Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
60063497 | 60 | #include <linux/atomic.h> |
e93ad19d | 61 | #include <linux/cpuset.h> |
a79a908f AK |
62 | #include <linux/proc_ns.h> |
63 | #include <linux/nsproxy.h> | |
1f3fe7eb | 64 | #include <linux/file.h> |
bd1060a1 | 65 | #include <net/sock.h> |
ddbcc7e8 | 66 | |
ed1777de TH |
67 | #define CREATE_TRACE_POINTS |
68 | #include <trace/events/cgroup.h> | |
69 | ||
b1a21367 TH |
70 | /* |
71 | * pidlists linger the following amount before being destroyed. The goal | |
72 | * is avoiding frequent destruction in the middle of consecutive read calls | |
73 | * Expiring in the middle is a performance problem not a correctness one. | |
74 | * 1 sec should be enough. | |
75 | */ | |
76 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
77 | ||
8d7e6fb0 TH |
78 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
79 | MAX_CFTYPE_NAME + 2) | |
80 | ||
e25e2cbb TH |
81 | /* |
82 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
83 | * hierarchy must be performed while holding it. | |
84 | * | |
f0d9a5f1 | 85 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 86 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 87 | * |
0e1d768f TH |
88 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
89 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 90 | */ |
2219449a TH |
91 | #ifdef CONFIG_PROVE_RCU |
92 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 93 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 94 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 95 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 96 | #else |
81a6a5cd | 97 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 98 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
99 | #endif |
100 | ||
6fa4918d | 101 | /* |
15a4c835 TH |
102 | * Protects cgroup_idr and css_idr so that IDs can be released without |
103 | * grabbing cgroup_mutex. | |
6fa4918d TH |
104 | */ |
105 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
106 | ||
34c06254 TH |
107 | /* |
108 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
109 | * against file removal/re-creation across css hiding. | |
110 | */ | |
111 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
112 | ||
69e943b7 TH |
113 | /* |
114 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
115 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
116 | */ | |
117 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 118 | |
1ed13287 TH |
119 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
120 | ||
8353da1f | 121 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
122 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
123 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 124 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 125 | |
e5fca243 TH |
126 | /* |
127 | * cgroup destruction makes heavy use of work items and there can be a lot | |
128 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
129 | * destruction work items don't end up filling up max_active of system_wq | |
130 | * which may lead to deadlock. | |
131 | */ | |
132 | static struct workqueue_struct *cgroup_destroy_wq; | |
133 | ||
b1a21367 TH |
134 | /* |
135 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
136 | * separate workqueue as flush domain. | |
137 | */ | |
138 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
139 | ||
3ed80a62 | 140 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 141 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 142 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
143 | #include <linux/cgroup_subsys.h> |
144 | }; | |
073219e9 TH |
145 | #undef SUBSYS |
146 | ||
147 | /* array of cgroup subsystem names */ | |
148 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
149 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
150 | #include <linux/cgroup_subsys.h> |
151 | }; | |
073219e9 | 152 | #undef SUBSYS |
ddbcc7e8 | 153 | |
49d1dc4b TH |
154 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
155 | #define SUBSYS(_x) \ | |
156 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
157 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
158 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
159 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
160 | #include <linux/cgroup_subsys.h> | |
161 | #undef SUBSYS | |
162 | ||
163 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
164 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
165 | #include <linux/cgroup_subsys.h> | |
166 | }; | |
167 | #undef SUBSYS | |
168 | ||
169 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
170 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
171 | #include <linux/cgroup_subsys.h> | |
172 | }; | |
173 | #undef SUBSYS | |
174 | ||
ddbcc7e8 | 175 | /* |
3dd06ffa | 176 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
177 | * unattached - it never has more than a single cgroup, and all tasks are |
178 | * part of that cgroup. | |
ddbcc7e8 | 179 | */ |
a2dd4247 | 180 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 181 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 182 | |
a2dd4247 TH |
183 | /* |
184 | * The default hierarchy always exists but is hidden until mounted for the | |
185 | * first time. This is for backward compatibility. | |
186 | */ | |
a7165264 | 187 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 188 | |
223ffb29 | 189 | /* Controllers blocked by the commandline in v1 */ |
6e5c8307 | 190 | static u16 cgroup_no_v1_mask; |
223ffb29 | 191 | |
5533e011 | 192 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 193 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 194 | |
f6d635ad TH |
195 | /* some controllers are implicitly enabled on the default hierarchy */ |
196 | static unsigned long cgrp_dfl_implicit_ss_mask; | |
197 | ||
ddbcc7e8 PM |
198 | /* The list of hierarchy roots */ |
199 | ||
9871bf95 TH |
200 | static LIST_HEAD(cgroup_roots); |
201 | static int cgroup_root_count; | |
ddbcc7e8 | 202 | |
3417ae1f | 203 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 204 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 205 | |
794611a1 | 206 | /* |
0cb51d71 TH |
207 | * Assign a monotonically increasing serial number to csses. It guarantees |
208 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
209 | * Also, as csses are always appended to the parent's ->children list, it | |
210 | * guarantees that sibling csses are always sorted in the ascending serial | |
211 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 212 | */ |
0cb51d71 | 213 | static u64 css_serial_nr_next = 1; |
794611a1 | 214 | |
cb4a3167 AS |
215 | /* |
216 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
217 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
218 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 219 | */ |
6e5c8307 TH |
220 | static u16 have_fork_callback __read_mostly; |
221 | static u16 have_exit_callback __read_mostly; | |
222 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 223 | |
a79a908f AK |
224 | /* cgroup namespace for init task */ |
225 | struct cgroup_namespace init_cgroup_ns = { | |
226 | .count = { .counter = 2, }, | |
227 | .user_ns = &init_user_ns, | |
228 | .ns.ops = &cgroupns_operations, | |
229 | .ns.inum = PROC_CGROUP_INIT_INO, | |
230 | .root_cset = &init_css_set, | |
231 | }; | |
232 | ||
7e47682e | 233 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 234 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 235 | |
67e9c74b | 236 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
237 | static struct cftype cgroup_dfl_base_files[]; |
238 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 239 | |
6e5c8307 | 240 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
945ba199 | 241 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
334c3679 TH |
242 | static int cgroup_apply_control(struct cgroup *cgrp); |
243 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 244 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 245 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
246 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
247 | struct cgroup_subsys *ss); | |
9d755d33 | 248 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 249 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
250 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
251 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 252 | bool is_add); |
42809dd4 | 253 | |
fc5ed1e9 TH |
254 | /** |
255 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
256 | * @ssid: subsys ID of interest | |
257 | * | |
258 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
259 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
260 | * is slower static_key_enabled() based test indexed by @ssid. | |
261 | */ | |
262 | static bool cgroup_ssid_enabled(int ssid) | |
263 | { | |
cfe02a8a AB |
264 | if (CGROUP_SUBSYS_COUNT == 0) |
265 | return false; | |
266 | ||
fc5ed1e9 TH |
267 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
268 | } | |
269 | ||
223ffb29 JW |
270 | static bool cgroup_ssid_no_v1(int ssid) |
271 | { | |
272 | return cgroup_no_v1_mask & (1 << ssid); | |
273 | } | |
274 | ||
9e10a130 TH |
275 | /** |
276 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
277 | * @cgrp: the cgroup of interest | |
278 | * | |
279 | * The default hierarchy is the v2 interface of cgroup and this function | |
280 | * can be used to test whether a cgroup is on the default hierarchy for | |
281 | * cases where a subsystem should behave differnetly depending on the | |
282 | * interface version. | |
283 | * | |
284 | * The set of behaviors which change on the default hierarchy are still | |
285 | * being determined and the mount option is prefixed with __DEVEL__. | |
286 | * | |
287 | * List of changed behaviors: | |
288 | * | |
289 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
290 | * and "name" are disallowed. | |
291 | * | |
292 | * - When mounting an existing superblock, mount options should match. | |
293 | * | |
294 | * - Remount is disallowed. | |
295 | * | |
296 | * - rename(2) is disallowed. | |
297 | * | |
298 | * - "tasks" is removed. Everything should be at process granularity. Use | |
299 | * "cgroup.procs" instead. | |
300 | * | |
301 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
302 | * recycled inbetween reads. | |
303 | * | |
304 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
305 | * notification mechanism will be implemented. | |
306 | * | |
307 | * - "cgroup.clone_children" is removed. | |
308 | * | |
309 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
310 | * and its descendants contain no task; otherwise, 1. The file also | |
311 | * generates kernfs notification which can be monitored through poll and | |
312 | * [di]notify when the value of the file changes. | |
313 | * | |
314 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
315 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
316 | * moved to an ancestor. | |
317 | * | |
318 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
319 | * masks of ancestors. | |
320 | * | |
321 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
322 | * is not created. | |
323 | * | |
324 | * - blkcg: blk-throttle becomes properly hierarchical. | |
325 | * | |
326 | * - debug: disallowed on the default hierarchy. | |
327 | */ | |
328 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
329 | { | |
330 | return cgrp->root == &cgrp_dfl_root; | |
331 | } | |
332 | ||
6fa4918d TH |
333 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
334 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
335 | gfp_t gfp_mask) | |
336 | { | |
337 | int ret; | |
338 | ||
339 | idr_preload(gfp_mask); | |
54504e97 | 340 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 341 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 342 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
343 | idr_preload_end(); |
344 | return ret; | |
345 | } | |
346 | ||
347 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
348 | { | |
349 | void *ret; | |
350 | ||
54504e97 | 351 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 352 | ret = idr_replace(idr, ptr, id); |
54504e97 | 353 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
354 | return ret; |
355 | } | |
356 | ||
357 | static void cgroup_idr_remove(struct idr *idr, int id) | |
358 | { | |
54504e97 | 359 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 360 | idr_remove(idr, id); |
54504e97 | 361 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
362 | } |
363 | ||
d51f39b0 TH |
364 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
365 | { | |
366 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
367 | ||
368 | if (parent_css) | |
369 | return container_of(parent_css, struct cgroup, self); | |
370 | return NULL; | |
371 | } | |
372 | ||
5531dc91 TH |
373 | /* subsystems visibly enabled on a cgroup */ |
374 | static u16 cgroup_control(struct cgroup *cgrp) | |
375 | { | |
376 | struct cgroup *parent = cgroup_parent(cgrp); | |
377 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
378 | ||
379 | if (parent) | |
380 | return parent->subtree_control; | |
381 | ||
382 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
383 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
384 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
385 | return root_ss_mask; |
386 | } | |
387 | ||
388 | /* subsystems enabled on a cgroup */ | |
389 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
390 | { | |
391 | struct cgroup *parent = cgroup_parent(cgrp); | |
392 | ||
393 | if (parent) | |
394 | return parent->subtree_ss_mask; | |
395 | ||
396 | return cgrp->root->subsys_mask; | |
397 | } | |
398 | ||
95109b62 TH |
399 | /** |
400 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
401 | * @cgrp: the cgroup of interest | |
9d800df1 | 402 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 403 | * |
ca8bdcaf TH |
404 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
405 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
406 | * the caller is responsible for pinning the returned css if it wants to | |
407 | * keep accessing it outside the said locks. This function may return | |
408 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
409 | */ |
410 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 411 | struct cgroup_subsys *ss) |
95109b62 | 412 | { |
ca8bdcaf | 413 | if (ss) |
aec25020 | 414 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 415 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 416 | else |
9d800df1 | 417 | return &cgrp->self; |
95109b62 | 418 | } |
42809dd4 | 419 | |
aec3dfcb TH |
420 | /** |
421 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
422 | * @cgrp: the cgroup of interest | |
9d800df1 | 423 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 424 | * |
d0f702e6 | 425 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
426 | * as the matching css of the nearest ancestor including self which has @ss |
427 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
428 | * function is guaranteed to return non-NULL css. | |
429 | */ | |
430 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
431 | struct cgroup_subsys *ss) | |
432 | { | |
433 | lockdep_assert_held(&cgroup_mutex); | |
434 | ||
435 | if (!ss) | |
9d800df1 | 436 | return &cgrp->self; |
aec3dfcb | 437 | |
eeecbd19 TH |
438 | /* |
439 | * This function is used while updating css associations and thus | |
5531dc91 | 440 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 441 | */ |
5531dc91 | 442 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 443 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
444 | if (!cgrp) |
445 | return NULL; | |
446 | } | |
aec3dfcb TH |
447 | |
448 | return cgroup_css(cgrp, ss); | |
95109b62 | 449 | } |
42809dd4 | 450 | |
eeecbd19 TH |
451 | /** |
452 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
453 | * @cgrp: the cgroup of interest | |
454 | * @ss: the subsystem of interest | |
455 | * | |
456 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
457 | * defined as the matching css of the nearest ancestor including self which | |
458 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
459 | * the root css is returned, so this function always returns a valid css. | |
460 | * The returned css must be put using css_put(). | |
461 | */ | |
462 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
463 | struct cgroup_subsys *ss) | |
464 | { | |
465 | struct cgroup_subsys_state *css; | |
466 | ||
467 | rcu_read_lock(); | |
468 | ||
469 | do { | |
470 | css = cgroup_css(cgrp, ss); | |
471 | ||
472 | if (css && css_tryget_online(css)) | |
473 | goto out_unlock; | |
474 | cgrp = cgroup_parent(cgrp); | |
475 | } while (cgrp); | |
476 | ||
477 | css = init_css_set.subsys[ss->id]; | |
478 | css_get(css); | |
479 | out_unlock: | |
480 | rcu_read_unlock(); | |
481 | return css; | |
482 | } | |
483 | ||
ddbcc7e8 | 484 | /* convenient tests for these bits */ |
54766d4a | 485 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 486 | { |
184faf32 | 487 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
488 | } |
489 | ||
052c3f3a TH |
490 | static void cgroup_get(struct cgroup *cgrp) |
491 | { | |
492 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
493 | css_get(&cgrp->self); | |
494 | } | |
495 | ||
496 | static bool cgroup_tryget(struct cgroup *cgrp) | |
497 | { | |
498 | return css_tryget(&cgrp->self); | |
499 | } | |
500 | ||
b4168640 | 501 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 502 | { |
2bd59d48 | 503 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 504 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
505 | |
506 | /* | |
507 | * This is open and unprotected implementation of cgroup_css(). | |
508 | * seq_css() is only called from a kernfs file operation which has | |
509 | * an active reference on the file. Because all the subsystem | |
510 | * files are drained before a css is disassociated with a cgroup, | |
511 | * the matching css from the cgroup's subsys table is guaranteed to | |
512 | * be and stay valid until the enclosing operation is complete. | |
513 | */ | |
514 | if (cft->ss) | |
515 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
516 | else | |
9d800df1 | 517 | return &cgrp->self; |
59f5296b | 518 | } |
b4168640 | 519 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 520 | |
e9685a03 | 521 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 522 | { |
bd89aabc | 523 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
524 | } |
525 | ||
1c6727af TH |
526 | /** |
527 | * for_each_css - iterate all css's of a cgroup | |
528 | * @css: the iteration cursor | |
529 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
530 | * @cgrp: the target cgroup to iterate css's of | |
531 | * | |
aec3dfcb | 532 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
533 | */ |
534 | #define for_each_css(css, ssid, cgrp) \ | |
535 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
536 | if (!((css) = rcu_dereference_check( \ | |
537 | (cgrp)->subsys[(ssid)], \ | |
538 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
539 | else | |
540 | ||
aec3dfcb TH |
541 | /** |
542 | * for_each_e_css - iterate all effective css's of a cgroup | |
543 | * @css: the iteration cursor | |
544 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
545 | * @cgrp: the target cgroup to iterate css's of | |
546 | * | |
547 | * Should be called under cgroup_[tree_]mutex. | |
548 | */ | |
549 | #define for_each_e_css(css, ssid, cgrp) \ | |
550 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
551 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
552 | ; \ | |
553 | else | |
554 | ||
30159ec7 | 555 | /** |
3ed80a62 | 556 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 557 | * @ss: the iteration cursor |
780cd8b3 | 558 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 559 | */ |
780cd8b3 | 560 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
561 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
562 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 563 | |
cb4a3167 | 564 | /** |
b4e0eeaf | 565 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
566 | * @ss: the iteration cursor |
567 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 568 | * @ss_mask: the bitmask |
cb4a3167 AS |
569 | * |
570 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 571 | * @ss_mask is set. |
cb4a3167 | 572 | */ |
b4e0eeaf TH |
573 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
574 | unsigned long __ss_mask = (ss_mask); \ | |
575 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 576 | (ssid) = 0; \ |
b4e0eeaf TH |
577 | break; \ |
578 | } \ | |
579 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
580 | (ss) = cgroup_subsys[ssid]; \ | |
581 | { | |
582 | ||
583 | #define while_each_subsys_mask() \ | |
584 | } \ | |
585 | } \ | |
586 | } while (false) | |
cb4a3167 | 587 | |
985ed670 TH |
588 | /* iterate across the hierarchies */ |
589 | #define for_each_root(root) \ | |
5549c497 | 590 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 591 | |
f8f22e53 TH |
592 | /* iterate over child cgrps, lock should be held throughout iteration */ |
593 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 594 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 595 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
596 | cgroup_is_dead(child); })) \ |
597 | ; \ | |
598 | else | |
7ae1bad9 | 599 | |
ce3f1d9d TH |
600 | /* walk live descendants in preorder */ |
601 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
602 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
603 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
604 | (dsct) = (d_css)->cgroup; \ | |
605 | cgroup_is_dead(dsct); })) \ | |
606 | ; \ | |
607 | else | |
608 | ||
609 | /* walk live descendants in postorder */ | |
610 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
611 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
612 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
613 | (dsct) = (d_css)->cgroup; \ | |
614 | cgroup_is_dead(dsct); })) \ | |
615 | ; \ | |
616 | else | |
617 | ||
81a6a5cd | 618 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 619 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 620 | |
69d0206c TH |
621 | /* |
622 | * A cgroup can be associated with multiple css_sets as different tasks may | |
623 | * belong to different cgroups on different hierarchies. In the other | |
624 | * direction, a css_set is naturally associated with multiple cgroups. | |
625 | * This M:N relationship is represented by the following link structure | |
