]>
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> | |
1f3fe7eb | 64 | #include <linux/file.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 | ||
8c8a5502 | 1163 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1164 | } |
1165 | ||
3dd06ffa | 1166 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1167 | { |
1168 | if (root) { | |
f2e85d57 TH |
1169 | idr_destroy(&root->cgroup_idr); |
1170 | kfree(root); | |
1171 | } | |
1172 | } | |
1173 | ||
3dd06ffa | 1174 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1175 | { |
3dd06ffa | 1176 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1177 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1178 | |
334c3679 | 1179 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1180 | |
776f02fa | 1181 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1182 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1183 | |
f2e85d57 | 1184 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1185 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1186 | |
7717f7ba | 1187 | /* |
f2e85d57 TH |
1188 | * Release all the links from cset_links to this hierarchy's |
1189 | * root cgroup | |
7717f7ba | 1190 | */ |
82d6489d | 1191 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1192 | |
1193 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1194 | list_del(&link->cset_link); | |
1195 | list_del(&link->cgrp_link); | |
1196 | kfree(link); | |
1197 | } | |
f0d9a5f1 | 1198 | |
82d6489d | 1199 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1200 | |
1201 | if (!list_empty(&root->root_list)) { | |
1202 | list_del(&root->root_list); | |
1203 | cgroup_root_count--; | |
1204 | } | |
1205 | ||
1206 | cgroup_exit_root_id(root); | |
1207 | ||
1208 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1209 | |
2bd59d48 | 1210 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1211 | cgroup_free_root(root); |
1212 | } | |
1213 | ||
4f41fc59 SH |
1214 | /* |
1215 | * look up cgroup associated with current task's cgroup namespace on the | |
1216 | * specified hierarchy | |
1217 | */ | |
1218 | static struct cgroup * | |
1219 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1220 | { | |
1221 | struct cgroup *res = NULL; | |
1222 | struct css_set *cset; | |
1223 | ||
1224 | lockdep_assert_held(&css_set_lock); | |
1225 | ||
1226 | rcu_read_lock(); | |
1227 | ||
1228 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1229 | if (cset == &init_css_set) { | |
1230 | res = &root->cgrp; | |
1231 | } else { | |
1232 | struct cgrp_cset_link *link; | |
1233 | ||
1234 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1235 | struct cgroup *c = link->cgrp; | |
1236 | ||
1237 | if (c->root == root) { | |
1238 | res = c; | |
1239 | break; | |
1240 | } | |
1241 | } | |
1242 | } | |
1243 | rcu_read_unlock(); | |
1244 | ||
1245 | BUG_ON(!res); | |
1246 | return res; | |
1247 | } | |
1248 | ||
ceb6a081 TH |
1249 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1250 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1251 | struct cgroup_root *root) |
7717f7ba | 1252 | { |
7717f7ba PM |
1253 | struct cgroup *res = NULL; |
1254 | ||
96d365e0 | 1255 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1256 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1257 | |
5abb8855 | 1258 | if (cset == &init_css_set) { |
3dd06ffa | 1259 | res = &root->cgrp; |
7717f7ba | 1260 | } else { |
69d0206c TH |
1261 | struct cgrp_cset_link *link; |
1262 | ||
1263 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1264 | struct cgroup *c = link->cgrp; |
69d0206c | 1265 | |
7717f7ba PM |
1266 | if (c->root == root) { |
1267 | res = c; | |
1268 | break; | |
1269 | } | |
1270 | } | |
1271 | } | |
96d365e0 | 1272 | |
7717f7ba PM |
1273 | BUG_ON(!res); |
1274 | return res; | |
1275 | } | |
1276 | ||
ddbcc7e8 | 1277 | /* |
ceb6a081 | 1278 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1279 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1280 | */ |
1281 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1282 | struct cgroup_root *root) |
ceb6a081 TH |
1283 | { |
1284 | /* | |
1285 | * No need to lock the task - since we hold cgroup_mutex the | |
1286 | * task can't change groups, so the only thing that can happen | |
1287 | * is that it exits and its css is set back to init_css_set. | |
1288 | */ | |
1289 | return cset_cgroup_from_root(task_css_set(task), root); | |
1290 | } | |
1291 | ||
ddbcc7e8 | 1292 | /* |
ddbcc7e8 PM |
1293 | * A task must hold cgroup_mutex to modify cgroups. |
1294 | * | |
1295 | * Any task can increment and decrement the count field without lock. | |
1296 | * So in general, code holding cgroup_mutex can't rely on the count | |
1297 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1298 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1299 | * means that no tasks are currently attached, therefore there is no |
1300 | * way a task attached to that cgroup can fork (the other way to | |
1301 | * increment the count). So code holding cgroup_mutex can safely | |
1302 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1303 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1304 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1305 | * needs that mutex. | |
1306 | * | |
ddbcc7e8 PM |
1307 | * A cgroup can only be deleted if both its 'count' of using tasks |
1308 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1309 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1310 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1311 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1312 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1313 | * |
1314 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1315 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1316 | */ |
1317 | ||
2bd59d48 | 1318 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1319 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1320 | |
8d7e6fb0 TH |
1321 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1322 | char *buf) | |
ddbcc7e8 | 1323 | { |
3e1d2eed TH |
1324 | struct cgroup_subsys *ss = cft->ss; |
1325 | ||
8d7e6fb0 TH |
1326 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1327 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1328 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1329 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1330 | cft->name); | |
8d7e6fb0 TH |
1331 | else |
1332 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1333 | return buf; | |
ddbcc7e8 PM |
1334 | } |
1335 | ||
f2e85d57 TH |
1336 | /** |
1337 | * cgroup_file_mode - deduce file mode of a control file | |
1338 | * @cft: the control file in question | |
1339 | * | |
7dbdb199 | 1340 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1341 | */ |
1342 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1343 | { |
f2e85d57 | 1344 | umode_t mode = 0; |
65dff759 | 1345 | |
f2e85d57 TH |
1346 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1347 | mode |= S_IRUGO; | |
1348 | ||
7dbdb199 TH |
1349 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1350 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1351 | mode |= S_IWUGO; | |
1352 | else | |
1353 | mode |= S_IWUSR; | |
1354 | } | |
f2e85d57 TH |
1355 | |
1356 | return mode; | |
65dff759 LZ |
1357 | } |
1358 | ||
af0ba678 | 1359 | /** |
8699b776 | 1360 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1361 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1362 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1363 | * |
1364 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1365 | * enabled together through its ->depends_on mask. In such cases, more | |
1366 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1367 | * | |
0f060deb | 1368 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1369 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1370 | */ |
5ced2518 | 1371 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1372 | { |
6e5c8307 | 1373 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1374 | struct cgroup_subsys *ss; |
1375 | int ssid; | |
1376 | ||
1377 | lockdep_assert_held(&cgroup_mutex); | |
1378 | ||
f6d635ad TH |
1379 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1380 | ||
af0ba678 | 1381 | while (true) { |
6e5c8307 | 1382 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1383 | |
b4e0eeaf | 1384 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1385 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1386 | } while_each_subsys_mask(); |
af0ba678 TH |
1387 | |
1388 | /* | |
1389 | * Mask out subsystems which aren't available. This can | |
1390 | * happen only if some depended-upon subsystems were bound | |
1391 | * to non-default hierarchies. | |
1392 | */ | |
5ced2518 | 1393 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1394 | |
1395 | if (new_ss_mask == cur_ss_mask) | |
1396 | break; | |
1397 | cur_ss_mask = new_ss_mask; | |
1398 | } | |
1399 | ||
0f060deb TH |
1400 | return cur_ss_mask; |
1401 | } | |
1402 | ||
a9746d8d TH |
1403 | /** |
1404 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1405 | * @kn: the kernfs_node being serviced | |
1406 | * | |
1407 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1408 | * the method finishes if locking succeeded. Note that once this function | |
1409 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1410 | * inaccessible any time. If the caller intends to continue to access the | |
1411 | * cgroup, it should pin it before invoking this function. | |
1412 | */ | |
1413 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1414 | { |
a9746d8d TH |
1415 | struct cgroup *cgrp; |
1416 | ||
1417 | if (kernfs_type(kn) == KERNFS_DIR) | |
1418 | cgrp = kn->priv; | |
1419 | else | |
1420 | cgrp = kn->parent->priv; | |
1421 | ||
1422 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1423 | |
1424 | kernfs_unbreak_active_protection(kn); | |
1425 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1426 | } |
1427 | ||
a9746d8d TH |
1428 | /** |
1429 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1430 | * @kn: the kernfs_node being serviced | |
945ba199 | 1431 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1432 | * |
1433 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1434 | * @kn. It breaks the active protection, performs cgroup locking and | |
1435 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1436 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1437 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1438 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1439 | * |
1440 | * Any cgroup kernfs method implementation which requires locking the | |
1441 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1442 | * locking under kernfs active protection and allows all kernfs operations | |
1443 | * including self-removal. | |
1444 | */ | |
945ba199 TH |
1445 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1446 | bool drain_offline) | |
05ef1d7c | 1447 | { |
a9746d8d TH |
1448 | struct cgroup *cgrp; |
1449 | ||
1450 | if (kernfs_type(kn) == KERNFS_DIR) | |
1451 | cgrp = kn->priv; | |
1452 | else | |
1453 | cgrp = kn->parent->priv; | |
05ef1d7c | 1454 | |
2739d3cc | 1455 | /* |
01f6474c | 1456 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1457 | * active_ref. cgroup liveliness check alone provides enough |
1458 | * protection against removal. Ensure @cgrp stays accessible and | |
1459 | * break the active_ref protection. | |
2739d3cc | 1460 | */ |
aa32362f LZ |
1461 | if (!cgroup_tryget(cgrp)) |
1462 | return NULL; | |
a9746d8d TH |
1463 | kernfs_break_active_protection(kn); |
1464 | ||
945ba199 TH |
1465 | if (drain_offline) |
1466 | cgroup_lock_and_drain_offline(cgrp); | |
1467 | else | |
1468 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1469 | |
a9746d8d TH |
1470 | if (!cgroup_is_dead(cgrp)) |
1471 | return cgrp; | |
1472 | ||
1473 | cgroup_kn_unlock(kn); | |
1474 | return NULL; | |
ddbcc7e8 | 1475 | } |
05ef1d7c | 1476 | |
2739d3cc | 1477 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1478 | { |
2bd59d48 | 1479 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1480 | |
01f6474c | 1481 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1482 | |
1483 | if (cft->file_offset) { | |
1484 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1485 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1486 | ||
1487 | spin_lock_irq(&cgroup_file_kn_lock); | |
1488 | cfile->kn = NULL; | |
1489 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1490 | } | |
1491 | ||
2bd59d48 | 1492 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1493 | } |
1494 | ||
13af07df | 1495 | /** |
4df8dc90 TH |
1496 | * css_clear_dir - remove subsys files in a cgroup directory |
1497 | * @css: taget css | |
13af07df | 1498 | */ |
334c3679 | 1499 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1500 | { |
334c3679 | 1501 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1502 | struct cftype *cfts; |
05ef1d7c | 1503 | |
88cb04b9 TH |
1504 | if (!(css->flags & CSS_VISIBLE)) |
1505 | return; | |
1506 | ||
1507 | css->flags &= ~CSS_VISIBLE; | |
1508 | ||
4df8dc90 TH |
1509 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1510 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1511 | } |
1512 | ||
ccdca218 | 1513 | /** |
4df8dc90 TH |
1514 | * css_populate_dir - create subsys files in a cgroup directory |
1515 | * @css: target css | |
ccdca218 TH |
1516 | * |
1517 | * On failure, no file is added. | |
1518 | */ | |
334c3679 | 1519 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1520 | { |
334c3679 | 1521 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1522 | struct cftype *cfts, *failed_cfts; |
1523 | int ret; | |
ccdca218 | 1524 | |
03970d3c | 1525 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1526 | return 0; |
1527 | ||
4df8dc90 TH |
1528 | if (!css->ss) { |
1529 | if (cgroup_on_dfl(cgrp)) | |
1530 | cfts = cgroup_dfl_base_files; | |
1531 | else | |
1532 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1533 | |
4df8dc90 TH |
1534 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1535 | } | |
ccdca218 | 1536 | |
4df8dc90 TH |
1537 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1538 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1539 | if (ret < 0) { | |
1540 | failed_cfts = cfts; | |
1541 | goto err; | |
ccdca218 TH |
1542 | } |
1543 | } | |
88cb04b9 TH |
1544 | |
1545 | css->flags |= CSS_VISIBLE; | |
1546 | ||
ccdca218 TH |
1547 | return 0; |
1548 | err: | |
4df8dc90 TH |
1549 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1550 | if (cfts == failed_cfts) | |
1551 | break; | |
1552 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1553 | } | |
ccdca218 TH |
1554 | return ret; |
1555 | } | |
1556 | ||
6e5c8307 | 1557 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1558 | { |
1ada4838 | 1559 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1560 | struct cgroup_subsys *ss; |
2d8f243a | 1561 | int ssid, i, ret; |
ddbcc7e8 | 1562 | |
ace2bee8 | 1563 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1564 | |
b4e0eeaf | 1565 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1566 | /* |
1567 | * If @ss has non-root csses attached to it, can't move. | |
1568 | * If @ss is an implicit controller, it is exempt from this | |
1569 | * rule and can be stolen. | |
1570 | */ | |
1571 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1572 | !ss->implicit_on_dfl) | |
3ed80a62 | 1573 | return -EBUSY; |
1d5be6b2 | 1574 | |
5df36032 | 1575 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1576 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1577 | return -EBUSY; |
b4e0eeaf | 1578 | } while_each_subsys_mask(); |
ddbcc7e8 | 1579 | |
b4e0eeaf | 1580 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1581 | struct cgroup_root *src_root = ss->root; |
1582 | struct cgroup *scgrp = &src_root->cgrp; | |
1583 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1584 | struct css_set *cset; |
a8a648c4 | 1585 | |
1ada4838 | 1586 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1587 | |
334c3679 TH |
1588 | /* disable from the source */ |
1589 | src_root->subsys_mask &= ~(1 << ssid); | |
1590 | WARN_ON(cgroup_apply_control(scgrp)); | |
1591 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1592 | |
334c3679 | 1593 | /* rebind */ |
1ada4838 TH |
1594 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1595 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1596 | ss->root = dst_root; |
1ada4838 | 1597 | css->cgroup = dcgrp; |
73e80ed8 | 1598 | |
82d6489d | 1599 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1600 | hash_for_each(css_set_table, i, cset, hlist) |
1601 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1602 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1603 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1604 | |
bd53d617 | 1605 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1606 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1607 | if (dst_root == &cgrp_dfl_root) { |
1608 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1609 | } else { | |
1ada4838 | 1610 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1611 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1612 | } |
a8a648c4 | 1613 | |
334c3679 TH |
1614 | ret = cgroup_apply_control(dcgrp); |
1615 | if (ret) | |
1616 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1617 | ss->name, ret); | |
1618 | ||
5df36032 TH |
1619 | if (ss->bind) |
1620 | ss->bind(css); | |
b4e0eeaf | 1621 | } while_each_subsys_mask(); |
ddbcc7e8 | 1622 | |
1ada4838 | 1623 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1624 | return 0; |
1625 | } | |
1626 | ||
4f41fc59 SH |
1627 | static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1628 | struct kernfs_root *kf_root) | |
1629 | { | |
09be4c82 | 1630 | int len = 0; |
4f41fc59 SH |
1631 | char *buf = NULL; |
1632 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1633 | struct cgroup *ns_cgroup; | |
1634 | ||
1635 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1636 | if (!buf) | |
1637 | return -ENOMEM; | |
1638 | ||
82d6489d | 1639 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1640 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1641 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1642 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1643 | |
1644 | if (len >= PATH_MAX) | |
1645 | len = -ERANGE; | |
1646 | else if (len > 0) { | |
1647 | seq_escape(sf, buf, " \t\n\\"); | |
1648 | len = 0; | |
1649 | } | |
1650 | kfree(buf); | |
1651 | return len; | |
1652 | } | |
1653 | ||
2bd59d48 TH |
1654 | static int cgroup_show_options(struct seq_file *seq, |
1655 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1656 | { |
3dd06ffa | 1657 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1658 | struct cgroup_subsys *ss; |
b85d2040 | 1659 | int ssid; |
ddbcc7e8 | 1660 | |
d98817d4 TH |
1661 | if (root != &cgrp_dfl_root) |
1662 | for_each_subsys(ss, ssid) | |
1663 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1664 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1665 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1666 | seq_puts(seq, ",noprefix"); |
93438629 | 1667 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1668 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1669 | |
1670 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1671 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1672 | seq_show_option(seq, "release_agent", |
1673 | root->release_agent_path); | |
69e943b7 TH |
1674 | spin_unlock(&release_agent_path_lock); |
1675 | ||
3dd06ffa | 1676 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1677 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1678 | if (strlen(root->name)) |
a068acf2 | 1679 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1680 | return 0; |
1681 | } | |
1682 | ||
1683 | struct cgroup_sb_opts { | |
6e5c8307 | 1684 | u16 subsys_mask; |
69dfa00c | 1685 | unsigned int flags; |
81a6a5cd | 1686 | char *release_agent; |
2260e7fc | 1687 | bool cpuset_clone_children; |
c6d57f33 | 1688 | char *name; |
2c6ab6d2 PM |
1689 | /* User explicitly requested empty subsystem */ |
1690 | bool none; | |
ddbcc7e8 PM |
1691 | }; |
1692 | ||
cf5d5941 | 1693 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1694 | { |
32a8cf23 DL |
1695 | char *token, *o = data; |
1696 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1697 | u16 mask = U16_MAX; |
30159ec7 | 1698 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1699 | int nr_opts = 0; |
30159ec7 | 1700 | int i; |
f9ab5b5b LZ |
1701 | |
1702 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1703 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1704 | #endif |
ddbcc7e8 | 1705 | |
c6d57f33 | 1706 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1707 | |
1708 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1709 | nr_opts++; |
1710 | ||
ddbcc7e8 PM |
1711 | if (!*token) |
1712 | return -EINVAL; | |
32a8cf23 | 1713 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1714 | /* Explicitly have no subsystems */ |
1715 | opts->none = true; | |
32a8cf23 DL |
1716 | continue; |
1717 | } | |
1718 | if (!strcmp(token, "all")) { | |
1719 | /* Mutually exclusive option 'all' + subsystem name */ | |
1720 | if (one_ss) | |
1721 | return -EINVAL; | |
1722 | all_ss = true; | |
1723 | continue; | |
1724 | } | |
1725 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1726 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1727 | continue; |
1728 | } | |
1729 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1730 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1731 | continue; |
1732 | } | |
03b1cde6 | 1733 | if (!strcmp(token, "xattr")) { |
93438629 | 1734 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1735 | continue; |
1736 | } | |
32a8cf23 | 1737 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1738 | /* Specifying two release agents is forbidden */ |
1739 | if (opts->release_agent) | |
1740 | return -EINVAL; | |
c6d57f33 | 1741 | opts->release_agent = |
e400c285 | 1742 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1743 | if (!opts->release_agent) |
1744 | return -ENOMEM; | |
32a8cf23 DL |
1745 | continue; |
1746 | } | |
1747 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1748 | const char *name = token + 5; |
1749 | /* Can't specify an empty name */ | |
1750 | if (!strlen(name)) | |
1751 | return -EINVAL; | |
1752 | /* Must match [\w.-]+ */ | |
1753 | for (i = 0; i < strlen(name); i++) { | |
1754 | char c = name[i]; | |
1755 | if (isalnum(c)) | |
1756 | continue; | |
1757 | if ((c == '.') || (c == '-') || (c == '_')) | |
1758 | continue; | |
1759 | return -EINVAL; | |
1760 | } | |
1761 | /* Specifying two names is forbidden */ | |
1762 | if (opts->name) | |
1763 | return -EINVAL; | |
1764 | opts->name = kstrndup(name, | |
e400c285 | 1765 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1766 | GFP_KERNEL); |
1767 | if (!opts->name) | |
1768 | return -ENOMEM; | |
32a8cf23 DL |
1769 | |
1770 | continue; | |
1771 | } | |
1772 | ||
30159ec7 | 1773 | for_each_subsys(ss, i) { |
3e1d2eed | 1774 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1775 | continue; |
fc5ed1e9 | 1776 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1777 | continue; |
223ffb29 JW |
1778 | if (cgroup_ssid_no_v1(i)) |
1779 | continue; | |
32a8cf23 DL |
1780 | |
1781 | /* Mutually exclusive option 'all' + subsystem name */ | |
1782 | if (all_ss) | |
1783 | return -EINVAL; | |
69dfa00c | 1784 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1785 | one_ss = true; |
1786 | ||
1787 | break; | |
1788 | } | |
1789 | if (i == CGROUP_SUBSYS_COUNT) | |
1790 | return -ENOENT; | |
1791 | } | |
1792 | ||
7b9a6ba5 TH |
1793 | /* |
1794 | * If the 'all' option was specified select all the subsystems, | |
1795 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1796 | * not specified, let's default to 'all' | |
1797 | */ | |
1798 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1799 | for_each_subsys(ss, i) | |
223ffb29 | 1800 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1801 | opts->subsys_mask |= (1 << i); |
1802 | ||
1803 | /* | |
1804 | * We either have to specify by name or by subsystems. (So all | |
1805 | * empty hierarchies must have a name). | |
1806 | */ | |
1807 | if (!opts->subsys_mask && !opts->name) | |
1808 | return -EINVAL; | |
1809 | ||
f9ab5b5b LZ |
1810 | /* |
1811 | * Option noprefix was introduced just for backward compatibility | |
1812 | * with the old cpuset, so we allow noprefix only if mounting just | |
1813 | * the cpuset subsystem. | |
1814 | */ | |
93438629 | 1815 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1816 | return -EINVAL; |
1817 | ||
2c6ab6d2 | 1818 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1819 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1820 | return -EINVAL; |
1821 | ||
ddbcc7e8 PM |
1822 | return 0; |
1823 | } | |
1824 | ||
2bd59d48 | 1825 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1826 | { |
1827 | int ret = 0; | |
3dd06ffa | 1828 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1829 | struct cgroup_sb_opts opts; |
6e5c8307 | 1830 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1831 | |
aa6ec29b TH |
1832 | if (root == &cgrp_dfl_root) { |
1833 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1834 | return -EINVAL; |
1835 | } | |
1836 | ||
334c3679 | 1837 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1838 | |
1839 | /* See what subsystems are wanted */ | |
1840 | ret = parse_cgroupfs_options(data, &opts); | |
1841 | if (ret) | |
1842 | goto out_unlock; | |
1843 | ||
f392e51c | 1844 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1845 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1846 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1847 | |
f392e51c TH |
1848 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1849 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1850 | |
cf5d5941 | 1851 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1852 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1853 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1854 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1855 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1856 | ret = -EINVAL; |
