Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
60063497 | 60 | #include <linux/atomic.h> |
e93ad19d | 61 | #include <linux/cpuset.h> |
a79a908f AK |
62 | #include <linux/proc_ns.h> |
63 | #include <linux/nsproxy.h> | |
64 | #include <linux/proc_ns.h> | |
bd1060a1 | 65 | #include <net/sock.h> |
ddbcc7e8 | 66 | |
b1a21367 TH |
67 | /* |
68 | * pidlists linger the following amount before being destroyed. The goal | |
69 | * is avoiding frequent destruction in the middle of consecutive read calls | |
70 | * Expiring in the middle is a performance problem not a correctness one. | |
71 | * 1 sec should be enough. | |
72 | */ | |
73 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
74 | ||
8d7e6fb0 TH |
75 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
76 | MAX_CFTYPE_NAME + 2) | |
77 | ||
e25e2cbb TH |
78 | /* |
79 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
80 | * hierarchy must be performed while holding it. | |
81 | * | |
f0d9a5f1 | 82 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 83 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 84 | * |
0e1d768f TH |
85 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
86 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 87 | */ |
2219449a TH |
88 | #ifdef CONFIG_PROVE_RCU |
89 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 90 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 91 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 92 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 93 | #else |
81a6a5cd | 94 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 95 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
96 | #endif |
97 | ||
6fa4918d | 98 | /* |
15a4c835 TH |
99 | * Protects cgroup_idr and css_idr so that IDs can be released without |
100 | * grabbing cgroup_mutex. | |
6fa4918d TH |
101 | */ |
102 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
103 | ||
34c06254 TH |
104 | /* |
105 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
106 | * against file removal/re-creation across css hiding. | |
107 | */ | |
108 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
109 | ||
69e943b7 TH |
110 | /* |
111 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
112 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
113 | */ | |
114 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 115 | |
1ed13287 TH |
116 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
117 | ||
8353da1f | 118 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
119 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
120 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 121 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 122 | |
e5fca243 TH |
123 | /* |
124 | * cgroup destruction makes heavy use of work items and there can be a lot | |
125 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
126 | * destruction work items don't end up filling up max_active of system_wq | |
127 | * which may lead to deadlock. | |
128 | */ | |
129 | static struct workqueue_struct *cgroup_destroy_wq; | |
130 | ||
b1a21367 TH |
131 | /* |
132 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
133 | * separate workqueue as flush domain. | |
134 | */ | |
135 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
136 | ||
3ed80a62 | 137 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 138 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 139 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
140 | #include <linux/cgroup_subsys.h> |
141 | }; | |
073219e9 TH |
142 | #undef SUBSYS |
143 | ||
144 | /* array of cgroup subsystem names */ | |
145 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
146 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
147 | #include <linux/cgroup_subsys.h> |
148 | }; | |
073219e9 | 149 | #undef SUBSYS |
ddbcc7e8 | 150 | |
49d1dc4b TH |
151 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
152 | #define SUBSYS(_x) \ | |
153 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
154 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
155 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
156 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
157 | #include <linux/cgroup_subsys.h> | |
158 | #undef SUBSYS | |
159 | ||
160 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
161 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
162 | #include <linux/cgroup_subsys.h> | |
163 | }; | |
164 | #undef SUBSYS | |
165 | ||
166 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
167 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
168 | #include <linux/cgroup_subsys.h> | |
169 | }; | |
170 | #undef SUBSYS | |
171 | ||
ddbcc7e8 | 172 | /* |
3dd06ffa | 173 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
174 | * unattached - it never has more than a single cgroup, and all tasks are |
175 | * part of that cgroup. | |
ddbcc7e8 | 176 | */ |
a2dd4247 | 177 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 178 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 179 | |
a2dd4247 TH |
180 | /* |
181 | * The default hierarchy always exists but is hidden until mounted for the | |
182 | * first time. This is for backward compatibility. | |
183 | */ | |
a7165264 | 184 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 185 | |
223ffb29 | 186 | /* Controllers blocked by the commandline in v1 */ |
6e5c8307 | 187 | static u16 cgroup_no_v1_mask; |
223ffb29 | 188 | |
5533e011 | 189 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 190 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 191 | |
f6d635ad TH |
192 | /* some controllers are implicitly enabled on the default hierarchy */ |
193 | static unsigned long cgrp_dfl_implicit_ss_mask; | |
194 | ||
ddbcc7e8 PM |
195 | /* The list of hierarchy roots */ |
196 | ||
9871bf95 TH |
197 | static LIST_HEAD(cgroup_roots); |
198 | static int cgroup_root_count; | |
ddbcc7e8 | 199 | |
3417ae1f | 200 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 201 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 202 | |
794611a1 | 203 | /* |
0cb51d71 TH |
204 | * Assign a monotonically increasing serial number to csses. It guarantees |
205 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
206 | * Also, as csses are always appended to the parent's ->children list, it | |
207 | * guarantees that sibling csses are always sorted in the ascending serial | |
208 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 209 | */ |
0cb51d71 | 210 | static u64 css_serial_nr_next = 1; |
794611a1 | 211 | |
cb4a3167 AS |
212 | /* |
213 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
214 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
215 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 216 | */ |
6e5c8307 TH |
217 | static u16 have_fork_callback __read_mostly; |
218 | static u16 have_exit_callback __read_mostly; | |
219 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 220 | |
a79a908f AK |
221 | /* cgroup namespace for init task */ |
222 | struct cgroup_namespace init_cgroup_ns = { | |
223 | .count = { .counter = 2, }, | |
224 | .user_ns = &init_user_ns, | |
225 | .ns.ops = &cgroupns_operations, | |
226 | .ns.inum = PROC_CGROUP_INIT_INO, | |
227 | .root_cset = &init_css_set, | |
228 | }; | |
229 | ||
7e47682e | 230 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 231 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 232 | |
67e9c74b | 233 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
234 | static struct cftype cgroup_dfl_base_files[]; |
235 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 236 | |
6e5c8307 | 237 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
945ba199 | 238 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
334c3679 TH |
239 | static int cgroup_apply_control(struct cgroup *cgrp); |
240 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 241 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 242 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
243 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
244 | struct cgroup_subsys *ss); | |
9d755d33 | 245 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 246 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
247 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
248 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 249 | bool is_add); |
42809dd4 | 250 | |
fc5ed1e9 TH |
251 | /** |
252 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
253 | * @ssid: subsys ID of interest | |
254 | * | |
255 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
256 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
257 | * is slower static_key_enabled() based test indexed by @ssid. | |
258 | */ | |
259 | static bool cgroup_ssid_enabled(int ssid) | |
260 | { | |
cfe02a8a AB |
261 | if (CGROUP_SUBSYS_COUNT == 0) |
262 | return false; | |
263 | ||
fc5ed1e9 TH |
264 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
265 | } | |
266 | ||
223ffb29 JW |
267 | static bool cgroup_ssid_no_v1(int ssid) |
268 | { | |
269 | return cgroup_no_v1_mask & (1 << ssid); | |
270 | } | |
271 | ||
9e10a130 TH |
272 | /** |
273 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
274 | * @cgrp: the cgroup of interest | |
275 | * | |
276 | * The default hierarchy is the v2 interface of cgroup and this function | |
277 | * can be used to test whether a cgroup is on the default hierarchy for | |
278 | * cases where a subsystem should behave differnetly depending on the | |
279 | * interface version. | |
280 | * | |
281 | * The set of behaviors which change on the default hierarchy are still | |
282 | * being determined and the mount option is prefixed with __DEVEL__. | |
283 | * | |
284 | * List of changed behaviors: | |
285 | * | |
286 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
287 | * and "name" are disallowed. | |
288 | * | |
289 | * - When mounting an existing superblock, mount options should match. | |
290 | * | |
291 | * - Remount is disallowed. | |
292 | * | |
293 | * - rename(2) is disallowed. | |
294 | * | |
295 | * - "tasks" is removed. Everything should be at process granularity. Use | |
296 | * "cgroup.procs" instead. | |
297 | * | |
298 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
299 | * recycled inbetween reads. | |
300 | * | |
301 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
302 | * notification mechanism will be implemented. | |
303 | * | |
304 | * - "cgroup.clone_children" is removed. | |
305 | * | |
306 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
307 | * and its descendants contain no task; otherwise, 1. The file also | |
308 | * generates kernfs notification which can be monitored through poll and | |
309 | * [di]notify when the value of the file changes. | |
310 | * | |
311 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
312 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
313 | * moved to an ancestor. | |
314 | * | |
315 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
316 | * masks of ancestors. | |
317 | * | |
318 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
319 | * is not created. | |
320 | * | |
321 | * - blkcg: blk-throttle becomes properly hierarchical. | |
322 | * | |
323 | * - debug: disallowed on the default hierarchy. | |
324 | */ | |
325 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
326 | { | |
327 | return cgrp->root == &cgrp_dfl_root; | |
328 | } | |
329 | ||
6fa4918d TH |
330 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
331 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
332 | gfp_t gfp_mask) | |
333 | { | |
334 | int ret; | |
335 | ||
336 | idr_preload(gfp_mask); | |
54504e97 | 337 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 338 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 339 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
340 | idr_preload_end(); |
341 | return ret; | |
342 | } | |
343 | ||
344 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
345 | { | |
346 | void *ret; | |
347 | ||
54504e97 | 348 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 349 | ret = idr_replace(idr, ptr, id); |
54504e97 | 350 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
351 | return ret; |
352 | } | |
353 | ||
354 | static void cgroup_idr_remove(struct idr *idr, int id) | |
355 | { | |
54504e97 | 356 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 357 | idr_remove(idr, id); |
54504e97 | 358 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
359 | } |
360 | ||
d51f39b0 TH |
361 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
362 | { | |
363 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
364 | ||
365 | if (parent_css) | |
366 | return container_of(parent_css, struct cgroup, self); | |
367 | return NULL; | |
368 | } | |
369 | ||
5531dc91 TH |
370 | /* subsystems visibly enabled on a cgroup */ |
371 | static u16 cgroup_control(struct cgroup *cgrp) | |
372 | { | |
373 | struct cgroup *parent = cgroup_parent(cgrp); | |
374 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
375 | ||
376 | if (parent) | |
377 | return parent->subtree_control; | |
378 | ||
379 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
380 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
381 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
382 | return root_ss_mask; |
383 | } | |
384 | ||
385 | /* subsystems enabled on a cgroup */ | |
386 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
387 | { | |
388 | struct cgroup *parent = cgroup_parent(cgrp); | |
389 | ||
390 | if (parent) | |
391 | return parent->subtree_ss_mask; | |
392 | ||
393 | return cgrp->root->subsys_mask; | |
394 | } | |
395 | ||
95109b62 TH |
396 | /** |
397 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
398 | * @cgrp: the cgroup of interest | |
9d800df1 | 399 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 400 | * |
ca8bdcaf TH |
401 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
402 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
403 | * the caller is responsible for pinning the returned css if it wants to | |
404 | * keep accessing it outside the said locks. This function may return | |
405 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
406 | */ |
407 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 408 | struct cgroup_subsys *ss) |
95109b62 | 409 | { |
ca8bdcaf | 410 | if (ss) |
aec25020 | 411 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 412 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 413 | else |
9d800df1 | 414 | return &cgrp->self; |
95109b62 | 415 | } |
42809dd4 | 416 | |
aec3dfcb TH |
417 | /** |
418 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
419 | * @cgrp: the cgroup of interest | |
9d800df1 | 420 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 421 | * |
d0f702e6 | 422 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
423 | * as the matching css of the nearest ancestor including self which has @ss |
424 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
425 | * function is guaranteed to return non-NULL css. | |
426 | */ | |
427 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
428 | struct cgroup_subsys *ss) | |
429 | { | |
430 | lockdep_assert_held(&cgroup_mutex); | |
431 | ||
432 | if (!ss) | |
9d800df1 | 433 | return &cgrp->self; |
aec3dfcb | 434 | |
eeecbd19 TH |
435 | /* |
436 | * This function is used while updating css associations and thus | |
5531dc91 | 437 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 438 | */ |
5531dc91 | 439 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 440 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
441 | if (!cgrp) |
442 | return NULL; | |
443 | } | |
aec3dfcb TH |
444 | |
445 | return cgroup_css(cgrp, ss); | |
95109b62 | 446 | } |
42809dd4 | 447 | |
eeecbd19 TH |
448 | /** |
449 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
450 | * @cgrp: the cgroup of interest | |
451 | * @ss: the subsystem of interest | |
452 | * | |
453 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
454 | * defined as the matching css of the nearest ancestor including self which | |
455 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
456 | * the root css is returned, so this function always returns a valid css. | |
457 | * The returned css must be put using css_put(). | |
458 | */ | |
459 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
460 | struct cgroup_subsys *ss) | |
461 | { | |
462 | struct cgroup_subsys_state *css; | |
463 | ||
464 | rcu_read_lock(); | |
465 | ||
466 | do { | |
467 | css = cgroup_css(cgrp, ss); | |
468 | ||
469 | if (css && css_tryget_online(css)) | |
470 | goto out_unlock; | |
471 | cgrp = cgroup_parent(cgrp); | |
472 | } while (cgrp); | |
473 | ||
474 | css = init_css_set.subsys[ss->id]; | |
475 | css_get(css); | |
476 | out_unlock: | |
477 | rcu_read_unlock(); | |
478 | return css; | |
479 | } | |
480 | ||
ddbcc7e8 | 481 | /* convenient tests for these bits */ |
54766d4a | 482 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 483 | { |
184faf32 | 484 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
485 | } |
486 | ||
052c3f3a TH |
487 | static void cgroup_get(struct cgroup *cgrp) |
488 | { | |
489 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
490 | css_get(&cgrp->self); | |
491 | } | |
492 | ||
493 | static bool cgroup_tryget(struct cgroup *cgrp) | |
494 | { | |
495 | return css_tryget(&cgrp->self); | |
496 | } | |
497 | ||
b4168640 | 498 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 499 | { |
2bd59d48 | 500 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 501 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
502 | |
503 | /* | |
504 | * This is open and unprotected implementation of cgroup_css(). | |
505 | * seq_css() is only called from a kernfs file operation which has | |
506 | * an active reference on the file. Because all the subsystem | |
507 | * files are drained before a css is disassociated with a cgroup, | |
508 | * the matching css from the cgroup's subsys table is guaranteed to | |
509 | * be and stay valid until the enclosing operation is complete. | |
510 | */ | |
511 | if (cft->ss) | |
512 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
513 | else | |
9d800df1 | 514 | return &cgrp->self; |
59f5296b | 515 | } |
b4168640 | 516 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 517 | |
e9685a03 | 518 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 519 | { |
bd89aabc | 520 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
521 | } |
522 | ||
1c6727af TH |
523 | /** |
524 | * for_each_css - iterate all css's of a cgroup | |
525 | * @css: the iteration cursor | |
526 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
527 | * @cgrp: the target cgroup to iterate css's of | |
528 | * | |
aec3dfcb | 529 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
530 | */ |
531 | #define for_each_css(css, ssid, cgrp) \ | |
532 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
533 | if (!((css) = rcu_dereference_check( \ | |
534 | (cgrp)->subsys[(ssid)], \ | |
535 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
536 | else | |
537 | ||
aec3dfcb TH |
538 | /** |
539 | * for_each_e_css - iterate all effective css's of a cgroup | |
540 | * @css: the iteration cursor | |
541 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
542 | * @cgrp: the target cgroup to iterate css's of | |
543 | * | |
544 | * Should be called under cgroup_[tree_]mutex. | |
545 | */ | |
546 | #define for_each_e_css(css, ssid, cgrp) \ | |
547 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
548 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
549 | ; \ | |
550 | else | |
551 | ||
30159ec7 | 552 | /** |
3ed80a62 | 553 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 554 | * @ss: the iteration cursor |
780cd8b3 | 555 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 556 | */ |
780cd8b3 | 557 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
558 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
559 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 560 | |
cb4a3167 | 561 | /** |
b4e0eeaf | 562 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
563 | * @ss: the iteration cursor |
564 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 565 | * @ss_mask: the bitmask |
cb4a3167 AS |
566 | * |
567 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 568 | * @ss_mask is set. |
cb4a3167 | 569 | */ |
b4e0eeaf TH |
570 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
571 | unsigned long __ss_mask = (ss_mask); \ | |
572 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 573 | (ssid) = 0; \ |
b4e0eeaf TH |
574 | break; \ |
575 | } \ | |
576 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
577 | (ss) = cgroup_subsys[ssid]; \ | |
578 | { | |
579 | ||
580 | #define while_each_subsys_mask() \ | |
581 | } \ | |
582 | } \ | |
583 | } while (false) | |
cb4a3167 | 584 | |
985ed670 TH |
585 | /* iterate across the hierarchies */ |
586 | #define for_each_root(root) \ | |
5549c497 | 587 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 588 | |
f8f22e53 TH |
589 | /* iterate over child cgrps, lock should be held throughout iteration */ |
590 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 591 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 592 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
593 | cgroup_is_dead(child); })) \ |
594 | ; \ | |
595 | else | |
7ae1bad9 | 596 | |
ce3f1d9d TH |
597 | /* walk live descendants in preorder */ |
598 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
599 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
600 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
601 | (dsct) = (d_css)->cgroup; \ | |
602 | cgroup_is_dead(dsct); })) \ | |
603 | ; \ | |
604 | else | |
605 | ||
606 | /* walk live descendants in postorder */ | |
607 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
608 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
609 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
610 | (dsct) = (d_css)->cgroup; \ | |
611 | cgroup_is_dead(dsct); })) \ | |
612 | ; \ | |
613 | else | |
614 | ||
81a6a5cd | 615 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 616 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 617 | |
69d0206c TH |
618 | /* |
619 | * A cgroup can be associated with multiple css_sets as different tasks may | |
620 | * belong to different cgroups on different hierarchies. In the other | |
621 | * direction, a css_set is naturally associated with multiple cgroups. | |
622 | * This M:N relationship is represented by the following link structure | |
623 | * which exists for each association and allows traversing the associations | |
624 | * from both sides. | |
625 | */ | |
626 | struct cgrp_cset_link { | |
627 | /* the cgroup and css_set this link associates */ | |
628 | struct cgroup *cgrp; | |
629 | struct css_set *cset; | |
630 | ||
631 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
632 | struct list_head cset_link; | |
633 | ||
634 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
635 | struct list_head cgrp_link; | |
817929ec PM |
636 | }; |
637 | ||
172a2c06 TH |
638 | /* |
639 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
640 | * hierarchies being mounted. It contains a pointer to the root state |
641 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
642 | * reference-counted, to improve performance when child cgroups | |
643 | * haven't been created. | |
644 | */ | |
5024ae29 | 645 | struct css_set init_css_set = { |
172a2c06 TH |
646 | .refcount = ATOMIC_INIT(1), |
647 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
648 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
649 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
650 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
651 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 652 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 653 | }; |
817929ec | 654 | |
172a2c06 | 655 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 656 | |
0de0942d TH |
657 | /** |
658 | * css_set_populated - does a css_set contain any tasks? | |
659 | * @cset: target css_set | |
660 | */ | |
661 | static bool css_set_populated(struct css_set *cset) | |
662 | { | |
f0d9a5f1 | 663 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
664 | |
665 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
666 | } | |
667 | ||
842b597e TH |
668 | /** |
669 | * cgroup_update_populated - updated populated count of a cgroup | |
670 | * @cgrp: the target cgroup | |
671 | * @populated: inc or dec populated count | |
672 | * | |
0de0942d TH |
673 | * One of the css_sets associated with @cgrp is either getting its first |
674 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
675 | * count is propagated towards root so that a given cgroup's populated_cnt | |
676 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
677 | * |
678 | * @cgrp's interface file "cgroup.populated" is zero if | |
679 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
680 | * changes from or to zero, userland is notified that the content of the | |
681 | * interface file has changed. This can be used to detect when @cgrp and | |
682 | * its descendants become populated or empty. | |
683 | */ | |
684 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
685 | { | |
f0d9a5f1 | 686 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
687 | |
688 | do { | |
689 | bool trigger; | |
690 | ||
691 | if (populated) | |
692 | trigger = !cgrp->populated_cnt++; | |
693 | else | |
694 | trigger = !--cgrp->populated_cnt; | |
695 | ||
696 | if (!trigger) | |
697 | break; | |
698 | ||
ad2ed2b3 | 699 | check_for_release(cgrp); |
6f60eade TH |
700 | cgroup_file_notify(&cgrp->events_file); |
701 | ||
d51f39b0 | 702 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
703 | } while (cgrp); |
704 | } | |
705 | ||
0de0942d TH |
706 | /** |
707 | * css_set_update_populated - update populated state of a css_set | |
708 | * @cset: target css_set | |
709 | * @populated: whether @cset is populated or depopulated | |
710 | * | |
711 | * @cset is either getting the first task or losing the last. Update the | |
712 | * ->populated_cnt of all associated cgroups accordingly. | |
713 | */ | |
714 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
715 | { | |
716 | struct cgrp_cset_link *link; | |
717 | ||
f0d9a5f1 | 718 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
719 | |
720 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
721 | cgroup_update_populated(link->cgrp, populated); | |
722 | } | |
723 | ||
f6d7d049 TH |
724 | /** |
725 | * css_set_move_task - move a task from one css_set to another | |
726 | * @task: task being moved | |
727 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
728 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
729 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
730 | * | |
731 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
732 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
733 | * instead of moved, @to_cset can be NULL. | |
734 | * | |
ed27b9f7 TH |
735 | * This function automatically handles populated_cnt updates and |
736 | * css_task_iter adjustments but the caller is responsible for managing | |
737 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
738 | */ |
739 | static void css_set_move_task(struct task_struct *task, | |
740 | struct css_set *from_cset, struct css_set *to_cset, | |
741 | bool use_mg_tasks) | |
742 | { | |
f0d9a5f1 | 743 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 744 | |
20b454a6 TH |
745 | if (to_cset && !css_set_populated(to_cset)) |
746 | css_set_update_populated(to_cset, true); | |
747 | ||
f6d7d049 | 748 | if (from_cset) { |
ed27b9f7 TH |
749 | struct css_task_iter *it, *pos; |
750 | ||
f6d7d049 | 751 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
752 | |
753 | /* | |
754 | * @task is leaving, advance task iterators which are | |
755 | * pointing to it so that they can resume at the next | |
756 | * position. Advancing an iterator might remove it from | |
757 | * the list, use safe walk. See css_task_iter_advance*() | |
758 | * for details. | |
759 | */ | |
760 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
761 | iters_node) | |
762 | if (it->task_pos == &task->cg_list) | |
763 | css_task_iter_advance(it); | |
764 | ||
f6d7d049 TH |
765 | list_del_init(&task->cg_list); |
766 | if (!css_set_populated(from_cset)) | |
767 | css_set_update_populated(from_cset, false); | |
768 | } else { | |
769 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
770 | } | |
771 | ||
772 | if (to_cset) { | |
773 | /* | |
774 | * We are synchronized through cgroup_threadgroup_rwsem | |
775 | * against PF_EXITING setting such that we can't race | |
776 | * against cgroup_exit() changing the css_set to | |
777 | * init_css_set and dropping the old one. | |
778 | */ | |
779 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
780 | ||
f6d7d049 TH |
781 | rcu_assign_pointer(task->cgroups, to_cset); |
782 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
783 | &to_cset->tasks); | |
784 | } | |
785 | } | |
786 | ||
7717f7ba PM |
787 | /* |
788 | * hash table for cgroup groups. This improves the performance to find | |
789 | * an existing css_set. This hash doesn't (currently) take into | |
790 | * account cgroups in empty hierarchies. | |
791 | */ | |
472b1053 | 792 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 793 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 794 | |
0ac801fe | 795 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 796 | { |
0ac801fe | 797 | unsigned long key = 0UL; |
30159ec7 TH |
798 | struct cgroup_subsys *ss; |
799 | int i; | |
472b1053 | 800 | |
30159ec7 | 801 | for_each_subsys(ss, i) |
0ac801fe LZ |
802 | key += (unsigned long)css[i]; |
803 | key = (key >> 16) ^ key; | |
472b1053 | 804 | |
0ac801fe | 805 | return key; |
472b1053 LZ |
806 | } |
807 | ||
a25eb52e | 808 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 809 | { |
69d0206c | 810 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
811 | struct cgroup_subsys *ss; |
812 | int ssid; | |
5abb8855 | 813 | |
f0d9a5f1 | 814 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
815 | |
816 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 817 | return; |
81a6a5cd | 818 | |
53254f90 TH |
819 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
820 | for_each_subsys(ss, ssid) { | |
2d8f243a | 821 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
822 | css_put(cset->subsys[ssid]); |
823 | } | |
5abb8855 | 824 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
825 | css_set_count--; |
826 | ||
69d0206c | 827 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
828 | list_del(&link->cset_link); |
829 | list_del(&link->cgrp_link); | |
2ceb231b TH |
830 | if (cgroup_parent(link->cgrp)) |
831 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 832 | kfree(link); |
81a6a5cd | 833 | } |
2c6ab6d2 | 834 | |
5abb8855 | 835 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
836 | } |
837 | ||
a25eb52e | 838 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
839 | { |
840 | /* | |