626 | * which exists for each association and allows traversing the associations | |
627 | * from both sides. | |
628 | */ | |
629 | struct cgrp_cset_link { | |
630 | /* the cgroup and css_set this link associates */ | |
631 | struct cgroup *cgrp; | |
632 | struct css_set *cset; | |
633 | ||
634 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
635 | struct list_head cset_link; | |
636 | ||
637 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
638 | struct list_head cgrp_link; | |
817929ec PM |
639 | }; |
640 | ||
172a2c06 TH |
641 | /* |
642 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
643 | * hierarchies being mounted. It contains a pointer to the root state |
644 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
645 | * reference-counted, to improve performance when child cgroups | |
646 | * haven't been created. | |
647 | */ | |
5024ae29 | 648 | struct css_set init_css_set = { |
172a2c06 TH |
649 | .refcount = ATOMIC_INIT(1), |
650 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
651 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
652 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
653 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
654 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 655 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 656 | }; |
817929ec | 657 | |
172a2c06 | 658 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 659 | |
0de0942d TH |
660 | /** |
661 | * css_set_populated - does a css_set contain any tasks? | |
662 | * @cset: target css_set | |
663 | */ | |
664 | static bool css_set_populated(struct css_set *cset) | |
665 | { | |
f0d9a5f1 | 666 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
667 | |
668 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
669 | } | |
670 | ||
842b597e TH |
671 | /** |
672 | * cgroup_update_populated - updated populated count of a cgroup | |
673 | * @cgrp: the target cgroup | |
674 | * @populated: inc or dec populated count | |
675 | * | |
0de0942d TH |
676 | * One of the css_sets associated with @cgrp is either getting its first |
677 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
678 | * count is propagated towards root so that a given cgroup's populated_cnt | |
679 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
680 | * |
681 | * @cgrp's interface file "cgroup.populated" is zero if | |
682 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
683 | * changes from or to zero, userland is notified that the content of the | |
684 | * interface file has changed. This can be used to detect when @cgrp and | |
685 | * its descendants become populated or empty. | |
686 | */ | |
687 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
688 | { | |
f0d9a5f1 | 689 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
690 | |
691 | do { | |
692 | bool trigger; | |
693 | ||
694 | if (populated) | |
695 | trigger = !cgrp->populated_cnt++; | |
696 | else | |
697 | trigger = !--cgrp->populated_cnt; | |
698 | ||
699 | if (!trigger) | |
700 | break; | |
701 | ||
ad2ed2b3 | 702 | check_for_release(cgrp); |
6f60eade TH |
703 | cgroup_file_notify(&cgrp->events_file); |
704 | ||
d51f39b0 | 705 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
706 | } while (cgrp); |
707 | } | |
708 | ||
0de0942d TH |
709 | /** |
710 | * css_set_update_populated - update populated state of a css_set | |
711 | * @cset: target css_set | |
712 | * @populated: whether @cset is populated or depopulated | |
713 | * | |
714 | * @cset is either getting the first task or losing the last. Update the | |
715 | * ->populated_cnt of all associated cgroups accordingly. | |
716 | */ | |
717 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
718 | { | |
719 | struct cgrp_cset_link *link; | |
720 | ||
f0d9a5f1 | 721 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
722 | |
723 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
724 | cgroup_update_populated(link->cgrp, populated); | |
725 | } | |
726 | ||
f6d7d049 TH |
727 | /** |
728 | * css_set_move_task - move a task from one css_set to another | |
729 | * @task: task being moved | |
730 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
731 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
732 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
733 | * | |
734 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
735 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
736 | * instead of moved, @to_cset can be NULL. | |
737 | * | |
ed27b9f7 TH |
738 | * This function automatically handles populated_cnt updates and |
739 | * css_task_iter adjustments but the caller is responsible for managing | |
740 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
741 | */ |
742 | static void css_set_move_task(struct task_struct *task, | |
743 | struct css_set *from_cset, struct css_set *to_cset, | |
744 | bool use_mg_tasks) | |
745 | { | |
f0d9a5f1 | 746 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 747 | |
20b454a6 TH |
748 | if (to_cset && !css_set_populated(to_cset)) |
749 | css_set_update_populated(to_cset, true); | |
750 | ||
f6d7d049 | 751 | if (from_cset) { |
ed27b9f7 TH |
752 | struct css_task_iter *it, *pos; |
753 | ||
f6d7d049 | 754 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
755 | |
756 | /* | |
757 | * @task is leaving, advance task iterators which are | |
758 | * pointing to it so that they can resume at the next | |
759 | * position. Advancing an iterator might remove it from | |
760 | * the list, use safe walk. See css_task_iter_advance*() | |
761 | * for details. | |
762 | */ | |
763 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
764 | iters_node) | |
765 | if (it->task_pos == &task->cg_list) | |
766 | css_task_iter_advance(it); | |
767 | ||
f6d7d049 TH |
768 | list_del_init(&task->cg_list); |
769 | if (!css_set_populated(from_cset)) | |
770 | css_set_update_populated(from_cset, false); | |
771 | } else { | |
772 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
773 | } | |
774 | ||
775 | if (to_cset) { | |
776 | /* | |
777 | * We are synchronized through cgroup_threadgroup_rwsem | |
778 | * against PF_EXITING setting such that we can't race | |
779 | * against cgroup_exit() changing the css_set to | |
780 | * init_css_set and dropping the old one. | |
781 | */ | |
782 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
783 | ||
f6d7d049 TH |
784 | rcu_assign_pointer(task->cgroups, to_cset); |
785 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
786 | &to_cset->tasks); | |
787 | } | |
788 | } | |
789 | ||
7717f7ba PM |
790 | /* |
791 | * hash table for cgroup groups. This improves the performance to find | |
792 | * an existing css_set. This hash doesn't (currently) take into | |
793 | * account cgroups in empty hierarchies. | |
794 | */ | |
472b1053 | 795 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 796 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 797 | |
0ac801fe | 798 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 799 | { |
0ac801fe | 800 | unsigned long key = 0UL; |
30159ec7 TH |
801 | struct cgroup_subsys *ss; |
802 | int i; | |
472b1053 | 803 | |
30159ec7 | 804 | for_each_subsys(ss, i) |
0ac801fe LZ |
805 | key += (unsigned long)css[i]; |
806 | key = (key >> 16) ^ key; | |
472b1053 | 807 | |
0ac801fe | 808 | return key; |
472b1053 LZ |
809 | } |
810 | ||
a25eb52e | 811 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 812 | { |
69d0206c | 813 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
814 | struct cgroup_subsys *ss; |
815 | int ssid; | |
5abb8855 | 816 | |
f0d9a5f1 | 817 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
818 | |
819 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 820 | return; |
81a6a5cd | 821 | |
53254f90 TH |
822 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
823 | for_each_subsys(ss, ssid) { | |
2d8f243a | 824 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
825 | css_put(cset->subsys[ssid]); |
826 | } | |
5abb8855 | 827 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
828 | css_set_count--; |
829 | ||
69d0206c | 830 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
831 | list_del(&link->cset_link); |
832 | list_del(&link->cgrp_link); | |
2ceb231b TH |
833 | if (cgroup_parent(link->cgrp)) |
834 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 835 | kfree(link); |
81a6a5cd | 836 | } |
2c6ab6d2 | 837 | |
5abb8855 | 838 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
839 | } |
840 | ||
a25eb52e | 841 | static void put_css_set(struct css_set *cset) |
89c5509b | 842 | { |
82d6489d DBO |
843 | unsigned long flags; |
844 | ||
89c5509b TH |
845 | /* |
846 | * Ensure that the refcount doesn't hit zero while any readers | |
847 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
848 | * rwlock | |
849 | */ | |
850 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
851 | return; | |
852 | ||
82d6489d | 853 | spin_lock_irqsave(&css_set_lock, flags); |
a25eb52e | 854 | put_css_set_locked(cset); |
82d6489d | 855 | spin_unlock_irqrestore(&css_set_lock, flags); |
89c5509b TH |
856 | } |
857 | ||
817929ec PM |
858 | /* |
859 | * refcounted get/put for css_set objects | |
860 | */ | |
5abb8855 | 861 | static inline void get_css_set(struct css_set *cset) |
817929ec | 862 | { |
5abb8855 | 863 | atomic_inc(&cset->refcount); |
817929ec PM |
864 | } |
865 | ||
b326f9d0 | 866 | /** |
7717f7ba | 867 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
868 | * @cset: candidate css_set being tested |
869 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
870 | * @new_cgrp: cgroup that's being entered by the task |
871 | * @template: desired set of css pointers in css_set (pre-calculated) | |
872 | * | |
6f4b7e63 | 873 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
874 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
875 | */ | |
5abb8855 TH |
876 | static bool compare_css_sets(struct css_set *cset, |
877 | struct css_set *old_cset, | |
7717f7ba PM |
878 | struct cgroup *new_cgrp, |
879 | struct cgroup_subsys_state *template[]) | |
880 | { | |
881 | struct list_head *l1, *l2; | |
882 | ||
aec3dfcb TH |
883 | /* |
884 | * On the default hierarchy, there can be csets which are | |
885 | * associated with the same set of cgroups but different csses. | |
886 | * Let's first ensure that csses match. | |
887 | */ | |
888 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 889 | return false; |
7717f7ba PM |
890 | |
891 | /* | |
892 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
893 | * different cgroups in hierarchies. As different cgroups may |
894 | * share the same effective css, this comparison is always | |
895 | * necessary. | |
7717f7ba | 896 | */ |
69d0206c TH |
897 | l1 = &cset->cgrp_links; |
898 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 899 | while (1) { |
69d0206c | 900 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 901 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
902 | |
903 | l1 = l1->next; | |
904 | l2 = l2->next; | |
905 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
906 | if (l1 == &cset->cgrp_links) { |
907 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
908 | break; |
909 | } else { | |
69d0206c | 910 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
911 | } |
912 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
913 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
914 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
915 | cgrp1 = link1->cgrp; | |
916 | cgrp2 = link2->cgrp; | |
7717f7ba | 917 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 918 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
919 | |
920 | /* | |
921 | * If this hierarchy is the hierarchy of the cgroup | |
922 | * that's changing, then we need to check that this | |
923 | * css_set points to the new cgroup; if it's any other | |
924 | * hierarchy, then this css_set should point to the | |
925 | * same cgroup as the old css_set. | |
926 | */ | |
5abb8855 TH |
927 | if (cgrp1->root == new_cgrp->root) { |
928 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
929 | return false; |
930 | } else { | |
5abb8855 | 931 | if (cgrp1 != cgrp2) |
7717f7ba PM |
932 | return false; |
933 | } | |
934 | } | |
935 | return true; | |
936 | } | |
937 | ||
b326f9d0 TH |
938 | /** |
939 | * find_existing_css_set - init css array and find the matching css_set | |
940 | * @old_cset: the css_set that we're using before the cgroup transition | |
941 | * @cgrp: the cgroup that we're moving into | |
942 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 943 | */ |
5abb8855 TH |
944 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
945 | struct cgroup *cgrp, | |
946 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 947 | { |
3dd06ffa | 948 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 949 | struct cgroup_subsys *ss; |
5abb8855 | 950 | struct css_set *cset; |
0ac801fe | 951 | unsigned long key; |
b326f9d0 | 952 | int i; |
817929ec | 953 | |
aae8aab4 BB |
954 | /* |
955 | * Build the set of subsystem state objects that we want to see in the | |
956 | * new css_set. while subsystems can change globally, the entries here | |
957 | * won't change, so no need for locking. | |
958 | */ | |
30159ec7 | 959 | for_each_subsys(ss, i) { |
f392e51c | 960 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
961 | /* |
962 | * @ss is in this hierarchy, so we want the | |
963 | * effective css from @cgrp. | |
964 | */ | |
965 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 966 | } else { |
aec3dfcb TH |
967 | /* |
968 | * @ss is not in this hierarchy, so we don't want | |
969 | * to change the css. | |
970 | */ | |
5abb8855 | 971 | template[i] = old_cset->subsys[i]; |
817929ec PM |
972 | } |
973 | } | |
974 | ||
0ac801fe | 975 | key = css_set_hash(template); |
5abb8855 TH |
976 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
977 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
978 | continue; |
979 | ||
980 | /* This css_set matches what we need */ | |
5abb8855 | 981 | return cset; |
472b1053 | 982 | } |
817929ec PM |
983 | |
984 | /* No existing cgroup group matched */ | |
985 | return NULL; | |
986 | } | |
987 | ||
69d0206c | 988 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 989 | { |
69d0206c | 990 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 991 | |
69d0206c TH |
992 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
993 | list_del(&link->cset_link); | |
36553434 LZ |
994 | kfree(link); |
995 | } | |
996 | } | |
997 | ||
69d0206c TH |
998 | /** |
999 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
1000 | * @count: the number of links to allocate | |
1001 | * @tmp_links: list_head the allocated links are put on | |
1002 | * | |
1003 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1004 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1005 | */ |
69d0206c | 1006 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1007 | { |
69d0206c | 1008 | struct cgrp_cset_link *link; |
817929ec | 1009 | int i; |
69d0206c TH |
1010 | |
1011 | INIT_LIST_HEAD(tmp_links); | |
1012 | ||
817929ec | 1013 | for (i = 0; i < count; i++) { |
f4f4be2b | 1014 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1015 | if (!link) { |
69d0206c | 1016 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1017 | return -ENOMEM; |
1018 | } | |
69d0206c | 1019 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1020 | } |
1021 | return 0; | |
1022 | } | |
1023 | ||
c12f65d4 LZ |
1024 | /** |
1025 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1026 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1027 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1028 | * @cgrp: the destination cgroup |
1029 | */ | |
69d0206c TH |
1030 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1031 | struct cgroup *cgrp) | |
c12f65d4 | 1032 | { |
69d0206c | 1033 | struct cgrp_cset_link *link; |
c12f65d4 | 1034 | |
69d0206c | 1035 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1036 | |
1037 | if (cgroup_on_dfl(cgrp)) | |
1038 | cset->dfl_cgrp = cgrp; | |
1039 | ||
69d0206c TH |
1040 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1041 | link->cset = cset; | |
7717f7ba | 1042 | link->cgrp = cgrp; |
842b597e | 1043 | |
7717f7ba | 1044 | /* |
389b9c1b TH |
1045 | * Always add links to the tail of the lists so that the lists are |
1046 | * in choronological order. | |
7717f7ba | 1047 | */ |
389b9c1b | 1048 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1049 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1050 | |
1051 | if (cgroup_parent(cgrp)) | |
1052 | cgroup_get(cgrp); | |
c12f65d4 LZ |
1053 | } |
1054 | ||
b326f9d0 TH |
1055 | /** |
1056 | * find_css_set - return a new css_set with one cgroup updated | |
1057 | * @old_cset: the baseline css_set | |
1058 | * @cgrp: the cgroup to be updated | |
1059 | * | |
1060 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1061 | * substituted into the appropriate hierarchy. | |
817929ec | 1062 | */ |
5abb8855 TH |
1063 | static struct css_set *find_css_set(struct css_set *old_cset, |
1064 | struct cgroup *cgrp) | |
817929ec | 1065 | { |
b326f9d0 | 1066 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1067 | struct css_set *cset; |
69d0206c TH |
1068 | struct list_head tmp_links; |
1069 | struct cgrp_cset_link *link; | |
2d8f243a | 1070 | struct cgroup_subsys *ss; |
0ac801fe | 1071 | unsigned long key; |
2d8f243a | 1072 | int ssid; |
472b1053 | 1073 | |
b326f9d0 TH |
1074 | lockdep_assert_held(&cgroup_mutex); |
1075 | ||
817929ec PM |
1076 | /* First see if we already have a cgroup group that matches |
1077 | * the desired set */ | |
82d6489d | 1078 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1079 | cset = find_existing_css_set(old_cset, cgrp, template); |
1080 | if (cset) | |
1081 | get_css_set(cset); | |
82d6489d | 1082 | spin_unlock_irq(&css_set_lock); |
817929ec | 1083 | |
5abb8855 TH |
1084 | if (cset) |
1085 | return cset; | |
817929ec | 1086 | |
f4f4be2b | 1087 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1088 | if (!cset) |
817929ec PM |
1089 | return NULL; |
1090 | ||
69d0206c | 1091 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1092 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1093 | kfree(cset); |
817929ec PM |
1094 | return NULL; |
1095 | } | |
1096 | ||
5abb8855 | 1097 | atomic_set(&cset->refcount, 1); |
69d0206c | 1098 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1099 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1100 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1101 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1102 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1103 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1104 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1105 | |
1106 | /* Copy the set of subsystem state objects generated in | |
1107 | * find_existing_css_set() */ | |
5abb8855 | 1108 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1109 | |
82d6489d | 1110 | spin_lock_irq(&css_set_lock); |
817929ec | 1111 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1112 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1113 | struct cgroup *c = link->cgrp; |
69d0206c | 1114 | |
7717f7ba PM |
1115 | if (c->root == cgrp->root) |
1116 | c = cgrp; | |
69d0206c | 1117 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1118 | } |
817929ec | 1119 | |
69d0206c | 1120 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1121 | |
817929ec | 1122 | css_set_count++; |
472b1053 | 1123 | |
2d8f243a | 1124 | /* Add @cset to the hash table */ |
5abb8855 TH |
1125 | key = css_set_hash(cset->subsys); |
1126 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1127 | |
53254f90 TH |
1128 | for_each_subsys(ss, ssid) { |
1129 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1130 | ||
2d8f243a | 1131 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1132 | &css->cgroup->e_csets[ssid]); |
1133 | css_get(css); | |
1134 | } | |
2d8f243a | 1135 | |
82d6489d | 1136 | spin_unlock_irq(&css_set_lock); |
817929ec | 1137 | |
5abb8855 | 1138 | return cset; |
b4f48b63 PM |
1139 | } |
1140 | ||
3dd06ffa | 1141 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1142 | { |
3dd06ffa | 1143 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1144 | |
3dd06ffa | 1145 | return root_cgrp->root; |
2bd59d48 TH |
1146 | } |
1147 | ||
3dd06ffa | 1148 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1149 | { |
1150 | int id; | |
1151 | ||
1152 | lockdep_assert_held(&cgroup_mutex); | |
1153 | ||
985ed670 | 1154 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1155 | if (id < 0) |
1156 | return id; | |
1157 | ||
1158 | root->hierarchy_id = id; | |
1159 | return 0; | |
1160 | } | |
1161 | ||
3dd06ffa | 1162 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1163 | { |
1164 | lockdep_assert_held(&cgroup_mutex); | |
1165 | ||
8c8a5502 | 1166 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1167 | } |
1168 | ||
3dd06ffa | 1169 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1170 | { |
1171 | if (root) { | |
f2e85d57 TH |
1172 | idr_destroy(&root->cgroup_idr); |
1173 | kfree(root); | |
1174 | } | |
1175 | } | |
1176 | ||
3dd06ffa | 1177 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1178 | { |
3dd06ffa | 1179 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1180 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1181 | |
ed1777de TH |
1182 | trace_cgroup_destroy_root(root); |
1183 | ||
334c3679 | 1184 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1185 | |
776f02fa | 1186 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1187 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1188 | |
f2e85d57 | 1189 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1190 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1191 | |
7717f7ba | 1192 | /* |
f2e85d57 TH |
1193 | * Release all the links from cset_links to this hierarchy's |
1194 | * root cgroup | |
7717f7ba | 1195 | */ |
82d6489d | 1196 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1197 | |
1198 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1199 | list_del(&link->cset_link); | |
1200 | list_del(&link->cgrp_link); | |
1201 | kfree(link); | |
1202 | } | |
f0d9a5f1 | 1203 | |
82d6489d | 1204 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1205 | |
1206 | if (!list_empty(&root->root_list)) { | |
1207 | list_del(&root->root_list); | |
1208 | cgroup_root_count--; | |
1209 | } | |
1210 | ||
1211 | cgroup_exit_root_id(root); | |
1212 | ||
1213 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1214 | |
2bd59d48 | 1215 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1216 | cgroup_free_root(root); |
1217 | } | |