1857 | goto out_unlock; | |
1858 | } | |
1859 | ||
f172e67c | 1860 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1861 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1862 | ret = -EBUSY; |
0670e08b | 1863 | goto out_unlock; |
cf5d5941 | 1864 | } |
ddbcc7e8 | 1865 | |
5df36032 | 1866 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1867 | if (ret) |
0670e08b | 1868 | goto out_unlock; |
ddbcc7e8 | 1869 | |
334c3679 | 1870 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1871 | |
69e943b7 TH |
1872 | if (opts.release_agent) { |
1873 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1874 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1875 | spin_unlock(&release_agent_path_lock); |
1876 | } | |
ddbcc7e8 | 1877 | out_unlock: |
66bdc9cf | 1878 | kfree(opts.release_agent); |
c6d57f33 | 1879 | kfree(opts.name); |
ddbcc7e8 | 1880 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1881 | return ret; |
1882 | } | |
1883 | ||
afeb0f9f TH |
1884 | /* |
1885 | * To reduce the fork() overhead for systems that are not actually using | |
1886 | * their cgroups capability, we don't maintain the lists running through | |
1887 | * each css_set to its tasks until we see the list actually used - in other | |
1888 | * words after the first mount. | |
1889 | */ | |
1890 | static bool use_task_css_set_links __read_mostly; | |
1891 | ||
1892 | static void cgroup_enable_task_cg_lists(void) | |
1893 | { | |
1894 | struct task_struct *p, *g; | |
1895 | ||
82d6489d | 1896 | spin_lock_irq(&css_set_lock); |
afeb0f9f TH |
1897 | |
1898 | if (use_task_css_set_links) | |
1899 | goto out_unlock; | |
1900 | ||
1901 | use_task_css_set_links = true; | |
1902 | ||
1903 | /* | |
1904 | * We need tasklist_lock because RCU is not safe against | |
1905 | * while_each_thread(). Besides, a forking task that has passed | |
1906 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1907 | * is not guaranteed to have its child immediately visible in the | |
1908 | * tasklist if we walk through it with RCU. | |
1909 | */ | |
1910 | read_lock(&tasklist_lock); | |
1911 | do_each_thread(g, p) { | |
afeb0f9f TH |
1912 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1913 | task_css_set(p) != &init_css_set); | |
1914 | ||
1915 | /* | |
1916 | * We should check if the process is exiting, otherwise | |
1917 | * it will race with cgroup_exit() in that the list | |
1918 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1919 | * Do it while holding siglock so that we don't end up |
1920 | * racing against cgroup_exit(). | |
82d6489d DBO |
1921 | * |
1922 | * Interrupts were already disabled while acquiring | |
1923 | * the css_set_lock, so we do not need to disable it | |
1924 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1925 | */ |
82d6489d | 1926 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1927 | if (!(p->flags & PF_EXITING)) { |
1928 | struct css_set *cset = task_css_set(p); | |
1929 | ||
0de0942d TH |
1930 | if (!css_set_populated(cset)) |
1931 | css_set_update_populated(cset, true); | |
389b9c1b | 1932 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1933 | get_css_set(cset); |
1934 | } | |
82d6489d | 1935 | spin_unlock(&p->sighand->siglock); |
afeb0f9f TH |
1936 | } while_each_thread(g, p); |
1937 | read_unlock(&tasklist_lock); | |
1938 | out_unlock: | |
82d6489d | 1939 | spin_unlock_irq(&css_set_lock); |
afeb0f9f | 1940 | } |
ddbcc7e8 | 1941 | |
cc31edce PM |
1942 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1943 | { | |
2d8f243a TH |
1944 | struct cgroup_subsys *ss; |
1945 | int ssid; | |
1946 | ||
d5c419b6 TH |
1947 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1948 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1949 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1950 | INIT_LIST_HEAD(&cgrp->pidlists); |
1951 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1952 | cgrp->self.cgroup = cgrp; |
184faf32 | 1953 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1954 | |
1955 | for_each_subsys(ss, ssid) | |
1956 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1957 | |
1958 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1959 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1960 | } |
c6d57f33 | 1961 | |
3dd06ffa | 1962 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1963 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1964 | { |
3dd06ffa | 1965 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1966 | |
ddbcc7e8 | 1967 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1968 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1969 | cgrp->root = root; |
cc31edce | 1970 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1971 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1972 | |
c6d57f33 PM |
1973 | root->flags = opts->flags; |
1974 | if (opts->release_agent) | |
1975 | strcpy(root->release_agent_path, opts->release_agent); | |
1976 | if (opts->name) | |
1977 | strcpy(root->name, opts->name); | |
2260e7fc | 1978 | if (opts->cpuset_clone_children) |
3dd06ffa | 1979 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1980 | } |
1981 | ||
6e5c8307 | 1982 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1983 | { |
d427dfeb | 1984 | LIST_HEAD(tmp_links); |
3dd06ffa | 1985 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1986 | struct css_set *cset; |
d427dfeb | 1987 | int i, ret; |
2c6ab6d2 | 1988 | |
d427dfeb | 1989 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1990 | |
cf780b7d | 1991 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1992 | if (ret < 0) |
2bd59d48 | 1993 | goto out; |
d427dfeb | 1994 | root_cgrp->id = ret; |
b11cfb58 | 1995 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1996 | |
2aad2a86 TH |
1997 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1998 | GFP_KERNEL); | |
9d755d33 TH |
1999 | if (ret) |
2000 | goto out; | |
2001 | ||
d427dfeb | 2002 | /* |
f0d9a5f1 | 2003 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 2004 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
2005 | * cgroup_lock, and that's us. Later rebinding may disable |
2006 | * controllers on the default hierarchy and thus create new csets, | |
2007 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 2008 | */ |
04313591 | 2009 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 2010 | if (ret) |
9d755d33 | 2011 | goto cancel_ref; |
ddbcc7e8 | 2012 | |
985ed670 | 2013 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 2014 | if (ret) |
9d755d33 | 2015 | goto cancel_ref; |
ddbcc7e8 | 2016 | |
2bd59d48 TH |
2017 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
2018 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
2019 | root_cgrp); | |
2020 | if (IS_ERR(root->kf_root)) { | |
2021 | ret = PTR_ERR(root->kf_root); | |
2022 | goto exit_root_id; | |
2023 | } | |
2024 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 2025 | |
334c3679 | 2026 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 2027 | if (ret) |
2bd59d48 | 2028 | goto destroy_root; |
ddbcc7e8 | 2029 | |
5df36032 | 2030 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 2031 | if (ret) |
2bd59d48 | 2032 | goto destroy_root; |
ddbcc7e8 | 2033 | |
d427dfeb TH |
2034 | /* |
2035 | * There must be no failure case after here, since rebinding takes | |
2036 | * care of subsystems' refcounts, which are explicitly dropped in | |
2037 | * the failure exit path. | |
2038 | */ | |
2039 | list_add(&root->root_list, &cgroup_roots); | |
2040 | cgroup_root_count++; | |
0df6a63f | 2041 | |
d427dfeb | 2042 | /* |
3dd06ffa | 2043 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2044 | * objects. |
2045 | */ | |
82d6489d | 2046 | spin_lock_irq(&css_set_lock); |
0de0942d | 2047 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2048 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2049 | if (css_set_populated(cset)) |
2050 | cgroup_update_populated(root_cgrp, true); | |
2051 | } | |
82d6489d | 2052 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 2053 | |
d5c419b6 | 2054 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2055 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2056 | |
2bd59d48 | 2057 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2058 | ret = 0; |
2bd59d48 | 2059 | goto out; |
d427dfeb | 2060 | |
2bd59d48 TH |
2061 | destroy_root: |
2062 | kernfs_destroy_root(root->kf_root); | |
2063 | root->kf_root = NULL; | |
2064 | exit_root_id: | |
d427dfeb | 2065 | cgroup_exit_root_id(root); |
9d755d33 | 2066 | cancel_ref: |
9a1049da | 2067 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2068 | out: |
d427dfeb TH |
2069 | free_cgrp_cset_links(&tmp_links); |
2070 | return ret; | |
ddbcc7e8 PM |
2071 | } |
2072 | ||
f7e83571 | 2073 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2074 | int flags, const char *unused_dev_name, |
f7e83571 | 2075 | void *data) |
ddbcc7e8 | 2076 | { |
67e9c74b | 2077 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 2078 | struct super_block *pinned_sb = NULL; |
ed82571b | 2079 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
970317aa | 2080 | struct cgroup_subsys *ss; |
3dd06ffa | 2081 | struct cgroup_root *root; |
ddbcc7e8 | 2082 | struct cgroup_sb_opts opts; |
2bd59d48 | 2083 | struct dentry *dentry; |
8e30e2b8 | 2084 | int ret; |
970317aa | 2085 | int i; |
c6b3d5bc | 2086 | bool new_sb; |
ddbcc7e8 | 2087 | |
ed82571b SH |
2088 | get_cgroup_ns(ns); |
2089 | ||
2090 | /* Check if the caller has permission to mount. */ | |
2091 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2092 | put_cgroup_ns(ns); | |
2093 | return ERR_PTR(-EPERM); | |
2094 | } | |
2095 | ||
56fde9e0 TH |
2096 | /* |
2097 | * The first time anyone tries to mount a cgroup, enable the list | |
2098 | * linking each css_set to its tasks and fix up all existing tasks. | |
2099 | */ | |
2100 | if (!use_task_css_set_links) | |
2101 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2102 | |
67e9c74b TH |
2103 | if (is_v2) { |
2104 | if (data) { | |
2105 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
ed82571b | 2106 | put_cgroup_ns(ns); |
67e9c74b TH |
2107 | return ERR_PTR(-EINVAL); |
2108 | } | |
a7165264 | 2109 | cgrp_dfl_visible = true; |
67e9c74b TH |
2110 | root = &cgrp_dfl_root; |
2111 | cgroup_get(&root->cgrp); | |
2112 | goto out_mount; | |
2113 | } | |
2114 | ||
334c3679 | 2115 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2116 | |
2117 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2118 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2119 | if (ret) |
8e30e2b8 | 2120 | goto out_unlock; |
a015edd2 | 2121 | |
970317aa LZ |
2122 | /* |
2123 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2124 | * still be dying after the previous unmount. Let's drain the | |
2125 | * dying subsystems. We just need to ensure that the ones | |
2126 | * unmounted previously finish dying and don't care about new ones | |
2127 | * starting. Testing ref liveliness is good enough. | |
2128 | */ | |
2129 | for_each_subsys(ss, i) { | |
2130 | if (!(opts.subsys_mask & (1 << i)) || | |
2131 | ss->root == &cgrp_dfl_root) | |
2132 | continue; | |
2133 | ||
2134 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2135 | mutex_unlock(&cgroup_mutex); | |
2136 | msleep(10); | |
2137 | ret = restart_syscall(); | |
2138 | goto out_free; | |
2139 | } | |
2140 | cgroup_put(&ss->root->cgrp); | |
2141 | } | |
2142 | ||
985ed670 | 2143 | for_each_root(root) { |
2bd59d48 | 2144 | bool name_match = false; |
3126121f | 2145 | |
3dd06ffa | 2146 | if (root == &cgrp_dfl_root) |
985ed670 | 2147 | continue; |
3126121f | 2148 | |
cf5d5941 | 2149 | /* |
2bd59d48 TH |
2150 | * If we asked for a name then it must match. Also, if |
2151 | * name matches but sybsys_mask doesn't, we should fail. | |
2152 | * Remember whether name matched. | |
cf5d5941 | 2153 | */ |
2bd59d48 TH |
2154 | if (opts.name) { |
2155 | if (strcmp(opts.name, root->name)) | |
2156 | continue; | |
2157 | name_match = true; | |
2158 | } | |
ddbcc7e8 | 2159 | |
c6d57f33 | 2160 | /* |
2bd59d48 TH |
2161 | * If we asked for subsystems (or explicitly for no |
2162 | * subsystems) then they must match. | |
c6d57f33 | 2163 | */ |
2bd59d48 | 2164 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2165 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2166 | if (!name_match) |
2167 | continue; | |
2168 | ret = -EBUSY; | |
2169 | goto out_unlock; | |
2170 | } | |
873fe09e | 2171 | |
7b9a6ba5 TH |
2172 | if (root->flags ^ opts.flags) |
2173 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2174 | |
776f02fa | 2175 | /* |
3a32bd72 LZ |
2176 | * We want to reuse @root whose lifetime is governed by its |
2177 | * ->cgrp. Let's check whether @root is alive and keep it | |
2178 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2179 | * want to block it by pinning the sb so that @root doesn't | |
2180 | * get killed before mount is complete. | |
2181 | * | |
2182 | * With the sb pinned, tryget_live can reliably indicate | |
2183 | * whether @root can be reused. If it's being killed, | |
2184 | * drain it. We can use wait_queue for the wait but this | |
2185 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2186 | */ |
3a32bd72 LZ |
2187 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2188 | if (IS_ERR(pinned_sb) || | |
2189 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2190 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2191 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2192 | deactivate_super(pinned_sb); | |
776f02fa | 2193 | msleep(10); |
a015edd2 TH |
2194 | ret = restart_syscall(); |
2195 | goto out_free; | |
776f02fa | 2196 | } |
ddbcc7e8 | 2197 | |
776f02fa | 2198 | ret = 0; |
2bd59d48 | 2199 | goto out_unlock; |
ddbcc7e8 | 2200 | } |
ddbcc7e8 | 2201 | |
817929ec | 2202 | /* |
172a2c06 TH |
2203 | * No such thing, create a new one. name= matching without subsys |
2204 | * specification is allowed for already existing hierarchies but we | |
2205 | * can't create new one without subsys specification. | |
817929ec | 2206 | */ |
172a2c06 TH |
2207 | if (!opts.subsys_mask && !opts.none) { |
2208 | ret = -EINVAL; | |
2209 | goto out_unlock; | |
817929ec | 2210 | } |
817929ec | 2211 | |
726a4994 EB |
2212 | /* Hierarchies may only be created in the initial cgroup namespace. */ |
2213 | if (ns != &init_cgroup_ns) { | |
ed82571b SH |
2214 | ret = -EPERM; |
2215 | goto out_unlock; | |
2216 | } | |
2217 | ||
172a2c06 TH |
2218 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2219 | if (!root) { | |
2220 | ret = -ENOMEM; | |
2bd59d48 | 2221 | goto out_unlock; |
839ec545 | 2222 | } |
e5f6a860 | 2223 | |
172a2c06 TH |
2224 | init_cgroup_root(root, &opts); |
2225 | ||
35585573 | 2226 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2227 | if (ret) |
2228 | cgroup_free_root(root); | |
fa3ca07e | 2229 | |
8e30e2b8 | 2230 | out_unlock: |
ddbcc7e8 | 2231 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2232 | out_free: |
c6d57f33 PM |
2233 | kfree(opts.release_agent); |
2234 | kfree(opts.name); | |
03b1cde6 | 2235 | |
ed82571b SH |
2236 | if (ret) { |
2237 | put_cgroup_ns(ns); | |
8e30e2b8 | 2238 | return ERR_PTR(ret); |
ed82571b | 2239 | } |
67e9c74b | 2240 | out_mount: |
c9482a5b | 2241 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2242 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2243 | &new_sb); | |
ed82571b SH |
2244 | |
2245 | /* | |
2246 | * In non-init cgroup namespace, instead of root cgroup's | |
2247 | * dentry, we return the dentry corresponding to the | |
2248 | * cgroupns->root_cgrp. | |
2249 | */ | |
2250 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { | |
2251 | struct dentry *nsdentry; | |
2252 | struct cgroup *cgrp; | |
2253 | ||
2254 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2255 | spin_lock_irq(&css_set_lock); |
ed82571b SH |
2256 | |
2257 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2258 | ||
82d6489d | 2259 | spin_unlock_irq(&css_set_lock); |
ed82571b SH |
2260 | mutex_unlock(&cgroup_mutex); |
2261 | ||
2262 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); | |
2263 | dput(dentry); | |
2264 | dentry = nsdentry; | |
2265 | } | |
2266 | ||
c6b3d5bc | 2267 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2268 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2269 | |
2270 | /* | |
2271 | * If @pinned_sb, we're reusing an existing root and holding an | |
2272 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2273 | */ | |
2274 | if (pinned_sb) { | |
2275 | WARN_ON(new_sb); | |
2276 | deactivate_super(pinned_sb); | |
2277 | } | |
2278 | ||
ed82571b | 2279 | put_cgroup_ns(ns); |
2bd59d48 TH |
2280 | return dentry; |
2281 | } | |
2282 | ||
2283 | static void cgroup_kill_sb(struct super_block *sb) | |
2284 | { | |
2285 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2286 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2287 | |
9d755d33 TH |
2288 | /* |
2289 | * If @root doesn't have any mounts or children, start killing it. | |
2290 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2291 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2292 | * |
2293 | * And don't kill the default root. | |
9d755d33 | 2294 | */ |
3c606d35 | 2295 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2296 | root == &cgrp_dfl_root) |
9d755d33 TH |
2297 | cgroup_put(&root->cgrp); |
2298 | else | |
2299 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2300 | ||
2bd59d48 | 2301 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2302 | } |
2303 | ||
2304 | static struct file_system_type cgroup_fs_type = { | |
2305 | .name = "cgroup", | |
f7e83571 | 2306 | .mount = cgroup_mount, |
ddbcc7e8 | 2307 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2308 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 PM |
2309 | }; |
2310 | ||
67e9c74b TH |
2311 | static struct file_system_type cgroup2_fs_type = { |
2312 | .name = "cgroup2", | |
2313 | .mount = cgroup_mount, | |
2314 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2315 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b TH |
2316 | }; |
2317 | ||
a79a908f AK |
2318 | static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2319 | struct cgroup_namespace *ns) | |
2320 | { | |
2321 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
2322 | int ret; | |
2323 | ||
2324 | ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); | |
2325 | if (ret < 0 || ret >= buflen) | |
2326 | return NULL; | |
2327 | return buf; | |
2328 | } | |
2329 | ||
2330 | char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, | |
2331 | struct cgroup_namespace *ns) | |
2332 | { | |
2333 | char *ret; | |
2334 | ||
2335 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2336 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2337 | |
2338 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2339 | ||
82d6489d | 2340 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2341 | mutex_unlock(&cgroup_mutex); |
2342 | ||
2343 | return ret; | |
2344 | } | |
2345 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2346 | ||
857a2beb | 2347 | /** |
913ffdb5 | 2348 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2349 | * @task: target task |
857a2beb TH |
2350 | * @buf: the buffer to write the path into |
2351 | * @buflen: the length of the buffer | |
2352 | * | |
913ffdb5 TH |
2353 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2354 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2355 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2356 | * cgroup controller callbacks. | |
2357 | * | |
e61734c5 | 2358 | * Return value is the same as kernfs_path(). |
857a2beb | 2359 | */ |
e61734c5 | 2360 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2361 | { |
3dd06ffa | 2362 | struct cgroup_root *root; |
913ffdb5 | 2363 | struct cgroup *cgrp; |
e61734c5 TH |
2364 | int hierarchy_id = 1; |
2365 | char *path = NULL; | |
857a2beb TH |
2366 | |
2367 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2368 | spin_lock_irq(&css_set_lock); |
857a2beb | 2369 | |
913ffdb5 TH |
2370 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2371 | ||
857a2beb TH |
2372 | if (root) { |
2373 | cgrp = task_cgroup_from_root(task, root); | |
a79a908f | 2374 | path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2375 | } else { |
2376 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2377 | if (strlcpy(buf, "/", buflen) < buflen) |
2378 | path = buf; | |
857a2beb TH |
2379 | } |
2380 | ||
82d6489d | 2381 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2382 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2383 | return path; |
857a2beb | 2384 | } |
913ffdb5 | 2385 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2386 | |
b3dc094e | 2387 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2388 | struct cgroup_taskset { |
b3dc094e TH |
2389 | /* the src and dst cset list running through cset->mg_node */ |
2390 | struct list_head src_csets; | |
2391 | struct list_head dst_csets; | |
2392 | ||
1f7dd3e5 TH |
2393 | /* the subsys currently being processed */ |
2394 | int ssid; | |
2395 | ||
b3dc094e TH |
2396 | /* |
2397 | * Fields for cgroup_taskset_*() iteration. | |
2398 | * | |
2399 | * Before migration is committed, the target migration tasks are on | |
2400 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2401 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2402 | * or ->dst_csets depending on whether migration is committed. | |
2403 | * | |
2404 | * ->cur_csets and ->cur_task point to the current task position | |
2405 | * during iteration. | |
2406 | */ | |
2407 | struct list_head *csets; | |
2408 | struct css_set *cur_cset; | |
2409 | struct task_struct *cur_task; | |
2f7ee569 TH |
2410 | }; |
2411 | ||
adaae5dc TH |
2412 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2413 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2414 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2415 | .csets = &tset.src_csets, \ | |
2416 | } | |
2417 | ||
2418 | /** | |
2419 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2420 | * @task: target task | |
2421 | * @tset: target taskset | |
2422 | * | |
2423 | * Add @task, which is a migration target, to @tset. This function becomes | |
2424 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2425 | * been added as a migration source and @task->cg_list will be moved from | |
2426 | * the css_set's tasks list to mg_tasks one. | |
2427 | */ | |
2428 | static void cgroup_taskset_add(struct task_struct *task, | |
2429 | struct cgroup_taskset *tset) | |
2430 | { | |
2431 | struct css_set *cset; | |
2432 | ||
f0d9a5f1 | 2433 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2434 | |
2435 | /* @task either already exited or can't exit until the end */ | |
2436 | if (task->flags & PF_EXITING) | |
2437 | return; | |
2438 | ||
2439 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2440 | if (list_empty(&task->cg_list)) | |
2441 | return; | |
2442 | ||
2443 | cset = task_css_set(task); | |
2444 | if (!cset->mg_src_cgrp) | |
2445 | return; | |
2446 | ||
2447 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2448 | if (list_empty(&cset->mg_node)) | |
2449 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2450 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2451 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2452 | &tset->dst_csets); | |
2453 | } | |
2454 | ||
2f7ee569 TH |
2455 | /** |
2456 | * cgroup_taskset_first - reset taskset and return the first task | |
2457 | * @tset: taskset of interest | |
1f7dd3e5 | 2458 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2459 | * |
2460 | * @tset iteration is initialized and the first task is returned. | |
2461 | */ | |
1f7dd3e5 TH |
2462 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2463 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2464 | { |
b3dc094e TH |
2465 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2466 | tset->cur_task = NULL; | |
2467 | ||
1f7dd3e5 | 2468 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2469 | } |
2f7ee569 TH |
2470 | |
2471 | /** | |
2472 | * cgroup_taskset_next - iterate to the next task in taskset | |
2473 | * @tset: taskset of interest | |
1f7dd3e5 | 2474 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2475 | * |
2476 | * Return the next task in @tset. Iteration must have been initialized | |
2477 | * with cgroup_taskset_first(). | |
2478 | */ | |
1f7dd3e5 TH |
2479 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2480 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2481 | { |
b3dc094e TH |
2482 | struct css_set *cset = tset->cur_cset; |
2483 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2484 | |
b3dc094e TH |
2485 | while (&cset->mg_node != tset->csets) { |
2486 | if (!task) | |
2487 | task = list_first_entry(&cset->mg_tasks, | |
2488 | struct task_struct, cg_list); | |
2489 | else | |
2490 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2491 | |
b3dc094e TH |
2492 | if (&task->cg_list != &cset->mg_tasks) { |
2493 | tset->cur_cset = cset; | |
2494 | tset->cur_task = task; | |
1f7dd3e5 TH |
2495 | |
2496 | /* | |
2497 | * This function may be called both before and | |
2498 | * after cgroup_taskset_migrate(). The two cases | |
2499 | * can be distinguished by looking at whether @cset | |
2500 | * has its ->mg_dst_cset set. | |
2501 | */ | |
2502 | if (cset->mg_dst_cset) | |
2503 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2504 | else | |
2505 | *dst_cssp = cset->subsys[tset->ssid]; | |
2506 | ||
b3dc094e TH |
2507 | return task; |
2508 | } | |
2f7ee569 | 2509 | |
b3dc094e TH |
2510 | cset = list_next_entry(cset, mg_node); |
2511 | task = NULL; | |
2512 | } | |
2f7ee569 | 2513 | |
b3dc094e | 2514 | return NULL; |
2f7ee569 | 2515 | } |
2f7ee569 | 2516 | |
adaae5dc | 2517 | /** |
37ff9f8f | 2518 | * cgroup_taskset_migrate - migrate a taskset |
adaae5dc | 2519 | * @tset: taget taskset |
37ff9f8f | 2520 | * @root: cgroup root the migration is taking place on |
adaae5dc | 2521 | * |
37ff9f8f TH |
2522 | * Migrate tasks in @tset as setup by migration preparation functions. |
2523 | * This function fails iff one of the ->can_attach callbacks fails and | |
