
2bd59d48 | 2038 | goto destroy_root; |
ddbcc7e8 | 2039 | |
d427dfeb TH |
2040 | /* |
2041 | * There must be no failure case after here, since rebinding takes | |
2042 | * care of subsystems' refcounts, which are explicitly dropped in | |
2043 | * the failure exit path. | |
2044 | */ | |
2045 | list_add(&root->root_list, &cgroup_roots); | |
2046 | cgroup_root_count++; | |
0df6a63f | 2047 | |
d427dfeb | 2048 | /* |
3dd06ffa | 2049 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2050 | * objects. |
2051 | */ | |
82d6489d | 2052 | spin_lock_irq(&css_set_lock); |
0de0942d | 2053 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2054 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2055 | if (css_set_populated(cset)) |
2056 | cgroup_update_populated(root_cgrp, true); | |
2057 | } | |
82d6489d | 2058 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 2059 | |
d5c419b6 | 2060 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2061 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2062 | |
2bd59d48 | 2063 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2064 | ret = 0; |
2bd59d48 | 2065 | goto out; |
d427dfeb | 2066 | |
2bd59d48 TH |
2067 | destroy_root: |
2068 | kernfs_destroy_root(root->kf_root); | |
2069 | root->kf_root = NULL; | |
2070 | exit_root_id: | |
d427dfeb | 2071 | cgroup_exit_root_id(root); |
9d755d33 | 2072 | cancel_ref: |
9a1049da | 2073 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2074 | out: |
d427dfeb TH |
2075 | free_cgrp_cset_links(&tmp_links); |
2076 | return ret; | |
ddbcc7e8 PM |
2077 | } |
2078 | ||
f7e83571 | 2079 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2080 | int flags, const char *unused_dev_name, |
f7e83571 | 2081 | void *data) |
ddbcc7e8 | 2082 | { |
67e9c74b | 2083 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 2084 | struct super_block *pinned_sb = NULL; |
ed82571b | 2085 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
970317aa | 2086 | struct cgroup_subsys *ss; |
3dd06ffa | 2087 | struct cgroup_root *root; |
ddbcc7e8 | 2088 | struct cgroup_sb_opts opts; |
2bd59d48 | 2089 | struct dentry *dentry; |
8e30e2b8 | 2090 | int ret; |
970317aa | 2091 | int i; |
c6b3d5bc | 2092 | bool new_sb; |
ddbcc7e8 | 2093 | |
ed82571b SH |
2094 | get_cgroup_ns(ns); |
2095 | ||
2096 | /* Check if the caller has permission to mount. */ | |
2097 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2098 | put_cgroup_ns(ns); | |
2099 | return ERR_PTR(-EPERM); | |
2100 | } | |
2101 | ||
56fde9e0 TH |
2102 | /* |
2103 | * The first time anyone tries to mount a cgroup, enable the list | |
2104 | * linking each css_set to its tasks and fix up all existing tasks. | |
2105 | */ | |
2106 | if (!use_task_css_set_links) | |
2107 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2108 | |
67e9c74b TH |
2109 | if (is_v2) { |
2110 | if (data) { | |
2111 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
ed82571b | 2112 | put_cgroup_ns(ns); |
67e9c74b TH |
2113 | return ERR_PTR(-EINVAL); |
2114 | } | |
a7165264 | 2115 | cgrp_dfl_visible = true; |
67e9c74b TH |
2116 | root = &cgrp_dfl_root; |
2117 | cgroup_get(&root->cgrp); | |
2118 | goto out_mount; | |
2119 | } | |
2120 | ||
334c3679 | 2121 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2122 | |
2123 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2124 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2125 | if (ret) |
8e30e2b8 | 2126 | goto out_unlock; |
a015edd2 | 2127 | |
970317aa LZ |
2128 | /* |
2129 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2130 | * still be dying after the previous unmount. Let's drain the | |
2131 | * dying subsystems. We just need to ensure that the ones | |
2132 | * unmounted previously finish dying and don't care about new ones | |
2133 | * starting. Testing ref liveliness is good enough. | |
2134 | */ | |
2135 | for_each_subsys(ss, i) { | |
2136 | if (!(opts.subsys_mask & (1 << i)) || | |
2137 | ss->root == &cgrp_dfl_root) | |
2138 | continue; | |
2139 | ||
2140 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2141 | mutex_unlock(&cgroup_mutex); | |
2142 | msleep(10); | |
2143 | ret = restart_syscall(); | |
2144 | goto out_free; | |
2145 | } | |
2146 | cgroup_put(&ss->root->cgrp); | |
2147 | } | |
2148 | ||
985ed670 | 2149 | for_each_root(root) { |
2bd59d48 | 2150 | bool name_match = false; |
3126121f | 2151 | |
3dd06ffa | 2152 | if (root == &cgrp_dfl_root) |
985ed670 | 2153 | continue; |
3126121f | 2154 | |
cf5d5941 | 2155 | /* |
2bd59d48 TH |
2156 | * If we asked for a name then it must match. Also, if |
2157 | * name matches but sybsys_mask doesn't, we should fail. | |
2158 | * Remember whether name matched. | |
cf5d5941 | 2159 | */ |
2bd59d48 TH |
2160 | if (opts.name) { |
2161 | if (strcmp(opts.name, root->name)) | |
2162 | continue; | |
2163 | name_match = true; | |
2164 | } | |
ddbcc7e8 | 2165 | |
c6d57f33 | 2166 | /* |
2bd59d48 TH |
2167 | * If we asked for subsystems (or explicitly for no |
2168 | * subsystems) then they must match. | |
c6d57f33 | 2169 | */ |
2bd59d48 | 2170 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2171 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2172 | if (!name_match) |
2173 | continue; | |
2174 | ret = -EBUSY; | |
2175 | goto out_unlock; | |
2176 | } | |
873fe09e | 2177 | |
7b9a6ba5 TH |
2178 | if (root->flags ^ opts.flags) |
2179 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2180 | |
776f02fa | 2181 | /* |
3a32bd72 LZ |
2182 | * We want to reuse @root whose lifetime is governed by its |
2183 | * ->cgrp. Let's check whether @root is alive and keep it | |
2184 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2185 | * want to block it by pinning the sb so that @root doesn't | |
2186 | * get killed before mount is complete. | |
2187 | * | |
2188 | * With the sb pinned, tryget_live can reliably indicate | |
2189 | * whether @root can be reused. If it's being killed, | |
2190 | * drain it. We can use wait_queue for the wait but this | |
2191 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2192 | */ |
3a32bd72 LZ |
2193 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2194 | if (IS_ERR(pinned_sb) || | |
2195 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2196 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2197 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2198 | deactivate_super(pinned_sb); | |
776f02fa | 2199 | msleep(10); |
a015edd2 TH |
2200 | ret = restart_syscall(); |
2201 | goto out_free; | |
776f02fa | 2202 | } |
ddbcc7e8 | 2203 | |
776f02fa | 2204 | ret = 0; |
2bd59d48 | 2205 | goto out_unlock; |
ddbcc7e8 | 2206 | } |
ddbcc7e8 | 2207 | |
817929ec | 2208 | /* |
172a2c06 TH |
2209 | * No such thing, create a new one. name= matching without subsys |
2210 | * specification is allowed for already existing hierarchies but we | |
2211 | * can't create new one without subsys specification. | |
817929ec | 2212 | */ |
172a2c06 TH |
2213 | if (!opts.subsys_mask && !opts.none) { |
2214 | ret = -EINVAL; | |
2215 | goto out_unlock; | |
817929ec | 2216 | } |
817929ec | 2217 | |
ed82571b SH |
2218 | /* |
2219 | * We know this subsystem has not yet been bound. Users in a non-init | |
2220 | * user namespace may only mount hierarchies with no bound subsystems, | |
2221 | * i.e. 'none,name=user1' | |
2222 | */ | |
2223 | if (!opts.none && !capable(CAP_SYS_ADMIN)) { | |
2224 | ret = -EPERM; | |
2225 | goto out_unlock; | |
2226 | } | |
2227 | ||
172a2c06 TH |
2228 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2229 | if (!root) { | |
2230 | ret = -ENOMEM; | |
2bd59d48 | 2231 | goto out_unlock; |
839ec545 | 2232 | } |
e5f6a860 | 2233 | |
172a2c06 TH |
2234 | init_cgroup_root(root, &opts); |
2235 | ||
35585573 | 2236 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2237 | if (ret) |
2238 | cgroup_free_root(root); | |
fa3ca07e | 2239 | |
8e30e2b8 | 2240 | out_unlock: |
ddbcc7e8 | 2241 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2242 | out_free: |
c6d57f33 PM |
2243 | kfree(opts.release_agent); |
2244 | kfree(opts.name); | |
03b1cde6 | 2245 | |
ed82571b SH |
2246 | if (ret) { |
2247 | put_cgroup_ns(ns); | |
8e30e2b8 | 2248 | return ERR_PTR(ret); |
ed82571b | 2249 | } |
67e9c74b | 2250 | out_mount: |
c9482a5b | 2251 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2252 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2253 | &new_sb); | |
ed82571b SH |
2254 | |
2255 | /* | |
2256 | * In non-init cgroup namespace, instead of root cgroup's | |
2257 | * dentry, we return the dentry corresponding to the | |
2258 | * cgroupns->root_cgrp. | |
2259 | */ | |
2260 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { | |
2261 | struct dentry *nsdentry; | |
2262 | struct cgroup *cgrp; | |
2263 | ||
2264 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2265 | spin_lock_irq(&css_set_lock); |
ed82571b SH |
2266 | |
2267 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2268 | ||
82d6489d | 2269 | spin_unlock_irq(&css_set_lock); |
ed82571b SH |
2270 | mutex_unlock(&cgroup_mutex); |
2271 | ||
2272 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); | |
2273 | dput(dentry); | |
2274 | dentry = nsdentry; | |
2275 | } | |
2276 | ||
c6b3d5bc | 2277 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2278 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2279 | |
2280 | /* | |
2281 | * If @pinned_sb, we're reusing an existing root and holding an | |
2282 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2283 | */ | |
2284 | if (pinned_sb) { | |
2285 | WARN_ON(new_sb); | |
2286 | deactivate_super(pinned_sb); | |
2287 | } | |
2288 | ||
ed82571b | 2289 | put_cgroup_ns(ns); |
2bd59d48 TH |
2290 | return dentry; |
2291 | } | |
2292 | ||
2293 | static void cgroup_kill_sb(struct super_block *sb) | |
2294 | { | |
2295 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2296 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2297 | |
9d755d33 TH |
2298 | /* |
2299 | * If @root doesn't have any mounts or children, start killing it. | |
2300 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2301 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2302 | * |
2303 | * And don't kill the default root. | |
9d755d33 | 2304 | */ |
3c606d35 | 2305 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2306 | root == &cgrp_dfl_root) |
9d755d33 TH |
2307 | cgroup_put(&root->cgrp); |
2308 | else | |
2309 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2310 | ||
2bd59d48 | 2311 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2312 | } |
2313 | ||
2314 | static struct file_system_type cgroup_fs_type = { | |
2315 | .name = "cgroup", | |
f7e83571 | 2316 | .mount = cgroup_mount, |
ddbcc7e8 | 2317 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2318 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 PM |
2319 | }; |
2320 | ||
67e9c74b TH |
2321 | static struct file_system_type cgroup2_fs_type = { |
2322 | .name = "cgroup2", | |
2323 | .mount = cgroup_mount, | |
2324 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2325 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b TH |
2326 | }; |
2327 | ||
a79a908f AK |
2328 | static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2329 | struct cgroup_namespace *ns) | |
2330 | { | |
2331 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
2332 | int ret; | |
2333 | ||
2334 | ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); | |
2335 | if (ret < 0 || ret >= buflen) | |
2336 | return NULL; | |
2337 | return buf; | |
2338 | } | |
2339 | ||
2340 | char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, | |
2341 | struct cgroup_namespace *ns) | |
2342 | { | |
2343 | char *ret; | |
2344 | ||
2345 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2346 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2347 | |
2348 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2349 | ||
82d6489d | 2350 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2351 | mutex_unlock(&cgroup_mutex); |
2352 | ||
2353 | return ret; | |
2354 | } | |
2355 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2356 | ||
857a2beb | 2357 | /** |
913ffdb5 | 2358 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2359 | * @task: target task |
857a2beb TH |
2360 | * @buf: the buffer to write the path into |
2361 | * @buflen: the length of the buffer | |
2362 | * | |
913ffdb5 TH |
2363 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2364 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2365 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2366 | * cgroup controller callbacks. | |
2367 | * | |
e61734c5 | 2368 | * Return value is the same as kernfs_path(). |
857a2beb | 2369 | */ |
e61734c5 | 2370 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2371 | { |
3dd06ffa | 2372 | struct cgroup_root *root; |
913ffdb5 | 2373 | struct cgroup *cgrp; |
e61734c5 TH |
2374 | int hierarchy_id = 1; |
2375 | char *path = NULL; | |
857a2beb TH |
2376 | |
2377 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2378 | spin_lock_irq(&css_set_lock); |
857a2beb | 2379 | |
913ffdb5 TH |
2380 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2381 | ||
857a2beb TH |
2382 | if (root) { |
2383 | cgrp = task_cgroup_from_root(task, root); | |
a79a908f | 2384 | path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2385 | } else { |
2386 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2387 | if (strlcpy(buf, "/", buflen) < buflen) |
2388 | path = buf; | |
857a2beb TH |
2389 | } |
2390 | ||
82d6489d | 2391 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2392 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2393 | return path; |
857a2beb | 2394 | } |
913ffdb5 | 2395 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2396 | |
b3dc094e | 2397 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2398 | struct cgroup_taskset { |
b3dc094e TH |
2399 | /* the src and dst cset list running through cset->mg_node */ |
2400 | struct list_head src_csets; | |
2401 | struct list_head dst_csets; | |
2402 | ||
1f7dd3e5 TH |
2403 | /* the subsys currently being processed */ |
2404 | int ssid; | |
2405 | ||
b3dc094e TH |
2406 | /* |
2407 | * Fields for cgroup_taskset_*() iteration. | |
2408 | * | |
2409 | * Before migration is committed, the target migration tasks are on | |
2410 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2411 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2412 | * or ->dst_csets depending on whether migration is committed. | |
2413 | * | |
2414 | * ->cur_csets and ->cur_task point to the current task position | |
2415 | * during iteration. | |
2416 | */ | |
2417 | struct list_head *csets; | |
2418 | struct css_set *cur_cset; | |
2419 | struct task_struct *cur_task; | |
2f7ee569 TH |
2420 | }; |
2421 | ||
adaae5dc TH |
2422 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2423 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2424 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2425 | .csets = &tset.src_csets, \ | |
2426 | } | |
2427 | ||
2428 | /** | |
2429 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2430 | * @task: target task | |
2431 | * @tset: target taskset | |
2432 | * | |
2433 | * Add @task, which is a migration target, to @tset. This function becomes | |
2434 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2435 | * been added as a migration source and @task->cg_list will be moved from | |
2436 | * the css_set's tasks list to mg_tasks one. | |
2437 | */ | |
2438 | static void cgroup_taskset_add(struct task_struct *task, | |
2439 | struct cgroup_taskset *tset) | |
2440 | { | |
2441 | struct css_set *cset; | |
2442 | ||
f0d9a5f1 | 2443 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2444 | |
2445 | /* @task either already exited or can't exit until the end */ | |
2446 | if (task->flags & PF_EXITING) | |
2447 | return; | |
2448 | ||
2449 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2450 | if (list_empty(&task->cg_list)) | |
2451 | return; | |
2452 | ||
2453 | cset = task_css_set(task); | |
2454 | if (!cset->mg_src_cgrp) | |
2455 | return; | |
2456 | ||
2457 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2458 | if (list_empty(&cset->mg_node)) | |
2459 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2460 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2461 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2462 | &tset->dst_csets); | |
2463 | } | |
2464 | ||
2f7ee569 TH |
2465 | /** |
2466 | * cgroup_taskset_first - reset taskset and return the first task | |
2467 | * @tset: taskset of interest | |
1f7dd3e5 | 2468 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2469 | * |
2470 | * @tset iteration is initialized and the first task is returned. | |
2471 | */ | |
1f7dd3e5 TH |
2472 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2473 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2474 | { |
b3dc094e TH |
2475 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2476 | tset->cur_task = NULL; | |
2477 | ||
1f7dd3e5 | 2478 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2479 | } |
2f7ee569 TH |
2480 | |
2481 | /** | |
2482 | * cgroup_taskset_next - iterate to the next task in taskset | |
2483 | * @tset: taskset of interest | |
1f7dd3e5 | 2484 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2485 | * |
2486 | * Return the next task in @tset. Iteration must have been initialized | |
2487 | * with cgroup_taskset_first(). | |
2488 | */ | |
1f7dd3e5 TH |
2489 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2490 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2491 | { |
b3dc094e TH |
2492 | struct css_set *cset = tset->cur_cset; |
2493 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2494 | |
b3dc094e TH |
2495 | while (&cset->mg_node != tset->csets) { |
2496 | if (!task) | |
2497 | task = list_first_entry(&cset->mg_tasks, | |
2498 | struct task_struct, cg_list); | |
2499 | else | |
2500 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2501 | |
b3dc094e TH |
2502 | if (&task->cg_list != &cset->mg_tasks) { |
2503 | tset->cur_cset = cset; | |
2504 | tset->cur_task = task; | |
1f7dd3e5 TH |
2505 | |
2506 | /* | |
2507 | * This function may be called both before and | |
2508 | * after cgroup_taskset_migrate(). The two cases | |
2509 | * can be distinguished by looking at whether @cset | |
2510 | * has its ->mg_dst_cset set. | |
2511 | */ | |
2512 | if (cset->mg_dst_cset) | |
2513 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2514 | else | |
2515 | *dst_cssp = cset->subsys[tset->ssid]; | |
2516 | ||
b3dc094e TH |
2517 | return task; |
2518 | } | |
2f7ee569 | 2519 | |
b3dc094e TH |
2520 | cset = list_next_entry(cset, mg_node); |
2521 | task = NULL; | |
2522 | } | |
2f7ee569 | 2523 | |
b3dc094e | 2524 | return NULL; |
2f7ee569 | 2525 | } |
2f7ee569 | 2526 | |
adaae5dc | 2527 | /** |
37ff9f8f | 2528 | * cgroup_taskset_migrate - migrate a taskset |
adaae5dc | 2529 | * @tset: taget taskset |
37ff9f8f | 2530 | * @root: cgroup root the migration is taking place on |
adaae5dc | 2531 | * |
37ff9f8f TH |
2532 | * Migrate tasks in @tset as setup by migration preparation functions. |
2533 | * This function fails iff one of the ->can_attach callbacks fails and | |
2534 | * guarantees that either all or none of the tasks in @tset are migrated. | |
2535 | * @tset is consumed regardless of success. | |
adaae5dc TH |
2536 | */ |
2537 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
37ff9f8f | 2538 | struct cgroup_root *root) |
adaae5dc | 2539 | { |
37ff9f8f | 2540 | struct cgroup_subsys *ss; |
adaae5dc TH |
2541 | struct task_struct *task, *tmp_task; |
2542 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2543 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2544 | |
2545 | /* methods shouldn't be called if no task is actually migrating */ | |
2546 | if (list_empty(&tset->src_csets)) | |
2547 | return 0; | |
2548 | ||
2549 | /* check that we can legitimately attach to the cgroup */ | |
37ff9f8f TH |
2550 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2551 | if (ss->can_attach) { | |
2552 | tset->ssid = ssid; | |
2553 | ret = ss->can_attach(tset); | |
adaae5dc | 2554 | if (ret) { |
37ff9f8f | 2555 | failed_ssid = ssid; |
adaae5dc TH |
2556 | goto out_cancel_attach; |
2557 | } | |
2558 | } | |
37ff9f8f | 2559 | } while_each_subsys_mask(); |
adaae5dc TH |
2560 | |
2561 | /* | |
2562 | * Now that we're guaranteed success, proceed to move all tasks to | |
2563 | * the new cgroup. There are no failure cases after here, so this | |
2564 | * is the commit point. | |
2565 | */ | |
82d6489d | 2566 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2567 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2568 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2569 | struct css_set *from_cset = task_css_set(task); | |
2570 | struct css_set *to_cset = cset->mg_dst_cset; | |
2571 | ||
2572 | get_css_set(to_cset); | |
2573 | css_set_move_task(task, from_cset, to_cset, true); | |
2574 | put_css_set_locked(from_cset); | |
2575 | } | |
adaae5dc | 2576 | } |
82d6489d | 2577 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2578 | |
2579 | /* | |
2580 | * Migration is committed, all target tasks are now on dst_csets. | |
2581 | * Nothing is sensitive to fork() after this point. Notify | |
2582 | * controllers that migration is complete. | |
2583 | */ | |
2584 | tset->csets = &tset->dst_csets; | |
2585 | ||
37ff9f8f TH |
2586 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2587 | if (ss->attach) { | |
2588 | tset->ssid = ssid; | |
2589 | ss->attach(tset); | |
1f7dd3e5 | 2590 | } |
37ff9f8f | 2591 | } while_each_subsys_mask(); |
adaae5dc TH |
2592 | |