1218 | ||
4f41fc59 SH |
1219 | /* |
1220 | * look up cgroup associated with current task's cgroup namespace on the | |
1221 | * specified hierarchy | |
1222 | */ | |
1223 | static struct cgroup * | |
1224 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1225 | { | |
1226 | struct cgroup *res = NULL; | |
1227 | struct css_set *cset; | |
1228 | ||
1229 | lockdep_assert_held(&css_set_lock); | |
1230 | ||
1231 | rcu_read_lock(); | |
1232 | ||
1233 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1234 | if (cset == &init_css_set) { | |
1235 | res = &root->cgrp; | |
1236 | } else { | |
1237 | struct cgrp_cset_link *link; | |
1238 | ||
1239 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1240 | struct cgroup *c = link->cgrp; | |
1241 | ||
1242 | if (c->root == root) { | |
1243 | res = c; | |
1244 | break; | |
1245 | } | |
1246 | } | |
1247 | } | |
1248 | rcu_read_unlock(); | |
1249 | ||
1250 | BUG_ON(!res); | |
1251 | return res; | |
1252 | } | |
1253 | ||
ceb6a081 TH |
1254 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1255 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1256 | struct cgroup_root *root) |
7717f7ba | 1257 | { |
7717f7ba PM |
1258 | struct cgroup *res = NULL; |
1259 | ||
96d365e0 | 1260 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1261 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1262 | |
5abb8855 | 1263 | if (cset == &init_css_set) { |
3dd06ffa | 1264 | res = &root->cgrp; |
7717f7ba | 1265 | } else { |
69d0206c TH |
1266 | struct cgrp_cset_link *link; |
1267 | ||
1268 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1269 | struct cgroup *c = link->cgrp; |
69d0206c | 1270 | |
7717f7ba PM |
1271 | if (c->root == root) { |
1272 | res = c; | |
1273 | break; | |
1274 | } | |
1275 | } | |
1276 | } | |
96d365e0 | 1277 | |
7717f7ba PM |
1278 | BUG_ON(!res); |
1279 | return res; | |
1280 | } | |
1281 | ||
ddbcc7e8 | 1282 | /* |
ceb6a081 | 1283 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1284 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1285 | */ |
1286 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1287 | struct cgroup_root *root) |
ceb6a081 TH |
1288 | { |
1289 | /* | |
1290 | * No need to lock the task - since we hold cgroup_mutex the | |
1291 | * task can't change groups, so the only thing that can happen | |
1292 | * is that it exits and its css is set back to init_css_set. | |
1293 | */ | |
1294 | return cset_cgroup_from_root(task_css_set(task), root); | |
1295 | } | |
1296 | ||
ddbcc7e8 | 1297 | /* |
ddbcc7e8 PM |
1298 | * A task must hold cgroup_mutex to modify cgroups. |
1299 | * | |
1300 | * Any task can increment and decrement the count field without lock. | |
1301 | * So in general, code holding cgroup_mutex can't rely on the count | |
1302 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1303 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1304 | * means that no tasks are currently attached, therefore there is no |
1305 | * way a task attached to that cgroup can fork (the other way to | |
1306 | * increment the count). So code holding cgroup_mutex can safely | |
1307 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1308 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1309 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1310 | * needs that mutex. | |
1311 | * | |
ddbcc7e8 PM |
1312 | * A cgroup can only be deleted if both its 'count' of using tasks |
1313 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1314 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1315 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1316 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1317 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1318 | * |
1319 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1320 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1321 | */ |
1322 | ||
2bd59d48 | 1323 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1324 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1325 | |
8d7e6fb0 TH |
1326 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1327 | char *buf) | |
ddbcc7e8 | 1328 | { |
3e1d2eed TH |
1329 | struct cgroup_subsys *ss = cft->ss; |
1330 | ||
8d7e6fb0 TH |
1331 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1332 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1333 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1334 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1335 | cft->name); | |
8d7e6fb0 TH |
1336 | else |
1337 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1338 | return buf; | |
ddbcc7e8 PM |
1339 | } |
1340 | ||
f2e85d57 TH |
1341 | /** |
1342 | * cgroup_file_mode - deduce file mode of a control file | |
1343 | * @cft: the control file in question | |
1344 | * | |
7dbdb199 | 1345 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1346 | */ |
1347 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1348 | { |
f2e85d57 | 1349 | umode_t mode = 0; |
65dff759 | 1350 | |
f2e85d57 TH |
1351 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1352 | mode |= S_IRUGO; | |
1353 | ||
7dbdb199 TH |
1354 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1355 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1356 | mode |= S_IWUGO; | |
1357 | else | |
1358 | mode |= S_IWUSR; | |
1359 | } | |
f2e85d57 TH |
1360 | |
1361 | return mode; | |
65dff759 LZ |
1362 | } |
1363 | ||
af0ba678 | 1364 | /** |
8699b776 | 1365 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1366 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1367 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1368 | * |
1369 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1370 | * enabled together through its ->depends_on mask. In such cases, more | |
1371 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1372 | * | |
0f060deb | 1373 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1374 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1375 | */ |
5ced2518 | 1376 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1377 | { |
6e5c8307 | 1378 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1379 | struct cgroup_subsys *ss; |
1380 | int ssid; | |
1381 | ||
1382 | lockdep_assert_held(&cgroup_mutex); | |
1383 | ||
f6d635ad TH |
1384 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1385 | ||
af0ba678 | 1386 | while (true) { |
6e5c8307 | 1387 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1388 | |
b4e0eeaf | 1389 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1390 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1391 | } while_each_subsys_mask(); |
af0ba678 TH |
1392 | |
1393 | /* | |
1394 | * Mask out subsystems which aren't available. This can | |
1395 | * happen only if some depended-upon subsystems were bound | |
1396 | * to non-default hierarchies. | |
1397 | */ | |
5ced2518 | 1398 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1399 | |
1400 | if (new_ss_mask == cur_ss_mask) | |
1401 | break; | |
1402 | cur_ss_mask = new_ss_mask; | |
1403 | } | |
1404 | ||
0f060deb TH |
1405 | return cur_ss_mask; |
1406 | } | |
1407 | ||
a9746d8d TH |
1408 | /** |
1409 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1410 | * @kn: the kernfs_node being serviced | |
1411 | * | |
1412 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1413 | * the method finishes if locking succeeded. Note that once this function | |
1414 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1415 | * inaccessible any time. If the caller intends to continue to access the | |
1416 | * cgroup, it should pin it before invoking this function. | |
1417 | */ | |
1418 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1419 | { |
a9746d8d TH |
1420 | struct cgroup *cgrp; |
1421 | ||
1422 | if (kernfs_type(kn) == KERNFS_DIR) | |
1423 | cgrp = kn->priv; | |
1424 | else | |
1425 | cgrp = kn->parent->priv; | |
1426 | ||
1427 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1428 | |
1429 | kernfs_unbreak_active_protection(kn); | |
1430 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1431 | } |
1432 | ||
a9746d8d TH |
1433 | /** |
1434 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1435 | * @kn: the kernfs_node being serviced | |
945ba199 | 1436 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1437 | * |
1438 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1439 | * @kn. It breaks the active protection, performs cgroup locking and | |
1440 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1441 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1442 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1443 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1444 | * |
1445 | * Any cgroup kernfs method implementation which requires locking the | |
1446 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1447 | * locking under kernfs active protection and allows all kernfs operations | |
1448 | * including self-removal. | |
1449 | */ | |
945ba199 TH |
1450 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1451 | bool drain_offline) | |
05ef1d7c | 1452 | { |
a9746d8d TH |
1453 | struct cgroup *cgrp; |
1454 | ||
1455 | if (kernfs_type(kn) == KERNFS_DIR) | |
1456 | cgrp = kn->priv; | |
1457 | else | |
1458 | cgrp = kn->parent->priv; | |
05ef1d7c | 1459 | |
2739d3cc | 1460 | /* |
01f6474c | 1461 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1462 | * active_ref. cgroup liveliness check alone provides enough |
1463 | * protection against removal. Ensure @cgrp stays accessible and | |
1464 | * break the active_ref protection. | |
2739d3cc | 1465 | */ |
aa32362f LZ |
1466 | if (!cgroup_tryget(cgrp)) |
1467 | return NULL; | |
a9746d8d TH |
1468 | kernfs_break_active_protection(kn); |
1469 | ||
945ba199 TH |
1470 | if (drain_offline) |
1471 | cgroup_lock_and_drain_offline(cgrp); | |
1472 | else | |
1473 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1474 | |
a9746d8d TH |
1475 | if (!cgroup_is_dead(cgrp)) |
1476 | return cgrp; | |
1477 | ||
1478 | cgroup_kn_unlock(kn); | |
1479 | return NULL; | |
ddbcc7e8 | 1480 | } |
05ef1d7c | 1481 | |
2739d3cc | 1482 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1483 | { |
2bd59d48 | 1484 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1485 | |
01f6474c | 1486 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1487 | |
1488 | if (cft->file_offset) { | |
1489 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1490 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1491 | ||
1492 | spin_lock_irq(&cgroup_file_kn_lock); | |
1493 | cfile->kn = NULL; | |
1494 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1495 | } | |
1496 | ||
2bd59d48 | 1497 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1498 | } |
1499 | ||
13af07df | 1500 | /** |
4df8dc90 TH |
1501 | * css_clear_dir - remove subsys files in a cgroup directory |
1502 | * @css: taget css | |
13af07df | 1503 | */ |
334c3679 | 1504 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1505 | { |
334c3679 | 1506 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1507 | struct cftype *cfts; |
05ef1d7c | 1508 | |
88cb04b9 TH |
1509 | if (!(css->flags & CSS_VISIBLE)) |
1510 | return; | |
1511 | ||
1512 | css->flags &= ~CSS_VISIBLE; | |
1513 | ||
4df8dc90 TH |
1514 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1515 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1516 | } |
1517 | ||
ccdca218 | 1518 | /** |
4df8dc90 TH |
1519 | * css_populate_dir - create subsys files in a cgroup directory |
1520 | * @css: target css | |
ccdca218 TH |
1521 | * |
1522 | * On failure, no file is added. | |
1523 | */ | |
334c3679 | 1524 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1525 | { |
334c3679 | 1526 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1527 | struct cftype *cfts, *failed_cfts; |
1528 | int ret; | |
ccdca218 | 1529 | |
03970d3c | 1530 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1531 | return 0; |
1532 | ||
4df8dc90 TH |
1533 | if (!css->ss) { |
1534 | if (cgroup_on_dfl(cgrp)) | |
1535 | cfts = cgroup_dfl_base_files; | |
1536 | else | |
1537 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1538 | |
4df8dc90 TH |
1539 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1540 | } | |
ccdca218 | 1541 | |
4df8dc90 TH |
1542 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1543 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1544 | if (ret < 0) { | |
1545 | failed_cfts = cfts; | |
1546 | goto err; | |
ccdca218 TH |
1547 | } |
1548 | } | |
88cb04b9 TH |
1549 | |
1550 | css->flags |= CSS_VISIBLE; | |
1551 | ||
ccdca218 TH |
1552 | return 0; |
1553 | err: | |
4df8dc90 TH |
1554 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1555 | if (cfts == failed_cfts) | |
1556 | break; | |
1557 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1558 | } | |
ccdca218 TH |
1559 | return ret; |
1560 | } | |
1561 | ||
6e5c8307 | 1562 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1563 | { |
1ada4838 | 1564 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1565 | struct cgroup_subsys *ss; |
2d8f243a | 1566 | int ssid, i, ret; |
ddbcc7e8 | 1567 | |
ace2bee8 | 1568 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1569 | |
b4e0eeaf | 1570 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1571 | /* |
1572 | * If @ss has non-root csses attached to it, can't move. | |
1573 | * If @ss is an implicit controller, it is exempt from this | |
1574 | * rule and can be stolen. | |
1575 | */ | |
1576 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1577 | !ss->implicit_on_dfl) | |
3ed80a62 | 1578 | return -EBUSY; |
1d5be6b2 | 1579 | |
5df36032 | 1580 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1581 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1582 | return -EBUSY; |
b4e0eeaf | 1583 | } while_each_subsys_mask(); |
ddbcc7e8 | 1584 | |
b4e0eeaf | 1585 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1586 | struct cgroup_root *src_root = ss->root; |
1587 | struct cgroup *scgrp = &src_root->cgrp; | |
1588 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1589 | struct css_set *cset; |
a8a648c4 | 1590 | |
1ada4838 | 1591 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1592 | |
334c3679 TH |
1593 | /* disable from the source */ |
1594 | src_root->subsys_mask &= ~(1 << ssid); | |
1595 | WARN_ON(cgroup_apply_control(scgrp)); | |
1596 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1597 | |
334c3679 | 1598 | /* rebind */ |
1ada4838 TH |
1599 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1600 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1601 | ss->root = dst_root; |
1ada4838 | 1602 | css->cgroup = dcgrp; |
73e80ed8 | 1603 | |
82d6489d | 1604 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1605 | hash_for_each(css_set_table, i, cset, hlist) |
1606 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1607 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1608 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1609 | |
bd53d617 | 1610 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1611 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1612 | if (dst_root == &cgrp_dfl_root) { |
1613 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1614 | } else { | |
1ada4838 | 1615 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1616 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1617 | } |
a8a648c4 | 1618 | |
334c3679 TH |
1619 | ret = cgroup_apply_control(dcgrp); |
1620 | if (ret) | |
1621 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1622 | ss->name, ret); | |
1623 | ||
5df36032 TH |
1624 | if (ss->bind) |
1625 | ss->bind(css); | |
b4e0eeaf | 1626 | } while_each_subsys_mask(); |
ddbcc7e8 | 1627 | |
1ada4838 | 1628 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1629 | return 0; |
1630 | } | |
1631 | ||
4f41fc59 SH |
1632 | static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1633 | struct kernfs_root *kf_root) | |
1634 | { | |
09be4c82 | 1635 | int len = 0; |
4f41fc59 SH |
1636 | char *buf = NULL; |
1637 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1638 | struct cgroup *ns_cgroup; | |
1639 | ||
1640 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1641 | if (!buf) | |
1642 | return -ENOMEM; | |
1643 | ||
82d6489d | 1644 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1645 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1646 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1647 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1648 | |
1649 | if (len >= PATH_MAX) | |
1650 | len = -ERANGE; | |
1651 | else if (len > 0) { | |
1652 | seq_escape(sf, buf, " \t\n\\"); | |
1653 | len = 0; | |
1654 | } | |
1655 | kfree(buf); | |
1656 | return len; | |
1657 | } | |
1658 | ||
2bd59d48 TH |
1659 | static int cgroup_show_options(struct seq_file *seq, |
1660 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1661 | { |
3dd06ffa | 1662 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1663 | struct cgroup_subsys *ss; |
b85d2040 | 1664 | int ssid; |
ddbcc7e8 | 1665 | |
d98817d4 TH |
1666 | if (root != &cgrp_dfl_root) |
1667 | for_each_subsys(ss, ssid) | |
1668 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1669 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1670 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1671 | seq_puts(seq, ",noprefix"); |
93438629 | 1672 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1673 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1674 | |
1675 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1676 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1677 | seq_show_option(seq, "release_agent", |
1678 | root->release_agent_path); | |
69e943b7 TH |
1679 | spin_unlock(&release_agent_path_lock); |
1680 | ||
3dd06ffa | 1681 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1682 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1683 | if (strlen(root->name)) |
a068acf2 | 1684 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1685 | return 0; |
1686 | } | |
1687 | ||
1688 | struct cgroup_sb_opts { | |
6e5c8307 | 1689 | u16 subsys_mask; |
69dfa00c | 1690 | unsigned int flags; |
81a6a5cd | 1691 | char *release_agent; |
2260e7fc | 1692 | bool cpuset_clone_children; |
c6d57f33 | 1693 | char *name; |
2c6ab6d2 PM |
1694 | /* User explicitly requested empty subsystem */ |
1695 | bool none; | |
ddbcc7e8 PM |
1696 | }; |
1697 | ||
cf5d5941 | 1698 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1699 | { |
32a8cf23 DL |
1700 | char *token, *o = data; |
1701 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1702 | u16 mask = U16_MAX; |
30159ec7 | 1703 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1704 | int nr_opts = 0; |
30159ec7 | 1705 | int i; |
f9ab5b5b LZ |
1706 | |
1707 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1708 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1709 | #endif |
ddbcc7e8 | 1710 | |
c6d57f33 | 1711 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1712 | |
1713 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1714 | nr_opts++; |
1715 | ||
ddbcc7e8 PM |
1716 | if (!*token) |
1717 | return -EINVAL; | |
32a8cf23 | 1718 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1719 | /* Explicitly have no subsystems */ |
1720 | opts->none = true; | |
32a8cf23 DL |
1721 | continue; |
1722 | } | |
1723 | if (!strcmp(token, "all")) { | |
1724 | /* Mutually exclusive option 'all' + subsystem name */ | |
1725 | if (one_ss) | |
1726 | return -EINVAL; | |
1727 | all_ss = true; | |
1728 | continue; | |
1729 | } | |
1730 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1731 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1732 | continue; |
1733 | } | |
1734 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1735 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1736 | continue; |
1737 | } | |
03b1cde6 | 1738 | if (!strcmp(token, "xattr")) { |
93438629 | 1739 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1740 | continue; |
1741 | } | |
32a8cf23 | 1742 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1743 | /* Specifying two release agents is forbidden */ |
1744 | if (opts->release_agent) | |
1745 | return -EINVAL; | |
c6d57f33 | 1746 | opts->release_agent = |
e400c285 | 1747 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1748 | if (!opts->release_agent) |
1749 | return -ENOMEM; | |
32a8cf23 DL |
1750 | continue; |
1751 | } | |
1752 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1753 | const char *name = token + 5; |
1754 | /* Can't specify an empty name */ | |
1755 | if (!strlen(name)) | |
1756 | return -EINVAL; | |
1757 | /* Must match [\w.-]+ */ | |
1758 | for (i = 0; i < strlen(name); i++) { | |
1759 | char c = name[i]; | |
1760 | if (isalnum(c)) | |
1761 | continue; | |
1762 | if ((c == '.') || (c == '-') || (c == '_')) | |
1763 | continue; | |
1764 | return -EINVAL; | |
1765 | } | |
1766 | /* Specifying two names is forbidden */ | |
1767 | if (opts->name) | |
1768 | return -EINVAL; | |
1769 | opts->name = kstrndup(name, | |
e400c285 | 1770 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1771 | GFP_KERNEL); |
1772 | if (!opts->name) | |
1773 | return -ENOMEM; | |
32a8cf23 DL |
1774 | |
1775 | continue; | |
1776 | } | |
1777 | ||
30159ec7 | 1778 | for_each_subsys(ss, i) { |
3e1d2eed | 1779 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1780 | continue; |
fc5ed1e9 | 1781 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1782 | continue; |
223ffb29 JW |
1783 | if (cgroup_ssid_no_v1(i)) |
1784 | continue; | |
32a8cf23 DL |
1785 | |
1786 | /* Mutually exclusive option 'all' + subsystem name */ | |
1787 | if (all_ss) | |
1788 | return -EINVAL; | |
69dfa00c | 1789 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1790 | one_ss = true; |
1791 | ||
1792 | break; | |
1793 | } | |
1794 | if (i == CGROUP_SUBSYS_COUNT) | |
1795 | return -ENOENT; | |
1796 | } | |
1797 | ||
7b9a6ba5 TH |
1798 | /* |
1799 | * If the 'all' option was specified select all the subsystems, | |
1800 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1801 | * not specified, let's default to 'all' | |
1802 | */ | |
1803 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1804 | for_each_subsys(ss, i) | |
223ffb29 | 1805 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1806 | opts->subsys_mask |= (1 << i); |
1807 | ||
1808 | /* | |
1809 | * We either have to specify by name or by subsystems. (So all | |
1810 | * empty hierarchies must have a name). | |
1811 | */ | |
1812 | if (!opts->subsys_mask && !opts->name) | |
1813 | return -EINVAL; | |
1814 | ||
f9ab5b5b LZ |
1815 | /* |
1816 | * Option noprefix was introduced just for backward compatibility | |
1817 | * with the old cpuset, so we allow noprefix only if mounting just | |
1818 | * the cpuset subsystem. | |