2524 | * guarantees that either all or none of the tasks in @tset are migrated. | |
2525 | * @tset is consumed regardless of success. | |
adaae5dc TH |
2526 | */ |
2527 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
37ff9f8f | 2528 | struct cgroup_root *root) |
adaae5dc | 2529 | { |
37ff9f8f | 2530 | struct cgroup_subsys *ss; |
adaae5dc TH |
2531 | struct task_struct *task, *tmp_task; |
2532 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2533 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2534 | |
2535 | /* methods shouldn't be called if no task is actually migrating */ | |
2536 | if (list_empty(&tset->src_csets)) | |
2537 | return 0; | |
2538 | ||
2539 | /* check that we can legitimately attach to the cgroup */ | |
37ff9f8f TH |
2540 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2541 | if (ss->can_attach) { | |
2542 | tset->ssid = ssid; | |
2543 | ret = ss->can_attach(tset); | |
adaae5dc | 2544 | if (ret) { |
37ff9f8f | 2545 | failed_ssid = ssid; |
adaae5dc TH |
2546 | goto out_cancel_attach; |
2547 | } | |
2548 | } | |
37ff9f8f | 2549 | } while_each_subsys_mask(); |
adaae5dc TH |
2550 | |
2551 | /* | |
2552 | * Now that we're guaranteed success, proceed to move all tasks to | |
2553 | * the new cgroup. There are no failure cases after here, so this | |
2554 | * is the commit point. | |
2555 | */ | |
82d6489d | 2556 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2557 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2558 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2559 | struct css_set *from_cset = task_css_set(task); | |
2560 | struct css_set *to_cset = cset->mg_dst_cset; | |
2561 | ||
2562 | get_css_set(to_cset); | |
2563 | css_set_move_task(task, from_cset, to_cset, true); | |
2564 | put_css_set_locked(from_cset); | |
2565 | } | |
adaae5dc | 2566 | } |
82d6489d | 2567 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2568 | |
2569 | /* | |
2570 | * Migration is committed, all target tasks are now on dst_csets. | |
2571 | * Nothing is sensitive to fork() after this point. Notify | |
2572 | * controllers that migration is complete. | |
2573 | */ | |
2574 | tset->csets = &tset->dst_csets; | |
2575 | ||
37ff9f8f TH |
2576 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2577 | if (ss->attach) { | |
2578 | tset->ssid = ssid; | |
2579 | ss->attach(tset); | |
1f7dd3e5 | 2580 | } |
37ff9f8f | 2581 | } while_each_subsys_mask(); |
adaae5dc TH |
2582 | |
2583 | ret = 0; | |
2584 | goto out_release_tset; | |
2585 | ||
2586 | out_cancel_attach: | |
37ff9f8f TH |
2587 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2588 | if (ssid == failed_ssid) | |
adaae5dc | 2589 | break; |
37ff9f8f TH |
2590 | if (ss->cancel_attach) { |
2591 | tset->ssid = ssid; | |
2592 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2593 | } |
37ff9f8f | 2594 | } while_each_subsys_mask(); |
adaae5dc | 2595 | out_release_tset: |
82d6489d | 2596 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2597 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2598 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2599 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2600 | list_del_init(&cset->mg_node); | |
2601 | } | |
82d6489d | 2602 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2603 | return ret; |
2604 | } | |
2605 | ||
6c694c88 TH |
2606 | /** |
2607 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2608 | * @dst_cgrp: destination cgroup to test | |
2609 | * | |
2610 | * On the default hierarchy, except for the root, subtree_control must be | |
2611 | * zero for migration destination cgroups with tasks so that child cgroups | |
2612 | * don't compete against tasks. | |
2613 | */ | |
2614 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) | |
2615 | { | |
2616 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2617 | !dst_cgrp->subtree_control; | |
2618 | } | |
2619 | ||
a043e3b2 | 2620 | /** |
1958d2d5 TH |
2621 | * cgroup_migrate_finish - cleanup after attach |
2622 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2623 | * |
1958d2d5 TH |
2624 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2625 | * those functions for details. | |
74a1166d | 2626 | */ |
1958d2d5 | 2627 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2628 | { |
1958d2d5 | 2629 | struct css_set *cset, *tmp_cset; |
74a1166d | 2630 | |
1958d2d5 TH |
2631 | lockdep_assert_held(&cgroup_mutex); |
2632 | ||
82d6489d | 2633 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2634 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2635 | cset->mg_src_cgrp = NULL; | |
e4857982 | 2636 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2637 | cset->mg_dst_cset = NULL; |
2638 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2639 | put_css_set_locked(cset); |
1958d2d5 | 2640 | } |
82d6489d | 2641 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2642 | } |
2643 | ||
2644 | /** | |
2645 | * cgroup_migrate_add_src - add a migration source css_set | |
2646 | * @src_cset: the source css_set to add | |
2647 | * @dst_cgrp: the destination cgroup | |
2648 | * @preloaded_csets: list of preloaded css_sets | |
2649 | * | |
2650 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2651 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2652 | * up by cgroup_migrate_finish(). | |
2653 | * | |
1ed13287 TH |
2654 | * This function may be called without holding cgroup_threadgroup_rwsem |
2655 | * even if the target is a process. Threads may be created and destroyed | |
2656 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2657 | * into play and the preloaded css_sets are guaranteed to cover all | |
2658 | * migrations. | |
1958d2d5 TH |
2659 | */ |
2660 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2661 | struct cgroup *dst_cgrp, | |
2662 | struct list_head *preloaded_csets) | |
2663 | { | |
2664 | struct cgroup *src_cgrp; | |
2665 | ||
2666 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2667 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2668 | |
2b021cbf TH |
2669 | /* |
2670 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2671 | * and doesn't contain any migratable tasks. Ignore it early so | |
2672 | * that the rest of migration path doesn't get confused by it. | |
2673 | */ | |
2674 | if (src_cset->dead) | |
2675 | return; | |
2676 | ||
1958d2d5 TH |
2677 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2678 | ||
1958d2d5 TH |
2679 | if (!list_empty(&src_cset->mg_preload_node)) |
2680 | return; | |
2681 | ||
2682 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2683 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2684 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2685 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2686 | ||
2687 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2688 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 TH |
2689 | get_css_set(src_cset); |
2690 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2691 | } | |
2692 | ||
2693 | /** | |
2694 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1958d2d5 TH |
2695 | * @preloaded_csets: list of preloaded source css_sets |
2696 | * | |
e4857982 TH |
2697 | * Tasks are about to be moved and all the source css_sets have been |
2698 | * preloaded to @preloaded_csets. This function looks up and pins all | |
2699 | * destination css_sets, links each to its source, and append them to | |
2700 | * @preloaded_csets. | |
1958d2d5 TH |
2701 | * |
2702 | * This function must be called after cgroup_migrate_add_src() has been | |
2703 | * called on each migration source css_set. After migration is performed | |
2704 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2705 | * @preloaded_csets. | |
2706 | */ | |
e4857982 | 2707 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
1958d2d5 TH |
2708 | { |
2709 | LIST_HEAD(csets); | |
f817de98 | 2710 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2711 | |
2712 | lockdep_assert_held(&cgroup_mutex); | |
2713 | ||
2714 | /* look up the dst cset for each src cset and link it to src */ | |
f817de98 | 2715 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2716 | struct css_set *dst_cset; |
2717 | ||
e4857982 | 2718 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2719 | if (!dst_cset) |
2720 | goto err; | |
2721 | ||
2722 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2723 | |
2724 | /* | |
2725 | * If src cset equals dst, it's noop. Drop the src. | |
2726 | * cgroup_migrate() will skip the cset too. Note that we | |
2727 | * can't handle src == dst as some nodes are used by both. | |
2728 | */ | |
2729 | if (src_cset == dst_cset) { | |
2730 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2731 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2732 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2733 | put_css_set(src_cset); |
2734 | put_css_set(dst_cset); | |
f817de98 TH |
2735 | continue; |
2736 | } | |
2737 | ||
1958d2d5 TH |
2738 | src_cset->mg_dst_cset = dst_cset; |
2739 | ||
2740 | if (list_empty(&dst_cset->mg_preload_node)) | |
2741 | list_add(&dst_cset->mg_preload_node, &csets); | |
2742 | else | |
a25eb52e | 2743 | put_css_set(dst_cset); |
1958d2d5 TH |
2744 | } |
2745 | ||
f817de98 | 2746 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2747 | return 0; |
2748 | err: | |
2749 | cgroup_migrate_finish(&csets); | |
2750 | return -ENOMEM; | |
2751 | } | |
2752 | ||
2753 | /** | |
2754 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2755 | * @leader: the leader of the process or the task to migrate |
2756 | * @threadgroup: whether @leader points to the whole process or a single task | |
37ff9f8f | 2757 | * @root: cgroup root migration is taking place on |
1958d2d5 | 2758 | * |
37ff9f8f TH |
2759 | * Migrate a process or task denoted by @leader. If migrating a process, |
2760 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2761 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2762 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2763 | * function and following up with cgroup_migrate_finish(). | |
2764 | * | |
2765 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2766 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2767 | * failure, when migrating multiple targets, the success or failure can be | |
2768 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2769 | * actually starting migrating. | |
2770 | */ | |
9af2ec45 | 2771 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
37ff9f8f | 2772 | struct cgroup_root *root) |
74a1166d | 2773 | { |
adaae5dc TH |
2774 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2775 | struct task_struct *task; | |
74a1166d | 2776 | |
fb5d2b4c MSB |
2777 | /* |
2778 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2779 | * already PF_EXITING could be freed from underneath us unless we | |
2780 | * take an rcu_read_lock. | |
2781 | */ | |
82d6489d | 2782 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2783 | rcu_read_lock(); |
9db8de37 | 2784 | task = leader; |
74a1166d | 2785 | do { |
adaae5dc | 2786 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2787 | if (!threadgroup) |
2788 | break; | |
9db8de37 | 2789 | } while_each_thread(leader, task); |
fb5d2b4c | 2790 | rcu_read_unlock(); |
82d6489d | 2791 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2792 | |
37ff9f8f | 2793 | return cgroup_taskset_migrate(&tset, root); |
74a1166d BB |
2794 | } |
2795 | ||
1958d2d5 TH |
2796 | /** |
2797 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2798 | * @dst_cgrp: the cgroup to attach to | |
2799 | * @leader: the task or the leader of the threadgroup to be attached | |
2800 | * @threadgroup: attach the whole threadgroup? | |
2801 | * | |
1ed13287 | 2802 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2803 | */ |
2804 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2805 | struct task_struct *leader, bool threadgroup) | |
2806 | { | |
2807 | LIST_HEAD(preloaded_csets); | |
2808 | struct task_struct *task; | |
2809 | int ret; | |
2810 | ||
6c694c88 TH |
2811 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2812 | return -EBUSY; | |
2813 | ||
1958d2d5 | 2814 | /* look up all src csets */ |
82d6489d | 2815 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2816 | rcu_read_lock(); |
2817 | task = leader; | |
2818 | do { | |
2819 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2820 | &preloaded_csets); | |
2821 | if (!threadgroup) | |
2822 | break; | |
2823 | } while_each_thread(leader, task); | |
2824 | rcu_read_unlock(); | |
82d6489d | 2825 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2826 | |
2827 | /* prepare dst csets and commit */ | |
e4857982 | 2828 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
1958d2d5 | 2829 | if (!ret) |
37ff9f8f | 2830 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
1958d2d5 TH |
2831 | |
2832 | cgroup_migrate_finish(&preloaded_csets); | |
2833 | return ret; | |
74a1166d BB |
2834 | } |
2835 | ||
187fe840 TH |
2836 | static int cgroup_procs_write_permission(struct task_struct *task, |
2837 | struct cgroup *dst_cgrp, | |
2838 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2839 | { |
2840 | const struct cred *cred = current_cred(); | |
2841 | const struct cred *tcred = get_task_cred(task); | |
2842 | int ret = 0; | |
2843 | ||
2844 | /* | |
2845 | * even if we're attaching all tasks in the thread group, we only | |
2846 | * need to check permissions on one of them. | |
2847 | */ | |
2848 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2849 | !uid_eq(cred->euid, tcred->uid) && | |
2850 | !uid_eq(cred->euid, tcred->suid)) | |
2851 | ret = -EACCES; | |
2852 | ||
187fe840 TH |
2853 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2854 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2855 | struct cgroup *cgrp; | |
2856 | struct inode *inode; | |
2857 | ||
82d6489d | 2858 | spin_lock_irq(&css_set_lock); |
187fe840 | 2859 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
82d6489d | 2860 | spin_unlock_irq(&css_set_lock); |
187fe840 TH |
2861 | |
2862 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2863 | cgrp = cgroup_parent(cgrp); | |
2864 | ||
2865 | ret = -ENOMEM; | |
6f60eade | 2866 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2867 | if (inode) { |
2868 | ret = inode_permission(inode, MAY_WRITE); | |
2869 | iput(inode); | |
2870 | } | |
2871 | } | |
2872 | ||
dedf22e9 TH |
2873 | put_cred(tcred); |
2874 | return ret; | |
2875 | } | |
2876 | ||
74a1166d BB |
2877 | /* |
2878 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2879 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2880 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2881 | */ |
acbef755 TH |
2882 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2883 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2884 | { |
bbcb81d0 | 2885 | struct task_struct *tsk; |
5cf1cacb | 2886 | struct cgroup_subsys *ss; |
e76ecaee | 2887 | struct cgroup *cgrp; |
acbef755 | 2888 | pid_t pid; |
5cf1cacb | 2889 | int ssid, ret; |
bbcb81d0 | 2890 | |
acbef755 TH |
2891 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2892 | return -EINVAL; | |
2893 | ||
945ba199 | 2894 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2895 | if (!cgrp) |
74a1166d BB |
2896 | return -ENODEV; |
2897 | ||
3014dde7 | 2898 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2899 | rcu_read_lock(); |
bbcb81d0 | 2900 | if (pid) { |
73507f33 | 2901 | tsk = find_task_by_vpid(pid); |
74a1166d | 2902 | if (!tsk) { |
dd4b0a46 | 2903 | ret = -ESRCH; |
3014dde7 | 2904 | goto out_unlock_rcu; |
bbcb81d0 | 2905 | } |
dedf22e9 | 2906 | } else { |
b78949eb | 2907 | tsk = current; |
dedf22e9 | 2908 | } |
cd3d0952 TH |
2909 | |
2910 | if (threadgroup) | |
b78949eb | 2911 | tsk = tsk->group_leader; |
c4c27fbd MG |
2912 | |
2913 | /* | |
14a40ffc | 2914 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2915 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2916 | * with no rt_runtime allocated. Just say no. | |
2917 | */ | |
14a40ffc | 2918 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2919 | ret = -EINVAL; |
3014dde7 | 2920 | goto out_unlock_rcu; |
c4c27fbd MG |
2921 | } |
2922 | ||
b78949eb MSB |
2923 | get_task_struct(tsk); |
2924 | rcu_read_unlock(); | |
2925 | ||
187fe840 | 2926 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2927 | if (!ret) |
2928 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2929 | |
f9f9e7b7 | 2930 | put_task_struct(tsk); |
3014dde7 TH |
2931 | goto out_unlock_threadgroup; |
2932 | ||
2933 | out_unlock_rcu: | |
2934 | rcu_read_unlock(); | |
2935 | out_unlock_threadgroup: | |
2936 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2937 | for_each_subsys(ss, ssid) |
2938 | if (ss->post_attach) | |
2939 | ss->post_attach(); | |
e76ecaee | 2940 | cgroup_kn_unlock(of->kn); |
acbef755 | 2941 | return ret ?: nbytes; |
bbcb81d0 PM |
2942 | } |
2943 | ||
7ae1bad9 TH |
2944 | /** |
2945 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2946 | * @from: attach to all cgroups of a given task | |
2947 | * @tsk: the task to be attached | |
2948 | */ | |
2949 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2950 | { | |
3dd06ffa | 2951 | struct cgroup_root *root; |
7ae1bad9 TH |
2952 | int retval = 0; |
2953 | ||
47cfcd09 | 2954 | mutex_lock(&cgroup_mutex); |
eedd0f4c | 2955 | percpu_down_write(&cgroup_threadgroup_rwsem); |
985ed670 | 2956 | for_each_root(root) { |
96d365e0 TH |
2957 | struct cgroup *from_cgrp; |
2958 | ||
3dd06ffa | 2959 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2960 | continue; |
2961 | ||
82d6489d | 2962 | spin_lock_irq(&css_set_lock); |
96d365e0 | 2963 | from_cgrp = task_cgroup_from_root(from, root); |
82d6489d | 2964 | spin_unlock_irq(&css_set_lock); |
7ae1bad9 | 2965 | |
6f4b7e63 | 2966 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2967 | if (retval) |
2968 | break; | |
2969 | } | |
eedd0f4c | 2970 | percpu_up_write(&cgroup_threadgroup_rwsem); |
47cfcd09 | 2971 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2972 | |
2973 | return retval; | |
2974 | } | |
2975 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2976 | ||
acbef755 TH |
2977 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2978 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2979 | { |
acbef755 | 2980 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2981 | } |
2982 | ||
acbef755 TH |
2983 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2984 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2985 | { |
acbef755 | 2986 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2987 | } |
2988 | ||
451af504 TH |
2989 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2990 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2991 | { |
e76ecaee | 2992 | struct cgroup *cgrp; |
5f469907 | 2993 | |
e76ecaee | 2994 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2995 | |
945ba199 | 2996 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2997 | if (!cgrp) |
e788e066 | 2998 | return -ENODEV; |
69e943b7 | 2999 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
3000 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
3001 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 3002 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 3003 | cgroup_kn_unlock(of->kn); |
451af504 | 3004 | return nbytes; |
e788e066 PM |
3005 | } |
3006 | ||
2da8ca82 | 3007 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 3008 | { |
2da8ca82 | 3009 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 3010 | |
46cfeb04 | 3011 | spin_lock(&release_agent_path_lock); |
e788e066 | 3012 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 3013 | spin_unlock(&release_agent_path_lock); |
e788e066 | 3014 | seq_putc(seq, '\n'); |
e788e066 PM |
3015 | return 0; |
3016 | } | |
3017 | ||
2da8ca82 | 3018 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 3019 | { |
c1d5d42e | 3020 | seq_puts(seq, "0\n"); |
e788e066 PM |
3021 | return 0; |
3022 | } | |
3023 | ||
6e5c8307 | 3024 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 3025 | { |
f8f22e53 TH |
3026 | struct cgroup_subsys *ss; |
3027 | bool printed = false; | |
3028 | int ssid; | |
a742c59d | 3029 | |
b4e0eeaf | 3030 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
3031 | if (printed) |
3032 | seq_putc(seq, ' '); | |
3033 | seq_printf(seq, "%s", ss->name); | |
3034 | printed = true; | |
b4e0eeaf | 3035 | } while_each_subsys_mask(); |
f8f22e53 TH |
3036 | if (printed) |
3037 | seq_putc(seq, '\n'); | |
355e0c48 PM |
3038 | } |
3039 | ||
f8f22e53 TH |
3040 | /* show controllers which are enabled from the parent */ |
3041 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3042 | { |
f8f22e53 TH |
3043 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3044 | ||
5531dc91 | 3045 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 3046 | return 0; |
ddbcc7e8 PM |
3047 | } |
3048 | ||
f8f22e53 TH |
3049 | /* show controllers which are enabled for a given cgroup's children */ |
3050 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3051 | { |
f8f22e53 TH |
3052 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3053 | ||
667c2491 | 3054 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
3055 | return 0; |
3056 | } | |
3057 | ||
3058 | /** | |
3059 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
3060 | * @cgrp: root of the subtree to update csses for | |
3061 | * | |
54962604 TH |
3062 | * @cgrp's control masks have changed and its subtree's css associations |
3063 | * need to be updated accordingly. This function looks up all css_sets | |
3064 | * which are attached to the subtree, creates the matching updated css_sets | |
3065 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
3066 | */ |
3067 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
3068 | { | |
3069 | LIST_HEAD(preloaded_csets); | |
10265075 | 3070 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
3071 | struct cgroup_subsys_state *d_css; |
3072 | struct cgroup *dsct; | |
f8f22e53 TH |
3073 | struct css_set *src_cset; |
3074 | int ret; | |
3075 | ||
f8f22e53 TH |
3076 | lockdep_assert_held(&cgroup_mutex); |
3077 | ||
3014dde7 TH |
3078 | percpu_down_write(&cgroup_threadgroup_rwsem); |
3079 | ||
f8f22e53 | 3080 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 3081 | spin_lock_irq(&css_set_lock); |
54962604 | 3082 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
3083 | struct cgrp_cset_link *link; |
3084 | ||
54962604 | 3085 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 3086 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
3087 | &preloaded_csets); |
3088 | } | |
82d6489d | 3089 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
3090 | |
3091 | /* NULL dst indicates self on default hierarchy */ | |
e4857982 | 3092 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
f8f22e53 TH |
3093 | if (ret) |
3094 | goto out_finish; | |
3095 | ||
82d6489d | 3096 | spin_lock_irq(&css_set_lock); |
f8f22e53 | 3097 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 3098 | struct task_struct *task, *ntask; |
f8f22e53 TH |
3099 | |
3100 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
3101 | if (!src_cset->mg_src_cgrp) | |
3102 | break; | |
3103 | ||
10265075 TH |
3104 | /* all tasks in src_csets need to be migrated */ |
3105 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
3106 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 3107 | } |
82d6489d | 3108 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 3109 | |
37ff9f8f | 3110 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
f8f22e53 TH |
3111 | out_finish: |
3112 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 3113 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
3114 | return ret; |
3115 | } | |
3116 | ||
1b9b96a1 | 3117 | /** |
945ba199 | 3118 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 3119 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
3120 | * |
3121 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
3122 | * controller while the previous css is still around. This function grabs |
3123 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 3124 | */ |
945ba199 TH |
3125 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
3126 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
3127 | { |
3128 | struct cgroup *dsct; | |
ce3f1d9d | 3129 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
3130 | struct cgroup_subsys *ss; |
3131 | int ssid; | |
3132 | ||
945ba199 TH |
3133 | restart: |
3134 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 3135 | |
ce3f1d9d | 3136 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
3137 | for_each_subsys(ss, ssid) { |
3138 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3139 | DEFINE_WAIT(wait); | |
3140 | ||
ce3f1d9d | 3141 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
3142 | continue; |
3143 | ||
3144 | cgroup_get(dsct); | |
3145 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
3146 | TASK_UNINTERRUPTIBLE); | |
3147 | ||
3148 | mutex_unlock(&cgroup_mutex); | |
3149 | schedule(); | |
3150 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
3151 | |
3152 | cgroup_put(dsct); | |
945ba199 | 3153 | goto restart; |
1b9b96a1 TH |
3154 | } |
3155 | } | |
1b9b96a1 TH |
3156 | } |
3157 | ||
15a27c36 TH |
3158 | /** |
3159 | * cgroup_save_control - save control masks of a subtree | |
3160 | * @cgrp: root of the target subtree | |
3161 | * | |
3162 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
3163 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3164 | */ | |
3165 | static void cgroup_save_control(struct cgroup *cgrp) | |
3166 | { | |
3167 | struct cgroup *dsct; | |
3168 | struct cgroup_subsys_state *d_css; | |
3169 | ||
3170 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3171 | dsct->old_subtree_control = dsct->subtree_control; | |
3172 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
3173 | } | |
3174 | } | |
3175 | ||
3176 | /** | |
3177 | * cgroup_propagate_control - refresh control masks of a subtree | |
3178 | * @cgrp: root of the target subtree | |
3179 | * | |
3180 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3181 | * ->subtree_control and propagate controller availability through the | |
3182 | * subtree so that descendants don't have unavailable controllers enabled. | |