2593 | ret = 0; | |
2594 | goto out_release_tset; | |
2595 | ||
2596 | out_cancel_attach: | |
37ff9f8f TH |
2597 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2598 | if (ssid == failed_ssid) | |
adaae5dc | 2599 | break; |
37ff9f8f TH |
2600 | if (ss->cancel_attach) { |
2601 | tset->ssid = ssid; | |
2602 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2603 | } |
37ff9f8f | 2604 | } while_each_subsys_mask(); |
adaae5dc | 2605 | out_release_tset: |
82d6489d | 2606 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2607 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2608 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2609 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2610 | list_del_init(&cset->mg_node); | |
2611 | } | |
82d6489d | 2612 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2613 | return ret; |
2614 | } | |
2615 | ||
6c694c88 TH |
2616 | /** |
2617 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2618 | * @dst_cgrp: destination cgroup to test | |
2619 | * | |
2620 | * On the default hierarchy, except for the root, subtree_control must be | |
2621 | * zero for migration destination cgroups with tasks so that child cgroups | |
2622 | * don't compete against tasks. | |
2623 | */ | |
2624 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) | |
2625 | { | |
2626 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2627 | !dst_cgrp->subtree_control; | |
2628 | } | |
2629 | ||
a043e3b2 | 2630 | /** |
1958d2d5 TH |
2631 | * cgroup_migrate_finish - cleanup after attach |
2632 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2633 | * |
1958d2d5 TH |
2634 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2635 | * those functions for details. | |
74a1166d | 2636 | */ |
1958d2d5 | 2637 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2638 | { |
1958d2d5 | 2639 | struct css_set *cset, *tmp_cset; |
74a1166d | 2640 | |
1958d2d5 TH |
2641 | lockdep_assert_held(&cgroup_mutex); |
2642 | ||
82d6489d | 2643 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2644 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2645 | cset->mg_src_cgrp = NULL; | |
e4857982 | 2646 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2647 | cset->mg_dst_cset = NULL; |
2648 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2649 | put_css_set_locked(cset); |
1958d2d5 | 2650 | } |
82d6489d | 2651 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2652 | } |
2653 | ||
2654 | /** | |
2655 | * cgroup_migrate_add_src - add a migration source css_set | |
2656 | * @src_cset: the source css_set to add | |
2657 | * @dst_cgrp: the destination cgroup | |
2658 | * @preloaded_csets: list of preloaded css_sets | |
2659 | * | |
2660 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2661 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2662 | * up by cgroup_migrate_finish(). | |
2663 | * | |
1ed13287 TH |
2664 | * This function may be called without holding cgroup_threadgroup_rwsem |
2665 | * even if the target is a process. Threads may be created and destroyed | |
2666 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2667 | * into play and the preloaded css_sets are guaranteed to cover all | |
2668 | * migrations. | |
1958d2d5 TH |
2669 | */ |
2670 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2671 | struct cgroup *dst_cgrp, | |
2672 | struct list_head *preloaded_csets) | |
2673 | { | |
2674 | struct cgroup *src_cgrp; | |
2675 | ||
2676 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2677 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2678 | |
2b021cbf TH |
2679 | /* |
2680 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2681 | * and doesn't contain any migratable tasks. Ignore it early so | |
2682 | * that the rest of migration path doesn't get confused by it. | |
2683 | */ | |
2684 | if (src_cset->dead) | |
2685 | return; | |
2686 | ||
1958d2d5 TH |
2687 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2688 | ||
1958d2d5 TH |
2689 | if (!list_empty(&src_cset->mg_preload_node)) |
2690 | return; | |
2691 | ||
2692 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2693 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2694 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2695 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2696 | ||
2697 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2698 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 TH |
2699 | get_css_set(src_cset); |
2700 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2701 | } | |
2702 | ||
2703 | /** | |
2704 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1958d2d5 TH |
2705 | * @preloaded_csets: list of preloaded source css_sets |
2706 | * | |
e4857982 TH |
2707 | * Tasks are about to be moved and all the source css_sets have been |
2708 | * preloaded to @preloaded_csets. This function looks up and pins all | |
2709 | * destination css_sets, links each to its source, and append them to | |
2710 | * @preloaded_csets. | |
1958d2d5 TH |
2711 | * |
2712 | * This function must be called after cgroup_migrate_add_src() has been | |
2713 | * called on each migration source css_set. After migration is performed | |
2714 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2715 | * @preloaded_csets. | |
2716 | */ | |
e4857982 | 2717 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
1958d2d5 TH |
2718 | { |
2719 | LIST_HEAD(csets); | |
f817de98 | 2720 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2721 | |
2722 | lockdep_assert_held(&cgroup_mutex); | |
2723 | ||
2724 | /* look up the dst cset for each src cset and link it to src */ | |
f817de98 | 2725 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2726 | struct css_set *dst_cset; |
2727 | ||
e4857982 | 2728 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2729 | if (!dst_cset) |
2730 | goto err; | |
2731 | ||
2732 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2733 | |
2734 | /* | |
2735 | * If src cset equals dst, it's noop. Drop the src. | |
2736 | * cgroup_migrate() will skip the cset too. Note that we | |
2737 | * can't handle src == dst as some nodes are used by both. | |
2738 | */ | |
2739 | if (src_cset == dst_cset) { | |
2740 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2741 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2742 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2743 | put_css_set(src_cset); |
2744 | put_css_set(dst_cset); | |
f817de98 TH |
2745 | continue; |
2746 | } | |
2747 | ||
1958d2d5 TH |
2748 | src_cset->mg_dst_cset = dst_cset; |
2749 | ||
2750 | if (list_empty(&dst_cset->mg_preload_node)) | |
2751 | list_add(&dst_cset->mg_preload_node, &csets); | |
2752 | else | |
a25eb52e | 2753 | put_css_set(dst_cset); |
1958d2d5 TH |
2754 | } |
2755 | ||
f817de98 | 2756 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2757 | return 0; |
2758 | err: | |
2759 | cgroup_migrate_finish(&csets); | |
2760 | return -ENOMEM; | |
2761 | } | |
2762 | ||
2763 | /** | |
2764 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2765 | * @leader: the leader of the process or the task to migrate |
2766 | * @threadgroup: whether @leader points to the whole process or a single task | |
37ff9f8f | 2767 | * @root: cgroup root migration is taking place on |
1958d2d5 | 2768 | * |
37ff9f8f TH |
2769 | * Migrate a process or task denoted by @leader. If migrating a process, |
2770 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2771 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2772 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2773 | * function and following up with cgroup_migrate_finish(). | |
2774 | * | |
2775 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2776 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2777 | * failure, when migrating multiple targets, the success or failure can be | |
2778 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2779 | * actually starting migrating. | |
2780 | */ | |
9af2ec45 | 2781 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
37ff9f8f | 2782 | struct cgroup_root *root) |
74a1166d | 2783 | { |
adaae5dc TH |
2784 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2785 | struct task_struct *task; | |
74a1166d | 2786 | |
fb5d2b4c MSB |
2787 | /* |
2788 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2789 | * already PF_EXITING could be freed from underneath us unless we | |
2790 | * take an rcu_read_lock. | |
2791 | */ | |
82d6489d | 2792 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2793 | rcu_read_lock(); |
9db8de37 | 2794 | task = leader; |
74a1166d | 2795 | do { |
adaae5dc | 2796 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2797 | if (!threadgroup) |
2798 | break; | |
9db8de37 | 2799 | } while_each_thread(leader, task); |
fb5d2b4c | 2800 | rcu_read_unlock(); |
82d6489d | 2801 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2802 | |
37ff9f8f | 2803 | return cgroup_taskset_migrate(&tset, root); |
74a1166d BB |
2804 | } |
2805 | ||
1958d2d5 TH |
2806 | /** |
2807 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2808 | * @dst_cgrp: the cgroup to attach to | |
2809 | * @leader: the task or the leader of the threadgroup to be attached | |
2810 | * @threadgroup: attach the whole threadgroup? | |
2811 | * | |
1ed13287 | 2812 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2813 | */ |
2814 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2815 | struct task_struct *leader, bool threadgroup) | |
2816 | { | |
2817 | LIST_HEAD(preloaded_csets); | |
2818 | struct task_struct *task; | |
2819 | int ret; | |
2820 | ||
6c694c88 TH |
2821 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2822 | return -EBUSY; | |
2823 | ||
1958d2d5 | 2824 | /* look up all src csets */ |
82d6489d | 2825 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2826 | rcu_read_lock(); |
2827 | task = leader; | |
2828 | do { | |
2829 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2830 | &preloaded_csets); | |
2831 | if (!threadgroup) | |
2832 | break; | |
2833 | } while_each_thread(leader, task); | |
2834 | rcu_read_unlock(); | |
82d6489d | 2835 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2836 | |
2837 | /* prepare dst csets and commit */ | |
e4857982 | 2838 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
1958d2d5 | 2839 | if (!ret) |
37ff9f8f | 2840 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
1958d2d5 TH |
2841 | |
2842 | cgroup_migrate_finish(&preloaded_csets); | |
2843 | return ret; | |
74a1166d BB |
2844 | } |
2845 | ||
187fe840 TH |
2846 | static int cgroup_procs_write_permission(struct task_struct *task, |
2847 | struct cgroup *dst_cgrp, | |
2848 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2849 | { |
2850 | const struct cred *cred = current_cred(); | |
2851 | const struct cred *tcred = get_task_cred(task); | |
2852 | int ret = 0; | |
2853 | ||
2854 | /* | |
2855 | * even if we're attaching all tasks in the thread group, we only | |
2856 | * need to check permissions on one of them. | |
2857 | */ | |
2858 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2859 | !uid_eq(cred->euid, tcred->uid) && | |
2860 | !uid_eq(cred->euid, tcred->suid)) | |
2861 | ret = -EACCES; | |
2862 | ||
187fe840 TH |
2863 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2864 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2865 | struct cgroup *cgrp; | |
2866 | struct inode *inode; | |
2867 | ||
82d6489d | 2868 | spin_lock_irq(&css_set_lock); |
187fe840 | 2869 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
82d6489d | 2870 | spin_unlock_irq(&css_set_lock); |
187fe840 TH |
2871 | |
2872 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2873 | cgrp = cgroup_parent(cgrp); | |
2874 | ||
2875 | ret = -ENOMEM; | |
6f60eade | 2876 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2877 | if (inode) { |
2878 | ret = inode_permission(inode, MAY_WRITE); | |
2879 | iput(inode); | |
2880 | } | |
2881 | } | |
2882 | ||
dedf22e9 TH |
2883 | put_cred(tcred); |
2884 | return ret; | |
2885 | } | |
2886 | ||
74a1166d BB |
2887 | /* |
2888 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2889 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2890 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2891 | */ |
acbef755 TH |
2892 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2893 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2894 | { |
bbcb81d0 | 2895 | struct task_struct *tsk; |
5cf1cacb | 2896 | struct cgroup_subsys *ss; |
e76ecaee | 2897 | struct cgroup *cgrp; |
acbef755 | 2898 | pid_t pid; |
5cf1cacb | 2899 | int ssid, ret; |
bbcb81d0 | 2900 | |
acbef755 TH |
2901 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2902 | return -EINVAL; | |
2903 | ||
945ba199 | 2904 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2905 | if (!cgrp) |
74a1166d BB |
2906 | return -ENODEV; |
2907 | ||
3014dde7 | 2908 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2909 | rcu_read_lock(); |
bbcb81d0 | 2910 | if (pid) { |
73507f33 | 2911 | tsk = find_task_by_vpid(pid); |
74a1166d | 2912 | if (!tsk) { |
dd4b0a46 | 2913 | ret = -ESRCH; |
3014dde7 | 2914 | goto out_unlock_rcu; |
bbcb81d0 | 2915 | } |
dedf22e9 | 2916 | } else { |
b78949eb | 2917 | tsk = current; |
dedf22e9 | 2918 | } |
cd3d0952 TH |
2919 | |
2920 | if (threadgroup) | |
b78949eb | 2921 | tsk = tsk->group_leader; |
c4c27fbd MG |
2922 | |
2923 | /* | |
14a40ffc | 2924 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2925 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2926 | * with no rt_runtime allocated. Just say no. | |
2927 | */ | |
14a40ffc | 2928 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2929 | ret = -EINVAL; |
3014dde7 | 2930 | goto out_unlock_rcu; |
c4c27fbd MG |
2931 | } |
2932 | ||
b78949eb MSB |
2933 | get_task_struct(tsk); |
2934 | rcu_read_unlock(); | |
2935 | ||
187fe840 | 2936 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2937 | if (!ret) |
2938 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2939 | |
f9f9e7b7 | 2940 | put_task_struct(tsk); |
3014dde7 TH |
2941 | goto out_unlock_threadgroup; |
2942 | ||
2943 | out_unlock_rcu: | |
2944 | rcu_read_unlock(); | |
2945 | out_unlock_threadgroup: | |
2946 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2947 | for_each_subsys(ss, ssid) |
2948 | if (ss->post_attach) | |
2949 | ss->post_attach(); | |
e76ecaee | 2950 | cgroup_kn_unlock(of->kn); |
acbef755 | 2951 | return ret ?: nbytes; |
bbcb81d0 PM |
2952 | } |
2953 | ||
7ae1bad9 TH |
2954 | /** |
2955 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2956 | * @from: attach to all cgroups of a given task | |
2957 | * @tsk: the task to be attached | |
2958 | */ | |
2959 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2960 | { | |
3dd06ffa | 2961 | struct cgroup_root *root; |
7ae1bad9 TH |
2962 | int retval = 0; |
2963 | ||
47cfcd09 | 2964 | mutex_lock(&cgroup_mutex); |
985ed670 | 2965 | for_each_root(root) { |
96d365e0 TH |
2966 | struct cgroup *from_cgrp; |
2967 | ||
3dd06ffa | 2968 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2969 | continue; |
2970 | ||
82d6489d | 2971 | spin_lock_irq(&css_set_lock); |
96d365e0 | 2972 | from_cgrp = task_cgroup_from_root(from, root); |
82d6489d | 2973 | spin_unlock_irq(&css_set_lock); |
7ae1bad9 | 2974 | |
6f4b7e63 | 2975 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2976 | if (retval) |
2977 | break; | |
2978 | } | |
47cfcd09 | 2979 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2980 | |
2981 | return retval; | |
2982 | } | |
2983 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2984 | ||
acbef755 TH |
2985 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2986 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2987 | { |
acbef755 | 2988 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2989 | } |
2990 | ||
acbef755 TH |
2991 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2992 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2993 | { |
acbef755 | 2994 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2995 | } |
2996 | ||
451af504 TH |
2997 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2998 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2999 | { |
e76ecaee | 3000 | struct cgroup *cgrp; |
5f469907 | 3001 | |
e76ecaee | 3002 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 3003 | |
945ba199 | 3004 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 3005 | if (!cgrp) |
e788e066 | 3006 | return -ENODEV; |
69e943b7 | 3007 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
3008 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
3009 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 3010 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 3011 | cgroup_kn_unlock(of->kn); |
451af504 | 3012 | return nbytes; |
e788e066 PM |
3013 | } |
3014 | ||
2da8ca82 | 3015 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 3016 | { |
2da8ca82 | 3017 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 3018 | |
46cfeb04 | 3019 | spin_lock(&release_agent_path_lock); |
e788e066 | 3020 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 3021 | spin_unlock(&release_agent_path_lock); |
e788e066 | 3022 | seq_putc(seq, '\n'); |
e788e066 PM |
3023 | return 0; |
3024 | } | |
3025 | ||
2da8ca82 | 3026 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 3027 | { |
c1d5d42e | 3028 | seq_puts(seq, "0\n"); |
e788e066 PM |
3029 | return 0; |
3030 | } | |
3031 | ||
6e5c8307 | 3032 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 3033 | { |
f8f22e53 TH |
3034 | struct cgroup_subsys *ss; |
3035 | bool printed = false; | |
3036 | int ssid; | |
a742c59d | 3037 | |
b4e0eeaf | 3038 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
3039 | if (printed) |
3040 | seq_putc(seq, ' '); | |
3041 | seq_printf(seq, "%s", ss->name); | |
3042 | printed = true; | |
b4e0eeaf | 3043 | } while_each_subsys_mask(); |
f8f22e53 TH |
3044 | if (printed) |
3045 | seq_putc(seq, '\n'); | |
355e0c48 PM |
3046 | } |
3047 | ||
f8f22e53 TH |
3048 | /* show controllers which are enabled from the parent */ |
3049 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3050 | { |
f8f22e53 TH |
3051 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3052 | ||
5531dc91 | 3053 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 3054 | return 0; |
ddbcc7e8 PM |
3055 | } |
3056 | ||
f8f22e53 TH |
3057 | /* show controllers which are enabled for a given cgroup's children */ |
3058 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3059 | { |
f8f22e53 TH |
3060 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3061 | ||
667c2491 | 3062 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
3063 | return 0; |
3064 | } | |
3065 | ||
3066 | /** | |
3067 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
3068 | * @cgrp: root of the subtree to update csses for | |
3069 | * | |
54962604 TH |
3070 | * @cgrp's control masks have changed and its subtree's css associations |
3071 | * need to be updated accordingly. This function looks up all css_sets | |
3072 | * which are attached to the subtree, creates the matching updated css_sets | |
3073 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
3074 | */ |
3075 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
3076 | { | |
3077 | LIST_HEAD(preloaded_csets); | |
10265075 | 3078 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
3079 | struct cgroup_subsys_state *d_css; |
3080 | struct cgroup *dsct; | |
f8f22e53 TH |
3081 | struct css_set *src_cset; |
3082 | int ret; | |
3083 | ||
f8f22e53 TH |
3084 | lockdep_assert_held(&cgroup_mutex); |
3085 | ||
3014dde7 TH |
3086 | percpu_down_write(&cgroup_threadgroup_rwsem); |
3087 | ||
f8f22e53 | 3088 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 3089 | spin_lock_irq(&css_set_lock); |
54962604 | 3090 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
3091 | struct cgrp_cset_link *link; |
3092 | ||
54962604 | 3093 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 3094 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
3095 | &preloaded_csets); |
3096 | } | |
82d6489d | 3097 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
3098 | |
3099 | /* NULL dst indicates self on default hierarchy */ | |
e4857982 | 3100 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
f8f22e53 TH |
3101 | if (ret) |
3102 | goto out_finish; | |
3103 | ||
82d6489d | 3104 | spin_lock_irq(&css_set_lock); |
f8f22e53 | 3105 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 3106 | struct task_struct *task, *ntask; |
f8f22e53 TH |
3107 | |
3108 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
3109 | if (!src_cset->mg_src_cgrp) | |
3110 | break; | |
3111 | ||
10265075 TH |
3112 | /* all tasks in src_csets need to be migrated */ |
3113 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
3114 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 3115 | } |
82d6489d | 3116 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 3117 | |
37ff9f8f | 3118 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
f8f22e53 TH |
3119 | out_finish: |
3120 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 3121 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
3122 | return ret; |
3123 | } | |
3124 | ||
1b9b96a1 | 3125 | /** |
945ba199 | 3126 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 3127 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
3128 | * |
3129 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
3130 | * controller while the previous css is still around. This function grabs |
3131 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 3132 | */ |
945ba199 TH |
3133 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