841 | * Ensure that the refcount doesn't hit zero while any readers | |
842 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
843 | * rwlock | |
844 | */ | |
845 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
846 | return; | |
847 | ||
f0d9a5f1 | 848 | spin_lock_bh(&css_set_lock); |
a25eb52e | 849 | put_css_set_locked(cset); |
f0d9a5f1 | 850 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
851 | } |
852 | ||
817929ec PM |
853 | /* |
854 | * refcounted get/put for css_set objects | |
855 | */ | |
5abb8855 | 856 | static inline void get_css_set(struct css_set *cset) |
817929ec | 857 | { |
5abb8855 | 858 | atomic_inc(&cset->refcount); |
817929ec PM |
859 | } |
860 | ||
b326f9d0 | 861 | /** |
7717f7ba | 862 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
863 | * @cset: candidate css_set being tested |
864 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
865 | * @new_cgrp: cgroup that's being entered by the task |
866 | * @template: desired set of css pointers in css_set (pre-calculated) | |
867 | * | |
6f4b7e63 | 868 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
869 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
870 | */ | |
5abb8855 TH |
871 | static bool compare_css_sets(struct css_set *cset, |
872 | struct css_set *old_cset, | |
7717f7ba PM |
873 | struct cgroup *new_cgrp, |
874 | struct cgroup_subsys_state *template[]) | |
875 | { | |
876 | struct list_head *l1, *l2; | |
877 | ||
aec3dfcb TH |
878 | /* |
879 | * On the default hierarchy, there can be csets which are | |
880 | * associated with the same set of cgroups but different csses. | |
881 | * Let's first ensure that csses match. | |
882 | */ | |
883 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 884 | return false; |
7717f7ba PM |
885 | |
886 | /* | |
887 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
888 | * different cgroups in hierarchies. As different cgroups may |
889 | * share the same effective css, this comparison is always | |
890 | * necessary. | |
7717f7ba | 891 | */ |
69d0206c TH |
892 | l1 = &cset->cgrp_links; |
893 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 894 | while (1) { |
69d0206c | 895 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 896 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
897 | |
898 | l1 = l1->next; | |
899 | l2 = l2->next; | |
900 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
901 | if (l1 == &cset->cgrp_links) { |
902 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
903 | break; |
904 | } else { | |
69d0206c | 905 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
906 | } |
907 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
908 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
909 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
910 | cgrp1 = link1->cgrp; | |
911 | cgrp2 = link2->cgrp; | |
7717f7ba | 912 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 913 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
914 | |
915 | /* | |
916 | * If this hierarchy is the hierarchy of the cgroup | |
917 | * that's changing, then we need to check that this | |
918 | * css_set points to the new cgroup; if it's any other | |
919 | * hierarchy, then this css_set should point to the | |
920 | * same cgroup as the old css_set. | |
921 | */ | |
5abb8855 TH |
922 | if (cgrp1->root == new_cgrp->root) { |
923 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
924 | return false; |
925 | } else { | |
5abb8855 | 926 | if (cgrp1 != cgrp2) |
7717f7ba PM |
927 | return false; |
928 | } | |
929 | } | |
930 | return true; | |
931 | } | |
932 | ||
b326f9d0 TH |
933 | /** |
934 | * find_existing_css_set - init css array and find the matching css_set | |
935 | * @old_cset: the css_set that we're using before the cgroup transition | |
936 | * @cgrp: the cgroup that we're moving into | |
937 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 938 | */ |
5abb8855 TH |
939 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
940 | struct cgroup *cgrp, | |
941 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 942 | { |
3dd06ffa | 943 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 944 | struct cgroup_subsys *ss; |
5abb8855 | 945 | struct css_set *cset; |
0ac801fe | 946 | unsigned long key; |
b326f9d0 | 947 | int i; |
817929ec | 948 | |
aae8aab4 BB |
949 | /* |
950 | * Build the set of subsystem state objects that we want to see in the | |
951 | * new css_set. while subsystems can change globally, the entries here | |
952 | * won't change, so no need for locking. | |
953 | */ | |
30159ec7 | 954 | for_each_subsys(ss, i) { |
f392e51c | 955 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
956 | /* |
957 | * @ss is in this hierarchy, so we want the | |
958 | * effective css from @cgrp. | |
959 | */ | |
960 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 961 | } else { |
aec3dfcb TH |
962 | /* |
963 | * @ss is not in this hierarchy, so we don't want | |
964 | * to change the css. | |
965 | */ | |
5abb8855 | 966 | template[i] = old_cset->subsys[i]; |
817929ec PM |
967 | } |
968 | } | |
969 | ||
0ac801fe | 970 | key = css_set_hash(template); |
5abb8855 TH |
971 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
972 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
973 | continue; |
974 | ||
975 | /* This css_set matches what we need */ | |
5abb8855 | 976 | return cset; |
472b1053 | 977 | } |
817929ec PM |
978 | |
979 | /* No existing cgroup group matched */ | |
980 | return NULL; | |
981 | } | |
982 | ||
69d0206c | 983 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 984 | { |
69d0206c | 985 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 986 | |
69d0206c TH |
987 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
988 | list_del(&link->cset_link); | |
36553434 LZ |
989 | kfree(link); |
990 | } | |
991 | } | |
992 | ||
69d0206c TH |
993 | /** |
994 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
995 | * @count: the number of links to allocate | |
996 | * @tmp_links: list_head the allocated links are put on | |
997 | * | |
998 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
999 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1000 | */ |
69d0206c | 1001 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1002 | { |
69d0206c | 1003 | struct cgrp_cset_link *link; |
817929ec | 1004 | int i; |
69d0206c TH |
1005 | |
1006 | INIT_LIST_HEAD(tmp_links); | |
1007 | ||
817929ec | 1008 | for (i = 0; i < count; i++) { |
f4f4be2b | 1009 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1010 | if (!link) { |
69d0206c | 1011 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1012 | return -ENOMEM; |
1013 | } | |
69d0206c | 1014 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1015 | } |
1016 | return 0; | |
1017 | } | |
1018 | ||
c12f65d4 LZ |
1019 | /** |
1020 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1021 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1022 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1023 | * @cgrp: the destination cgroup |
1024 | */ | |
69d0206c TH |
1025 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1026 | struct cgroup *cgrp) | |
c12f65d4 | 1027 | { |
69d0206c | 1028 | struct cgrp_cset_link *link; |
c12f65d4 | 1029 | |
69d0206c | 1030 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1031 | |
1032 | if (cgroup_on_dfl(cgrp)) | |
1033 | cset->dfl_cgrp = cgrp; | |
1034 | ||
69d0206c TH |
1035 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1036 | link->cset = cset; | |
7717f7ba | 1037 | link->cgrp = cgrp; |
842b597e | 1038 | |
7717f7ba | 1039 | /* |
389b9c1b TH |
1040 | * Always add links to the tail of the lists so that the lists are |
1041 | * in choronological order. | |
7717f7ba | 1042 | */ |
389b9c1b | 1043 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1044 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1045 | |
1046 | if (cgroup_parent(cgrp)) | |
1047 | cgroup_get(cgrp); | |
c12f65d4 LZ |
1048 | } |
1049 | ||
b326f9d0 TH |
1050 | /** |
1051 | * find_css_set - return a new css_set with one cgroup updated | |
1052 | * @old_cset: the baseline css_set | |
1053 | * @cgrp: the cgroup to be updated | |
1054 | * | |
1055 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1056 | * substituted into the appropriate hierarchy. | |
817929ec | 1057 | */ |
5abb8855 TH |
1058 | static struct css_set *find_css_set(struct css_set *old_cset, |
1059 | struct cgroup *cgrp) | |
817929ec | 1060 | { |
b326f9d0 | 1061 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1062 | struct css_set *cset; |
69d0206c TH |
1063 | struct list_head tmp_links; |
1064 | struct cgrp_cset_link *link; | |
2d8f243a | 1065 | struct cgroup_subsys *ss; |
0ac801fe | 1066 | unsigned long key; |
2d8f243a | 1067 | int ssid; |
472b1053 | 1068 | |
b326f9d0 TH |
1069 | lockdep_assert_held(&cgroup_mutex); |
1070 | ||
817929ec PM |
1071 | /* First see if we already have a cgroup group that matches |
1072 | * the desired set */ | |
f0d9a5f1 | 1073 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1074 | cset = find_existing_css_set(old_cset, cgrp, template); |
1075 | if (cset) | |
1076 | get_css_set(cset); | |
f0d9a5f1 | 1077 | spin_unlock_bh(&css_set_lock); |
817929ec | 1078 | |
5abb8855 TH |
1079 | if (cset) |
1080 | return cset; | |
817929ec | 1081 | |
f4f4be2b | 1082 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1083 | if (!cset) |
817929ec PM |
1084 | return NULL; |
1085 | ||
69d0206c | 1086 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1087 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1088 | kfree(cset); |
817929ec PM |
1089 | return NULL; |
1090 | } | |
1091 | ||
5abb8855 | 1092 | atomic_set(&cset->refcount, 1); |
69d0206c | 1093 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1094 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1095 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1096 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1097 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1098 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1099 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1100 | |
1101 | /* Copy the set of subsystem state objects generated in | |
1102 | * find_existing_css_set() */ | |
5abb8855 | 1103 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1104 | |
f0d9a5f1 | 1105 | spin_lock_bh(&css_set_lock); |
817929ec | 1106 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1107 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1108 | struct cgroup *c = link->cgrp; |
69d0206c | 1109 | |
7717f7ba PM |
1110 | if (c->root == cgrp->root) |
1111 | c = cgrp; | |
69d0206c | 1112 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1113 | } |
817929ec | 1114 | |
69d0206c | 1115 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1116 | |
817929ec | 1117 | css_set_count++; |
472b1053 | 1118 | |
2d8f243a | 1119 | /* Add @cset to the hash table */ |
5abb8855 TH |
1120 | key = css_set_hash(cset->subsys); |
1121 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1122 | |
53254f90 TH |
1123 | for_each_subsys(ss, ssid) { |
1124 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1125 | ||
2d8f243a | 1126 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1127 | &css->cgroup->e_csets[ssid]); |
1128 | css_get(css); | |
1129 | } | |
2d8f243a | 1130 | |
f0d9a5f1 | 1131 | spin_unlock_bh(&css_set_lock); |
817929ec | 1132 | |
5abb8855 | 1133 | return cset; |
b4f48b63 PM |
1134 | } |
1135 | ||
3dd06ffa | 1136 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1137 | { |
3dd06ffa | 1138 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1139 | |
3dd06ffa | 1140 | return root_cgrp->root; |
2bd59d48 TH |
1141 | } |
1142 | ||
3dd06ffa | 1143 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1144 | { |
1145 | int id; | |
1146 | ||
1147 | lockdep_assert_held(&cgroup_mutex); | |
1148 | ||
985ed670 | 1149 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1150 | if (id < 0) |
1151 | return id; | |
1152 | ||
1153 | root->hierarchy_id = id; | |
1154 | return 0; | |
1155 | } | |
1156 | ||
3dd06ffa | 1157 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1158 | { |
1159 | lockdep_assert_held(&cgroup_mutex); | |
1160 | ||
8c8a5502 | 1161 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1162 | } |
1163 | ||
3dd06ffa | 1164 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1165 | { |
1166 | if (root) { | |
f2e85d57 TH |
1167 | idr_destroy(&root->cgroup_idr); |
1168 | kfree(root); | |
1169 | } | |
1170 | } | |
1171 | ||
3dd06ffa | 1172 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1173 | { |
3dd06ffa | 1174 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1175 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1176 | |
334c3679 | 1177 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1178 | |
776f02fa | 1179 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1180 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1181 | |
f2e85d57 | 1182 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1183 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1184 | |
7717f7ba | 1185 | /* |
f2e85d57 TH |
1186 | * Release all the links from cset_links to this hierarchy's |
1187 | * root cgroup | |
7717f7ba | 1188 | */ |
f0d9a5f1 | 1189 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1190 | |
1191 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1192 | list_del(&link->cset_link); | |
1193 | list_del(&link->cgrp_link); | |
1194 | kfree(link); | |
1195 | } | |
f0d9a5f1 TH |
1196 | |
1197 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1198 | |
1199 | if (!list_empty(&root->root_list)) { | |
1200 | list_del(&root->root_list); | |
1201 | cgroup_root_count--; | |
1202 | } | |
1203 | ||
1204 | cgroup_exit_root_id(root); | |
1205 | ||
1206 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1207 | |
2bd59d48 | 1208 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1209 | cgroup_free_root(root); |
1210 | } | |
1211 | ||
4f41fc59 SH |
1212 | /* |
1213 | * look up cgroup associated with current task's cgroup namespace on the | |
1214 | * specified hierarchy | |
1215 | */ | |
1216 | static struct cgroup * | |
1217 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1218 | { | |
1219 | struct cgroup *res = NULL; | |
1220 | struct css_set *cset; | |
1221 | ||
1222 | lockdep_assert_held(&css_set_lock); | |
1223 | ||
1224 | rcu_read_lock(); | |
1225 | ||
1226 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1227 | if (cset == &init_css_set) { | |
1228 | res = &root->cgrp; | |
1229 | } else { | |
1230 | struct cgrp_cset_link *link; | |
1231 | ||
1232 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1233 | struct cgroup *c = link->cgrp; | |
1234 | ||
1235 | if (c->root == root) { | |
1236 | res = c; | |
1237 | break; | |
1238 | } | |
1239 | } | |
1240 | } | |
1241 | rcu_read_unlock(); | |
1242 | ||
1243 | BUG_ON(!res); | |
1244 | return res; | |
1245 | } | |
1246 | ||
ceb6a081 TH |
1247 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1248 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1249 | struct cgroup_root *root) |
7717f7ba | 1250 | { |
7717f7ba PM |
1251 | struct cgroup *res = NULL; |
1252 | ||
96d365e0 | 1253 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1254 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1255 | |
5abb8855 | 1256 | if (cset == &init_css_set) { |
3dd06ffa | 1257 | res = &root->cgrp; |
7717f7ba | 1258 | } else { |
69d0206c TH |
1259 | struct cgrp_cset_link *link; |
1260 | ||
1261 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1262 | struct cgroup *c = link->cgrp; |
69d0206c | 1263 | |
7717f7ba PM |
1264 | if (c->root == root) { |
1265 | res = c; | |
1266 | break; | |
1267 | } | |
1268 | } | |
1269 | } | |
96d365e0 | 1270 | |
7717f7ba PM |
1271 | BUG_ON(!res); |
1272 | return res; | |
1273 | } | |
1274 | ||
ddbcc7e8 | 1275 | /* |
ceb6a081 | 1276 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1277 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1278 | */ |
1279 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1280 | struct cgroup_root *root) |
ceb6a081 TH |
1281 | { |
1282 | /* | |
1283 | * No need to lock the task - since we hold cgroup_mutex the | |
1284 | * task can't change groups, so the only thing that can happen | |
1285 | * is that it exits and its css is set back to init_css_set. | |
1286 | */ | |
1287 | return cset_cgroup_from_root(task_css_set(task), root); | |
1288 | } | |
1289 | ||
ddbcc7e8 | 1290 | /* |
ddbcc7e8 PM |
1291 | * A task must hold cgroup_mutex to modify cgroups. |
1292 | * | |
1293 | * Any task can increment and decrement the count field without lock. | |
1294 | * So in general, code holding cgroup_mutex can't rely on the count | |
1295 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1296 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1297 | * means that no tasks are currently attached, therefore there is no |
1298 | * way a task attached to that cgroup can fork (the other way to | |
1299 | * increment the count). So code holding cgroup_mutex can safely | |
1300 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1301 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1302 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1303 | * needs that mutex. | |
1304 | * | |
ddbcc7e8 PM |
1305 | * A cgroup can only be deleted if both its 'count' of using tasks |
1306 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1307 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1308 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1309 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1310 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1311 | * |
1312 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1313 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1314 | */ |
1315 | ||
2bd59d48 | 1316 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1317 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1318 | |
8d7e6fb0 TH |
1319 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1320 | char *buf) | |
ddbcc7e8 | 1321 | { |
3e1d2eed TH |
1322 | struct cgroup_subsys *ss = cft->ss; |
1323 | ||
8d7e6fb0 TH |
1324 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1325 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1326 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1327 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1328 | cft->name); | |
8d7e6fb0 TH |
1329 | else |
1330 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1331 | return buf; | |
ddbcc7e8 PM |
1332 | } |
1333 | ||
f2e85d57 TH |
1334 | /** |
1335 | * cgroup_file_mode - deduce file mode of a control file | |
1336 | * @cft: the control file in question | |
1337 | * | |
7dbdb199 | 1338 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1339 | */ |
1340 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1341 | { |
f2e85d57 | 1342 | umode_t mode = 0; |
65dff759 | 1343 | |
f2e85d57 TH |
1344 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1345 | mode |= S_IRUGO; | |
1346 | ||
7dbdb199 TH |
1347 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1348 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1349 | mode |= S_IWUGO; | |
1350 | else | |
1351 | mode |= S_IWUSR; | |
1352 | } | |
f2e85d57 TH |
1353 | |
1354 | return mode; | |
65dff759 LZ |
1355 | } |
1356 | ||
af0ba678 | 1357 | /** |
8699b776 | 1358 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1359 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1360 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1361 | * |
1362 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1363 | * enabled together through its ->depends_on mask. In such cases, more | |
1364 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1365 | * | |
0f060deb | 1366 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1367 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1368 | */ |
5ced2518 | 1369 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1370 | { |
6e5c8307 | 1371 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1372 | struct cgroup_subsys *ss; |
1373 | int ssid; | |
1374 | ||
1375 | lockdep_assert_held(&cgroup_mutex); | |
1376 | ||
f6d635ad TH |
1377 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1378 | ||
af0ba678 | 1379 | while (true) { |
6e5c8307 | 1380 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1381 | |
b4e0eeaf | 1382 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1383 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1384 | } while_each_subsys_mask(); |
af0ba678 TH |
1385 | |
1386 | /* | |
1387 | * Mask out subsystems which aren't available. This can | |
1388 | * happen only if some depended-upon subsystems were bound | |
1389 | * to non-default hierarchies. | |
1390 | */ | |
5ced2518 | 1391 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1392 | |
1393 | if (new_ss_mask == cur_ss_mask) | |
1394 | break; | |
1395 | cur_ss_mask = new_ss_mask; | |
1396 | } | |
1397 | ||
0f060deb TH |
1398 | return cur_ss_mask; |
1399 | } | |
1400 | ||
a9746d8d TH |
1401 | /** |
1402 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1403 | * @kn: the kernfs_node being serviced | |
1404 | * | |
1405 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1406 | * the method finishes if locking succeeded. Note that once this function | |
1407 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1408 | * inaccessible any time. If the caller intends to continue to access the | |
1409 | * cgroup, it should pin it before invoking this function. | |
1410 | */ | |
1411 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1412 | { |
a9746d8d TH |
1413 | struct cgroup *cgrp; |
1414 | ||
1415 | if (kernfs_type(kn) == KERNFS_DIR) | |
1416 | cgrp = kn->priv; | |
1417 | else | |
1418 | cgrp = kn->parent->priv; | |
1419 | ||
1420 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1421 | |
1422 | kernfs_unbreak_active_protection(kn); | |
1423 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1424 | } |
1425 | ||
a9746d8d TH |
1426 | /** |
1427 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1428 | * @kn: the kernfs_node being serviced | |
945ba199 | 1429 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1430 | * |
1431 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1432 | * @kn. It breaks the active protection, performs cgroup locking and | |
1433 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1434 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1435 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1436 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1437 | * |
1438 | * Any cgroup kernfs method implementation which requires locking the | |
1439 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1440 | * locking under kernfs active protection and allows all kernfs operations | |
1441 | * including self-removal. | |
1442 | */ | |
945ba199 TH |
1443 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1444 | bool drain_offline) | |
05ef1d7c | 1445 | { |
a9746d8d TH |
1446 | struct cgroup *cgrp; |
1447 | ||
1448 | if (kernfs_type(kn) == KERNFS_DIR) | |
1449 | cgrp = kn->priv; | |
1450 | else | |
1451 | cgrp = kn->parent->priv; | |
05ef1d7c | 1452 | |
2739d3cc | 1453 | /* |
01f6474c | 1454 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1455 | * active_ref. cgroup liveliness check alone provides enough |
1456 | * protection against removal. Ensure @cgrp stays accessible and | |
1457 | * break the active_ref protection. | |
2739d3cc | 1458 | */ |
aa32362f LZ |
1459 | if (!cgroup_tryget(cgrp)) |
1460 | return NULL; | |
a9746d8d TH |
1461 | kernfs_break_active_protection(kn); |
1462 | ||
945ba199 TH |
1463 | if (drain_offline) |
1464 | cgroup_lock_and_drain_offline(cgrp); | |
1465 | else | |
1466 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1467 | |
a9746d8d TH |
1468 | if (!cgroup_is_dead(cgrp)) |
1469 | return cgrp; | |
1470 | ||
1471 | cgroup_kn_unlock(kn); | |
1472 | return NULL; | |
ddbcc7e8 | 1473 | } |
05ef1d7c | 1474 | |
2739d3cc | 1475 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1476 | { |
2bd59d48 | 1477 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1478 | |
01f6474c | 1479 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1480 | |
1481 | if (cft->file_offset) { | |
1482 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1483 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1484 | ||
1485 | spin_lock_irq(&cgroup_file_kn_lock); | |
1486 | cfile->kn = NULL; | |
1487 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1488 | } | |
1489 | ||
2bd59d48 | 1490 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1491 | } |
1492 | ||
13af07df | 1493 | /** |
4df8dc90 TH |
1494 | * css_clear_dir - remove subsys files in a cgroup directory |
1495 | * @css: taget css | |
13af07df | 1496 | */ |
334c3679 | 1497 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1498 | { |
334c3679 | 1499 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1500 | struct cftype *cfts; |
05ef1d7c | 1501 | |
88cb04b9 TH |
1502 | if (!(css->flags & CSS_VISIBLE)) |
1503 | return; | |
1504 | ||
1505 | css->flags &= ~CSS_VISIBLE; | |
1506 | ||
4df8dc90 TH |
1507 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1508 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1509 | } |
1510 | ||
ccdca218 | 1511 | /** |
4df8dc90 TH |
1512 | * css_populate_dir - create subsys files in a cgroup directory |
1513 | * @css: target css | |
ccdca218 TH |
1514 | * |
1515 | * On failure, no file is added. | |
1516 | */ | |
334c3679 | 1517 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1518 | { |
334c3679 | 1519 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1520 | struct cftype *cfts, *failed_cfts; |
1521 | int ret; | |
ccdca218 | 1522 | |
03970d3c | 1523 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1524 | return 0; |
1525 | ||
4df8dc90 TH |
1526 | if (!css->ss) { |
1527 | if (cgroup_on_dfl(cgrp)) | |
1528 | cfts = cgroup_dfl_base_files; | |
1529 | else | |
1530 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1531 | |
4df8dc90 TH |
1532 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1533 | } | |
ccdca218 | 1534 | |
4df8dc90 TH |
1535 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1536 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1537 | if (ret < 0) { | |
1538 | failed_cfts = cfts; | |
1539 | goto err; | |
ccdca218 TH |
1540 | } |
1541 | } | |
88cb04b9 TH |
1542 | |
1543 | css->flags |= CSS_VISIBLE; | |
1544 | ||
ccdca218 TH |
1545 | return 0; |
1546 | err: | |
4df8dc90 TH |
1547 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1548 | if (cfts == failed_cfts) | |
1549 | break; | |
1550 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1551 | } | |
ccdca218 TH |
1552 | return ret; |
1553 | } | |
1554 | ||
6e5c8307 | 1555 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1556 | { |
1ada4838 | 1557 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1558 | struct cgroup_subsys *ss; |
2d8f243a | 1559 | int ssid, i, ret; |
ddbcc7e8 | 1560 | |
ace2bee8 | 1561 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1562 | |
b4e0eeaf | 1563 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1564 | /* |
1565 | * If @ss has non-root csses attached to it, can't move. | |
1566 | * If @ss is an implicit controller, it is exempt from this | |
1567 | * rule and can be stolen. | |
1568 | */ | |
1569 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1570 | !ss->implicit_on_dfl) | |
3ed80a62 | 1571 | return -EBUSY; |
1d5be6b2 | 1572 | |
5df36032 | 1573 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1574 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1575 | return -EBUSY; |
b4e0eeaf | 1576 | } while_each_subsys_mask(); |
ddbcc7e8 | 1577 | |
b4e0eeaf | 1578 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1579 | struct cgroup_root *src_root = ss->root; |
1580 | struct cgroup *scgrp = &src_root->cgrp; | |
1581 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1582 | struct css_set *cset; |
a8a648c4 | 1583 | |
1ada4838 | 1584 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1585 | |
334c3679 TH |
1586 | /* disable from the source */ |
1587 | src_root->subsys_mask &= ~(1 << ssid); | |
1588 | WARN_ON(cgroup_apply_control(scgrp)); | |
1589 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1590 | |
334c3679 | 1591 | /* rebind */ |
1ada4838 TH |
1592 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1593 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1594 | ss->root = dst_root; |
1ada4838 | 1595 | css->cgroup = dcgrp; |
73e80ed8 | 1596 | |
f0d9a5f1 | 1597 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1598 | hash_for_each(css_set_table, i, cset, hlist) |
1599 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1600 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1601 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1602 | |
bd53d617 | 1603 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1604 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1605 | if (dst_root == &cgrp_dfl_root) { |
1606 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1607 | } else { | |
1ada4838 | 1608 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1609 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1610 | } |
a8a648c4 | 1611 | |
334c3679 TH |
1612 | ret = cgroup_apply_control(dcgrp); |
1613 | if (ret) | |
1614 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1615 | ss->name, ret); | |
1616 | ||
5df36032 TH |
1617 | if (ss->bind) |
1618 | ss->bind(css); | |
b4e0eeaf | 1619 | } while_each_subsys_mask(); |
ddbcc7e8 | 1620 | |
1ada4838 | 1621 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1622 | return 0; |
1623 | } | |
1624 | ||
4f41fc59 SH |
1625 | static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1626 | struct kernfs_root *kf_root) | |
1627 | { | |
09be4c82 | 1628 | int len = 0; |
4f41fc59 SH |
1629 | char *buf = NULL; |
1630 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1631 | struct cgroup *ns_cgroup; | |
1632 | ||
1633 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1634 | if (!buf) | |
1635 | return -ENOMEM; | |
1636 | ||
1637 | spin_lock_bh(&css_set_lock); | |
1638 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); | |
1639 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
1640 | spin_unlock_bh(&css_set_lock); | |
1641 | ||
1642 | if (len >= PATH_MAX) | |
1643 | len = -ERANGE; | |
1644 | else if (len > 0) { | |
1645 | seq_escape(sf, buf, " \t\n\\"); | |
1646 | len = 0; | |
1647 | } | |
1648 | kfree(buf); | |
1649 | return len; | |
1650 | } | |
1651 | ||
2bd59d48 TH |
1652 | static int cgroup_show_options(struct seq_file *seq, |
1653 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1654 | { |
3dd06ffa | 1655 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1656 | struct cgroup_subsys *ss; |
b85d2040 | 1657 | int ssid; |
ddbcc7e8 | 1658 | |
d98817d4 TH |
1659 | if (root != &cgrp_dfl_root) |
1660 | for_each_subsys(ss, ssid) | |
1661 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1662 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1663 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1664 | seq_puts(seq, ",noprefix"); |