3183 | */ | |
3184 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3185 | { | |
3186 | struct cgroup *dsct; | |
3187 | struct cgroup_subsys_state *d_css; | |
3188 | ||
3189 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3190 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3191 | dsct->subtree_ss_mask = |
3192 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3193 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3194 | } |
3195 | } | |
3196 | ||
3197 | /** | |
3198 | * cgroup_restore_control - restore control masks of a subtree | |
3199 | * @cgrp: root of the target subtree | |
3200 | * | |
3201 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3202 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3203 | */ | |
3204 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3205 | { | |
3206 | struct cgroup *dsct; | |
3207 | struct cgroup_subsys_state *d_css; | |
3208 | ||
3209 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3210 | dsct->subtree_control = dsct->old_subtree_control; | |
3211 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3212 | } | |
3213 | } | |
3214 | ||
f6d635ad TH |
3215 | static bool css_visible(struct cgroup_subsys_state *css) |
3216 | { | |
3217 | struct cgroup_subsys *ss = css->ss; | |
3218 | struct cgroup *cgrp = css->cgroup; | |
3219 | ||
3220 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3221 | return true; | |
3222 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3223 | return false; | |
3224 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3225 | } | |
3226 | ||
bdb53bd7 TH |
3227 | /** |
3228 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3229 | * @cgrp: root of the target subtree |
bdb53bd7 | 3230 | * |
ce3f1d9d | 3231 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3232 | * visible. A css is created invisible if it's being implicitly enabled |
3233 | * through dependency. An invisible css is made visible when the userland | |
3234 | * explicitly enables it. | |
3235 | * | |
3236 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3237 | * been processed already aren't cleaned up. The caller is responsible for | |
3238 | * cleaning up with cgroup_apply_control_disble(). | |
3239 | */ | |
3240 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3241 | { | |
3242 | struct cgroup *dsct; | |
ce3f1d9d | 3243 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3244 | struct cgroup_subsys *ss; |
3245 | int ssid, ret; | |
3246 | ||
ce3f1d9d | 3247 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3248 | for_each_subsys(ss, ssid) { |
3249 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3250 | ||
945ba199 TH |
3251 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3252 | ||
bdb53bd7 TH |
3253 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3254 | continue; | |
3255 | ||
3256 | if (!css) { | |
3257 | css = css_create(dsct, ss); | |
3258 | if (IS_ERR(css)) | |
3259 | return PTR_ERR(css); | |
3260 | } | |
3261 | ||
f6d635ad | 3262 | if (css_visible(css)) { |
334c3679 | 3263 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3264 | if (ret) |
3265 | return ret; | |
3266 | } | |
3267 | } | |
3268 | } | |
3269 | ||
3270 | return 0; | |
3271 | } | |
3272 | ||
12b3bb6a TH |
3273 | /** |
3274 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3275 | * @cgrp: root of the target subtree |
12b3bb6a | 3276 | * |
ce3f1d9d | 3277 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3278 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3279 | * | |
3280 | * A css is hidden when the userland requests it to be disabled while other | |
3281 | * subsystems are still depending on it. The css must not actively control | |
3282 | * resources and be in the vanilla state if it's made visible again later. | |
3283 | * Controllers which may be depended upon should provide ->css_reset() for | |
3284 | * this purpose. | |
3285 | */ | |
3286 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3287 | { | |
3288 | struct cgroup *dsct; | |
ce3f1d9d | 3289 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3290 | struct cgroup_subsys *ss; |
3291 | int ssid; | |
3292 | ||
ce3f1d9d | 3293 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3294 | for_each_subsys(ss, ssid) { |
3295 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3296 | ||
945ba199 TH |
3297 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3298 | ||
12b3bb6a TH |
3299 | if (!css) |
3300 | continue; | |
3301 | ||
334c3679 TH |
3302 | if (css->parent && |
3303 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3304 | kill_css(css); |
f6d635ad | 3305 | } else if (!css_visible(css)) { |
334c3679 | 3306 | css_clear_dir(css); |
12b3bb6a TH |
3307 | if (ss->css_reset) |
3308 | ss->css_reset(css); | |
3309 | } | |
3310 | } | |
3311 | } | |
3312 | } | |
3313 | ||
f7b2814b TH |
3314 | /** |
3315 | * cgroup_apply_control - apply control mask updates to the subtree | |
3316 | * @cgrp: root of the target subtree | |
3317 | * | |
3318 | * subsystems can be enabled and disabled in a subtree using the following | |
3319 | * steps. | |
3320 | * | |
3321 | * 1. Call cgroup_save_control() to stash the current state. | |
3322 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3323 | * 3. Call cgroup_apply_control() to apply the changes. | |
3324 | * 4. Optionally perform other related operations. | |
3325 | * 5. Call cgroup_finalize_control() to finish up. | |
3326 | * | |
3327 | * This function implements step 3 and propagates the mask changes | |
3328 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3329 | * process migrations. | |
3330 | */ | |
3331 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3332 | { | |
3333 | int ret; | |
3334 | ||
3335 | cgroup_propagate_control(cgrp); | |
3336 | ||
3337 | ret = cgroup_apply_control_enable(cgrp); | |
3338 | if (ret) | |
3339 | return ret; | |
3340 | ||
3341 | /* | |
3342 | * At this point, cgroup_e_css() results reflect the new csses | |
3343 | * making the following cgroup_update_dfl_csses() properly update | |
3344 | * css associations of all tasks in the subtree. | |
3345 | */ | |
3346 | ret = cgroup_update_dfl_csses(cgrp); | |
3347 | if (ret) | |
3348 | return ret; | |
3349 | ||
3350 | return 0; | |
3351 | } | |
3352 | ||
3353 | /** | |
3354 | * cgroup_finalize_control - finalize control mask update | |
3355 | * @cgrp: root of the target subtree | |
3356 | * @ret: the result of the update | |
3357 | * | |
3358 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3359 | */ | |
3360 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3361 | { | |
3362 | if (ret) { | |
3363 | cgroup_restore_control(cgrp); | |
3364 | cgroup_propagate_control(cgrp); | |
3365 | } | |
3366 | ||
3367 | cgroup_apply_control_disable(cgrp); | |
3368 | } | |
3369 | ||
f8f22e53 | 3370 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3371 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3372 | char *buf, size_t nbytes, | |
3373 | loff_t off) | |
f8f22e53 | 3374 | { |
6e5c8307 | 3375 | u16 enable = 0, disable = 0; |
a9746d8d | 3376 | struct cgroup *cgrp, *child; |
f8f22e53 | 3377 | struct cgroup_subsys *ss; |
451af504 | 3378 | char *tok; |
f8f22e53 TH |
3379 | int ssid, ret; |
3380 | ||
3381 | /* | |
d37167ab TH |
3382 | * Parse input - space separated list of subsystem names prefixed |
3383 | * with either + or -. | |
f8f22e53 | 3384 | */ |
451af504 TH |
3385 | buf = strstrip(buf); |
3386 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3387 | if (tok[0] == '\0') |
3388 | continue; | |
a7165264 | 3389 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3390 | if (!cgroup_ssid_enabled(ssid) || |
3391 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3392 | continue; |
3393 | ||
3394 | if (*tok == '+') { | |
7d331fa9 TH |
3395 | enable |= 1 << ssid; |
3396 | disable &= ~(1 << ssid); | |
f8f22e53 | 3397 | } else if (*tok == '-') { |
7d331fa9 TH |
3398 | disable |= 1 << ssid; |
3399 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3400 | } else { |
3401 | return -EINVAL; | |
3402 | } | |
3403 | break; | |
b4e0eeaf | 3404 | } while_each_subsys_mask(); |
f8f22e53 TH |
3405 | if (ssid == CGROUP_SUBSYS_COUNT) |
3406 | return -EINVAL; | |
3407 | } | |
3408 | ||
945ba199 | 3409 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3410 | if (!cgrp) |
3411 | return -ENODEV; | |
f8f22e53 TH |
3412 | |
3413 | for_each_subsys(ss, ssid) { | |
3414 | if (enable & (1 << ssid)) { | |
667c2491 | 3415 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3416 | enable &= ~(1 << ssid); |
3417 | continue; | |
3418 | } | |
3419 | ||
5531dc91 | 3420 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3421 | ret = -ENOENT; |
3422 | goto out_unlock; | |
3423 | } | |
f8f22e53 | 3424 | } else if (disable & (1 << ssid)) { |
667c2491 | 3425 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3426 | disable &= ~(1 << ssid); |
3427 | continue; | |
3428 | } | |
3429 | ||
3430 | /* a child has it enabled? */ | |
3431 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3432 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3433 | ret = -EBUSY; |
ddab2b6e | 3434 | goto out_unlock; |
f8f22e53 TH |
3435 | } |
3436 | } | |
3437 | } | |
3438 | } | |
3439 | ||
3440 | if (!enable && !disable) { | |
3441 | ret = 0; | |
ddab2b6e | 3442 | goto out_unlock; |
f8f22e53 TH |
3443 | } |
3444 | ||
3445 | /* | |
667c2491 | 3446 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3447 | * with tasks so that child cgroups don't compete against tasks. |
3448 | */ | |
9157056d TH |
3449 | if (enable && cgroup_parent(cgrp)) { |
3450 | struct cgrp_cset_link *link; | |
3451 | ||
3452 | /* | |
3453 | * Because namespaces pin csets too, @cgrp->cset_links | |
3454 | * might not be empty even when @cgrp is empty. Walk and | |
3455 | * verify each cset. | |
3456 | */ | |
3457 | spin_lock_irq(&css_set_lock); | |
3458 | ||
3459 | ret = 0; | |
3460 | list_for_each_entry(link, &cgrp->cset_links, cset_link) { | |
3461 | if (css_set_populated(link->cset)) { | |
3462 | ret = -EBUSY; | |
3463 | break; | |
3464 | } | |
3465 | } | |
3466 | ||
3467 | spin_unlock_irq(&css_set_lock); | |
3468 | ||
3469 | if (ret) | |
3470 | goto out_unlock; | |
f8f22e53 TH |
3471 | } |
3472 | ||
15a27c36 TH |
3473 | /* save and update control masks and prepare csses */ |
3474 | cgroup_save_control(cgrp); | |
f63070d3 | 3475 | |
15a27c36 TH |
3476 | cgrp->subtree_control |= enable; |
3477 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3478 | |
f7b2814b | 3479 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3480 | |
f7b2814b | 3481 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3482 | |
3483 | kernfs_activate(cgrp->kn); | |
3484 | ret = 0; | |
3485 | out_unlock: | |
a9746d8d | 3486 | cgroup_kn_unlock(of->kn); |
451af504 | 3487 | return ret ?: nbytes; |
f8f22e53 TH |
3488 | } |
3489 | ||
4a07c222 | 3490 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3491 | { |
4a07c222 | 3492 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3493 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3494 | return 0; |
3495 | } | |
3496 | ||
2bd59d48 TH |
3497 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3498 | size_t nbytes, loff_t off) | |
355e0c48 | 3499 | { |
2bd59d48 TH |
3500 | struct cgroup *cgrp = of->kn->parent->priv; |
3501 | struct cftype *cft = of->kn->priv; | |
3502 | struct cgroup_subsys_state *css; | |
a742c59d | 3503 | int ret; |
355e0c48 | 3504 | |
b4168640 TH |
3505 | if (cft->write) |
3506 | return cft->write(of, buf, nbytes, off); | |
3507 | ||
2bd59d48 TH |
3508 | /* |
3509 | * kernfs guarantees that a file isn't deleted with operations in | |
3510 | * flight, which means that the matching css is and stays alive and | |
3511 | * doesn't need to be pinned. The RCU locking is not necessary | |
3512 | * either. It's just for the convenience of using cgroup_css(). | |
3513 | */ | |
3514 | rcu_read_lock(); | |
3515 | css = cgroup_css(cgrp, cft->ss); | |
3516 | rcu_read_unlock(); | |
a742c59d | 3517 | |
451af504 | 3518 | if (cft->write_u64) { |
a742c59d TH |
3519 | unsigned long long v; |
3520 | ret = kstrtoull(buf, 0, &v); | |
3521 | if (!ret) | |
3522 | ret = cft->write_u64(css, cft, v); | |
3523 | } else if (cft->write_s64) { | |
3524 | long long v; | |
3525 | ret = kstrtoll(buf, 0, &v); | |
3526 | if (!ret) | |
3527 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3528 | } else { |
a742c59d | 3529 | ret = -EINVAL; |
e73d2c61 | 3530 | } |
2bd59d48 | 3531 | |
a742c59d | 3532 | return ret ?: nbytes; |
355e0c48 PM |
3533 | } |
3534 | ||
6612f05b | 3535 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3536 | { |
2bd59d48 | 3537 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3538 | } |
3539 | ||
6612f05b | 3540 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3541 | { |
2bd59d48 | 3542 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3543 | } |
3544 | ||
6612f05b | 3545 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3546 | { |
2bd59d48 | 3547 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3548 | } |
3549 | ||
91796569 | 3550 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3551 | { |
7da11279 TH |
3552 | struct cftype *cft = seq_cft(m); |
3553 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3554 | |
2da8ca82 TH |
3555 | if (cft->seq_show) |
3556 | return cft->seq_show(m, arg); | |
e73d2c61 | 3557 | |
f4c753b7 | 3558 | if (cft->read_u64) |
896f5199 TH |
3559 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3560 | else if (cft->read_s64) | |
3561 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3562 | else | |
3563 | return -EINVAL; | |
3564 | return 0; | |
91796569 PM |
3565 | } |
3566 | ||
2bd59d48 TH |
3567 | static struct kernfs_ops cgroup_kf_single_ops = { |
3568 | .atomic_write_len = PAGE_SIZE, | |
3569 | .write = cgroup_file_write, | |
3570 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3571 | }; |
3572 | ||
2bd59d48 TH |
3573 | static struct kernfs_ops cgroup_kf_ops = { |
3574 | .atomic_write_len = PAGE_SIZE, | |
3575 | .write = cgroup_file_write, | |
3576 | .seq_start = cgroup_seqfile_start, | |
3577 | .seq_next = cgroup_seqfile_next, | |
3578 | .seq_stop = cgroup_seqfile_stop, | |
3579 | .seq_show = cgroup_seqfile_show, | |
3580 | }; | |
ddbcc7e8 PM |
3581 | |
3582 | /* | |
3583 | * cgroup_rename - Only allow simple rename of directories in place. | |
3584 | */ | |
2bd59d48 TH |
3585 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3586 | const char *new_name_str) | |
ddbcc7e8 | 3587 | { |
2bd59d48 | 3588 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3589 | int ret; |
65dff759 | 3590 | |
2bd59d48 | 3591 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3592 | return -ENOTDIR; |
2bd59d48 | 3593 | if (kn->parent != new_parent) |
ddbcc7e8 | 3594 | return -EIO; |
65dff759 | 3595 | |
6db8e85c TH |
3596 | /* |
3597 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3598 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3599 | */ |
aa6ec29b | 3600 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3601 | return -EPERM; |
099fca32 | 3602 | |
e1b2dc17 | 3603 | /* |
8353da1f | 3604 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3605 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3606 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3607 | */ |
3608 | kernfs_break_active_protection(new_parent); | |
3609 | kernfs_break_active_protection(kn); | |
099fca32 | 3610 | |
2bd59d48 | 3611 | mutex_lock(&cgroup_mutex); |
099fca32 | 3612 | |
2bd59d48 | 3613 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3614 | |
2bd59d48 | 3615 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3616 | |
3617 | kernfs_unbreak_active_protection(kn); | |
3618 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3619 | return ret; |
099fca32 LZ |
3620 | } |
3621 | ||
49957f8e TH |
3622 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3623 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3624 | { | |
3625 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3626 | .ia_uid = current_fsuid(), | |
3627 | .ia_gid = current_fsgid(), }; | |
3628 | ||
3629 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3630 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3631 | return 0; | |
3632 | ||
3633 | return kernfs_setattr(kn, &iattr); | |
3634 | } | |
3635 | ||
4df8dc90 TH |
3636 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3637 | struct cftype *cft) | |
ddbcc7e8 | 3638 | { |
8d7e6fb0 | 3639 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3640 | struct kernfs_node *kn; |
3641 | struct lock_class_key *key = NULL; | |
49957f8e | 3642 | int ret; |
05ef1d7c | 3643 | |
2bd59d48 TH |
3644 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3645 | key = &cft->lockdep_key; | |
3646 | #endif | |
3647 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3648 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3649 | NULL, key); |
49957f8e TH |
3650 | if (IS_ERR(kn)) |
3651 | return PTR_ERR(kn); | |
3652 | ||
3653 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3654 | if (ret) { |
49957f8e | 3655 | kernfs_remove(kn); |
f8f22e53 TH |
3656 | return ret; |
3657 | } | |
3658 | ||
6f60eade TH |
3659 | if (cft->file_offset) { |
3660 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3661 | ||
34c06254 | 3662 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3663 | cfile->kn = kn; |
34c06254 | 3664 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3665 | } |
3666 | ||
f8f22e53 | 3667 | return 0; |
ddbcc7e8 PM |
3668 | } |
3669 | ||
b1f28d31 TH |
3670 | /** |
3671 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3672 | * @css: the target css |
3673 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3674 | * @cfts: array of cftypes to be added |
3675 | * @is_add: whether to add or remove | |
3676 | * | |
3677 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3678 | * For removals, this function never fails. |
b1f28d31 | 3679 | */ |
4df8dc90 TH |
3680 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3681 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3682 | bool is_add) |
ddbcc7e8 | 3683 | { |
6732ed85 | 3684 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3685 | int ret = 0; |
b1f28d31 | 3686 | |
01f6474c | 3687 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3688 | |
6732ed85 TH |
3689 | restart: |
3690 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3691 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3692 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3693 | continue; |
05ebb6e6 | 3694 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3695 | continue; |
d51f39b0 | 3696 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3697 | continue; |
d51f39b0 | 3698 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3699 | continue; |
3700 | ||
2739d3cc | 3701 | if (is_add) { |
4df8dc90 | 3702 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3703 | if (ret) { |
ed3d261b JP |
3704 | pr_warn("%s: failed to add %s, err=%d\n", |
3705 | __func__, cft->name, ret); | |
6732ed85 TH |
3706 | cft_end = cft; |
3707 | is_add = false; | |
3708 | goto restart; | |
b1f28d31 | 3709 | } |
2739d3cc LZ |
3710 | } else { |
3711 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3712 | } |
ddbcc7e8 | 3713 | } |
b598dde3 | 3714 | return ret; |
ddbcc7e8 PM |
3715 | } |
3716 | ||
21a2d343 | 3717 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3718 | { |
3719 | LIST_HEAD(pending); | |
2bb566cb | 3720 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3721 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3722 | struct cgroup_subsys_state *css; |
9ccece80 | 3723 | int ret = 0; |
8e3f6541 | 3724 | |
01f6474c | 3725 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3726 | |
e8c82d20 | 3727 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3728 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3729 | struct cgroup *cgrp = css->cgroup; |
3730 | ||
88cb04b9 | 3731 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3732 | continue; |
3733 | ||
4df8dc90 | 3734 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3735 | if (ret) |
3736 | break; | |
8e3f6541 | 3737 | } |
21a2d343 TH |
3738 | |
3739 | if (is_add && !ret) | |
3740 | kernfs_activate(root->kn); | |
9ccece80 | 3741 | return ret; |
8e3f6541 TH |
3742 | } |
3743 | ||
2da440a2 | 3744 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3745 | { |
2bb566cb | 3746 | struct cftype *cft; |
8e3f6541 | 3747 | |
2bd59d48 TH |
3748 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3749 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3750 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3751 | kfree(cft->kf_ops); | |
3752 | cft->kf_ops = NULL; | |
2da440a2 | 3753 | cft->ss = NULL; |
a8ddc821 TH |
3754 | |
3755 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3756 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3757 | } |
2da440a2 TH |
3758 | } |
3759 | ||
2bd59d48 | 3760 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3761 | { |
3762 | struct cftype *cft; | |
3763 | ||
2bd59d48 TH |
3764 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3765 | struct kernfs_ops *kf_ops; | |
3766 | ||
0adb0704 TH |
3767 | WARN_ON(cft->ss || cft->kf_ops); |
3768 | ||
2bd59d48 TH |
3769 | if (cft->seq_start) |
3770 | kf_ops = &cgroup_kf_ops; | |
3771 | else | |
3772 | kf_ops = &cgroup_kf_single_ops; | |
3773 | ||
3774 | /* | |
3775 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3776 | * make a copy of kf_ops to set its atomic_write_len. | |
3777 | */ | |
3778 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3779 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3780 | if (!kf_ops) { | |
3781 | cgroup_exit_cftypes(cfts); | |
3782 | return -ENOMEM; | |
3783 | } | |
3784 | kf_ops->atomic_write_len = cft->max_write_len; | |
3785 | } | |
8e3f6541 | 3786 | |
2bd59d48 | 3787 | cft->kf_ops = kf_ops; |
2bb566cb | 3788 | cft->ss = ss; |
2bd59d48 | 3789 | } |
2bb566cb | 3790 | |
2bd59d48 | 3791 | return 0; |
2da440a2 TH |
3792 | } |
3793 | ||
21a2d343 TH |
3794 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3795 | { | |
01f6474c | 3796 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3797 | |
3798 | if (!cfts || !cfts[0].ss) | |
3799 | return -ENOENT; | |
3800 | ||
3801 | list_del(&cfts->node); | |
3802 | cgroup_apply_cftypes(cfts, false); | |
3803 | cgroup_exit_cftypes(cfts); | |
3804 | return 0; | |
8e3f6541 | 3805 | } |
8e3f6541 | 3806 | |
79578621 TH |
3807 | /** |
3808 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3809 | * @cfts: zero-length name terminated array of cftypes |
3810 | * | |
2bb566cb TH |
3811 | * Unregister @cfts. Files described by @cfts are removed from all |
3812 | * existing cgroups and all future cgroups won't have them either. This | |
3813 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3814 | * |
3815 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3816 | * registered. |
79578621 | 3817 | */ |
2bb566cb | 3818 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3819 | { |
21a2d343 | 3820 | int ret; |
79578621 | 3821 | |
01f6474c | 3822 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3823 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3824 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3825 | return ret; |
80b13586 TH |
3826 | } |
3827 | ||
8e3f6541 TH |
3828 | /** |
3829 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3830 | * @ss: target cgroup subsystem | |
3831 | * @cfts: zero-length name terminated array of cftypes | |
3832 | * | |
3833 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3834 | * existing cgroups to which @ss is attached and all future cgroups will | |
3835 | * have them too. This function can be called anytime whether @ss is | |
3836 | * attached or not. | |
3837 | * | |
3838 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3839 | * function currently returns 0 as long as @cfts registration is successful | |
3840 | * even if some file creation attempts on existing cgroups fail. | |
3841 | */ | |
2cf669a5 | 3842 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3843 | { |
9ccece80 | 3844 | int ret; |
8e3f6541 | 3845 | |
fc5ed1e9 | 3846 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3847 | return 0; |
3848 | ||
dc5736ed LZ |
3849 | if (!cfts || cfts[0].name[0] == '\0') |
3850 | return 0; | |
2bb566cb | 3851 | |
2bd59d48 TH |
3852 | ret = cgroup_init_cftypes(ss, cfts); |
3853 | if (ret) | |
3854 | return ret; | |
79578621 | 3855 | |
01f6474c | 3856 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3857 | |
0adb0704 | 3858 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3859 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3860 | if (ret) |
21a2d343 | 3861 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3862 | |
01f6474c | 3863 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3864 | return ret; |
79578621 TH |
3865 | } |
3866 | ||
a8ddc821 TH |
3867 | /** |
3868 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3869 | * @ss: target cgroup subsystem | |
3870 | * @cfts: zero-length name terminated array of cftypes | |
3871 | * | |
3872 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3873 | * the default hierarchy. | |
3874 | */ | |
3875 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3876 | { | |
3877 | struct cftype *cft; | |
3878 | ||
3879 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3880 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3881 | return cgroup_add_cftypes(ss, cfts); |
3882 | } | |
3883 | ||
3884 | /** | |
3885 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3886 | * @ss: target cgroup subsystem | |
3887 | * @cfts: zero-length name terminated array of cftypes | |
3888 | * | |
3889 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3890 | * the legacy hierarchies. | |
3891 | */ | |
2cf669a5 TH |
3892 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3893 | { | |
a8ddc821 TH |
3894 | struct cftype *cft; |
3895 | ||
e4b7037c TH |
3896 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3897 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3898 | return cgroup_add_cftypes(ss, cfts); |
3899 | } | |
3900 | ||
34c06254 TH |
3901 | /** |