3134 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
3135 | { |
3136 | struct cgroup *dsct; | |
ce3f1d9d | 3137 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
3138 | struct cgroup_subsys *ss; |
3139 | int ssid; | |
3140 | ||
945ba199 TH |
3141 | restart: |
3142 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 3143 | |
ce3f1d9d | 3144 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
3145 | for_each_subsys(ss, ssid) { |
3146 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3147 | DEFINE_WAIT(wait); | |
3148 | ||
ce3f1d9d | 3149 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
3150 | continue; |
3151 | ||
3152 | cgroup_get(dsct); | |
3153 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
3154 | TASK_UNINTERRUPTIBLE); | |
3155 | ||
3156 | mutex_unlock(&cgroup_mutex); | |
3157 | schedule(); | |
3158 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
3159 | |
3160 | cgroup_put(dsct); | |
945ba199 | 3161 | goto restart; |
1b9b96a1 TH |
3162 | } |
3163 | } | |
1b9b96a1 TH |
3164 | } |
3165 | ||
15a27c36 TH |
3166 | /** |
3167 | * cgroup_save_control - save control masks of a subtree | |
3168 | * @cgrp: root of the target subtree | |
3169 | * | |
3170 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
3171 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3172 | */ | |
3173 | static void cgroup_save_control(struct cgroup *cgrp) | |
3174 | { | |
3175 | struct cgroup *dsct; | |
3176 | struct cgroup_subsys_state *d_css; | |
3177 | ||
3178 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3179 | dsct->old_subtree_control = dsct->subtree_control; | |
3180 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
3181 | } | |
3182 | } | |
3183 | ||
3184 | /** | |
3185 | * cgroup_propagate_control - refresh control masks of a subtree | |
3186 | * @cgrp: root of the target subtree | |
3187 | * | |
3188 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3189 | * ->subtree_control and propagate controller availability through the | |
3190 | * subtree so that descendants don't have unavailable controllers enabled. | |
3191 | */ | |
3192 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3193 | { | |
3194 | struct cgroup *dsct; | |
3195 | struct cgroup_subsys_state *d_css; | |
3196 | ||
3197 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3198 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3199 | dsct->subtree_ss_mask = |
3200 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3201 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3202 | } |
3203 | } | |
3204 | ||
3205 | /** | |
3206 | * cgroup_restore_control - restore control masks of a subtree | |
3207 | * @cgrp: root of the target subtree | |
3208 | * | |
3209 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3210 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3211 | */ | |
3212 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3213 | { | |
3214 | struct cgroup *dsct; | |
3215 | struct cgroup_subsys_state *d_css; | |
3216 | ||
3217 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3218 | dsct->subtree_control = dsct->old_subtree_control; | |
3219 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3220 | } | |
3221 | } | |
3222 | ||
f6d635ad TH |
3223 | static bool css_visible(struct cgroup_subsys_state *css) |
3224 | { | |
3225 | struct cgroup_subsys *ss = css->ss; | |
3226 | struct cgroup *cgrp = css->cgroup; | |
3227 | ||
3228 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3229 | return true; | |
3230 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3231 | return false; | |
3232 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3233 | } | |
3234 | ||
bdb53bd7 TH |
3235 | /** |
3236 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3237 | * @cgrp: root of the target subtree |
bdb53bd7 | 3238 | * |
ce3f1d9d | 3239 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3240 | * visible. A css is created invisible if it's being implicitly enabled |
3241 | * through dependency. An invisible css is made visible when the userland | |
3242 | * explicitly enables it. | |
3243 | * | |
3244 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3245 | * been processed already aren't cleaned up. The caller is responsible for | |
3246 | * cleaning up with cgroup_apply_control_disble(). | |
3247 | */ | |
3248 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3249 | { | |
3250 | struct cgroup *dsct; | |
ce3f1d9d | 3251 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3252 | struct cgroup_subsys *ss; |
3253 | int ssid, ret; | |
3254 | ||
ce3f1d9d | 3255 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3256 | for_each_subsys(ss, ssid) { |
3257 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3258 | ||
945ba199 TH |
3259 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3260 | ||
bdb53bd7 TH |
3261 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3262 | continue; | |
3263 | ||
3264 | if (!css) { | |
3265 | css = css_create(dsct, ss); | |
3266 | if (IS_ERR(css)) | |
3267 | return PTR_ERR(css); | |
3268 | } | |
3269 | ||
f6d635ad | 3270 | if (css_visible(css)) { |
334c3679 | 3271 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3272 | if (ret) |
3273 | return ret; | |
3274 | } | |
3275 | } | |
3276 | } | |
3277 | ||
3278 | return 0; | |
3279 | } | |
3280 | ||
12b3bb6a TH |
3281 | /** |
3282 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3283 | * @cgrp: root of the target subtree |
12b3bb6a | 3284 | * |
ce3f1d9d | 3285 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3286 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3287 | * | |
3288 | * A css is hidden when the userland requests it to be disabled while other | |
3289 | * subsystems are still depending on it. The css must not actively control | |
3290 | * resources and be in the vanilla state if it's made visible again later. | |
3291 | * Controllers which may be depended upon should provide ->css_reset() for | |
3292 | * this purpose. | |
3293 | */ | |
3294 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3295 | { | |
3296 | struct cgroup *dsct; | |
ce3f1d9d | 3297 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3298 | struct cgroup_subsys *ss; |
3299 | int ssid; | |
3300 | ||
ce3f1d9d | 3301 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3302 | for_each_subsys(ss, ssid) { |
3303 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3304 | ||
945ba199 TH |
3305 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3306 | ||
12b3bb6a TH |
3307 | if (!css) |
3308 | continue; | |
3309 | ||
334c3679 TH |
3310 | if (css->parent && |
3311 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3312 | kill_css(css); |
f6d635ad | 3313 | } else if (!css_visible(css)) { |
334c3679 | 3314 | css_clear_dir(css); |
12b3bb6a TH |
3315 | if (ss->css_reset) |
3316 | ss->css_reset(css); | |
3317 | } | |
3318 | } | |
3319 | } | |
3320 | } | |
3321 | ||
f7b2814b TH |
3322 | /** |
3323 | * cgroup_apply_control - apply control mask updates to the subtree | |
3324 | * @cgrp: root of the target subtree | |
3325 | * | |
3326 | * subsystems can be enabled and disabled in a subtree using the following | |
3327 | * steps. | |
3328 | * | |
3329 | * 1. Call cgroup_save_control() to stash the current state. | |
3330 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3331 | * 3. Call cgroup_apply_control() to apply the changes. | |
3332 | * 4. Optionally perform other related operations. | |
3333 | * 5. Call cgroup_finalize_control() to finish up. | |
3334 | * | |
3335 | * This function implements step 3 and propagates the mask changes | |
3336 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3337 | * process migrations. | |
3338 | */ | |
3339 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3340 | { | |
3341 | int ret; | |
3342 | ||
3343 | cgroup_propagate_control(cgrp); | |
3344 | ||
3345 | ret = cgroup_apply_control_enable(cgrp); | |
3346 | if (ret) | |
3347 | return ret; | |
3348 | ||
3349 | /* | |
3350 | * At this point, cgroup_e_css() results reflect the new csses | |
3351 | * making the following cgroup_update_dfl_csses() properly update | |
3352 | * css associations of all tasks in the subtree. | |
3353 | */ | |
3354 | ret = cgroup_update_dfl_csses(cgrp); | |
3355 | if (ret) | |
3356 | return ret; | |
3357 | ||
3358 | return 0; | |
3359 | } | |
3360 | ||
3361 | /** | |
3362 | * cgroup_finalize_control - finalize control mask update | |
3363 | * @cgrp: root of the target subtree | |
3364 | * @ret: the result of the update | |
3365 | * | |
3366 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3367 | */ | |
3368 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3369 | { | |
3370 | if (ret) { | |
3371 | cgroup_restore_control(cgrp); | |
3372 | cgroup_propagate_control(cgrp); | |
3373 | } | |
3374 | ||
3375 | cgroup_apply_control_disable(cgrp); | |
3376 | } | |
3377 | ||
f8f22e53 | 3378 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3379 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3380 | char *buf, size_t nbytes, | |
3381 | loff_t off) | |
f8f22e53 | 3382 | { |
6e5c8307 | 3383 | u16 enable = 0, disable = 0; |
a9746d8d | 3384 | struct cgroup *cgrp, *child; |
f8f22e53 | 3385 | struct cgroup_subsys *ss; |
451af504 | 3386 | char *tok; |
f8f22e53 TH |
3387 | int ssid, ret; |
3388 | ||
3389 | /* | |
d37167ab TH |
3390 | * Parse input - space separated list of subsystem names prefixed |
3391 | * with either + or -. | |
f8f22e53 | 3392 | */ |
451af504 TH |
3393 | buf = strstrip(buf); |
3394 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3395 | if (tok[0] == '\0') |
3396 | continue; | |
a7165264 | 3397 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3398 | if (!cgroup_ssid_enabled(ssid) || |
3399 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3400 | continue; |
3401 | ||
3402 | if (*tok == '+') { | |
7d331fa9 TH |
3403 | enable |= 1 << ssid; |
3404 | disable &= ~(1 << ssid); | |
f8f22e53 | 3405 | } else if (*tok == '-') { |
7d331fa9 TH |
3406 | disable |= 1 << ssid; |
3407 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3408 | } else { |
3409 | return -EINVAL; | |
3410 | } | |
3411 | break; | |
b4e0eeaf | 3412 | } while_each_subsys_mask(); |
f8f22e53 TH |
3413 | if (ssid == CGROUP_SUBSYS_COUNT) |
3414 | return -EINVAL; | |
3415 | } | |
3416 | ||
945ba199 | 3417 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3418 | if (!cgrp) |
3419 | return -ENODEV; | |
f8f22e53 TH |
3420 | |
3421 | for_each_subsys(ss, ssid) { | |
3422 | if (enable & (1 << ssid)) { | |
667c2491 | 3423 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3424 | enable &= ~(1 << ssid); |
3425 | continue; | |
3426 | } | |
3427 | ||
5531dc91 | 3428 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3429 | ret = -ENOENT; |
3430 | goto out_unlock; | |
3431 | } | |
f8f22e53 | 3432 | } else if (disable & (1 << ssid)) { |
667c2491 | 3433 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3434 | disable &= ~(1 << ssid); |
3435 | continue; | |
3436 | } | |
3437 | ||
3438 | /* a child has it enabled? */ | |
3439 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3440 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3441 | ret = -EBUSY; |
ddab2b6e | 3442 | goto out_unlock; |
f8f22e53 TH |
3443 | } |
3444 | } | |
3445 | } | |
3446 | } | |
3447 | ||
3448 | if (!enable && !disable) { | |
3449 | ret = 0; | |
ddab2b6e | 3450 | goto out_unlock; |
f8f22e53 TH |
3451 | } |
3452 | ||
3453 | /* | |
667c2491 | 3454 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3455 | * with tasks so that child cgroups don't compete against tasks. |
3456 | */ | |
d51f39b0 | 3457 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3458 | ret = -EBUSY; |
3459 | goto out_unlock; | |
3460 | } | |
3461 | ||
15a27c36 TH |
3462 | /* save and update control masks and prepare csses */ |
3463 | cgroup_save_control(cgrp); | |
f63070d3 | 3464 | |
15a27c36 TH |
3465 | cgrp->subtree_control |= enable; |
3466 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3467 | |
f7b2814b | 3468 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3469 | |
f7b2814b | 3470 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3471 | |
3472 | kernfs_activate(cgrp->kn); | |
3473 | ret = 0; | |
3474 | out_unlock: | |
a9746d8d | 3475 | cgroup_kn_unlock(of->kn); |
451af504 | 3476 | return ret ?: nbytes; |
f8f22e53 TH |
3477 | } |
3478 | ||
4a07c222 | 3479 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3480 | { |
4a07c222 | 3481 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3482 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3483 | return 0; |
3484 | } | |
3485 | ||
2bd59d48 TH |
3486 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3487 | size_t nbytes, loff_t off) | |
355e0c48 | 3488 | { |
2bd59d48 TH |
3489 | struct cgroup *cgrp = of->kn->parent->priv; |
3490 | struct cftype *cft = of->kn->priv; | |
3491 | struct cgroup_subsys_state *css; | |
a742c59d | 3492 | int ret; |
355e0c48 | 3493 | |
b4168640 TH |
3494 | if (cft->write) |
3495 | return cft->write(of, buf, nbytes, off); | |
3496 | ||
2bd59d48 TH |
3497 | /* |
3498 | * kernfs guarantees that a file isn't deleted with operations in | |
3499 | * flight, which means that the matching css is and stays alive and | |
3500 | * doesn't need to be pinned. The RCU locking is not necessary | |
3501 | * either. It's just for the convenience of using cgroup_css(). | |
3502 | */ | |
3503 | rcu_read_lock(); | |
3504 | css = cgroup_css(cgrp, cft->ss); | |
3505 | rcu_read_unlock(); | |
a742c59d | 3506 | |
451af504 | 3507 | if (cft->write_u64) { |
a742c59d TH |
3508 | unsigned long long v; |
3509 | ret = kstrtoull(buf, 0, &v); | |
3510 | if (!ret) | |
3511 | ret = cft->write_u64(css, cft, v); | |
3512 | } else if (cft->write_s64) { | |
3513 | long long v; | |
3514 | ret = kstrtoll(buf, 0, &v); | |
3515 | if (!ret) | |
3516 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3517 | } else { |
a742c59d | 3518 | ret = -EINVAL; |
e73d2c61 | 3519 | } |
2bd59d48 | 3520 | |
a742c59d | 3521 | return ret ?: nbytes; |
355e0c48 PM |
3522 | } |
3523 | ||
6612f05b | 3524 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3525 | { |
2bd59d48 | 3526 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3527 | } |
3528 | ||
6612f05b | 3529 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3530 | { |
2bd59d48 | 3531 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3532 | } |
3533 | ||
6612f05b | 3534 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3535 | { |
2bd59d48 | 3536 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3537 | } |
3538 | ||
91796569 | 3539 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3540 | { |
7da11279 TH |
3541 | struct cftype *cft = seq_cft(m); |
3542 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3543 | |
2da8ca82 TH |
3544 | if (cft->seq_show) |
3545 | return cft->seq_show(m, arg); | |
e73d2c61 | 3546 | |
f4c753b7 | 3547 | if (cft->read_u64) |
896f5199 TH |
3548 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3549 | else if (cft->read_s64) | |
3550 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3551 | else | |
3552 | return -EINVAL; | |
3553 | return 0; | |
91796569 PM |
3554 | } |
3555 | ||
2bd59d48 TH |
3556 | static struct kernfs_ops cgroup_kf_single_ops = { |
3557 | .atomic_write_len = PAGE_SIZE, | |
3558 | .write = cgroup_file_write, | |
3559 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3560 | }; |
3561 | ||
2bd59d48 TH |
3562 | static struct kernfs_ops cgroup_kf_ops = { |
3563 | .atomic_write_len = PAGE_SIZE, | |
3564 | .write = cgroup_file_write, | |
3565 | .seq_start = cgroup_seqfile_start, | |
3566 | .seq_next = cgroup_seqfile_next, | |
3567 | .seq_stop = cgroup_seqfile_stop, | |
3568 | .seq_show = cgroup_seqfile_show, | |
3569 | }; | |
ddbcc7e8 PM |
3570 | |
3571 | /* | |
3572 | * cgroup_rename - Only allow simple rename of directories in place. | |
3573 | */ | |
2bd59d48 TH |
3574 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3575 | const char *new_name_str) | |
ddbcc7e8 | 3576 | { |
2bd59d48 | 3577 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3578 | int ret; |
65dff759 | 3579 | |
2bd59d48 | 3580 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3581 | return -ENOTDIR; |
2bd59d48 | 3582 | if (kn->parent != new_parent) |
ddbcc7e8 | 3583 | return -EIO; |
65dff759 | 3584 | |
6db8e85c TH |
3585 | /* |
3586 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3587 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3588 | */ |
aa6ec29b | 3589 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3590 | return -EPERM; |
099fca32 | 3591 | |
e1b2dc17 | 3592 | /* |
8353da1f | 3593 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3594 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3595 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3596 | */ |
3597 | kernfs_break_active_protection(new_parent); | |
3598 | kernfs_break_active_protection(kn); | |
099fca32 | 3599 | |
2bd59d48 | 3600 | mutex_lock(&cgroup_mutex); |
099fca32 | 3601 | |
2bd59d48 | 3602 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3603 | |
2bd59d48 | 3604 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3605 | |
3606 | kernfs_unbreak_active_protection(kn); | |
3607 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3608 | return ret; |
099fca32 LZ |
3609 | } |
3610 | ||
49957f8e TH |
3611 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3612 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3613 | { | |
3614 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3615 | .ia_uid = current_fsuid(), | |
3616 | .ia_gid = current_fsgid(), }; | |
3617 | ||
3618 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3619 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3620 | return 0; | |
3621 | ||
3622 | return kernfs_setattr(kn, &iattr); | |
3623 | } | |
3624 | ||
4df8dc90 TH |
3625 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3626 | struct cftype *cft) | |
ddbcc7e8 | 3627 | { |
8d7e6fb0 | 3628 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3629 | struct kernfs_node *kn; |
3630 | struct lock_class_key *key = NULL; | |
49957f8e | 3631 | int ret; |
05ef1d7c | 3632 | |
2bd59d48 TH |
3633 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3634 | key = &cft->lockdep_key; | |
3635 | #endif | |
3636 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3637 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3638 | NULL, key); |
49957f8e TH |
3639 | if (IS_ERR(kn)) |
3640 | return PTR_ERR(kn); | |
3641 | ||
3642 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3643 | if (ret) { |
49957f8e | 3644 | kernfs_remove(kn); |
f8f22e53 TH |
3645 | return ret; |
3646 | } | |
3647 | ||
6f60eade TH |
3648 | if (cft->file_offset) { |
3649 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3650 | ||
34c06254 | 3651 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3652 | cfile->kn = kn; |
34c06254 | 3653 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3654 | } |
3655 | ||
f8f22e53 | 3656 | return 0; |
ddbcc7e8 PM |
3657 | } |
3658 | ||
b1f28d31 TH |
3659 | /** |
3660 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3661 | * @css: the target css |
3662 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3663 | * @cfts: array of cftypes to be added |
3664 | * @is_add: whether to add or remove | |
3665 | * | |
3666 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3667 | * For removals, this function never fails. |
b1f28d31 | 3668 | */ |
4df8dc90 TH |
3669 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3670 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3671 | bool is_add) |
ddbcc7e8 | 3672 | { |
6732ed85 | 3673 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3674 | int ret = 0; |
b1f28d31 | 3675 | |
01f6474c | 3676 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3677 | |
6732ed85 TH |
3678 | restart: |
3679 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3680 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3681 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3682 | continue; |
05ebb6e6 | 3683 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3684 | continue; |
d51f39b0 | 3685 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3686 | continue; |
d51f39b0 | 3687 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3688 | continue; |
3689 | ||
2739d3cc | 3690 | if (is_add) { |
4df8dc90 | 3691 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3692 | if (ret) { |
ed3d261b JP |
3693 | pr_warn("%s: failed to add %s, err=%d\n", |
3694 | __func__, cft->name, ret); | |
6732ed85 TH |
3695 | cft_end = cft; |
3696 | is_add = false; | |
3697 | goto restart; | |
b1f28d31 | 3698 | } |
2739d3cc LZ |
3699 | } else { |
3700 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3701 | } |
ddbcc7e8 | 3702 | } |
b598dde3 | 3703 | return ret; |
ddbcc7e8 PM |
3704 | } |
3705 | ||
21a2d343 | 3706 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3707 | { |
3708 | LIST_HEAD(pending); | |
2bb566cb | 3709 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3710 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3711 | struct cgroup_subsys_state *css; |
9ccece80 | 3712 | int ret = 0; |
8e3f6541 | 3713 | |
01f6474c | 3714 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3715 | |
e8c82d20 | 3716 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3717 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3718 | struct cgroup *cgrp = css->cgroup; |
3719 | ||
88cb04b9 | 3720 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3721 | continue; |
3722 | ||
4df8dc90 | 3723 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3724 | if (ret) |