93438629 | 1665 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1666 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1667 | |
1668 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1669 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1670 | seq_show_option(seq, "release_agent", |
1671 | root->release_agent_path); | |
69e943b7 TH |
1672 | spin_unlock(&release_agent_path_lock); |
1673 | ||
3dd06ffa | 1674 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1675 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1676 | if (strlen(root->name)) |
a068acf2 | 1677 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1678 | return 0; |
1679 | } | |
1680 | ||
1681 | struct cgroup_sb_opts { | |
6e5c8307 | 1682 | u16 subsys_mask; |
69dfa00c | 1683 | unsigned int flags; |
81a6a5cd | 1684 | char *release_agent; |
2260e7fc | 1685 | bool cpuset_clone_children; |
c6d57f33 | 1686 | char *name; |
2c6ab6d2 PM |
1687 | /* User explicitly requested empty subsystem */ |
1688 | bool none; | |
ddbcc7e8 PM |
1689 | }; |
1690 | ||
cf5d5941 | 1691 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1692 | { |
32a8cf23 DL |
1693 | char *token, *o = data; |
1694 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1695 | u16 mask = U16_MAX; |
30159ec7 | 1696 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1697 | int nr_opts = 0; |
30159ec7 | 1698 | int i; |
f9ab5b5b LZ |
1699 | |
1700 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1701 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1702 | #endif |
ddbcc7e8 | 1703 | |
c6d57f33 | 1704 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1705 | |
1706 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1707 | nr_opts++; |
1708 | ||
ddbcc7e8 PM |
1709 | if (!*token) |
1710 | return -EINVAL; | |
32a8cf23 | 1711 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1712 | /* Explicitly have no subsystems */ |
1713 | opts->none = true; | |
32a8cf23 DL |
1714 | continue; |
1715 | } | |
1716 | if (!strcmp(token, "all")) { | |
1717 | /* Mutually exclusive option 'all' + subsystem name */ | |
1718 | if (one_ss) | |
1719 | return -EINVAL; | |
1720 | all_ss = true; | |
1721 | continue; | |
1722 | } | |
1723 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1724 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1725 | continue; |
1726 | } | |
1727 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1728 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1729 | continue; |
1730 | } | |
03b1cde6 | 1731 | if (!strcmp(token, "xattr")) { |
93438629 | 1732 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1733 | continue; |
1734 | } | |
32a8cf23 | 1735 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1736 | /* Specifying two release agents is forbidden */ |
1737 | if (opts->release_agent) | |
1738 | return -EINVAL; | |
c6d57f33 | 1739 | opts->release_agent = |
e400c285 | 1740 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1741 | if (!opts->release_agent) |
1742 | return -ENOMEM; | |
32a8cf23 DL |
1743 | continue; |
1744 | } | |
1745 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1746 | const char *name = token + 5; |
1747 | /* Can't specify an empty name */ | |
1748 | if (!strlen(name)) | |
1749 | return -EINVAL; | |
1750 | /* Must match [\w.-]+ */ | |
1751 | for (i = 0; i < strlen(name); i++) { | |
1752 | char c = name[i]; | |
1753 | if (isalnum(c)) | |
1754 | continue; | |
1755 | if ((c == '.') || (c == '-') || (c == '_')) | |
1756 | continue; | |
1757 | return -EINVAL; | |
1758 | } | |
1759 | /* Specifying two names is forbidden */ | |
1760 | if (opts->name) | |
1761 | return -EINVAL; | |
1762 | opts->name = kstrndup(name, | |
e400c285 | 1763 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1764 | GFP_KERNEL); |
1765 | if (!opts->name) | |
1766 | return -ENOMEM; | |
32a8cf23 DL |
1767 | |
1768 | continue; | |
1769 | } | |
1770 | ||
30159ec7 | 1771 | for_each_subsys(ss, i) { |
3e1d2eed | 1772 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1773 | continue; |
fc5ed1e9 | 1774 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1775 | continue; |
223ffb29 JW |
1776 | if (cgroup_ssid_no_v1(i)) |
1777 | continue; | |
32a8cf23 DL |
1778 | |
1779 | /* Mutually exclusive option 'all' + subsystem name */ | |
1780 | if (all_ss) | |
1781 | return -EINVAL; | |
69dfa00c | 1782 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1783 | one_ss = true; |
1784 | ||
1785 | break; | |
1786 | } | |
1787 | if (i == CGROUP_SUBSYS_COUNT) | |
1788 | return -ENOENT; | |
1789 | } | |
1790 | ||
7b9a6ba5 TH |
1791 | /* |
1792 | * If the 'all' option was specified select all the subsystems, | |
1793 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1794 | * not specified, let's default to 'all' | |
1795 | */ | |
1796 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1797 | for_each_subsys(ss, i) | |
223ffb29 | 1798 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1799 | opts->subsys_mask |= (1 << i); |
1800 | ||
1801 | /* | |
1802 | * We either have to specify by name or by subsystems. (So all | |
1803 | * empty hierarchies must have a name). | |
1804 | */ | |
1805 | if (!opts->subsys_mask && !opts->name) | |
1806 | return -EINVAL; | |
1807 | ||
f9ab5b5b LZ |
1808 | /* |
1809 | * Option noprefix was introduced just for backward compatibility | |
1810 | * with the old cpuset, so we allow noprefix only if mounting just | |
1811 | * the cpuset subsystem. | |
1812 | */ | |
93438629 | 1813 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1814 | return -EINVAL; |
1815 | ||
2c6ab6d2 | 1816 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1817 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1818 | return -EINVAL; |
1819 | ||
ddbcc7e8 PM |
1820 | return 0; |
1821 | } | |
1822 | ||
2bd59d48 | 1823 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1824 | { |
1825 | int ret = 0; | |
3dd06ffa | 1826 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1827 | struct cgroup_sb_opts opts; |
6e5c8307 | 1828 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1829 | |
aa6ec29b TH |
1830 | if (root == &cgrp_dfl_root) { |
1831 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1832 | return -EINVAL; |
1833 | } | |
1834 | ||
334c3679 | 1835 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1836 | |
1837 | /* See what subsystems are wanted */ | |
1838 | ret = parse_cgroupfs_options(data, &opts); | |
1839 | if (ret) | |
1840 | goto out_unlock; | |
1841 | ||
f392e51c | 1842 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1843 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1844 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1845 | |
f392e51c TH |
1846 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1847 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1848 | |
cf5d5941 | 1849 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1850 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1851 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1852 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1853 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1854 | ret = -EINVAL; |
1855 | goto out_unlock; | |
1856 | } | |
1857 | ||
f172e67c | 1858 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1859 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1860 | ret = -EBUSY; |
0670e08b | 1861 | goto out_unlock; |
cf5d5941 | 1862 | } |
ddbcc7e8 | 1863 | |
5df36032 | 1864 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1865 | if (ret) |
0670e08b | 1866 | goto out_unlock; |
ddbcc7e8 | 1867 | |
334c3679 | 1868 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1869 | |
69e943b7 TH |
1870 | if (opts.release_agent) { |
1871 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1872 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1873 | spin_unlock(&release_agent_path_lock); |
1874 | } | |
ddbcc7e8 | 1875 | out_unlock: |
66bdc9cf | 1876 | kfree(opts.release_agent); |
c6d57f33 | 1877 | kfree(opts.name); |
ddbcc7e8 | 1878 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1879 | return ret; |
1880 | } | |
1881 | ||
afeb0f9f TH |
1882 | /* |
1883 | * To reduce the fork() overhead for systems that are not actually using | |
1884 | * their cgroups capability, we don't maintain the lists running through | |
1885 | * each css_set to its tasks until we see the list actually used - in other | |
1886 | * words after the first mount. | |
1887 | */ | |
1888 | static bool use_task_css_set_links __read_mostly; | |
1889 | ||
1890 | static void cgroup_enable_task_cg_lists(void) | |
1891 | { | |
1892 | struct task_struct *p, *g; | |
1893 | ||
f0d9a5f1 | 1894 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1895 | |
1896 | if (use_task_css_set_links) | |
1897 | goto out_unlock; | |
1898 | ||
1899 | use_task_css_set_links = true; | |
1900 | ||
1901 | /* | |
1902 | * We need tasklist_lock because RCU is not safe against | |
1903 | * while_each_thread(). Besides, a forking task that has passed | |
1904 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1905 | * is not guaranteed to have its child immediately visible in the | |
1906 | * tasklist if we walk through it with RCU. | |
1907 | */ | |
1908 | read_lock(&tasklist_lock); | |
1909 | do_each_thread(g, p) { | |
afeb0f9f TH |
1910 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1911 | task_css_set(p) != &init_css_set); | |
1912 | ||
1913 | /* | |
1914 | * We should check if the process is exiting, otherwise | |
1915 | * it will race with cgroup_exit() in that the list | |
1916 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1917 | * Do it while holding siglock so that we don't end up |
1918 | * racing against cgroup_exit(). | |
afeb0f9f | 1919 | */ |
f153ad11 | 1920 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1921 | if (!(p->flags & PF_EXITING)) { |
1922 | struct css_set *cset = task_css_set(p); | |
1923 | ||
0de0942d TH |
1924 | if (!css_set_populated(cset)) |
1925 | css_set_update_populated(cset, true); | |
389b9c1b | 1926 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1927 | get_css_set(cset); |
1928 | } | |
f153ad11 | 1929 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1930 | } while_each_thread(g, p); |
1931 | read_unlock(&tasklist_lock); | |
1932 | out_unlock: | |
f0d9a5f1 | 1933 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1934 | } |
ddbcc7e8 | 1935 | |
cc31edce PM |
1936 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1937 | { | |
2d8f243a TH |
1938 | struct cgroup_subsys *ss; |
1939 | int ssid; | |
1940 | ||
d5c419b6 TH |
1941 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1942 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1943 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1944 | INIT_LIST_HEAD(&cgrp->pidlists); |
1945 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1946 | cgrp->self.cgroup = cgrp; |
184faf32 | 1947 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1948 | |
1949 | for_each_subsys(ss, ssid) | |
1950 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1951 | |
1952 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1953 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1954 | } |
c6d57f33 | 1955 | |
3dd06ffa | 1956 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1957 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1958 | { |
3dd06ffa | 1959 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1960 | |
ddbcc7e8 | 1961 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1962 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1963 | cgrp->root = root; |
cc31edce | 1964 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1965 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1966 | |
c6d57f33 PM |
1967 | root->flags = opts->flags; |
1968 | if (opts->release_agent) | |
1969 | strcpy(root->release_agent_path, opts->release_agent); | |
1970 | if (opts->name) | |
1971 | strcpy(root->name, opts->name); | |
2260e7fc | 1972 | if (opts->cpuset_clone_children) |
3dd06ffa | 1973 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1974 | } |
1975 | ||
6e5c8307 | 1976 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1977 | { |
d427dfeb | 1978 | LIST_HEAD(tmp_links); |
3dd06ffa | 1979 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1980 | struct css_set *cset; |
d427dfeb | 1981 | int i, ret; |
2c6ab6d2 | 1982 | |
d427dfeb | 1983 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1984 | |
cf780b7d | 1985 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1986 | if (ret < 0) |
2bd59d48 | 1987 | goto out; |
d427dfeb | 1988 | root_cgrp->id = ret; |
b11cfb58 | 1989 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1990 | |
2aad2a86 TH |
1991 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1992 | GFP_KERNEL); | |
9d755d33 TH |
1993 | if (ret) |
1994 | goto out; | |
1995 | ||
d427dfeb | 1996 | /* |
f0d9a5f1 | 1997 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1998 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1999 | * cgroup_lock, and that's us. Later rebinding may disable |
2000 | * controllers on the default hierarchy and thus create new csets, | |
2001 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 2002 | */ |
04313591 | 2003 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 2004 | if (ret) |
9d755d33 | 2005 | goto cancel_ref; |
ddbcc7e8 | 2006 | |
985ed670 | 2007 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 2008 | if (ret) |
9d755d33 | 2009 | goto cancel_ref; |
ddbcc7e8 | 2010 | |
2bd59d48 TH |
2011 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
2012 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
2013 | root_cgrp); | |
2014 | if (IS_ERR(root->kf_root)) { | |
2015 | ret = PTR_ERR(root->kf_root); | |
2016 | goto exit_root_id; | |
2017 | } | |
2018 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 2019 | |
334c3679 | 2020 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 2021 | if (ret) |
2bd59d48 | 2022 | goto destroy_root; |
ddbcc7e8 | 2023 | |
5df36032 | 2024 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 2025 | if (ret) |
2bd59d48 | 2026 | goto destroy_root; |
ddbcc7e8 | 2027 | |
d427dfeb TH |
2028 | /* |
2029 | * There must be no failure case after here, since rebinding takes | |
2030 | * care of subsystems' refcounts, which are explicitly dropped in | |
2031 | * the failure exit path. | |
2032 | */ | |
2033 | list_add(&root->root_list, &cgroup_roots); | |
2034 | cgroup_root_count++; | |
0df6a63f | 2035 | |
d427dfeb | 2036 | /* |
3dd06ffa | 2037 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2038 | * objects. |
2039 | */ | |
f0d9a5f1 | 2040 | spin_lock_bh(&css_set_lock); |
0de0942d | 2041 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2042 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2043 | if (css_set_populated(cset)) |
2044 | cgroup_update_populated(root_cgrp, true); | |
2045 | } | |
f0d9a5f1 | 2046 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 2047 | |
d5c419b6 | 2048 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2049 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2050 | |
2bd59d48 | 2051 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2052 | ret = 0; |
2bd59d48 | 2053 | goto out; |
d427dfeb | 2054 | |
2bd59d48 TH |
2055 | destroy_root: |
2056 | kernfs_destroy_root(root->kf_root); | |
2057 | root->kf_root = NULL; | |
2058 | exit_root_id: | |
d427dfeb | 2059 | cgroup_exit_root_id(root); |
9d755d33 | 2060 | cancel_ref: |
9a1049da | 2061 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2062 | out: |
d427dfeb TH |
2063 | free_cgrp_cset_links(&tmp_links); |
2064 | return ret; | |
ddbcc7e8 PM |
2065 | } |
2066 | ||
f7e83571 | 2067 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2068 | int flags, const char *unused_dev_name, |
f7e83571 | 2069 | void *data) |
ddbcc7e8 | 2070 | { |
67e9c74b | 2071 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 2072 | struct super_block *pinned_sb = NULL; |
ed82571b | 2073 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
970317aa | 2074 | struct cgroup_subsys *ss; |
3dd06ffa | 2075 | struct cgroup_root *root; |
ddbcc7e8 | 2076 | struct cgroup_sb_opts opts; |
2bd59d48 | 2077 | struct dentry *dentry; |
8e30e2b8 | 2078 | int ret; |
970317aa | 2079 | int i; |
c6b3d5bc | 2080 | bool new_sb; |
ddbcc7e8 | 2081 | |
ed82571b SH |
2082 | get_cgroup_ns(ns); |
2083 | ||
2084 | /* Check if the caller has permission to mount. */ | |
2085 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2086 | put_cgroup_ns(ns); | |
2087 | return ERR_PTR(-EPERM); | |
2088 | } | |
2089 | ||
56fde9e0 TH |
2090 | /* |
2091 | * The first time anyone tries to mount a cgroup, enable the list | |
2092 | * linking each css_set to its tasks and fix up all existing tasks. | |
2093 | */ | |
2094 | if (!use_task_css_set_links) | |
2095 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2096 | |
67e9c74b TH |
2097 | if (is_v2) { |
2098 | if (data) { | |
2099 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
ed82571b | 2100 | put_cgroup_ns(ns); |
67e9c74b TH |
2101 | return ERR_PTR(-EINVAL); |
2102 | } | |
a7165264 | 2103 | cgrp_dfl_visible = true; |
67e9c74b TH |
2104 | root = &cgrp_dfl_root; |
2105 | cgroup_get(&root->cgrp); | |
2106 | goto out_mount; | |
2107 | } | |
2108 | ||
334c3679 | 2109 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2110 | |
2111 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2112 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2113 | if (ret) |
8e30e2b8 | 2114 | goto out_unlock; |
a015edd2 | 2115 | |
970317aa LZ |
2116 | /* |
2117 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2118 | * still be dying after the previous unmount. Let's drain the | |
2119 | * dying subsystems. We just need to ensure that the ones | |
2120 | * unmounted previously finish dying and don't care about new ones | |
2121 | * starting. Testing ref liveliness is good enough. | |
2122 | */ | |
2123 | for_each_subsys(ss, i) { | |
2124 | if (!(opts.subsys_mask & (1 << i)) || | |
2125 | ss->root == &cgrp_dfl_root) | |
2126 | continue; | |
2127 | ||
2128 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2129 | mutex_unlock(&cgroup_mutex); | |
2130 | msleep(10); | |
2131 | ret = restart_syscall(); | |
2132 | goto out_free; | |
2133 | } | |
2134 | cgroup_put(&ss->root->cgrp); | |
2135 | } | |
2136 | ||
985ed670 | 2137 | for_each_root(root) { |
2bd59d48 | 2138 | bool name_match = false; |
3126121f | 2139 | |
3dd06ffa | 2140 | if (root == &cgrp_dfl_root) |
985ed670 | 2141 | continue; |
3126121f | 2142 | |
cf5d5941 | 2143 | /* |
2bd59d48 TH |
2144 | * If we asked for a name then it must match. Also, if |
2145 | * name matches but sybsys_mask doesn't, we should fail. | |
2146 | * Remember whether name matched. | |
cf5d5941 | 2147 | */ |
2bd59d48 TH |
2148 | if (opts.name) { |
2149 | if (strcmp(opts.name, root->name)) | |
2150 | continue; | |
2151 | name_match = true; | |
2152 | } | |
ddbcc7e8 | 2153 | |
c6d57f33 | 2154 | /* |
2bd59d48 TH |
2155 | * If we asked for subsystems (or explicitly for no |
2156 | * subsystems) then they must match. | |
c6d57f33 | 2157 | */ |
2bd59d48 | 2158 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2159 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2160 | if (!name_match) |
2161 | continue; | |
2162 | ret = -EBUSY; | |
2163 | goto out_unlock; | |
2164 | } | |
873fe09e | 2165 | |
7b9a6ba5 TH |
2166 | if (root->flags ^ opts.flags) |
2167 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2168 | |
776f02fa | 2169 | /* |
3a32bd72 LZ |
2170 | * We want to reuse @root whose lifetime is governed by its |
2171 | * ->cgrp. Let's check whether @root is alive and keep it | |
2172 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2173 | * want to block it by pinning the sb so that @root doesn't | |
2174 | * get killed before mount is complete. | |
2175 | * | |
2176 | * With the sb pinned, tryget_live can reliably indicate | |
2177 | * whether @root can be reused. If it's being killed, | |
2178 | * drain it. We can use wait_queue for the wait but this | |
2179 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2180 | */ |
3a32bd72 LZ |
2181 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2182 | if (IS_ERR(pinned_sb) || | |
2183 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2184 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2185 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2186 | deactivate_super(pinned_sb); | |
776f02fa | 2187 | msleep(10); |
a015edd2 TH |
2188 | ret = restart_syscall(); |
2189 | goto out_free; | |
776f02fa | 2190 | } |
ddbcc7e8 | 2191 | |
776f02fa | 2192 | ret = 0; |
2bd59d48 | 2193 | goto out_unlock; |
ddbcc7e8 | 2194 | } |
ddbcc7e8 | 2195 | |
817929ec | 2196 | /* |
172a2c06 TH |
2197 | * No such thing, create a new one. name= matching without subsys |
2198 | * specification is allowed for already existing hierarchies but we | |
2199 | * can't create new one without subsys specification. | |
817929ec | 2200 | */ |
172a2c06 TH |
2201 | if (!opts.subsys_mask && !opts.none) { |
2202 | ret = -EINVAL; | |
2203 | goto out_unlock; | |
817929ec | 2204 | } |
817929ec | 2205 | |
ed82571b SH |
2206 | /* |
2207 | * We know this subsystem has not yet been bound. Users in a non-init | |
2208 | * user namespace may only mount hierarchies with no bound subsystems, | |
2209 | * i.e. 'none,name=user1' | |
2210 | */ | |
2211 | if (!opts.none && !capable(CAP_SYS_ADMIN)) { | |
2212 | ret = -EPERM; | |
2213 | goto out_unlock; | |
2214 | } | |
2215 | ||
172a2c06 TH |
2216 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2217 | if (!root) { | |
2218 | ret = -ENOMEM; | |
2bd59d48 | 2219 | goto out_unlock; |
839ec545 | 2220 | } |
e5f6a860 | 2221 | |
172a2c06 TH |
2222 | init_cgroup_root(root, &opts); |
2223 | ||
35585573 | 2224 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2225 | if (ret) |
2226 | cgroup_free_root(root); | |
fa3ca07e | 2227 | |
8e30e2b8 | 2228 | out_unlock: |
ddbcc7e8 | 2229 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2230 | out_free: |
c6d57f33 PM |
2231 | kfree(opts.release_agent); |
2232 | kfree(opts.name); | |
03b1cde6 | 2233 | |
ed82571b SH |
2234 | if (ret) { |
2235 | put_cgroup_ns(ns); | |
8e30e2b8 | 2236 | return ERR_PTR(ret); |
ed82571b | 2237 | } |
67e9c74b | 2238 | out_mount: |
c9482a5b | 2239 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2240 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2241 | &new_sb); | |
ed82571b SH |
2242 | |
2243 | /* | |
2244 | * In non-init cgroup namespace, instead of root cgroup's | |
2245 | * dentry, we return the dentry corresponding to the | |
2246 | * cgroupns->root_cgrp. | |
2247 | */ | |
2248 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { | |
2249 | struct dentry *nsdentry; | |
2250 | struct cgroup *cgrp; | |
2251 | ||
2252 | mutex_lock(&cgroup_mutex); | |
2253 | spin_lock_bh(&css_set_lock); | |
2254 | ||
2255 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2256 | ||
2257 | spin_unlock_bh(&css_set_lock); | |
2258 | mutex_unlock(&cgroup_mutex); | |
2259 | ||
2260 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); | |
2261 | dput(dentry); | |
2262 | dentry = nsdentry; | |
2263 | } | |
2264 | ||
c6b3d5bc | 2265 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2266 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2267 | |
2268 | /* | |
2269 | * If @pinned_sb, we're reusing an existing root and holding an | |
2270 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2271 | */ | |
2272 | if (pinned_sb) { | |
2273 | WARN_ON(new_sb); | |
2274 | deactivate_super(pinned_sb); | |
2275 | } | |
2276 | ||
ed82571b | 2277 | put_cgroup_ns(ns); |
2bd59d48 TH |
2278 | return dentry; |
2279 | } | |
2280 | ||
2281 | static void cgroup_kill_sb(struct super_block *sb) | |
2282 | { | |
2283 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2284 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2285 | |
9d755d33 TH |
2286 | /* |
2287 | * If @root doesn't have any mounts or children, start killing it. | |
2288 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2289 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2290 | * |
2291 | * And don't kill the default root. | |
9d755d33 | 2292 | */ |
3c606d35 | 2293 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2294 | root == &cgrp_dfl_root) |
9d755d33 TH |
2295 | cgroup_put(&root->cgrp); |
2296 | else | |
2297 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2298 | ||
2bd59d48 | 2299 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2300 | } |
2301 | ||
2302 | static struct file_system_type cgroup_fs_type = { | |
2303 | .name = "cgroup", | |
f7e83571 | 2304 | .mount = cgroup_mount, |
ddbcc7e8 | 2305 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2306 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 PM |
2307 | }; |
2308 | ||
67e9c74b TH |
2309 | static struct file_system_type cgroup2_fs_type = { |
2310 | .name = "cgroup2", | |
2311 | .mount = cgroup_mount, | |
2312 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2313 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b TH |
2314 | }; |
2315 | ||
a79a908f AK |
2316 | static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2317 | struct cgroup_namespace *ns) | |
2318 | { | |
2319 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
2320 | int ret; | |
2321 | ||
2322 | ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); | |
2323 | if (ret < 0 || ret >= buflen) | |
2324 | return NULL; | |
2325 | return buf; | |
2326 | } | |
2327 | ||
2328 | char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, | |
2329 | struct cgroup_namespace *ns) | |
2330 | { | |
2331 | char *ret; | |
2332 | ||
2333 | mutex_lock(&cgroup_mutex); | |
2334 | spin_lock_bh(&css_set_lock); | |
2335 | ||
2336 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2337 | ||
2338 | spin_unlock_bh(&css_set_lock); | |
2339 | mutex_unlock(&cgroup_mutex); | |
2340 | ||
2341 | return ret; | |
2342 | } | |
2343 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2344 | ||
857a2beb | 2345 | /** |
913ffdb5 | 2346 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2347 | * @task: target task |
857a2beb TH |
2348 | * @buf: the buffer to write the path into |
2349 | * @buflen: the length of the buffer | |
2350 | * | |
913ffdb5 TH |
2351 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2352 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2353 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2354 | * cgroup controller callbacks. | |
2355 | * | |
e61734c5 | 2356 | * Return value is the same as kernfs_path(). |
857a2beb | 2357 | */ |
e61734c5 | 2358 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2359 | { |
3dd06ffa | 2360 | struct cgroup_root *root; |
913ffdb5 | 2361 | struct cgroup *cgrp; |
e61734c5 TH |
2362 | int hierarchy_id = 1; |
2363 | char *path = NULL; | |
857a2beb TH |
2364 | |
2365 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2366 | spin_lock_bh(&css_set_lock); |
857a2beb | 2367 | |
913ffdb5 TH |
2368 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2369 | ||
857a2beb TH |
2370 | if (root) { |
2371 | cgrp = task_cgroup_from_root(task, root); | |
a79a908f | 2372 | path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2373 | } else { |
2374 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2375 | if (strlcpy(buf, "/", buflen) < buflen) |
2376 | path = buf; | |
857a2beb TH |
2377 | } |
2378 | ||
f0d9a5f1 | 2379 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2380 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2381 | return path; |
857a2beb | 2382 | } |
913ffdb5 | 2383 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2384 | |
b3dc094e | 2385 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2386 | struct cgroup_taskset { |
b3dc094e TH |
2387 | /* the src and dst cset list running through cset->mg_node */ |
2388 | struct list_head src_csets; | |
2389 | struct list_head dst_csets; | |
2390 | ||
1f7dd3e5 TH |
2391 | /* the subsys currently being processed */ |
2392 | int ssid; | |
2393 | ||
b3dc094e TH |
2394 | /* |
2395 | * Fields for cgroup_taskset_*() iteration. | |
2396 | * | |
2397 | * Before migration is committed, the target migration tasks are on | |
2398 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2399 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2400 | * or ->dst_csets depending on whether migration is committed. | |
2401 | * | |
2402 | * ->cur_csets and ->cur_task point to the current task position | |
2403 | * during iteration. | |
2404 | */ | |
2405 | struct list_head *csets; | |
2406 | struct css_set *cur_cset; | |
2407 | struct task_struct *cur_task; | |
2f7ee569 TH |
2408 | }; |
2409 | ||
adaae5dc TH |
2410 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2411 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2412 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2413 | .csets = &tset.src_csets, \ | |
2414 | } | |
2415 | ||
2416 | /** | |
2417 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2418 | * @task: target task | |
2419 | * @tset: target taskset | |
2420 | * | |
2421 | * Add @task, which is a migration target, to @tset. This function becomes | |
2422 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2423 | * been added as a migration source and @task->cg_list will be moved from | |
2424 | * the css_set's tasks list to mg_tasks one. | |
2425 | */ | |
2426 | static void cgroup_taskset_add(struct task_struct *task, | |
2427 | struct cgroup_taskset *tset) | |
2428 | { | |
2429 | struct css_set *cset; | |
2430 | ||
f0d9a5f1 | 2431 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2432 | |
2433 | /* @task either already exited or can't exit until the end */ | |
2434 | if (task->flags & PF_EXITING) | |
2435 | return; | |
2436 | ||
2437 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2438 | if (list_empty(&task->cg_list)) | |
2439 | return; | |
2440 | ||
2441 | cset = task_css_set(task); | |
2442 | if (!cset->mg_src_cgrp) | |
2443 | return; | |
2444 | ||
2445 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2446 | if (list_empty(&cset->mg_node)) | |
2447 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2448 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2449 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2450 | &tset->dst_csets); | |
2451 | } | |
2452 | ||
2f7ee569 TH |
2453 | /** |
2454 | * cgroup_taskset_first - reset taskset and return the first task | |
2455 | * @tset: taskset of interest | |
1f7dd3e5 | 2456 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2457 | * |
2458 | * @tset iteration is initialized and the first task is returned. | |
2459 | */ | |
1f7dd3e5 TH |
2460 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2461 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2462 | { |
b3dc094e TH |
2463 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2464 | tset->cur_task = NULL; | |
2465 | ||
1f7dd3e5 | 2466 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2467 | } |
2f7ee569 TH |
2468 | |
2469 | /** | |
2470 | * cgroup_taskset_next - iterate to the next task in taskset | |
2471 | * @tset: taskset of interest | |