3902 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3903 | * @cfile: target cgroup_file | |
3904 | * | |
3905 | * @cfile must have been obtained by setting cftype->file_offset. | |
3906 | */ | |
3907 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3908 | { | |
3909 | unsigned long flags; | |
3910 | ||
3911 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3912 | if (cfile->kn) | |
3913 | kernfs_notify(cfile->kn); | |
3914 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3915 | } | |
3916 | ||
a043e3b2 LZ |
3917 | /** |
3918 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3919 | * @cgrp: the cgroup in question | |
3920 | * | |
9157056d TH |
3921 | * Return the number of tasks in the cgroup. The returned number can be |
3922 | * higher than the actual number of tasks due to css_set references from | |
3923 | * namespace roots and temporary usages. | |
a043e3b2 | 3924 | */ |
07bc356e | 3925 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3926 | { |
3927 | int count = 0; | |
69d0206c | 3928 | struct cgrp_cset_link *link; |
817929ec | 3929 | |
82d6489d | 3930 | spin_lock_irq(&css_set_lock); |
69d0206c TH |
3931 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3932 | count += atomic_read(&link->cset->refcount); | |
82d6489d | 3933 | spin_unlock_irq(&css_set_lock); |
bbcb81d0 PM |
3934 | return count; |
3935 | } | |
3936 | ||
53fa5261 | 3937 | /** |
492eb21b | 3938 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3939 | * @pos: the current position (%NULL to initiate traversal) |
3940 | * @parent: css whose children to walk | |
53fa5261 | 3941 | * |
c2931b70 | 3942 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3943 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3944 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3945 | * be returned regardless of their states. | |
3946 | * | |
3947 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3948 | * css which finished ->css_online() is guaranteed to be visible in the | |
3949 | * future iterations and will stay visible until the last reference is put. | |
3950 | * A css which hasn't finished ->css_online() or already finished | |
3951 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3952 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3953 | */ |
c2931b70 TH |
3954 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3955 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3956 | { |
c2931b70 | 3957 | struct cgroup_subsys_state *next; |
53fa5261 | 3958 | |
8353da1f | 3959 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3960 | |
3961 | /* | |
de3f0341 TH |
3962 | * @pos could already have been unlinked from the sibling list. |
3963 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3964 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3965 | * @pos is taken off list, at which time its next pointer is valid, | |
3966 | * and, as releases are serialized, the one pointed to by the next | |
3967 | * pointer is guaranteed to not have started release yet. This | |
3968 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3969 | * critical section, the one pointed to by its next pointer is | |
3970 | * guaranteed to not have finished its RCU grace period even if we | |
3971 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3972 | * |
de3f0341 TH |
3973 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3974 | * dereferenced; however, as each css is given a monotonically | |
3975 | * increasing unique serial number and always appended to the | |
3976 | * sibling list, the next one can be found by walking the parent's | |
3977 | * children until the first css with higher serial number than | |
3978 | * @pos's. While this path can be slower, it happens iff iteration | |
3979 | * races against release and the race window is very small. | |
53fa5261 | 3980 | */ |
3b287a50 | 3981 | if (!pos) { |
c2931b70 TH |
3982 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3983 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3984 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3985 | } else { |
c2931b70 | 3986 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3987 | if (next->serial_nr > pos->serial_nr) |
3988 | break; | |
53fa5261 TH |
3989 | } |
3990 | ||
3b281afb TH |
3991 | /* |
3992 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3993 | * the next sibling. |
3b281afb | 3994 | */ |
c2931b70 TH |
3995 | if (&next->sibling != &parent->children) |
3996 | return next; | |
3b281afb | 3997 | return NULL; |
53fa5261 | 3998 | } |
53fa5261 | 3999 | |
574bd9f7 | 4000 | /** |
492eb21b | 4001 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 4002 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4003 | * @root: css whose descendants to walk |
574bd9f7 | 4004 | * |
492eb21b | 4005 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
4006 | * to visit for pre-order traversal of @root's descendants. @root is |
4007 | * included in the iteration and the first node to be visited. | |
75501a6d | 4008 | * |
87fb54f1 TH |
4009 | * While this function requires cgroup_mutex or RCU read locking, it |
4010 | * doesn't require the whole traversal to be contained in a single critical | |
4011 | * section. This function will return the correct next descendant as long | |
4012 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
4013 | * |
4014 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4015 | * css which finished ->css_online() is guaranteed to be visible in the | |
4016 | * future iterations and will stay visible until the last reference is put. | |
4017 | * A css which hasn't finished ->css_online() or already finished | |
4018 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4019 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4020 | */ |
492eb21b TH |
4021 | struct cgroup_subsys_state * |
4022 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
4023 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4024 | { |
492eb21b | 4025 | struct cgroup_subsys_state *next; |
574bd9f7 | 4026 | |
8353da1f | 4027 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4028 | |
bd8815a6 | 4029 | /* if first iteration, visit @root */ |
7805d000 | 4030 | if (!pos) |
bd8815a6 | 4031 | return root; |
574bd9f7 TH |
4032 | |
4033 | /* visit the first child if exists */ | |
492eb21b | 4034 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4035 | if (next) |
4036 | return next; | |
4037 | ||
4038 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4039 | while (pos != root) { |
5c9d535b | 4040 | next = css_next_child(pos, pos->parent); |
75501a6d | 4041 | if (next) |
574bd9f7 | 4042 | return next; |
5c9d535b | 4043 | pos = pos->parent; |
7805d000 | 4044 | } |
574bd9f7 TH |
4045 | |
4046 | return NULL; | |
4047 | } | |
574bd9f7 | 4048 | |
12a9d2fe | 4049 | /** |
492eb21b TH |
4050 | * css_rightmost_descendant - return the rightmost descendant of a css |
4051 | * @pos: css of interest | |
12a9d2fe | 4052 | * |
492eb21b TH |
4053 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4054 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4055 | * subtree of @pos. |
75501a6d | 4056 | * |
87fb54f1 TH |
4057 | * While this function requires cgroup_mutex or RCU read locking, it |
4058 | * doesn't require the whole traversal to be contained in a single critical | |
4059 | * section. This function will return the correct rightmost descendant as | |
4060 | * long as @pos is accessible. | |
12a9d2fe | 4061 | */ |
492eb21b TH |
4062 | struct cgroup_subsys_state * |
4063 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4064 | { |
492eb21b | 4065 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4066 | |
8353da1f | 4067 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4068 | |
4069 | do { | |
4070 | last = pos; | |
4071 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4072 | pos = NULL; | |
492eb21b | 4073 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4074 | pos = tmp; |
4075 | } while (pos); | |
4076 | ||
4077 | return last; | |
4078 | } | |
12a9d2fe | 4079 | |
492eb21b TH |
4080 | static struct cgroup_subsys_state * |
4081 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4082 | { |
492eb21b | 4083 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4084 | |
4085 | do { | |
4086 | last = pos; | |
492eb21b | 4087 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4088 | } while (pos); |
4089 | ||
4090 | return last; | |
4091 | } | |
4092 | ||
4093 | /** | |
492eb21b | 4094 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4095 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4096 | * @root: css whose descendants to walk |
574bd9f7 | 4097 | * |
492eb21b | 4098 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4099 | * to visit for post-order traversal of @root's descendants. @root is |
4100 | * included in the iteration and the last node to be visited. | |
75501a6d | 4101 | * |
87fb54f1 TH |
4102 | * While this function requires cgroup_mutex or RCU read locking, it |
4103 | * doesn't require the whole traversal to be contained in a single critical | |
4104 | * section. This function will return the correct next descendant as long | |
4105 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4106 | * @cgroup. | |
c2931b70 TH |
4107 | * |
4108 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4109 | * css which finished ->css_online() is guaranteed to be visible in the | |
4110 | * future iterations and will stay visible until the last reference is put. | |
4111 | * A css which hasn't finished ->css_online() or already finished | |
4112 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4113 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4114 | */ |
492eb21b TH |
4115 | struct cgroup_subsys_state * |
4116 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4117 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4118 | { |
492eb21b | 4119 | struct cgroup_subsys_state *next; |
574bd9f7 | 4120 | |
8353da1f | 4121 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4122 | |
58b79a91 TH |
4123 | /* if first iteration, visit leftmost descendant which may be @root */ |
4124 | if (!pos) | |
4125 | return css_leftmost_descendant(root); | |
574bd9f7 | 4126 | |
bd8815a6 TH |
4127 | /* if we visited @root, we're done */ |
4128 | if (pos == root) | |
4129 | return NULL; | |
4130 | ||
574bd9f7 | 4131 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4132 | next = css_next_child(pos, pos->parent); |
75501a6d | 4133 | if (next) |
492eb21b | 4134 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4135 | |
4136 | /* no sibling left, visit parent */ | |
5c9d535b | 4137 | return pos->parent; |
574bd9f7 | 4138 | } |
574bd9f7 | 4139 | |
f3d46500 TH |
4140 | /** |
4141 | * css_has_online_children - does a css have online children | |
4142 | * @css: the target css | |
4143 | * | |
4144 | * Returns %true if @css has any online children; otherwise, %false. This | |
4145 | * function can be called from any context but the caller is responsible | |
4146 | * for synchronizing against on/offlining as necessary. | |
4147 | */ | |
4148 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4149 | { |
f3d46500 TH |
4150 | struct cgroup_subsys_state *child; |
4151 | bool ret = false; | |
cbc125ef TH |
4152 | |
4153 | rcu_read_lock(); | |
f3d46500 | 4154 | css_for_each_child(child, css) { |
99bae5f9 | 4155 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4156 | ret = true; |
4157 | break; | |
cbc125ef TH |
4158 | } |
4159 | } | |
4160 | rcu_read_unlock(); | |
f3d46500 | 4161 | return ret; |
574bd9f7 | 4162 | } |
574bd9f7 | 4163 | |
0942eeee | 4164 | /** |
ecb9d535 | 4165 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4166 | * @it: the iterator to advance |
4167 | * | |
4168 | * Advance @it to the next css_set to walk. | |
d515876e | 4169 | */ |
ecb9d535 | 4170 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4171 | { |
0f0a2b4f | 4172 | struct list_head *l = it->cset_pos; |
d515876e TH |
4173 | struct cgrp_cset_link *link; |
4174 | struct css_set *cset; | |
4175 | ||
f0d9a5f1 | 4176 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4177 | |
d515876e TH |
4178 | /* Advance to the next non-empty css_set */ |
4179 | do { | |
4180 | l = l->next; | |
0f0a2b4f TH |
4181 | if (l == it->cset_head) { |
4182 | it->cset_pos = NULL; | |
ecb9d535 | 4183 | it->task_pos = NULL; |
d515876e TH |
4184 | return; |
4185 | } | |
3ebb2b6e TH |
4186 | |
4187 | if (it->ss) { | |
4188 | cset = container_of(l, struct css_set, | |
4189 | e_cset_node[it->ss->id]); | |
4190 | } else { | |
4191 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4192 | cset = link->cset; | |
4193 | } | |
0de0942d | 4194 | } while (!css_set_populated(cset)); |
c7561128 | 4195 | |
0f0a2b4f | 4196 | it->cset_pos = l; |
c7561128 TH |
4197 | |
4198 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 4199 | it->task_pos = cset->tasks.next; |
c7561128 | 4200 | else |
0f0a2b4f TH |
4201 | it->task_pos = cset->mg_tasks.next; |
4202 | ||
4203 | it->tasks_head = &cset->tasks; | |
4204 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4205 | |
4206 | /* | |
4207 | * We don't keep css_sets locked across iteration steps and thus | |
4208 | * need to take steps to ensure that iteration can be resumed after | |
4209 | * the lock is re-acquired. Iteration is performed at two levels - | |
4210 | * css_sets and tasks in them. | |
4211 | * | |
4212 | * Once created, a css_set never leaves its cgroup lists, so a | |
4213 | * pinned css_set is guaranteed to stay put and we can resume | |
4214 | * iteration afterwards. | |
4215 | * | |
4216 | * Tasks may leave @cset across iteration steps. This is resolved | |
4217 | * by registering each iterator with the css_set currently being | |
4218 | * walked and making css_set_move_task() advance iterators whose | |
4219 | * next task is leaving. | |
4220 | */ | |
4221 | if (it->cur_cset) { | |
4222 | list_del(&it->iters_node); | |
4223 | put_css_set_locked(it->cur_cset); | |
4224 | } | |
4225 | get_css_set(cset); | |
4226 | it->cur_cset = cset; | |
4227 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4228 | } |
4229 | ||
ecb9d535 TH |
4230 | static void css_task_iter_advance(struct css_task_iter *it) |
4231 | { | |
4232 | struct list_head *l = it->task_pos; | |
4233 | ||
f0d9a5f1 | 4234 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4235 | WARN_ON_ONCE(!l); |
4236 | ||
4237 | /* | |
4238 | * Advance iterator to find next entry. cset->tasks is consumed | |
4239 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4240 | * next cset. | |
4241 | */ | |
4242 | l = l->next; | |
4243 | ||
4244 | if (l == it->tasks_head) | |
4245 | l = it->mg_tasks_head->next; | |
4246 | ||
4247 | if (l == it->mg_tasks_head) | |
4248 | css_task_iter_advance_css_set(it); | |
4249 | else | |
4250 | it->task_pos = l; | |
4251 | } | |
4252 | ||
0942eeee | 4253 | /** |
72ec7029 TH |
4254 | * css_task_iter_start - initiate task iteration |
4255 | * @css: the css to walk tasks of | |
0942eeee TH |
4256 | * @it: the task iterator to use |
4257 | * | |
72ec7029 TH |
4258 | * Initiate iteration through the tasks of @css. The caller can call |
4259 | * css_task_iter_next() to walk through the tasks until the function | |
4260 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4261 | * called. | |
0942eeee | 4262 | */ |
72ec7029 TH |
4263 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4264 | struct css_task_iter *it) | |
817929ec | 4265 | { |
56fde9e0 TH |
4266 | /* no one should try to iterate before mounting cgroups */ |
4267 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4268 | |
ed27b9f7 TH |
4269 | memset(it, 0, sizeof(*it)); |
4270 | ||
82d6489d | 4271 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4272 | |
3ebb2b6e TH |
4273 | it->ss = css->ss; |
4274 | ||
4275 | if (it->ss) | |
4276 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4277 | else | |
4278 | it->cset_pos = &css->cgroup->cset_links; | |
4279 | ||
0f0a2b4f | 4280 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4281 | |
ecb9d535 | 4282 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4283 | |
82d6489d | 4284 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4285 | } |
4286 | ||
0942eeee | 4287 | /** |
72ec7029 | 4288 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4289 | * @it: the task iterator being iterated |
4290 | * | |
4291 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4292 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4293 | * reaches the end. | |
0942eeee | 4294 | */ |
72ec7029 | 4295 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4296 | { |
d5745675 | 4297 | if (it->cur_task) { |
ed27b9f7 | 4298 | put_task_struct(it->cur_task); |
d5745675 TH |
4299 | it->cur_task = NULL; |
4300 | } | |
ed27b9f7 | 4301 | |
82d6489d | 4302 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4303 | |
d5745675 TH |
4304 | if (it->task_pos) { |
4305 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4306 | cg_list); | |
4307 | get_task_struct(it->cur_task); | |
4308 | css_task_iter_advance(it); | |
4309 | } | |
ed27b9f7 | 4310 | |
82d6489d | 4311 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4312 | |
4313 | return it->cur_task; | |
817929ec PM |
4314 | } |
4315 | ||
0942eeee | 4316 | /** |
72ec7029 | 4317 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4318 | * @it: the task iterator to finish |
4319 | * | |
72ec7029 | 4320 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4321 | */ |
72ec7029 | 4322 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4323 | { |
ed27b9f7 | 4324 | if (it->cur_cset) { |
82d6489d | 4325 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4326 | list_del(&it->iters_node); |
4327 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4328 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4329 | } |
4330 | ||
4331 | if (it->cur_task) | |
4332 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4333 | } |
4334 | ||
4335 | /** | |
8cc99345 TH |
4336 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4337 | * @to: cgroup to which the tasks will be moved | |
4338 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4339 | * |
eaf797ab TH |
4340 | * Locking rules between cgroup_post_fork() and the migration path |
4341 | * guarantee that, if a task is forking while being migrated, the new child | |
4342 | * is guaranteed to be either visible in the source cgroup after the | |
4343 | * parent's migration is complete or put into the target cgroup. No task | |
4344 | * can slip out of migration through forking. | |
31a7df01 | 4345 | */ |
8cc99345 | 4346 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4347 | { |
952aaa12 TH |
4348 | LIST_HEAD(preloaded_csets); |
4349 | struct cgrp_cset_link *link; | |
72ec7029 | 4350 | struct css_task_iter it; |
e406d1cf | 4351 | struct task_struct *task; |
952aaa12 | 4352 | int ret; |
31a7df01 | 4353 | |
6c694c88 TH |
4354 | if (!cgroup_may_migrate_to(to)) |
4355 | return -EBUSY; | |
4356 | ||
952aaa12 | 4357 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4358 | |
eedd0f4c EB |
4359 | percpu_down_write(&cgroup_threadgroup_rwsem); |
4360 | ||
952aaa12 | 4361 | /* all tasks in @from are being moved, all csets are source */ |
82d6489d | 4362 | spin_lock_irq(&css_set_lock); |
952aaa12 TH |
4363 | list_for_each_entry(link, &from->cset_links, cset_link) |
4364 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
82d6489d | 4365 | spin_unlock_irq(&css_set_lock); |
31a7df01 | 4366 | |
e4857982 | 4367 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
952aaa12 TH |
4368 | if (ret) |
4369 | goto out_err; | |
8cc99345 | 4370 | |
952aaa12 | 4371 | /* |
2cfa2b19 | 4372 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4373 | * ->can_attach() fails. |
4374 | */ | |
e406d1cf | 4375 | do { |
9d800df1 | 4376 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4377 | task = css_task_iter_next(&it); |
4378 | if (task) | |
4379 | get_task_struct(task); | |
4380 | css_task_iter_end(&it); | |
4381 | ||
4382 | if (task) { | |
37ff9f8f | 4383 | ret = cgroup_migrate(task, false, to->root); |
e406d1cf TH |
4384 | put_task_struct(task); |
4385 | } | |
4386 | } while (task && !ret); | |
952aaa12 TH |
4387 | out_err: |
4388 | cgroup_migrate_finish(&preloaded_csets); | |
eedd0f4c | 4389 | percpu_up_write(&cgroup_threadgroup_rwsem); |
47cfcd09 | 4390 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4391 | return ret; |
8cc99345 TH |
4392 | } |
4393 | ||
bbcb81d0 | 4394 | /* |
102a775e | 4395 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4396 | * |
4397 | * Reading this file can return large amounts of data if a cgroup has | |
4398 | * *lots* of attached tasks. So it may need several calls to read(), | |
4399 | * but we cannot guarantee that the information we produce is correct | |
4400 | * unless we produce it entirely atomically. | |
4401 | * | |
bbcb81d0 | 4402 | */ |
bbcb81d0 | 4403 | |
24528255 LZ |
4404 | /* which pidlist file are we talking about? */ |
4405 | enum cgroup_filetype { | |
4406 | CGROUP_FILE_PROCS, | |
4407 | CGROUP_FILE_TASKS, | |
4408 | }; | |
4409 | ||
4410 | /* | |
4411 | * A pidlist is a list of pids that virtually represents the contents of one | |
4412 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4413 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4414 | * to the cgroup. | |
4415 | */ | |
4416 | struct cgroup_pidlist { | |
4417 | /* | |
4418 | * used to find which pidlist is wanted. doesn't change as long as | |
4419 | * this particular list stays in the list. | |
4420 | */ | |
4421 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4422 | /* array of xids */ | |
4423 | pid_t *list; | |
4424 | /* how many elements the above list has */ | |
4425 | int length; | |
24528255 LZ |
4426 | /* each of these stored in a list by its cgroup */ |
4427 | struct list_head links; | |
4428 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4429 | struct cgroup *owner; | |
b1a21367 TH |
4430 | /* for delayed destruction */ |
4431 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4432 | }; |
4433 | ||
d1d9fd33 BB |
4434 | /* |
4435 | * The following two functions "fix" the issue where there are more pids | |
4436 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4437 | * TODO: replace with a kernel-wide solution to this problem | |
4438 | */ | |
4439 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4440 | static void *pidlist_allocate(int count) | |
4441 | { | |
4442 | if (PIDLIST_TOO_LARGE(count)) | |
4443 | return vmalloc(count * sizeof(pid_t)); | |
4444 | else | |
4445 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4446 | } | |
b1a21367 | 4447 | |
d1d9fd33 BB |
4448 | static void pidlist_free(void *p) |
4449 | { | |
58794514 | 4450 | kvfree(p); |
d1d9fd33 | 4451 | } |
d1d9fd33 | 4452 | |
b1a21367 TH |
4453 | /* |
4454 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4455 | * should be left afterwards. | |
4456 | */ | |
4457 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4458 | { | |
4459 | struct cgroup_pidlist *l, *tmp_l; | |
4460 | ||
4461 | mutex_lock(&cgrp->pidlist_mutex); | |
4462 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4463 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4464 | mutex_unlock(&cgrp->pidlist_mutex); | |
4465 | ||
4466 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4467 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4468 | } | |
4469 | ||
4470 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4471 | { | |
4472 | struct delayed_work *dwork = to_delayed_work(work); | |
4473 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4474 | destroy_dwork); | |
4475 | struct cgroup_pidlist *tofree = NULL; | |
4476 | ||
4477 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4478 | |
4479 | /* | |
04502365 TH |
4480 | * Destroy iff we didn't get queued again. The state won't change |
4481 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4482 | */ |
04502365 | 4483 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4484 | list_del(&l->links); |
4485 | pidlist_free(l->list); | |
4486 | put_pid_ns(l->key.ns); | |
4487 | tofree = l; | |
4488 | } | |
4489 | ||
b1a21367 TH |
4490 | mutex_unlock(&l->owner->pidlist_mutex); |
4491 | kfree(tofree); | |
4492 | } | |
4493 | ||
bbcb81d0 | 4494 | /* |
102a775e | 4495 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4496 | * Returns the number of unique elements. |
bbcb81d0 | 4497 | */ |
6ee211ad | 4498 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4499 | { |
102a775e | 4500 | int src, dest = 1; |
102a775e BB |
4501 | |
4502 | /* | |
4503 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4504 | * edge cases first; no work needs to be done for either | |
4505 | */ | |
4506 | if (length == 0 || length == 1) | |
4507 | return length; | |
4508 | /* src and dest walk down the list; dest counts unique elements */ | |
4509 | for (src = 1; src < length; src++) { | |
4510 | /* find next unique element */ | |
4511 | while (list[src] == list[src-1]) { | |
4512 | src++; | |
4513 | if (src == length) | |
4514 | goto after; | |
4515 | } | |
4516 | /* dest always points to where the next unique element goes */ | |
4517 | list[dest] = list[src]; | |
4518 | dest++; | |
4519 | } | |
4520 | after: | |
102a775e BB |
4521 | return dest; |
4522 | } | |
4523 | ||
afb2bc14 TH |
4524 | /* |
4525 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4526 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4527 | * different per namespace, each namespace needs differently sorted list, | |
4528 | * making it impossible to use, for example, single rbtree of member tasks | |
4529 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4530 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4531 | * pidlists keyed by cgroup and namespace. | |
4532 | * | |
4533 | * All this extra complexity was caused by the original implementation | |
4534 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4535 | * want to do away with it. Explicitly scramble sort order if on the |
4536 | * default hierarchy so that no such expectation exists in the new | |
4537 | * interface. | |
afb2bc14 TH |
4538 | * |
4539 | * Scrambling is done by swapping every two consecutive bits, which is | |