1819 | */ | |
93438629 | 1820 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1821 | return -EINVAL; |
1822 | ||
2c6ab6d2 | 1823 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1824 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1825 | return -EINVAL; |
1826 | ||
ddbcc7e8 PM |
1827 | return 0; |
1828 | } | |
1829 | ||
2bd59d48 | 1830 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1831 | { |
1832 | int ret = 0; | |
3dd06ffa | 1833 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1834 | struct cgroup_sb_opts opts; |
6e5c8307 | 1835 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1836 | |
aa6ec29b TH |
1837 | if (root == &cgrp_dfl_root) { |
1838 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1839 | return -EINVAL; |
1840 | } | |
1841 | ||
334c3679 | 1842 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1843 | |
1844 | /* See what subsystems are wanted */ | |
1845 | ret = parse_cgroupfs_options(data, &opts); | |
1846 | if (ret) | |
1847 | goto out_unlock; | |
1848 | ||
f392e51c | 1849 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1850 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1851 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1852 | |
f392e51c TH |
1853 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1854 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1855 | |
cf5d5941 | 1856 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1857 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1858 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1859 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1860 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1861 | ret = -EINVAL; |
1862 | goto out_unlock; | |
1863 | } | |
1864 | ||
f172e67c | 1865 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1866 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1867 | ret = -EBUSY; |
0670e08b | 1868 | goto out_unlock; |
cf5d5941 | 1869 | } |
ddbcc7e8 | 1870 | |
5df36032 | 1871 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1872 | if (ret) |
0670e08b | 1873 | goto out_unlock; |
ddbcc7e8 | 1874 | |
334c3679 | 1875 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1876 | |
69e943b7 TH |
1877 | if (opts.release_agent) { |
1878 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1879 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1880 | spin_unlock(&release_agent_path_lock); |
1881 | } | |
ed1777de TH |
1882 | |
1883 | trace_cgroup_remount(root); | |
1884 | ||
ddbcc7e8 | 1885 | out_unlock: |
66bdc9cf | 1886 | kfree(opts.release_agent); |
c6d57f33 | 1887 | kfree(opts.name); |
ddbcc7e8 | 1888 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1889 | return ret; |
1890 | } | |
1891 | ||
afeb0f9f TH |
1892 | /* |
1893 | * To reduce the fork() overhead for systems that are not actually using | |
1894 | * their cgroups capability, we don't maintain the lists running through | |
1895 | * each css_set to its tasks until we see the list actually used - in other | |
1896 | * words after the first mount. | |
1897 | */ | |
1898 | static bool use_task_css_set_links __read_mostly; | |
1899 | ||
1900 | static void cgroup_enable_task_cg_lists(void) | |
1901 | { | |
1902 | struct task_struct *p, *g; | |
1903 | ||
82d6489d | 1904 | spin_lock_irq(&css_set_lock); |
afeb0f9f TH |
1905 | |
1906 | if (use_task_css_set_links) | |
1907 | goto out_unlock; | |
1908 | ||
1909 | use_task_css_set_links = true; | |
1910 | ||
1911 | /* | |
1912 | * We need tasklist_lock because RCU is not safe against | |
1913 | * while_each_thread(). Besides, a forking task that has passed | |
1914 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1915 | * is not guaranteed to have its child immediately visible in the | |
1916 | * tasklist if we walk through it with RCU. | |
1917 | */ | |
1918 | read_lock(&tasklist_lock); | |
1919 | do_each_thread(g, p) { | |
afeb0f9f TH |
1920 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1921 | task_css_set(p) != &init_css_set); | |
1922 | ||
1923 | /* | |
1924 | * We should check if the process is exiting, otherwise | |
1925 | * it will race with cgroup_exit() in that the list | |
1926 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1927 | * Do it while holding siglock so that we don't end up |
1928 | * racing against cgroup_exit(). | |
82d6489d DBO |
1929 | * |
1930 | * Interrupts were already disabled while acquiring | |
1931 | * the css_set_lock, so we do not need to disable it | |
1932 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1933 | */ |
82d6489d | 1934 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1935 | if (!(p->flags & PF_EXITING)) { |
1936 | struct css_set *cset = task_css_set(p); | |
1937 | ||
0de0942d TH |
1938 | if (!css_set_populated(cset)) |
1939 | css_set_update_populated(cset, true); | |
389b9c1b | 1940 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1941 | get_css_set(cset); |
1942 | } | |
82d6489d | 1943 | spin_unlock(&p->sighand->siglock); |
afeb0f9f TH |
1944 | } while_each_thread(g, p); |
1945 | read_unlock(&tasklist_lock); | |
1946 | out_unlock: | |
82d6489d | 1947 | spin_unlock_irq(&css_set_lock); |
afeb0f9f | 1948 | } |
ddbcc7e8 | 1949 | |
cc31edce PM |
1950 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1951 | { | |
2d8f243a TH |
1952 | struct cgroup_subsys *ss; |
1953 | int ssid; | |
1954 | ||
d5c419b6 TH |
1955 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1956 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1957 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1958 | INIT_LIST_HEAD(&cgrp->pidlists); |
1959 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1960 | cgrp->self.cgroup = cgrp; |
184faf32 | 1961 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1962 | |
1963 | for_each_subsys(ss, ssid) | |
1964 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1965 | |
1966 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1967 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1968 | } |
c6d57f33 | 1969 | |
3dd06ffa | 1970 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1971 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1972 | { |
3dd06ffa | 1973 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1974 | |
ddbcc7e8 | 1975 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1976 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1977 | cgrp->root = root; |
cc31edce | 1978 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1979 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1980 | |
c6d57f33 PM |
1981 | root->flags = opts->flags; |
1982 | if (opts->release_agent) | |
1983 | strcpy(root->release_agent_path, opts->release_agent); | |
1984 | if (opts->name) | |
1985 | strcpy(root->name, opts->name); | |
2260e7fc | 1986 | if (opts->cpuset_clone_children) |
3dd06ffa | 1987 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1988 | } |
1989 | ||
6e5c8307 | 1990 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1991 | { |
d427dfeb | 1992 | LIST_HEAD(tmp_links); |
3dd06ffa | 1993 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1994 | struct css_set *cset; |
d427dfeb | 1995 | int i, ret; |
2c6ab6d2 | 1996 | |
d427dfeb | 1997 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1998 | |
cf780b7d | 1999 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 2000 | if (ret < 0) |
2bd59d48 | 2001 | goto out; |
d427dfeb | 2002 | root_cgrp->id = ret; |
b11cfb58 | 2003 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 2004 | |
2aad2a86 TH |
2005 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
2006 | GFP_KERNEL); | |
9d755d33 TH |
2007 | if (ret) |
2008 | goto out; | |
2009 | ||
d427dfeb | 2010 | /* |
f0d9a5f1 | 2011 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 2012 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
2013 | * cgroup_lock, and that's us. Later rebinding may disable |
2014 | * controllers on the default hierarchy and thus create new csets, | |
2015 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 2016 | */ |
04313591 | 2017 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 2018 | if (ret) |
9d755d33 | 2019 | goto cancel_ref; |
ddbcc7e8 | 2020 | |
985ed670 | 2021 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 2022 | if (ret) |
9d755d33 | 2023 | goto cancel_ref; |
ddbcc7e8 | 2024 | |
2bd59d48 TH |
2025 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
2026 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
2027 | root_cgrp); | |
2028 | if (IS_ERR(root->kf_root)) { | |
2029 | ret = PTR_ERR(root->kf_root); | |
2030 | goto exit_root_id; | |
2031 | } | |
2032 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 2033 | |
334c3679 | 2034 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 2035 | if (ret) |
2bd59d48 | 2036 | goto destroy_root; |
ddbcc7e8 | 2037 | |
5df36032 | 2038 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 2039 | if (ret) |
2bd59d48 | 2040 | goto destroy_root; |
ddbcc7e8 | 2041 | |
ed1777de TH |
2042 | trace_cgroup_setup_root(root); |
2043 | ||
d427dfeb TH |
2044 | /* |
2045 | * There must be no failure case after here, since rebinding takes | |
2046 | * care of subsystems' refcounts, which are explicitly dropped in | |
2047 | * the failure exit path. | |
2048 | */ | |
2049 | list_add(&root->root_list, &cgroup_roots); | |
2050 | cgroup_root_count++; | |
0df6a63f | 2051 | |
d427dfeb | 2052 | /* |
3dd06ffa | 2053 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2054 | * objects. |
2055 | */ | |
82d6489d | 2056 | spin_lock_irq(&css_set_lock); |
0de0942d | 2057 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2058 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2059 | if (css_set_populated(cset)) |
2060 | cgroup_update_populated(root_cgrp, true); | |
2061 | } | |
82d6489d | 2062 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 2063 | |
d5c419b6 | 2064 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2065 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2066 | |
2bd59d48 | 2067 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2068 | ret = 0; |
2bd59d48 | 2069 | goto out; |
d427dfeb | 2070 | |
2bd59d48 TH |
2071 | destroy_root: |
2072 | kernfs_destroy_root(root->kf_root); | |
2073 | root->kf_root = NULL; | |
2074 | exit_root_id: | |
d427dfeb | 2075 | cgroup_exit_root_id(root); |
9d755d33 | 2076 | cancel_ref: |
9a1049da | 2077 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2078 | out: |
d427dfeb TH |
2079 | free_cgrp_cset_links(&tmp_links); |
2080 | return ret; | |
ddbcc7e8 PM |
2081 | } |
2082 | ||
f7e83571 | 2083 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2084 | int flags, const char *unused_dev_name, |
f7e83571 | 2085 | void *data) |
ddbcc7e8 | 2086 | { |
67e9c74b | 2087 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 2088 | struct super_block *pinned_sb = NULL; |
ed82571b | 2089 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
970317aa | 2090 | struct cgroup_subsys *ss; |
3dd06ffa | 2091 | struct cgroup_root *root; |
ddbcc7e8 | 2092 | struct cgroup_sb_opts opts; |
2bd59d48 | 2093 | struct dentry *dentry; |
8e30e2b8 | 2094 | int ret; |
970317aa | 2095 | int i; |
c6b3d5bc | 2096 | bool new_sb; |
ddbcc7e8 | 2097 | |
ed82571b SH |
2098 | get_cgroup_ns(ns); |
2099 | ||
2100 | /* Check if the caller has permission to mount. */ | |
2101 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2102 | put_cgroup_ns(ns); | |
2103 | return ERR_PTR(-EPERM); | |
2104 | } | |
2105 | ||
56fde9e0 TH |
2106 | /* |
2107 | * The first time anyone tries to mount a cgroup, enable the list | |
2108 | * linking each css_set to its tasks and fix up all existing tasks. | |
2109 | */ | |
2110 | if (!use_task_css_set_links) | |
2111 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2112 | |
67e9c74b TH |
2113 | if (is_v2) { |
2114 | if (data) { | |
2115 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
ed82571b | 2116 | put_cgroup_ns(ns); |
67e9c74b TH |
2117 | return ERR_PTR(-EINVAL); |
2118 | } | |
a7165264 | 2119 | cgrp_dfl_visible = true; |
67e9c74b TH |
2120 | root = &cgrp_dfl_root; |
2121 | cgroup_get(&root->cgrp); | |
2122 | goto out_mount; | |
2123 | } | |
2124 | ||
334c3679 | 2125 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2126 | |
2127 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2128 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2129 | if (ret) |
8e30e2b8 | 2130 | goto out_unlock; |
a015edd2 | 2131 | |
970317aa LZ |
2132 | /* |
2133 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2134 | * still be dying after the previous unmount. Let's drain the | |
2135 | * dying subsystems. We just need to ensure that the ones | |
2136 | * unmounted previously finish dying and don't care about new ones | |
2137 | * starting. Testing ref liveliness is good enough. | |
2138 | */ | |
2139 | for_each_subsys(ss, i) { | |
2140 | if (!(opts.subsys_mask & (1 << i)) || | |
2141 | ss->root == &cgrp_dfl_root) | |
2142 | continue; | |
2143 | ||
2144 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2145 | mutex_unlock(&cgroup_mutex); | |
2146 | msleep(10); | |
2147 | ret = restart_syscall(); | |
2148 | goto out_free; | |
2149 | } | |
2150 | cgroup_put(&ss->root->cgrp); | |
2151 | } | |
2152 | ||
985ed670 | 2153 | for_each_root(root) { |
2bd59d48 | 2154 | bool name_match = false; |
3126121f | 2155 | |
3dd06ffa | 2156 | if (root == &cgrp_dfl_root) |
985ed670 | 2157 | continue; |
3126121f | 2158 | |
cf5d5941 | 2159 | /* |
2bd59d48 TH |
2160 | * If we asked for a name then it must match. Also, if |
2161 | * name matches but sybsys_mask doesn't, we should fail. | |
2162 | * Remember whether name matched. | |
cf5d5941 | 2163 | */ |
2bd59d48 TH |
2164 | if (opts.name) { |
2165 | if (strcmp(opts.name, root->name)) | |
2166 | continue; | |
2167 | name_match = true; | |
2168 | } | |
ddbcc7e8 | 2169 | |
c6d57f33 | 2170 | /* |
2bd59d48 TH |
2171 | * If we asked for subsystems (or explicitly for no |
2172 | * subsystems) then they must match. | |
c6d57f33 | 2173 | */ |
2bd59d48 | 2174 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2175 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2176 | if (!name_match) |
2177 | continue; | |
2178 | ret = -EBUSY; | |
2179 | goto out_unlock; | |
2180 | } | |
873fe09e | 2181 | |
7b9a6ba5 TH |
2182 | if (root->flags ^ opts.flags) |
2183 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2184 | |
776f02fa | 2185 | /* |
3a32bd72 LZ |
2186 | * We want to reuse @root whose lifetime is governed by its |
2187 | * ->cgrp. Let's check whether @root is alive and keep it | |
2188 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2189 | * want to block it by pinning the sb so that @root doesn't | |
2190 | * get killed before mount is complete. | |
2191 | * | |
2192 | * With the sb pinned, tryget_live can reliably indicate | |
2193 | * whether @root can be reused. If it's being killed, | |
2194 | * drain it. We can use wait_queue for the wait but this | |
2195 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2196 | */ |
3a32bd72 LZ |
2197 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2198 | if (IS_ERR(pinned_sb) || | |
2199 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2200 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2201 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2202 | deactivate_super(pinned_sb); | |
776f02fa | 2203 | msleep(10); |
a015edd2 TH |
2204 | ret = restart_syscall(); |
2205 | goto out_free; | |
776f02fa | 2206 | } |
ddbcc7e8 | 2207 | |
776f02fa | 2208 | ret = 0; |
2bd59d48 | 2209 | goto out_unlock; |
ddbcc7e8 | 2210 | } |
ddbcc7e8 | 2211 | |
817929ec | 2212 | /* |
172a2c06 TH |
2213 | * No such thing, create a new one. name= matching without subsys |
2214 | * specification is allowed for already existing hierarchies but we | |
2215 | * can't create new one without subsys specification. | |
817929ec | 2216 | */ |
172a2c06 TH |
2217 | if (!opts.subsys_mask && !opts.none) { |
2218 | ret = -EINVAL; | |
2219 | goto out_unlock; | |
817929ec | 2220 | } |
817929ec | 2221 | |
726a4994 EB |
2222 | /* Hierarchies may only be created in the initial cgroup namespace. */ |
2223 | if (ns != &init_cgroup_ns) { | |
ed82571b SH |
2224 | ret = -EPERM; |
2225 | goto out_unlock; | |
2226 | } | |
2227 | ||
172a2c06 TH |
2228 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2229 | if (!root) { | |
2230 | ret = -ENOMEM; | |
2bd59d48 | 2231 | goto out_unlock; |
839ec545 | 2232 | } |
e5f6a860 | 2233 | |
172a2c06 TH |
2234 | init_cgroup_root(root, &opts); |
2235 | ||
35585573 | 2236 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2237 | if (ret) |
2238 | cgroup_free_root(root); | |
fa3ca07e | 2239 | |
8e30e2b8 | 2240 | out_unlock: |
ddbcc7e8 | 2241 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2242 | out_free: |
c6d57f33 PM |
2243 | kfree(opts.release_agent); |
2244 | kfree(opts.name); | |
03b1cde6 | 2245 | |
ed82571b SH |
2246 | if (ret) { |
2247 | put_cgroup_ns(ns); | |
8e30e2b8 | 2248 | return ERR_PTR(ret); |
ed82571b | 2249 | } |
67e9c74b | 2250 | out_mount: |
c9482a5b | 2251 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2252 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2253 | &new_sb); | |
ed82571b SH |
2254 | |
2255 | /* | |
2256 | * In non-init cgroup namespace, instead of root cgroup's | |
2257 | * dentry, we return the dentry corresponding to the | |
2258 | * cgroupns->root_cgrp. | |
2259 | */ | |
2260 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { | |
2261 | struct dentry *nsdentry; | |
2262 | struct cgroup *cgrp; | |
2263 | ||
2264 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2265 | spin_lock_irq(&css_set_lock); |
ed82571b SH |
2266 | |
2267 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2268 | ||
82d6489d | 2269 | spin_unlock_irq(&css_set_lock); |
ed82571b SH |
2270 | mutex_unlock(&cgroup_mutex); |
2271 | ||
2272 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); | |
2273 | dput(dentry); | |
2274 | dentry = nsdentry; | |
2275 | } | |
2276 | ||
c6b3d5bc | 2277 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2278 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2279 | |
2280 | /* | |
2281 | * If @pinned_sb, we're reusing an existing root and holding an | |
2282 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2283 | */ | |
2284 | if (pinned_sb) { | |
2285 | WARN_ON(new_sb); | |
2286 | deactivate_super(pinned_sb); | |
2287 | } | |
2288 | ||
ed82571b | 2289 | put_cgroup_ns(ns); |
2bd59d48 TH |
2290 | return dentry; |
2291 | } | |
2292 | ||
2293 | static void cgroup_kill_sb(struct super_block *sb) | |
2294 | { | |
2295 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2296 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2297 | |
9d755d33 TH |
2298 | /* |
2299 | * If @root doesn't have any mounts or children, start killing it. | |
2300 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2301 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2302 | * |
2303 | * And don't kill the default root. | |
9d755d33 | 2304 | */ |
3c606d35 | 2305 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2306 | root == &cgrp_dfl_root) |
9d755d33 TH |
2307 | cgroup_put(&root->cgrp); |
2308 | else | |
2309 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2310 | ||
2bd59d48 | 2311 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2312 | } |
2313 | ||
2314 | static struct file_system_type cgroup_fs_type = { | |
2315 | .name = "cgroup", | |
f7e83571 | 2316 | .mount = cgroup_mount, |
ddbcc7e8 | 2317 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2318 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 PM |
2319 | }; |
2320 | ||
67e9c74b TH |
2321 | static struct file_system_type cgroup2_fs_type = { |
2322 | .name = "cgroup2", | |
2323 | .mount = cgroup_mount, | |
2324 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2325 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b TH |
2326 | }; |
2327 | ||
4c737b41 TH |
2328 | static int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2329 | struct cgroup_namespace *ns) | |
a79a908f AK |
2330 | { |
2331 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 2332 | |
4c737b41 | 2333 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
2334 | } |
2335 | ||
4c737b41 TH |
2336 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
2337 | struct cgroup_namespace *ns) | |
a79a908f | 2338 | { |
4c737b41 | 2339 | int ret; |
a79a908f AK |
2340 | |
2341 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2342 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2343 | |
2344 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2345 | ||
82d6489d | 2346 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2347 | mutex_unlock(&cgroup_mutex); |
2348 | ||
2349 | return ret; | |
2350 | } | |
2351 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2352 | ||
857a2beb | 2353 | /** |
913ffdb5 | 2354 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2355 | * @task: target task |
857a2beb TH |
2356 | * @buf: the buffer to write the path into |
2357 | * @buflen: the length of the buffer | |
2358 | * | |
913ffdb5 TH |
2359 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2360 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2361 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2362 | * cgroup controller callbacks. | |
2363 | * | |
e61734c5 | 2364 | * Return value is the same as kernfs_path(). |
857a2beb | 2365 | */ |
4c737b41 | 2366 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2367 | { |
3dd06ffa | 2368 | struct cgroup_root *root; |
913ffdb5 | 2369 | struct cgroup *cgrp; |
e61734c5 | 2370 | int hierarchy_id = 1; |
4c737b41 | 2371 | int ret; |
857a2beb TH |
2372 | |
2373 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2374 | spin_lock_irq(&css_set_lock); |
857a2beb | 2375 | |
913ffdb5 TH |
2376 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2377 | ||
857a2beb TH |
2378 | if (root) { |
2379 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 2380 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2381 | } else { |
2382 | /* if no hierarchy exists, everyone is in "/" */ | |
4c737b41 | 2383 | ret = strlcpy(buf, "/", buflen); |
857a2beb TH |
2384 | } |
2385 | ||
82d6489d | 2386 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2387 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 2388 | return ret; |
857a2beb | 2389 | } |
913ffdb5 | 2390 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2391 | |
b3dc094e | 2392 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2393 | struct cgroup_taskset { |
b3dc094e TH |
2394 | /* the src and dst cset list running through cset->mg_node */ |
2395 | struct list_head src_csets; | |
2396 | struct list_head dst_csets; | |
2397 | ||