1f7dd3e5 | 2472 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2473 | * |
2474 | * Return the next task in @tset. Iteration must have been initialized | |
2475 | * with cgroup_taskset_first(). | |
2476 | */ | |
1f7dd3e5 TH |
2477 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2478 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2479 | { |
b3dc094e TH |
2480 | struct css_set *cset = tset->cur_cset; |
2481 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2482 | |
b3dc094e TH |
2483 | while (&cset->mg_node != tset->csets) { |
2484 | if (!task) | |
2485 | task = list_first_entry(&cset->mg_tasks, | |
2486 | struct task_struct, cg_list); | |
2487 | else | |
2488 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2489 | |
b3dc094e TH |
2490 | if (&task->cg_list != &cset->mg_tasks) { |
2491 | tset->cur_cset = cset; | |
2492 | tset->cur_task = task; | |
1f7dd3e5 TH |
2493 | |
2494 | /* | |
2495 | * This function may be called both before and | |
2496 | * after cgroup_taskset_migrate(). The two cases | |
2497 | * can be distinguished by looking at whether @cset | |
2498 | * has its ->mg_dst_cset set. | |
2499 | */ | |
2500 | if (cset->mg_dst_cset) | |
2501 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2502 | else | |
2503 | *dst_cssp = cset->subsys[tset->ssid]; | |
2504 | ||
b3dc094e TH |
2505 | return task; |
2506 | } | |
2f7ee569 | 2507 | |
b3dc094e TH |
2508 | cset = list_next_entry(cset, mg_node); |
2509 | task = NULL; | |
2510 | } | |
2f7ee569 | 2511 | |
b3dc094e | 2512 | return NULL; |
2f7ee569 | 2513 | } |
2f7ee569 | 2514 | |
adaae5dc | 2515 | /** |
37ff9f8f | 2516 | * cgroup_taskset_migrate - migrate a taskset |
adaae5dc | 2517 | * @tset: taget taskset |
37ff9f8f | 2518 | * @root: cgroup root the migration is taking place on |
adaae5dc | 2519 | * |
37ff9f8f TH |
2520 | * Migrate tasks in @tset as setup by migration preparation functions. |
2521 | * This function fails iff one of the ->can_attach callbacks fails and | |
2522 | * guarantees that either all or none of the tasks in @tset are migrated. | |
2523 | * @tset is consumed regardless of success. | |
adaae5dc TH |
2524 | */ |
2525 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
37ff9f8f | 2526 | struct cgroup_root *root) |
adaae5dc | 2527 | { |
37ff9f8f | 2528 | struct cgroup_subsys *ss; |
adaae5dc TH |
2529 | struct task_struct *task, *tmp_task; |
2530 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2531 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2532 | |
2533 | /* methods shouldn't be called if no task is actually migrating */ | |
2534 | if (list_empty(&tset->src_csets)) | |
2535 | return 0; | |
2536 | ||
2537 | /* check that we can legitimately attach to the cgroup */ | |
37ff9f8f TH |
2538 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2539 | if (ss->can_attach) { | |
2540 | tset->ssid = ssid; | |
2541 | ret = ss->can_attach(tset); | |
adaae5dc | 2542 | if (ret) { |
37ff9f8f | 2543 | failed_ssid = ssid; |
adaae5dc TH |
2544 | goto out_cancel_attach; |
2545 | } | |
2546 | } | |
37ff9f8f | 2547 | } while_each_subsys_mask(); |
adaae5dc TH |
2548 | |
2549 | /* | |
2550 | * Now that we're guaranteed success, proceed to move all tasks to | |
2551 | * the new cgroup. There are no failure cases after here, so this | |
2552 | * is the commit point. | |
2553 | */ | |
f0d9a5f1 | 2554 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2555 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2556 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2557 | struct css_set *from_cset = task_css_set(task); | |
2558 | struct css_set *to_cset = cset->mg_dst_cset; | |
2559 | ||
2560 | get_css_set(to_cset); | |
2561 | css_set_move_task(task, from_cset, to_cset, true); | |
2562 | put_css_set_locked(from_cset); | |
2563 | } | |
adaae5dc | 2564 | } |
f0d9a5f1 | 2565 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2566 | |
2567 | /* | |
2568 | * Migration is committed, all target tasks are now on dst_csets. | |
2569 | * Nothing is sensitive to fork() after this point. Notify | |
2570 | * controllers that migration is complete. | |
2571 | */ | |
2572 | tset->csets = &tset->dst_csets; | |
2573 | ||
37ff9f8f TH |
2574 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2575 | if (ss->attach) { | |
2576 | tset->ssid = ssid; | |
2577 | ss->attach(tset); | |
1f7dd3e5 | 2578 | } |
37ff9f8f | 2579 | } while_each_subsys_mask(); |
adaae5dc TH |
2580 | |
2581 | ret = 0; | |
2582 | goto out_release_tset; | |
2583 | ||
2584 | out_cancel_attach: | |
37ff9f8f TH |
2585 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2586 | if (ssid == failed_ssid) | |
adaae5dc | 2587 | break; |
37ff9f8f TH |
2588 | if (ss->cancel_attach) { |
2589 | tset->ssid = ssid; | |
2590 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2591 | } |
37ff9f8f | 2592 | } while_each_subsys_mask(); |
adaae5dc | 2593 | out_release_tset: |
f0d9a5f1 | 2594 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2595 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2596 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2597 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2598 | list_del_init(&cset->mg_node); | |
2599 | } | |
f0d9a5f1 | 2600 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2601 | return ret; |
2602 | } | |
2603 | ||
6c694c88 TH |
2604 | /** |
2605 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2606 | * @dst_cgrp: destination cgroup to test | |
2607 | * | |
2608 | * On the default hierarchy, except for the root, subtree_control must be | |
2609 | * zero for migration destination cgroups with tasks so that child cgroups | |
2610 | * don't compete against tasks. | |
2611 | */ | |
2612 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) | |
2613 | { | |
2614 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2615 | !dst_cgrp->subtree_control; | |
2616 | } | |
2617 | ||
a043e3b2 | 2618 | /** |
1958d2d5 TH |
2619 | * cgroup_migrate_finish - cleanup after attach |
2620 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2621 | * |
1958d2d5 TH |
2622 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2623 | * those functions for details. | |
74a1166d | 2624 | */ |
1958d2d5 | 2625 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2626 | { |
1958d2d5 | 2627 | struct css_set *cset, *tmp_cset; |
74a1166d | 2628 | |
1958d2d5 TH |
2629 | lockdep_assert_held(&cgroup_mutex); |
2630 | ||
f0d9a5f1 | 2631 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2632 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2633 | cset->mg_src_cgrp = NULL; | |
e4857982 | 2634 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2635 | cset->mg_dst_cset = NULL; |
2636 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2637 | put_css_set_locked(cset); |
1958d2d5 | 2638 | } |
f0d9a5f1 | 2639 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2640 | } |
2641 | ||
2642 | /** | |
2643 | * cgroup_migrate_add_src - add a migration source css_set | |
2644 | * @src_cset: the source css_set to add | |
2645 | * @dst_cgrp: the destination cgroup | |
2646 | * @preloaded_csets: list of preloaded css_sets | |
2647 | * | |
2648 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2649 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2650 | * up by cgroup_migrate_finish(). | |
2651 | * | |
1ed13287 TH |
2652 | * This function may be called without holding cgroup_threadgroup_rwsem |
2653 | * even if the target is a process. Threads may be created and destroyed | |
2654 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2655 | * into play and the preloaded css_sets are guaranteed to cover all | |
2656 | * migrations. | |
1958d2d5 TH |
2657 | */ |
2658 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2659 | struct cgroup *dst_cgrp, | |
2660 | struct list_head *preloaded_csets) | |
2661 | { | |
2662 | struct cgroup *src_cgrp; | |
2663 | ||
2664 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2665 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2666 | |
2b021cbf TH |
2667 | /* |
2668 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2669 | * and doesn't contain any migratable tasks. Ignore it early so | |
2670 | * that the rest of migration path doesn't get confused by it. | |
2671 | */ | |
2672 | if (src_cset->dead) | |
2673 | return; | |
2674 | ||
1958d2d5 TH |
2675 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2676 | ||
1958d2d5 TH |
2677 | if (!list_empty(&src_cset->mg_preload_node)) |
2678 | return; | |
2679 | ||
2680 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2681 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2682 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2683 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2684 | ||
2685 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2686 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 TH |
2687 | get_css_set(src_cset); |
2688 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2689 | } | |
2690 | ||
2691 | /** | |
2692 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1958d2d5 TH |
2693 | * @preloaded_csets: list of preloaded source css_sets |
2694 | * | |
e4857982 TH |
2695 | * Tasks are about to be moved and all the source css_sets have been |
2696 | * preloaded to @preloaded_csets. This function looks up and pins all | |
2697 | * destination css_sets, links each to its source, and append them to | |
2698 | * @preloaded_csets. | |
1958d2d5 TH |
2699 | * |
2700 | * This function must be called after cgroup_migrate_add_src() has been | |
2701 | * called on each migration source css_set. After migration is performed | |
2702 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2703 | * @preloaded_csets. | |
2704 | */ | |
e4857982 | 2705 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
1958d2d5 TH |
2706 | { |
2707 | LIST_HEAD(csets); | |
f817de98 | 2708 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2709 | |
2710 | lockdep_assert_held(&cgroup_mutex); | |
2711 | ||
2712 | /* look up the dst cset for each src cset and link it to src */ | |
f817de98 | 2713 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2714 | struct css_set *dst_cset; |
2715 | ||
e4857982 | 2716 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2717 | if (!dst_cset) |
2718 | goto err; | |
2719 | ||
2720 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2721 | |
2722 | /* | |
2723 | * If src cset equals dst, it's noop. Drop the src. | |
2724 | * cgroup_migrate() will skip the cset too. Note that we | |
2725 | * can't handle src == dst as some nodes are used by both. | |
2726 | */ | |
2727 | if (src_cset == dst_cset) { | |
2728 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2729 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2730 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2731 | put_css_set(src_cset); |
2732 | put_css_set(dst_cset); | |
f817de98 TH |
2733 | continue; |
2734 | } | |
2735 | ||
1958d2d5 TH |
2736 | src_cset->mg_dst_cset = dst_cset; |
2737 | ||
2738 | if (list_empty(&dst_cset->mg_preload_node)) | |
2739 | list_add(&dst_cset->mg_preload_node, &csets); | |
2740 | else | |
a25eb52e | 2741 | put_css_set(dst_cset); |
1958d2d5 TH |
2742 | } |
2743 | ||
f817de98 | 2744 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2745 | return 0; |
2746 | err: | |
2747 | cgroup_migrate_finish(&csets); | |
2748 | return -ENOMEM; | |
2749 | } | |
2750 | ||
2751 | /** | |
2752 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2753 | * @leader: the leader of the process or the task to migrate |
2754 | * @threadgroup: whether @leader points to the whole process or a single task | |
37ff9f8f | 2755 | * @root: cgroup root migration is taking place on |
1958d2d5 | 2756 | * |
37ff9f8f TH |
2757 | * Migrate a process or task denoted by @leader. If migrating a process, |
2758 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2759 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2760 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2761 | * function and following up with cgroup_migrate_finish(). | |
2762 | * | |
2763 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2764 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2765 | * failure, when migrating multiple targets, the success or failure can be | |
2766 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2767 | * actually starting migrating. | |
2768 | */ | |
9af2ec45 | 2769 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
37ff9f8f | 2770 | struct cgroup_root *root) |
74a1166d | 2771 | { |
adaae5dc TH |
2772 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2773 | struct task_struct *task; | |
74a1166d | 2774 | |
fb5d2b4c MSB |
2775 | /* |
2776 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2777 | * already PF_EXITING could be freed from underneath us unless we | |
2778 | * take an rcu_read_lock. | |
2779 | */ | |
f0d9a5f1 | 2780 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2781 | rcu_read_lock(); |
9db8de37 | 2782 | task = leader; |
74a1166d | 2783 | do { |
adaae5dc | 2784 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2785 | if (!threadgroup) |
2786 | break; | |
9db8de37 | 2787 | } while_each_thread(leader, task); |
fb5d2b4c | 2788 | rcu_read_unlock(); |
f0d9a5f1 | 2789 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2790 | |
37ff9f8f | 2791 | return cgroup_taskset_migrate(&tset, root); |
74a1166d BB |
2792 | } |
2793 | ||
1958d2d5 TH |
2794 | /** |
2795 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2796 | * @dst_cgrp: the cgroup to attach to | |
2797 | * @leader: the task or the leader of the threadgroup to be attached | |
2798 | * @threadgroup: attach the whole threadgroup? | |
2799 | * | |
1ed13287 | 2800 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2801 | */ |
2802 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2803 | struct task_struct *leader, bool threadgroup) | |
2804 | { | |
2805 | LIST_HEAD(preloaded_csets); | |
2806 | struct task_struct *task; | |
2807 | int ret; | |
2808 | ||
6c694c88 TH |
2809 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2810 | return -EBUSY; | |
2811 | ||
1958d2d5 | 2812 | /* look up all src csets */ |
f0d9a5f1 | 2813 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2814 | rcu_read_lock(); |
2815 | task = leader; | |
2816 | do { | |
2817 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2818 | &preloaded_csets); | |
2819 | if (!threadgroup) | |
2820 | break; | |
2821 | } while_each_thread(leader, task); | |
2822 | rcu_read_unlock(); | |
f0d9a5f1 | 2823 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2824 | |
2825 | /* prepare dst csets and commit */ | |
e4857982 | 2826 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
1958d2d5 | 2827 | if (!ret) |
37ff9f8f | 2828 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
1958d2d5 TH |
2829 | |
2830 | cgroup_migrate_finish(&preloaded_csets); | |
2831 | return ret; | |
74a1166d BB |
2832 | } |
2833 | ||
187fe840 TH |
2834 | static int cgroup_procs_write_permission(struct task_struct *task, |
2835 | struct cgroup *dst_cgrp, | |
2836 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2837 | { |
2838 | const struct cred *cred = current_cred(); | |
2839 | const struct cred *tcred = get_task_cred(task); | |
2840 | int ret = 0; | |
2841 | ||
2842 | /* | |
2843 | * even if we're attaching all tasks in the thread group, we only | |
2844 | * need to check permissions on one of them. | |
2845 | */ | |
2846 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2847 | !uid_eq(cred->euid, tcred->uid) && | |
2848 | !uid_eq(cred->euid, tcred->suid)) | |
2849 | ret = -EACCES; | |
2850 | ||
187fe840 TH |
2851 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2852 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2853 | struct cgroup *cgrp; | |
2854 | struct inode *inode; | |
2855 | ||
f0d9a5f1 | 2856 | spin_lock_bh(&css_set_lock); |
187fe840 | 2857 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2858 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2859 | |
2860 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2861 | cgrp = cgroup_parent(cgrp); | |
2862 | ||
2863 | ret = -ENOMEM; | |
6f60eade | 2864 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2865 | if (inode) { |
2866 | ret = inode_permission(inode, MAY_WRITE); | |
2867 | iput(inode); | |
2868 | } | |
2869 | } | |
2870 | ||
dedf22e9 TH |
2871 | put_cred(tcred); |
2872 | return ret; | |
2873 | } | |
2874 | ||
74a1166d BB |
2875 | /* |
2876 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2877 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2878 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2879 | */ |
acbef755 TH |
2880 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2881 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2882 | { |
bbcb81d0 | 2883 | struct task_struct *tsk; |
5cf1cacb | 2884 | struct cgroup_subsys *ss; |
e76ecaee | 2885 | struct cgroup *cgrp; |
acbef755 | 2886 | pid_t pid; |
5cf1cacb | 2887 | int ssid, ret; |
bbcb81d0 | 2888 | |
acbef755 TH |
2889 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2890 | return -EINVAL; | |
2891 | ||
945ba199 | 2892 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2893 | if (!cgrp) |
74a1166d BB |
2894 | return -ENODEV; |
2895 | ||
3014dde7 | 2896 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2897 | rcu_read_lock(); |
bbcb81d0 | 2898 | if (pid) { |
73507f33 | 2899 | tsk = find_task_by_vpid(pid); |
74a1166d | 2900 | if (!tsk) { |
dd4b0a46 | 2901 | ret = -ESRCH; |
3014dde7 | 2902 | goto out_unlock_rcu; |
bbcb81d0 | 2903 | } |
dedf22e9 | 2904 | } else { |
b78949eb | 2905 | tsk = current; |
dedf22e9 | 2906 | } |
cd3d0952 TH |
2907 | |
2908 | if (threadgroup) | |
b78949eb | 2909 | tsk = tsk->group_leader; |
c4c27fbd MG |
2910 | |
2911 | /* | |
14a40ffc | 2912 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2913 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2914 | * with no rt_runtime allocated. Just say no. | |
2915 | */ | |
14a40ffc | 2916 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2917 | ret = -EINVAL; |
3014dde7 | 2918 | goto out_unlock_rcu; |
c4c27fbd MG |
2919 | } |
2920 | ||
b78949eb MSB |
2921 | get_task_struct(tsk); |
2922 | rcu_read_unlock(); | |
2923 | ||
187fe840 | 2924 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2925 | if (!ret) |
2926 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2927 | |
f9f9e7b7 | 2928 | put_task_struct(tsk); |
3014dde7 TH |
2929 | goto out_unlock_threadgroup; |
2930 | ||
2931 | out_unlock_rcu: | |
2932 | rcu_read_unlock(); | |
2933 | out_unlock_threadgroup: | |
2934 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2935 | for_each_subsys(ss, ssid) |
2936 | if (ss->post_attach) | |
2937 | ss->post_attach(); | |
e76ecaee | 2938 | cgroup_kn_unlock(of->kn); |
acbef755 | 2939 | return ret ?: nbytes; |
bbcb81d0 PM |
2940 | } |
2941 | ||
7ae1bad9 TH |
2942 | /** |
2943 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2944 | * @from: attach to all cgroups of a given task | |
2945 | * @tsk: the task to be attached | |
2946 | */ | |
2947 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2948 | { | |
3dd06ffa | 2949 | struct cgroup_root *root; |
7ae1bad9 TH |
2950 | int retval = 0; |
2951 | ||
47cfcd09 | 2952 | mutex_lock(&cgroup_mutex); |
985ed670 | 2953 | for_each_root(root) { |
96d365e0 TH |
2954 | struct cgroup *from_cgrp; |
2955 | ||
3dd06ffa | 2956 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2957 | continue; |
2958 | ||
f0d9a5f1 | 2959 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2960 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2961 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2962 | |
6f4b7e63 | 2963 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2964 | if (retval) |
2965 | break; | |
2966 | } | |
47cfcd09 | 2967 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2968 | |
2969 | return retval; | |
2970 | } | |
2971 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2972 | ||
acbef755 TH |
2973 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2974 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2975 | { |
acbef755 | 2976 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2977 | } |
2978 | ||
acbef755 TH |
2979 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2980 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2981 | { |
acbef755 | 2982 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2983 | } |
2984 | ||
451af504 TH |
2985 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2986 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2987 | { |
e76ecaee | 2988 | struct cgroup *cgrp; |
5f469907 | 2989 | |
e76ecaee | 2990 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2991 | |
945ba199 | 2992 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2993 | if (!cgrp) |
e788e066 | 2994 | return -ENODEV; |
69e943b7 | 2995 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2996 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2997 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2998 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2999 | cgroup_kn_unlock(of->kn); |
451af504 | 3000 | return nbytes; |
e788e066 PM |
3001 | } |
3002 | ||
2da8ca82 | 3003 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 3004 | { |
2da8ca82 | 3005 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 3006 | |
46cfeb04 | 3007 | spin_lock(&release_agent_path_lock); |
e788e066 | 3008 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 3009 | spin_unlock(&release_agent_path_lock); |
e788e066 | 3010 | seq_putc(seq, '\n'); |
e788e066 PM |
3011 | return 0; |
3012 | } | |
3013 | ||
2da8ca82 | 3014 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 3015 | { |
c1d5d42e | 3016 | seq_puts(seq, "0\n"); |
e788e066 PM |
3017 | return 0; |
3018 | } | |
3019 | ||
6e5c8307 | 3020 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 3021 | { |
f8f22e53 TH |
3022 | struct cgroup_subsys *ss; |
3023 | bool printed = false; | |
3024 | int ssid; | |
a742c59d | 3025 | |
b4e0eeaf | 3026 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
3027 | if (printed) |
3028 | seq_putc(seq, ' '); | |
3029 | seq_printf(seq, "%s", ss->name); | |
3030 | printed = true; | |
b4e0eeaf | 3031 | } while_each_subsys_mask(); |
f8f22e53 TH |
3032 | if (printed) |
3033 | seq_putc(seq, '\n'); | |
355e0c48 PM |
3034 | } |
3035 | ||
f8f22e53 TH |
3036 | /* show controllers which are enabled from the parent */ |
3037 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3038 | { |
f8f22e53 TH |
3039 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3040 | ||
5531dc91 | 3041 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 3042 | return 0; |
ddbcc7e8 PM |
3043 | } |
3044 | ||
f8f22e53 TH |
3045 | /* show controllers which are enabled for a given cgroup's children */ |
3046 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3047 | { |
f8f22e53 TH |
3048 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3049 | ||
667c2491 | 3050 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
3051 | return 0; |
3052 | } | |
3053 | ||
3054 | /** | |
3055 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
3056 | * @cgrp: root of the subtree to update csses for | |
3057 | * | |
54962604 TH |
3058 | * @cgrp's control masks have changed and its subtree's css associations |
3059 | * need to be updated accordingly. This function looks up all css_sets | |
3060 | * which are attached to the subtree, creates the matching updated css_sets | |
3061 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
3062 | */ |
3063 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
3064 | { | |
3065 | LIST_HEAD(preloaded_csets); | |
10265075 | 3066 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
3067 | struct cgroup_subsys_state *d_css; |
3068 | struct cgroup *dsct; | |
f8f22e53 TH |
3069 | struct css_set *src_cset; |
3070 | int ret; | |
3071 | ||
f8f22e53 TH |
3072 | lockdep_assert_held(&cgroup_mutex); |
3073 | ||
3014dde7 TH |
3074 | percpu_down_write(&cgroup_threadgroup_rwsem); |
3075 | ||
f8f22e53 | 3076 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 3077 | spin_lock_bh(&css_set_lock); |
54962604 | 3078 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
3079 | struct cgrp_cset_link *link; |
3080 | ||
54962604 | 3081 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 3082 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
3083 | &preloaded_csets); |
3084 | } | |
f0d9a5f1 | 3085 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
3086 | |
3087 | /* NULL dst indicates self on default hierarchy */ | |
e4857982 | 3088 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
f8f22e53 TH |
3089 | if (ret) |
3090 | goto out_finish; | |
3091 | ||
f0d9a5f1 | 3092 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 3093 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 3094 | struct task_struct *task, *ntask; |
f8f22e53 TH |
3095 | |
3096 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
3097 | if (!src_cset->mg_src_cgrp) | |
3098 | break; | |
3099 | ||
10265075 TH |
3100 | /* all tasks in src_csets need to be migrated */ |
3101 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
3102 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 3103 | } |
f0d9a5f1 | 3104 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 3105 | |
37ff9f8f | 3106 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
f8f22e53 TH |
3107 | out_finish: |
3108 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 3109 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
3110 | return ret; |
3111 | } | |
3112 | ||
1b9b96a1 | 3113 | /** |
945ba199 | 3114 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 3115 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
3116 | * |
3117 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
3118 | * controller while the previous css is still around. This function grabs |
3119 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 3120 | */ |
945ba199 TH |
3121 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
3122 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
3123 | { |
3124 | struct cgroup *dsct; | |
ce3f1d9d | 3125 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
3126 | struct cgroup_subsys *ss; |
3127 | int ssid; | |
3128 | ||
945ba199 TH |
3129 | restart: |
3130 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 3131 | |
ce3f1d9d | 3132 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
3133 | for_each_subsys(ss, ssid) { |
3134 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3135 | DEFINE_WAIT(wait); | |
3136 | ||
ce3f1d9d | 3137 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
3138 | continue; |
3139 | ||
3140 | cgroup_get(dsct); | |
3141 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
3142 | TASK_UNINTERRUPTIBLE); | |
3143 | ||
3144 | mutex_unlock(&cgroup_mutex); | |
3145 | schedule(); | |
3146 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
3147 | |
3148 | cgroup_put(dsct); | |
945ba199 | 3149 | goto restart; |
1b9b96a1 TH |
3150 | } |
3151 | } | |
1b9b96a1 TH |
3152 | } |
3153 | ||
15a27c36 TH |
3154 | /** |
3155 | * cgroup_save_control - save control masks of a subtree | |
3156 | * @cgrp: root of the target subtree | |
3157 | * | |
3158 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
3159 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3160 | */ | |
3161 | static void cgroup_save_control(struct cgroup *cgrp) | |
3162 | { | |
3163 | struct cgroup *dsct; | |
3164 | struct cgroup_subsys_state *d_css; | |
3165 | ||
3166 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3167 | dsct->old_subtree_control = dsct->subtree_control; | |
3168 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
3169 | } | |
3170 | } | |
3171 | ||
3172 | /** | |
3173 | * cgroup_propagate_control - refresh control masks of a subtree | |
3174 | * @cgrp: root of the target subtree | |
3175 | * | |
3176 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3177 | * ->subtree_control and propagate controller availability through the | |
3178 | * subtree so that descendants don't have unavailable controllers enabled. | |
3179 | */ | |
3180 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3181 | { | |
3182 | struct cgroup *dsct; | |
3183 | struct cgroup_subsys_state *d_css; | |
3184 | ||
3185 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3186 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3187 | dsct->subtree_ss_mask = |
3188 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3189 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3190 | } |
3191 | } | |
3192 | ||
3193 | /** | |
3194 | * cgroup_restore_control - restore control masks of a subtree | |
3195 | * @cgrp: root of the target subtree | |
3196 | * | |
3197 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3198 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3199 | */ | |
3200 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3201 | { | |
3202 | struct cgroup *dsct; | |
3203 | struct cgroup_subsys_state *d_css; | |
3204 | ||
3205 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3206 | dsct->subtree_control = dsct->old_subtree_control; | |
3207 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3208 | } | |
3209 | } | |
3210 | ||
f6d635ad TH |
3211 | static bool css_visible(struct cgroup_subsys_state *css) |
3212 | { | |
3213 | struct cgroup_subsys *ss = css->ss; | |
3214 | struct cgroup *cgrp = css->cgroup; | |
3215 | ||
3216 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3217 | return true; | |
3218 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3219 | return false; | |
3220 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3221 | } | |
3222 | ||
bdb53bd7 TH |
3223 | /** |
3224 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3225 | * @cgrp: root of the target subtree |
bdb53bd7 | 3226 | * |
ce3f1d9d | 3227 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3228 | * visible. A css is created invisible if it's being implicitly enabled |
3229 | * through dependency. An invisible css is made visible when the userland | |
3230 | * explicitly enables it. | |
3231 | * | |
3232 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3233 | * been processed already aren't cleaned up. The caller is responsible for | |
3234 | * cleaning up with cgroup_apply_control_disble(). | |
3235 | */ | |
3236 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3237 | { | |
3238 | struct cgroup *dsct; | |
ce3f1d9d | 3239 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3240 | struct cgroup_subsys *ss; |
3241 | int ssid, ret; | |
3242 | ||
ce3f1d9d | 3243 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3244 | for_each_subsys(ss, ssid) { |
3245 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3246 | ||
945ba199 TH |
3247 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3248 | ||
bdb53bd7 TH |
3249 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3250 | continue; | |
3251 | ||
3252 | if (!css) { | |
3253 | css = css_create(dsct, ss); | |
3254 | if (IS_ERR(css)) | |
3255 | return PTR_ERR(css); | |
3256 | } | |
3257 | ||
f6d635ad | 3258 | if (css_visible(css)) { |