4540 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4541 | */ | |
4542 | static pid_t pid_fry(pid_t pid) | |
4543 | { | |
4544 | unsigned a = pid & 0x55555555; | |
4545 | unsigned b = pid & 0xAAAAAAAA; | |
4546 | ||
4547 | return (a << 1) | (b >> 1); | |
4548 | } | |
4549 | ||
4550 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4551 | { | |
aa6ec29b | 4552 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4553 | return pid_fry(pid); |
4554 | else | |
4555 | return pid; | |
4556 | } | |
4557 | ||
102a775e BB |
4558 | static int cmppid(const void *a, const void *b) |
4559 | { | |
4560 | return *(pid_t *)a - *(pid_t *)b; | |
4561 | } | |
4562 | ||
afb2bc14 TH |
4563 | static int fried_cmppid(const void *a, const void *b) |
4564 | { | |
4565 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4566 | } | |
4567 | ||
e6b81710 TH |
4568 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4569 | enum cgroup_filetype type) | |
4570 | { | |
4571 | struct cgroup_pidlist *l; | |
4572 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4573 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4574 | ||
4575 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4576 | ||
4577 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4578 | if (l->key.type == type && l->key.ns == ns) | |
4579 | return l; | |
4580 | return NULL; | |
4581 | } | |
4582 | ||
72a8cb30 BB |
4583 | /* |
4584 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4585 | * returns with the lock on that pidlist already held, and takes care | |
4586 | * of the use count, or returns NULL with no locks held if we're out of | |
4587 | * memory. | |
4588 | */ | |
e6b81710 TH |
4589 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4590 | enum cgroup_filetype type) | |
72a8cb30 BB |
4591 | { |
4592 | struct cgroup_pidlist *l; | |
b70cc5fd | 4593 | |
e6b81710 TH |
4594 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4595 | ||
4596 | l = cgroup_pidlist_find(cgrp, type); | |
4597 | if (l) | |
4598 | return l; | |
4599 | ||
72a8cb30 | 4600 | /* entry not found; create a new one */ |
f4f4be2b | 4601 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4602 | if (!l) |
72a8cb30 | 4603 | return l; |
e6b81710 | 4604 | |
b1a21367 | 4605 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4606 | l->key.type = type; |
e6b81710 TH |
4607 | /* don't need task_nsproxy() if we're looking at ourself */ |
4608 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4609 | l->owner = cgrp; |
4610 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4611 | return l; |
4612 | } | |
4613 | ||
102a775e BB |
4614 | /* |
4615 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4616 | */ | |
72a8cb30 BB |
4617 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4618 | struct cgroup_pidlist **lp) | |
102a775e BB |
4619 | { |
4620 | pid_t *array; | |
4621 | int length; | |
4622 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4623 | struct css_task_iter it; |
817929ec | 4624 | struct task_struct *tsk; |
102a775e BB |
4625 | struct cgroup_pidlist *l; |
4626 | ||
4bac00d1 TH |
4627 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4628 | ||
102a775e BB |
4629 | /* |
4630 | * If cgroup gets more users after we read count, we won't have | |
4631 | * enough space - tough. This race is indistinguishable to the | |
4632 | * caller from the case that the additional cgroup users didn't | |
4633 | * show up until sometime later on. | |
4634 | */ | |
4635 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4636 | array = pidlist_allocate(length); |
102a775e BB |
4637 | if (!array) |
4638 | return -ENOMEM; | |
4639 | /* now, populate the array */ | |
9d800df1 | 4640 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4641 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4642 | if (unlikely(n == length)) |
817929ec | 4643 | break; |
102a775e | 4644 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4645 | if (type == CGROUP_FILE_PROCS) |
4646 | pid = task_tgid_vnr(tsk); | |
4647 | else | |
4648 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4649 | if (pid > 0) /* make sure to only use valid results */ |
4650 | array[n++] = pid; | |
817929ec | 4651 | } |
72ec7029 | 4652 | css_task_iter_end(&it); |
102a775e BB |
4653 | length = n; |
4654 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4655 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4656 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4657 | else | |
4658 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4659 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4660 | length = pidlist_uniq(array, length); |
e6b81710 | 4661 | |
e6b81710 | 4662 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4663 | if (!l) { |
d1d9fd33 | 4664 | pidlist_free(array); |
72a8cb30 | 4665 | return -ENOMEM; |
102a775e | 4666 | } |
e6b81710 TH |
4667 | |
4668 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4669 | pidlist_free(l->list); |
102a775e BB |
4670 | l->list = array; |
4671 | l->length = length; | |
72a8cb30 | 4672 | *lp = l; |
102a775e | 4673 | return 0; |
bbcb81d0 PM |
4674 | } |
4675 | ||
846c7bb0 | 4676 | /** |
a043e3b2 | 4677 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4678 | * @stats: cgroupstats to fill information into |
4679 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4680 | * been requested. | |
a043e3b2 LZ |
4681 | * |
4682 | * Build and fill cgroupstats so that taskstats can export it to user | |
4683 | * space. | |
846c7bb0 BS |
4684 | */ |
4685 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4686 | { | |
2bd59d48 | 4687 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4688 | struct cgroup *cgrp; |
72ec7029 | 4689 | struct css_task_iter it; |
846c7bb0 | 4690 | struct task_struct *tsk; |
33d283be | 4691 | |
2bd59d48 TH |
4692 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4693 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4694 | kernfs_type(kn) != KERNFS_DIR) | |
4695 | return -EINVAL; | |
4696 | ||
bad34660 LZ |
4697 | mutex_lock(&cgroup_mutex); |
4698 | ||
846c7bb0 | 4699 | /* |
2bd59d48 | 4700 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4701 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4702 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4703 | */ |
2bd59d48 TH |
4704 | rcu_read_lock(); |
4705 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4706 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4707 | rcu_read_unlock(); |
bad34660 | 4708 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4709 | return -ENOENT; |
4710 | } | |
bad34660 | 4711 | rcu_read_unlock(); |
846c7bb0 | 4712 | |
9d800df1 | 4713 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4714 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4715 | switch (tsk->state) { |
4716 | case TASK_RUNNING: | |
4717 | stats->nr_running++; | |
4718 | break; | |
4719 | case TASK_INTERRUPTIBLE: | |
4720 | stats->nr_sleeping++; | |
4721 | break; | |
4722 | case TASK_UNINTERRUPTIBLE: | |
4723 | stats->nr_uninterruptible++; | |
4724 | break; | |
4725 | case TASK_STOPPED: | |
4726 | stats->nr_stopped++; | |
4727 | break; | |
4728 | default: | |
4729 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4730 | stats->nr_io_wait++; | |
4731 | break; | |
4732 | } | |
4733 | } | |
72ec7029 | 4734 | css_task_iter_end(&it); |
846c7bb0 | 4735 | |
bad34660 | 4736 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4737 | return 0; |
846c7bb0 BS |
4738 | } |
4739 | ||
8f3ff208 | 4740 | |
bbcb81d0 | 4741 | /* |
102a775e | 4742 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4743 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4744 | * in the cgroup->l->list array. |
bbcb81d0 | 4745 | */ |
cc31edce | 4746 | |
102a775e | 4747 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4748 | { |
cc31edce PM |
4749 | /* |
4750 | * Initially we receive a position value that corresponds to | |
4751 | * one more than the last pid shown (or 0 on the first call or | |
4752 | * after a seek to the start). Use a binary-search to find the | |
4753 | * next pid to display, if any | |
4754 | */ | |
2bd59d48 | 4755 | struct kernfs_open_file *of = s->private; |
7da11279 | 4756 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4757 | struct cgroup_pidlist *l; |
7da11279 | 4758 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4759 | int index = 0, pid = *pos; |
4bac00d1 TH |
4760 | int *iter, ret; |
4761 | ||
4762 | mutex_lock(&cgrp->pidlist_mutex); | |
4763 | ||
4764 | /* | |
5d22444f | 4765 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4766 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4767 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4768 | * could already have been destroyed. |
4769 | */ | |
5d22444f TH |
4770 | if (of->priv) |
4771 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4772 | |
4773 | /* | |
4774 | * Either this is the first start() after open or the matching | |
4775 | * pidlist has been destroyed inbetween. Create a new one. | |
4776 | */ | |
5d22444f TH |
4777 | if (!of->priv) { |
4778 | ret = pidlist_array_load(cgrp, type, | |
4779 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4780 | if (ret) |
4781 | return ERR_PTR(ret); | |
4782 | } | |
5d22444f | 4783 | l = of->priv; |
cc31edce | 4784 | |
cc31edce | 4785 | if (pid) { |
102a775e | 4786 | int end = l->length; |
20777766 | 4787 | |
cc31edce PM |
4788 | while (index < end) { |
4789 | int mid = (index + end) / 2; | |
afb2bc14 | 4790 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4791 | index = mid; |
4792 | break; | |
afb2bc14 | 4793 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4794 | index = mid + 1; |
4795 | else | |
4796 | end = mid; | |
4797 | } | |
4798 | } | |
4799 | /* If we're off the end of the array, we're done */ | |
102a775e | 4800 | if (index >= l->length) |
cc31edce PM |
4801 | return NULL; |
4802 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4803 | iter = l->list + index; |
afb2bc14 | 4804 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4805 | return iter; |
4806 | } | |
4807 | ||
102a775e | 4808 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4809 | { |
2bd59d48 | 4810 | struct kernfs_open_file *of = s->private; |
5d22444f | 4811 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4812 | |
5d22444f TH |
4813 | if (l) |
4814 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4815 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4816 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4817 | } |
4818 | ||
102a775e | 4819 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4820 | { |
2bd59d48 | 4821 | struct kernfs_open_file *of = s->private; |
5d22444f | 4822 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4823 | pid_t *p = v; |
4824 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4825 | /* |
4826 | * Advance to the next pid in the array. If this goes off the | |
4827 | * end, we're done | |
4828 | */ | |
4829 | p++; | |
4830 | if (p >= end) { | |
4831 | return NULL; | |
4832 | } else { | |
7da11279 | 4833 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4834 | return p; |
4835 | } | |
4836 | } | |
4837 | ||
102a775e | 4838 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4839 | { |
94ff212d JP |
4840 | seq_printf(s, "%d\n", *(int *)v); |
4841 | ||
4842 | return 0; | |
cc31edce | 4843 | } |
bbcb81d0 | 4844 | |
182446d0 TH |
4845 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4846 | struct cftype *cft) | |
81a6a5cd | 4847 | { |
182446d0 | 4848 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4849 | } |
4850 | ||
182446d0 TH |
4851 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4852 | struct cftype *cft, u64 val) | |
6379c106 | 4853 | { |
6379c106 | 4854 | if (val) |
182446d0 | 4855 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4856 | else |
182446d0 | 4857 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4858 | return 0; |
4859 | } | |
4860 | ||
182446d0 TH |
4861 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4862 | struct cftype *cft) | |
97978e6d | 4863 | { |
182446d0 | 4864 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4865 | } |
4866 | ||
182446d0 TH |
4867 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4868 | struct cftype *cft, u64 val) | |
97978e6d DL |
4869 | { |
4870 | if (val) | |
182446d0 | 4871 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4872 | else |
182446d0 | 4873 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4874 | return 0; |
4875 | } | |
4876 | ||
a14c6874 TH |
4877 | /* cgroup core interface files for the default hierarchy */ |
4878 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4879 | { |
d5c56ced | 4880 | .name = "cgroup.procs", |
6f60eade | 4881 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4882 | .seq_start = cgroup_pidlist_start, |
4883 | .seq_next = cgroup_pidlist_next, | |
4884 | .seq_stop = cgroup_pidlist_stop, | |
4885 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4886 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4887 | .write = cgroup_procs_write, |
102a775e | 4888 | }, |
f8f22e53 TH |
4889 | { |
4890 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4891 | .seq_show = cgroup_controllers_show, |
4892 | }, | |
4893 | { | |
4894 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4895 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4896 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4897 | }, |
842b597e | 4898 | { |
4a07c222 | 4899 | .name = "cgroup.events", |
a14c6874 | 4900 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4901 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4902 | .seq_show = cgroup_events_show, |
842b597e | 4903 | }, |
a14c6874 TH |
4904 | { } /* terminate */ |
4905 | }; | |
d5c56ced | 4906 | |
a14c6874 TH |
4907 | /* cgroup core interface files for the legacy hierarchies */ |
4908 | static struct cftype cgroup_legacy_base_files[] = { | |
4909 | { | |
4910 | .name = "cgroup.procs", | |
4911 | .seq_start = cgroup_pidlist_start, | |
4912 | .seq_next = cgroup_pidlist_next, | |
4913 | .seq_stop = cgroup_pidlist_stop, | |
4914 | .seq_show = cgroup_pidlist_show, | |
4915 | .private = CGROUP_FILE_PROCS, | |
4916 | .write = cgroup_procs_write, | |
a14c6874 TH |
4917 | }, |
4918 | { | |
4919 | .name = "cgroup.clone_children", | |
4920 | .read_u64 = cgroup_clone_children_read, | |
4921 | .write_u64 = cgroup_clone_children_write, | |
4922 | }, | |
4923 | { | |
4924 | .name = "cgroup.sane_behavior", | |
4925 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4926 | .seq_show = cgroup_sane_behavior_show, | |
4927 | }, | |
d5c56ced TH |
4928 | { |
4929 | .name = "tasks", | |
6612f05b TH |
4930 | .seq_start = cgroup_pidlist_start, |
4931 | .seq_next = cgroup_pidlist_next, | |
4932 | .seq_stop = cgroup_pidlist_stop, | |
4933 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4934 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4935 | .write = cgroup_tasks_write, |
d5c56ced TH |
4936 | }, |
4937 | { | |
4938 | .name = "notify_on_release", | |
d5c56ced TH |
4939 | .read_u64 = cgroup_read_notify_on_release, |
4940 | .write_u64 = cgroup_write_notify_on_release, | |
4941 | }, | |
6e6ff25b TH |
4942 | { |
4943 | .name = "release_agent", | |
a14c6874 | 4944 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4945 | .seq_show = cgroup_release_agent_show, |
451af504 | 4946 | .write = cgroup_release_agent_write, |
5f469907 | 4947 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4948 | }, |
db0416b6 | 4949 | { } /* terminate */ |
bbcb81d0 PM |
4950 | }; |
4951 | ||
0c21ead1 TH |
4952 | /* |
4953 | * css destruction is four-stage process. | |
4954 | * | |
4955 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4956 | * Implemented in kill_css(). | |
4957 | * | |
4958 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4959 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4960 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4961 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4962 | * |
4963 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4964 | * accessors are inside RCU read sections. css_release() schedules the | |
4965 | * RCU callback. | |
4966 | * | |
4967 | * 4. After the grace period, the css can be freed. Implemented in | |
4968 | * css_free_work_fn(). | |
4969 | * | |
4970 | * It is actually hairier because both step 2 and 4 require process context | |
4971 | * and thus involve punting to css->destroy_work adding two additional | |
4972 | * steps to the already complex sequence. | |
4973 | */ | |
35ef10da | 4974 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4975 | { |
4976 | struct cgroup_subsys_state *css = | |
35ef10da | 4977 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4978 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4979 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4980 | |
9a1049da TH |
4981 | percpu_ref_exit(&css->refcnt); |
4982 | ||
01e58659 | 4983 | if (ss) { |
9d755d33 | 4984 | /* css free path */ |
8bb5ef79 | 4985 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4986 | int id = css->id; |
4987 | ||
01e58659 VD |
4988 | ss->css_free(css); |
4989 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4990 | cgroup_put(cgrp); |
8bb5ef79 TH |
4991 | |
4992 | if (parent) | |
4993 | css_put(parent); | |
9d755d33 TH |
4994 | } else { |
4995 | /* cgroup free path */ | |
4996 | atomic_dec(&cgrp->root->nr_cgrps); | |
4997 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4998 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4999 | |
d51f39b0 | 5000 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
5001 | /* |
5002 | * We get a ref to the parent, and put the ref when | |
5003 | * this cgroup is being freed, so it's guaranteed | |
5004 | * that the parent won't be destroyed before its | |
5005 | * children. | |
5006 | */ | |
d51f39b0 | 5007 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
5008 | kernfs_put(cgrp->kn); |
5009 | kfree(cgrp); | |
5010 | } else { | |
5011 | /* | |
5012 | * This is root cgroup's refcnt reaching zero, | |
5013 | * which indicates that the root should be | |
5014 | * released. | |
5015 | */ | |
5016 | cgroup_destroy_root(cgrp->root); | |
5017 | } | |
5018 | } | |
48ddbe19 TH |
5019 | } |
5020 | ||
0c21ead1 | 5021 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
5022 | { |
5023 | struct cgroup_subsys_state *css = | |
0c21ead1 | 5024 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 5025 | |
35ef10da | 5026 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 5027 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
5028 | } |
5029 | ||
25e15d83 | 5030 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
5031 | { |
5032 | struct cgroup_subsys_state *css = | |
25e15d83 | 5033 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 5034 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 5035 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 5036 | |
1fed1b2e TH |
5037 | mutex_lock(&cgroup_mutex); |
5038 | ||
de3f0341 | 5039 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
5040 | list_del_rcu(&css->sibling); |
5041 | ||
9d755d33 TH |
5042 | if (ss) { |
5043 | /* css release path */ | |
01e58659 | 5044 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
5045 | if (ss->css_released) |
5046 | ss->css_released(css); | |
9d755d33 TH |
5047 | } else { |
5048 | /* cgroup release path */ | |
9d755d33 TH |
5049 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
5050 | cgrp->id = -1; | |
a4189487 LZ |
5051 | |
5052 | /* | |
5053 | * There are two control paths which try to determine | |
5054 | * cgroup from dentry without going through kernfs - | |
5055 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
5056 | * Those are supported by RCU protecting clearing of | |
5057 | * cgrp->kn->priv backpointer. | |
5058 | */ | |
6cd0f5bb TH |
5059 | if (cgrp->kn) |
5060 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
5061 | NULL); | |
9d755d33 | 5062 | } |
d3daf28d | 5063 | |
1fed1b2e TH |
5064 | mutex_unlock(&cgroup_mutex); |
5065 | ||
0c21ead1 | 5066 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
5067 | } |
5068 | ||
d3daf28d TH |
5069 | static void css_release(struct percpu_ref *ref) |
5070 | { | |
5071 | struct cgroup_subsys_state *css = | |
5072 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5073 | ||
25e15d83 TH |
5074 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5075 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5076 | } |
5077 | ||
ddfcadab TH |
5078 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5079 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5080 | { |
0cb51d71 TH |
5081 | lockdep_assert_held(&cgroup_mutex); |
5082 | ||
ddfcadab TH |
5083 | cgroup_get(cgrp); |
5084 | ||
d5c419b6 | 5085 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5086 | css->cgroup = cgrp; |
72c97e54 | 5087 | css->ss = ss; |
8fa3b8d6 | 5088 | css->id = -1; |
d5c419b6 TH |
5089 | INIT_LIST_HEAD(&css->sibling); |
5090 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 5091 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5092 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5093 | |
d51f39b0 TH |
5094 | if (cgroup_parent(cgrp)) { |
5095 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5096 | css_get(css->parent); |
ddfcadab | 5097 | } |
48ddbe19 | 5098 | |
ca8bdcaf | 5099 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5100 | } |
5101 | ||
2a4ac633 | 5102 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5103 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5104 | { |
623f926b | 5105 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5106 | int ret = 0; |
5107 | ||
a31f2d3f TH |
5108 | lockdep_assert_held(&cgroup_mutex); |
5109 | ||
92fb9748 | 5110 | if (ss->css_online) |
eb95419b | 5111 | ret = ss->css_online(css); |
ae7f164a | 5112 | if (!ret) { |
eb95419b | 5113 | css->flags |= CSS_ONLINE; |
aec25020 | 5114 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5115 | |
5116 | atomic_inc(&css->online_cnt); | |
5117 | if (css->parent) | |
5118 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5119 | } |
b1929db4 | 5120 | return ret; |
a31f2d3f TH |
5121 | } |
5122 | ||
2a4ac633 | 5123 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5124 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5125 | { |
623f926b | 5126 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5127 | |
5128 | lockdep_assert_held(&cgroup_mutex); | |
5129 | ||
5130 | if (!(css->flags & CSS_ONLINE)) | |
5131 | return; | |
5132 | ||
fa06235b VD |
5133 | if (ss->css_reset) |
5134 | ss->css_reset(css); | |
5135 | ||
d7eeac19 | 5136 | if (ss->css_offline) |
eb95419b | 5137 | ss->css_offline(css); |
a31f2d3f | 5138 | |
eb95419b | 5139 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5140 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5141 | |
5142 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5143 | } |
5144 | ||
c81c925a | 5145 | /** |
6cd0f5bb | 5146 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5147 | * @cgrp: the cgroup new css will be associated with |
5148 | * @ss: the subsys of new css | |
5149 | * | |
5150 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5151 | * css is online and installed in @cgrp. This function doesn't create the |
5152 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5153 | */ |
6cd0f5bb TH |
5154 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5155 | struct cgroup_subsys *ss) | |
c81c925a | 5156 | { |
d51f39b0 | 5157 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5158 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5159 | struct cgroup_subsys_state *css; |
5160 | int err; | |
5161 | ||
c81c925a TH |
5162 | lockdep_assert_held(&cgroup_mutex); |
5163 | ||
1fed1b2e | 5164 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
5165 | if (!css) |
5166 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 5167 | if (IS_ERR(css)) |
6cd0f5bb | 5168 | return css; |
c81c925a | 5169 | |
ddfcadab | 5170 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5171 | |
2aad2a86 | 5172 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5173 | if (err) |
3eb59ec6 | 5174 | goto err_free_css; |
c81c925a | 5175 | |
cf780b7d | 5176 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 5177 | if (err < 0) |
b00c52da | 5178 | goto err_free_css; |
15a4c835 | 5179 | css->id = err; |
c81c925a | 5180 | |
15a4c835 | 5181 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5182 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5183 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5184 | |
5185 | err = online_css(css); | |
5186 | if (err) | |
1fed1b2e | 5187 | goto err_list_del; |
94419627 | 5188 | |
c81c925a | 5189 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5190 | cgroup_parent(parent)) { |
ed3d261b | 5191 | 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 | 5192 | current->comm, current->pid, ss->name); |
c81c925a | 5193 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5194 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5195 | ss->warned_broken_hierarchy = true; |
5196 | } | |
5197 | ||
6cd0f5bb | 5198 | return css; |
c81c925a | 5199 | |
1fed1b2e TH |
5200 | err_list_del: |
5201 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 5202 | err_free_css: |
a2bed820 | 5203 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 5204 | return ERR_PTR(err); |
c81c925a TH |
5205 | } |
5206 | ||
a5bca215 | 5207 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 5208 | { |
a5bca215 | 5209 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
5210 | struct cgroup *cgrp, *tcgrp; |
5211 | int level = parent->level + 1; | |
03970d3c | 5212 | int ret; |
ddbcc7e8 | 5213 | |
0a950f65 | 5214 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
5215 | cgrp = kzalloc(sizeof(*cgrp) + |
5216 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
5217 | if (!cgrp) |
5218 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5219 | |
2aad2a86 | 5220 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5221 | if (ret) |
5222 | goto out_free_cgrp; | |
5223 | ||
0ab02ca8 LZ |
5224 | /* |
5225 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5226 | * a half-baked cgroup. | |