1f7dd3e5 TH |
2398 | /* the subsys currently being processed */ |
2399 | int ssid; | |
2400 | ||
b3dc094e TH |
2401 | /* |
2402 | * Fields for cgroup_taskset_*() iteration. | |
2403 | * | |
2404 | * Before migration is committed, the target migration tasks are on | |
2405 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2406 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2407 | * or ->dst_csets depending on whether migration is committed. | |
2408 | * | |
2409 | * ->cur_csets and ->cur_task point to the current task position | |
2410 | * during iteration. | |
2411 | */ | |
2412 | struct list_head *csets; | |
2413 | struct css_set *cur_cset; | |
2414 | struct task_struct *cur_task; | |
2f7ee569 TH |
2415 | }; |
2416 | ||
adaae5dc TH |
2417 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2418 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2419 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2420 | .csets = &tset.src_csets, \ | |
2421 | } | |
2422 | ||
2423 | /** | |
2424 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2425 | * @task: target task | |
2426 | * @tset: target taskset | |
2427 | * | |
2428 | * Add @task, which is a migration target, to @tset. This function becomes | |
2429 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2430 | * been added as a migration source and @task->cg_list will be moved from | |
2431 | * the css_set's tasks list to mg_tasks one. | |
2432 | */ | |
2433 | static void cgroup_taskset_add(struct task_struct *task, | |
2434 | struct cgroup_taskset *tset) | |
2435 | { | |
2436 | struct css_set *cset; | |
2437 | ||
f0d9a5f1 | 2438 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2439 | |
2440 | /* @task either already exited or can't exit until the end */ | |
2441 | if (task->flags & PF_EXITING) | |
2442 | return; | |
2443 | ||
2444 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2445 | if (list_empty(&task->cg_list)) | |
2446 | return; | |
2447 | ||
2448 | cset = task_css_set(task); | |
2449 | if (!cset->mg_src_cgrp) | |
2450 | return; | |
2451 | ||
2452 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2453 | if (list_empty(&cset->mg_node)) | |
2454 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2455 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2456 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2457 | &tset->dst_csets); | |
2458 | } | |
2459 | ||
2f7ee569 TH |
2460 | /** |
2461 | * cgroup_taskset_first - reset taskset and return the first task | |
2462 | * @tset: taskset of interest | |
1f7dd3e5 | 2463 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2464 | * |
2465 | * @tset iteration is initialized and the first task is returned. | |
2466 | */ | |
1f7dd3e5 TH |
2467 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2468 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2469 | { |
b3dc094e TH |
2470 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2471 | tset->cur_task = NULL; | |
2472 | ||
1f7dd3e5 | 2473 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2474 | } |
2f7ee569 TH |
2475 | |
2476 | /** | |
2477 | * cgroup_taskset_next - iterate to the next task in taskset | |
2478 | * @tset: taskset of interest | |
1f7dd3e5 | 2479 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2480 | * |
2481 | * Return the next task in @tset. Iteration must have been initialized | |
2482 | * with cgroup_taskset_first(). | |
2483 | */ | |
1f7dd3e5 TH |
2484 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2485 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2486 | { |
b3dc094e TH |
2487 | struct css_set *cset = tset->cur_cset; |
2488 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2489 | |
b3dc094e TH |
2490 | while (&cset->mg_node != tset->csets) { |
2491 | if (!task) | |
2492 | task = list_first_entry(&cset->mg_tasks, | |
2493 | struct task_struct, cg_list); | |
2494 | else | |
2495 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2496 | |
b3dc094e TH |
2497 | if (&task->cg_list != &cset->mg_tasks) { |
2498 | tset->cur_cset = cset; | |
2499 | tset->cur_task = task; | |
1f7dd3e5 TH |
2500 | |
2501 | /* | |
2502 | * This function may be called both before and | |
2503 | * after cgroup_taskset_migrate(). The two cases | |
2504 | * can be distinguished by looking at whether @cset | |
2505 | * has its ->mg_dst_cset set. | |
2506 | */ | |
2507 | if (cset->mg_dst_cset) | |
2508 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2509 | else | |
2510 | *dst_cssp = cset->subsys[tset->ssid]; | |
2511 | ||
b3dc094e TH |
2512 | return task; |
2513 | } | |
2f7ee569 | 2514 | |
b3dc094e TH |
2515 | cset = list_next_entry(cset, mg_node); |
2516 | task = NULL; | |
2517 | } | |
2f7ee569 | 2518 | |
b3dc094e | 2519 | return NULL; |
2f7ee569 | 2520 | } |
2f7ee569 | 2521 | |
adaae5dc | 2522 | /** |
37ff9f8f | 2523 | * cgroup_taskset_migrate - migrate a taskset |
adaae5dc | 2524 | * @tset: taget taskset |
37ff9f8f | 2525 | * @root: cgroup root the migration is taking place on |
adaae5dc | 2526 | * |
37ff9f8f TH |
2527 | * Migrate tasks in @tset as setup by migration preparation functions. |
2528 | * This function fails iff one of the ->can_attach callbacks fails and | |
2529 | * guarantees that either all or none of the tasks in @tset are migrated. | |
2530 | * @tset is consumed regardless of success. | |
adaae5dc TH |
2531 | */ |
2532 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
37ff9f8f | 2533 | struct cgroup_root *root) |
adaae5dc | 2534 | { |
37ff9f8f | 2535 | struct cgroup_subsys *ss; |
adaae5dc TH |
2536 | struct task_struct *task, *tmp_task; |
2537 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2538 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2539 | |
2540 | /* methods shouldn't be called if no task is actually migrating */ | |
2541 | if (list_empty(&tset->src_csets)) | |
2542 | return 0; | |
2543 | ||
2544 | /* check that we can legitimately attach to the cgroup */ | |
37ff9f8f TH |
2545 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2546 | if (ss->can_attach) { | |
2547 | tset->ssid = ssid; | |
2548 | ret = ss->can_attach(tset); | |
adaae5dc | 2549 | if (ret) { |
37ff9f8f | 2550 | failed_ssid = ssid; |
adaae5dc TH |
2551 | goto out_cancel_attach; |
2552 | } | |
2553 | } | |
37ff9f8f | 2554 | } while_each_subsys_mask(); |
adaae5dc TH |
2555 | |
2556 | /* | |
2557 | * Now that we're guaranteed success, proceed to move all tasks to | |
2558 | * the new cgroup. There are no failure cases after here, so this | |
2559 | * is the commit point. | |
2560 | */ | |
82d6489d | 2561 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2562 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2563 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2564 | struct css_set *from_cset = task_css_set(task); | |
2565 | struct css_set *to_cset = cset->mg_dst_cset; | |
2566 | ||
2567 | get_css_set(to_cset); | |
2568 | css_set_move_task(task, from_cset, to_cset, true); | |
2569 | put_css_set_locked(from_cset); | |
2570 | } | |
adaae5dc | 2571 | } |
82d6489d | 2572 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2573 | |
2574 | /* | |
2575 | * Migration is committed, all target tasks are now on dst_csets. | |
2576 | * Nothing is sensitive to fork() after this point. Notify | |
2577 | * controllers that migration is complete. | |
2578 | */ | |
2579 | tset->csets = &tset->dst_csets; | |
2580 | ||
37ff9f8f TH |
2581 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2582 | if (ss->attach) { | |
2583 | tset->ssid = ssid; | |
2584 | ss->attach(tset); | |
1f7dd3e5 | 2585 | } |
37ff9f8f | 2586 | } while_each_subsys_mask(); |
adaae5dc TH |
2587 | |
2588 | ret = 0; | |
2589 | goto out_release_tset; | |
2590 | ||
2591 | out_cancel_attach: | |
37ff9f8f TH |
2592 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2593 | if (ssid == failed_ssid) | |
adaae5dc | 2594 | break; |
37ff9f8f TH |
2595 | if (ss->cancel_attach) { |
2596 | tset->ssid = ssid; | |
2597 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2598 | } |
37ff9f8f | 2599 | } while_each_subsys_mask(); |
adaae5dc | 2600 | out_release_tset: |
82d6489d | 2601 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2602 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2603 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2604 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2605 | list_del_init(&cset->mg_node); | |
2606 | } | |
82d6489d | 2607 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2608 | return ret; |
2609 | } | |
2610 | ||
6c694c88 TH |
2611 | /** |
2612 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2613 | * @dst_cgrp: destination cgroup to test | |
2614 | * | |
2615 | * On the default hierarchy, except for the root, subtree_control must be | |
2616 | * zero for migration destination cgroups with tasks so that child cgroups | |
2617 | * don't compete against tasks. | |
2618 | */ | |
2619 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) | |
2620 | { | |
2621 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2622 | !dst_cgrp->subtree_control; | |
2623 | } | |
2624 | ||
a043e3b2 | 2625 | /** |
1958d2d5 TH |
2626 | * cgroup_migrate_finish - cleanup after attach |
2627 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2628 | * |
1958d2d5 TH |
2629 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2630 | * those functions for details. | |
74a1166d | 2631 | */ |
1958d2d5 | 2632 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2633 | { |
1958d2d5 | 2634 | struct css_set *cset, *tmp_cset; |
74a1166d | 2635 | |
1958d2d5 TH |
2636 | lockdep_assert_held(&cgroup_mutex); |
2637 | ||
82d6489d | 2638 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2639 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2640 | cset->mg_src_cgrp = NULL; | |
e4857982 | 2641 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2642 | cset->mg_dst_cset = NULL; |
2643 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2644 | put_css_set_locked(cset); |
1958d2d5 | 2645 | } |
82d6489d | 2646 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2647 | } |
2648 | ||
2649 | /** | |
2650 | * cgroup_migrate_add_src - add a migration source css_set | |
2651 | * @src_cset: the source css_set to add | |
2652 | * @dst_cgrp: the destination cgroup | |
2653 | * @preloaded_csets: list of preloaded css_sets | |
2654 | * | |
2655 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2656 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2657 | * up by cgroup_migrate_finish(). | |
2658 | * | |
1ed13287 TH |
2659 | * This function may be called without holding cgroup_threadgroup_rwsem |
2660 | * even if the target is a process. Threads may be created and destroyed | |
2661 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2662 | * into play and the preloaded css_sets are guaranteed to cover all | |
2663 | * migrations. | |
1958d2d5 TH |
2664 | */ |
2665 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2666 | struct cgroup *dst_cgrp, | |
2667 | struct list_head *preloaded_csets) | |
2668 | { | |
2669 | struct cgroup *src_cgrp; | |
2670 | ||
2671 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2672 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2673 | |
2b021cbf TH |
2674 | /* |
2675 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2676 | * and doesn't contain any migratable tasks. Ignore it early so | |
2677 | * that the rest of migration path doesn't get confused by it. | |
2678 | */ | |
2679 | if (src_cset->dead) | |
2680 | return; | |
2681 | ||
1958d2d5 TH |
2682 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2683 | ||
1958d2d5 TH |
2684 | if (!list_empty(&src_cset->mg_preload_node)) |
2685 | return; | |
2686 | ||
2687 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2688 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2689 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2690 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2691 | ||
2692 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2693 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 TH |
2694 | get_css_set(src_cset); |
2695 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2696 | } | |
2697 | ||
2698 | /** | |
2699 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1958d2d5 TH |
2700 | * @preloaded_csets: list of preloaded source css_sets |
2701 | * | |
e4857982 TH |
2702 | * Tasks are about to be moved and all the source css_sets have been |
2703 | * preloaded to @preloaded_csets. This function looks up and pins all | |
2704 | * destination css_sets, links each to its source, and append them to | |
2705 | * @preloaded_csets. | |
1958d2d5 TH |
2706 | * |
2707 | * This function must be called after cgroup_migrate_add_src() has been | |
2708 | * called on each migration source css_set. After migration is performed | |
2709 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2710 | * @preloaded_csets. | |
2711 | */ | |
e4857982 | 2712 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
1958d2d5 TH |
2713 | { |
2714 | LIST_HEAD(csets); | |
f817de98 | 2715 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2716 | |
2717 | lockdep_assert_held(&cgroup_mutex); | |
2718 | ||
2719 | /* look up the dst cset for each src cset and link it to src */ | |
f817de98 | 2720 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2721 | struct css_set *dst_cset; |
2722 | ||
e4857982 | 2723 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2724 | if (!dst_cset) |
2725 | goto err; | |
2726 | ||
2727 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2728 | |
2729 | /* | |
2730 | * If src cset equals dst, it's noop. Drop the src. | |
2731 | * cgroup_migrate() will skip the cset too. Note that we | |
2732 | * can't handle src == dst as some nodes are used by both. | |
2733 | */ | |
2734 | if (src_cset == dst_cset) { | |
2735 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2736 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2737 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2738 | put_css_set(src_cset); |
2739 | put_css_set(dst_cset); | |
f817de98 TH |
2740 | continue; |
2741 | } | |
2742 | ||
1958d2d5 TH |
2743 | src_cset->mg_dst_cset = dst_cset; |
2744 | ||
2745 | if (list_empty(&dst_cset->mg_preload_node)) | |
2746 | list_add(&dst_cset->mg_preload_node, &csets); | |
2747 | else | |
a25eb52e | 2748 | put_css_set(dst_cset); |
1958d2d5 TH |
2749 | } |
2750 | ||
f817de98 | 2751 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2752 | return 0; |
2753 | err: | |
2754 | cgroup_migrate_finish(&csets); | |
2755 | return -ENOMEM; | |
2756 | } | |
2757 | ||
2758 | /** | |
2759 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2760 | * @leader: the leader of the process or the task to migrate |
2761 | * @threadgroup: whether @leader points to the whole process or a single task | |
37ff9f8f | 2762 | * @root: cgroup root migration is taking place on |
1958d2d5 | 2763 | * |
37ff9f8f TH |
2764 | * Migrate a process or task denoted by @leader. If migrating a process, |
2765 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2766 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2767 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2768 | * function and following up with cgroup_migrate_finish(). | |
2769 | * | |
2770 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2771 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2772 | * failure, when migrating multiple targets, the success or failure can be | |
2773 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2774 | * actually starting migrating. | |
2775 | */ | |
9af2ec45 | 2776 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
37ff9f8f | 2777 | struct cgroup_root *root) |
74a1166d | 2778 | { |
adaae5dc TH |
2779 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2780 | struct task_struct *task; | |
74a1166d | 2781 | |
fb5d2b4c MSB |
2782 | /* |
2783 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2784 | * already PF_EXITING could be freed from underneath us unless we | |
2785 | * take an rcu_read_lock. | |
2786 | */ | |
82d6489d | 2787 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2788 | rcu_read_lock(); |
9db8de37 | 2789 | task = leader; |
74a1166d | 2790 | do { |
adaae5dc | 2791 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2792 | if (!threadgroup) |
2793 | break; | |
9db8de37 | 2794 | } while_each_thread(leader, task); |
fb5d2b4c | 2795 | rcu_read_unlock(); |
82d6489d | 2796 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2797 | |
37ff9f8f | 2798 | return cgroup_taskset_migrate(&tset, root); |
74a1166d BB |
2799 | } |
2800 | ||
1958d2d5 TH |
2801 | /** |
2802 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2803 | * @dst_cgrp: the cgroup to attach to | |
2804 | * @leader: the task or the leader of the threadgroup to be attached | |
2805 | * @threadgroup: attach the whole threadgroup? | |
2806 | * | |
1ed13287 | 2807 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2808 | */ |
2809 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2810 | struct task_struct *leader, bool threadgroup) | |
2811 | { | |
2812 | LIST_HEAD(preloaded_csets); | |
2813 | struct task_struct *task; | |
2814 | int ret; | |
2815 | ||
6c694c88 TH |
2816 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2817 | return -EBUSY; | |
2818 | ||
1958d2d5 | 2819 | /* look up all src csets */ |
82d6489d | 2820 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2821 | rcu_read_lock(); |
2822 | task = leader; | |
2823 | do { | |
2824 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2825 | &preloaded_csets); | |
2826 | if (!threadgroup) | |
2827 | break; | |
2828 | } while_each_thread(leader, task); | |
2829 | rcu_read_unlock(); | |
82d6489d | 2830 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2831 | |
2832 | /* prepare dst csets and commit */ | |
e4857982 | 2833 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
1958d2d5 | 2834 | if (!ret) |
37ff9f8f | 2835 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
1958d2d5 TH |
2836 | |
2837 | cgroup_migrate_finish(&preloaded_csets); | |
ed1777de TH |
2838 | |
2839 | if (!ret) | |
2840 | trace_cgroup_attach_task(dst_cgrp, leader, threadgroup); | |
2841 | ||
1958d2d5 | 2842 | return ret; |
74a1166d BB |
2843 | } |
2844 | ||
187fe840 TH |
2845 | static int cgroup_procs_write_permission(struct task_struct *task, |
2846 | struct cgroup *dst_cgrp, | |
2847 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2848 | { |
2849 | const struct cred *cred = current_cred(); | |
2850 | const struct cred *tcred = get_task_cred(task); | |
2851 | int ret = 0; | |
2852 | ||
2853 | /* | |
2854 | * even if we're attaching all tasks in the thread group, we only | |
2855 | * need to check permissions on one of them. | |
2856 | */ | |
2857 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2858 | !uid_eq(cred->euid, tcred->uid) && | |
2859 | !uid_eq(cred->euid, tcred->suid)) | |
2860 | ret = -EACCES; | |
2861 | ||
187fe840 TH |
2862 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2863 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2864 | struct cgroup *cgrp; | |
2865 | struct inode *inode; | |
2866 | ||
82d6489d | 2867 | spin_lock_irq(&css_set_lock); |
187fe840 | 2868 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
82d6489d | 2869 | spin_unlock_irq(&css_set_lock); |
187fe840 TH |
2870 | |
2871 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2872 | cgrp = cgroup_parent(cgrp); | |
2873 | ||
2874 | ret = -ENOMEM; | |
6f60eade | 2875 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2876 | if (inode) { |
2877 | ret = inode_permission(inode, MAY_WRITE); | |
2878 | iput(inode); | |
2879 | } | |
2880 | } | |
2881 | ||
dedf22e9 TH |
2882 | put_cred(tcred); |
2883 | return ret; | |
2884 | } | |
2885 | ||
74a1166d BB |
2886 | /* |
2887 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2888 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2889 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2890 | */ |
acbef755 TH |
2891 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2892 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2893 | { |
bbcb81d0 | 2894 | struct task_struct *tsk; |
5cf1cacb | 2895 | struct cgroup_subsys *ss; |
e76ecaee | 2896 | struct cgroup *cgrp; |
acbef755 | 2897 | pid_t pid; |
5cf1cacb | 2898 | int ssid, ret; |
bbcb81d0 | 2899 | |
acbef755 TH |
2900 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2901 | return -EINVAL; | |
2902 | ||
945ba199 | 2903 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2904 | if (!cgrp) |
74a1166d BB |
2905 | return -ENODEV; |
2906 | ||
3014dde7 | 2907 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2908 | rcu_read_lock(); |
bbcb81d0 | 2909 | if (pid) { |
73507f33 | 2910 | tsk = find_task_by_vpid(pid); |
74a1166d | 2911 | if (!tsk) { |
dd4b0a46 | 2912 | ret = -ESRCH; |
3014dde7 | 2913 | goto out_unlock_rcu; |
bbcb81d0 | 2914 | } |
dedf22e9 | 2915 | } else { |
b78949eb | 2916 | tsk = current; |
dedf22e9 | 2917 | } |
cd3d0952 TH |
2918 | |
2919 | if (threadgroup) | |
b78949eb | 2920 | tsk = tsk->group_leader; |
c4c27fbd MG |
2921 | |
2922 | /* | |
14a40ffc | 2923 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2924 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2925 | * with no rt_runtime allocated. Just say no. | |
2926 | */ | |
14a40ffc | 2927 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2928 | ret = -EINVAL; |
3014dde7 | 2929 | goto out_unlock_rcu; |
c4c27fbd MG |
2930 | } |
2931 | ||
b78949eb MSB |
2932 | get_task_struct(tsk); |
2933 | rcu_read_unlock(); | |
2934 | ||
187fe840 | 2935 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2936 | if (!ret) |
2937 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2938 | |
f9f9e7b7 | 2939 | put_task_struct(tsk); |
3014dde7 TH |
2940 | goto out_unlock_threadgroup; |
2941 | ||
2942 | out_unlock_rcu: | |
2943 | rcu_read_unlock(); | |
2944 | out_unlock_threadgroup: | |
2945 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2946 | for_each_subsys(ss, ssid) |
2947 | if (ss->post_attach) | |
2948 | ss->post_attach(); | |
e76ecaee | 2949 | cgroup_kn_unlock(of->kn); |
acbef755 | 2950 | return ret ?: nbytes; |
bbcb81d0 PM |
2951 | } |
2952 | ||
7ae1bad9 TH |
2953 | /** |
2954 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2955 | * @from: attach to all cgroups of a given task | |
2956 | * @tsk: the task to be attached | |
2957 | */ | |
2958 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2959 | { | |
3dd06ffa | 2960 | struct cgroup_root *root; |
7ae1bad9 TH |
2961 | int retval = 0; |
2962 | ||
47cfcd09 | 2963 | mutex_lock(&cgroup_mutex); |
eedd0f4c | 2964 | percpu_down_write(&cgroup_threadgroup_rwsem); |