334c3679 | 3259 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3260 | if (ret) |
3261 | return ret; | |
3262 | } | |
3263 | } | |
3264 | } | |
3265 | ||
3266 | return 0; | |
3267 | } | |
3268 | ||
12b3bb6a TH |
3269 | /** |
3270 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3271 | * @cgrp: root of the target subtree |
12b3bb6a | 3272 | * |
ce3f1d9d | 3273 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3274 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3275 | * | |
3276 | * A css is hidden when the userland requests it to be disabled while other | |
3277 | * subsystems are still depending on it. The css must not actively control | |
3278 | * resources and be in the vanilla state if it's made visible again later. | |
3279 | * Controllers which may be depended upon should provide ->css_reset() for | |
3280 | * this purpose. | |
3281 | */ | |
3282 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3283 | { | |
3284 | struct cgroup *dsct; | |
ce3f1d9d | 3285 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3286 | struct cgroup_subsys *ss; |
3287 | int ssid; | |
3288 | ||
ce3f1d9d | 3289 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3290 | for_each_subsys(ss, ssid) { |
3291 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3292 | ||
945ba199 TH |
3293 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3294 | ||
12b3bb6a TH |
3295 | if (!css) |
3296 | continue; | |
3297 | ||
334c3679 TH |
3298 | if (css->parent && |
3299 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3300 | kill_css(css); |
f6d635ad | 3301 | } else if (!css_visible(css)) { |
334c3679 | 3302 | css_clear_dir(css); |
12b3bb6a TH |
3303 | if (ss->css_reset) |
3304 | ss->css_reset(css); | |
3305 | } | |
3306 | } | |
3307 | } | |
3308 | } | |
3309 | ||
f7b2814b TH |
3310 | /** |
3311 | * cgroup_apply_control - apply control mask updates to the subtree | |
3312 | * @cgrp: root of the target subtree | |
3313 | * | |
3314 | * subsystems can be enabled and disabled in a subtree using the following | |
3315 | * steps. | |
3316 | * | |
3317 | * 1. Call cgroup_save_control() to stash the current state. | |
3318 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3319 | * 3. Call cgroup_apply_control() to apply the changes. | |
3320 | * 4. Optionally perform other related operations. | |
3321 | * 5. Call cgroup_finalize_control() to finish up. | |
3322 | * | |
3323 | * This function implements step 3 and propagates the mask changes | |
3324 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3325 | * process migrations. | |
3326 | */ | |
3327 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3328 | { | |
3329 | int ret; | |
3330 | ||
3331 | cgroup_propagate_control(cgrp); | |
3332 | ||
3333 | ret = cgroup_apply_control_enable(cgrp); | |
3334 | if (ret) | |
3335 | return ret; | |
3336 | ||
3337 | /* | |
3338 | * At this point, cgroup_e_css() results reflect the new csses | |
3339 | * making the following cgroup_update_dfl_csses() properly update | |
3340 | * css associations of all tasks in the subtree. | |
3341 | */ | |
3342 | ret = cgroup_update_dfl_csses(cgrp); | |
3343 | if (ret) | |
3344 | return ret; | |
3345 | ||
3346 | return 0; | |
3347 | } | |
3348 | ||
3349 | /** | |
3350 | * cgroup_finalize_control - finalize control mask update | |
3351 | * @cgrp: root of the target subtree | |
3352 | * @ret: the result of the update | |
3353 | * | |
3354 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3355 | */ | |
3356 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3357 | { | |
3358 | if (ret) { | |
3359 | cgroup_restore_control(cgrp); | |
3360 | cgroup_propagate_control(cgrp); | |
3361 | } | |
3362 | ||
3363 | cgroup_apply_control_disable(cgrp); | |
3364 | } | |
3365 | ||
f8f22e53 | 3366 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3367 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3368 | char *buf, size_t nbytes, | |
3369 | loff_t off) | |
f8f22e53 | 3370 | { |
6e5c8307 | 3371 | u16 enable = 0, disable = 0; |
a9746d8d | 3372 | struct cgroup *cgrp, *child; |
f8f22e53 | 3373 | struct cgroup_subsys *ss; |
451af504 | 3374 | char *tok; |
f8f22e53 TH |
3375 | int ssid, ret; |
3376 | ||
3377 | /* | |
d37167ab TH |
3378 | * Parse input - space separated list of subsystem names prefixed |
3379 | * with either + or -. | |
f8f22e53 | 3380 | */ |
451af504 TH |
3381 | buf = strstrip(buf); |
3382 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3383 | if (tok[0] == '\0') |
3384 | continue; | |
a7165264 | 3385 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3386 | if (!cgroup_ssid_enabled(ssid) || |
3387 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3388 | continue; |
3389 | ||
3390 | if (*tok == '+') { | |
7d331fa9 TH |
3391 | enable |= 1 << ssid; |
3392 | disable &= ~(1 << ssid); | |
f8f22e53 | 3393 | } else if (*tok == '-') { |
7d331fa9 TH |
3394 | disable |= 1 << ssid; |
3395 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3396 | } else { |
3397 | return -EINVAL; | |
3398 | } | |
3399 | break; | |
b4e0eeaf | 3400 | } while_each_subsys_mask(); |
f8f22e53 TH |
3401 | if (ssid == CGROUP_SUBSYS_COUNT) |
3402 | return -EINVAL; | |
3403 | } | |
3404 | ||
945ba199 | 3405 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3406 | if (!cgrp) |
3407 | return -ENODEV; | |
f8f22e53 TH |
3408 | |
3409 | for_each_subsys(ss, ssid) { | |
3410 | if (enable & (1 << ssid)) { | |
667c2491 | 3411 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3412 | enable &= ~(1 << ssid); |
3413 | continue; | |
3414 | } | |
3415 | ||
5531dc91 | 3416 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3417 | ret = -ENOENT; |
3418 | goto out_unlock; | |
3419 | } | |
f8f22e53 | 3420 | } else if (disable & (1 << ssid)) { |
667c2491 | 3421 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3422 | disable &= ~(1 << ssid); |
3423 | continue; | |
3424 | } | |
3425 | ||
3426 | /* a child has it enabled? */ | |
3427 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3428 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3429 | ret = -EBUSY; |
ddab2b6e | 3430 | goto out_unlock; |
f8f22e53 TH |
3431 | } |
3432 | } | |
3433 | } | |
3434 | } | |
3435 | ||
3436 | if (!enable && !disable) { | |
3437 | ret = 0; | |
ddab2b6e | 3438 | goto out_unlock; |
f8f22e53 TH |
3439 | } |
3440 | ||
3441 | /* | |
667c2491 | 3442 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3443 | * with tasks so that child cgroups don't compete against tasks. |
3444 | */ | |
d51f39b0 | 3445 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3446 | ret = -EBUSY; |
3447 | goto out_unlock; | |
3448 | } | |
3449 | ||
15a27c36 TH |
3450 | /* save and update control masks and prepare csses */ |
3451 | cgroup_save_control(cgrp); | |
f63070d3 | 3452 | |
15a27c36 TH |
3453 | cgrp->subtree_control |= enable; |
3454 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3455 | |
f7b2814b | 3456 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3457 | |
f7b2814b | 3458 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3459 | |
3460 | kernfs_activate(cgrp->kn); | |
3461 | ret = 0; | |
3462 | out_unlock: | |
a9746d8d | 3463 | cgroup_kn_unlock(of->kn); |
451af504 | 3464 | return ret ?: nbytes; |
f8f22e53 TH |
3465 | } |
3466 | ||
4a07c222 | 3467 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3468 | { |
4a07c222 | 3469 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3470 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3471 | return 0; |
3472 | } | |
3473 | ||
2bd59d48 TH |
3474 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3475 | size_t nbytes, loff_t off) | |
355e0c48 | 3476 | { |
2bd59d48 TH |
3477 | struct cgroup *cgrp = of->kn->parent->priv; |
3478 | struct cftype *cft = of->kn->priv; | |
3479 | struct cgroup_subsys_state *css; | |
a742c59d | 3480 | int ret; |
355e0c48 | 3481 | |
b4168640 TH |
3482 | if (cft->write) |
3483 | return cft->write(of, buf, nbytes, off); | |
3484 | ||
2bd59d48 TH |
3485 | /* |
3486 | * kernfs guarantees that a file isn't deleted with operations in | |
3487 | * flight, which means that the matching css is and stays alive and | |
3488 | * doesn't need to be pinned. The RCU locking is not necessary | |
3489 | * either. It's just for the convenience of using cgroup_css(). | |
3490 | */ | |
3491 | rcu_read_lock(); | |
3492 | css = cgroup_css(cgrp, cft->ss); | |
3493 | rcu_read_unlock(); | |
a742c59d | 3494 | |
451af504 | 3495 | if (cft->write_u64) { |
a742c59d TH |
3496 | unsigned long long v; |
3497 | ret = kstrtoull(buf, 0, &v); | |
3498 | if (!ret) | |
3499 | ret = cft->write_u64(css, cft, v); | |
3500 | } else if (cft->write_s64) { | |
3501 | long long v; | |
3502 | ret = kstrtoll(buf, 0, &v); | |
3503 | if (!ret) | |
3504 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3505 | } else { |
a742c59d | 3506 | ret = -EINVAL; |
e73d2c61 | 3507 | } |
2bd59d48 | 3508 | |
a742c59d | 3509 | return ret ?: nbytes; |
355e0c48 PM |
3510 | } |
3511 | ||
6612f05b | 3512 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3513 | { |
2bd59d48 | 3514 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3515 | } |
3516 | ||
6612f05b | 3517 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3518 | { |
2bd59d48 | 3519 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3520 | } |
3521 | ||
6612f05b | 3522 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3523 | { |
2bd59d48 | 3524 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3525 | } |
3526 | ||
91796569 | 3527 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3528 | { |
7da11279 TH |
3529 | struct cftype *cft = seq_cft(m); |
3530 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3531 | |
2da8ca82 TH |
3532 | if (cft->seq_show) |
3533 | return cft->seq_show(m, arg); | |
e73d2c61 | 3534 | |
f4c753b7 | 3535 | if (cft->read_u64) |
896f5199 TH |
3536 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3537 | else if (cft->read_s64) | |
3538 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3539 | else | |
3540 | return -EINVAL; | |
3541 | return 0; | |
91796569 PM |
3542 | } |
3543 | ||
2bd59d48 TH |
3544 | static struct kernfs_ops cgroup_kf_single_ops = { |
3545 | .atomic_write_len = PAGE_SIZE, | |
3546 | .write = cgroup_file_write, | |
3547 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3548 | }; |
3549 | ||
2bd59d48 TH |
3550 | static struct kernfs_ops cgroup_kf_ops = { |
3551 | .atomic_write_len = PAGE_SIZE, | |
3552 | .write = cgroup_file_write, | |
3553 | .seq_start = cgroup_seqfile_start, | |
3554 | .seq_next = cgroup_seqfile_next, | |
3555 | .seq_stop = cgroup_seqfile_stop, | |
3556 | .seq_show = cgroup_seqfile_show, | |
3557 | }; | |
ddbcc7e8 PM |
3558 | |
3559 | /* | |
3560 | * cgroup_rename - Only allow simple rename of directories in place. | |
3561 | */ | |
2bd59d48 TH |
3562 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3563 | const char *new_name_str) | |
ddbcc7e8 | 3564 | { |
2bd59d48 | 3565 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3566 | int ret; |
65dff759 | 3567 | |
2bd59d48 | 3568 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3569 | return -ENOTDIR; |
2bd59d48 | 3570 | if (kn->parent != new_parent) |
ddbcc7e8 | 3571 | return -EIO; |
65dff759 | 3572 | |
6db8e85c TH |
3573 | /* |
3574 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3575 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3576 | */ |
aa6ec29b | 3577 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3578 | return -EPERM; |
099fca32 | 3579 | |
e1b2dc17 | 3580 | /* |
8353da1f | 3581 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3582 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3583 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3584 | */ |
3585 | kernfs_break_active_protection(new_parent); | |
3586 | kernfs_break_active_protection(kn); | |
099fca32 | 3587 | |
2bd59d48 | 3588 | mutex_lock(&cgroup_mutex); |
099fca32 | 3589 | |
2bd59d48 | 3590 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3591 | |
2bd59d48 | 3592 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3593 | |
3594 | kernfs_unbreak_active_protection(kn); | |
3595 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3596 | return ret; |
099fca32 LZ |
3597 | } |
3598 | ||
49957f8e TH |
3599 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3600 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3601 | { | |
3602 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3603 | .ia_uid = current_fsuid(), | |
3604 | .ia_gid = current_fsgid(), }; | |
3605 | ||
3606 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3607 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3608 | return 0; | |
3609 | ||
3610 | return kernfs_setattr(kn, &iattr); | |
3611 | } | |
3612 | ||
4df8dc90 TH |
3613 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3614 | struct cftype *cft) | |
ddbcc7e8 | 3615 | { |
8d7e6fb0 | 3616 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3617 | struct kernfs_node *kn; |
3618 | struct lock_class_key *key = NULL; | |
49957f8e | 3619 | int ret; |
05ef1d7c | 3620 | |
2bd59d48 TH |
3621 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3622 | key = &cft->lockdep_key; | |
3623 | #endif | |
3624 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3625 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3626 | NULL, key); |
49957f8e TH |
3627 | if (IS_ERR(kn)) |
3628 | return PTR_ERR(kn); | |
3629 | ||
3630 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3631 | if (ret) { |
49957f8e | 3632 | kernfs_remove(kn); |
f8f22e53 TH |
3633 | return ret; |
3634 | } | |
3635 | ||
6f60eade TH |
3636 | if (cft->file_offset) { |
3637 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3638 | ||
34c06254 | 3639 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3640 | cfile->kn = kn; |
34c06254 | 3641 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3642 | } |
3643 | ||
f8f22e53 | 3644 | return 0; |
ddbcc7e8 PM |
3645 | } |
3646 | ||
b1f28d31 TH |
3647 | /** |
3648 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3649 | * @css: the target css |
3650 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3651 | * @cfts: array of cftypes to be added |
3652 | * @is_add: whether to add or remove | |
3653 | * | |
3654 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3655 | * For removals, this function never fails. |
b1f28d31 | 3656 | */ |
4df8dc90 TH |
3657 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3658 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3659 | bool is_add) |
ddbcc7e8 | 3660 | { |
6732ed85 | 3661 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3662 | int ret = 0; |
b1f28d31 | 3663 | |
01f6474c | 3664 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3665 | |
6732ed85 TH |
3666 | restart: |
3667 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3668 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3669 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3670 | continue; |
05ebb6e6 | 3671 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3672 | continue; |
d51f39b0 | 3673 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3674 | continue; |
d51f39b0 | 3675 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3676 | continue; |
3677 | ||
2739d3cc | 3678 | if (is_add) { |
4df8dc90 | 3679 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3680 | if (ret) { |
ed3d261b JP |
3681 | pr_warn("%s: failed to add %s, err=%d\n", |
3682 | __func__, cft->name, ret); | |
6732ed85 TH |
3683 | cft_end = cft; |
3684 | is_add = false; | |
3685 | goto restart; | |
b1f28d31 | 3686 | } |
2739d3cc LZ |
3687 | } else { |
3688 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3689 | } |
ddbcc7e8 | 3690 | } |
b598dde3 | 3691 | return ret; |
ddbcc7e8 PM |
3692 | } |
3693 | ||
21a2d343 | 3694 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3695 | { |
3696 | LIST_HEAD(pending); | |
2bb566cb | 3697 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3698 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3699 | struct cgroup_subsys_state *css; |
9ccece80 | 3700 | int ret = 0; |
8e3f6541 | 3701 | |
01f6474c | 3702 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3703 | |
e8c82d20 | 3704 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3705 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3706 | struct cgroup *cgrp = css->cgroup; |
3707 | ||
88cb04b9 | 3708 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3709 | continue; |
3710 | ||
4df8dc90 | 3711 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3712 | if (ret) |
3713 | break; | |
8e3f6541 | 3714 | } |
21a2d343 TH |
3715 | |
3716 | if (is_add && !ret) | |
3717 | kernfs_activate(root->kn); | |
9ccece80 | 3718 | return ret; |
8e3f6541 TH |
3719 | } |
3720 | ||
2da440a2 | 3721 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3722 | { |
2bb566cb | 3723 | struct cftype *cft; |
8e3f6541 | 3724 | |
2bd59d48 TH |
3725 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3726 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3727 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3728 | kfree(cft->kf_ops); | |
3729 | cft->kf_ops = NULL; | |
2da440a2 | 3730 | cft->ss = NULL; |
a8ddc821 TH |
3731 | |
3732 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3733 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3734 | } |
2da440a2 TH |
3735 | } |
3736 | ||
2bd59d48 | 3737 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3738 | { |
3739 | struct cftype *cft; | |
3740 | ||
2bd59d48 TH |
3741 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3742 | struct kernfs_ops *kf_ops; | |
3743 | ||
0adb0704 TH |
3744 | WARN_ON(cft->ss || cft->kf_ops); |
3745 | ||
2bd59d48 TH |
3746 | if (cft->seq_start) |
3747 | kf_ops = &cgroup_kf_ops; | |
3748 | else | |
3749 | kf_ops = &cgroup_kf_single_ops; | |
3750 | ||
3751 | /* | |
3752 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3753 | * make a copy of kf_ops to set its atomic_write_len. | |
3754 | */ | |
3755 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3756 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3757 | if (!kf_ops) { | |
3758 | cgroup_exit_cftypes(cfts); | |
3759 | return -ENOMEM; | |
3760 | } | |
3761 | kf_ops->atomic_write_len = cft->max_write_len; | |
3762 | } | |
8e3f6541 | 3763 | |
2bd59d48 | 3764 | cft->kf_ops = kf_ops; |
2bb566cb | 3765 | cft->ss = ss; |
2bd59d48 | 3766 | } |
2bb566cb | 3767 | |
2bd59d48 | 3768 | return 0; |
2da440a2 TH |
3769 | } |
3770 | ||
21a2d343 TH |
3771 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3772 | { | |
01f6474c | 3773 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3774 | |
3775 | if (!cfts || !cfts[0].ss) | |
3776 | return -ENOENT; | |
3777 | ||
3778 | list_del(&cfts->node); | |
3779 | cgroup_apply_cftypes(cfts, false); | |
3780 | cgroup_exit_cftypes(cfts); | |
3781 | return 0; | |
8e3f6541 | 3782 | } |
8e3f6541 | 3783 | |
79578621 TH |
3784 | /** |
3785 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3786 | * @cfts: zero-length name terminated array of cftypes |
3787 | * | |
2bb566cb TH |
3788 | * Unregister @cfts. Files described by @cfts are removed from all |
3789 | * existing cgroups and all future cgroups won't have them either. This | |
3790 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3791 | * |
3792 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3793 | * registered. |
79578621 | 3794 | */ |
2bb566cb | 3795 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3796 | { |
21a2d343 | 3797 | int ret; |
79578621 | 3798 | |
01f6474c | 3799 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3800 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3801 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3802 | return ret; |
80b13586 TH |
3803 | } |
3804 | ||
8e3f6541 TH |
3805 | /** |
3806 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3807 | * @ss: target cgroup subsystem | |
3808 | * @cfts: zero-length name terminated array of cftypes | |
3809 | * | |
3810 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3811 | * existing cgroups to which @ss is attached and all future cgroups will | |
3812 | * have them too. This function can be called anytime whether @ss is | |
3813 | * attached or not. | |
3814 | * | |
3815 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3816 | * function currently returns 0 as long as @cfts registration is successful | |
3817 | * even if some file creation attempts on existing cgroups fail. | |
3818 | */ | |
2cf669a5 | 3819 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3820 | { |
9ccece80 | 3821 | int ret; |
8e3f6541 | 3822 | |
fc5ed1e9 | 3823 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3824 | return 0; |
3825 | ||
dc5736ed LZ |
3826 | if (!cfts || cfts[0].name[0] == '\0') |
3827 | return 0; | |
2bb566cb | 3828 | |
2bd59d48 TH |
3829 | ret = cgroup_init_cftypes(ss, cfts); |
3830 | if (ret) | |
3831 | return ret; | |
79578621 | 3832 | |
01f6474c | 3833 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3834 | |
0adb0704 | 3835 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3836 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3837 | if (ret) |
21a2d343 | 3838 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3839 | |
01f6474c | 3840 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3841 | return ret; |
79578621 TH |
3842 | } |
3843 | ||
a8ddc821 TH |
3844 | /** |
3845 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3846 | * @ss: target cgroup subsystem | |
3847 | * @cfts: zero-length name terminated array of cftypes | |
3848 | * | |
3849 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3850 | * the default hierarchy. | |
3851 | */ | |
3852 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3853 | { | |
3854 | struct cftype *cft; | |
3855 | ||
3856 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3857 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3858 | return cgroup_add_cftypes(ss, cfts); |
3859 | } | |
3860 | ||
3861 | /** | |
3862 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3863 | * @ss: target cgroup subsystem | |
3864 | * @cfts: zero-length name terminated array of cftypes | |
3865 | * | |
3866 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3867 | * the legacy hierarchies. | |
3868 | */ | |
2cf669a5 TH |
3869 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3870 | { | |
a8ddc821 TH |
3871 | struct cftype *cft; |
3872 | ||
e4b7037c TH |
3873 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3874 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3875 | return cgroup_add_cftypes(ss, cfts); |
3876 | } | |
3877 | ||
34c06254 TH |
3878 | /** |
3879 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3880 | * @cfile: target cgroup_file | |
3881 | * | |
3882 | * @cfile must have been obtained by setting cftype->file_offset. | |
3883 | */ | |
3884 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3885 | { | |
3886 | unsigned long flags; | |
3887 | ||
3888 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3889 | if (cfile->kn) | |
3890 | kernfs_notify(cfile->kn); | |
3891 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3892 | } | |
3893 | ||
a043e3b2 LZ |
3894 | /** |
3895 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3896 | * @cgrp: the cgroup in question | |
3897 | * | |
3898 | * Return the number of tasks in the cgroup. | |
3899 | */ | |
07bc356e | 3900 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3901 | { |
3902 | int count = 0; | |
69d0206c | 3903 | struct cgrp_cset_link *link; |
817929ec | 3904 | |
f0d9a5f1 | 3905 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3906 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3907 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3908 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3909 | return count; |
3910 | } | |
3911 | ||
53fa5261 | 3912 | /** |
492eb21b | 3913 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3914 | * @pos: the current position (%NULL to initiate traversal) |
3915 | * @parent: css whose children to walk | |
53fa5261 | 3916 | * |
c2931b70 | 3917 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3918 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3919 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3920 | * be returned regardless of their states. | |
3921 | * | |
3922 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3923 | * css which finished ->css_online() is guaranteed to be visible in the | |
3924 | * future iterations and will stay visible until the last reference is put. | |
3925 | * A css which hasn't finished ->css_online() or already finished | |
3926 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3927 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3928 | */ |
c2931b70 TH |
3929 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3930 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3931 | { |
c2931b70 | 3932 | struct cgroup_subsys_state *next; |
53fa5261 | 3933 | |
8353da1f | 3934 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3935 | |
3936 | /* | |
de3f0341 TH |
3937 | * @pos could already have been unlinked from the sibling list. |
3938 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3939 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3940 | * @pos is taken off list, at which time its next pointer is valid, | |
3941 | * and, as releases are serialized, the one pointed to by the next | |
3942 | * pointer is guaranteed to not have started release yet. This | |
3943 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3944 | * critical section, the one pointed to by its next pointer is | |
3945 | * guaranteed to not have finished its RCU grace period even if we | |
3946 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3947 | * |
de3f0341 TH |
3948 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3949 | * dereferenced; however, as each css is given a monotonically | |
3950 | * increasing unique serial number and always appended to the | |
3951 | * sibling list, the next one can be found by walking the parent's | |
3952 | * children until the first css with higher serial number than | |
3953 | * @pos's. While this path can be slower, it happens iff iteration | |
3954 | * races against release and the race window is very small. | |
53fa5261 | 3955 | */ |
3b287a50 | 3956 | if (!pos) { |
c2931b70 TH |
3957 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3958 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3959 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3960 | } else { |
c2931b70 | 3961 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3962 | if (next->serial_nr > pos->serial_nr) |
3963 | break; | |
53fa5261 TH |
3964 | } |
3965 | ||
3b281afb TH |
3966 | /* |
3967 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3968 | * the next sibling. |
3b281afb | 3969 | */ |
c2931b70 TH |
3970 | if (&next->sibling != &parent->children) |
3971 | return next; | |
3b281afb | 3972 | return NULL; |
53fa5261 | 3973 | } |
53fa5261 | 3974 | |
574bd9f7 | 3975 | /** |
492eb21b | 3976 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3977 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3978 | * @root: css whose descendants to walk |
574bd9f7 | 3979 | * |
492eb21b | 3980 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3981 | * to visit for pre-order traversal of @root's descendants. @root is |
3982 | * included in the iteration and the first node to be visited. | |
75501a6d | 3983 | * |
87fb54f1 TH |
3984 | * While this function requires cgroup_mutex or RCU read locking, it |
3985 | * doesn't require the whole traversal to be contained in a single critical | |
3986 | * section. This function will return the correct next descendant as long | |
3987 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3988 | * |
3989 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3990 | * css which finished ->css_online() is guaranteed to be visible in the | |
3991 | * future iterations and will stay visible until the last reference is put. | |
3992 | * A css which hasn't finished ->css_online() or already finished | |
3993 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3994 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3995 | */ |
492eb21b TH |
3996 | struct cgroup_subsys_state * |
3997 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3998 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3999 | { |
492eb21b | 4000 | struct cgroup_subsys_state *next; |
574bd9f7 | 4001 | |
8353da1f | 4002 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4003 | |
bd8815a6 | 4004 | /* if first iteration, visit @root */ |
7805d000 | 4005 | if (!pos) |
bd8815a6 | 4006 | return root; |
574bd9f7 TH |
4007 | |
4008 | /* visit the first child if exists */ | |
492eb21b | 4009 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4010 | if (next) |
4011 | return next; | |
4012 | ||
4013 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4014 | while (pos != root) { |
5c9d535b | 4015 | next = css_next_child(pos, pos->parent); |
75501a6d | 4016 | if (next) |
574bd9f7 | 4017 | return next; |
5c9d535b | 4018 | pos = pos->parent; |
7805d000 | 4019 | } |
574bd9f7 TH |
4020 | |
4021 | return NULL; | |
4022 | } | |
574bd9f7 | 4023 | |
12a9d2fe | 4024 | /** |
492eb21b TH |
4025 | * css_rightmost_descendant - return the rightmost descendant of a css |
4026 | * @pos: css of interest | |
12a9d2fe | 4027 | * |
492eb21b TH |
4028 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4029 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4030 | * subtree of @pos. |
75501a6d | 4031 | * |
87fb54f1 TH |
4032 | * While this function requires cgroup_mutex or RCU read locking, it |
4033 | * doesn't require the whole traversal to be contained in a single critical | |
4034 | * section. This function will return the correct rightmost descendant as | |
4035 | * long as @pos is accessible. | |
12a9d2fe | 4036 | */ |
492eb21b TH |
4037 | struct cgroup_subsys_state * |
4038 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4039 | { |
492eb21b | 4040 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4041 | |
8353da1f | 4042 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4043 | |
4044 | do { | |
4045 | last = pos; | |
4046 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4047 | pos = NULL; | |
492eb21b | 4048 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4049 | pos = tmp; |
4050 | } while (pos); | |
4051 | ||
4052 | return last; | |
4053 | } | |
12a9d2fe | 4054 | |
492eb21b TH |
4055 | static struct cgroup_subsys_state * |
4056 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4057 | { |
492eb21b | 4058 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4059 | |
4060 | do { | |
4061 | last = pos; | |
492eb21b | 4062 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4063 | } while (pos); |
4064 | ||
4065 | return last; | |
4066 | } | |
4067 | ||
4068 | /** | |
492eb21b | 4069 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4070 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4071 | * @root: css whose descendants to walk |
574bd9f7 | 4072 | * |