3725 | break; | |
8e3f6541 | 3726 | } |
21a2d343 TH |
3727 | |
3728 | if (is_add && !ret) | |
3729 | kernfs_activate(root->kn); | |
9ccece80 | 3730 | return ret; |
8e3f6541 TH |
3731 | } |
3732 | ||
2da440a2 | 3733 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3734 | { |
2bb566cb | 3735 | struct cftype *cft; |
8e3f6541 | 3736 | |
2bd59d48 TH |
3737 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3738 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3739 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3740 | kfree(cft->kf_ops); | |
3741 | cft->kf_ops = NULL; | |
2da440a2 | 3742 | cft->ss = NULL; |
a8ddc821 TH |
3743 | |
3744 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3745 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3746 | } |
2da440a2 TH |
3747 | } |
3748 | ||
2bd59d48 | 3749 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3750 | { |
3751 | struct cftype *cft; | |
3752 | ||
2bd59d48 TH |
3753 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3754 | struct kernfs_ops *kf_ops; | |
3755 | ||
0adb0704 TH |
3756 | WARN_ON(cft->ss || cft->kf_ops); |
3757 | ||
2bd59d48 TH |
3758 | if (cft->seq_start) |
3759 | kf_ops = &cgroup_kf_ops; | |
3760 | else | |
3761 | kf_ops = &cgroup_kf_single_ops; | |
3762 | ||
3763 | /* | |
3764 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3765 | * make a copy of kf_ops to set its atomic_write_len. | |
3766 | */ | |
3767 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3768 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3769 | if (!kf_ops) { | |
3770 | cgroup_exit_cftypes(cfts); | |
3771 | return -ENOMEM; | |
3772 | } | |
3773 | kf_ops->atomic_write_len = cft->max_write_len; | |
3774 | } | |
8e3f6541 | 3775 | |
2bd59d48 | 3776 | cft->kf_ops = kf_ops; |
2bb566cb | 3777 | cft->ss = ss; |
2bd59d48 | 3778 | } |
2bb566cb | 3779 | |
2bd59d48 | 3780 | return 0; |
2da440a2 TH |
3781 | } |
3782 | ||
21a2d343 TH |
3783 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3784 | { | |
01f6474c | 3785 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3786 | |
3787 | if (!cfts || !cfts[0].ss) | |
3788 | return -ENOENT; | |
3789 | ||
3790 | list_del(&cfts->node); | |
3791 | cgroup_apply_cftypes(cfts, false); | |
3792 | cgroup_exit_cftypes(cfts); | |
3793 | return 0; | |
8e3f6541 | 3794 | } |
8e3f6541 | 3795 | |
79578621 TH |
3796 | /** |
3797 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3798 | * @cfts: zero-length name terminated array of cftypes |
3799 | * | |
2bb566cb TH |
3800 | * Unregister @cfts. Files described by @cfts are removed from all |
3801 | * existing cgroups and all future cgroups won't have them either. This | |
3802 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3803 | * |
3804 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3805 | * registered. |
79578621 | 3806 | */ |
2bb566cb | 3807 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3808 | { |
21a2d343 | 3809 | int ret; |
79578621 | 3810 | |
01f6474c | 3811 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3812 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3813 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3814 | return ret; |
80b13586 TH |
3815 | } |
3816 | ||
8e3f6541 TH |
3817 | /** |
3818 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3819 | * @ss: target cgroup subsystem | |
3820 | * @cfts: zero-length name terminated array of cftypes | |
3821 | * | |
3822 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3823 | * existing cgroups to which @ss is attached and all future cgroups will | |
3824 | * have them too. This function can be called anytime whether @ss is | |
3825 | * attached or not. | |
3826 | * | |
3827 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3828 | * function currently returns 0 as long as @cfts registration is successful | |
3829 | * even if some file creation attempts on existing cgroups fail. | |
3830 | */ | |
2cf669a5 | 3831 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3832 | { |
9ccece80 | 3833 | int ret; |
8e3f6541 | 3834 | |
fc5ed1e9 | 3835 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3836 | return 0; |
3837 | ||
dc5736ed LZ |
3838 | if (!cfts || cfts[0].name[0] == '\0') |
3839 | return 0; | |
2bb566cb | 3840 | |
2bd59d48 TH |
3841 | ret = cgroup_init_cftypes(ss, cfts); |
3842 | if (ret) | |
3843 | return ret; | |
79578621 | 3844 | |
01f6474c | 3845 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3846 | |
0adb0704 | 3847 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3848 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3849 | if (ret) |
21a2d343 | 3850 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3851 | |
01f6474c | 3852 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3853 | return ret; |
79578621 TH |
3854 | } |
3855 | ||
a8ddc821 TH |
3856 | /** |
3857 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3858 | * @ss: target cgroup subsystem | |
3859 | * @cfts: zero-length name terminated array of cftypes | |
3860 | * | |
3861 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3862 | * the default hierarchy. | |
3863 | */ | |
3864 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3865 | { | |
3866 | struct cftype *cft; | |
3867 | ||
3868 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3869 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3870 | return cgroup_add_cftypes(ss, cfts); |
3871 | } | |
3872 | ||
3873 | /** | |
3874 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3875 | * @ss: target cgroup subsystem | |
3876 | * @cfts: zero-length name terminated array of cftypes | |
3877 | * | |
3878 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3879 | * the legacy hierarchies. | |
3880 | */ | |
2cf669a5 TH |
3881 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3882 | { | |
a8ddc821 TH |
3883 | struct cftype *cft; |
3884 | ||
e4b7037c TH |
3885 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3886 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3887 | return cgroup_add_cftypes(ss, cfts); |
3888 | } | |
3889 | ||
34c06254 TH |
3890 | /** |
3891 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3892 | * @cfile: target cgroup_file | |
3893 | * | |
3894 | * @cfile must have been obtained by setting cftype->file_offset. | |
3895 | */ | |
3896 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3897 | { | |
3898 | unsigned long flags; | |
3899 | ||
3900 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3901 | if (cfile->kn) | |
3902 | kernfs_notify(cfile->kn); | |
3903 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3904 | } | |
3905 | ||
a043e3b2 LZ |
3906 | /** |
3907 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3908 | * @cgrp: the cgroup in question | |
3909 | * | |
3910 | * Return the number of tasks in the cgroup. | |
3911 | */ | |
07bc356e | 3912 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3913 | { |
3914 | int count = 0; | |
69d0206c | 3915 | struct cgrp_cset_link *link; |
817929ec | 3916 | |
82d6489d | 3917 | spin_lock_irq(&css_set_lock); |
69d0206c TH |
3918 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3919 | count += atomic_read(&link->cset->refcount); | |
82d6489d | 3920 | spin_unlock_irq(&css_set_lock); |
bbcb81d0 PM |
3921 | return count; |
3922 | } | |
3923 | ||
53fa5261 | 3924 | /** |
492eb21b | 3925 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3926 | * @pos: the current position (%NULL to initiate traversal) |
3927 | * @parent: css whose children to walk | |
53fa5261 | 3928 | * |
c2931b70 | 3929 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3930 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3931 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3932 | * be returned regardless of their states. | |
3933 | * | |
3934 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3935 | * css which finished ->css_online() is guaranteed to be visible in the | |
3936 | * future iterations and will stay visible until the last reference is put. | |
3937 | * A css which hasn't finished ->css_online() or already finished | |
3938 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3939 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3940 | */ |
c2931b70 TH |
3941 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3942 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3943 | { |
c2931b70 | 3944 | struct cgroup_subsys_state *next; |
53fa5261 | 3945 | |
8353da1f | 3946 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3947 | |
3948 | /* | |
de3f0341 TH |
3949 | * @pos could already have been unlinked from the sibling list. |
3950 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3951 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3952 | * @pos is taken off list, at which time its next pointer is valid, | |
3953 | * and, as releases are serialized, the one pointed to by the next | |
3954 | * pointer is guaranteed to not have started release yet. This | |
3955 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3956 | * critical section, the one pointed to by its next pointer is | |
3957 | * guaranteed to not have finished its RCU grace period even if we | |
3958 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3959 | * |
de3f0341 TH |
3960 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3961 | * dereferenced; however, as each css is given a monotonically | |
3962 | * increasing unique serial number and always appended to the | |
3963 | * sibling list, the next one can be found by walking the parent's | |
3964 | * children until the first css with higher serial number than | |
3965 | * @pos's. While this path can be slower, it happens iff iteration | |
3966 | * races against release and the race window is very small. | |
53fa5261 | 3967 | */ |
3b287a50 | 3968 | if (!pos) { |
c2931b70 TH |
3969 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3970 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3971 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3972 | } else { |
c2931b70 | 3973 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3974 | if (next->serial_nr > pos->serial_nr) |
3975 | break; | |
53fa5261 TH |
3976 | } |
3977 | ||
3b281afb TH |
3978 | /* |
3979 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3980 | * the next sibling. |
3b281afb | 3981 | */ |
c2931b70 TH |
3982 | if (&next->sibling != &parent->children) |
3983 | return next; | |
3b281afb | 3984 | return NULL; |
53fa5261 | 3985 | } |
53fa5261 | 3986 | |
574bd9f7 | 3987 | /** |
492eb21b | 3988 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3989 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3990 | * @root: css whose descendants to walk |
574bd9f7 | 3991 | * |
492eb21b | 3992 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3993 | * to visit for pre-order traversal of @root's descendants. @root is |
3994 | * included in the iteration and the first node to be visited. | |
75501a6d | 3995 | * |
87fb54f1 TH |
3996 | * While this function requires cgroup_mutex or RCU read locking, it |
3997 | * doesn't require the whole traversal to be contained in a single critical | |
3998 | * section. This function will return the correct next descendant as long | |
3999 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
4000 | * |
4001 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4002 | * css which finished ->css_online() is guaranteed to be visible in the | |
4003 | * future iterations and will stay visible until the last reference is put. | |
4004 | * A css which hasn't finished ->css_online() or already finished | |
4005 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4006 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4007 | */ |
492eb21b TH |
4008 | struct cgroup_subsys_state * |
4009 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
4010 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4011 | { |
492eb21b | 4012 | struct cgroup_subsys_state *next; |
574bd9f7 | 4013 | |
8353da1f | 4014 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4015 | |
bd8815a6 | 4016 | /* if first iteration, visit @root */ |
7805d000 | 4017 | if (!pos) |
bd8815a6 | 4018 | return root; |
574bd9f7 TH |
4019 | |
4020 | /* visit the first child if exists */ | |
492eb21b | 4021 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4022 | if (next) |
4023 | return next; | |
4024 | ||
4025 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4026 | while (pos != root) { |
5c9d535b | 4027 | next = css_next_child(pos, pos->parent); |
75501a6d | 4028 | if (next) |
574bd9f7 | 4029 | return next; |
5c9d535b | 4030 | pos = pos->parent; |
7805d000 | 4031 | } |
574bd9f7 TH |
4032 | |
4033 | return NULL; | |
4034 | } | |
574bd9f7 | 4035 | |
12a9d2fe | 4036 | /** |
492eb21b TH |
4037 | * css_rightmost_descendant - return the rightmost descendant of a css |
4038 | * @pos: css of interest | |
12a9d2fe | 4039 | * |
492eb21b TH |
4040 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4041 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4042 | * subtree of @pos. |
75501a6d | 4043 | * |
87fb54f1 TH |
4044 | * While this function requires cgroup_mutex or RCU read locking, it |
4045 | * doesn't require the whole traversal to be contained in a single critical | |
4046 | * section. This function will return the correct rightmost descendant as | |
4047 | * long as @pos is accessible. | |
12a9d2fe | 4048 | */ |
492eb21b TH |
4049 | struct cgroup_subsys_state * |
4050 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4051 | { |
492eb21b | 4052 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4053 | |
8353da1f | 4054 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4055 | |
4056 | do { | |
4057 | last = pos; | |
4058 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4059 | pos = NULL; | |
492eb21b | 4060 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4061 | pos = tmp; |
4062 | } while (pos); | |
4063 | ||
4064 | return last; | |
4065 | } | |
12a9d2fe | 4066 | |
492eb21b TH |
4067 | static struct cgroup_subsys_state * |
4068 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4069 | { |
492eb21b | 4070 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4071 | |
4072 | do { | |
4073 | last = pos; | |
492eb21b | 4074 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4075 | } while (pos); |
4076 | ||
4077 | return last; | |
4078 | } | |
4079 | ||
4080 | /** | |
492eb21b | 4081 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4082 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4083 | * @root: css whose descendants to walk |
574bd9f7 | 4084 | * |
492eb21b | 4085 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4086 | * to visit for post-order traversal of @root's descendants. @root is |
4087 | * included in the iteration and the last node to be visited. | |
75501a6d | 4088 | * |
87fb54f1 TH |
4089 | * While this function requires cgroup_mutex or RCU read locking, it |
4090 | * doesn't require the whole traversal to be contained in a single critical | |
4091 | * section. This function will return the correct next descendant as long | |
4092 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4093 | * @cgroup. | |
c2931b70 TH |
4094 | * |
4095 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4096 | * css which finished ->css_online() is guaranteed to be visible in the | |
4097 | * future iterations and will stay visible until the last reference is put. | |
4098 | * A css which hasn't finished ->css_online() or already finished | |
4099 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4100 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4101 | */ |
492eb21b TH |
4102 | struct cgroup_subsys_state * |
4103 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4104 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4105 | { |
492eb21b | 4106 | struct cgroup_subsys_state *next; |
574bd9f7 | 4107 | |
8353da1f | 4108 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4109 | |
58b79a91 TH |
4110 | /* if first iteration, visit leftmost descendant which may be @root */ |
4111 | if (!pos) | |
4112 | return css_leftmost_descendant(root); | |
574bd9f7 | 4113 | |
bd8815a6 TH |
4114 | /* if we visited @root, we're done */ |
4115 | if (pos == root) | |
4116 | return NULL; | |
4117 | ||
574bd9f7 | 4118 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4119 | next = css_next_child(pos, pos->parent); |
75501a6d | 4120 | if (next) |
492eb21b | 4121 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4122 | |
4123 | /* no sibling left, visit parent */ | |
5c9d535b | 4124 | return pos->parent; |
574bd9f7 | 4125 | } |
574bd9f7 | 4126 | |
f3d46500 TH |
4127 | /** |
4128 | * css_has_online_children - does a css have online children | |
4129 | * @css: the target css | |
4130 | * | |
4131 | * Returns %true if @css has any online children; otherwise, %false. This | |
4132 | * function can be called from any context but the caller is responsible | |
4133 | * for synchronizing against on/offlining as necessary. | |
4134 | */ | |
4135 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4136 | { |
f3d46500 TH |
4137 | struct cgroup_subsys_state *child; |
4138 | bool ret = false; | |
cbc125ef TH |
4139 | |
4140 | rcu_read_lock(); | |
f3d46500 | 4141 | css_for_each_child(child, css) { |
99bae5f9 | 4142 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4143 | ret = true; |
4144 | break; | |
cbc125ef TH |
4145 | } |
4146 | } | |
4147 | rcu_read_unlock(); | |
f3d46500 | 4148 | return ret; |
574bd9f7 | 4149 | } |
574bd9f7 | 4150 | |
0942eeee | 4151 | /** |
ecb9d535 | 4152 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4153 | * @it: the iterator to advance |
4154 | * | |
4155 | * Advance @it to the next css_set to walk. | |
d515876e | 4156 | */ |
ecb9d535 | 4157 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4158 | { |
0f0a2b4f | 4159 | struct list_head *l = it->cset_pos; |
d515876e TH |
4160 | struct cgrp_cset_link *link; |
4161 | struct css_set *cset; | |
4162 | ||
f0d9a5f1 | 4163 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4164 | |
d515876e TH |
4165 | /* Advance to the next non-empty css_set */ |
4166 | do { | |
4167 | l = l->next; | |
0f0a2b4f TH |
4168 | if (l == it->cset_head) { |
4169 | it->cset_pos = NULL; | |
ecb9d535 | 4170 | it->task_pos = NULL; |
d515876e TH |
4171 | return; |
4172 | } | |
3ebb2b6e TH |
4173 | |
4174 | if (it->ss) { | |
4175 | cset = container_of(l, struct css_set, | |
4176 | e_cset_node[it->ss->id]); | |
4177 | } else { | |
4178 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4179 | cset = link->cset; | |
4180 | } | |
0de0942d | 4181 | } while (!css_set_populated(cset)); |
c7561128 | 4182 | |
0f0a2b4f | 4183 | it->cset_pos = l; |
c7561128 TH |
4184 | |
4185 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 4186 | it->task_pos = cset->tasks.next; |
c7561128 | 4187 | else |
0f0a2b4f TH |
4188 | it->task_pos = cset->mg_tasks.next; |
4189 | ||
4190 | it->tasks_head = &cset->tasks; | |
4191 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4192 | |
4193 | /* | |
4194 | * We don't keep css_sets locked across iteration steps and thus | |
4195 | * need to take steps to ensure that iteration can be resumed after | |
4196 | * the lock is re-acquired. Iteration is performed at two levels - | |
4197 | * css_sets and tasks in them. | |
4198 | * | |
4199 | * Once created, a css_set never leaves its cgroup lists, so a | |
4200 | * pinned css_set is guaranteed to stay put and we can resume | |
4201 | * iteration afterwards. | |
4202 | * | |
4203 | * Tasks may leave @cset across iteration steps. This is resolved | |
4204 | * by registering each iterator with the css_set currently being | |
4205 | * walked and making css_set_move_task() advance iterators whose | |
4206 | * next task is leaving. | |
4207 | */ | |
4208 | if (it->cur_cset) { | |
4209 | list_del(&it->iters_node); | |
4210 | put_css_set_locked(it->cur_cset); | |
4211 | } | |
4212 | get_css_set(cset); | |
4213 | it->cur_cset = cset; | |
4214 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4215 | } |
4216 | ||
ecb9d535 TH |
4217 | static void css_task_iter_advance(struct css_task_iter *it) |
4218 | { | |
4219 | struct list_head *l = it->task_pos; | |
4220 | ||
f0d9a5f1 | 4221 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4222 | WARN_ON_ONCE(!l); |
4223 | ||
4224 | /* | |
4225 | * Advance iterator to find next entry. cset->tasks is consumed | |
4226 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4227 | * next cset. | |
4228 | */ | |
4229 | l = l->next; | |
4230 | ||
4231 | if (l == it->tasks_head) | |
4232 | l = it->mg_tasks_head->next; | |
4233 | ||
4234 | if (l == it->mg_tasks_head) | |
4235 | css_task_iter_advance_css_set(it); | |
4236 | else | |
4237 | it->task_pos = l; | |
4238 | } | |
4239 | ||
0942eeee | 4240 | /** |
72ec7029 TH |
4241 | * css_task_iter_start - initiate task iteration |
4242 | * @css: the css to walk tasks of | |
0942eeee TH |
4243 | * @it: the task iterator to use |
4244 | * | |
72ec7029 TH |
4245 | * Initiate iteration through the tasks of @css. The caller can call |
4246 | * css_task_iter_next() to walk through the tasks until the function | |
4247 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4248 | * called. | |
0942eeee | 4249 | */ |
72ec7029 TH |
4250 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4251 | struct css_task_iter *it) | |
817929ec | 4252 | { |
56fde9e0 TH |
4253 | /* no one should try to iterate before mounting cgroups */ |
4254 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4255 | |
ed27b9f7 TH |
4256 | memset(it, 0, sizeof(*it)); |
4257 | ||