985ed670 | 2965 | for_each_root(root) { |
96d365e0 TH |
2966 | struct cgroup *from_cgrp; |
2967 | ||
3dd06ffa | 2968 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2969 | continue; |
2970 | ||
82d6489d | 2971 | spin_lock_irq(&css_set_lock); |
96d365e0 | 2972 | from_cgrp = task_cgroup_from_root(from, root); |
82d6489d | 2973 | spin_unlock_irq(&css_set_lock); |
7ae1bad9 | 2974 | |
6f4b7e63 | 2975 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2976 | if (retval) |
2977 | break; | |
2978 | } | |
eedd0f4c | 2979 | percpu_up_write(&cgroup_threadgroup_rwsem); |
47cfcd09 | 2980 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2981 | |
2982 | return retval; | |
2983 | } | |
2984 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2985 | ||
acbef755 TH |
2986 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2987 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2988 | { |
acbef755 | 2989 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2990 | } |
2991 | ||
acbef755 TH |
2992 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2993 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2994 | { |
acbef755 | 2995 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2996 | } |
2997 | ||
451af504 TH |
2998 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2999 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 3000 | { |
e76ecaee | 3001 | struct cgroup *cgrp; |
5f469907 | 3002 | |
e76ecaee | 3003 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 3004 | |
945ba199 | 3005 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 3006 | if (!cgrp) |
e788e066 | 3007 | return -ENODEV; |
69e943b7 | 3008 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
3009 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
3010 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 3011 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 3012 | cgroup_kn_unlock(of->kn); |
451af504 | 3013 | return nbytes; |
e788e066 PM |
3014 | } |
3015 | ||
2da8ca82 | 3016 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 3017 | { |
2da8ca82 | 3018 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 3019 | |
46cfeb04 | 3020 | spin_lock(&release_agent_path_lock); |
e788e066 | 3021 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 3022 | spin_unlock(&release_agent_path_lock); |
e788e066 | 3023 | seq_putc(seq, '\n'); |
e788e066 PM |
3024 | return 0; |
3025 | } | |
3026 | ||
2da8ca82 | 3027 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 3028 | { |
c1d5d42e | 3029 | seq_puts(seq, "0\n"); |
e788e066 PM |
3030 | return 0; |
3031 | } | |
3032 | ||
6e5c8307 | 3033 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 3034 | { |
f8f22e53 TH |
3035 | struct cgroup_subsys *ss; |
3036 | bool printed = false; | |
3037 | int ssid; | |
a742c59d | 3038 | |
b4e0eeaf | 3039 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
3040 | if (printed) |
3041 | seq_putc(seq, ' '); | |
3042 | seq_printf(seq, "%s", ss->name); | |
3043 | printed = true; | |
b4e0eeaf | 3044 | } while_each_subsys_mask(); |
f8f22e53 TH |
3045 | if (printed) |
3046 | seq_putc(seq, '\n'); | |
355e0c48 PM |
3047 | } |
3048 | ||
f8f22e53 TH |
3049 | /* show controllers which are enabled from the parent */ |
3050 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3051 | { |
f8f22e53 TH |
3052 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3053 | ||
5531dc91 | 3054 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 3055 | return 0; |
ddbcc7e8 PM |
3056 | } |
3057 | ||
f8f22e53 TH |
3058 | /* show controllers which are enabled for a given cgroup's children */ |
3059 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3060 | { |
f8f22e53 TH |
3061 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3062 | ||
667c2491 | 3063 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
3064 | return 0; |
3065 | } | |
3066 | ||
3067 | /** | |
3068 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
3069 | * @cgrp: root of the subtree to update csses for | |
3070 | * | |
54962604 TH |
3071 | * @cgrp's control masks have changed and its subtree's css associations |
3072 | * need to be updated accordingly. This function looks up all css_sets | |
3073 | * which are attached to the subtree, creates the matching updated css_sets | |
3074 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
3075 | */ |
3076 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
3077 | { | |
3078 | LIST_HEAD(preloaded_csets); | |
10265075 | 3079 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
3080 | struct cgroup_subsys_state *d_css; |
3081 | struct cgroup *dsct; | |
f8f22e53 TH |
3082 | struct css_set *src_cset; |
3083 | int ret; | |
3084 | ||
f8f22e53 TH |
3085 | lockdep_assert_held(&cgroup_mutex); |
3086 | ||
3014dde7 TH |
3087 | percpu_down_write(&cgroup_threadgroup_rwsem); |
3088 | ||
f8f22e53 | 3089 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 3090 | spin_lock_irq(&css_set_lock); |
54962604 | 3091 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
3092 | struct cgrp_cset_link *link; |
3093 | ||
54962604 | 3094 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 3095 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
3096 | &preloaded_csets); |
3097 | } | |
82d6489d | 3098 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
3099 | |
3100 | /* NULL dst indicates self on default hierarchy */ | |
e4857982 | 3101 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
f8f22e53 TH |
3102 | if (ret) |
3103 | goto out_finish; | |
3104 | ||
82d6489d | 3105 | spin_lock_irq(&css_set_lock); |
f8f22e53 | 3106 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 3107 | struct task_struct *task, *ntask; |
f8f22e53 TH |
3108 | |
3109 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
3110 | if (!src_cset->mg_src_cgrp) | |
3111 | break; | |
3112 | ||
10265075 TH |
3113 | /* all tasks in src_csets need to be migrated */ |
3114 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
3115 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 3116 | } |
82d6489d | 3117 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 3118 | |
37ff9f8f | 3119 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
f8f22e53 TH |
3120 | out_finish: |
3121 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 3122 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
3123 | return ret; |
3124 | } | |
3125 | ||
1b9b96a1 | 3126 | /** |
945ba199 | 3127 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 3128 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
3129 | * |
3130 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
3131 | * controller while the previous css is still around. This function grabs |
3132 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 3133 | */ |
945ba199 TH |
3134 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
3135 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
3136 | { |
3137 | struct cgroup *dsct; | |
ce3f1d9d | 3138 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
3139 | struct cgroup_subsys *ss; |
3140 | int ssid; | |
3141 | ||
945ba199 TH |
3142 | restart: |
3143 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 3144 | |
ce3f1d9d | 3145 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
3146 | for_each_subsys(ss, ssid) { |
3147 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3148 | DEFINE_WAIT(wait); | |
3149 | ||
ce3f1d9d | 3150 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
3151 | continue; |
3152 | ||
3153 | cgroup_get(dsct); | |
3154 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
3155 | TASK_UNINTERRUPTIBLE); | |
3156 | ||
3157 | mutex_unlock(&cgroup_mutex); | |
3158 | schedule(); | |
3159 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
3160 | |
3161 | cgroup_put(dsct); | |
945ba199 | 3162 | goto restart; |
1b9b96a1 TH |
3163 | } |
3164 | } | |
1b9b96a1 TH |
3165 | } |
3166 | ||
15a27c36 TH |
3167 | /** |
3168 | * cgroup_save_control - save control masks of a subtree | |
3169 | * @cgrp: root of the target subtree | |
3170 | * | |
3171 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
3172 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3173 | */ | |
3174 | static void cgroup_save_control(struct cgroup *cgrp) | |
3175 | { | |
3176 | struct cgroup *dsct; | |
3177 | struct cgroup_subsys_state *d_css; | |
3178 | ||
3179 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3180 | dsct->old_subtree_control = dsct->subtree_control; | |
3181 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
3182 | } | |
3183 | } | |
3184 | ||
3185 | /** | |
3186 | * cgroup_propagate_control - refresh control masks of a subtree | |
3187 | * @cgrp: root of the target subtree | |
3188 | * | |
3189 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3190 | * ->subtree_control and propagate controller availability through the | |
3191 | * subtree so that descendants don't have unavailable controllers enabled. | |
3192 | */ | |
3193 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3194 | { | |
3195 | struct cgroup *dsct; | |
3196 | struct cgroup_subsys_state *d_css; | |
3197 | ||
3198 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3199 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3200 | dsct->subtree_ss_mask = |
3201 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3202 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3203 | } |
3204 | } | |
3205 | ||
3206 | /** | |
3207 | * cgroup_restore_control - restore control masks of a subtree | |
3208 | * @cgrp: root of the target subtree | |
3209 | * | |
3210 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3211 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3212 | */ | |
3213 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3214 | { | |
3215 | struct cgroup *dsct; | |
3216 | struct cgroup_subsys_state *d_css; | |
3217 | ||
3218 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3219 | dsct->subtree_control = dsct->old_subtree_control; | |
3220 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3221 | } | |
3222 | } | |
3223 | ||
f6d635ad TH |
3224 | static bool css_visible(struct cgroup_subsys_state *css) |
3225 | { | |
3226 | struct cgroup_subsys *ss = css->ss; | |
3227 | struct cgroup *cgrp = css->cgroup; | |
3228 | ||
3229 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3230 | return true; | |
3231 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3232 | return false; | |
3233 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3234 | } | |
3235 | ||
bdb53bd7 TH |
3236 | /** |
3237 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3238 | * @cgrp: root of the target subtree |
bdb53bd7 | 3239 | * |
ce3f1d9d | 3240 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3241 | * visible. A css is created invisible if it's being implicitly enabled |
3242 | * through dependency. An invisible css is made visible when the userland | |
3243 | * explicitly enables it. | |
3244 | * | |
3245 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3246 | * been processed already aren't cleaned up. The caller is responsible for | |
3247 | * cleaning up with cgroup_apply_control_disble(). | |
3248 | */ | |
3249 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3250 | { | |
3251 | struct cgroup *dsct; | |
ce3f1d9d | 3252 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3253 | struct cgroup_subsys *ss; |
3254 | int ssid, ret; | |
3255 | ||
ce3f1d9d | 3256 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3257 | for_each_subsys(ss, ssid) { |
3258 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3259 | ||
945ba199 TH |
3260 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3261 | ||
bdb53bd7 TH |
3262 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3263 | continue; | |
3264 | ||
3265 | if (!css) { | |
3266 | css = css_create(dsct, ss); | |
3267 | if (IS_ERR(css)) | |
3268 | return PTR_ERR(css); | |
3269 | } | |
3270 | ||
f6d635ad | 3271 | if (css_visible(css)) { |
334c3679 | 3272 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3273 | if (ret) |
3274 | return ret; | |
3275 | } | |
3276 | } | |
3277 | } | |
3278 | ||
3279 | return 0; | |
3280 | } | |
3281 | ||
12b3bb6a TH |
3282 | /** |
3283 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3284 | * @cgrp: root of the target subtree |
12b3bb6a | 3285 | * |
ce3f1d9d | 3286 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3287 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3288 | * | |
3289 | * A css is hidden when the userland requests it to be disabled while other | |
3290 | * subsystems are still depending on it. The css must not actively control | |
3291 | * resources and be in the vanilla state if it's made visible again later. | |
3292 | * Controllers which may be depended upon should provide ->css_reset() for | |
3293 | * this purpose. | |
3294 | */ | |
3295 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3296 | { | |
3297 | struct cgroup *dsct; | |
ce3f1d9d | 3298 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3299 | struct cgroup_subsys *ss; |
3300 | int ssid; | |
3301 | ||
ce3f1d9d | 3302 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3303 | for_each_subsys(ss, ssid) { |
3304 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3305 | ||
945ba199 TH |
3306 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3307 | ||
12b3bb6a TH |
3308 | if (!css) |
3309 | continue; | |
3310 | ||
334c3679 TH |
3311 | if (css->parent && |
3312 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3313 | kill_css(css); |
f6d635ad | 3314 | } else if (!css_visible(css)) { |
334c3679 | 3315 | css_clear_dir(css); |
12b3bb6a TH |
3316 | if (ss->css_reset) |
3317 | ss->css_reset(css); | |
3318 | } | |
3319 | } | |
3320 | } | |
3321 | } | |
3322 | ||
f7b2814b TH |
3323 | /** |
3324 | * cgroup_apply_control - apply control mask updates to the subtree | |
3325 | * @cgrp: root of the target subtree | |
3326 | * | |
3327 | * subsystems can be enabled and disabled in a subtree using the following | |
3328 | * steps. | |
3329 | * | |
3330 | * 1. Call cgroup_save_control() to stash the current state. | |
3331 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3332 | * 3. Call cgroup_apply_control() to apply the changes. | |
3333 | * 4. Optionally perform other related operations. | |
3334 | * 5. Call cgroup_finalize_control() to finish up. | |
3335 | * | |
3336 | * This function implements step 3 and propagates the mask changes | |
3337 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3338 | * process migrations. | |
3339 | */ | |
3340 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3341 | { | |
3342 | int ret; | |
3343 | ||
3344 | cgroup_propagate_control(cgrp); | |
3345 | ||
3346 | ret = cgroup_apply_control_enable(cgrp); | |
3347 | if (ret) | |
3348 | return ret; | |
3349 | ||
3350 | /* | |
3351 | * At this point, cgroup_e_css() results reflect the new csses | |
3352 | * making the following cgroup_update_dfl_csses() properly update | |
3353 | * css associations of all tasks in the subtree. | |
3354 | */ | |
3355 | ret = cgroup_update_dfl_csses(cgrp); | |
3356 | if (ret) | |
3357 | return ret; | |
3358 | ||
3359 | return 0; | |
3360 | } | |
3361 | ||
3362 | /** | |
3363 | * cgroup_finalize_control - finalize control mask update | |
3364 | * @cgrp: root of the target subtree | |
3365 | * @ret: the result of the update | |
3366 | * | |
3367 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3368 | */ | |
3369 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3370 | { | |
3371 | if (ret) { | |
3372 | cgroup_restore_control(cgrp); | |
3373 | cgroup_propagate_control(cgrp); | |
3374 | } | |
3375 | ||
3376 | cgroup_apply_control_disable(cgrp); | |
3377 | } | |
3378 | ||
f8f22e53 | 3379 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3380 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3381 | char *buf, size_t nbytes, | |
3382 | loff_t off) | |
f8f22e53 | 3383 | { |
6e5c8307 | 3384 | u16 enable = 0, disable = 0; |
a9746d8d | 3385 | struct cgroup *cgrp, *child; |
f8f22e53 | 3386 | struct cgroup_subsys *ss; |
451af504 | 3387 | char *tok; |
f8f22e53 TH |
3388 | int ssid, ret; |
3389 | ||
3390 | /* | |
d37167ab TH |
3391 | * Parse input - space separated list of subsystem names prefixed |
3392 | * with either + or -. | |
f8f22e53 | 3393 | */ |
451af504 TH |
3394 | buf = strstrip(buf); |
3395 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3396 | if (tok[0] == '\0') |
3397 | continue; | |
a7165264 | 3398 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3399 | if (!cgroup_ssid_enabled(ssid) || |
3400 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3401 | continue; |
3402 | ||
3403 | if (*tok == '+') { | |
7d331fa9 TH |
3404 | enable |= 1 << ssid; |
3405 | disable &= ~(1 << ssid); | |
f8f22e53 | 3406 | } else if (*tok == '-') { |
7d331fa9 TH |
3407 | disable |= 1 << ssid; |
3408 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3409 | } else { |
3410 | return -EINVAL; | |
3411 | } | |
3412 | break; | |
b4e0eeaf | 3413 | } while_each_subsys_mask(); |
f8f22e53 TH |
3414 | if (ssid == CGROUP_SUBSYS_COUNT) |
3415 | return -EINVAL; | |
3416 | } | |
3417 | ||
945ba199 | 3418 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3419 | if (!cgrp) |
3420 | return -ENODEV; | |
f8f22e53 TH |
3421 | |
3422 | for_each_subsys(ss, ssid) { | |
3423 | if (enable & (1 << ssid)) { | |
667c2491 | 3424 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3425 | enable &= ~(1 << ssid); |
3426 | continue; | |
3427 | } | |
3428 | ||
5531dc91 | 3429 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3430 | ret = -ENOENT; |
3431 | goto out_unlock; | |
3432 | } | |
f8f22e53 | 3433 | } else if (disable & (1 << ssid)) { |
667c2491 | 3434 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3435 | disable &= ~(1 << ssid); |
3436 | continue; | |
3437 | } | |
3438 | ||
3439 | /* a child has it enabled? */ | |
3440 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3441 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3442 | ret = -EBUSY; |
ddab2b6e | 3443 | goto out_unlock; |
f8f22e53 TH |
3444 | } |
3445 | } | |
3446 | } | |
3447 | } | |
3448 | ||
3449 | if (!enable && !disable) { | |
3450 | ret = 0; | |
ddab2b6e | 3451 | goto out_unlock; |
f8f22e53 TH |
3452 | } |
3453 | ||
3454 | /* | |
667c2491 | 3455 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3456 | * with tasks so that child cgroups don't compete against tasks. |
3457 | */ | |
9157056d TH |
3458 | if (enable && cgroup_parent(cgrp)) { |
3459 | struct cgrp_cset_link *link; | |
3460 | ||
3461 | /* | |
3462 | * Because namespaces pin csets too, @cgrp->cset_links | |
3463 | * might not be empty even when @cgrp is empty. Walk and | |
3464 | * verify each cset. | |
3465 | */ | |
3466 | spin_lock_irq(&css_set_lock); | |
3467 | ||
3468 | ret = 0; | |
3469 | list_for_each_entry(link, &cgrp->cset_links, cset_link) { | |
3470 | if (css_set_populated(link->cset)) { | |
3471 | ret = -EBUSY; | |
3472 | break; | |
3473 | } | |
3474 | } | |
3475 | ||
3476 | spin_unlock_irq(&css_set_lock); | |
3477 | ||
3478 | if (ret) | |
3479 | goto out_unlock; | |
f8f22e53 TH |
3480 | } |
3481 | ||
15a27c36 TH |
3482 | /* save and update control masks and prepare csses */ |
3483 | cgroup_save_control(cgrp); | |
f63070d3 | 3484 | |
15a27c36 TH |
3485 | cgrp->subtree_control |= enable; |
3486 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3487 | |
f7b2814b | 3488 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3489 | |
f7b2814b | 3490 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3491 | |
3492 | kernfs_activate(cgrp->kn); | |
3493 | ret = 0; | |
3494 | out_unlock: | |
a9746d8d | 3495 | cgroup_kn_unlock(of->kn); |
451af504 | 3496 | return ret ?: nbytes; |
f8f22e53 TH |
3497 | } |
3498 | ||
4a07c222 | 3499 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3500 | { |
4a07c222 | 3501 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3502 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3503 | return 0; |
3504 | } | |
3505 | ||
2bd59d48 TH |
3506 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3507 | size_t nbytes, loff_t off) | |
355e0c48 | 3508 | { |
2bd59d48 TH |
3509 | struct cgroup *cgrp = of->kn->parent->priv; |
3510 | struct cftype *cft = of->kn->priv; | |
3511 | struct cgroup_subsys_state *css; | |
a742c59d | 3512 | int ret; |
355e0c48 | 3513 | |
b4168640 TH |
3514 | if (cft->write) |
3515 | return cft->write(of, buf, nbytes, off); | |
3516 | ||
2bd59d48 TH |
3517 | /* |
3518 | * kernfs guarantees that a file isn't deleted with operations in | |
3519 | * flight, which means that the matching css is and stays alive and | |
3520 | * doesn't need to be pinned. The RCU locking is not necessary | |
3521 | * either. It's just for the convenience of using cgroup_css(). | |
3522 | */ | |
3523 | rcu_read_lock(); | |
3524 | css = cgroup_css(cgrp, cft->ss); | |
3525 | rcu_read_unlock(); | |
a742c59d | 3526 | |
451af504 | 3527 | if (cft->write_u64) { |
a742c59d TH |
3528 | unsigned long long v; |
3529 | ret = kstrtoull(buf, 0, &v); | |
3530 | if (!ret) | |
3531 | ret = cft->write_u64(css, cft, v); | |
3532 | } else if (cft->write_s64) { | |
3533 | long long v; | |
3534 | ret = kstrtoll(buf, 0, &v); | |
3535 | if (!ret) | |
3536 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3537 | } else { |
a742c59d | 3538 | ret = -EINVAL; |
e73d2c61 | 3539 | } |
2bd59d48 | 3540 | |
a742c59d | 3541 | return ret ?: nbytes; |
355e0c48 PM |
3542 | } |
3543 | ||
6612f05b | 3544 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3545 | { |
2bd59d48 | 3546 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3547 | } |
3548 | ||
6612f05b | 3549 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3550 | { |
2bd59d48 | 3551 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3552 | } |
3553 | ||
6612f05b | 3554 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3555 | { |
2bd59d48 | 3556 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3557 | } |
3558 | ||
91796569 | 3559 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3560 | { |
7da11279 TH |
3561 | struct cftype *cft = seq_cft(m); |
3562 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3563 | |
2da8ca82 TH |
3564 | if (cft->seq_show) |
3565 | return cft->seq_show(m, arg); | |
e73d2c61 | 3566 | |
f4c753b7 | 3567 | if (cft->read_u64) |
896f5199 TH |
3568 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3569 | else if (cft->read_s64) | |
3570 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3571 | else | |
3572 | return -EINVAL; | |
3573 | return 0; | |
91796569 PM |
3574 | } |
3575 | ||
2bd59d48 TH |
3576 | static struct kernfs_ops cgroup_kf_single_ops = { |
3577 | .atomic_write_len = PAGE_SIZE, | |