5227 | */ | |
cf780b7d | 5228 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5229 | if (cgrp->id < 0) { |
ba0f4d76 | 5230 | ret = -ENOMEM; |
9d755d33 | 5231 | goto out_cancel_ref; |
976c06bc TH |
5232 | } |
5233 | ||
cc31edce | 5234 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5235 | |
9d800df1 | 5236 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5237 | cgrp->root = root; |
b11cfb58 TH |
5238 | cgrp->level = level; |
5239 | ||
5240 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5241 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5242 | |
b6abdb0e LZ |
5243 | if (notify_on_release(parent)) |
5244 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5245 | ||
2260e7fc TH |
5246 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5247 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5248 | |
0cb51d71 | 5249 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5250 | |
4e139afc | 5251 | /* allocation complete, commit to creation */ |
d5c419b6 | 5252 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5253 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5254 | cgroup_get(parent); |
415cf07a | 5255 | |
0d80255e TH |
5256 | /* |
5257 | * @cgrp is now fully operational. If something fails after this | |
5258 | * point, it'll be released via the normal destruction path. | |
5259 | */ | |
6fa4918d | 5260 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5261 | |
bd53d617 TH |
5262 | /* |
5263 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5264 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5265 | */ |
03970d3c | 5266 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5267 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5268 | |
5269 | cgroup_propagate_control(cgrp); | |
5270 | ||
5271 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5272 | ret = cgroup_apply_control_enable(cgrp); | |
5273 | if (ret) | |
5274 | goto out_destroy; | |
2bd59d48 | 5275 | |
a5bca215 TH |
5276 | return cgrp; |
5277 | ||
5278 | out_cancel_ref: | |
5279 | percpu_ref_exit(&cgrp->self.refcnt); | |
5280 | out_free_cgrp: | |
5281 | kfree(cgrp); | |
5282 | return ERR_PTR(ret); | |
5283 | out_destroy: | |
5284 | cgroup_destroy_locked(cgrp); | |
5285 | return ERR_PTR(ret); | |
5286 | } | |
5287 | ||
5288 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5289 | umode_t mode) | |
5290 | { | |
5291 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5292 | struct kernfs_node *kn; |
03970d3c | 5293 | int ret; |
a5bca215 TH |
5294 | |
5295 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5296 | if (strchr(name, '\n')) | |
5297 | return -EINVAL; | |
5298 | ||
945ba199 | 5299 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5300 | if (!parent) |
5301 | return -ENODEV; | |
5302 | ||
5303 | cgrp = cgroup_create(parent); | |
5304 | if (IS_ERR(cgrp)) { | |
5305 | ret = PTR_ERR(cgrp); | |
5306 | goto out_unlock; | |
5307 | } | |
5308 | ||
195e9b6c TH |
5309 | /* create the directory */ |
5310 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5311 | if (IS_ERR(kn)) { | |
5312 | ret = PTR_ERR(kn); | |
5313 | goto out_destroy; | |
5314 | } | |
5315 | cgrp->kn = kn; | |
5316 | ||
5317 | /* | |
5318 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5319 | * that @cgrp->kn is always accessible. | |
5320 | */ | |
5321 | kernfs_get(kn); | |
5322 | ||
5323 | ret = cgroup_kn_set_ugid(kn); | |
5324 | if (ret) | |
5325 | goto out_destroy; | |
5326 | ||
334c3679 | 5327 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5328 | if (ret) |
5329 | goto out_destroy; | |
5330 | ||
03970d3c TH |
5331 | ret = cgroup_apply_control_enable(cgrp); |
5332 | if (ret) | |
5333 | goto out_destroy; | |
195e9b6c TH |
5334 | |
5335 | /* let's create and online css's */ | |
2bd59d48 | 5336 | kernfs_activate(kn); |
ddbcc7e8 | 5337 | |
ba0f4d76 TH |
5338 | ret = 0; |
5339 | goto out_unlock; | |
ddbcc7e8 | 5340 | |
a5bca215 TH |
5341 | out_destroy: |
5342 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5343 | out_unlock: |
a9746d8d | 5344 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5345 | return ret; |
ddbcc7e8 PM |
5346 | } |
5347 | ||
223dbc38 TH |
5348 | /* |
5349 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5350 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5351 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5352 | */ |
5353 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5354 | { |
223dbc38 TH |
5355 | struct cgroup_subsys_state *css = |
5356 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5357 | |
f20104de | 5358 | mutex_lock(&cgroup_mutex); |
09a503ea | 5359 | |
aa226ff4 TH |
5360 | do { |
5361 | offline_css(css); | |
5362 | css_put(css); | |
5363 | /* @css can't go away while we're holding cgroup_mutex */ | |
5364 | css = css->parent; | |
5365 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5366 | ||
5367 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5368 | } |
5369 | ||
223dbc38 TH |
5370 | /* css kill confirmation processing requires process context, bounce */ |
5371 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5372 | { |
5373 | struct cgroup_subsys_state *css = | |
5374 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5375 | ||
aa226ff4 TH |
5376 | if (atomic_dec_and_test(&css->online_cnt)) { |
5377 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5378 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5379 | } | |
d3daf28d TH |
5380 | } |
5381 | ||
f392e51c TH |
5382 | /** |
5383 | * kill_css - destroy a css | |
5384 | * @css: css to destroy | |
5385 | * | |
5386 | * This function initiates destruction of @css by removing cgroup interface | |
5387 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5388 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5389 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5390 | */ |
5391 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5392 | { |
01f6474c | 5393 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5394 | |
2bd59d48 TH |
5395 | /* |
5396 | * This must happen before css is disassociated with its cgroup. | |
5397 | * See seq_css() for details. | |
5398 | */ | |
334c3679 | 5399 | css_clear_dir(css); |
3c14f8b4 | 5400 | |
edae0c33 TH |
5401 | /* |
5402 | * Killing would put the base ref, but we need to keep it alive | |
5403 | * until after ->css_offline(). | |
5404 | */ | |
5405 | css_get(css); | |
5406 | ||
5407 | /* | |
5408 | * cgroup core guarantees that, by the time ->css_offline() is | |
5409 | * invoked, no new css reference will be given out via | |
ec903c0c | 5410 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5411 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5412 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5413 | * | |
5414 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5415 | * css is confirmed to be seen as killed on all CPUs. | |
5416 | */ | |
5417 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5418 | } |
5419 | ||
5420 | /** | |
5421 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5422 | * @cgrp: cgroup to be destroyed | |
5423 | * | |
5424 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5425 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5426 | * guarantee that css_tryget_online() won't succeed by the time |
5427 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5428 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5429 | * |
5430 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5431 | * userland visible parts and start killing the percpu refcnts of | |
5432 | * css's. Set up so that the next stage will be kicked off once all | |
5433 | * the percpu refcnts are confirmed to be killed. | |
5434 | * | |
5435 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5436 | * rest of destruction. Once all cgroup references are gone, the | |
5437 | * cgroup is RCU-freed. | |
5438 | * | |
5439 | * This function implements s1. After this step, @cgrp is gone as far as | |
5440 | * the userland is concerned and a new cgroup with the same name may be | |
5441 | * created. As cgroup doesn't care about the names internally, this | |
5442 | * doesn't cause any problem. | |
5443 | */ | |
42809dd4 TH |
5444 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5445 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5446 | { |
2bd59d48 | 5447 | struct cgroup_subsys_state *css; |
2b021cbf | 5448 | struct cgrp_cset_link *link; |
1c6727af | 5449 | int ssid; |
ddbcc7e8 | 5450 | |
42809dd4 TH |
5451 | lockdep_assert_held(&cgroup_mutex); |
5452 | ||
91486f61 TH |
5453 | /* |
5454 | * Only migration can raise populated from zero and we're already | |
5455 | * holding cgroup_mutex. | |
5456 | */ | |
5457 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5458 | return -EBUSY; |
a043e3b2 | 5459 | |
bb78a92f | 5460 | /* |
d5c419b6 TH |
5461 | * Make sure there's no live children. We can't test emptiness of |
5462 | * ->self.children as dead children linger on it while being | |
5463 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5464 | */ |
f3d46500 | 5465 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5466 | return -EBUSY; |
5467 | ||
455050d2 | 5468 | /* |
2b021cbf TH |
5469 | * Mark @cgrp and the associated csets dead. The former prevents |
5470 | * further task migration and child creation by disabling | |
5471 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5472 | * the migration path. | |
455050d2 | 5473 | */ |
184faf32 | 5474 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5475 | |
82d6489d | 5476 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5477 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5478 | link->cset->dead = true; | |
82d6489d | 5479 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5480 | |
249f3468 | 5481 | /* initiate massacre of all css's */ |
1c6727af TH |
5482 | for_each_css(css, ssid, cgrp) |
5483 | kill_css(css); | |
455050d2 | 5484 | |
455050d2 | 5485 | /* |
01f6474c TH |
5486 | * Remove @cgrp directory along with the base files. @cgrp has an |
5487 | * extra ref on its kn. | |
f20104de | 5488 | */ |
01f6474c | 5489 | kernfs_remove(cgrp->kn); |
f20104de | 5490 | |
d51f39b0 | 5491 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5492 | |
249f3468 | 5493 | /* put the base reference */ |
9d755d33 | 5494 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5495 | |
ea15f8cc TH |
5496 | return 0; |
5497 | }; | |
5498 | ||
2bd59d48 | 5499 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5500 | { |
a9746d8d | 5501 | struct cgroup *cgrp; |
2bd59d48 | 5502 | int ret = 0; |
42809dd4 | 5503 | |
945ba199 | 5504 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5505 | if (!cgrp) |
5506 | return 0; | |
42809dd4 | 5507 | |
a9746d8d | 5508 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5509 | |
a9746d8d | 5510 | cgroup_kn_unlock(kn); |
42809dd4 | 5511 | return ret; |
8e3f6541 TH |
5512 | } |
5513 | ||
2bd59d48 TH |
5514 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5515 | .remount_fs = cgroup_remount, | |
5516 | .show_options = cgroup_show_options, | |
5517 | .mkdir = cgroup_mkdir, | |
5518 | .rmdir = cgroup_rmdir, | |
5519 | .rename = cgroup_rename, | |
4f41fc59 | 5520 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5521 | }; |
5522 | ||
15a4c835 | 5523 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5524 | { |
ddbcc7e8 | 5525 | struct cgroup_subsys_state *css; |
cfe36bde | 5526 | |
a5ae9899 | 5527 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5528 | |
648bb56d TH |
5529 | mutex_lock(&cgroup_mutex); |
5530 | ||
15a4c835 | 5531 | idr_init(&ss->css_idr); |
0adb0704 | 5532 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5533 | |
3dd06ffa TH |
5534 | /* Create the root cgroup state for this subsystem */ |
5535 | ss->root = &cgrp_dfl_root; | |
5536 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5537 | /* We don't handle early failures gracefully */ |
5538 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5539 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5540 | |
5541 | /* | |
5542 | * Root csses are never destroyed and we can't initialize | |
5543 | * percpu_ref during early init. Disable refcnting. | |
5544 | */ | |
5545 | css->flags |= CSS_NO_REF; | |
5546 | ||
15a4c835 | 5547 | if (early) { |
9395a450 | 5548 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5549 | css->id = 1; |
5550 | } else { | |
5551 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5552 | BUG_ON(css->id < 0); | |
5553 | } | |
ddbcc7e8 | 5554 | |
e8d55fde | 5555 | /* Update the init_css_set to contain a subsys |
817929ec | 5556 | * pointer to this state - since the subsystem is |
e8d55fde | 5557 | * newly registered, all tasks and hence the |
3dd06ffa | 5558 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5559 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5560 | |
cb4a3167 AS |
5561 | have_fork_callback |= (bool)ss->fork << ss->id; |
5562 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5563 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5564 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5565 | |
e8d55fde LZ |
5566 | /* At system boot, before all subsystems have been |
5567 | * registered, no tasks have been forked, so we don't | |
5568 | * need to invoke fork callbacks here. */ | |
5569 | BUG_ON(!list_empty(&init_task.tasks)); | |
5570 | ||
ae7f164a | 5571 | BUG_ON(online_css(css)); |
a8638030 | 5572 | |
cf5d5941 BB |
5573 | mutex_unlock(&cgroup_mutex); |
5574 | } | |
cf5d5941 | 5575 | |
ddbcc7e8 | 5576 | /** |
a043e3b2 LZ |
5577 | * cgroup_init_early - cgroup initialization at system boot |
5578 | * | |
5579 | * Initialize cgroups at system boot, and initialize any | |
5580 | * subsystems that request early init. | |
ddbcc7e8 PM |
5581 | */ |
5582 | int __init cgroup_init_early(void) | |
5583 | { | |
7b9a6ba5 | 5584 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5585 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5586 | int i; |
30159ec7 | 5587 | |
3dd06ffa | 5588 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5589 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5590 | ||
a4ea1cc9 | 5591 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5592 | |
3ed80a62 | 5593 | for_each_subsys(ss, i) { |
aec25020 | 5594 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5595 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5596 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5597 | ss->id, ss->name); |
073219e9 TH |
5598 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5599 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5600 | ||
aec25020 | 5601 | ss->id = i; |
073219e9 | 5602 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5603 | if (!ss->legacy_name) |
5604 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5605 | |
5606 | if (ss->early_init) | |
15a4c835 | 5607 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5608 | } |
5609 | return 0; | |
5610 | } | |
5611 | ||
6e5c8307 | 5612 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5613 | |
ddbcc7e8 | 5614 | /** |
a043e3b2 LZ |
5615 | * cgroup_init - cgroup initialization |
5616 | * | |
5617 | * Register cgroup filesystem and /proc file, and initialize | |
5618 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5619 | */ |
5620 | int __init cgroup_init(void) | |
5621 | { | |
30159ec7 | 5622 | struct cgroup_subsys *ss; |
035f4f51 | 5623 | int ssid; |
ddbcc7e8 | 5624 | |
6e5c8307 | 5625 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5626 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5627 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5628 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5629 | |
3942a9bd PZ |
5630 | /* |
5631 | * The latency of the synchronize_sched() is too high for cgroups, | |
5632 | * avoid it at the cost of forcing all readers into the slow path. | |
5633 | */ | |
5634 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
5635 | ||
a79a908f AK |
5636 | get_user_ns(init_cgroup_ns.user_ns); |
5637 | ||
54e7b4eb | 5638 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5639 | |
2378d8b8 TH |
5640 | /* |
5641 | * Add init_css_set to the hash table so that dfl_root can link to | |
5642 | * it during init. | |
5643 | */ | |
5644 | hash_add(css_set_table, &init_css_set.hlist, | |
5645 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5646 | |
3dd06ffa | 5647 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5648 | |
54e7b4eb TH |
5649 | mutex_unlock(&cgroup_mutex); |
5650 | ||
172a2c06 | 5651 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5652 | if (ss->early_init) { |
5653 | struct cgroup_subsys_state *css = | |
5654 | init_css_set.subsys[ss->id]; | |
5655 | ||
5656 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5657 | GFP_KERNEL); | |
5658 | BUG_ON(css->id < 0); | |
5659 | } else { | |
5660 | cgroup_init_subsys(ss, false); | |
5661 | } | |
172a2c06 | 5662 | |
2d8f243a TH |
5663 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5664 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5665 | |
5666 | /* | |
c731ae1d LZ |
5667 | * Setting dfl_root subsys_mask needs to consider the |
5668 | * disabled flag and cftype registration needs kmalloc, | |
5669 | * both of which aren't available during early_init. | |
172a2c06 | 5670 | */ |
a3e72739 TH |
5671 | if (cgroup_disable_mask & (1 << ssid)) { |
5672 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5673 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5674 | ss->name); | |
a8ddc821 | 5675 | continue; |
a3e72739 | 5676 | } |
a8ddc821 | 5677 | |
223ffb29 JW |
5678 | if (cgroup_ssid_no_v1(ssid)) |
5679 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5680 | ss->name); | |
5681 | ||
a8ddc821 TH |
5682 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5683 | ||
f6d635ad TH |
5684 | if (ss->implicit_on_dfl) |
5685 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5686 | else if (!ss->dfl_cftypes) | |
a7165264 | 5687 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5688 | |
a8ddc821 TH |
5689 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5690 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5691 | } else { | |
5692 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5693 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5694 | } |
295458e6 VD |
5695 | |
5696 | if (ss->bind) | |
5697 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5698 | } |
5699 | ||
2378d8b8 TH |
5700 | /* init_css_set.subsys[] has been updated, re-hash */ |
5701 | hash_del(&init_css_set.hlist); | |
5702 | hash_add(css_set_table, &init_css_set.hlist, | |
5703 | css_set_hash(init_css_set.subsys)); | |
5704 | ||
035f4f51 TH |
5705 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5706 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5707 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5708 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5709 | |
2bd59d48 | 5710 | return 0; |
ddbcc7e8 | 5711 | } |
b4f48b63 | 5712 | |
e5fca243 TH |
5713 | static int __init cgroup_wq_init(void) |
5714 | { | |
5715 | /* | |
5716 | * There isn't much point in executing destruction path in | |
5717 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5718 | * Use 1 for @max_active. |
e5fca243 TH |
5719 | * |
5720 | * We would prefer to do this in cgroup_init() above, but that | |
5721 | * is called before init_workqueues(): so leave this until after. | |
5722 | */ | |
1a11533f | 5723 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5724 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5725 | |
5726 | /* | |
5727 | * Used to destroy pidlists and separate to serve as flush domain. | |
5728 | * Cap @max_active to 1 too. | |
5729 | */ | |
5730 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5731 | 0, 1); | |
5732 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5733 | ||
e5fca243 TH |
5734 | return 0; |
5735 | } | |
5736 | core_initcall(cgroup_wq_init); | |
5737 | ||
a424316c PM |
5738 | /* |
5739 | * proc_cgroup_show() | |
5740 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5741 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5742 | */ |
006f4ac4 ZL |
5743 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5744 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5745 | { |
e61734c5 | 5746 | char *buf, *path; |
a424316c | 5747 | int retval; |
3dd06ffa | 5748 | struct cgroup_root *root; |
a424316c PM |
5749 | |
5750 | retval = -ENOMEM; | |
e61734c5 | 5751 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5752 | if (!buf) |
5753 | goto out; | |
5754 | ||
a424316c | 5755 | mutex_lock(&cgroup_mutex); |
82d6489d | 5756 | spin_lock_irq(&css_set_lock); |
a424316c | 5757 | |
985ed670 | 5758 | for_each_root(root) { |
a424316c | 5759 | struct cgroup_subsys *ss; |
bd89aabc | 5760 | struct cgroup *cgrp; |
b85d2040 | 5761 | int ssid, count = 0; |
a424316c | 5762 | |
a7165264 | 5763 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5764 | continue; |
5765 | ||
2c6ab6d2 | 5766 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5767 | if (root != &cgrp_dfl_root) |
5768 | for_each_subsys(ss, ssid) | |
5769 | if (root->subsys_mask & (1 << ssid)) | |
5770 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5771 | ss->legacy_name); |
c6d57f33 PM |
5772 | if (strlen(root->name)) |
5773 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5774 | root->name); | |
a424316c | 5775 | seq_putc(m, ':'); |
2e91fa7f | 5776 | |
7717f7ba | 5777 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5778 | |
5779 | /* | |
5780 | * On traditional hierarchies, all zombie tasks show up as | |
5781 | * belonging to the root cgroup. On the default hierarchy, | |
5782 | * while a zombie doesn't show up in "cgroup.procs" and | |
5783 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5784 | * reporting the cgroup it belonged to before exiting. If | |
5785 | * the cgroup is removed before the zombie is reaped, | |
5786 | * " (deleted)" is appended to the cgroup path. | |
5787 | */ | |
5788 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
a79a908f AK |
5789 | path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
5790 | current->nsproxy->cgroup_ns); | |
2e91fa7f TH |
5791 | if (!path) { |
5792 | retval = -ENAMETOOLONG; | |
5793 | goto out_unlock; | |
5794 | } | |
5795 | } else { | |
5796 | path = "/"; | |
e61734c5 | 5797 | } |
2e91fa7f | 5798 | |
e61734c5 | 5799 | seq_puts(m, path); |
2e91fa7f TH |
5800 | |
5801 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5802 | seq_puts(m, " (deleted)\n"); | |
5803 | else | |
5804 | seq_putc(m, '\n'); | |
a424316c PM |
5805 | } |
5806 | ||
006f4ac4 | 5807 | retval = 0; |
a424316c | 5808 | out_unlock: |
82d6489d | 5809 | spin_unlock_irq(&css_set_lock); |
a424316c | 5810 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5811 | kfree(buf); |
5812 | out: | |
5813 | return retval; | |
5814 | } | |
5815 | ||
a424316c PM |
5816 | /* Display information about each subsystem and each hierarchy */ |
5817 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5818 | { | |
30159ec7 | 5819 | struct cgroup_subsys *ss; |
a424316c | 5820 | int i; |
a424316c | 5821 | |
8bab8dde | 5822 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5823 | /* |
5824 | * ideally we don't want subsystems moving around while we do this. | |
5825 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5826 | * subsys/hierarchy state. | |
5827 | */ | |
a424316c | 5828 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5829 | |
5830 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5831 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5832 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5833 | atomic_read(&ss->root->nr_cgrps), |
5834 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5835 | |
a424316c PM |
5836 | mutex_unlock(&cgroup_mutex); |
5837 | return 0; | |
5838 | } | |
5839 | ||
5840 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5841 | { | |
9dce07f1 | 5842 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5843 | } |
5844 | ||
828c0950 | 5845 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5846 | .open = cgroupstats_open, |
5847 | .read = seq_read, | |
5848 | .llseek = seq_lseek, | |
5849 | .release = single_release, | |
5850 | }; | |
5851 | ||
b4f48b63 | 5852 | /** |
eaf797ab | 5853 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5854 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5855 | * |
eaf797ab TH |
5856 | * A task is associated with the init_css_set until cgroup_post_fork() |
5857 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5858 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5859 | */ |
5860 | void cgroup_fork(struct task_struct *child) | |
5861 | { | |
eaf797ab | 5862 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5863 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5864 | } |
5865 | ||
7e47682e AS |
5866 | /** |
5867 | * cgroup_can_fork - called on a new task before the process is exposed | |
5868 | * @child: the task in question. | |
5869 | * | |
5870 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5871 | * returns an error, the fork aborts with that error code. This allows for | |
5872 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5873 | */ | |
b53202e6 | 5874 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5875 | { |
5876 | struct cgroup_subsys *ss; | |
5877 | int i, j, ret; | |
5878 | ||
b4e0eeaf | 5879 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5880 | ret = ss->can_fork(child); |
7e47682e AS |
5881 | if (ret) |
5882 | goto out_revert; | |
b4e0eeaf | 5883 | } while_each_subsys_mask(); |
7e47682e AS |
5884 | |
5885 | return 0; | |
5886 | ||
5887 | out_revert: | |
5888 | for_each_subsys(ss, j) { | |
5889 | if (j >= i) | |
5890 | break; | |
5891 | if (ss->cancel_fork) | |
b53202e6 | 5892 | ss->cancel_fork(child); |
7e47682e AS |
5893 | } |
5894 | ||
5895 | return ret; | |
5896 | } | |
5897 | ||
5898 | /** | |
5899 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5900 | * @child: the task in question | |
5901 | * | |
5902 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5903 | * cgroup_can_fork() succeded. | |
5904 | */ | |
b53202e6 | 5905 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5906 | { |