82d6489d | 4258 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4259 | |
3ebb2b6e TH |
4260 | it->ss = css->ss; |
4261 | ||
4262 | if (it->ss) | |
4263 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4264 | else | |
4265 | it->cset_pos = &css->cgroup->cset_links; | |
4266 | ||
0f0a2b4f | 4267 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4268 | |
ecb9d535 | 4269 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4270 | |
82d6489d | 4271 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4272 | } |
4273 | ||
0942eeee | 4274 | /** |
72ec7029 | 4275 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4276 | * @it: the task iterator being iterated |
4277 | * | |
4278 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4279 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4280 | * reaches the end. | |
0942eeee | 4281 | */ |
72ec7029 | 4282 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4283 | { |
d5745675 | 4284 | if (it->cur_task) { |
ed27b9f7 | 4285 | put_task_struct(it->cur_task); |
d5745675 TH |
4286 | it->cur_task = NULL; |
4287 | } | |
ed27b9f7 | 4288 | |
82d6489d | 4289 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4290 | |
d5745675 TH |
4291 | if (it->task_pos) { |
4292 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4293 | cg_list); | |
4294 | get_task_struct(it->cur_task); | |
4295 | css_task_iter_advance(it); | |
4296 | } | |
ed27b9f7 | 4297 | |
82d6489d | 4298 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4299 | |
4300 | return it->cur_task; | |
817929ec PM |
4301 | } |
4302 | ||
0942eeee | 4303 | /** |
72ec7029 | 4304 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4305 | * @it: the task iterator to finish |
4306 | * | |
72ec7029 | 4307 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4308 | */ |
72ec7029 | 4309 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4310 | { |
ed27b9f7 | 4311 | if (it->cur_cset) { |
82d6489d | 4312 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4313 | list_del(&it->iters_node); |
4314 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4315 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4316 | } |
4317 | ||
4318 | if (it->cur_task) | |
4319 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4320 | } |
4321 | ||
4322 | /** | |
8cc99345 TH |
4323 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4324 | * @to: cgroup to which the tasks will be moved | |
4325 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4326 | * |
eaf797ab TH |
4327 | * Locking rules between cgroup_post_fork() and the migration path |
4328 | * guarantee that, if a task is forking while being migrated, the new child | |
4329 | * is guaranteed to be either visible in the source cgroup after the | |
4330 | * parent's migration is complete or put into the target cgroup. No task | |
4331 | * can slip out of migration through forking. | |
31a7df01 | 4332 | */ |
8cc99345 | 4333 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4334 | { |
952aaa12 TH |
4335 | LIST_HEAD(preloaded_csets); |
4336 | struct cgrp_cset_link *link; | |
72ec7029 | 4337 | struct css_task_iter it; |
e406d1cf | 4338 | struct task_struct *task; |
952aaa12 | 4339 | int ret; |
31a7df01 | 4340 | |
6c694c88 TH |
4341 | if (!cgroup_may_migrate_to(to)) |
4342 | return -EBUSY; | |
4343 | ||
952aaa12 | 4344 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4345 | |
952aaa12 | 4346 | /* all tasks in @from are being moved, all csets are source */ |
82d6489d | 4347 | spin_lock_irq(&css_set_lock); |
952aaa12 TH |
4348 | list_for_each_entry(link, &from->cset_links, cset_link) |
4349 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
82d6489d | 4350 | spin_unlock_irq(&css_set_lock); |
31a7df01 | 4351 | |
e4857982 | 4352 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
952aaa12 TH |
4353 | if (ret) |
4354 | goto out_err; | |
8cc99345 | 4355 | |
952aaa12 | 4356 | /* |
2cfa2b19 | 4357 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4358 | * ->can_attach() fails. |
4359 | */ | |
e406d1cf | 4360 | do { |
9d800df1 | 4361 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4362 | task = css_task_iter_next(&it); |
4363 | if (task) | |
4364 | get_task_struct(task); | |
4365 | css_task_iter_end(&it); | |
4366 | ||
4367 | if (task) { | |
37ff9f8f | 4368 | ret = cgroup_migrate(task, false, to->root); |
e406d1cf TH |
4369 | put_task_struct(task); |
4370 | } | |
4371 | } while (task && !ret); | |
952aaa12 TH |
4372 | out_err: |
4373 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4374 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4375 | return ret; |
8cc99345 TH |
4376 | } |
4377 | ||
bbcb81d0 | 4378 | /* |
102a775e | 4379 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4380 | * |
4381 | * Reading this file can return large amounts of data if a cgroup has | |
4382 | * *lots* of attached tasks. So it may need several calls to read(), | |
4383 | * but we cannot guarantee that the information we produce is correct | |
4384 | * unless we produce it entirely atomically. | |
4385 | * | |
bbcb81d0 | 4386 | */ |
bbcb81d0 | 4387 | |
24528255 LZ |
4388 | /* which pidlist file are we talking about? */ |
4389 | enum cgroup_filetype { | |
4390 | CGROUP_FILE_PROCS, | |
4391 | CGROUP_FILE_TASKS, | |
4392 | }; | |
4393 | ||
4394 | /* | |
4395 | * A pidlist is a list of pids that virtually represents the contents of one | |
4396 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4397 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4398 | * to the cgroup. | |
4399 | */ | |
4400 | struct cgroup_pidlist { | |
4401 | /* | |
4402 | * used to find which pidlist is wanted. doesn't change as long as | |
4403 | * this particular list stays in the list. | |
4404 | */ | |
4405 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4406 | /* array of xids */ | |
4407 | pid_t *list; | |
4408 | /* how many elements the above list has */ | |
4409 | int length; | |
24528255 LZ |
4410 | /* each of these stored in a list by its cgroup */ |
4411 | struct list_head links; | |
4412 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4413 | struct cgroup *owner; | |
b1a21367 TH |
4414 | /* for delayed destruction */ |
4415 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4416 | }; |
4417 | ||
d1d9fd33 BB |
4418 | /* |
4419 | * The following two functions "fix" the issue where there are more pids | |
4420 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4421 | * TODO: replace with a kernel-wide solution to this problem | |
4422 | */ | |
4423 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4424 | static void *pidlist_allocate(int count) | |
4425 | { | |
4426 | if (PIDLIST_TOO_LARGE(count)) | |
4427 | return vmalloc(count * sizeof(pid_t)); | |
4428 | else | |
4429 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4430 | } | |
b1a21367 | 4431 | |
d1d9fd33 BB |
4432 | static void pidlist_free(void *p) |
4433 | { | |
58794514 | 4434 | kvfree(p); |
d1d9fd33 | 4435 | } |
d1d9fd33 | 4436 | |
b1a21367 TH |
4437 | /* |
4438 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4439 | * should be left afterwards. | |
4440 | */ | |
4441 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4442 | { | |
4443 | struct cgroup_pidlist *l, *tmp_l; | |
4444 | ||
4445 | mutex_lock(&cgrp->pidlist_mutex); | |
4446 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4447 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4448 | mutex_unlock(&cgrp->pidlist_mutex); | |
4449 | ||
4450 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4451 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4452 | } | |
4453 | ||
4454 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4455 | { | |
4456 | struct delayed_work *dwork = to_delayed_work(work); | |
4457 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4458 | destroy_dwork); | |
4459 | struct cgroup_pidlist *tofree = NULL; | |
4460 | ||
4461 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4462 | |
4463 | /* | |
04502365 TH |
4464 | * Destroy iff we didn't get queued again. The state won't change |
4465 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4466 | */ |
04502365 | 4467 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4468 | list_del(&l->links); |
4469 | pidlist_free(l->list); | |
4470 | put_pid_ns(l->key.ns); | |
4471 | tofree = l; | |
4472 | } | |
4473 | ||
b1a21367 TH |
4474 | mutex_unlock(&l->owner->pidlist_mutex); |
4475 | kfree(tofree); | |
4476 | } | |
4477 | ||
bbcb81d0 | 4478 | /* |
102a775e | 4479 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4480 | * Returns the number of unique elements. |
bbcb81d0 | 4481 | */ |
6ee211ad | 4482 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4483 | { |
102a775e | 4484 | int src, dest = 1; |
102a775e BB |
4485 | |
4486 | /* | |
4487 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4488 | * edge cases first; no work needs to be done for either | |
4489 | */ | |
4490 | if (length == 0 || length == 1) | |
4491 | return length; | |
4492 | /* src and dest walk down the list; dest counts unique elements */ | |
4493 | for (src = 1; src < length; src++) { | |
4494 | /* find next unique element */ | |
4495 | while (list[src] == list[src-1]) { | |
4496 | src++; | |
4497 | if (src == length) | |
4498 | goto after; | |
4499 | } | |
4500 | /* dest always points to where the next unique element goes */ | |
4501 | list[dest] = list[src]; | |
4502 | dest++; | |
4503 | } | |
4504 | after: | |
102a775e BB |
4505 | return dest; |
4506 | } | |
4507 | ||
afb2bc14 TH |
4508 | /* |
4509 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4510 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4511 | * different per namespace, each namespace needs differently sorted list, | |
4512 | * making it impossible to use, for example, single rbtree of member tasks | |
4513 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4514 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4515 | * pidlists keyed by cgroup and namespace. | |
4516 | * | |
4517 | * All this extra complexity was caused by the original implementation | |
4518 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4519 | * want to do away with it. Explicitly scramble sort order if on the |
4520 | * default hierarchy so that no such expectation exists in the new | |
4521 | * interface. | |
afb2bc14 TH |
4522 | * |
4523 | * Scrambling is done by swapping every two consecutive bits, which is | |
4524 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4525 | */ | |
4526 | static pid_t pid_fry(pid_t pid) | |
4527 | { | |
4528 | unsigned a = pid & 0x55555555; | |
4529 | unsigned b = pid & 0xAAAAAAAA; | |
4530 | ||
4531 | return (a << 1) | (b >> 1); | |
4532 | } | |
4533 | ||
4534 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4535 | { | |
aa6ec29b | 4536 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4537 | return pid_fry(pid); |
4538 | else | |
4539 | return pid; | |
4540 | } | |
4541 | ||
102a775e BB |
4542 | static int cmppid(const void *a, const void *b) |
4543 | { | |
4544 | return *(pid_t *)a - *(pid_t *)b; | |
4545 | } | |
4546 | ||
afb2bc14 TH |
4547 | static int fried_cmppid(const void *a, const void *b) |
4548 | { | |
4549 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4550 | } | |
4551 | ||
e6b81710 TH |
4552 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4553 | enum cgroup_filetype type) | |
4554 | { | |
4555 | struct cgroup_pidlist *l; | |
4556 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4557 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4558 | ||
4559 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4560 | ||
4561 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4562 | if (l->key.type == type && l->key.ns == ns) | |
4563 | return l; | |
4564 | return NULL; | |
4565 | } | |
4566 | ||
72a8cb30 BB |
4567 | /* |
4568 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4569 | * returns with the lock on that pidlist already held, and takes care | |
4570 | * of the use count, or returns NULL with no locks held if we're out of | |
4571 | * memory. | |
4572 | */ | |
e6b81710 TH |
4573 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4574 | enum cgroup_filetype type) | |
72a8cb30 BB |
4575 | { |
4576 | struct cgroup_pidlist *l; | |
b70cc5fd | 4577 | |
e6b81710 TH |
4578 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4579 | ||
4580 | l = cgroup_pidlist_find(cgrp, type); | |
4581 | if (l) | |
4582 | return l; | |
4583 | ||
72a8cb30 | 4584 | /* entry not found; create a new one */ |
f4f4be2b | 4585 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4586 | if (!l) |
72a8cb30 | 4587 | return l; |
e6b81710 | 4588 | |
b1a21367 | 4589 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4590 | l->key.type = type; |
e6b81710 TH |
4591 | /* don't need task_nsproxy() if we're looking at ourself */ |
4592 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4593 | l->owner = cgrp; |
4594 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4595 | return l; |
4596 | } | |
4597 | ||
102a775e BB |
4598 | /* |
4599 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4600 | */ | |
72a8cb30 BB |
4601 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4602 | struct cgroup_pidlist **lp) | |
102a775e BB |
4603 | { |
4604 | pid_t *array; | |
4605 | int length; | |
4606 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4607 | struct css_task_iter it; |
817929ec | 4608 | struct task_struct *tsk; |
102a775e BB |
4609 | struct cgroup_pidlist *l; |
4610 | ||
4bac00d1 TH |
4611 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4612 | ||
102a775e BB |
4613 | /* |
4614 | * If cgroup gets more users after we read count, we won't have | |
4615 | * enough space - tough. This race is indistinguishable to the | |
4616 | * caller from the case that the additional cgroup users didn't | |
4617 | * show up until sometime later on. | |
4618 | */ | |
4619 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4620 | array = pidlist_allocate(length); |
102a775e BB |
4621 | if (!array) |
4622 | return -ENOMEM; | |
4623 | /* now, populate the array */ | |
9d800df1 | 4624 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4625 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4626 | if (unlikely(n == length)) |
817929ec | 4627 | break; |
102a775e | 4628 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4629 | if (type == CGROUP_FILE_PROCS) |
4630 | pid = task_tgid_vnr(tsk); | |
4631 | else | |
4632 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4633 | if (pid > 0) /* make sure to only use valid results */ |
4634 | array[n++] = pid; | |
817929ec | 4635 | } |
72ec7029 | 4636 | css_task_iter_end(&it); |
102a775e BB |
4637 | length = n; |
4638 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4639 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4640 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4641 | else | |
4642 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4643 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4644 | length = pidlist_uniq(array, length); |
e6b81710 | 4645 | |
e6b81710 | 4646 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4647 | if (!l) { |
d1d9fd33 | 4648 | pidlist_free(array); |
72a8cb30 | 4649 | return -ENOMEM; |
102a775e | 4650 | } |
e6b81710 TH |
4651 | |
4652 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4653 | pidlist_free(l->list); |
102a775e BB |
4654 | l->list = array; |
4655 | l->length = length; | |
72a8cb30 | 4656 | *lp = l; |
102a775e | 4657 | return 0; |
bbcb81d0 PM |
4658 | } |
4659 | ||
846c7bb0 | 4660 | /** |
a043e3b2 | 4661 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4662 | * @stats: cgroupstats to fill information into |
4663 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4664 | * been requested. | |
a043e3b2 LZ |
4665 | * |
4666 | * Build and fill cgroupstats so that taskstats can export it to user | |
4667 | * space. | |
846c7bb0 BS |
4668 | */ |
4669 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4670 | { | |
2bd59d48 | 4671 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4672 | struct cgroup *cgrp; |
72ec7029 | 4673 | struct css_task_iter it; |
846c7bb0 | 4674 | struct task_struct *tsk; |
33d283be | 4675 | |
2bd59d48 TH |
4676 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4677 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4678 | kernfs_type(kn) != KERNFS_DIR) | |
4679 | return -EINVAL; | |
4680 | ||
bad34660 LZ |
4681 | mutex_lock(&cgroup_mutex); |
4682 | ||
846c7bb0 | 4683 | /* |
2bd59d48 | 4684 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4685 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4686 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4687 | */ |
2bd59d48 TH |
4688 | rcu_read_lock(); |
4689 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4690 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4691 | rcu_read_unlock(); |
bad34660 | 4692 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4693 | return -ENOENT; |
4694 | } | |
bad34660 | 4695 | rcu_read_unlock(); |
846c7bb0 | 4696 | |
9d800df1 | 4697 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4698 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4699 | switch (tsk->state) { |
4700 | case TASK_RUNNING: | |
4701 | stats->nr_running++; | |
4702 | break; | |
4703 | case TASK_INTERRUPTIBLE: | |
4704 | stats->nr_sleeping++; | |
4705 | break; | |
4706 | case TASK_UNINTERRUPTIBLE: | |
4707 | stats->nr_uninterruptible++; | |
4708 | break; | |
4709 | case TASK_STOPPED: | |
4710 | stats->nr_stopped++; | |
4711 | break; | |
4712 | default: | |
4713 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4714 | stats->nr_io_wait++; | |
4715 | break; | |
4716 | } | |
4717 | } | |
72ec7029 | 4718 | css_task_iter_end(&it); |
846c7bb0 | 4719 | |
bad34660 | 4720 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4721 | return 0; |
846c7bb0 BS |
4722 | } |
4723 | ||
8f3ff208 | 4724 | |
bbcb81d0 | 4725 | /* |
102a775e | 4726 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4727 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4728 | * in the cgroup->l->list array. |
bbcb81d0 | 4729 | */ |
cc31edce | 4730 | |
102a775e | 4731 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4732 | { |
cc31edce PM |
4733 | /* |
4734 | * Initially we receive a position value that corresponds to | |
4735 | * one more than the last pid shown (or 0 on the first call or | |
4736 | * after a seek to the start). Use a binary-search to find the | |
4737 | * next pid to display, if any | |
4738 | */ | |
2bd59d48 | 4739 | struct kernfs_open_file *of = s->private; |
7da11279 | 4740 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4741 | struct cgroup_pidlist *l; |
7da11279 | 4742 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4743 | int index = 0, pid = *pos; |
4bac00d1 TH |
4744 | int *iter, ret; |
4745 | ||
4746 | mutex_lock(&cgrp->pidlist_mutex); | |
4747 | ||
4748 | /* | |
5d22444f | 4749 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4750 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4751 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4752 | * could already have been destroyed. |
4753 | */ | |
5d22444f TH |
4754 | if (of->priv) |
4755 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4756 | |
4757 | /* | |
4758 | * Either this is the first start() after open or the matching | |
4759 | * pidlist has been destroyed inbetween. Create a new one. | |
4760 | */ | |
5d22444f TH |
4761 | if (!of->priv) { |
4762 | ret = pidlist_array_load(cgrp, type, | |
4763 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4764 | if (ret) |
4765 | return ERR_PTR(ret); | |
4766 | } | |
5d22444f | 4767 | l = of->priv; |
cc31edce | 4768 | |
cc31edce | 4769 | if (pid) { |
102a775e | 4770 | int end = l->length; |
20777766 | 4771 | |
cc31edce PM |
4772 | while (index < end) { |
4773 | int mid = (index + end) / 2; | |
afb2bc14 | 4774 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4775 | index = mid; |
4776 | break; | |
afb2bc14 | 4777 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4778 | index = mid + 1; |
4779 | else | |
4780 | end = mid; | |
4781 | } | |
4782 | } | |
4783 | /* If we're off the end of the array, we're done */ | |
102a775e | 4784 | if (index >= l->length) |
cc31edce PM |
4785 | return NULL; |
4786 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4787 | iter = l->list + index; |
afb2bc14 | 4788 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4789 | return iter; |
4790 | } | |
4791 | ||
102a775e | 4792 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4793 | { |
2bd59d48 | 4794 | struct kernfs_open_file *of = s->private; |
5d22444f | 4795 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4796 | |
5d22444f TH |
4797 | if (l) |
4798 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4799 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4800 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4801 | } |
4802 | ||
102a775e | 4803 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4804 | { |
2bd59d48 | 4805 | struct kernfs_open_file *of = s->private; |
5d22444f | 4806 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4807 | pid_t *p = v; |
4808 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4809 | /* |
4810 | * Advance to the next pid in the array. If this goes off the | |
4811 | * end, we're done | |
4812 | */ | |
4813 | p++; | |
4814 | if (p >= end) { | |
4815 | return NULL; | |
4816 | } else { | |
7da11279 | 4817 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4818 | return p; |
4819 | } | |
4820 | } | |
4821 | ||
102a775e | 4822 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4823 | { |