492eb21b | 4073 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4074 | * to visit for post-order traversal of @root's descendants. @root is |
4075 | * included in the iteration and the last node to be visited. | |
75501a6d | 4076 | * |
87fb54f1 TH |
4077 | * While this function requires cgroup_mutex or RCU read locking, it |
4078 | * doesn't require the whole traversal to be contained in a single critical | |
4079 | * section. This function will return the correct next descendant as long | |
4080 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4081 | * @cgroup. | |
c2931b70 TH |
4082 | * |
4083 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4084 | * css which finished ->css_online() is guaranteed to be visible in the | |
4085 | * future iterations and will stay visible until the last reference is put. | |
4086 | * A css which hasn't finished ->css_online() or already finished | |
4087 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4088 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4089 | */ |
492eb21b TH |
4090 | struct cgroup_subsys_state * |
4091 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4092 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4093 | { |
492eb21b | 4094 | struct cgroup_subsys_state *next; |
574bd9f7 | 4095 | |
8353da1f | 4096 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4097 | |
58b79a91 TH |
4098 | /* if first iteration, visit leftmost descendant which may be @root */ |
4099 | if (!pos) | |
4100 | return css_leftmost_descendant(root); | |
574bd9f7 | 4101 | |
bd8815a6 TH |
4102 | /* if we visited @root, we're done */ |
4103 | if (pos == root) | |
4104 | return NULL; | |
4105 | ||
574bd9f7 | 4106 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4107 | next = css_next_child(pos, pos->parent); |
75501a6d | 4108 | if (next) |
492eb21b | 4109 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4110 | |
4111 | /* no sibling left, visit parent */ | |
5c9d535b | 4112 | return pos->parent; |
574bd9f7 | 4113 | } |
574bd9f7 | 4114 | |
f3d46500 TH |
4115 | /** |
4116 | * css_has_online_children - does a css have online children | |
4117 | * @css: the target css | |
4118 | * | |
4119 | * Returns %true if @css has any online children; otherwise, %false. This | |
4120 | * function can be called from any context but the caller is responsible | |
4121 | * for synchronizing against on/offlining as necessary. | |
4122 | */ | |
4123 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4124 | { |
f3d46500 TH |
4125 | struct cgroup_subsys_state *child; |
4126 | bool ret = false; | |
cbc125ef TH |
4127 | |
4128 | rcu_read_lock(); | |
f3d46500 | 4129 | css_for_each_child(child, css) { |
99bae5f9 | 4130 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4131 | ret = true; |
4132 | break; | |
cbc125ef TH |
4133 | } |
4134 | } | |
4135 | rcu_read_unlock(); | |
f3d46500 | 4136 | return ret; |
574bd9f7 | 4137 | } |
574bd9f7 | 4138 | |
0942eeee | 4139 | /** |
ecb9d535 | 4140 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4141 | * @it: the iterator to advance |
4142 | * | |
4143 | * Advance @it to the next css_set to walk. | |
d515876e | 4144 | */ |
ecb9d535 | 4145 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4146 | { |
0f0a2b4f | 4147 | struct list_head *l = it->cset_pos; |
d515876e TH |
4148 | struct cgrp_cset_link *link; |
4149 | struct css_set *cset; | |
4150 | ||
f0d9a5f1 | 4151 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4152 | |
d515876e TH |
4153 | /* Advance to the next non-empty css_set */ |
4154 | do { | |
4155 | l = l->next; | |
0f0a2b4f TH |
4156 | if (l == it->cset_head) { |
4157 | it->cset_pos = NULL; | |
ecb9d535 | 4158 | it->task_pos = NULL; |
d515876e TH |
4159 | return; |
4160 | } | |
3ebb2b6e TH |
4161 | |
4162 | if (it->ss) { | |
4163 | cset = container_of(l, struct css_set, | |
4164 | e_cset_node[it->ss->id]); | |
4165 | } else { | |
4166 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4167 | cset = link->cset; | |
4168 | } | |
0de0942d | 4169 | } while (!css_set_populated(cset)); |
c7561128 | 4170 | |
0f0a2b4f | 4171 | it->cset_pos = l; |
c7561128 TH |
4172 | |
4173 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 4174 | it->task_pos = cset->tasks.next; |
c7561128 | 4175 | else |
0f0a2b4f TH |
4176 | it->task_pos = cset->mg_tasks.next; |
4177 | ||
4178 | it->tasks_head = &cset->tasks; | |
4179 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4180 | |
4181 | /* | |
4182 | * We don't keep css_sets locked across iteration steps and thus | |
4183 | * need to take steps to ensure that iteration can be resumed after | |
4184 | * the lock is re-acquired. Iteration is performed at two levels - | |
4185 | * css_sets and tasks in them. | |
4186 | * | |
4187 | * Once created, a css_set never leaves its cgroup lists, so a | |
4188 | * pinned css_set is guaranteed to stay put and we can resume | |
4189 | * iteration afterwards. | |
4190 | * | |
4191 | * Tasks may leave @cset across iteration steps. This is resolved | |
4192 | * by registering each iterator with the css_set currently being | |
4193 | * walked and making css_set_move_task() advance iterators whose | |
4194 | * next task is leaving. | |
4195 | */ | |
4196 | if (it->cur_cset) { | |
4197 | list_del(&it->iters_node); | |
4198 | put_css_set_locked(it->cur_cset); | |
4199 | } | |
4200 | get_css_set(cset); | |
4201 | it->cur_cset = cset; | |
4202 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4203 | } |
4204 | ||
ecb9d535 TH |
4205 | static void css_task_iter_advance(struct css_task_iter *it) |
4206 | { | |
4207 | struct list_head *l = it->task_pos; | |
4208 | ||
f0d9a5f1 | 4209 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4210 | WARN_ON_ONCE(!l); |
4211 | ||
4212 | /* | |
4213 | * Advance iterator to find next entry. cset->tasks is consumed | |
4214 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4215 | * next cset. | |
4216 | */ | |
4217 | l = l->next; | |
4218 | ||
4219 | if (l == it->tasks_head) | |
4220 | l = it->mg_tasks_head->next; | |
4221 | ||
4222 | if (l == it->mg_tasks_head) | |
4223 | css_task_iter_advance_css_set(it); | |
4224 | else | |
4225 | it->task_pos = l; | |
4226 | } | |
4227 | ||
0942eeee | 4228 | /** |
72ec7029 TH |
4229 | * css_task_iter_start - initiate task iteration |
4230 | * @css: the css to walk tasks of | |
0942eeee TH |
4231 | * @it: the task iterator to use |
4232 | * | |
72ec7029 TH |
4233 | * Initiate iteration through the tasks of @css. The caller can call |
4234 | * css_task_iter_next() to walk through the tasks until the function | |
4235 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4236 | * called. | |
0942eeee | 4237 | */ |
72ec7029 TH |
4238 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4239 | struct css_task_iter *it) | |
817929ec | 4240 | { |
56fde9e0 TH |
4241 | /* no one should try to iterate before mounting cgroups */ |
4242 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4243 | |
ed27b9f7 TH |
4244 | memset(it, 0, sizeof(*it)); |
4245 | ||
f0d9a5f1 | 4246 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 4247 | |
3ebb2b6e TH |
4248 | it->ss = css->ss; |
4249 | ||
4250 | if (it->ss) | |
4251 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4252 | else | |
4253 | it->cset_pos = &css->cgroup->cset_links; | |
4254 | ||
0f0a2b4f | 4255 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4256 | |
ecb9d535 | 4257 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4258 | |
f0d9a5f1 | 4259 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
4260 | } |
4261 | ||
0942eeee | 4262 | /** |
72ec7029 | 4263 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4264 | * @it: the task iterator being iterated |
4265 | * | |
4266 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4267 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4268 | * reaches the end. | |
0942eeee | 4269 | */ |
72ec7029 | 4270 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4271 | { |
d5745675 | 4272 | if (it->cur_task) { |
ed27b9f7 | 4273 | put_task_struct(it->cur_task); |
d5745675 TH |
4274 | it->cur_task = NULL; |
4275 | } | |
ed27b9f7 | 4276 | |
f0d9a5f1 | 4277 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 4278 | |
d5745675 TH |
4279 | if (it->task_pos) { |
4280 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4281 | cg_list); | |
4282 | get_task_struct(it->cur_task); | |
4283 | css_task_iter_advance(it); | |
4284 | } | |
ed27b9f7 | 4285 | |
f0d9a5f1 | 4286 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4287 | |
4288 | return it->cur_task; | |
817929ec PM |
4289 | } |
4290 | ||
0942eeee | 4291 | /** |
72ec7029 | 4292 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4293 | * @it: the task iterator to finish |
4294 | * | |
72ec7029 | 4295 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4296 | */ |
72ec7029 | 4297 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4298 | { |
ed27b9f7 | 4299 | if (it->cur_cset) { |
f0d9a5f1 | 4300 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
4301 | list_del(&it->iters_node); |
4302 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 4303 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4304 | } |
4305 | ||
4306 | if (it->cur_task) | |
4307 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4308 | } |
4309 | ||
4310 | /** | |
8cc99345 TH |
4311 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4312 | * @to: cgroup to which the tasks will be moved | |
4313 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4314 | * |
eaf797ab TH |
4315 | * Locking rules between cgroup_post_fork() and the migration path |
4316 | * guarantee that, if a task is forking while being migrated, the new child | |
4317 | * is guaranteed to be either visible in the source cgroup after the | |
4318 | * parent's migration is complete or put into the target cgroup. No task | |
4319 | * can slip out of migration through forking. | |
31a7df01 | 4320 | */ |
8cc99345 | 4321 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4322 | { |
952aaa12 TH |
4323 | LIST_HEAD(preloaded_csets); |
4324 | struct cgrp_cset_link *link; | |
72ec7029 | 4325 | struct css_task_iter it; |
e406d1cf | 4326 | struct task_struct *task; |
952aaa12 | 4327 | int ret; |
31a7df01 | 4328 | |
6c694c88 TH |
4329 | if (!cgroup_may_migrate_to(to)) |
4330 | return -EBUSY; | |
4331 | ||
952aaa12 | 4332 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4333 | |
952aaa12 | 4334 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4335 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4336 | list_for_each_entry(link, &from->cset_links, cset_link) |
4337 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4338 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4339 | |
e4857982 | 4340 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
952aaa12 TH |
4341 | if (ret) |
4342 | goto out_err; | |
8cc99345 | 4343 | |
952aaa12 | 4344 | /* |
2cfa2b19 | 4345 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4346 | * ->can_attach() fails. |
4347 | */ | |
e406d1cf | 4348 | do { |
9d800df1 | 4349 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4350 | task = css_task_iter_next(&it); |
4351 | if (task) | |
4352 | get_task_struct(task); | |
4353 | css_task_iter_end(&it); | |
4354 | ||
4355 | if (task) { | |
37ff9f8f | 4356 | ret = cgroup_migrate(task, false, to->root); |
e406d1cf TH |
4357 | put_task_struct(task); |
4358 | } | |
4359 | } while (task && !ret); | |
952aaa12 TH |
4360 | out_err: |
4361 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4362 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4363 | return ret; |
8cc99345 TH |
4364 | } |
4365 | ||
bbcb81d0 | 4366 | /* |
102a775e | 4367 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4368 | * |
4369 | * Reading this file can return large amounts of data if a cgroup has | |
4370 | * *lots* of attached tasks. So it may need several calls to read(), | |
4371 | * but we cannot guarantee that the information we produce is correct | |
4372 | * unless we produce it entirely atomically. | |
4373 | * | |
bbcb81d0 | 4374 | */ |
bbcb81d0 | 4375 | |
24528255 LZ |
4376 | /* which pidlist file are we talking about? */ |
4377 | enum cgroup_filetype { | |
4378 | CGROUP_FILE_PROCS, | |
4379 | CGROUP_FILE_TASKS, | |
4380 | }; | |
4381 | ||
4382 | /* | |
4383 | * A pidlist is a list of pids that virtually represents the contents of one | |
4384 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4385 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4386 | * to the cgroup. | |
4387 | */ | |
4388 | struct cgroup_pidlist { | |
4389 | /* | |
4390 | * used to find which pidlist is wanted. doesn't change as long as | |
4391 | * this particular list stays in the list. | |
4392 | */ | |
4393 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4394 | /* array of xids */ | |
4395 | pid_t *list; | |
4396 | /* how many elements the above list has */ | |
4397 | int length; | |
24528255 LZ |
4398 | /* each of these stored in a list by its cgroup */ |
4399 | struct list_head links; | |
4400 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4401 | struct cgroup *owner; | |
b1a21367 TH |
4402 | /* for delayed destruction */ |
4403 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4404 | }; |
4405 | ||
d1d9fd33 BB |
4406 | /* |
4407 | * The following two functions "fix" the issue where there are more pids | |
4408 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4409 | * TODO: replace with a kernel-wide solution to this problem | |
4410 | */ | |
4411 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4412 | static void *pidlist_allocate(int count) | |
4413 | { | |
4414 | if (PIDLIST_TOO_LARGE(count)) | |
4415 | return vmalloc(count * sizeof(pid_t)); | |
4416 | else | |
4417 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4418 | } | |
b1a21367 | 4419 | |
d1d9fd33 BB |
4420 | static void pidlist_free(void *p) |
4421 | { | |
58794514 | 4422 | kvfree(p); |
d1d9fd33 | 4423 | } |
d1d9fd33 | 4424 | |
b1a21367 TH |
4425 | /* |
4426 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4427 | * should be left afterwards. | |
4428 | */ | |
4429 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4430 | { | |
4431 | struct cgroup_pidlist *l, *tmp_l; | |
4432 | ||
4433 | mutex_lock(&cgrp->pidlist_mutex); | |
4434 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4435 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4436 | mutex_unlock(&cgrp->pidlist_mutex); | |
4437 | ||
4438 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4439 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4440 | } | |
4441 | ||
4442 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4443 | { | |
4444 | struct delayed_work *dwork = to_delayed_work(work); | |
4445 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4446 | destroy_dwork); | |
4447 | struct cgroup_pidlist *tofree = NULL; | |
4448 | ||
4449 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4450 | |
4451 | /* | |
04502365 TH |
4452 | * Destroy iff we didn't get queued again. The state won't change |
4453 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4454 | */ |
04502365 | 4455 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4456 | list_del(&l->links); |
4457 | pidlist_free(l->list); | |
4458 | put_pid_ns(l->key.ns); | |
4459 | tofree = l; | |
4460 | } | |
4461 | ||
b1a21367 TH |
4462 | mutex_unlock(&l->owner->pidlist_mutex); |
4463 | kfree(tofree); | |
4464 | } | |
4465 | ||
bbcb81d0 | 4466 | /* |
102a775e | 4467 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4468 | * Returns the number of unique elements. |
bbcb81d0 | 4469 | */ |
6ee211ad | 4470 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4471 | { |
102a775e | 4472 | int src, dest = 1; |
102a775e BB |
4473 | |
4474 | /* | |
4475 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4476 | * edge cases first; no work needs to be done for either | |
4477 | */ | |
4478 | if (length == 0 || length == 1) | |
4479 | return length; | |
4480 | /* src and dest walk down the list; dest counts unique elements */ | |
4481 | for (src = 1; src < length; src++) { | |
4482 | /* find next unique element */ | |
4483 | while (list[src] == list[src-1]) { | |
4484 | src++; | |
4485 | if (src == length) | |
4486 | goto after; | |
4487 | } | |
4488 | /* dest always points to where the next unique element goes */ | |
4489 | list[dest] = list[src]; | |
4490 | dest++; | |
4491 | } | |
4492 | after: | |
102a775e BB |
4493 | return dest; |
4494 | } | |
4495 | ||
afb2bc14 TH |
4496 | /* |
4497 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4498 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4499 | * different per namespace, each namespace needs differently sorted list, | |
4500 | * making it impossible to use, for example, single rbtree of member tasks | |
4501 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4502 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4503 | * pidlists keyed by cgroup and namespace. | |
4504 | * | |
4505 | * All this extra complexity was caused by the original implementation | |
4506 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4507 | * want to do away with it. Explicitly scramble sort order if on the |
4508 | * default hierarchy so that no such expectation exists in the new | |
4509 | * interface. | |
afb2bc14 TH |
4510 | * |
4511 | * Scrambling is done by swapping every two consecutive bits, which is | |
4512 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4513 | */ | |
4514 | static pid_t pid_fry(pid_t pid) | |
4515 | { | |
4516 | unsigned a = pid & 0x55555555; | |
4517 | unsigned b = pid & 0xAAAAAAAA; | |
4518 | ||
4519 | return (a << 1) | (b >> 1); | |
4520 | } | |
4521 | ||
4522 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4523 | { | |
aa6ec29b | 4524 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4525 | return pid_fry(pid); |
4526 | else | |
4527 | return pid; | |
4528 | } | |
4529 | ||
102a775e BB |
4530 | static int cmppid(const void *a, const void *b) |
4531 | { | |
4532 | return *(pid_t *)a - *(pid_t *)b; | |
4533 | } | |
4534 | ||
afb2bc14 TH |
4535 | static int fried_cmppid(const void *a, const void *b) |
4536 | { | |
4537 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4538 | } | |
4539 | ||
e6b81710 TH |
4540 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4541 | enum cgroup_filetype type) | |
4542 | { | |
4543 | struct cgroup_pidlist *l; | |
4544 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4545 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4546 | ||
4547 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4548 | ||
4549 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4550 | if (l->key.type == type && l->key.ns == ns) | |
4551 | return l; | |
4552 | return NULL; | |
4553 | } | |
4554 | ||
72a8cb30 BB |
4555 | /* |
4556 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4557 | * returns with the lock on that pidlist already held, and takes care | |
4558 | * of the use count, or returns NULL with no locks held if we're out of | |
4559 | * memory. | |
4560 | */ | |
e6b81710 TH |
4561 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4562 | enum cgroup_filetype type) | |
72a8cb30 BB |
4563 | { |
4564 | struct cgroup_pidlist *l; | |
b70cc5fd | 4565 | |
e6b81710 TH |
4566 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4567 | ||
4568 | l = cgroup_pidlist_find(cgrp, type); | |
4569 | if (l) | |
4570 | return l; | |
4571 | ||
72a8cb30 | 4572 | /* entry not found; create a new one */ |
f4f4be2b | 4573 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4574 | if (!l) |
72a8cb30 | 4575 | return l; |
e6b81710 | 4576 | |
b1a21367 | 4577 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4578 | l->key.type = type; |
e6b81710 TH |
4579 | /* don't need task_nsproxy() if we're looking at ourself */ |
4580 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4581 | l->owner = cgrp; |
4582 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4583 | return l; |
4584 | } | |
4585 | ||
102a775e BB |
4586 | /* |
4587 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4588 | */ | |
72a8cb30 BB |
4589 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4590 | struct cgroup_pidlist **lp) | |
102a775e BB |
4591 | { |
4592 | pid_t *array; | |
4593 | int length; | |
4594 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4595 | struct css_task_iter it; |
817929ec | 4596 | struct task_struct *tsk; |
102a775e BB |
4597 | struct cgroup_pidlist *l; |
4598 | ||
4bac00d1 TH |
4599 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4600 | ||
102a775e BB |
4601 | /* |
4602 | * If cgroup gets more users after we read count, we won't have | |
4603 | * enough space - tough. This race is indistinguishable to the | |
4604 | * caller from the case that the additional cgroup users didn't | |
4605 | * show up until sometime later on. | |
4606 | */ | |
4607 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4608 | array = pidlist_allocate(length); |
102a775e BB |
4609 | if (!array) |
4610 | return -ENOMEM; | |
4611 | /* now, populate the array */ | |
9d800df1 | 4612 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4613 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4614 | if (unlikely(n == length)) |
817929ec | 4615 | break; |
102a775e | 4616 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4617 | if (type == CGROUP_FILE_PROCS) |
4618 | pid = task_tgid_vnr(tsk); | |
4619 | else | |
4620 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4621 | if (pid > 0) /* make sure to only use valid results */ |
4622 | array[n++] = pid; | |
817929ec | 4623 | } |
72ec7029 | 4624 | css_task_iter_end(&it); |
102a775e BB |
4625 | length = n; |
4626 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4627 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4628 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4629 | else | |
4630 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4631 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4632 | length = pidlist_uniq(array, length); |
e6b81710 | 4633 | |
e6b81710 | 4634 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4635 | if (!l) { |
d1d9fd33 | 4636 | pidlist_free(array); |
72a8cb30 | 4637 | return -ENOMEM; |
102a775e | 4638 | } |
e6b81710 TH |
4639 | |
4640 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4641 | pidlist_free(l->list); |
102a775e BB |
4642 | l->list = array; |
4643 | l->length = length; | |
72a8cb30 | 4644 | *lp = l; |
102a775e | 4645 | return 0; |
bbcb81d0 PM |
4646 | } |
4647 | ||
846c7bb0 | 4648 | /** |
a043e3b2 | 4649 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4650 | * @stats: cgroupstats to fill information into |
4651 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4652 | * been requested. | |
a043e3b2 LZ |
4653 | * |
4654 | * Build and fill cgroupstats so that taskstats can export it to user | |
4655 | * space. | |
846c7bb0 BS |
4656 | */ |
4657 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4658 | { | |
2bd59d48 | 4659 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4660 | struct cgroup *cgrp; |
72ec7029 | 4661 | struct css_task_iter it; |
846c7bb0 | 4662 | struct task_struct *tsk; |
33d283be | 4663 | |
2bd59d48 TH |
4664 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4665 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4666 | kernfs_type(kn) != KERNFS_DIR) | |
4667 | return -EINVAL; | |
4668 | ||
bad34660 LZ |
4669 | mutex_lock(&cgroup_mutex); |
4670 | ||
846c7bb0 | 4671 | /* |
2bd59d48 | 4672 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4673 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4674 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4675 | */ |
2bd59d48 TH |
4676 | rcu_read_lock(); |
4677 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4678 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4679 | rcu_read_unlock(); |
bad34660 | 4680 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4681 | return -ENOENT; |
4682 | } | |
bad34660 | 4683 | rcu_read_unlock(); |
846c7bb0 | 4684 | |
9d800df1 | 4685 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4686 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4687 | switch (tsk->state) { |
4688 | case TASK_RUNNING: | |
4689 | stats->nr_running++; | |
4690 | break; | |
4691 | case TASK_INTERRUPTIBLE: | |
4692 | stats->nr_sleeping++; | |
4693 | break; | |
4694 | case TASK_UNINTERRUPTIBLE: | |
4695 | stats->nr_uninterruptible++; | |
4696 | break; | |
4697 | case TASK_STOPPED: | |
4698 | stats->nr_stopped++; | |
4699 | break; | |
4700 | default: | |
4701 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4702 | stats->nr_io_wait++; | |
4703 | break; | |
4704 | } | |
4705 | } | |
72ec7029 | 4706 | css_task_iter_end(&it); |
846c7bb0 | 4707 | |
bad34660 | 4708 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4709 | return 0; |
846c7bb0 BS |
4710 | } |
4711 | ||
8f3ff208 | 4712 | |
bbcb81d0 | 4713 | /* |
102a775e | 4714 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4715 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4716 | * in the cgroup->l->list array. |
bbcb81d0 | 4717 | */ |
cc31edce | 4718 | |
102a775e | 4719 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4720 | { |
cc31edce PM |
4721 | /* |
4722 | * Initially we receive a position value that corresponds to | |
4723 | * one more than the last pid shown (or 0 on the first call or | |
4724 | * after a seek to the start). Use a binary-search to find the | |
4725 | * next pid to display, if any | |
4726 | */ | |
2bd59d48 | 4727 | struct kernfs_open_file *of = s->private; |
7da11279 | 4728 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4729 | struct cgroup_pidlist *l; |
7da11279 | 4730 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4731 | int index = 0, pid = *pos; |
4bac00d1 TH |
4732 | int *iter, ret; |
4733 | ||
4734 | mutex_lock(&cgrp->pidlist_mutex); | |
4735 | ||
4736 | /* | |
5d22444f | 4737 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4738 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4739 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4740 | * could already have been destroyed. |
4741 | */ | |
5d22444f TH |
4742 | if (of->priv) |
4743 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4744 | |
4745 | /* | |
4746 | * Either this is the first start() after open or the matching | |
4747 | * pidlist has been destroyed inbetween. Create a new one. | |
4748 | */ | |
5d22444f TH |
4749 | if (!of->priv) { |
4750 | ret = pidlist_array_load(cgrp, type, | |
4751 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4752 | if (ret) |
4753 | return ERR_PTR(ret); | |
4754 | } | |
5d22444f | 4755 | l = of->priv; |
cc31edce | 4756 | |
cc31edce | 4757 | if (pid) { |
102a775e | 4758 | int end = l->length; |
20777766 | 4759 | |
cc31edce PM |
4760 | while (index < end) { |
4761 | int mid = (index + end) / 2; | |
afb2bc14 | 4762 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4763 | index = mid; |
4764 | break; | |
afb2bc14 | 4765 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4766 | index = mid + 1; |
4767 | else | |
4768 | end = mid; | |
4769 | } | |
4770 | } | |
4771 | /* If we're off the end of the array, we're done */ | |
102a775e | 4772 | if (index >= l->length) |
cc31edce PM |
4773 | return NULL; |
4774 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4775 | iter = l->list + index; |
afb2bc14 | 4776 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4777 | return iter; |
4778 | } | |
4779 | ||
102a775e | 4780 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4781 | { |
2bd59d48 | 4782 | struct kernfs_open_file *of = s->private; |
5d22444f | 4783 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4784 | |
5d22444f TH |
4785 | if (l) |
4786 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4787 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4788 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4789 | } |
4790 | ||
102a775e | 4791 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4792 | { |
2bd59d48 | 4793 | struct kernfs_open_file *of = s->private; |
5d22444f | 4794 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4795 | pid_t *p = v; |
4796 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4797 | /* |
4798 | * Advance to the next pid in the array. If this goes off the | |
4799 | * end, we're done | |
4800 | */ | |
4801 | p++; | |
4802 | if (p >= end) { | |
4803 | return NULL; | |
4804 | } else { | |
7da11279 | 4805 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4806 | return p; |
4807 | } | |
4808 | } | |
4809 | ||
102a775e | 4810 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4811 | { |
94ff212d JP |
4812 | seq_printf(s, "%d\n", *(int *)v); |
4813 | ||
4814 | return 0; | |
cc31edce | 4815 | } |
bbcb81d0 | 4816 | |
182446d0 TH |
4817 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4818 | struct cftype *cft) | |
81a6a5cd | 4819 | { |
182446d0 | 4820 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4821 | } |
4822 | ||
182446d0 TH |
4823 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4824 | struct cftype *cft, u64 val) | |
6379c106 | 4825 | { |
6379c106 | 4826 | if (val) |
182446d0 | 4827 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4828 | else |
182446d0 | 4829 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4830 | return 0; |
4831 | } | |
4832 | ||
182446d0 TH |
4833 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4834 | struct cftype *cft) | |
97978e6d | 4835 | { |
182446d0 | 4836 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4837 | } |
4838 | ||
182446d0 TH |
4839 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4840 | struct cftype *cft, u64 val) | |
97978e6d DL |
4841 | { |
4842 | if (val) | |
182446d0 | 4843 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4844 | else |
182446d0 | 4845 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4846 | return 0; |
4847 | } | |
4848 | ||
a14c6874 TH |
4849 | /* cgroup core interface files for the default hierarchy */ |
4850 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4851 | { |
d5c56ced | 4852 | .name = "cgroup.procs", |
6f60eade | 4853 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4854 | .seq_start = cgroup_pidlist_start, |
4855 | .seq_next = cgroup_pidlist_next, | |
4856 | .seq_stop = cgroup_pidlist_stop, | |
4857 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4858 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4859 | .write = cgroup_procs_write, |
102a775e | 4860 | }, |
f8f22e53 TH |
4861 | { |
4862 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4863 | .seq_show = cgroup_controllers_show, |
4864 | }, | |
4865 | { | |
4866 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4867 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4868 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4869 | }, |
842b597e | 4870 | { |
4a07c222 | 4871 | .name = "cgroup.events", |
a14c6874 | 4872 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4873 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4874 | .seq_show = cgroup_events_show, |
842b597e | 4875 | }, |
a14c6874 TH |
4876 | { } /* terminate */ |
4877 | }; | |
d5c56ced | 4878 | |
a14c6874 TH |