3578 | .write = cgroup_file_write, | |
3579 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3580 | }; |
3581 | ||
2bd59d48 TH |
3582 | static struct kernfs_ops cgroup_kf_ops = { |
3583 | .atomic_write_len = PAGE_SIZE, | |
3584 | .write = cgroup_file_write, | |
3585 | .seq_start = cgroup_seqfile_start, | |
3586 | .seq_next = cgroup_seqfile_next, | |
3587 | .seq_stop = cgroup_seqfile_stop, | |
3588 | .seq_show = cgroup_seqfile_show, | |
3589 | }; | |
ddbcc7e8 PM |
3590 | |
3591 | /* | |
3592 | * cgroup_rename - Only allow simple rename of directories in place. | |
3593 | */ | |
2bd59d48 TH |
3594 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3595 | const char *new_name_str) | |
ddbcc7e8 | 3596 | { |
2bd59d48 | 3597 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3598 | int ret; |
65dff759 | 3599 | |
2bd59d48 | 3600 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3601 | return -ENOTDIR; |
2bd59d48 | 3602 | if (kn->parent != new_parent) |
ddbcc7e8 | 3603 | return -EIO; |
65dff759 | 3604 | |
6db8e85c TH |
3605 | /* |
3606 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3607 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3608 | */ |
aa6ec29b | 3609 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3610 | return -EPERM; |
099fca32 | 3611 | |
e1b2dc17 | 3612 | /* |
8353da1f | 3613 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3614 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3615 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3616 | */ |
3617 | kernfs_break_active_protection(new_parent); | |
3618 | kernfs_break_active_protection(kn); | |
099fca32 | 3619 | |
2bd59d48 | 3620 | mutex_lock(&cgroup_mutex); |
099fca32 | 3621 | |
2bd59d48 | 3622 | ret = kernfs_rename(kn, new_parent, new_name_str); |
ed1777de TH |
3623 | if (!ret) |
3624 | trace_cgroup_rename(cgrp); | |
099fca32 | 3625 | |
2bd59d48 | 3626 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3627 | |
3628 | kernfs_unbreak_active_protection(kn); | |
3629 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3630 | return ret; |
099fca32 LZ |
3631 | } |
3632 | ||
49957f8e TH |
3633 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3634 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3635 | { | |
3636 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3637 | .ia_uid = current_fsuid(), | |
3638 | .ia_gid = current_fsgid(), }; | |
3639 | ||
3640 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3641 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3642 | return 0; | |
3643 | ||
3644 | return kernfs_setattr(kn, &iattr); | |
3645 | } | |
3646 | ||
4df8dc90 TH |
3647 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3648 | struct cftype *cft) | |
ddbcc7e8 | 3649 | { |
8d7e6fb0 | 3650 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3651 | struct kernfs_node *kn; |
3652 | struct lock_class_key *key = NULL; | |
49957f8e | 3653 | int ret; |
05ef1d7c | 3654 | |
2bd59d48 TH |
3655 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3656 | key = &cft->lockdep_key; | |
3657 | #endif | |
3658 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3659 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3660 | NULL, key); |
49957f8e TH |
3661 | if (IS_ERR(kn)) |
3662 | return PTR_ERR(kn); | |
3663 | ||
3664 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3665 | if (ret) { |
49957f8e | 3666 | kernfs_remove(kn); |
f8f22e53 TH |
3667 | return ret; |
3668 | } | |
3669 | ||
6f60eade TH |
3670 | if (cft->file_offset) { |
3671 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3672 | ||
34c06254 | 3673 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3674 | cfile->kn = kn; |
34c06254 | 3675 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3676 | } |
3677 | ||
f8f22e53 | 3678 | return 0; |
ddbcc7e8 PM |
3679 | } |
3680 | ||
b1f28d31 TH |
3681 | /** |
3682 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3683 | * @css: the target css |
3684 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3685 | * @cfts: array of cftypes to be added |
3686 | * @is_add: whether to add or remove | |
3687 | * | |
3688 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3689 | * For removals, this function never fails. |
b1f28d31 | 3690 | */ |
4df8dc90 TH |
3691 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3692 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3693 | bool is_add) |
ddbcc7e8 | 3694 | { |
6732ed85 | 3695 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3696 | int ret = 0; |
b1f28d31 | 3697 | |
01f6474c | 3698 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3699 | |
6732ed85 TH |
3700 | restart: |
3701 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3702 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3703 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3704 | continue; |
05ebb6e6 | 3705 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3706 | continue; |
d51f39b0 | 3707 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3708 | continue; |
d51f39b0 | 3709 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3710 | continue; |
3711 | ||
2739d3cc | 3712 | if (is_add) { |
4df8dc90 | 3713 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3714 | if (ret) { |
ed3d261b JP |
3715 | pr_warn("%s: failed to add %s, err=%d\n", |
3716 | __func__, cft->name, ret); | |
6732ed85 TH |
3717 | cft_end = cft; |
3718 | is_add = false; | |
3719 | goto restart; | |
b1f28d31 | 3720 | } |
2739d3cc LZ |
3721 | } else { |
3722 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3723 | } |
ddbcc7e8 | 3724 | } |
b598dde3 | 3725 | return ret; |
ddbcc7e8 PM |
3726 | } |
3727 | ||
21a2d343 | 3728 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3729 | { |
3730 | LIST_HEAD(pending); | |
2bb566cb | 3731 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3732 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3733 | struct cgroup_subsys_state *css; |
9ccece80 | 3734 | int ret = 0; |
8e3f6541 | 3735 | |
01f6474c | 3736 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3737 | |
e8c82d20 | 3738 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3739 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3740 | struct cgroup *cgrp = css->cgroup; |
3741 | ||
88cb04b9 | 3742 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3743 | continue; |
3744 | ||
4df8dc90 | 3745 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3746 | if (ret) |
3747 | break; | |
8e3f6541 | 3748 | } |
21a2d343 TH |
3749 | |
3750 | if (is_add && !ret) | |
3751 | kernfs_activate(root->kn); | |
9ccece80 | 3752 | return ret; |
8e3f6541 TH |
3753 | } |
3754 | ||
2da440a2 | 3755 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3756 | { |
2bb566cb | 3757 | struct cftype *cft; |
8e3f6541 | 3758 | |
2bd59d48 TH |
3759 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3760 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3761 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3762 | kfree(cft->kf_ops); | |
3763 | cft->kf_ops = NULL; | |
2da440a2 | 3764 | cft->ss = NULL; |
a8ddc821 TH |
3765 | |
3766 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3767 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3768 | } |
2da440a2 TH |
3769 | } |
3770 | ||
2bd59d48 | 3771 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3772 | { |
3773 | struct cftype *cft; | |
3774 | ||
2bd59d48 TH |
3775 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3776 | struct kernfs_ops *kf_ops; | |
3777 | ||
0adb0704 TH |
3778 | WARN_ON(cft->ss || cft->kf_ops); |
3779 | ||
2bd59d48 TH |
3780 | if (cft->seq_start) |
3781 | kf_ops = &cgroup_kf_ops; | |
3782 | else | |
3783 | kf_ops = &cgroup_kf_single_ops; | |
3784 | ||
3785 | /* | |
3786 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3787 | * make a copy of kf_ops to set its atomic_write_len. | |
3788 | */ | |
3789 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3790 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3791 | if (!kf_ops) { | |
3792 | cgroup_exit_cftypes(cfts); | |
3793 | return -ENOMEM; | |
3794 | } | |
3795 | kf_ops->atomic_write_len = cft->max_write_len; | |
3796 | } | |
8e3f6541 | 3797 | |
2bd59d48 | 3798 | cft->kf_ops = kf_ops; |
2bb566cb | 3799 | cft->ss = ss; |
2bd59d48 | 3800 | } |
2bb566cb | 3801 | |
2bd59d48 | 3802 | return 0; |
2da440a2 TH |
3803 | } |
3804 | ||
21a2d343 TH |
3805 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3806 | { | |
01f6474c | 3807 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3808 | |
3809 | if (!cfts || !cfts[0].ss) | |
3810 | return -ENOENT; | |
3811 | ||
3812 | list_del(&cfts->node); | |
3813 | cgroup_apply_cftypes(cfts, false); | |
3814 | cgroup_exit_cftypes(cfts); | |
3815 | return 0; | |
8e3f6541 | 3816 | } |
8e3f6541 | 3817 | |
79578621 TH |
3818 | /** |
3819 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3820 | * @cfts: zero-length name terminated array of cftypes |
3821 | * | |
2bb566cb TH |
3822 | * Unregister @cfts. Files described by @cfts are removed from all |
3823 | * existing cgroups and all future cgroups won't have them either. This | |
3824 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3825 | * |
3826 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3827 | * registered. |
79578621 | 3828 | */ |
2bb566cb | 3829 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3830 | { |
21a2d343 | 3831 | int ret; |
79578621 | 3832 | |
01f6474c | 3833 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3834 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3835 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3836 | return ret; |
80b13586 TH |
3837 | } |
3838 | ||
8e3f6541 TH |
3839 | /** |
3840 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3841 | * @ss: target cgroup subsystem | |
3842 | * @cfts: zero-length name terminated array of cftypes | |
3843 | * | |
3844 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3845 | * existing cgroups to which @ss is attached and all future cgroups will | |
3846 | * have them too. This function can be called anytime whether @ss is | |
3847 | * attached or not. | |
3848 | * | |
3849 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3850 | * function currently returns 0 as long as @cfts registration is successful | |
3851 | * even if some file creation attempts on existing cgroups fail. | |
3852 | */ | |
2cf669a5 | 3853 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3854 | { |
9ccece80 | 3855 | int ret; |
8e3f6541 | 3856 | |
fc5ed1e9 | 3857 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3858 | return 0; |
3859 | ||
dc5736ed LZ |
3860 | if (!cfts || cfts[0].name[0] == '\0') |
3861 | return 0; | |
2bb566cb | 3862 | |
2bd59d48 TH |
3863 | ret = cgroup_init_cftypes(ss, cfts); |
3864 | if (ret) | |
3865 | return ret; | |
79578621 | 3866 | |
01f6474c | 3867 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3868 | |
0adb0704 | 3869 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3870 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3871 | if (ret) |
21a2d343 | 3872 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3873 | |
01f6474c | 3874 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3875 | return ret; |
79578621 TH |
3876 | } |
3877 | ||
a8ddc821 TH |
3878 | /** |
3879 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3880 | * @ss: target cgroup subsystem | |
3881 | * @cfts: zero-length name terminated array of cftypes | |
3882 | * | |
3883 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3884 | * the default hierarchy. | |
3885 | */ | |
3886 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3887 | { | |
3888 | struct cftype *cft; | |
3889 | ||
3890 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3891 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3892 | return cgroup_add_cftypes(ss, cfts); |
3893 | } | |
3894 | ||
3895 | /** | |
3896 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3897 | * @ss: target cgroup subsystem | |
3898 | * @cfts: zero-length name terminated array of cftypes | |
3899 | * | |
3900 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3901 | * the legacy hierarchies. | |
3902 | */ | |
2cf669a5 TH |
3903 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3904 | { | |
a8ddc821 TH |
3905 | struct cftype *cft; |
3906 | ||
e4b7037c TH |
3907 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3908 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3909 | return cgroup_add_cftypes(ss, cfts); |
3910 | } | |
3911 | ||
34c06254 TH |
3912 | /** |
3913 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3914 | * @cfile: target cgroup_file | |
3915 | * | |
3916 | * @cfile must have been obtained by setting cftype->file_offset. | |
3917 | */ | |
3918 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3919 | { | |
3920 | unsigned long flags; | |
3921 | ||
3922 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3923 | if (cfile->kn) | |
3924 | kernfs_notify(cfile->kn); | |
3925 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3926 | } | |
3927 | ||
a043e3b2 LZ |
3928 | /** |
3929 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3930 | * @cgrp: the cgroup in question | |
3931 | * | |
9157056d TH |
3932 | * Return the number of tasks in the cgroup. The returned number can be |
3933 | * higher than the actual number of tasks due to css_set references from | |
3934 | * namespace roots and temporary usages. | |
a043e3b2 | 3935 | */ |
07bc356e | 3936 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3937 | { |
3938 | int count = 0; | |
69d0206c | 3939 | struct cgrp_cset_link *link; |
817929ec | 3940 | |
82d6489d | 3941 | spin_lock_irq(&css_set_lock); |
69d0206c TH |
3942 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3943 | count += atomic_read(&link->cset->refcount); | |
82d6489d | 3944 | spin_unlock_irq(&css_set_lock); |
bbcb81d0 PM |
3945 | return count; |
3946 | } | |
3947 | ||
53fa5261 | 3948 | /** |
492eb21b | 3949 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3950 | * @pos: the current position (%NULL to initiate traversal) |
3951 | * @parent: css whose children to walk | |
53fa5261 | 3952 | * |
c2931b70 | 3953 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3954 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3955 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3956 | * be returned regardless of their states. | |
3957 | * | |
3958 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3959 | * css which finished ->css_online() is guaranteed to be visible in the | |
3960 | * future iterations and will stay visible until the last reference is put. | |
3961 | * A css which hasn't finished ->css_online() or already finished | |
3962 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3963 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3964 | */ |
c2931b70 TH |
3965 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3966 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3967 | { |
c2931b70 | 3968 | struct cgroup_subsys_state *next; |
53fa5261 | 3969 | |
8353da1f | 3970 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3971 | |
3972 | /* | |
de3f0341 TH |
3973 | * @pos could already have been unlinked from the sibling list. |
3974 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3975 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3976 | * @pos is taken off list, at which time its next pointer is valid, | |
3977 | * and, as releases are serialized, the one pointed to by the next | |
3978 | * pointer is guaranteed to not have started release yet. This | |
3979 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3980 | * critical section, the one pointed to by its next pointer is | |
3981 | * guaranteed to not have finished its RCU grace period even if we | |
3982 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3983 | * |
de3f0341 TH |
3984 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3985 | * dereferenced; however, as each css is given a monotonically | |
3986 | * increasing unique serial number and always appended to the | |
3987 | * sibling list, the next one can be found by walking the parent's | |
3988 | * children until the first css with higher serial number than | |
3989 | * @pos's. While this path can be slower, it happens iff iteration | |
3990 | * races against release and the race window is very small. | |
53fa5261 | 3991 | */ |
3b287a50 | 3992 | if (!pos) { |
c2931b70 TH |
3993 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3994 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3995 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3996 | } else { |
c2931b70 | 3997 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3998 | if (next->serial_nr > pos->serial_nr) |
3999 | break; | |
53fa5261 TH |
4000 | } |
4001 | ||
3b281afb TH |
4002 | /* |
4003 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 4004 | * the next sibling. |
3b281afb | 4005 | */ |
c2931b70 TH |
4006 | if (&next->sibling != &parent->children) |
4007 | return next; | |
3b281afb | 4008 | return NULL; |
53fa5261 | 4009 | } |
53fa5261 | 4010 | |
574bd9f7 | 4011 | /** |
492eb21b | 4012 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 4013 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4014 | * @root: css whose descendants to walk |
574bd9f7 | 4015 | * |
492eb21b | 4016 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
4017 | * to visit for pre-order traversal of @root's descendants. @root is |
4018 | * included in the iteration and the first node to be visited. | |
75501a6d | 4019 | * |
87fb54f1 TH |
4020 | * While this function requires cgroup_mutex or RCU read locking, it |
4021 | * doesn't require the whole traversal to be contained in a single critical | |
4022 | * section. This function will return the correct next descendant as long | |
4023 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
4024 | * |
4025 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4026 | * css which finished ->css_online() is guaranteed to be visible in the | |
4027 | * future iterations and will stay visible until the last reference is put. | |
4028 | * A css which hasn't finished ->css_online() or already finished | |
4029 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4030 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4031 | */ |
492eb21b TH |
4032 | struct cgroup_subsys_state * |
4033 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
4034 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4035 | { |
492eb21b | 4036 | struct cgroup_subsys_state *next; |
574bd9f7 | 4037 | |
8353da1f | 4038 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4039 | |
bd8815a6 | 4040 | /* if first iteration, visit @root */ |
7805d000 | 4041 | if (!pos) |
bd8815a6 | 4042 | return root; |
574bd9f7 TH |
4043 | |
4044 | /* visit the first child if exists */ | |
492eb21b | 4045 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4046 | if (next) |
4047 | return next; | |
4048 | ||
4049 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4050 | while (pos != root) { |
5c9d535b | 4051 | next = css_next_child(pos, pos->parent); |
75501a6d | 4052 | if (next) |
574bd9f7 | 4053 | return next; |
5c9d535b | 4054 | pos = pos->parent; |
7805d000 | 4055 | } |
574bd9f7 TH |
4056 | |
4057 | return NULL; | |
4058 | } | |
574bd9f7 | 4059 | |
12a9d2fe | 4060 | /** |
492eb21b TH |
4061 | * css_rightmost_descendant - return the rightmost descendant of a css |
4062 | * @pos: css of interest | |
12a9d2fe | 4063 | * |
492eb21b TH |
4064 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4065 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4066 | * subtree of @pos. |
75501a6d | 4067 | * |
87fb54f1 TH |
4068 | * While this function requires cgroup_mutex or RCU read locking, it |
4069 | * doesn't require the whole traversal to be contained in a single critical | |
4070 | * section. This function will return the correct rightmost descendant as | |
4071 | * long as @pos is accessible. | |
12a9d2fe | 4072 | */ |
492eb21b TH |
4073 | struct cgroup_subsys_state * |
4074 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4075 | { |
492eb21b | 4076 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4077 | |
8353da1f | 4078 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4079 | |
4080 | do { | |
4081 | last = pos; | |
4082 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4083 | pos = NULL; | |
492eb21b | 4084 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4085 | pos = tmp; |
4086 | } while (pos); | |
4087 | ||
4088 | return last; | |
4089 | } | |
12a9d2fe | 4090 | |
492eb21b TH |
4091 | static struct cgroup_subsys_state * |
4092 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4093 | { |
492eb21b | 4094 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4095 | |
4096 | do { | |
4097 | last = pos; | |
492eb21b | 4098 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4099 | } while (pos); |
4100 | ||
4101 | return last; | |
4102 | } | |
4103 | ||
4104 | /** | |
492eb21b | 4105 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4106 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4107 | * @root: css whose descendants to walk |
574bd9f7 | 4108 | * |
492eb21b | 4109 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4110 | * to visit for post-order traversal of @root's descendants. @root is |
4111 | * included in the iteration and the last node to be visited. | |
75501a6d | 4112 | * |
87fb54f1 TH |
4113 | * While this function requires cgroup_mutex or RCU read locking, it |
4114 | * doesn't require the whole traversal to be contained in a single critical | |
4115 | * section. This function will return the correct next descendant as long | |
4116 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4117 | * @cgroup. | |
c2931b70 TH |
4118 | * |
4119 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4120 | * css which finished ->css_online() is guaranteed to be visible in the | |
4121 | * future iterations and will stay visible until the last reference is put. | |
4122 | * A css which hasn't finished ->css_online() or already finished | |
4123 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4124 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4125 | */ |
492eb21b TH |
4126 | struct cgroup_subsys_state * |
4127 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4128 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4129 | { |
492eb21b | 4130 | struct cgroup_subsys_state *next; |
574bd9f7 | 4131 | |
8353da1f | 4132 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4133 | |
58b79a91 TH |
4134 | /* if first iteration, visit leftmost descendant which may be @root */ |
4135 | if (!pos) | |
4136 | return css_leftmost_descendant(root); | |