94ff212d JP |
4824 | seq_printf(s, "%d\n", *(int *)v); |
4825 | ||
4826 | return 0; | |
cc31edce | 4827 | } |
bbcb81d0 | 4828 | |
182446d0 TH |
4829 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4830 | struct cftype *cft) | |
81a6a5cd | 4831 | { |
182446d0 | 4832 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4833 | } |
4834 | ||
182446d0 TH |
4835 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4836 | struct cftype *cft, u64 val) | |
6379c106 | 4837 | { |
6379c106 | 4838 | if (val) |
182446d0 | 4839 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4840 | else |
182446d0 | 4841 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4842 | return 0; |
4843 | } | |
4844 | ||
182446d0 TH |
4845 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4846 | struct cftype *cft) | |
97978e6d | 4847 | { |
182446d0 | 4848 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4849 | } |
4850 | ||
182446d0 TH |
4851 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4852 | struct cftype *cft, u64 val) | |
97978e6d DL |
4853 | { |
4854 | if (val) | |
182446d0 | 4855 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4856 | else |
182446d0 | 4857 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4858 | return 0; |
4859 | } | |
4860 | ||
a14c6874 TH |
4861 | /* cgroup core interface files for the default hierarchy */ |
4862 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4863 | { |
d5c56ced | 4864 | .name = "cgroup.procs", |
6f60eade | 4865 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4866 | .seq_start = cgroup_pidlist_start, |
4867 | .seq_next = cgroup_pidlist_next, | |
4868 | .seq_stop = cgroup_pidlist_stop, | |
4869 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4870 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4871 | .write = cgroup_procs_write, |
102a775e | 4872 | }, |
f8f22e53 TH |
4873 | { |
4874 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4875 | .seq_show = cgroup_controllers_show, |
4876 | }, | |
4877 | { | |
4878 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4879 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4880 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4881 | }, |
842b597e | 4882 | { |
4a07c222 | 4883 | .name = "cgroup.events", |
a14c6874 | 4884 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4885 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4886 | .seq_show = cgroup_events_show, |
842b597e | 4887 | }, |
a14c6874 TH |
4888 | { } /* terminate */ |
4889 | }; | |
d5c56ced | 4890 | |
a14c6874 TH |
4891 | /* cgroup core interface files for the legacy hierarchies */ |
4892 | static struct cftype cgroup_legacy_base_files[] = { | |
4893 | { | |
4894 | .name = "cgroup.procs", | |
4895 | .seq_start = cgroup_pidlist_start, | |
4896 | .seq_next = cgroup_pidlist_next, | |
4897 | .seq_stop = cgroup_pidlist_stop, | |
4898 | .seq_show = cgroup_pidlist_show, | |
4899 | .private = CGROUP_FILE_PROCS, | |
4900 | .write = cgroup_procs_write, | |
a14c6874 TH |
4901 | }, |
4902 | { | |
4903 | .name = "cgroup.clone_children", | |
4904 | .read_u64 = cgroup_clone_children_read, | |
4905 | .write_u64 = cgroup_clone_children_write, | |
4906 | }, | |
4907 | { | |
4908 | .name = "cgroup.sane_behavior", | |
4909 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4910 | .seq_show = cgroup_sane_behavior_show, | |
4911 | }, | |
d5c56ced TH |
4912 | { |
4913 | .name = "tasks", | |
6612f05b TH |
4914 | .seq_start = cgroup_pidlist_start, |
4915 | .seq_next = cgroup_pidlist_next, | |
4916 | .seq_stop = cgroup_pidlist_stop, | |
4917 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4918 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4919 | .write = cgroup_tasks_write, |
d5c56ced TH |
4920 | }, |
4921 | { | |
4922 | .name = "notify_on_release", | |
d5c56ced TH |
4923 | .read_u64 = cgroup_read_notify_on_release, |
4924 | .write_u64 = cgroup_write_notify_on_release, | |
4925 | }, | |
6e6ff25b TH |
4926 | { |
4927 | .name = "release_agent", | |
a14c6874 | 4928 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4929 | .seq_show = cgroup_release_agent_show, |
451af504 | 4930 | .write = cgroup_release_agent_write, |
5f469907 | 4931 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4932 | }, |
db0416b6 | 4933 | { } /* terminate */ |
bbcb81d0 PM |
4934 | }; |
4935 | ||
0c21ead1 TH |
4936 | /* |
4937 | * css destruction is four-stage process. | |
4938 | * | |
4939 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4940 | * Implemented in kill_css(). | |
4941 | * | |
4942 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4943 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4944 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4945 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4946 | * |
4947 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4948 | * accessors are inside RCU read sections. css_release() schedules the | |
4949 | * RCU callback. | |
4950 | * | |
4951 | * 4. After the grace period, the css can be freed. Implemented in | |
4952 | * css_free_work_fn(). | |
4953 | * | |
4954 | * It is actually hairier because both step 2 and 4 require process context | |
4955 | * and thus involve punting to css->destroy_work adding two additional | |
4956 | * steps to the already complex sequence. | |
4957 | */ | |
35ef10da | 4958 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4959 | { |
4960 | struct cgroup_subsys_state *css = | |
35ef10da | 4961 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4962 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4963 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4964 | |
9a1049da TH |
4965 | percpu_ref_exit(&css->refcnt); |
4966 | ||
01e58659 | 4967 | if (ss) { |
9d755d33 | 4968 | /* css free path */ |
8bb5ef79 | 4969 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4970 | int id = css->id; |
4971 | ||
01e58659 VD |
4972 | ss->css_free(css); |
4973 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4974 | cgroup_put(cgrp); |
8bb5ef79 TH |
4975 | |
4976 | if (parent) | |
4977 | css_put(parent); | |
9d755d33 TH |
4978 | } else { |
4979 | /* cgroup free path */ | |
4980 | atomic_dec(&cgrp->root->nr_cgrps); | |
4981 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4982 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4983 | |
d51f39b0 | 4984 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4985 | /* |
4986 | * We get a ref to the parent, and put the ref when | |
4987 | * this cgroup is being freed, so it's guaranteed | |
4988 | * that the parent won't be destroyed before its | |
4989 | * children. | |
4990 | */ | |
d51f39b0 | 4991 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4992 | kernfs_put(cgrp->kn); |
4993 | kfree(cgrp); | |
4994 | } else { | |
4995 | /* | |
4996 | * This is root cgroup's refcnt reaching zero, | |
4997 | * which indicates that the root should be | |
4998 | * released. | |
4999 | */ | |
5000 | cgroup_destroy_root(cgrp->root); | |
5001 | } | |
5002 | } | |
48ddbe19 TH |
5003 | } |
5004 | ||
0c21ead1 | 5005 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
5006 | { |
5007 | struct cgroup_subsys_state *css = | |
0c21ead1 | 5008 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 5009 | |
35ef10da | 5010 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 5011 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
5012 | } |
5013 | ||
25e15d83 | 5014 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
5015 | { |
5016 | struct cgroup_subsys_state *css = | |
25e15d83 | 5017 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 5018 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 5019 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 5020 | |
1fed1b2e TH |
5021 | mutex_lock(&cgroup_mutex); |
5022 | ||
de3f0341 | 5023 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
5024 | list_del_rcu(&css->sibling); |
5025 | ||
9d755d33 TH |
5026 | if (ss) { |
5027 | /* css release path */ | |
01e58659 | 5028 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
5029 | if (ss->css_released) |
5030 | ss->css_released(css); | |
9d755d33 TH |
5031 | } else { |
5032 | /* cgroup release path */ | |
9d755d33 TH |
5033 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
5034 | cgrp->id = -1; | |
a4189487 LZ |
5035 | |
5036 | /* | |
5037 | * There are two control paths which try to determine | |
5038 | * cgroup from dentry without going through kernfs - | |
5039 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
5040 | * Those are supported by RCU protecting clearing of | |
5041 | * cgrp->kn->priv backpointer. | |
5042 | */ | |
6cd0f5bb TH |
5043 | if (cgrp->kn) |
5044 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
5045 | NULL); | |
9d755d33 | 5046 | } |
d3daf28d | 5047 | |
1fed1b2e TH |
5048 | mutex_unlock(&cgroup_mutex); |
5049 | ||
0c21ead1 | 5050 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
5051 | } |
5052 | ||
d3daf28d TH |
5053 | static void css_release(struct percpu_ref *ref) |
5054 | { | |
5055 | struct cgroup_subsys_state *css = | |
5056 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5057 | ||
25e15d83 TH |
5058 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5059 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5060 | } |
5061 | ||
ddfcadab TH |
5062 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5063 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5064 | { |
0cb51d71 TH |
5065 | lockdep_assert_held(&cgroup_mutex); |
5066 | ||
ddfcadab TH |
5067 | cgroup_get(cgrp); |
5068 | ||
d5c419b6 | 5069 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5070 | css->cgroup = cgrp; |
72c97e54 | 5071 | css->ss = ss; |
8fa3b8d6 | 5072 | css->id = -1; |
d5c419b6 TH |
5073 | INIT_LIST_HEAD(&css->sibling); |
5074 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 5075 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5076 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5077 | |
d51f39b0 TH |
5078 | if (cgroup_parent(cgrp)) { |
5079 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5080 | css_get(css->parent); |
ddfcadab | 5081 | } |
48ddbe19 | 5082 | |
ca8bdcaf | 5083 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5084 | } |
5085 | ||
2a4ac633 | 5086 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5087 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5088 | { |
623f926b | 5089 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5090 | int ret = 0; |
5091 | ||
a31f2d3f TH |
5092 | lockdep_assert_held(&cgroup_mutex); |
5093 | ||
92fb9748 | 5094 | if (ss->css_online) |
eb95419b | 5095 | ret = ss->css_online(css); |
ae7f164a | 5096 | if (!ret) { |
eb95419b | 5097 | css->flags |= CSS_ONLINE; |
aec25020 | 5098 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5099 | |
5100 | atomic_inc(&css->online_cnt); | |
5101 | if (css->parent) | |
5102 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5103 | } |
b1929db4 | 5104 | return ret; |
a31f2d3f TH |
5105 | } |
5106 | ||
2a4ac633 | 5107 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5108 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5109 | { |
623f926b | 5110 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5111 | |
5112 | lockdep_assert_held(&cgroup_mutex); | |
5113 | ||
5114 | if (!(css->flags & CSS_ONLINE)) | |
5115 | return; | |
5116 | ||
fa06235b VD |
5117 | if (ss->css_reset) |
5118 | ss->css_reset(css); | |
5119 | ||
d7eeac19 | 5120 | if (ss->css_offline) |
eb95419b | 5121 | ss->css_offline(css); |
a31f2d3f | 5122 | |
eb95419b | 5123 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5124 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5125 | |
5126 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5127 | } |
5128 | ||
c81c925a | 5129 | /** |
6cd0f5bb | 5130 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5131 | * @cgrp: the cgroup new css will be associated with |
5132 | * @ss: the subsys of new css | |
5133 | * | |
5134 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5135 | * css is online and installed in @cgrp. This function doesn't create the |
5136 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5137 | */ |
6cd0f5bb TH |
5138 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5139 | struct cgroup_subsys *ss) | |
c81c925a | 5140 | { |
d51f39b0 | 5141 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5142 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5143 | struct cgroup_subsys_state *css; |
5144 | int err; | |
5145 | ||
c81c925a TH |
5146 | lockdep_assert_held(&cgroup_mutex); |
5147 | ||
1fed1b2e | 5148 | css = ss->css_alloc(parent_css); |
c81c925a | 5149 | if (IS_ERR(css)) |
6cd0f5bb | 5150 | return css; |
c81c925a | 5151 | |
ddfcadab | 5152 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5153 | |
2aad2a86 | 5154 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5155 | if (err) |
3eb59ec6 | 5156 | goto err_free_css; |
c81c925a | 5157 | |
cf780b7d | 5158 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 5159 | if (err < 0) |
b00c52da | 5160 | goto err_free_css; |
15a4c835 | 5161 | css->id = err; |
c81c925a | 5162 | |
15a4c835 | 5163 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5164 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5165 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5166 | |
5167 | err = online_css(css); | |
5168 | if (err) | |
1fed1b2e | 5169 | goto err_list_del; |
94419627 | 5170 | |
c81c925a | 5171 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5172 | cgroup_parent(parent)) { |
ed3d261b | 5173 | 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 | 5174 | current->comm, current->pid, ss->name); |
c81c925a | 5175 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5176 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5177 | ss->warned_broken_hierarchy = true; |
5178 | } | |
5179 | ||
6cd0f5bb | 5180 | return css; |
c81c925a | 5181 | |
1fed1b2e TH |
5182 | err_list_del: |
5183 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 5184 | err_free_css: |
a2bed820 | 5185 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 5186 | return ERR_PTR(err); |
c81c925a TH |
5187 | } |
5188 | ||
a5bca215 | 5189 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 5190 | { |
a5bca215 | 5191 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
5192 | struct cgroup *cgrp, *tcgrp; |
5193 | int level = parent->level + 1; | |
03970d3c | 5194 | int ret; |
ddbcc7e8 | 5195 | |
0a950f65 | 5196 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
5197 | cgrp = kzalloc(sizeof(*cgrp) + |
5198 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
5199 | if (!cgrp) |
5200 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5201 | |
2aad2a86 | 5202 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5203 | if (ret) |
5204 | goto out_free_cgrp; | |
5205 | ||
0ab02ca8 LZ |
5206 | /* |
5207 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5208 | * a half-baked cgroup. | |
5209 | */ | |
cf780b7d | 5210 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5211 | if (cgrp->id < 0) { |
ba0f4d76 | 5212 | ret = -ENOMEM; |
9d755d33 | 5213 | goto out_cancel_ref; |
976c06bc TH |
5214 | } |
5215 | ||
cc31edce | 5216 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5217 | |
9d800df1 | 5218 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5219 | cgrp->root = root; |
b11cfb58 TH |
5220 | cgrp->level = level; |
5221 | ||
5222 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5223 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5224 | |
b6abdb0e LZ |
5225 | if (notify_on_release(parent)) |
5226 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5227 | ||
2260e7fc TH |
5228 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5229 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5230 | |
0cb51d71 | 5231 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5232 | |
4e139afc | 5233 | /* allocation complete, commit to creation */ |
d5c419b6 | 5234 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5235 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5236 | cgroup_get(parent); |
415cf07a | 5237 | |
0d80255e TH |
5238 | /* |
5239 | * @cgrp is now fully operational. If something fails after this | |
5240 | * point, it'll be released via the normal destruction path. | |
5241 | */ | |
6fa4918d | 5242 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5243 | |
bd53d617 TH |
5244 | /* |
5245 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5246 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5247 | */ |
03970d3c | 5248 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5249 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5250 | |
5251 | cgroup_propagate_control(cgrp); | |
5252 | ||
5253 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5254 | ret = cgroup_apply_control_enable(cgrp); | |
5255 | if (ret) | |
5256 | goto out_destroy; | |
2bd59d48 | 5257 | |
a5bca215 TH |
5258 | return cgrp; |
5259 | ||
5260 | out_cancel_ref: | |
5261 | percpu_ref_exit(&cgrp->self.refcnt); | |
5262 | out_free_cgrp: | |
5263 | kfree(cgrp); | |
5264 | return ERR_PTR(ret); | |
5265 | out_destroy: | |
5266 | cgroup_destroy_locked(cgrp); | |
5267 | return ERR_PTR(ret); | |
5268 | } | |
5269 | ||
5270 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5271 | umode_t mode) | |
5272 | { | |
5273 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5274 | struct kernfs_node *kn; |
03970d3c | 5275 | int ret; |
a5bca215 TH |
5276 | |
5277 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5278 | if (strchr(name, '\n')) | |
5279 | return -EINVAL; | |
5280 | ||
945ba199 | 5281 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5282 | if (!parent) |
5283 | return -ENODEV; | |
5284 | ||
5285 | cgrp = cgroup_create(parent); | |
5286 | if (IS_ERR(cgrp)) { | |
5287 | ret = PTR_ERR(cgrp); | |
5288 | goto out_unlock; | |
5289 | } | |
5290 | ||
195e9b6c TH |
5291 | /* create the directory */ |
5292 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5293 | if (IS_ERR(kn)) { | |
5294 | ret = PTR_ERR(kn); | |
5295 | goto out_destroy; | |
5296 | } | |
5297 | cgrp->kn = kn; | |
5298 | ||
5299 | /* | |
5300 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5301 | * that @cgrp->kn is always accessible. | |
5302 | */ | |
5303 | kernfs_get(kn); | |
5304 | ||
5305 | ret = cgroup_kn_set_ugid(kn); | |
5306 | if (ret) | |
5307 | goto out_destroy; | |
5308 | ||
334c3679 | 5309 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5310 | if (ret) |
5311 | goto out_destroy; | |
5312 | ||
03970d3c TH |
5313 | ret = cgroup_apply_control_enable(cgrp); |
5314 | if (ret) | |
5315 | goto out_destroy; | |
195e9b6c TH |
5316 | |
5317 | /* let's create and online css's */ | |
2bd59d48 | 5318 | kernfs_activate(kn); |
ddbcc7e8 | 5319 | |
ba0f4d76 TH |
5320 | ret = 0; |
5321 | goto out_unlock; | |
ddbcc7e8 | 5322 | |
a5bca215 TH |
5323 | out_destroy: |
5324 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5325 | out_unlock: |
a9746d8d | 5326 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5327 | return ret; |
ddbcc7e8 PM |
5328 | } |
5329 | ||
223dbc38 TH |
5330 | /* |
5331 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5332 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5333 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5334 | */ |
5335 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5336 | { |
223dbc38 TH |
5337 | struct cgroup_subsys_state *css = |
5338 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5339 | |
f20104de | 5340 | mutex_lock(&cgroup_mutex); |
09a503ea | 5341 | |
aa226ff4 TH |
5342 | do { |
5343 | offline_css(css); | |
5344 | css_put(css); | |
5345 | /* @css can't go away while we're holding cgroup_mutex */ | |
5346 | css = css->parent; | |
5347 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5348 | ||
5349 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5350 | } |
5351 | ||
223dbc38 TH |
5352 | /* css kill confirmation processing requires process context, bounce */ |
5353 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5354 | { |
5355 | struct cgroup_subsys_state *css = | |
5356 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5357 | ||
aa226ff4 TH |
5358 | if (atomic_dec_and_test(&css->online_cnt)) { |
5359 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5360 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5361 | } | |
d3daf28d TH |
5362 | } |
5363 | ||
f392e51c TH |
5364 | /** |
5365 | * kill_css - destroy a css | |
5366 | * @css: css to destroy | |
5367 | * | |
5368 | * This function initiates destruction of @css by removing cgroup interface | |
5369 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5370 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5371 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5372 | */ |
5373 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5374 | { |
01f6474c | 5375 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5376 | |
2bd59d48 TH |
5377 | /* |
5378 | * This must happen before css is disassociated with its cgroup. | |
5379 | * See seq_css() for details. | |
5380 | */ | |
334c3679 | 5381 | css_clear_dir(css); |
3c14f8b4 | 5382 | |
edae0c33 TH |
5383 | /* |
5384 | * Killing would put the base ref, but we need to keep it alive | |
5385 | * until after ->css_offline(). | |
5386 | */ | |
5387 | css_get(css); | |
5388 | ||
5389 | /* | |
5390 | * cgroup core guarantees that, by the time ->css_offline() is | |
5391 | * invoked, no new css reference will be given out via | |
ec903c0c | 5392 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5393 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5394 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5395 | * | |