4879 | /* cgroup core interface files for the legacy hierarchies */ |
4880 | static struct cftype cgroup_legacy_base_files[] = { | |
4881 | { | |
4882 | .name = "cgroup.procs", | |
4883 | .seq_start = cgroup_pidlist_start, | |
4884 | .seq_next = cgroup_pidlist_next, | |
4885 | .seq_stop = cgroup_pidlist_stop, | |
4886 | .seq_show = cgroup_pidlist_show, | |
4887 | .private = CGROUP_FILE_PROCS, | |
4888 | .write = cgroup_procs_write, | |
a14c6874 TH |
4889 | }, |
4890 | { | |
4891 | .name = "cgroup.clone_children", | |
4892 | .read_u64 = cgroup_clone_children_read, | |
4893 | .write_u64 = cgroup_clone_children_write, | |
4894 | }, | |
4895 | { | |
4896 | .name = "cgroup.sane_behavior", | |
4897 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4898 | .seq_show = cgroup_sane_behavior_show, | |
4899 | }, | |
d5c56ced TH |
4900 | { |
4901 | .name = "tasks", | |
6612f05b TH |
4902 | .seq_start = cgroup_pidlist_start, |
4903 | .seq_next = cgroup_pidlist_next, | |
4904 | .seq_stop = cgroup_pidlist_stop, | |
4905 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4906 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4907 | .write = cgroup_tasks_write, |
d5c56ced TH |
4908 | }, |
4909 | { | |
4910 | .name = "notify_on_release", | |
d5c56ced TH |
4911 | .read_u64 = cgroup_read_notify_on_release, |
4912 | .write_u64 = cgroup_write_notify_on_release, | |
4913 | }, | |
6e6ff25b TH |
4914 | { |
4915 | .name = "release_agent", | |
a14c6874 | 4916 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4917 | .seq_show = cgroup_release_agent_show, |
451af504 | 4918 | .write = cgroup_release_agent_write, |
5f469907 | 4919 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4920 | }, |
db0416b6 | 4921 | { } /* terminate */ |
bbcb81d0 PM |
4922 | }; |
4923 | ||
0c21ead1 TH |
4924 | /* |
4925 | * css destruction is four-stage process. | |
4926 | * | |
4927 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4928 | * Implemented in kill_css(). | |
4929 | * | |
4930 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4931 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4932 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4933 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4934 | * |
4935 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4936 | * accessors are inside RCU read sections. css_release() schedules the | |
4937 | * RCU callback. | |
4938 | * | |
4939 | * 4. After the grace period, the css can be freed. Implemented in | |
4940 | * css_free_work_fn(). | |
4941 | * | |
4942 | * It is actually hairier because both step 2 and 4 require process context | |
4943 | * and thus involve punting to css->destroy_work adding two additional | |
4944 | * steps to the already complex sequence. | |
4945 | */ | |
35ef10da | 4946 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4947 | { |
4948 | struct cgroup_subsys_state *css = | |
35ef10da | 4949 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4950 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4951 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4952 | |
9a1049da TH |
4953 | percpu_ref_exit(&css->refcnt); |
4954 | ||
01e58659 | 4955 | if (ss) { |
9d755d33 | 4956 | /* css free path */ |
8bb5ef79 | 4957 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4958 | int id = css->id; |
4959 | ||
01e58659 VD |
4960 | ss->css_free(css); |
4961 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4962 | cgroup_put(cgrp); |
8bb5ef79 TH |
4963 | |
4964 | if (parent) | |
4965 | css_put(parent); | |
9d755d33 TH |
4966 | } else { |
4967 | /* cgroup free path */ | |
4968 | atomic_dec(&cgrp->root->nr_cgrps); | |
4969 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4970 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4971 | |
d51f39b0 | 4972 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4973 | /* |
4974 | * We get a ref to the parent, and put the ref when | |
4975 | * this cgroup is being freed, so it's guaranteed | |
4976 | * that the parent won't be destroyed before its | |
4977 | * children. | |
4978 | */ | |
d51f39b0 | 4979 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4980 | kernfs_put(cgrp->kn); |
4981 | kfree(cgrp); | |
4982 | } else { | |
4983 | /* | |
4984 | * This is root cgroup's refcnt reaching zero, | |
4985 | * which indicates that the root should be | |
4986 | * released. | |
4987 | */ | |
4988 | cgroup_destroy_root(cgrp->root); | |
4989 | } | |
4990 | } | |
48ddbe19 TH |
4991 | } |
4992 | ||
0c21ead1 | 4993 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4994 | { |
4995 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4996 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4997 | |
35ef10da | 4998 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4999 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
5000 | } |
5001 | ||
25e15d83 | 5002 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
5003 | { |
5004 | struct cgroup_subsys_state *css = | |
25e15d83 | 5005 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 5006 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 5007 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 5008 | |
1fed1b2e TH |
5009 | mutex_lock(&cgroup_mutex); |
5010 | ||
de3f0341 | 5011 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
5012 | list_del_rcu(&css->sibling); |
5013 | ||
9d755d33 TH |
5014 | if (ss) { |
5015 | /* css release path */ | |
01e58659 | 5016 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
5017 | if (ss->css_released) |
5018 | ss->css_released(css); | |
9d755d33 TH |
5019 | } else { |
5020 | /* cgroup release path */ | |
9d755d33 TH |
5021 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
5022 | cgrp->id = -1; | |
a4189487 LZ |
5023 | |
5024 | /* | |
5025 | * There are two control paths which try to determine | |
5026 | * cgroup from dentry without going through kernfs - | |
5027 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
5028 | * Those are supported by RCU protecting clearing of | |
5029 | * cgrp->kn->priv backpointer. | |
5030 | */ | |
6cd0f5bb TH |
5031 | if (cgrp->kn) |
5032 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
5033 | NULL); | |
9d755d33 | 5034 | } |
d3daf28d | 5035 | |
1fed1b2e TH |
5036 | mutex_unlock(&cgroup_mutex); |
5037 | ||
0c21ead1 | 5038 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
5039 | } |
5040 | ||
d3daf28d TH |
5041 | static void css_release(struct percpu_ref *ref) |
5042 | { | |
5043 | struct cgroup_subsys_state *css = | |
5044 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5045 | ||
25e15d83 TH |
5046 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5047 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5048 | } |
5049 | ||
ddfcadab TH |
5050 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5051 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5052 | { |
0cb51d71 TH |
5053 | lockdep_assert_held(&cgroup_mutex); |
5054 | ||
ddfcadab TH |
5055 | cgroup_get(cgrp); |
5056 | ||
d5c419b6 | 5057 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5058 | css->cgroup = cgrp; |
72c97e54 | 5059 | css->ss = ss; |
d5c419b6 TH |
5060 | INIT_LIST_HEAD(&css->sibling); |
5061 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 5062 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5063 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5064 | |
d51f39b0 TH |
5065 | if (cgroup_parent(cgrp)) { |
5066 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5067 | css_get(css->parent); |
ddfcadab | 5068 | } |
48ddbe19 | 5069 | |
ca8bdcaf | 5070 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5071 | } |
5072 | ||
2a4ac633 | 5073 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5074 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5075 | { |
623f926b | 5076 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5077 | int ret = 0; |
5078 | ||
a31f2d3f TH |
5079 | lockdep_assert_held(&cgroup_mutex); |
5080 | ||
92fb9748 | 5081 | if (ss->css_online) |
eb95419b | 5082 | ret = ss->css_online(css); |
ae7f164a | 5083 | if (!ret) { |
eb95419b | 5084 | css->flags |= CSS_ONLINE; |
aec25020 | 5085 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5086 | |
5087 | atomic_inc(&css->online_cnt); | |
5088 | if (css->parent) | |
5089 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5090 | } |
b1929db4 | 5091 | return ret; |
a31f2d3f TH |
5092 | } |
5093 | ||
2a4ac633 | 5094 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5095 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5096 | { |
623f926b | 5097 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5098 | |
5099 | lockdep_assert_held(&cgroup_mutex); | |
5100 | ||
5101 | if (!(css->flags & CSS_ONLINE)) | |
5102 | return; | |
5103 | ||
fa06235b VD |
5104 | if (ss->css_reset) |
5105 | ss->css_reset(css); | |
5106 | ||
d7eeac19 | 5107 | if (ss->css_offline) |
eb95419b | 5108 | ss->css_offline(css); |
a31f2d3f | 5109 | |
eb95419b | 5110 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5111 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5112 | |
5113 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5114 | } |
5115 | ||
c81c925a | 5116 | /** |
6cd0f5bb | 5117 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5118 | * @cgrp: the cgroup new css will be associated with |
5119 | * @ss: the subsys of new css | |
5120 | * | |
5121 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5122 | * css is online and installed in @cgrp. This function doesn't create the |
5123 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5124 | */ |
6cd0f5bb TH |
5125 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5126 | struct cgroup_subsys *ss) | |
c81c925a | 5127 | { |
d51f39b0 | 5128 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5129 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5130 | struct cgroup_subsys_state *css; |
5131 | int err; | |
5132 | ||
c81c925a TH |
5133 | lockdep_assert_held(&cgroup_mutex); |
5134 | ||
1fed1b2e | 5135 | css = ss->css_alloc(parent_css); |
c81c925a | 5136 | if (IS_ERR(css)) |
6cd0f5bb | 5137 | return css; |
c81c925a | 5138 | |
ddfcadab | 5139 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5140 | |
2aad2a86 | 5141 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5142 | if (err) |
3eb59ec6 | 5143 | goto err_free_css; |
c81c925a | 5144 | |
cf780b7d | 5145 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
5146 | if (err < 0) |
5147 | goto err_free_percpu_ref; | |
5148 | css->id = err; | |
c81c925a | 5149 | |
15a4c835 | 5150 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5151 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5152 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5153 | |
5154 | err = online_css(css); | |
5155 | if (err) | |
1fed1b2e | 5156 | goto err_list_del; |
94419627 | 5157 | |
c81c925a | 5158 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5159 | cgroup_parent(parent)) { |
ed3d261b | 5160 | 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 | 5161 | current->comm, current->pid, ss->name); |
c81c925a | 5162 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5163 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5164 | ss->warned_broken_hierarchy = true; |
5165 | } | |
5166 | ||
6cd0f5bb | 5167 | return css; |
c81c925a | 5168 | |
1fed1b2e TH |
5169 | err_list_del: |
5170 | list_del_rcu(&css->sibling); | |
15a4c835 | 5171 | cgroup_idr_remove(&ss->css_idr, css->id); |
3eb59ec6 | 5172 | err_free_percpu_ref: |
9a1049da | 5173 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 5174 | err_free_css: |
a2bed820 | 5175 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 5176 | return ERR_PTR(err); |
c81c925a TH |
5177 | } |
5178 | ||
a5bca215 | 5179 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 5180 | { |
a5bca215 | 5181 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
5182 | struct cgroup *cgrp, *tcgrp; |
5183 | int level = parent->level + 1; | |
03970d3c | 5184 | int ret; |
ddbcc7e8 | 5185 | |
0a950f65 | 5186 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
5187 | cgrp = kzalloc(sizeof(*cgrp) + |
5188 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
5189 | if (!cgrp) |
5190 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5191 | |
2aad2a86 | 5192 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5193 | if (ret) |
5194 | goto out_free_cgrp; | |
5195 | ||
0ab02ca8 LZ |
5196 | /* |
5197 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5198 | * a half-baked cgroup. | |
5199 | */ | |
cf780b7d | 5200 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5201 | if (cgrp->id < 0) { |
ba0f4d76 | 5202 | ret = -ENOMEM; |
9d755d33 | 5203 | goto out_cancel_ref; |
976c06bc TH |
5204 | } |
5205 | ||
cc31edce | 5206 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5207 | |
9d800df1 | 5208 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5209 | cgrp->root = root; |
b11cfb58 TH |
5210 | cgrp->level = level; |
5211 | ||
5212 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5213 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5214 | |
b6abdb0e LZ |
5215 | if (notify_on_release(parent)) |
5216 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5217 | ||
2260e7fc TH |
5218 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5219 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5220 | |
0cb51d71 | 5221 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5222 | |
4e139afc | 5223 | /* allocation complete, commit to creation */ |
d5c419b6 | 5224 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5225 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5226 | cgroup_get(parent); |
415cf07a | 5227 | |
0d80255e TH |
5228 | /* |
5229 | * @cgrp is now fully operational. If something fails after this | |
5230 | * point, it'll be released via the normal destruction path. | |
5231 | */ | |
6fa4918d | 5232 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5233 | |
bd53d617 TH |
5234 | /* |
5235 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5236 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5237 | */ |
03970d3c | 5238 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5239 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5240 | |
5241 | cgroup_propagate_control(cgrp); | |
5242 | ||
5243 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5244 | ret = cgroup_apply_control_enable(cgrp); | |
5245 | if (ret) | |
5246 | goto out_destroy; | |
2bd59d48 | 5247 | |
a5bca215 TH |
5248 | return cgrp; |
5249 | ||
5250 | out_cancel_ref: | |
5251 | percpu_ref_exit(&cgrp->self.refcnt); | |
5252 | out_free_cgrp: | |
5253 | kfree(cgrp); | |
5254 | return ERR_PTR(ret); | |
5255 | out_destroy: | |
5256 | cgroup_destroy_locked(cgrp); | |
5257 | return ERR_PTR(ret); | |
5258 | } | |
5259 | ||
5260 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5261 | umode_t mode) | |
5262 | { | |
5263 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5264 | struct kernfs_node *kn; |
03970d3c | 5265 | int ret; |
a5bca215 TH |
5266 | |
5267 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5268 | if (strchr(name, '\n')) | |
5269 | return -EINVAL; | |
5270 | ||
945ba199 | 5271 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5272 | if (!parent) |
5273 | return -ENODEV; | |
5274 | ||
5275 | cgrp = cgroup_create(parent); | |
5276 | if (IS_ERR(cgrp)) { | |
5277 | ret = PTR_ERR(cgrp); | |
5278 | goto out_unlock; | |
5279 | } | |
5280 | ||
195e9b6c TH |
5281 | /* create the directory */ |
5282 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5283 | if (IS_ERR(kn)) { | |
5284 | ret = PTR_ERR(kn); | |
5285 | goto out_destroy; | |
5286 | } | |
5287 | cgrp->kn = kn; | |
5288 | ||
5289 | /* | |
5290 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5291 | * that @cgrp->kn is always accessible. | |
5292 | */ | |
5293 | kernfs_get(kn); | |
5294 | ||
5295 | ret = cgroup_kn_set_ugid(kn); | |
5296 | if (ret) | |
5297 | goto out_destroy; | |
5298 | ||
334c3679 | 5299 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5300 | if (ret) |
5301 | goto out_destroy; | |
5302 | ||
03970d3c TH |
5303 | ret = cgroup_apply_control_enable(cgrp); |
5304 | if (ret) | |
5305 | goto out_destroy; | |
195e9b6c TH |
5306 | |
5307 | /* let's create and online css's */ | |
2bd59d48 | 5308 | kernfs_activate(kn); |
ddbcc7e8 | 5309 | |
ba0f4d76 TH |
5310 | ret = 0; |
5311 | goto out_unlock; | |
ddbcc7e8 | 5312 | |
a5bca215 TH |
5313 | out_destroy: |
5314 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5315 | out_unlock: |
a9746d8d | 5316 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5317 | return ret; |
ddbcc7e8 PM |
5318 | } |
5319 | ||
223dbc38 TH |
5320 | /* |
5321 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5322 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5323 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5324 | */ |
5325 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5326 | { |
223dbc38 TH |
5327 | struct cgroup_subsys_state *css = |
5328 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5329 | |
f20104de | 5330 | mutex_lock(&cgroup_mutex); |
09a503ea | 5331 | |
aa226ff4 TH |
5332 | do { |
5333 | offline_css(css); | |
5334 | css_put(css); | |
5335 | /* @css can't go away while we're holding cgroup_mutex */ | |
5336 | css = css->parent; | |
5337 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5338 | ||
5339 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5340 | } |
5341 | ||
223dbc38 TH |
5342 | /* css kill confirmation processing requires process context, bounce */ |
5343 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5344 | { |
5345 | struct cgroup_subsys_state *css = | |
5346 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5347 | ||
aa226ff4 TH |
5348 | if (atomic_dec_and_test(&css->online_cnt)) { |
5349 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5350 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5351 | } | |
d3daf28d TH |
5352 | } |
5353 | ||
f392e51c TH |
5354 | /** |
5355 | * kill_css - destroy a css | |
5356 | * @css: css to destroy | |
5357 | * | |
5358 | * This function initiates destruction of @css by removing cgroup interface | |
5359 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5360 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5361 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5362 | */ |
5363 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5364 | { |
01f6474c | 5365 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5366 | |
2bd59d48 TH |
5367 | /* |
5368 | * This must happen before css is disassociated with its cgroup. | |
5369 | * See seq_css() for details. | |
5370 | */ | |
334c3679 | 5371 | css_clear_dir(css); |
3c14f8b4 | 5372 | |
edae0c33 TH |
5373 | /* |
5374 | * Killing would put the base ref, but we need to keep it alive | |
5375 | * until after ->css_offline(). | |
5376 | */ | |
5377 | css_get(css); | |
5378 | ||
5379 | /* | |
5380 | * cgroup core guarantees that, by the time ->css_offline() is | |
5381 | * invoked, no new css reference will be given out via | |
ec903c0c | 5382 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5383 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5384 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5385 | * | |
5386 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5387 | * css is confirmed to be seen as killed on all CPUs. | |
5388 | */ | |
5389 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5390 | } |
5391 | ||
5392 | /** | |
5393 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5394 | * @cgrp: cgroup to be destroyed | |
5395 | * | |
5396 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5397 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5398 | * guarantee that css_tryget_online() won't succeed by the time |
5399 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5400 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5401 | * |
5402 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5403 | * userland visible parts and start killing the percpu refcnts of | |
5404 | * css's. Set up so that the next stage will be kicked off once all | |
5405 | * the percpu refcnts are confirmed to be killed. | |
5406 | * | |
5407 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5408 | * rest of destruction. Once all cgroup references are gone, the | |
5409 | * cgroup is RCU-freed. | |
5410 | * | |
5411 | * This function implements s1. After this step, @cgrp is gone as far as | |
5412 | * the userland is concerned and a new cgroup with the same name may be | |
5413 | * created. As cgroup doesn't care about the names internally, this | |
5414 | * doesn't cause any problem. | |
5415 | */ | |
42809dd4 TH |
5416 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5417 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5418 | { |
2bd59d48 | 5419 | struct cgroup_subsys_state *css; |
2b021cbf | 5420 | struct cgrp_cset_link *link; |
1c6727af | 5421 | int ssid; |
ddbcc7e8 | 5422 | |
42809dd4 TH |
5423 | lockdep_assert_held(&cgroup_mutex); |
5424 | ||
91486f61 TH |
5425 | /* |
5426 | * Only migration can raise populated from zero and we're already | |
5427 | * holding cgroup_mutex. | |
5428 | */ | |
5429 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5430 | return -EBUSY; |
a043e3b2 | 5431 | |
bb78a92f | 5432 | /* |
d5c419b6 TH |
5433 | * Make sure there's no live children. We can't test emptiness of |
5434 | * ->self.children as dead children linger on it while being | |
5435 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5436 | */ |
f3d46500 | 5437 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5438 | return -EBUSY; |
5439 | ||
455050d2 | 5440 | /* |
2b021cbf TH |
5441 | * Mark @cgrp and the associated csets dead. The former prevents |
5442 | * further task migration and child creation by disabling | |
5443 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5444 | * the migration path. | |
455050d2 | 5445 | */ |
184faf32 | 5446 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5447 | |
2b021cbf TH |
5448 | spin_lock_bh(&css_set_lock); |
5449 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
5450 | link->cset->dead = true; | |
5451 | spin_unlock_bh(&css_set_lock); | |
5452 | ||
249f3468 | 5453 | /* initiate massacre of all css's */ |
1c6727af TH |
5454 | for_each_css(css, ssid, cgrp) |
5455 | kill_css(css); | |
455050d2 | 5456 | |
455050d2 | 5457 | /* |
01f6474c TH |
5458 | * Remove @cgrp directory along with the base files. @cgrp has an |
5459 | * extra ref on its kn. | |
f20104de | 5460 | */ |
01f6474c | 5461 | kernfs_remove(cgrp->kn); |
f20104de | 5462 | |
d51f39b0 | 5463 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5464 | |
249f3468 | 5465 | /* put the base reference */ |
9d755d33 | 5466 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5467 | |
ea15f8cc TH |
5468 | return 0; |
5469 | }; | |
5470 | ||
2bd59d48 | 5471 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5472 | { |
a9746d8d | 5473 | struct cgroup *cgrp; |
2bd59d48 | 5474 | int ret = 0; |
42809dd4 | 5475 | |
945ba199 | 5476 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5477 | if (!cgrp) |
5478 | return 0; | |
42809dd4 | 5479 | |
a9746d8d | 5480 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5481 | |
a9746d8d | 5482 | cgroup_kn_unlock(kn); |
42809dd4 | 5483 | return ret; |
8e3f6541 TH |
5484 | } |
5485 | ||
2bd59d48 TH |
5486 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5487 | .remount_fs = cgroup_remount, | |
5488 | .show_options = cgroup_show_options, | |
5489 | .mkdir = cgroup_mkdir, | |
5490 | .rmdir = cgroup_rmdir, | |
5491 | .rename = cgroup_rename, | |
4f41fc59 | 5492 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5493 | }; |
5494 | ||
15a4c835 | 5495 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5496 | { |
ddbcc7e8 | 5497 | struct cgroup_subsys_state *css; |
cfe36bde | 5498 | |
a5ae9899 | 5499 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5500 | |
648bb56d TH |
5501 | mutex_lock(&cgroup_mutex); |
5502 | ||
15a4c835 | 5503 | idr_init(&ss->css_idr); |
0adb0704 | 5504 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5505 | |
3dd06ffa TH |
5506 | /* Create the root cgroup state for this subsystem */ |
5507 | ss->root = &cgrp_dfl_root; | |
5508 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5509 | /* We don't handle early failures gracefully */ |
5510 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5511 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5512 | |
5513 | /* | |
5514 | * Root csses are never destroyed and we can't initialize | |
5515 | * percpu_ref during early init. Disable refcnting. | |
5516 | */ | |
5517 | css->flags |= CSS_NO_REF; | |
5518 | ||
15a4c835 | 5519 | if (early) { |
9395a450 | 5520 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5521 | css->id = 1; |
5522 | } else { | |
5523 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5524 | BUG_ON(css->id < 0); | |
5525 | } | |
ddbcc7e8 | 5526 | |
e8d55fde | 5527 | /* Update the init_css_set to contain a subsys |
817929ec | 5528 | * pointer to this state - since the subsystem is |
e8d55fde | 5529 | * newly registered, all tasks and hence the |
3dd06ffa | 5530 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5531 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5532 | |
cb4a3167 AS |
5533 | have_fork_callback |= (bool)ss->fork << ss->id; |
5534 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5535 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5536 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5537 | |
e8d55fde LZ |
5538 | /* At system boot, before all subsystems have been |
5539 | * registered, no tasks have been forked, so we don't | |
5540 | * need to invoke fork callbacks here. */ | |
5541 | BUG_ON(!list_empty(&init_task.tasks)); | |
5542 | ||
ae7f164a | 5543 | BUG_ON(online_css(css)); |
a8638030 | 5544 | |
cf5d5941 BB |
5545 | mutex_unlock(&cgroup_mutex); |
5546 | } | |
cf5d5941 | 5547 | |
ddbcc7e8 | 5548 | /** |
a043e3b2 LZ |
5549 | * cgroup_init_early - cgroup initialization at system boot |
5550 | * | |
5551 | * Initialize cgroups at system boot, and initialize any | |
5552 | * subsystems that request early init. | |
ddbcc7e8 PM |
5553 | */ |
5554 | int __init cgroup_init_early(void) | |
5555 | { | |
7b9a6ba5 | 5556 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5557 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5558 | int i; |
30159ec7 | 5559 | |
3dd06ffa | 5560 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5561 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5562 | ||
a4ea1cc9 | 5563 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5564 | |
3ed80a62 | 5565 | for_each_subsys(ss, i) { |
aec25020 | 5566 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5567 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5568 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5569 | ss->id, ss->name); |
073219e9 TH |
5570 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5571 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5572 | ||
aec25020 | 5573 | ss->id = i; |
073219e9 | 5574 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5575 | if (!ss->legacy_name) |
5576 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5577 | |
5578 | if (ss->early_init) | |
15a4c835 | 5579 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5580 | } |
5581 | return 0; | |
5582 | } | |
5583 | ||
6e5c8307 | 5584 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5585 | |
ddbcc7e8 | 5586 | /** |
a043e3b2 LZ |
5587 | * cgroup_init - cgroup initialization |
5588 | * | |
5589 | * Register cgroup filesystem and /proc file, and initialize | |
5590 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5591 | */ |
5592 | int __init cgroup_init(void) | |
5593 | { | |
30159ec7 | 5594 | struct cgroup_subsys *ss; |
035f4f51 | 5595 | int ssid; |
ddbcc7e8 | 5596 | |
6e5c8307 | 5597 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5598 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5599 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5600 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5601 | |
a79a908f AK |
5602 | get_user_ns(init_cgroup_ns.user_ns); |
5603 | ||
54e7b4eb | 5604 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5605 | |
2378d8b8 TH |
5606 | /* |
5607 | * Add init_css_set to the hash table so that dfl_root can link to | |
5608 | * it during init. | |
5609 | */ | |
5610 | hash_add(css_set_table, &init_css_set.hlist, | |
5611 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5612 | |
3dd06ffa | 5613 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5614 | |
54e7b4eb TH |
5615 | mutex_unlock(&cgroup_mutex); |
5616 | ||
172a2c06 | 5617 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5618 | if (ss->early_init) { |
5619 | struct cgroup_subsys_state *css = | |
5620 | init_css_set.subsys[ss->id]; | |
5621 | ||
5622 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5623 | GFP_KERNEL); | |
5624 | BUG_ON(css->id < 0); | |
5625 | } else { | |
5626 | cgroup_init_subsys(ss, false); | |
5627 | } | |
172a2c06 | 5628 | |
2d8f243a TH |
5629 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5630 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5631 | |
5632 | /* | |
c731ae1d LZ |
5633 | * Setting dfl_root subsys_mask needs to consider the |
5634 | * disabled flag and cftype registration needs kmalloc, | |
5635 | * both of which aren't available during early_init. | |
172a2c06 | 5636 | */ |
a3e72739 TH |
5637 | if (cgroup_disable_mask & (1 << ssid)) { |
5638 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5639 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5640 | ss->name); | |
a8ddc821 | 5641 | continue; |
a3e72739 | 5642 | } |
a8ddc821 | 5643 | |
223ffb29 JW |
5644 | if (cgroup_ssid_no_v1(ssid)) |
5645 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5646 | ss->name); | |
5647 | ||
a8ddc821 TH |
5648 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5649 | ||
f6d635ad TH |
5650 | if (ss->implicit_on_dfl) |
5651 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5652 | else if (!ss->dfl_cftypes) | |
a7165264 | 5653 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5654 | |
a8ddc821 TH |
5655 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5656 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5657 | } else { | |
5658 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5659 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5660 | } |
295458e6 VD |
5661 | |
5662 | if (ss->bind) | |
5663 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
5664 | } |
5665 | ||
2378d8b8 TH |
5666 | /* init_css_set.subsys[] has been updated, re-hash */ |
5667 | hash_del(&init_css_set.hlist); | |
5668 | hash_add(css_set_table, &init_css_set.hlist, | |
5669 | css_set_hash(init_css_set.subsys)); | |
5670 | ||
035f4f51 TH |
5671 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5672 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5673 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5674 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5675 | |
2bd59d48 | 5676 | return 0; |
ddbcc7e8 | 5677 | } |
b4f48b63 | 5678 | |