5907 | struct cgroup_subsys *ss; | |
5908 | int i; | |
5909 | ||
5910 | for_each_subsys(ss, i) | |
5911 | if (ss->cancel_fork) | |
b53202e6 | 5912 | ss->cancel_fork(child); |
7e47682e AS |
5913 | } |
5914 | ||
817929ec | 5915 | /** |
a043e3b2 LZ |
5916 | * cgroup_post_fork - called on a new task after adding it to the task list |
5917 | * @child: the task in question | |
5918 | * | |
5edee61e TH |
5919 | * Adds the task to the list running through its css_set if necessary and |
5920 | * call the subsystem fork() callbacks. Has to be after the task is | |
5921 | * visible on the task list in case we race with the first call to | |
0942eeee | 5922 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5923 | * list. |
a043e3b2 | 5924 | */ |
b53202e6 | 5925 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5926 | { |
30159ec7 | 5927 | struct cgroup_subsys *ss; |
5edee61e TH |
5928 | int i; |
5929 | ||
3ce3230a | 5930 | /* |
251f8c03 | 5931 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5932 | * function sets use_task_css_set_links before grabbing |
5933 | * tasklist_lock and we just went through tasklist_lock to add | |
5934 | * @child, it's guaranteed that either we see the set | |
5935 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5936 | * @child during its iteration. | |
5937 | * | |
5938 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5939 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5940 | * association is stable, and, on completion of the parent's |
5941 | * migration, @child is visible in the source of migration or | |
5942 | * already in the destination cgroup. This guarantee is necessary | |
5943 | * when implementing operations which need to migrate all tasks of | |
5944 | * a cgroup to another. | |
5945 | * | |
251f8c03 | 5946 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5947 | * will remain in init_css_set. This is safe because all tasks are |
5948 | * in the init_css_set before cg_links is enabled and there's no | |
5949 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5950 | */ |
817929ec | 5951 | if (use_task_css_set_links) { |
eaf797ab TH |
5952 | struct css_set *cset; |
5953 | ||
82d6489d | 5954 | spin_lock_irq(&css_set_lock); |
0e1d768f | 5955 | cset = task_css_set(current); |
eaf797ab | 5956 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5957 | get_css_set(cset); |
f6d7d049 | 5958 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5959 | } |
82d6489d | 5960 | spin_unlock_irq(&css_set_lock); |
817929ec | 5961 | } |
5edee61e TH |
5962 | |
5963 | /* | |
5964 | * Call ss->fork(). This must happen after @child is linked on | |
5965 | * css_set; otherwise, @child might change state between ->fork() | |
5966 | * and addition to css_set. | |
5967 | */ | |
b4e0eeaf | 5968 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5969 | ss->fork(child); |
b4e0eeaf | 5970 | } while_each_subsys_mask(); |
817929ec | 5971 | } |
5edee61e | 5972 | |
b4f48b63 PM |
5973 | /** |
5974 | * cgroup_exit - detach cgroup from exiting task | |
5975 | * @tsk: pointer to task_struct of exiting process | |
5976 | * | |
5977 | * Description: Detach cgroup from @tsk and release it. | |
5978 | * | |
5979 | * Note that cgroups marked notify_on_release force every task in | |
5980 | * them to take the global cgroup_mutex mutex when exiting. | |
5981 | * This could impact scaling on very large systems. Be reluctant to | |
5982 | * use notify_on_release cgroups where very high task exit scaling | |
5983 | * is required on large systems. | |
5984 | * | |
0e1d768f TH |
5985 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5986 | * call cgroup_exit() while the task is still competent to handle | |
5987 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5988 | * each hierarchy for the remainder of its exit. No need to bother with | |
5989 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5990 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5991 | */ |
1ec41830 | 5992 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5993 | { |
30159ec7 | 5994 | struct cgroup_subsys *ss; |
5abb8855 | 5995 | struct css_set *cset; |
d41d5a01 | 5996 | int i; |
817929ec PM |
5997 | |
5998 | /* | |
0e1d768f | 5999 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 6000 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 6001 | */ |
0de0942d TH |
6002 | cset = task_css_set(tsk); |
6003 | ||
817929ec | 6004 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 6005 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 6006 | css_set_move_task(tsk, cset, NULL, false); |
82d6489d | 6007 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
6008 | } else { |
6009 | get_css_set(cset); | |
817929ec PM |
6010 | } |
6011 | ||
cb4a3167 | 6012 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 6013 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 6014 | ss->exit(tsk); |
b4e0eeaf | 6015 | } while_each_subsys_mask(); |
2e91fa7f | 6016 | } |
30159ec7 | 6017 | |
2e91fa7f TH |
6018 | void cgroup_free(struct task_struct *task) |
6019 | { | |
6020 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
6021 | struct cgroup_subsys *ss; |
6022 | int ssid; | |
6023 | ||
b4e0eeaf | 6024 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 6025 | ss->free(task); |
b4e0eeaf | 6026 | } while_each_subsys_mask(); |
d41d5a01 | 6027 | |
2e91fa7f | 6028 | put_css_set(cset); |
b4f48b63 | 6029 | } |
697f4161 | 6030 | |
bd89aabc | 6031 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 6032 | { |
27bd4dbb | 6033 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
6034 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
6035 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
6036 | } |
6037 | ||
81a6a5cd PM |
6038 | /* |
6039 | * Notify userspace when a cgroup is released, by running the | |
6040 | * configured release agent with the name of the cgroup (path | |
6041 | * relative to the root of cgroup file system) as the argument. | |
6042 | * | |
6043 | * Most likely, this user command will try to rmdir this cgroup. | |
6044 | * | |
6045 | * This races with the possibility that some other task will be | |
6046 | * attached to this cgroup before it is removed, or that some other | |
6047 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
6048 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
6049 | * unused, and this cgroup will be reprieved from its death sentence, | |
6050 | * to continue to serve a useful existence. Next time it's released, | |
6051 | * we will get notified again, if it still has 'notify_on_release' set. | |
6052 | * | |
6053 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
6054 | * means only wait until the task is successfully execve()'d. The | |
6055 | * separate release agent task is forked by call_usermodehelper(), | |
6056 | * then control in this thread returns here, without waiting for the | |
6057 | * release agent task. We don't bother to wait because the caller of | |
6058 | * this routine has no use for the exit status of the release agent | |
6059 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 6060 | */ |
81a6a5cd PM |
6061 | static void cgroup_release_agent(struct work_struct *work) |
6062 | { | |
971ff493 ZL |
6063 | struct cgroup *cgrp = |
6064 | container_of(work, struct cgroup, release_agent_work); | |
6065 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
6066 | char *argv[3], *envp[3]; | |
6067 | ||
81a6a5cd | 6068 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
6069 | |
6070 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
6071 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
6072 | if (!pathbuf || !agentbuf) | |
6073 | goto out; | |
6074 | ||
82d6489d | 6075 | spin_lock_irq(&css_set_lock); |
a79a908f | 6076 | path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); |
82d6489d | 6077 | spin_unlock_irq(&css_set_lock); |
971ff493 ZL |
6078 | if (!path) |
6079 | goto out; | |
6080 | ||
6081 | argv[0] = agentbuf; | |
6082 | argv[1] = path; | |
6083 | argv[2] = NULL; | |
6084 | ||
6085 | /* minimal command environment */ | |
6086 | envp[0] = "HOME=/"; | |
6087 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
6088 | envp[2] = NULL; | |
6089 | ||
81a6a5cd | 6090 | mutex_unlock(&cgroup_mutex); |
971ff493 | 6091 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 6092 | goto out_free; |
971ff493 | 6093 | out: |
81a6a5cd | 6094 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 6095 | out_free: |
971ff493 ZL |
6096 | kfree(agentbuf); |
6097 | kfree(pathbuf); | |
81a6a5cd | 6098 | } |
8bab8dde PM |
6099 | |
6100 | static int __init cgroup_disable(char *str) | |
6101 | { | |
30159ec7 | 6102 | struct cgroup_subsys *ss; |
8bab8dde | 6103 | char *token; |
30159ec7 | 6104 | int i; |
8bab8dde PM |
6105 | |
6106 | while ((token = strsep(&str, ",")) != NULL) { | |
6107 | if (!*token) | |
6108 | continue; | |
be45c900 | 6109 | |
3ed80a62 | 6110 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6111 | if (strcmp(token, ss->name) && |
6112 | strcmp(token, ss->legacy_name)) | |
6113 | continue; | |
a3e72739 | 6114 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
6115 | } |
6116 | } | |
6117 | return 1; | |
6118 | } | |
6119 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6120 | |
223ffb29 JW |
6121 | static int __init cgroup_no_v1(char *str) |
6122 | { | |
6123 | struct cgroup_subsys *ss; | |
6124 | char *token; | |
6125 | int i; | |
6126 | ||
6127 | while ((token = strsep(&str, ",")) != NULL) { | |
6128 | if (!*token) | |
6129 | continue; | |
6130 | ||
6131 | if (!strcmp(token, "all")) { | |
6e5c8307 | 6132 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
6133 | break; |
6134 | } | |
6135 | ||
6136 | for_each_subsys(ss, i) { | |
6137 | if (strcmp(token, ss->name) && | |
6138 | strcmp(token, ss->legacy_name)) | |
6139 | continue; | |
6140 | ||
6141 | cgroup_no_v1_mask |= 1 << i; | |
6142 | } | |
6143 | } | |
6144 | return 1; | |
6145 | } | |
6146 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
6147 | ||
b77d7b60 | 6148 | /** |
ec903c0c | 6149 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6150 | * @dentry: directory dentry of interest |
6151 | * @ss: subsystem of interest | |
b77d7b60 | 6152 | * |
5a17f543 TH |
6153 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6154 | * to get the corresponding css and return it. If such css doesn't exist | |
6155 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6156 | */ |
ec903c0c TH |
6157 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6158 | struct cgroup_subsys *ss) | |
e5d1367f | 6159 | { |
2bd59d48 | 6160 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6161 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6162 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6163 | struct cgroup *cgrp; |
e5d1367f | 6164 | |
35cf0836 | 6165 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6166 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6167 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6168 | return ERR_PTR(-EBADF); |
6169 | ||
5a17f543 TH |
6170 | rcu_read_lock(); |
6171 | ||
2bd59d48 TH |
6172 | /* |
6173 | * This path doesn't originate from kernfs and @kn could already | |
6174 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6175 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
6176 | */ |
6177 | cgrp = rcu_dereference(kn->priv); | |
6178 | if (cgrp) | |
6179 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6180 | |
ec903c0c | 6181 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6182 | css = ERR_PTR(-ENOENT); |
6183 | ||
6184 | rcu_read_unlock(); | |
6185 | return css; | |
e5d1367f | 6186 | } |
e5d1367f | 6187 | |
1cb650b9 LZ |
6188 | /** |
6189 | * css_from_id - lookup css by id | |
6190 | * @id: the cgroup id | |
6191 | * @ss: cgroup subsys to be looked into | |
6192 | * | |
6193 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6194 | * Should be called under rcu_read_lock(). | |
6195 | */ | |
6196 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6197 | { | |
6fa4918d | 6198 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 6199 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
6200 | } |
6201 | ||
16af4396 TH |
6202 | /** |
6203 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6204 | * @path: path on the default hierarchy | |
6205 | * | |
6206 | * Find the cgroup at @path on the default hierarchy, increment its | |
6207 | * reference count and return it. Returns pointer to the found cgroup on | |
6208 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6209 | * if @path points to a non-directory. | |
6210 | */ | |
6211 | struct cgroup *cgroup_get_from_path(const char *path) | |
6212 | { | |
6213 | struct kernfs_node *kn; | |
6214 | struct cgroup *cgrp; | |
6215 | ||
6216 | mutex_lock(&cgroup_mutex); | |
6217 | ||
6218 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6219 | if (kn) { | |
6220 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6221 | cgrp = kn->priv; | |
6222 | cgroup_get(cgrp); | |
6223 | } else { | |
6224 | cgrp = ERR_PTR(-ENOTDIR); | |
6225 | } | |
6226 | kernfs_put(kn); | |
6227 | } else { | |
6228 | cgrp = ERR_PTR(-ENOENT); | |
6229 | } | |
6230 | ||
6231 | mutex_unlock(&cgroup_mutex); | |
6232 | return cgrp; | |
6233 | } | |
6234 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6235 | ||
1f3fe7eb MKL |
6236 | /** |
6237 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
6238 | * @fd: fd obtained by open(cgroup2_dir) | |
6239 | * | |
6240 | * Find the cgroup from a fd which should be obtained | |
6241 | * by opening a cgroup directory. Returns a pointer to the | |
6242 | * cgroup on success. ERR_PTR is returned if the cgroup | |
6243 | * cannot be found. | |
6244 | */ | |
6245 | struct cgroup *cgroup_get_from_fd(int fd) | |
6246 | { | |
6247 | struct cgroup_subsys_state *css; | |
6248 | struct cgroup *cgrp; | |
6249 | struct file *f; | |
6250 | ||
6251 | f = fget_raw(fd); | |
6252 | if (!f) | |
6253 | return ERR_PTR(-EBADF); | |
6254 | ||
6255 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
6256 | fput(f); | |
6257 | if (IS_ERR(css)) | |
6258 | return ERR_CAST(css); | |
6259 | ||
6260 | cgrp = css->cgroup; | |
6261 | if (!cgroup_on_dfl(cgrp)) { | |
6262 | cgroup_put(cgrp); | |
6263 | return ERR_PTR(-EBADF); | |
6264 | } | |
6265 | ||
6266 | return cgrp; | |
6267 | } | |
6268 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
6269 | ||
bd1060a1 TH |
6270 | /* |
6271 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6272 | * definition in cgroup-defs.h. | |
6273 | */ | |
6274 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6275 | ||
6276 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6277 | ||
3fa4cc9c | 6278 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6279 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6280 | ||
6281 | void cgroup_sk_alloc_disable(void) | |
6282 | { | |
6283 | if (cgroup_sk_alloc_disabled) | |
6284 | return; | |
6285 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6286 | cgroup_sk_alloc_disabled = true; | |
6287 | } | |
6288 | ||
6289 | #else | |
6290 | ||
6291 | #define cgroup_sk_alloc_disabled false | |
6292 | ||
6293 | #endif | |
6294 | ||
6295 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6296 | { | |
6297 | if (cgroup_sk_alloc_disabled) | |
6298 | return; | |
6299 | ||
d979a39d JW |
6300 | /* Socket clone path */ |
6301 | if (skcd->val) { | |
6302 | cgroup_get(sock_cgroup_ptr(skcd)); | |
6303 | return; | |
6304 | } | |
6305 | ||
bd1060a1 TH |
6306 | rcu_read_lock(); |
6307 | ||
6308 | while (true) { | |
6309 | struct css_set *cset; | |
6310 | ||
6311 | cset = task_css_set(current); | |
6312 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6313 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6314 | break; | |
6315 | } | |
6316 | cpu_relax(); | |
6317 | } | |
6318 | ||
6319 | rcu_read_unlock(); | |
6320 | } | |
6321 | ||
6322 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6323 | { | |
6324 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6325 | } | |
6326 | ||
6327 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6328 | ||
a79a908f AK |
6329 | /* cgroup namespaces */ |
6330 | ||
d08311dd EB |
6331 | static struct ucounts *inc_cgroup_namespaces(struct user_namespace *ns) |
6332 | { | |
6333 | return inc_ucount(ns, current_euid(), UCOUNT_CGROUP_NAMESPACES); | |
6334 | } | |
6335 | ||
6336 | static void dec_cgroup_namespaces(struct ucounts *ucounts) | |
6337 | { | |
6338 | dec_ucount(ucounts, UCOUNT_CGROUP_NAMESPACES); | |
6339 | } | |
6340 | ||
a79a908f AK |
6341 | static struct cgroup_namespace *alloc_cgroup_ns(void) |
6342 | { | |
6343 | struct cgroup_namespace *new_ns; | |
6344 | int ret; | |
6345 | ||
6346 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); | |
6347 | if (!new_ns) | |
6348 | return ERR_PTR(-ENOMEM); | |
6349 | ret = ns_alloc_inum(&new_ns->ns); | |
6350 | if (ret) { | |
6351 | kfree(new_ns); | |
6352 | return ERR_PTR(ret); | |
6353 | } | |
6354 | atomic_set(&new_ns->count, 1); | |
6355 | new_ns->ns.ops = &cgroupns_operations; | |
6356 | return new_ns; | |
6357 | } | |
6358 | ||
6359 | void free_cgroup_ns(struct cgroup_namespace *ns) | |
6360 | { | |
6361 | put_css_set(ns->root_cset); | |
d08311dd | 6362 | dec_cgroup_namespaces(ns->ucounts); |
a79a908f AK |
6363 | put_user_ns(ns->user_ns); |
6364 | ns_free_inum(&ns->ns); | |
6365 | kfree(ns); | |
6366 | } | |
6367 | EXPORT_SYMBOL(free_cgroup_ns); | |
6368 | ||
6369 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, | |
6370 | struct user_namespace *user_ns, | |
6371 | struct cgroup_namespace *old_ns) | |
6372 | { | |
fa5ff8a1 | 6373 | struct cgroup_namespace *new_ns; |
d08311dd | 6374 | struct ucounts *ucounts; |
fa5ff8a1 | 6375 | struct css_set *cset; |
a79a908f AK |
6376 | |
6377 | BUG_ON(!old_ns); | |
6378 | ||
6379 | if (!(flags & CLONE_NEWCGROUP)) { | |
6380 | get_cgroup_ns(old_ns); | |
6381 | return old_ns; | |
6382 | } | |
6383 | ||
6384 | /* Allow only sysadmin to create cgroup namespace. */ | |
a79a908f | 6385 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
fa5ff8a1 | 6386 | return ERR_PTR(-EPERM); |
a79a908f | 6387 | |
d08311dd EB |
6388 | ucounts = inc_cgroup_namespaces(user_ns); |
6389 | if (!ucounts) | |
df75e774 | 6390 | return ERR_PTR(-ENOSPC); |
d08311dd | 6391 | |
7bd88308 | 6392 | /* It is not safe to take cgroup_mutex here */ |
82d6489d | 6393 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
6394 | cset = task_css_set(current); |
6395 | get_css_set(cset); | |
82d6489d | 6396 | spin_unlock_irq(&css_set_lock); |
a79a908f | 6397 | |
a79a908f | 6398 | new_ns = alloc_cgroup_ns(); |
d2202557 | 6399 | if (IS_ERR(new_ns)) { |
fa5ff8a1 | 6400 | put_css_set(cset); |
d08311dd | 6401 | dec_cgroup_namespaces(ucounts); |
fa5ff8a1 | 6402 | return new_ns; |
d2202557 | 6403 | } |
a79a908f AK |
6404 | |
6405 | new_ns->user_ns = get_user_ns(user_ns); | |
d08311dd | 6406 | new_ns->ucounts = ucounts; |
a79a908f AK |
6407 | new_ns->root_cset = cset; |
6408 | ||
6409 | return new_ns; | |
a79a908f AK |
6410 | } |
6411 | ||
6412 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) | |
6413 | { | |
6414 | return container_of(ns, struct cgroup_namespace, ns); | |
6415 | } | |
6416 | ||
a0530e08 | 6417 | static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
a79a908f | 6418 | { |
a0530e08 AK |
6419 | struct cgroup_namespace *cgroup_ns = to_cg_ns(ns); |
6420 | ||
6421 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) || | |
6422 | !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN)) | |
6423 | return -EPERM; | |
6424 | ||
6425 | /* Don't need to do anything if we are attaching to our own cgroupns. */ | |
6426 | if (cgroup_ns == nsproxy->cgroup_ns) | |
6427 | return 0; | |
6428 | ||
6429 | get_cgroup_ns(cgroup_ns); | |
6430 | put_cgroup_ns(nsproxy->cgroup_ns); | |
6431 | nsproxy->cgroup_ns = cgroup_ns; | |
6432 | ||
6433 | return 0; | |
a79a908f AK |
6434 | } |
6435 | ||
6436 | static struct ns_common *cgroupns_get(struct task_struct *task) | |
6437 | { | |
6438 | struct cgroup_namespace *ns = NULL; | |
6439 | struct nsproxy *nsproxy; | |
6440 | ||
6441 | task_lock(task); | |
6442 | nsproxy = task->nsproxy; | |
6443 | if (nsproxy) { | |
6444 | ns = nsproxy->cgroup_ns; | |
6445 | get_cgroup_ns(ns); | |
6446 | } | |
6447 | task_unlock(task); | |
6448 | ||
6449 | return ns ? &ns->ns : NULL; | |
6450 | } | |
6451 | ||
6452 | static void cgroupns_put(struct ns_common *ns) | |
6453 | { | |
6454 | put_cgroup_ns(to_cg_ns(ns)); | |
6455 | } | |
6456 | ||
bcac25a5 AV |
6457 | static struct user_namespace *cgroupns_owner(struct ns_common *ns) |
6458 | { | |
6459 | return to_cg_ns(ns)->user_ns; | |
6460 | } | |
6461 | ||
a79a908f AK |
6462 | const struct proc_ns_operations cgroupns_operations = { |
6463 | .name = "cgroup", | |
6464 | .type = CLONE_NEWCGROUP, | |
6465 | .get = cgroupns_get, | |
6466 | .put = cgroupns_put, | |
6467 | .install = cgroupns_install, | |
bcac25a5 | 6468 | .owner = cgroupns_owner, |
a79a908f AK |
6469 | }; |
6470 | ||
6471 | static __init int cgroup_namespaces_init(void) | |
6472 | { | |
6473 | return 0; | |
6474 | } | |
6475 | subsys_initcall(cgroup_namespaces_init); | |
6476 | ||
fe693435 | 6477 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6478 | static struct cgroup_subsys_state * |
6479 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6480 | { |
6481 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6482 | ||
6483 | if (!css) | |
6484 | return ERR_PTR(-ENOMEM); | |
6485 | ||
6486 | return css; | |
6487 | } | |
6488 | ||
eb95419b | 6489 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6490 | { |
eb95419b | 6491 | kfree(css); |
fe693435 PM |
6492 | } |
6493 | ||
182446d0 TH |
6494 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6495 | struct cftype *cft) | |
fe693435 | 6496 | { |
182446d0 | 6497 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6498 | } |
6499 | ||
182446d0 TH |
6500 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6501 | struct cftype *cft) | |
fe693435 PM |
6502 | { |
6503 | return (u64)(unsigned long)current->cgroups; | |
6504 | } | |
6505 | ||
182446d0 | 6506 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6507 | struct cftype *cft) |
fe693435 PM |
6508 | { |
6509 | u64 count; | |
6510 | ||
6511 | rcu_read_lock(); | |
a8ad805c | 6512 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6513 | rcu_read_unlock(); |
6514 | return count; | |
6515 | } | |
6516 | ||
2da8ca82 | 6517 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6518 | { |
69d0206c | 6519 | struct cgrp_cset_link *link; |
5abb8855 | 6520 | struct css_set *cset; |
e61734c5 TH |
6521 | char *name_buf; |
6522 | ||
6523 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6524 | if (!name_buf) | |
6525 | return -ENOMEM; | |
7717f7ba | 6526 | |
82d6489d | 6527 | spin_lock_irq(&css_set_lock); |
7717f7ba | 6528 | rcu_read_lock(); |
5abb8855 | 6529 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6530 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6531 | struct cgroup *c = link->cgrp; |
7717f7ba | 6532 | |
a2dd4247 | 6533 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6534 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6535 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6536 | } |
6537 | rcu_read_unlock(); | |
82d6489d | 6538 | spin_unlock_irq(&css_set_lock); |
e61734c5 | 6539 | kfree(name_buf); |
7717f7ba PM |
6540 | return 0; |
6541 | } | |
6542 | ||
6543 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6544 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6545 | { |
2da8ca82 | 6546 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6547 | struct cgrp_cset_link *link; |
7717f7ba | 6548 | |
82d6489d | 6549 | spin_lock_irq(&css_set_lock); |
182446d0 | 6550 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6551 | struct css_set *cset = link->cset; |
7717f7ba PM |
6552 | struct task_struct *task; |
6553 | int count = 0; | |
c7561128 | 6554 | |
5abb8855 | 6555 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6556 | |
5abb8855 | 6557 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6558 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6559 | goto overflow; | |
6560 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6561 | } | |
6562 | ||
6563 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6564 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6565 | goto overflow; | |
6566 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6567 | } |
c7561128 TH |
6568 | continue; |
6569 | overflow: | |
6570 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6571 | } |
82d6489d | 6572 | spin_unlock_irq(&css_set_lock); |
7717f7ba PM |
6573 | return 0; |
6574 | } | |
6575 | ||
182446d0 | 6576 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6577 | { |
27bd4dbb | 6578 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6579 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6580 | } |
6581 | ||
6582 | static struct cftype debug_files[] = { | |
fe693435 PM |
6583 | { |
6584 | .name = "taskcount", | |
6585 | .read_u64 = debug_taskcount_read, | |
6586 | }, | |
6587 | ||
6588 | { | |
6589 | .name = "current_css_set", | |
6590 | .read_u64 = current_css_set_read, | |
6591 | }, | |
6592 | ||
6593 | { | |
6594 | .name = "current_css_set_refcount", | |
6595 | .read_u64 = current_css_set_refcount_read, | |
6596 | }, | |
6597 | ||
7717f7ba PM |
6598 | { |
6599 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6600 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6601 | }, |
6602 | ||
6603 | { | |
6604 | .name = "cgroup_css_links", | |
2da8ca82 | 6605 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6606 | }, |
6607 | ||
fe693435 PM |
6608 | { |
6609 | .name = "releasable", | |
6610 | .read_u64 = releasable_read, | |
6611 | }, | |
fe693435 | 6612 | |
4baf6e33 TH |
6613 | { } /* terminate */ |
6614 | }; | |
fe693435 | 6615 | |
073219e9 | 6616 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6617 | .css_alloc = debug_css_alloc, |
6618 | .css_free = debug_css_free, | |
5577964e | 6619 | .legacy_cftypes = debug_files, |
fe693435 PM |
6620 | }; |
6621 | #endif /* CONFIG_CGROUP_DEBUG */ |