5396 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5397 | * css is confirmed to be seen as killed on all CPUs. | |
5398 | */ | |
5399 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5400 | } |
5401 | ||
5402 | /** | |
5403 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5404 | * @cgrp: cgroup to be destroyed | |
5405 | * | |
5406 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5407 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5408 | * guarantee that css_tryget_online() won't succeed by the time |
5409 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5410 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5411 | * |
5412 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5413 | * userland visible parts and start killing the percpu refcnts of | |
5414 | * css's. Set up so that the next stage will be kicked off once all | |
5415 | * the percpu refcnts are confirmed to be killed. | |
5416 | * | |
5417 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5418 | * rest of destruction. Once all cgroup references are gone, the | |
5419 | * cgroup is RCU-freed. | |
5420 | * | |
5421 | * This function implements s1. After this step, @cgrp is gone as far as | |
5422 | * the userland is concerned and a new cgroup with the same name may be | |
5423 | * created. As cgroup doesn't care about the names internally, this | |
5424 | * doesn't cause any problem. | |
5425 | */ | |
42809dd4 TH |
5426 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5427 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5428 | { |
2bd59d48 | 5429 | struct cgroup_subsys_state *css; |
2b021cbf | 5430 | struct cgrp_cset_link *link; |
1c6727af | 5431 | int ssid; |
ddbcc7e8 | 5432 | |
42809dd4 TH |
5433 | lockdep_assert_held(&cgroup_mutex); |
5434 | ||
91486f61 TH |
5435 | /* |
5436 | * Only migration can raise populated from zero and we're already | |
5437 | * holding cgroup_mutex. | |
5438 | */ | |
5439 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5440 | return -EBUSY; |
a043e3b2 | 5441 | |
bb78a92f | 5442 | /* |
d5c419b6 TH |
5443 | * Make sure there's no live children. We can't test emptiness of |
5444 | * ->self.children as dead children linger on it while being | |
5445 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5446 | */ |
f3d46500 | 5447 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5448 | return -EBUSY; |
5449 | ||
455050d2 | 5450 | /* |
2b021cbf TH |
5451 | * Mark @cgrp and the associated csets dead. The former prevents |
5452 | * further task migration and child creation by disabling | |
5453 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5454 | * the migration path. | |
455050d2 | 5455 | */ |
184faf32 | 5456 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5457 | |
82d6489d | 5458 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5459 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5460 | link->cset->dead = true; | |
82d6489d | 5461 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5462 | |
249f3468 | 5463 | /* initiate massacre of all css's */ |
1c6727af TH |
5464 | for_each_css(css, ssid, cgrp) |
5465 | kill_css(css); | |
455050d2 | 5466 | |
455050d2 | 5467 | /* |
01f6474c TH |
5468 | * Remove @cgrp directory along with the base files. @cgrp has an |
5469 | * extra ref on its kn. | |
f20104de | 5470 | */ |
01f6474c | 5471 | kernfs_remove(cgrp->kn); |
f20104de | 5472 | |
d51f39b0 | 5473 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5474 | |
249f3468 | 5475 | /* put the base reference */ |
9d755d33 | 5476 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5477 | |
ea15f8cc TH |
5478 | return 0; |
5479 | }; | |
5480 | ||
2bd59d48 | 5481 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5482 | { |
a9746d8d | 5483 | struct cgroup *cgrp; |
2bd59d48 | 5484 | int ret = 0; |
42809dd4 | 5485 | |
945ba199 | 5486 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5487 | if (!cgrp) |
5488 | return 0; | |
42809dd4 | 5489 | |
a9746d8d | 5490 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5491 | |
a9746d8d | 5492 | cgroup_kn_unlock(kn); |
42809dd4 | 5493 | return ret; |
8e3f6541 TH |
5494 | } |
5495 | ||
2bd59d48 TH |
5496 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5497 | .remount_fs = cgroup_remount, | |
5498 | .show_options = cgroup_show_options, | |
5499 | .mkdir = cgroup_mkdir, | |
5500 | .rmdir = cgroup_rmdir, | |
5501 | .rename = cgroup_rename, | |
4f41fc59 | 5502 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5503 | }; |
5504 | ||
15a4c835 | 5505 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5506 | { |
ddbcc7e8 | 5507 | struct cgroup_subsys_state *css; |
cfe36bde | 5508 | |
a5ae9899 | 5509 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5510 | |
648bb56d TH |
5511 | mutex_lock(&cgroup_mutex); |
5512 | ||
15a4c835 | 5513 | idr_init(&ss->css_idr); |
0adb0704 | 5514 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5515 | |
3dd06ffa TH |
5516 | /* Create the root cgroup state for this subsystem */ |
5517 | ss->root = &cgrp_dfl_root; | |
5518 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5519 | /* We don't handle early failures gracefully */ |
5520 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5521 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5522 | |
5523 | /* | |
5524 | * Root csses are never destroyed and we can't initialize | |
5525 | * percpu_ref during early init. Disable refcnting. | |
5526 | */ | |
5527 | css->flags |= CSS_NO_REF; | |
5528 | ||
15a4c835 | 5529 | if (early) { |
9395a450 | 5530 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5531 | css->id = 1; |
5532 | } else { | |
5533 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5534 | BUG_ON(css->id < 0); | |
5535 | } | |
ddbcc7e8 | 5536 | |
e8d55fde | 5537 | /* Update the init_css_set to contain a subsys |
817929ec | 5538 | * pointer to this state - since the subsystem is |
e8d55fde | 5539 | * newly registered, all tasks and hence the |
3dd06ffa | 5540 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5541 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5542 | |
cb4a3167 AS |
5543 | have_fork_callback |= (bool)ss->fork << ss->id; |
5544 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5545 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5546 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5547 | |
e8d55fde LZ |
5548 | /* At system boot, before all subsystems have been |
5549 | * registered, no tasks have been forked, so we don't | |
5550 | * need to invoke fork callbacks here. */ | |
5551 | BUG_ON(!list_empty(&init_task.tasks)); | |
5552 | ||
ae7f164a | 5553 | BUG_ON(online_css(css)); |
a8638030 | 5554 | |
cf5d5941 BB |
5555 | mutex_unlock(&cgroup_mutex); |
5556 | } | |
cf5d5941 | 5557 | |
ddbcc7e8 | 5558 | /** |
a043e3b2 LZ |
5559 | * cgroup_init_early - cgroup initialization at system boot |
5560 | * | |
5561 | * Initialize cgroups at system boot, and initialize any | |
5562 | * subsystems that request early init. | |
ddbcc7e8 PM |
5563 | */ |
5564 | int __init cgroup_init_early(void) | |
5565 | { | |
7b9a6ba5 | 5566 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5567 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5568 | int i; |
30159ec7 | 5569 | |
3dd06ffa | 5570 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5571 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5572 | ||
a4ea1cc9 | 5573 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5574 | |
3ed80a62 | 5575 | for_each_subsys(ss, i) { |
aec25020 | 5576 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5577 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5578 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5579 | ss->id, ss->name); |
073219e9 TH |
5580 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5581 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5582 | ||
aec25020 | 5583 | ss->id = i; |
073219e9 | 5584 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5585 | if (!ss->legacy_name) |
5586 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5587 | |
5588 | if (ss->early_init) | |
15a4c835 | 5589 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5590 | } |
5591 | return 0; | |
5592 | } | |
5593 | ||
6e5c8307 | 5594 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5595 | |
ddbcc7e8 | 5596 | /** |
a043e3b2 LZ |
5597 | * cgroup_init - cgroup initialization |
5598 | * | |
5599 | * Register cgroup filesystem and /proc file, and initialize | |
5600 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5601 | */ |
5602 | int __init cgroup_init(void) | |
5603 | { | |
30159ec7 | 5604 | struct cgroup_subsys *ss; |
035f4f51 | 5605 | int ssid; |
ddbcc7e8 | 5606 | |
6e5c8307 | 5607 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5608 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5609 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5610 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5611 | |
a79a908f AK |
5612 | get_user_ns(init_cgroup_ns.user_ns); |
5613 | ||
54e7b4eb | 5614 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5615 | |
2378d8b8 TH |
5616 | /* |
5617 | * Add init_css_set to the hash table so that dfl_root can link to | |
5618 | * it during init. | |
5619 | */ | |
5620 | hash_add(css_set_table, &init_css_set.hlist, | |
5621 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5622 | |
3dd06ffa | 5623 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5624 | |
54e7b4eb TH |
5625 | mutex_unlock(&cgroup_mutex); |
5626 | ||
172a2c06 | 5627 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5628 | if (ss->early_init) { |
5629 | struct cgroup_subsys_state *css = | |
5630 | init_css_set.subsys[ss->id]; | |
5631 | ||
5632 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5633 | GFP_KERNEL); | |
5634 | BUG_ON(css->id < 0); | |
5635 | } else { | |
5636 | cgroup_init_subsys(ss, false); | |
5637 | } | |
172a2c06 | 5638 | |
2d8f243a TH |
5639 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5640 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5641 | |
5642 | /* | |
c731ae1d LZ |
5643 | * Setting dfl_root subsys_mask needs to consider the |
5644 | * disabled flag and cftype registration needs kmalloc, | |
5645 | * both of which aren't available during early_init. | |
172a2c06 | 5646 | */ |
a3e72739 TH |
5647 | if (cgroup_disable_mask & (1 << ssid)) { |
5648 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5649 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5650 | ss->name); | |
a8ddc821 | 5651 | continue; |
a3e72739 | 5652 | } |
a8ddc821 | 5653 | |
223ffb29 JW |
5654 | if (cgroup_ssid_no_v1(ssid)) |
5655 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5656 | ss->name); | |
5657 | ||
a8ddc821 TH |
5658 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5659 | ||
f6d635ad TH |
5660 | if (ss->implicit_on_dfl) |
5661 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5662 | else if (!ss->dfl_cftypes) | |
a7165264 | 5663 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5664 | |
a8ddc821 TH |
5665 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5666 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5667 | } else { | |
5668 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5669 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5670 | } |
295458e6 VD |
5671 | |
5672 | if (ss->bind) | |
5673 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
5674 | } |
5675 | ||
2378d8b8 TH |
5676 | /* init_css_set.subsys[] has been updated, re-hash */ |
5677 | hash_del(&init_css_set.hlist); | |
5678 | hash_add(css_set_table, &init_css_set.hlist, | |
5679 | css_set_hash(init_css_set.subsys)); | |
5680 | ||
035f4f51 TH |
5681 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5682 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5683 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5684 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5685 | |
2bd59d48 | 5686 | return 0; |
ddbcc7e8 | 5687 | } |
b4f48b63 | 5688 | |
e5fca243 TH |
5689 | static int __init cgroup_wq_init(void) |
5690 | { | |
5691 | /* | |
5692 | * There isn't much point in executing destruction path in | |
5693 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5694 | * Use 1 for @max_active. |
e5fca243 TH |
5695 | * |
5696 | * We would prefer to do this in cgroup_init() above, but that | |
5697 | * is called before init_workqueues(): so leave this until after. | |
5698 | */ | |
1a11533f | 5699 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5700 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5701 | |
5702 | /* | |
5703 | * Used to destroy pidlists and separate to serve as flush domain. | |
5704 | * Cap @max_active to 1 too. | |
5705 | */ | |
5706 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5707 | 0, 1); | |
5708 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5709 | ||
e5fca243 TH |
5710 | return 0; |
5711 | } | |
5712 | core_initcall(cgroup_wq_init); | |
5713 | ||
a424316c PM |
5714 | /* |
5715 | * proc_cgroup_show() | |
5716 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5717 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5718 | */ |
006f4ac4 ZL |
5719 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5720 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5721 | { |
e61734c5 | 5722 | char *buf, *path; |
a424316c | 5723 | int retval; |
3dd06ffa | 5724 | struct cgroup_root *root; |
a424316c PM |
5725 | |
5726 | retval = -ENOMEM; | |
e61734c5 | 5727 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5728 | if (!buf) |
5729 | goto out; | |
5730 | ||
a424316c | 5731 | mutex_lock(&cgroup_mutex); |
82d6489d | 5732 | spin_lock_irq(&css_set_lock); |
a424316c | 5733 | |
985ed670 | 5734 | for_each_root(root) { |
a424316c | 5735 | struct cgroup_subsys *ss; |
bd89aabc | 5736 | struct cgroup *cgrp; |
b85d2040 | 5737 | int ssid, count = 0; |
a424316c | 5738 | |
a7165264 | 5739 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5740 | continue; |
5741 | ||
2c6ab6d2 | 5742 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5743 | if (root != &cgrp_dfl_root) |
5744 | for_each_subsys(ss, ssid) | |
5745 | if (root->subsys_mask & (1 << ssid)) | |
5746 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5747 | ss->legacy_name); |
c6d57f33 PM |
5748 | if (strlen(root->name)) |
5749 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5750 | root->name); | |
a424316c | 5751 | seq_putc(m, ':'); |
2e91fa7f | 5752 | |
7717f7ba | 5753 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5754 | |
5755 | /* | |
5756 | * On traditional hierarchies, all zombie tasks show up as | |
5757 | * belonging to the root cgroup. On the default hierarchy, | |
5758 | * while a zombie doesn't show up in "cgroup.procs" and | |
5759 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5760 | * reporting the cgroup it belonged to before exiting. If | |
5761 | * the cgroup is removed before the zombie is reaped, | |
5762 | * " (deleted)" is appended to the cgroup path. | |
5763 | */ | |
5764 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
a79a908f AK |
5765 | path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
5766 | current->nsproxy->cgroup_ns); | |
2e91fa7f TH |
5767 | if (!path) { |
5768 | retval = -ENAMETOOLONG; | |
5769 | goto out_unlock; | |
5770 | } | |
5771 | } else { | |
5772 | path = "/"; | |
e61734c5 | 5773 | } |
2e91fa7f | 5774 | |
e61734c5 | 5775 | seq_puts(m, path); |
2e91fa7f TH |
5776 | |
5777 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5778 | seq_puts(m, " (deleted)\n"); | |
5779 | else | |
5780 | seq_putc(m, '\n'); | |
a424316c PM |
5781 | } |
5782 | ||
006f4ac4 | 5783 | retval = 0; |
a424316c | 5784 | out_unlock: |
82d6489d | 5785 | spin_unlock_irq(&css_set_lock); |
a424316c | 5786 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5787 | kfree(buf); |
5788 | out: | |
5789 | return retval; | |
5790 | } | |
5791 | ||
a424316c PM |
5792 | /* Display information about each subsystem and each hierarchy */ |
5793 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5794 | { | |
30159ec7 | 5795 | struct cgroup_subsys *ss; |
a424316c | 5796 | int i; |
a424316c | 5797 | |
8bab8dde | 5798 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5799 | /* |
5800 | * ideally we don't want subsystems moving around while we do this. | |
5801 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5802 | * subsys/hierarchy state. | |
5803 | */ | |
a424316c | 5804 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5805 | |
5806 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5807 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5808 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5809 | atomic_read(&ss->root->nr_cgrps), |
5810 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5811 | |
a424316c PM |
5812 | mutex_unlock(&cgroup_mutex); |
5813 | return 0; | |
5814 | } | |
5815 | ||
5816 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5817 | { | |
9dce07f1 | 5818 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5819 | } |
5820 | ||
828c0950 | 5821 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5822 | .open = cgroupstats_open, |
5823 | .read = seq_read, | |
5824 | .llseek = seq_lseek, | |
5825 | .release = single_release, | |
5826 | }; | |
5827 | ||
b4f48b63 | 5828 | /** |
eaf797ab | 5829 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5830 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5831 | * |
eaf797ab TH |
5832 | * A task is associated with the init_css_set until cgroup_post_fork() |
5833 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5834 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5835 | */ |
5836 | void cgroup_fork(struct task_struct *child) | |
5837 | { | |
eaf797ab | 5838 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5839 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5840 | } |
5841 | ||
7e47682e AS |
5842 | /** |
5843 | * cgroup_can_fork - called on a new task before the process is exposed | |
5844 | * @child: the task in question. | |
5845 | * | |
5846 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5847 | * returns an error, the fork aborts with that error code. This allows for | |
5848 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5849 | */ | |
b53202e6 | 5850 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5851 | { |
5852 | struct cgroup_subsys *ss; | |
5853 | int i, j, ret; | |
5854 | ||
b4e0eeaf | 5855 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5856 | ret = ss->can_fork(child); |
7e47682e AS |
5857 | if (ret) |
5858 | goto out_revert; | |
b4e0eeaf | 5859 | } while_each_subsys_mask(); |
7e47682e AS |
5860 | |
5861 | return 0; | |
5862 | ||
5863 | out_revert: | |
5864 | for_each_subsys(ss, j) { | |
5865 | if (j >= i) | |
5866 | break; | |
5867 | if (ss->cancel_fork) | |
b53202e6 | 5868 | ss->cancel_fork(child); |
7e47682e AS |
5869 | } |
5870 | ||
5871 | return ret; | |
5872 | } | |
5873 | ||
5874 | /** | |
5875 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5876 | * @child: the task in question | |
5877 | * | |
5878 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5879 | * cgroup_can_fork() succeded. | |
5880 | */ | |
b53202e6 | 5881 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5882 | { |
5883 | struct cgroup_subsys *ss; | |
5884 | int i; | |
5885 | ||
5886 | for_each_subsys(ss, i) | |
5887 | if (ss->cancel_fork) | |
b53202e6 | 5888 | ss->cancel_fork(child); |
7e47682e AS |
5889 | } |
5890 | ||
817929ec | 5891 | /** |
a043e3b2 LZ |
5892 | * cgroup_post_fork - called on a new task after adding it to the task list |
5893 | * @child: the task in question | |
5894 | * | |
5edee61e TH |
5895 | * Adds the task to the list running through its css_set if necessary and |
5896 | * call the subsystem fork() callbacks. Has to be after the task is | |
5897 | * visible on the task list in case we race with the first call to | |
0942eeee | 5898 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5899 | * list. |
a043e3b2 | 5900 | */ |
b53202e6 | 5901 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5902 | { |
30159ec7 | 5903 | struct cgroup_subsys *ss; |
5edee61e TH |
5904 | int i; |
5905 | ||
3ce3230a | 5906 | /* |
251f8c03 | 5907 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5908 | * function sets use_task_css_set_links before grabbing |
5909 | * tasklist_lock and we just went through tasklist_lock to add | |
5910 | * @child, it's guaranteed that either we see the set | |
5911 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5912 | * @child during its iteration. | |
5913 | * | |
5914 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5915 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5916 | * association is stable, and, on completion of the parent's |
5917 | * migration, @child is visible in the source of migration or | |
5918 | * already in the destination cgroup. This guarantee is necessary | |
5919 | * when implementing operations which need to migrate all tasks of | |
5920 | * a cgroup to another. | |
5921 | * | |
251f8c03 | 5922 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5923 | * will remain in init_css_set. This is safe because all tasks are |
5924 | * in the init_css_set before cg_links is enabled and there's no | |
5925 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5926 | */ |
817929ec | 5927 | if (use_task_css_set_links) { |
eaf797ab TH |
5928 | struct css_set *cset; |
5929 | ||
82d6489d | 5930 | spin_lock_irq(&css_set_lock); |
0e1d768f | 5931 | cset = task_css_set(current); |
eaf797ab | 5932 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5933 | get_css_set(cset); |