e5fca243 TH |
5679 | static int __init cgroup_wq_init(void) |
5680 | { | |
5681 | /* | |
5682 | * There isn't much point in executing destruction path in | |
5683 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5684 | * Use 1 for @max_active. |
e5fca243 TH |
5685 | * |
5686 | * We would prefer to do this in cgroup_init() above, but that | |
5687 | * is called before init_workqueues(): so leave this until after. | |
5688 | */ | |
1a11533f | 5689 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5690 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5691 | |
5692 | /* | |
5693 | * Used to destroy pidlists and separate to serve as flush domain. | |
5694 | * Cap @max_active to 1 too. | |
5695 | */ | |
5696 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5697 | 0, 1); | |
5698 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5699 | ||
e5fca243 TH |
5700 | return 0; |
5701 | } | |
5702 | core_initcall(cgroup_wq_init); | |
5703 | ||
a424316c PM |
5704 | /* |
5705 | * proc_cgroup_show() | |
5706 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5707 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5708 | */ |
006f4ac4 ZL |
5709 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5710 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5711 | { |
e61734c5 | 5712 | char *buf, *path; |
a424316c | 5713 | int retval; |
3dd06ffa | 5714 | struct cgroup_root *root; |
a424316c PM |
5715 | |
5716 | retval = -ENOMEM; | |
e61734c5 | 5717 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5718 | if (!buf) |
5719 | goto out; | |
5720 | ||
a424316c | 5721 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5722 | spin_lock_bh(&css_set_lock); |
a424316c | 5723 | |
985ed670 | 5724 | for_each_root(root) { |
a424316c | 5725 | struct cgroup_subsys *ss; |
bd89aabc | 5726 | struct cgroup *cgrp; |
b85d2040 | 5727 | int ssid, count = 0; |
a424316c | 5728 | |
a7165264 | 5729 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5730 | continue; |
5731 | ||
2c6ab6d2 | 5732 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5733 | if (root != &cgrp_dfl_root) |
5734 | for_each_subsys(ss, ssid) | |
5735 | if (root->subsys_mask & (1 << ssid)) | |
5736 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5737 | ss->legacy_name); |
c6d57f33 PM |
5738 | if (strlen(root->name)) |
5739 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5740 | root->name); | |
a424316c | 5741 | seq_putc(m, ':'); |
2e91fa7f | 5742 | |
7717f7ba | 5743 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5744 | |
5745 | /* | |
5746 | * On traditional hierarchies, all zombie tasks show up as | |
5747 | * belonging to the root cgroup. On the default hierarchy, | |
5748 | * while a zombie doesn't show up in "cgroup.procs" and | |
5749 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5750 | * reporting the cgroup it belonged to before exiting. If | |
5751 | * the cgroup is removed before the zombie is reaped, | |
5752 | * " (deleted)" is appended to the cgroup path. | |
5753 | */ | |
5754 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
a79a908f AK |
5755 | path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
5756 | current->nsproxy->cgroup_ns); | |
2e91fa7f TH |
5757 | if (!path) { |
5758 | retval = -ENAMETOOLONG; | |
5759 | goto out_unlock; | |
5760 | } | |
5761 | } else { | |
5762 | path = "/"; | |
e61734c5 | 5763 | } |
2e91fa7f | 5764 | |
e61734c5 | 5765 | seq_puts(m, path); |
2e91fa7f TH |
5766 | |
5767 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5768 | seq_puts(m, " (deleted)\n"); | |
5769 | else | |
5770 | seq_putc(m, '\n'); | |
a424316c PM |
5771 | } |
5772 | ||
006f4ac4 | 5773 | retval = 0; |
a424316c | 5774 | out_unlock: |
f0d9a5f1 | 5775 | spin_unlock_bh(&css_set_lock); |
a424316c | 5776 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5777 | kfree(buf); |
5778 | out: | |
5779 | return retval; | |
5780 | } | |
5781 | ||
a424316c PM |
5782 | /* Display information about each subsystem and each hierarchy */ |
5783 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5784 | { | |
30159ec7 | 5785 | struct cgroup_subsys *ss; |
a424316c | 5786 | int i; |
a424316c | 5787 | |
8bab8dde | 5788 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5789 | /* |
5790 | * ideally we don't want subsystems moving around while we do this. | |
5791 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5792 | * subsys/hierarchy state. | |
5793 | */ | |
a424316c | 5794 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5795 | |
5796 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5797 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5798 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5799 | atomic_read(&ss->root->nr_cgrps), |
5800 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5801 | |
a424316c PM |
5802 | mutex_unlock(&cgroup_mutex); |
5803 | return 0; | |
5804 | } | |
5805 | ||
5806 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5807 | { | |
9dce07f1 | 5808 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5809 | } |
5810 | ||
828c0950 | 5811 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5812 | .open = cgroupstats_open, |
5813 | .read = seq_read, | |
5814 | .llseek = seq_lseek, | |
5815 | .release = single_release, | |
5816 | }; | |
5817 | ||
b4f48b63 | 5818 | /** |
eaf797ab | 5819 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5820 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5821 | * |
eaf797ab TH |
5822 | * A task is associated with the init_css_set until cgroup_post_fork() |
5823 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5824 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5825 | */ |
5826 | void cgroup_fork(struct task_struct *child) | |
5827 | { | |
eaf797ab | 5828 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5829 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5830 | } |
5831 | ||
7e47682e AS |
5832 | /** |
5833 | * cgroup_can_fork - called on a new task before the process is exposed | |
5834 | * @child: the task in question. | |
5835 | * | |
5836 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5837 | * returns an error, the fork aborts with that error code. This allows for | |
5838 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5839 | */ | |
b53202e6 | 5840 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5841 | { |
5842 | struct cgroup_subsys *ss; | |
5843 | int i, j, ret; | |
5844 | ||
b4e0eeaf | 5845 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5846 | ret = ss->can_fork(child); |
7e47682e AS |
5847 | if (ret) |
5848 | goto out_revert; | |
b4e0eeaf | 5849 | } while_each_subsys_mask(); |
7e47682e AS |
5850 | |
5851 | return 0; | |
5852 | ||
5853 | out_revert: | |
5854 | for_each_subsys(ss, j) { | |
5855 | if (j >= i) | |
5856 | break; | |
5857 | if (ss->cancel_fork) | |
b53202e6 | 5858 | ss->cancel_fork(child); |
7e47682e AS |
5859 | } |
5860 | ||
5861 | return ret; | |
5862 | } | |
5863 | ||
5864 | /** | |
5865 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5866 | * @child: the task in question | |
5867 | * | |
5868 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5869 | * cgroup_can_fork() succeded. | |
5870 | */ | |
b53202e6 | 5871 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5872 | { |
5873 | struct cgroup_subsys *ss; | |
5874 | int i; | |
5875 | ||
5876 | for_each_subsys(ss, i) | |
5877 | if (ss->cancel_fork) | |
b53202e6 | 5878 | ss->cancel_fork(child); |
7e47682e AS |
5879 | } |
5880 | ||
817929ec | 5881 | /** |
a043e3b2 LZ |
5882 | * cgroup_post_fork - called on a new task after adding it to the task list |
5883 | * @child: the task in question | |
5884 | * | |
5edee61e TH |
5885 | * Adds the task to the list running through its css_set if necessary and |
5886 | * call the subsystem fork() callbacks. Has to be after the task is | |
5887 | * visible on the task list in case we race with the first call to | |
0942eeee | 5888 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5889 | * list. |
a043e3b2 | 5890 | */ |
b53202e6 | 5891 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5892 | { |
30159ec7 | 5893 | struct cgroup_subsys *ss; |
5edee61e TH |
5894 | int i; |
5895 | ||
3ce3230a | 5896 | /* |
251f8c03 | 5897 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5898 | * function sets use_task_css_set_links before grabbing |
5899 | * tasklist_lock and we just went through tasklist_lock to add | |
5900 | * @child, it's guaranteed that either we see the set | |
5901 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5902 | * @child during its iteration. | |
5903 | * | |
5904 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5905 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5906 | * association is stable, and, on completion of the parent's |
5907 | * migration, @child is visible in the source of migration or | |
5908 | * already in the destination cgroup. This guarantee is necessary | |
5909 | * when implementing operations which need to migrate all tasks of | |
5910 | * a cgroup to another. | |
5911 | * | |
251f8c03 | 5912 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5913 | * will remain in init_css_set. This is safe because all tasks are |
5914 | * in the init_css_set before cg_links is enabled and there's no | |
5915 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5916 | */ |
817929ec | 5917 | if (use_task_css_set_links) { |
eaf797ab TH |
5918 | struct css_set *cset; |
5919 | ||
f0d9a5f1 | 5920 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5921 | cset = task_css_set(current); |
eaf797ab | 5922 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5923 | get_css_set(cset); |
f6d7d049 | 5924 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5925 | } |
f0d9a5f1 | 5926 | spin_unlock_bh(&css_set_lock); |
817929ec | 5927 | } |
5edee61e TH |
5928 | |
5929 | /* | |
5930 | * Call ss->fork(). This must happen after @child is linked on | |
5931 | * css_set; otherwise, @child might change state between ->fork() | |
5932 | * and addition to css_set. | |
5933 | */ | |
b4e0eeaf | 5934 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5935 | ss->fork(child); |
b4e0eeaf | 5936 | } while_each_subsys_mask(); |
817929ec | 5937 | } |
5edee61e | 5938 | |
b4f48b63 PM |
5939 | /** |
5940 | * cgroup_exit - detach cgroup from exiting task | |
5941 | * @tsk: pointer to task_struct of exiting process | |
5942 | * | |
5943 | * Description: Detach cgroup from @tsk and release it. | |
5944 | * | |
5945 | * Note that cgroups marked notify_on_release force every task in | |
5946 | * them to take the global cgroup_mutex mutex when exiting. | |
5947 | * This could impact scaling on very large systems. Be reluctant to | |
5948 | * use notify_on_release cgroups where very high task exit scaling | |
5949 | * is required on large systems. | |
5950 | * | |
0e1d768f TH |
5951 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5952 | * call cgroup_exit() while the task is still competent to handle | |
5953 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5954 | * each hierarchy for the remainder of its exit. No need to bother with | |
5955 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5956 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5957 | */ |
1ec41830 | 5958 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5959 | { |
30159ec7 | 5960 | struct cgroup_subsys *ss; |
5abb8855 | 5961 | struct css_set *cset; |
d41d5a01 | 5962 | int i; |
817929ec PM |
5963 | |
5964 | /* | |
0e1d768f | 5965 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5966 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5967 | */ |
0de0942d TH |
5968 | cset = task_css_set(tsk); |
5969 | ||
817929ec | 5970 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5971 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5972 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5973 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5974 | } else { |
5975 | get_css_set(cset); | |
817929ec PM |
5976 | } |
5977 | ||
cb4a3167 | 5978 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5979 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5980 | ss->exit(tsk); |
b4e0eeaf | 5981 | } while_each_subsys_mask(); |
2e91fa7f | 5982 | } |
30159ec7 | 5983 | |
2e91fa7f TH |
5984 | void cgroup_free(struct task_struct *task) |
5985 | { | |
5986 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5987 | struct cgroup_subsys *ss; |
5988 | int ssid; | |
5989 | ||
b4e0eeaf | 5990 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 5991 | ss->free(task); |
b4e0eeaf | 5992 | } while_each_subsys_mask(); |
d41d5a01 | 5993 | |
2e91fa7f | 5994 | put_css_set(cset); |
b4f48b63 | 5995 | } |
697f4161 | 5996 | |
bd89aabc | 5997 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5998 | { |
27bd4dbb | 5999 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
6000 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
6001 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
6002 | } |
6003 | ||
81a6a5cd PM |
6004 | /* |
6005 | * Notify userspace when a cgroup is released, by running the | |
6006 | * configured release agent with the name of the cgroup (path | |
6007 | * relative to the root of cgroup file system) as the argument. | |
6008 | * | |
6009 | * Most likely, this user command will try to rmdir this cgroup. | |
6010 | * | |
6011 | * This races with the possibility that some other task will be | |
6012 | * attached to this cgroup before it is removed, or that some other | |
6013 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
6014 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
6015 | * unused, and this cgroup will be reprieved from its death sentence, | |
6016 | * to continue to serve a useful existence. Next time it's released, | |
6017 | * we will get notified again, if it still has 'notify_on_release' set. | |
6018 | * | |
6019 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
6020 | * means only wait until the task is successfully execve()'d. The | |
6021 | * separate release agent task is forked by call_usermodehelper(), | |
6022 | * then control in this thread returns here, without waiting for the | |
6023 | * release agent task. We don't bother to wait because the caller of | |
6024 | * this routine has no use for the exit status of the release agent | |
6025 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 6026 | */ |
81a6a5cd PM |
6027 | static void cgroup_release_agent(struct work_struct *work) |
6028 | { | |
971ff493 ZL |
6029 | struct cgroup *cgrp = |
6030 | container_of(work, struct cgroup, release_agent_work); | |
6031 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
6032 | char *argv[3], *envp[3]; | |
6033 | ||
81a6a5cd | 6034 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
6035 | |
6036 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
6037 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
6038 | if (!pathbuf || !agentbuf) | |
6039 | goto out; | |
6040 | ||
a79a908f AK |
6041 | spin_lock_bh(&css_set_lock); |
6042 | path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
6043 | spin_unlock_bh(&css_set_lock); | |
971ff493 ZL |
6044 | if (!path) |
6045 | goto out; | |
6046 | ||
6047 | argv[0] = agentbuf; | |
6048 | argv[1] = path; | |
6049 | argv[2] = NULL; | |
6050 | ||
6051 | /* minimal command environment */ | |
6052 | envp[0] = "HOME=/"; | |
6053 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
6054 | envp[2] = NULL; | |
6055 | ||
81a6a5cd | 6056 | mutex_unlock(&cgroup_mutex); |
971ff493 | 6057 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 6058 | goto out_free; |
971ff493 | 6059 | out: |
81a6a5cd | 6060 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 6061 | out_free: |
971ff493 ZL |
6062 | kfree(agentbuf); |
6063 | kfree(pathbuf); | |
81a6a5cd | 6064 | } |
8bab8dde PM |
6065 | |
6066 | static int __init cgroup_disable(char *str) | |
6067 | { | |
30159ec7 | 6068 | struct cgroup_subsys *ss; |
8bab8dde | 6069 | char *token; |
30159ec7 | 6070 | int i; |
8bab8dde PM |
6071 | |
6072 | while ((token = strsep(&str, ",")) != NULL) { | |
6073 | if (!*token) | |
6074 | continue; | |
be45c900 | 6075 | |
3ed80a62 | 6076 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6077 | if (strcmp(token, ss->name) && |
6078 | strcmp(token, ss->legacy_name)) | |
6079 | continue; | |
a3e72739 | 6080 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
6081 | } |
6082 | } | |
6083 | return 1; | |
6084 | } | |
6085 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6086 | |
223ffb29 JW |
6087 | static int __init cgroup_no_v1(char *str) |
6088 | { | |
6089 | struct cgroup_subsys *ss; | |
6090 | char *token; | |
6091 | int i; | |
6092 | ||
6093 | while ((token = strsep(&str, ",")) != NULL) { | |
6094 | if (!*token) | |
6095 | continue; | |
6096 | ||
6097 | if (!strcmp(token, "all")) { | |
6e5c8307 | 6098 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
6099 | break; |
6100 | } | |
6101 | ||
6102 | for_each_subsys(ss, i) { | |
6103 | if (strcmp(token, ss->name) && | |
6104 | strcmp(token, ss->legacy_name)) | |
6105 | continue; | |
6106 | ||
6107 | cgroup_no_v1_mask |= 1 << i; | |
6108 | } | |
6109 | } | |
6110 | return 1; | |
6111 | } | |
6112 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
6113 | ||
b77d7b60 | 6114 | /** |
ec903c0c | 6115 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6116 | * @dentry: directory dentry of interest |
6117 | * @ss: subsystem of interest | |
b77d7b60 | 6118 | * |
5a17f543 TH |
6119 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6120 | * to get the corresponding css and return it. If such css doesn't exist | |
6121 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6122 | */ |
ec903c0c TH |
6123 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6124 | struct cgroup_subsys *ss) | |
e5d1367f | 6125 | { |
2bd59d48 | 6126 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6127 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6128 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6129 | struct cgroup *cgrp; |
e5d1367f | 6130 | |
35cf0836 | 6131 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6132 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6133 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6134 | return ERR_PTR(-EBADF); |
6135 | ||
5a17f543 TH |
6136 | rcu_read_lock(); |
6137 | ||
2bd59d48 TH |
6138 | /* |
6139 | * This path doesn't originate from kernfs and @kn could already | |
6140 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6141 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
6142 | */ |
6143 | cgrp = rcu_dereference(kn->priv); | |
6144 | if (cgrp) | |
6145 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6146 | |
ec903c0c | 6147 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6148 | css = ERR_PTR(-ENOENT); |
6149 | ||
6150 | rcu_read_unlock(); | |
6151 | return css; | |
e5d1367f | 6152 | } |
e5d1367f | 6153 | |
1cb650b9 LZ |
6154 | /** |
6155 | * css_from_id - lookup css by id | |
6156 | * @id: the cgroup id | |
6157 | * @ss: cgroup subsys to be looked into | |
6158 | * | |
6159 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6160 | * Should be called under rcu_read_lock(). | |
6161 | */ | |
6162 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6163 | { | |
6fa4918d | 6164 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 6165 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
6166 | } |
6167 | ||
16af4396 TH |
6168 | /** |
6169 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6170 | * @path: path on the default hierarchy | |
6171 | * | |
6172 | * Find the cgroup at @path on the default hierarchy, increment its | |
6173 | * reference count and return it. Returns pointer to the found cgroup on | |
6174 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6175 | * if @path points to a non-directory. | |
6176 | */ | |
6177 | struct cgroup *cgroup_get_from_path(const char *path) | |
6178 | { | |
6179 | struct kernfs_node *kn; | |
6180 | struct cgroup *cgrp; | |
6181 | ||
6182 | mutex_lock(&cgroup_mutex); | |
6183 | ||
6184 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6185 | if (kn) { | |
6186 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6187 | cgrp = kn->priv; | |
6188 | cgroup_get(cgrp); | |
6189 | } else { | |
6190 | cgrp = ERR_PTR(-ENOTDIR); | |
6191 | } | |
6192 | kernfs_put(kn); | |
6193 | } else { | |
6194 | cgrp = ERR_PTR(-ENOENT); | |
6195 | } | |
6196 | ||
6197 | mutex_unlock(&cgroup_mutex); | |
6198 | return cgrp; | |
6199 | } | |
6200 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6201 | ||
bd1060a1 TH |
6202 | /* |
6203 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6204 | * definition in cgroup-defs.h. | |
6205 | */ | |
6206 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6207 | ||
6208 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6209 | ||
3fa4cc9c | 6210 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6211 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6212 | ||
6213 | void cgroup_sk_alloc_disable(void) | |
6214 | { | |
6215 | if (cgroup_sk_alloc_disabled) | |
6216 | return; | |
6217 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6218 | cgroup_sk_alloc_disabled = true; | |
6219 | } | |
6220 | ||
6221 | #else | |
6222 | ||
6223 | #define cgroup_sk_alloc_disabled false | |
6224 | ||
6225 | #endif | |
6226 | ||
6227 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6228 | { | |
6229 | if (cgroup_sk_alloc_disabled) | |
6230 | return; | |
6231 | ||
6232 | rcu_read_lock(); | |
6233 | ||
6234 | while (true) { | |
6235 | struct css_set *cset; | |
6236 | ||
6237 | cset = task_css_set(current); | |
6238 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6239 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6240 | break; | |
6241 | } | |
6242 | cpu_relax(); | |
6243 | } | |
6244 | ||
6245 | rcu_read_unlock(); | |
6246 | } | |
6247 | ||
6248 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6249 | { | |
6250 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6251 | } | |
6252 | ||
6253 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6254 | ||
a79a908f AK |
6255 | /* cgroup namespaces */ |
6256 | ||
6257 | static struct cgroup_namespace *alloc_cgroup_ns(void) | |
6258 | { | |
6259 | struct cgroup_namespace *new_ns; | |
6260 | int ret; | |
6261 | ||
6262 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); | |
6263 | if (!new_ns) | |
6264 | return ERR_PTR(-ENOMEM); | |
6265 | ret = ns_alloc_inum(&new_ns->ns); | |
6266 | if (ret) { | |
6267 | kfree(new_ns); | |
6268 | return ERR_PTR(ret); | |
6269 | } | |
6270 | atomic_set(&new_ns->count, 1); | |
6271 | new_ns->ns.ops = &cgroupns_operations; | |
6272 | return new_ns; | |
6273 | } | |
6274 | ||
6275 | void free_cgroup_ns(struct cgroup_namespace *ns) | |
6276 | { | |
6277 | put_css_set(ns->root_cset); | |
6278 | put_user_ns(ns->user_ns); | |
6279 | ns_free_inum(&ns->ns); | |
6280 | kfree(ns); | |
6281 | } | |
6282 | EXPORT_SYMBOL(free_cgroup_ns); | |
6283 | ||
6284 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, | |
6285 | struct user_namespace *user_ns, | |
6286 | struct cgroup_namespace *old_ns) | |
6287 | { | |
fa5ff8a1 TH |
6288 | struct cgroup_namespace *new_ns; |
6289 | struct css_set *cset; | |
a79a908f AK |
6290 | |
6291 | BUG_ON(!old_ns); | |
6292 | ||
6293 | if (!(flags & CLONE_NEWCGROUP)) { | |
6294 | get_cgroup_ns(old_ns); | |
6295 | return old_ns; | |
6296 | } | |
6297 | ||
6298 | /* Allow only sysadmin to create cgroup namespace. */ | |
a79a908f | 6299 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
fa5ff8a1 | 6300 | return ERR_PTR(-EPERM); |
a79a908f AK |
6301 | |
6302 | mutex_lock(&cgroup_mutex); | |
6303 | spin_lock_bh(&css_set_lock); | |
6304 | ||
6305 | cset = task_css_set(current); | |
6306 | get_css_set(cset); | |
6307 | ||
6308 | spin_unlock_bh(&css_set_lock); | |
6309 | mutex_unlock(&cgroup_mutex); | |
6310 | ||
a79a908f | 6311 | new_ns = alloc_cgroup_ns(); |
d2202557 | 6312 | if (IS_ERR(new_ns)) { |
fa5ff8a1 TH |
6313 | put_css_set(cset); |
6314 | return new_ns; | |
d2202557 | 6315 | } |
a79a908f AK |
6316 | |
6317 | new_ns->user_ns = get_user_ns(user_ns); | |
6318 | new_ns->root_cset = cset; | |
6319 | ||
6320 | return new_ns; | |
a79a908f AK |
6321 | } |
6322 | ||
6323 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) | |
6324 | { | |
6325 | return container_of(ns, struct cgroup_namespace, ns); | |
6326 | } | |
6327 | ||
a0530e08 | 6328 | static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
a79a908f | 6329 | { |
a0530e08 AK |
6330 | struct cgroup_namespace *cgroup_ns = to_cg_ns(ns); |
6331 | ||
6332 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) || | |
6333 | !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN)) | |
6334 | return -EPERM; | |
6335 | ||
6336 | /* Don't need to do anything if we are attaching to our own cgroupns. */ | |
6337 | if (cgroup_ns == nsproxy->cgroup_ns) | |
6338 | return 0; | |
6339 | ||
6340 | get_cgroup_ns(cgroup_ns); | |
6341 | put_cgroup_ns(nsproxy->cgroup_ns); | |
6342 | nsproxy->cgroup_ns = cgroup_ns; | |
6343 | ||
6344 | return 0; | |
a79a908f AK |
6345 | } |
6346 | ||
6347 | static struct ns_common *cgroupns_get(struct task_struct *task) | |
6348 | { | |
6349 | struct cgroup_namespace *ns = NULL; | |
6350 | struct nsproxy *nsproxy; | |
6351 | ||
6352 | task_lock(task); | |
6353 | nsproxy = task->nsproxy; | |
6354 | if (nsproxy) { | |
6355 | ns = nsproxy->cgroup_ns; | |
6356 | get_cgroup_ns(ns); | |
6357 | } | |
6358 | task_unlock(task); | |
6359 | ||
6360 | return ns ? &ns->ns : NULL; | |
6361 | } | |
6362 | ||
6363 | static void cgroupns_put(struct ns_common *ns) | |
6364 | { | |
6365 | put_cgroup_ns(to_cg_ns(ns)); | |
6366 | } | |
6367 | ||
6368 | const struct proc_ns_operations cgroupns_operations = { | |
6369 | .name = "cgroup", | |
6370 | .type = CLONE_NEWCGROUP, | |
6371 | .get = cgroupns_get, | |
6372 | .put = cgroupns_put, | |
6373 | .install = cgroupns_install, | |
6374 | }; | |
6375 | ||
6376 | static __init int cgroup_namespaces_init(void) | |
6377 | { | |
6378 | return 0; | |
6379 | } | |
6380 | subsys_initcall(cgroup_namespaces_init); | |
6381 | ||
fe693435 | 6382 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6383 | static struct cgroup_subsys_state * |
6384 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6385 | { |
6386 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6387 | ||
6388 | if (!css) | |
6389 | return ERR_PTR(-ENOMEM); | |
6390 | ||
6391 | return css; | |
6392 | } | |
6393 | ||
eb95419b | 6394 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6395 | { |
eb95419b | 6396 | kfree(css); |
fe693435 PM |
6397 | } |
6398 | ||
182446d0 TH |
6399 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6400 | struct cftype *cft) | |
fe693435 | 6401 | { |
182446d0 | 6402 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6403 | } |
6404 | ||
182446d0 TH |
6405 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6406 | struct cftype *cft) | |
fe693435 PM |
6407 | { |
6408 | return (u64)(unsigned long)current->cgroups; | |
6409 | } | |
6410 | ||
182446d0 | 6411 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6412 | struct cftype *cft) |
fe693435 PM |
6413 | { |
6414 | u64 count; | |
6415 | ||
6416 | rcu_read_lock(); | |
a8ad805c | 6417 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6418 | rcu_read_unlock(); |
6419 | return count; | |
6420 | } | |
6421 | ||
2da8ca82 | 6422 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6423 | { |
69d0206c | 6424 | struct cgrp_cset_link *link; |
5abb8855 | 6425 | struct css_set *cset; |
e61734c5 TH |
6426 | char *name_buf; |
6427 | ||
6428 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6429 | if (!name_buf) | |
6430 | return -ENOMEM; | |
7717f7ba | 6431 | |
f0d9a5f1 | 6432 | spin_lock_bh(&css_set_lock); |
7717f7ba | 6433 | rcu_read_lock(); |
5abb8855 | 6434 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6435 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6436 | struct cgroup *c = link->cgrp; |
7717f7ba | 6437 | |
a2dd4247 | 6438 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6439 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6440 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6441 | } |
6442 | rcu_read_unlock(); | |
f0d9a5f1 | 6443 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 6444 | kfree(name_buf); |
7717f7ba PM |
6445 | return 0; |
6446 | } | |
6447 | ||
6448 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6449 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6450 | { |
2da8ca82 | 6451 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6452 | struct cgrp_cset_link *link; |
7717f7ba | 6453 | |
f0d9a5f1 | 6454 | spin_lock_bh(&css_set_lock); |
182446d0 | 6455 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6456 | struct css_set *cset = link->cset; |
7717f7ba PM |
6457 | struct task_struct *task; |
6458 | int count = 0; | |
c7561128 | 6459 | |
5abb8855 | 6460 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6461 | |
5abb8855 | 6462 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6463 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6464 | goto overflow; | |
6465 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6466 | } | |
6467 | ||
6468 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6469 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6470 | goto overflow; | |
6471 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6472 | } |
c7561128 TH |
6473 | continue; |
6474 | overflow: | |
6475 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6476 | } |
f0d9a5f1 | 6477 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
6478 | return 0; |
6479 | } | |
6480 | ||
182446d0 | 6481 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6482 | { |
27bd4dbb | 6483 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6484 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6485 | } |
6486 | ||
6487 | static struct cftype debug_files[] = { | |
fe693435 PM |
6488 | { |
6489 | .name = "taskcount", | |
6490 | .read_u64 = debug_taskcount_read, | |
6491 | }, | |
6492 | ||
6493 | { | |
6494 | .name = "current_css_set", | |
6495 | .read_u64 = current_css_set_read, | |
6496 | }, | |
6497 | ||
6498 | { | |
6499 | .name = "current_css_set_refcount", | |
6500 | .read_u64 = current_css_set_refcount_read, | |
6501 | }, | |
6502 | ||
7717f7ba PM |
6503 | { |
6504 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6505 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6506 | }, |
6507 | ||
6508 | { | |
6509 | .name = "cgroup_css_links", | |
2da8ca82 | 6510 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6511 | }, |
6512 | ||
fe693435 PM |
6513 | { |
6514 | .name = "releasable", | |
6515 | .read_u64 = releasable_read, | |
6516 | }, | |
fe693435 | 6517 | |
4baf6e33 TH |
6518 | { } /* terminate */ |
6519 | }; | |
fe693435 | 6520 | |
073219e9 | 6521 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6522 | .css_alloc = debug_css_alloc, |
6523 | .css_free = debug_css_free, | |
5577964e | 6524 | .legacy_cftypes = debug_files, |
fe693435 PM |
6525 | }; |
6526 | #endif /* CONFIG_CGROUP_DEBUG */ |