f6d7d049 | 5934 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5935 | } |
82d6489d | 5936 | spin_unlock_irq(&css_set_lock); |
817929ec | 5937 | } |
5edee61e TH |
5938 | |
5939 | /* | |
5940 | * Call ss->fork(). This must happen after @child is linked on | |
5941 | * css_set; otherwise, @child might change state between ->fork() | |
5942 | * and addition to css_set. | |
5943 | */ | |
b4e0eeaf | 5944 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5945 | ss->fork(child); |
b4e0eeaf | 5946 | } while_each_subsys_mask(); |
817929ec | 5947 | } |
5edee61e | 5948 | |
b4f48b63 PM |
5949 | /** |
5950 | * cgroup_exit - detach cgroup from exiting task | |
5951 | * @tsk: pointer to task_struct of exiting process | |
5952 | * | |
5953 | * Description: Detach cgroup from @tsk and release it. | |
5954 | * | |
5955 | * Note that cgroups marked notify_on_release force every task in | |
5956 | * them to take the global cgroup_mutex mutex when exiting. | |
5957 | * This could impact scaling on very large systems. Be reluctant to | |
5958 | * use notify_on_release cgroups where very high task exit scaling | |
5959 | * is required on large systems. | |
5960 | * | |
0e1d768f TH |
5961 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5962 | * call cgroup_exit() while the task is still competent to handle | |
5963 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5964 | * each hierarchy for the remainder of its exit. No need to bother with | |
5965 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5966 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5967 | */ |
1ec41830 | 5968 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5969 | { |
30159ec7 | 5970 | struct cgroup_subsys *ss; |
5abb8855 | 5971 | struct css_set *cset; |
d41d5a01 | 5972 | int i; |
817929ec PM |
5973 | |
5974 | /* | |
0e1d768f | 5975 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5976 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5977 | */ |
0de0942d TH |
5978 | cset = task_css_set(tsk); |
5979 | ||
817929ec | 5980 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 5981 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 5982 | css_set_move_task(tsk, cset, NULL, false); |
82d6489d | 5983 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
5984 | } else { |
5985 | get_css_set(cset); | |
817929ec PM |
5986 | } |
5987 | ||
cb4a3167 | 5988 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5989 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5990 | ss->exit(tsk); |
b4e0eeaf | 5991 | } while_each_subsys_mask(); |
2e91fa7f | 5992 | } |
30159ec7 | 5993 | |
2e91fa7f TH |
5994 | void cgroup_free(struct task_struct *task) |
5995 | { | |
5996 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5997 | struct cgroup_subsys *ss; |
5998 | int ssid; | |
5999 | ||
b4e0eeaf | 6000 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 6001 | ss->free(task); |
b4e0eeaf | 6002 | } while_each_subsys_mask(); |
d41d5a01 | 6003 | |
2e91fa7f | 6004 | put_css_set(cset); |
b4f48b63 | 6005 | } |
697f4161 | 6006 | |
bd89aabc | 6007 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 6008 | { |
27bd4dbb | 6009 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
6010 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
6011 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
6012 | } |
6013 | ||
81a6a5cd PM |
6014 | /* |
6015 | * Notify userspace when a cgroup is released, by running the | |
6016 | * configured release agent with the name of the cgroup (path | |
6017 | * relative to the root of cgroup file system) as the argument. | |
6018 | * | |
6019 | * Most likely, this user command will try to rmdir this cgroup. | |
6020 | * | |
6021 | * This races with the possibility that some other task will be | |
6022 | * attached to this cgroup before it is removed, or that some other | |
6023 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
6024 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
6025 | * unused, and this cgroup will be reprieved from its death sentence, | |
6026 | * to continue to serve a useful existence. Next time it's released, | |
6027 | * we will get notified again, if it still has 'notify_on_release' set. | |
6028 | * | |
6029 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
6030 | * means only wait until the task is successfully execve()'d. The | |
6031 | * separate release agent task is forked by call_usermodehelper(), | |
6032 | * then control in this thread returns here, without waiting for the | |
6033 | * release agent task. We don't bother to wait because the caller of | |
6034 | * this routine has no use for the exit status of the release agent | |
6035 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 6036 | */ |
81a6a5cd PM |
6037 | static void cgroup_release_agent(struct work_struct *work) |
6038 | { | |
971ff493 ZL |
6039 | struct cgroup *cgrp = |
6040 | container_of(work, struct cgroup, release_agent_work); | |
6041 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
6042 | char *argv[3], *envp[3]; | |
6043 | ||
81a6a5cd | 6044 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
6045 | |
6046 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
6047 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
6048 | if (!pathbuf || !agentbuf) | |
6049 | goto out; | |
6050 | ||
82d6489d | 6051 | spin_lock_irq(&css_set_lock); |
a79a908f | 6052 | path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); |
82d6489d | 6053 | spin_unlock_irq(&css_set_lock); |
971ff493 ZL |
6054 | if (!path) |
6055 | goto out; | |
6056 | ||
6057 | argv[0] = agentbuf; | |
6058 | argv[1] = path; | |
6059 | argv[2] = NULL; | |
6060 | ||
6061 | /* minimal command environment */ | |
6062 | envp[0] = "HOME=/"; | |
6063 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
6064 | envp[2] = NULL; | |
6065 | ||
81a6a5cd | 6066 | mutex_unlock(&cgroup_mutex); |
971ff493 | 6067 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 6068 | goto out_free; |
971ff493 | 6069 | out: |
81a6a5cd | 6070 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 6071 | out_free: |
971ff493 ZL |
6072 | kfree(agentbuf); |
6073 | kfree(pathbuf); | |
81a6a5cd | 6074 | } |
8bab8dde PM |
6075 | |
6076 | static int __init cgroup_disable(char *str) | |
6077 | { | |
30159ec7 | 6078 | struct cgroup_subsys *ss; |
8bab8dde | 6079 | char *token; |
30159ec7 | 6080 | int i; |
8bab8dde PM |
6081 | |
6082 | while ((token = strsep(&str, ",")) != NULL) { | |
6083 | if (!*token) | |
6084 | continue; | |
be45c900 | 6085 | |
3ed80a62 | 6086 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6087 | if (strcmp(token, ss->name) && |
6088 | strcmp(token, ss->legacy_name)) | |
6089 | continue; | |
a3e72739 | 6090 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
6091 | } |
6092 | } | |
6093 | return 1; | |
6094 | } | |
6095 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6096 | |
223ffb29 JW |
6097 | static int __init cgroup_no_v1(char *str) |
6098 | { | |
6099 | struct cgroup_subsys *ss; | |
6100 | char *token; | |
6101 | int i; | |
6102 | ||
6103 | while ((token = strsep(&str, ",")) != NULL) { | |
6104 | if (!*token) | |
6105 | continue; | |
6106 | ||
6107 | if (!strcmp(token, "all")) { | |
6e5c8307 | 6108 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
6109 | break; |
6110 | } | |
6111 | ||
6112 | for_each_subsys(ss, i) { | |
6113 | if (strcmp(token, ss->name) && | |
6114 | strcmp(token, ss->legacy_name)) | |
6115 | continue; | |
6116 | ||
6117 | cgroup_no_v1_mask |= 1 << i; | |
6118 | } | |
6119 | } | |
6120 | return 1; | |
6121 | } | |
6122 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
6123 | ||
b77d7b60 | 6124 | /** |
ec903c0c | 6125 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6126 | * @dentry: directory dentry of interest |
6127 | * @ss: subsystem of interest | |
b77d7b60 | 6128 | * |
5a17f543 TH |
6129 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6130 | * to get the corresponding css and return it. If such css doesn't exist | |
6131 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6132 | */ |
ec903c0c TH |
6133 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6134 | struct cgroup_subsys *ss) | |
e5d1367f | 6135 | { |
2bd59d48 | 6136 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6137 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6138 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6139 | struct cgroup *cgrp; |
e5d1367f | 6140 | |
35cf0836 | 6141 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6142 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6143 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6144 | return ERR_PTR(-EBADF); |
6145 | ||
5a17f543 TH |
6146 | rcu_read_lock(); |
6147 | ||
2bd59d48 TH |
6148 | /* |
6149 | * This path doesn't originate from kernfs and @kn could already | |
6150 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6151 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
6152 | */ |
6153 | cgrp = rcu_dereference(kn->priv); | |
6154 | if (cgrp) | |
6155 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6156 | |
ec903c0c | 6157 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6158 | css = ERR_PTR(-ENOENT); |
6159 | ||
6160 | rcu_read_unlock(); | |
6161 | return css; | |
e5d1367f | 6162 | } |
e5d1367f | 6163 | |
1cb650b9 LZ |
6164 | /** |
6165 | * css_from_id - lookup css by id | |
6166 | * @id: the cgroup id | |
6167 | * @ss: cgroup subsys to be looked into | |
6168 | * | |
6169 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6170 | * Should be called under rcu_read_lock(). | |
6171 | */ | |
6172 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6173 | { | |
6fa4918d | 6174 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 6175 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
6176 | } |
6177 | ||
16af4396 TH |
6178 | /** |
6179 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6180 | * @path: path on the default hierarchy | |
6181 | * | |
6182 | * Find the cgroup at @path on the default hierarchy, increment its | |
6183 | * reference count and return it. Returns pointer to the found cgroup on | |
6184 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6185 | * if @path points to a non-directory. | |
6186 | */ | |
6187 | struct cgroup *cgroup_get_from_path(const char *path) | |
6188 | { | |
6189 | struct kernfs_node *kn; | |
6190 | struct cgroup *cgrp; | |
6191 | ||
6192 | mutex_lock(&cgroup_mutex); | |
6193 | ||
6194 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6195 | if (kn) { | |
6196 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6197 | cgrp = kn->priv; | |
6198 | cgroup_get(cgrp); | |
6199 | } else { | |
6200 | cgrp = ERR_PTR(-ENOTDIR); | |
6201 | } | |
6202 | kernfs_put(kn); | |
6203 | } else { | |
6204 | cgrp = ERR_PTR(-ENOENT); | |
6205 | } | |
6206 | ||
6207 | mutex_unlock(&cgroup_mutex); | |
6208 | return cgrp; | |
6209 | } | |
6210 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6211 | ||
bd1060a1 TH |
6212 | /* |
6213 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6214 | * definition in cgroup-defs.h. | |
6215 | */ | |
6216 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6217 | ||
6218 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6219 | ||
3fa4cc9c | 6220 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6221 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6222 | ||
6223 | void cgroup_sk_alloc_disable(void) | |
6224 | { | |
6225 | if (cgroup_sk_alloc_disabled) | |
6226 | return; | |
6227 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6228 | cgroup_sk_alloc_disabled = true; | |
6229 | } | |
6230 | ||
6231 | #else | |
6232 | ||
6233 | #define cgroup_sk_alloc_disabled false | |
6234 | ||
6235 | #endif | |
6236 | ||
6237 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6238 | { | |
6239 | if (cgroup_sk_alloc_disabled) | |
6240 | return; | |
6241 | ||
6242 | rcu_read_lock(); | |
6243 | ||
6244 | while (true) { | |
6245 | struct css_set *cset; | |
6246 | ||
6247 | cset = task_css_set(current); | |
6248 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6249 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6250 | break; | |
6251 | } | |
6252 | cpu_relax(); | |
6253 | } | |
6254 | ||
6255 | rcu_read_unlock(); | |
6256 | } | |
6257 | ||
6258 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6259 | { | |
6260 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6261 | } | |
6262 | ||
6263 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6264 | ||
a79a908f AK |
6265 | /* cgroup namespaces */ |
6266 | ||
6267 | static struct cgroup_namespace *alloc_cgroup_ns(void) | |
6268 | { | |
6269 | struct cgroup_namespace *new_ns; | |
6270 | int ret; | |
6271 | ||
6272 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); | |
6273 | if (!new_ns) | |
6274 | return ERR_PTR(-ENOMEM); | |
6275 | ret = ns_alloc_inum(&new_ns->ns); | |
6276 | if (ret) { | |
6277 | kfree(new_ns); | |
6278 | return ERR_PTR(ret); | |
6279 | } | |
6280 | atomic_set(&new_ns->count, 1); | |
6281 | new_ns->ns.ops = &cgroupns_operations; | |
6282 | return new_ns; | |
6283 | } | |
6284 | ||
6285 | void free_cgroup_ns(struct cgroup_namespace *ns) | |
6286 | { | |
6287 | put_css_set(ns->root_cset); | |
6288 | put_user_ns(ns->user_ns); | |
6289 | ns_free_inum(&ns->ns); | |
6290 | kfree(ns); | |
6291 | } | |
6292 | EXPORT_SYMBOL(free_cgroup_ns); | |
6293 | ||
6294 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, | |
6295 | struct user_namespace *user_ns, | |
6296 | struct cgroup_namespace *old_ns) | |
6297 | { | |
fa5ff8a1 TH |
6298 | struct cgroup_namespace *new_ns; |
6299 | struct css_set *cset; | |
a79a908f AK |
6300 | |
6301 | BUG_ON(!old_ns); | |
6302 | ||
6303 | if (!(flags & CLONE_NEWCGROUP)) { | |
6304 | get_cgroup_ns(old_ns); | |
6305 | return old_ns; | |
6306 | } | |
6307 | ||
6308 | /* Allow only sysadmin to create cgroup namespace. */ | |
a79a908f | 6309 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
fa5ff8a1 | 6310 | return ERR_PTR(-EPERM); |
a79a908f AK |
6311 | |
6312 | mutex_lock(&cgroup_mutex); | |
82d6489d | 6313 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
6314 | |
6315 | cset = task_css_set(current); | |
6316 | get_css_set(cset); | |
6317 | ||
82d6489d | 6318 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
6319 | mutex_unlock(&cgroup_mutex); |
6320 | ||
a79a908f | 6321 | new_ns = alloc_cgroup_ns(); |
d2202557 | 6322 | if (IS_ERR(new_ns)) { |
fa5ff8a1 TH |
6323 | put_css_set(cset); |
6324 | return new_ns; | |
d2202557 | 6325 | } |
a79a908f AK |
6326 | |
6327 | new_ns->user_ns = get_user_ns(user_ns); | |
6328 | new_ns->root_cset = cset; | |
6329 | ||
6330 | return new_ns; | |
a79a908f AK |
6331 | } |
6332 | ||
6333 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) | |
6334 | { | |
6335 | return container_of(ns, struct cgroup_namespace, ns); | |
6336 | } | |
6337 | ||
a0530e08 | 6338 | static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
a79a908f | 6339 | { |
a0530e08 AK |
6340 | struct cgroup_namespace *cgroup_ns = to_cg_ns(ns); |
6341 | ||
6342 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) || | |
6343 | !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN)) | |
6344 | return -EPERM; | |
6345 | ||
6346 | /* Don't need to do anything if we are attaching to our own cgroupns. */ | |
6347 | if (cgroup_ns == nsproxy->cgroup_ns) | |
6348 | return 0; | |
6349 | ||
6350 | get_cgroup_ns(cgroup_ns); | |
6351 | put_cgroup_ns(nsproxy->cgroup_ns); | |
6352 | nsproxy->cgroup_ns = cgroup_ns; | |
6353 | ||
6354 | return 0; | |
a79a908f AK |
6355 | } |
6356 | ||
6357 | static struct ns_common *cgroupns_get(struct task_struct *task) | |
6358 | { | |
6359 | struct cgroup_namespace *ns = NULL; | |
6360 | struct nsproxy *nsproxy; | |
6361 | ||
6362 | task_lock(task); | |
6363 | nsproxy = task->nsproxy; | |
6364 | if (nsproxy) { | |
6365 | ns = nsproxy->cgroup_ns; | |
6366 | get_cgroup_ns(ns); | |
6367 | } | |
6368 | task_unlock(task); | |
6369 | ||
6370 | return ns ? &ns->ns : NULL; | |
6371 | } | |
6372 | ||
6373 | static void cgroupns_put(struct ns_common *ns) | |
6374 | { | |
6375 | put_cgroup_ns(to_cg_ns(ns)); | |
6376 | } | |
6377 | ||
6378 | const struct proc_ns_operations cgroupns_operations = { | |
6379 | .name = "cgroup", | |
6380 | .type = CLONE_NEWCGROUP, | |
6381 | .get = cgroupns_get, | |
6382 | .put = cgroupns_put, | |
6383 | .install = cgroupns_install, | |
6384 | }; | |
6385 | ||
6386 | static __init int cgroup_namespaces_init(void) | |
6387 | { | |
6388 | return 0; | |
6389 | } | |
6390 | subsys_initcall(cgroup_namespaces_init); | |
6391 | ||
fe693435 | 6392 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6393 | static struct cgroup_subsys_state * |
6394 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6395 | { |
6396 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6397 | ||
6398 | if (!css) | |
6399 | return ERR_PTR(-ENOMEM); | |
6400 | ||
6401 | return css; | |
6402 | } | |
6403 | ||
eb95419b | 6404 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6405 | { |
eb95419b | 6406 | kfree(css); |
fe693435 PM |
6407 | } |
6408 | ||
182446d0 TH |
6409 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6410 | struct cftype *cft) | |
fe693435 | 6411 | { |
182446d0 | 6412 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6413 | } |
6414 | ||
182446d0 TH |
6415 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6416 | struct cftype *cft) | |
fe693435 PM |
6417 | { |
6418 | return (u64)(unsigned long)current->cgroups; | |
6419 | } | |
6420 | ||
182446d0 | 6421 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6422 | struct cftype *cft) |
fe693435 PM |
6423 | { |
6424 | u64 count; | |
6425 | ||
6426 | rcu_read_lock(); | |
a8ad805c | 6427 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6428 | rcu_read_unlock(); |
6429 | return count; | |
6430 | } | |
6431 | ||
2da8ca82 | 6432 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6433 | { |
69d0206c | 6434 | struct cgrp_cset_link *link; |
5abb8855 | 6435 | struct css_set *cset; |
e61734c5 TH |
6436 | char *name_buf; |
6437 | ||
6438 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6439 | if (!name_buf) | |
6440 | return -ENOMEM; | |
7717f7ba | 6441 | |
82d6489d | 6442 | spin_lock_irq(&css_set_lock); |
7717f7ba | 6443 | rcu_read_lock(); |
5abb8855 | 6444 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6445 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6446 | struct cgroup *c = link->cgrp; |
7717f7ba | 6447 | |
a2dd4247 | 6448 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6449 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6450 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6451 | } |
6452 | rcu_read_unlock(); | |
82d6489d | 6453 | spin_unlock_irq(&css_set_lock); |
e61734c5 | 6454 | kfree(name_buf); |
7717f7ba PM |
6455 | return 0; |
6456 | } | |
6457 | ||
6458 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6459 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6460 | { |
2da8ca82 | 6461 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6462 | struct cgrp_cset_link *link; |
7717f7ba | 6463 | |
82d6489d | 6464 | spin_lock_irq(&css_set_lock); |
182446d0 | 6465 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6466 | struct css_set *cset = link->cset; |
7717f7ba PM |
6467 | struct task_struct *task; |
6468 | int count = 0; | |
c7561128 | 6469 | |
5abb8855 | 6470 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6471 | |
5abb8855 | 6472 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6473 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6474 | goto overflow; | |
6475 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6476 | } | |
6477 | ||
6478 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6479 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6480 | goto overflow; | |
6481 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6482 | } |
c7561128 TH |
6483 | continue; |
6484 | overflow: | |
6485 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6486 | } |
82d6489d | 6487 | spin_unlock_irq(&css_set_lock); |
7717f7ba PM |
6488 | return 0; |
6489 | } | |
6490 | ||
182446d0 | 6491 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6492 | { |
27bd4dbb | 6493 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6494 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6495 | } |
6496 | ||
6497 | static struct cftype debug_files[] = { | |
fe693435 PM |
6498 | { |
6499 | .name = "taskcount", | |
6500 | .read_u64 = debug_taskcount_read, | |
6501 | }, | |
6502 | ||
6503 | { | |
6504 | .name = "current_css_set", | |
6505 | .read_u64 = current_css_set_read, | |
6506 | }, | |
6507 | ||
6508 | { | |
6509 | .name = "current_css_set_refcount", | |
6510 | .read_u64 = current_css_set_refcount_read, | |
6511 | }, | |
6512 | ||
7717f7ba PM |
6513 | { |
6514 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6515 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6516 | }, |
6517 | ||
6518 | { | |
6519 | .name = "cgroup_css_links", | |
2da8ca82 | 6520 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6521 | }, |
6522 | ||
fe693435 PM |
6523 | { |
6524 | .name = "releasable", | |
6525 | .read_u64 = releasable_read, | |
6526 | }, | |
fe693435 | 6527 | |
4baf6e33 TH |
6528 | { } /* terminate */ |
6529 | }; | |
fe693435 | 6530 | |
073219e9 | 6531 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6532 | .css_alloc = debug_css_alloc, |
6533 | .css_free = debug_css_free, | |
5577964e | 6534 | .legacy_cftypes = debug_files, |
fe693435 PM |
6535 | }; |
6536 | #endif /* CONFIG_CGROUP_DEBUG */ |