574bd9f7 | 4137 | |
bd8815a6 TH |
4138 | /* if we visited @root, we're done */ |
4139 | if (pos == root) | |
4140 | return NULL; | |
4141 | ||
574bd9f7 | 4142 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4143 | next = css_next_child(pos, pos->parent); |
75501a6d | 4144 | if (next) |
492eb21b | 4145 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4146 | |
4147 | /* no sibling left, visit parent */ | |
5c9d535b | 4148 | return pos->parent; |
574bd9f7 | 4149 | } |
574bd9f7 | 4150 | |
f3d46500 TH |
4151 | /** |
4152 | * css_has_online_children - does a css have online children | |
4153 | * @css: the target css | |
4154 | * | |
4155 | * Returns %true if @css has any online children; otherwise, %false. This | |
4156 | * function can be called from any context but the caller is responsible | |
4157 | * for synchronizing against on/offlining as necessary. | |
4158 | */ | |
4159 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4160 | { |
f3d46500 TH |
4161 | struct cgroup_subsys_state *child; |
4162 | bool ret = false; | |
cbc125ef TH |
4163 | |
4164 | rcu_read_lock(); | |
f3d46500 | 4165 | css_for_each_child(child, css) { |
99bae5f9 | 4166 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4167 | ret = true; |
4168 | break; | |
cbc125ef TH |
4169 | } |
4170 | } | |
4171 | rcu_read_unlock(); | |
f3d46500 | 4172 | return ret; |
574bd9f7 | 4173 | } |
574bd9f7 | 4174 | |
0942eeee | 4175 | /** |
ecb9d535 | 4176 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4177 | * @it: the iterator to advance |
4178 | * | |
4179 | * Advance @it to the next css_set to walk. | |
d515876e | 4180 | */ |
ecb9d535 | 4181 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4182 | { |
0f0a2b4f | 4183 | struct list_head *l = it->cset_pos; |
d515876e TH |
4184 | struct cgrp_cset_link *link; |
4185 | struct css_set *cset; | |
4186 | ||
f0d9a5f1 | 4187 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4188 | |
d515876e TH |
4189 | /* Advance to the next non-empty css_set */ |
4190 | do { | |
4191 | l = l->next; | |
0f0a2b4f TH |
4192 | if (l == it->cset_head) { |
4193 | it->cset_pos = NULL; | |
ecb9d535 | 4194 | it->task_pos = NULL; |
d515876e TH |
4195 | return; |
4196 | } | |
3ebb2b6e TH |
4197 | |
4198 | if (it->ss) { | |
4199 | cset = container_of(l, struct css_set, | |
4200 | e_cset_node[it->ss->id]); | |
4201 | } else { | |
4202 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4203 | cset = link->cset; | |
4204 | } | |
0de0942d | 4205 | } while (!css_set_populated(cset)); |
c7561128 | 4206 | |
0f0a2b4f | 4207 | it->cset_pos = l; |
c7561128 TH |
4208 | |
4209 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 4210 | it->task_pos = cset->tasks.next; |
c7561128 | 4211 | else |
0f0a2b4f TH |
4212 | it->task_pos = cset->mg_tasks.next; |
4213 | ||
4214 | it->tasks_head = &cset->tasks; | |
4215 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4216 | |
4217 | /* | |
4218 | * We don't keep css_sets locked across iteration steps and thus | |
4219 | * need to take steps to ensure that iteration can be resumed after | |
4220 | * the lock is re-acquired. Iteration is performed at two levels - | |
4221 | * css_sets and tasks in them. | |
4222 | * | |
4223 | * Once created, a css_set never leaves its cgroup lists, so a | |
4224 | * pinned css_set is guaranteed to stay put and we can resume | |
4225 | * iteration afterwards. | |
4226 | * | |
4227 | * Tasks may leave @cset across iteration steps. This is resolved | |
4228 | * by registering each iterator with the css_set currently being | |
4229 | * walked and making css_set_move_task() advance iterators whose | |
4230 | * next task is leaving. | |
4231 | */ | |
4232 | if (it->cur_cset) { | |
4233 | list_del(&it->iters_node); | |
4234 | put_css_set_locked(it->cur_cset); | |
4235 | } | |
4236 | get_css_set(cset); | |
4237 | it->cur_cset = cset; | |
4238 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4239 | } |
4240 | ||
ecb9d535 TH |
4241 | static void css_task_iter_advance(struct css_task_iter *it) |
4242 | { | |
4243 | struct list_head *l = it->task_pos; | |
4244 | ||
f0d9a5f1 | 4245 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4246 | WARN_ON_ONCE(!l); |
4247 | ||
4248 | /* | |
4249 | * Advance iterator to find next entry. cset->tasks is consumed | |
4250 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4251 | * next cset. | |
4252 | */ | |
4253 | l = l->next; | |
4254 | ||
4255 | if (l == it->tasks_head) | |
4256 | l = it->mg_tasks_head->next; | |
4257 | ||
4258 | if (l == it->mg_tasks_head) | |
4259 | css_task_iter_advance_css_set(it); | |
4260 | else | |
4261 | it->task_pos = l; | |
4262 | } | |
4263 | ||
0942eeee | 4264 | /** |
72ec7029 TH |
4265 | * css_task_iter_start - initiate task iteration |
4266 | * @css: the css to walk tasks of | |
0942eeee TH |
4267 | * @it: the task iterator to use |
4268 | * | |
72ec7029 TH |
4269 | * Initiate iteration through the tasks of @css. The caller can call |
4270 | * css_task_iter_next() to walk through the tasks until the function | |
4271 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4272 | * called. | |
0942eeee | 4273 | */ |
72ec7029 TH |
4274 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4275 | struct css_task_iter *it) | |
817929ec | 4276 | { |
56fde9e0 TH |
4277 | /* no one should try to iterate before mounting cgroups */ |
4278 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4279 | |
ed27b9f7 TH |
4280 | memset(it, 0, sizeof(*it)); |
4281 | ||
82d6489d | 4282 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4283 | |
3ebb2b6e TH |
4284 | it->ss = css->ss; |
4285 | ||
4286 | if (it->ss) | |
4287 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4288 | else | |
4289 | it->cset_pos = &css->cgroup->cset_links; | |
4290 | ||
0f0a2b4f | 4291 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4292 | |
ecb9d535 | 4293 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4294 | |
82d6489d | 4295 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4296 | } |
4297 | ||
0942eeee | 4298 | /** |
72ec7029 | 4299 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4300 | * @it: the task iterator being iterated |
4301 | * | |
4302 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4303 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4304 | * reaches the end. | |
0942eeee | 4305 | */ |
72ec7029 | 4306 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4307 | { |
d5745675 | 4308 | if (it->cur_task) { |
ed27b9f7 | 4309 | put_task_struct(it->cur_task); |
d5745675 TH |
4310 | it->cur_task = NULL; |
4311 | } | |
ed27b9f7 | 4312 | |
82d6489d | 4313 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4314 | |
d5745675 TH |
4315 | if (it->task_pos) { |
4316 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4317 | cg_list); | |
4318 | get_task_struct(it->cur_task); | |
4319 | css_task_iter_advance(it); | |
4320 | } | |
ed27b9f7 | 4321 | |
82d6489d | 4322 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4323 | |
4324 | return it->cur_task; | |
817929ec PM |
4325 | } |
4326 | ||
0942eeee | 4327 | /** |
72ec7029 | 4328 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4329 | * @it: the task iterator to finish |
4330 | * | |
72ec7029 | 4331 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4332 | */ |
72ec7029 | 4333 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4334 | { |
ed27b9f7 | 4335 | if (it->cur_cset) { |
82d6489d | 4336 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4337 | list_del(&it->iters_node); |
4338 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4339 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4340 | } |
4341 | ||
4342 | if (it->cur_task) | |
4343 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4344 | } |
4345 | ||
4346 | /** | |
8cc99345 TH |
4347 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4348 | * @to: cgroup to which the tasks will be moved | |
4349 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4350 | * |
eaf797ab TH |
4351 | * Locking rules between cgroup_post_fork() and the migration path |
4352 | * guarantee that, if a task is forking while being migrated, the new child | |
4353 | * is guaranteed to be either visible in the source cgroup after the | |
4354 | * parent's migration is complete or put into the target cgroup. No task | |
4355 | * can slip out of migration through forking. | |
31a7df01 | 4356 | */ |
8cc99345 | 4357 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4358 | { |
952aaa12 TH |
4359 | LIST_HEAD(preloaded_csets); |
4360 | struct cgrp_cset_link *link; | |
72ec7029 | 4361 | struct css_task_iter it; |
e406d1cf | 4362 | struct task_struct *task; |
952aaa12 | 4363 | int ret; |
31a7df01 | 4364 | |
6c694c88 TH |
4365 | if (!cgroup_may_migrate_to(to)) |
4366 | return -EBUSY; | |
4367 | ||
952aaa12 | 4368 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4369 | |
eedd0f4c EB |
4370 | percpu_down_write(&cgroup_threadgroup_rwsem); |
4371 | ||
952aaa12 | 4372 | /* all tasks in @from are being moved, all csets are source */ |
82d6489d | 4373 | spin_lock_irq(&css_set_lock); |
952aaa12 TH |
4374 | list_for_each_entry(link, &from->cset_links, cset_link) |
4375 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
82d6489d | 4376 | spin_unlock_irq(&css_set_lock); |
31a7df01 | 4377 | |
e4857982 | 4378 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
952aaa12 TH |
4379 | if (ret) |
4380 | goto out_err; | |
8cc99345 | 4381 | |
952aaa12 | 4382 | /* |
2cfa2b19 | 4383 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4384 | * ->can_attach() fails. |
4385 | */ | |
e406d1cf | 4386 | do { |
9d800df1 | 4387 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4388 | task = css_task_iter_next(&it); |
4389 | if (task) | |
4390 | get_task_struct(task); | |
4391 | css_task_iter_end(&it); | |
4392 | ||
4393 | if (task) { | |
37ff9f8f | 4394 | ret = cgroup_migrate(task, false, to->root); |
ed1777de TH |
4395 | if (!ret) |
4396 | trace_cgroup_transfer_tasks(to, task, false); | |
e406d1cf TH |
4397 | put_task_struct(task); |
4398 | } | |
4399 | } while (task && !ret); | |
952aaa12 TH |
4400 | out_err: |
4401 | cgroup_migrate_finish(&preloaded_csets); | |
eedd0f4c | 4402 | percpu_up_write(&cgroup_threadgroup_rwsem); |
47cfcd09 | 4403 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4404 | return ret; |
8cc99345 TH |
4405 | } |
4406 | ||
bbcb81d0 | 4407 | /* |
102a775e | 4408 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4409 | * |
4410 | * Reading this file can return large amounts of data if a cgroup has | |
4411 | * *lots* of attached tasks. So it may need several calls to read(), | |
4412 | * but we cannot guarantee that the information we produce is correct | |
4413 | * unless we produce it entirely atomically. | |
4414 | * | |
bbcb81d0 | 4415 | */ |
bbcb81d0 | 4416 | |
24528255 LZ |
4417 | /* which pidlist file are we talking about? */ |
4418 | enum cgroup_filetype { | |
4419 | CGROUP_FILE_PROCS, | |
4420 | CGROUP_FILE_TASKS, | |
4421 | }; | |
4422 | ||
4423 | /* | |
4424 | * A pidlist is a list of pids that virtually represents the contents of one | |
4425 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4426 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4427 | * to the cgroup. | |
4428 | */ | |
4429 | struct cgroup_pidlist { | |
4430 | /* | |
4431 | * used to find which pidlist is wanted. doesn't change as long as | |
4432 | * this particular list stays in the list. | |
4433 | */ | |
4434 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4435 | /* array of xids */ | |
4436 | pid_t *list; | |
4437 | /* how many elements the above list has */ | |
4438 | int length; | |
24528255 LZ |
4439 | /* each of these stored in a list by its cgroup */ |
4440 | struct list_head links; | |
4441 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4442 | struct cgroup *owner; | |
b1a21367 TH |
4443 | /* for delayed destruction */ |
4444 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4445 | }; |
4446 | ||
d1d9fd33 BB |
4447 | /* |
4448 | * The following two functions "fix" the issue where there are more pids | |
4449 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4450 | * TODO: replace with a kernel-wide solution to this problem | |
4451 | */ | |
4452 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4453 | static void *pidlist_allocate(int count) | |
4454 | { | |
4455 | if (PIDLIST_TOO_LARGE(count)) | |
4456 | return vmalloc(count * sizeof(pid_t)); | |
4457 | else | |
4458 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4459 | } | |
b1a21367 | 4460 | |
d1d9fd33 BB |
4461 | static void pidlist_free(void *p) |
4462 | { | |
58794514 | 4463 | kvfree(p); |
d1d9fd33 | 4464 | } |
d1d9fd33 | 4465 | |
b1a21367 TH |
4466 | /* |
4467 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4468 | * should be left afterwards. | |
4469 | */ | |
4470 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4471 | { | |
4472 | struct cgroup_pidlist *l, *tmp_l; | |
4473 | ||
4474 | mutex_lock(&cgrp->pidlist_mutex); | |
4475 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4476 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4477 | mutex_unlock(&cgrp->pidlist_mutex); | |
4478 | ||
4479 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4480 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4481 | } | |
4482 | ||
4483 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4484 | { | |
4485 | struct delayed_work *dwork = to_delayed_work(work); | |
4486 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4487 | destroy_dwork); | |
4488 | struct cgroup_pidlist *tofree = NULL; | |
4489 | ||
4490 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4491 | |
4492 | /* | |
04502365 TH |
4493 | * Destroy iff we didn't get queued again. The state won't change |
4494 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4495 | */ |
04502365 | 4496 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4497 | list_del(&l->links); |
4498 | pidlist_free(l->list); | |
4499 | put_pid_ns(l->key.ns); | |
4500 | tofree = l; | |
4501 | } | |
4502 | ||
b1a21367 TH |
4503 | mutex_unlock(&l->owner->pidlist_mutex); |
4504 | kfree(tofree); | |
4505 | } | |
4506 | ||
bbcb81d0 | 4507 | /* |
102a775e | 4508 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4509 | * Returns the number of unique elements. |
bbcb81d0 | 4510 | */ |
6ee211ad | 4511 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4512 | { |
102a775e | 4513 | int src, dest = 1; |
102a775e BB |
4514 | |
4515 | /* | |
4516 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4517 | * edge cases first; no work needs to be done for either | |
4518 | */ | |
4519 | if (length == 0 || length == 1) | |
4520 | return length; | |
4521 | /* src and dest walk down the list; dest counts unique elements */ | |
4522 | for (src = 1; src < length; src++) { | |
4523 | /* find next unique element */ | |
4524 | while (list[src] == list[src-1]) { | |
4525 | src++; | |
4526 | if (src == length) | |
4527 | goto after; | |
4528 | } | |
4529 | /* dest always points to where the next unique element goes */ | |
4530 | list[dest] = list[src]; | |
4531 | dest++; | |
4532 | } | |
4533 | after: | |
102a775e BB |
4534 | return dest; |
4535 | } | |
4536 | ||
afb2bc14 TH |
4537 | /* |
4538 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4539 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4540 | * different per namespace, each namespace needs differently sorted list, | |
4541 | * making it impossible to use, for example, single rbtree of member tasks | |
4542 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4543 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4544 | * pidlists keyed by cgroup and namespace. | |
4545 | * | |
4546 | * All this extra complexity was caused by the original implementation | |
4547 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4548 | * want to do away with it. Explicitly scramble sort order if on the |
4549 | * default hierarchy so that no such expectation exists in the new | |
4550 | * interface. | |
afb2bc14 TH |
4551 | * |
4552 | * Scrambling is done by swapping every two consecutive bits, which is | |
4553 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4554 | */ | |
4555 | static pid_t pid_fry(pid_t pid) | |
4556 | { | |
4557 | unsigned a = pid & 0x55555555; | |
4558 | unsigned b = pid & 0xAAAAAAAA; | |
4559 | ||
4560 | return (a << 1) | (b >> 1); | |
4561 | } | |
4562 | ||
4563 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4564 | { | |
aa6ec29b | 4565 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4566 | return pid_fry(pid); |
4567 | else | |
4568 | return pid; | |
4569 | } | |
4570 | ||
102a775e BB |
4571 | static int cmppid(const void *a, const void *b) |
4572 | { | |
4573 | return *(pid_t *)a - *(pid_t *)b; | |
4574 | } | |
4575 | ||
afb2bc14 TH |
4576 | static int fried_cmppid(const void *a, const void *b) |
4577 | { | |
4578 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4579 | } | |
4580 | ||
e6b81710 TH |
4581 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4582 | enum cgroup_filetype type) | |
4583 | { | |
4584 | struct cgroup_pidlist *l; | |
4585 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4586 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4587 | ||
4588 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4589 | ||
4590 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4591 | if (l->key.type == type && l->key.ns == ns) | |
4592 | return l; | |
4593 | return NULL; | |
4594 | } | |
4595 | ||
72a8cb30 BB |
4596 | /* |
4597 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4598 | * returns with the lock on that pidlist already held, and takes care | |
4599 | * of the use count, or returns NULL with no locks held if we're out of | |
4600 | * memory. | |
4601 | */ | |
e6b81710 TH |
4602 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4603 | enum cgroup_filetype type) | |
72a8cb30 BB |
4604 | { |
4605 | struct cgroup_pidlist *l; | |
b70cc5fd | 4606 | |
e6b81710 TH |
4607 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4608 | ||
4609 | l = cgroup_pidlist_find(cgrp, type); | |
4610 | if (l) | |
4611 | return l; | |
4612 | ||
72a8cb30 | 4613 | /* entry not found; create a new one */ |
f4f4be2b | 4614 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4615 | if (!l) |
72a8cb30 | 4616 | return l; |
e6b81710 | 4617 | |
b1a21367 | 4618 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4619 | l->key.type = type; |
e6b81710 TH |
4620 | /* don't need task_nsproxy() if we're looking at ourself */ |
4621 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4622 | l->owner = cgrp; |
4623 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4624 | return l; |
4625 | } | |
4626 | ||
102a775e BB |
4627 | /* |
4628 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4629 | */ | |
72a8cb30 BB |
4630 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4631 | struct cgroup_pidlist **lp) | |
102a775e BB |
4632 | { |
4633 | pid_t *array; | |
4634 | int length; | |
4635 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4636 | struct css_task_iter it; |
817929ec | 4637 | struct task_struct *tsk; |
102a775e BB |
4638 | struct cgroup_pidlist *l; |
4639 | ||
4bac00d1 TH |
4640 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4641 | ||
102a775e BB |
4642 | /* |
4643 | * If cgroup gets more users after we read count, we won't have | |
4644 | * enough space - tough. This race is indistinguishable to the | |
4645 | * caller from the case that the additional cgroup users didn't | |
4646 | * show up until sometime later on. | |
4647 | */ | |
4648 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4649 | array = pidlist_allocate(length); |
102a775e BB |
4650 | if (!array) |
4651 | return -ENOMEM; | |
4652 | /* now, populate the array */ | |
9d800df1 | 4653 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4654 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4655 | if (unlikely(n == length)) |
817929ec | 4656 | break; |
102a775e | 4657 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4658 | if (type == CGROUP_FILE_PROCS) |
4659 | pid = task_tgid_vnr(tsk); | |
4660 | else | |
4661 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4662 | if (pid > 0) /* make sure to only use valid results */ |
4663 | array[n++] = pid; | |
817929ec | 4664 | } |
72ec7029 | 4665 | css_task_iter_end(&it); |
102a775e BB |
4666 | length = n; |
4667 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4668 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4669 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4670 | else | |
4671 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4672 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4673 | length = pidlist_uniq(array, length); |
e6b81710 | 4674 | |
e6b81710 | 4675 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4676 | if (!l) { |
d1d9fd33 | 4677 | pidlist_free(array); |
72a8cb30 | 4678 | return -ENOMEM; |
102a775e | 4679 | } |
e6b81710 TH |
4680 | |
4681 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4682 | pidlist_free(l->list); |
102a775e BB |
4683 | l->list = array; |
4684 | l->length = length; | |
72a8cb30 | 4685 | *lp = l; |
102a775e | 4686 | return 0; |
bbcb81d0 PM |
4687 | } |
4688 | ||
846c7bb0 | 4689 | /** |
a043e3b2 | 4690 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4691 | * @stats: cgroupstats to fill information into |
4692 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4693 | * been requested. | |
a043e3b2 LZ |
4694 | * |
4695 | * Build and fill cgroupstats so that taskstats can export it to user | |
4696 | * space. | |
846c7bb0 BS |
4697 | */ |
4698 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4699 | { | |
2bd59d48 | 4700 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4701 | struct cgroup *cgrp; |
72ec7029 | 4702 | struct css_task_iter it; |
846c7bb0 | 4703 | struct task_struct *tsk; |
33d283be | 4704 | |
2bd59d48 TH |
4705 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4706 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4707 | kernfs_type(kn) != KERNFS_DIR) | |
4708 | return -EINVAL; | |
4709 | ||
bad34660 LZ |
4710 | mutex_lock(&cgroup_mutex); |
4711 | ||
846c7bb0 | 4712 | /* |
2bd59d48 | 4713 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4714 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4715 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4716 | */ |
2bd59d48 TH |
4717 | rcu_read_lock(); |
4718 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4719 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4720 | rcu_read_unlock(); |
bad34660 | 4721 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4722 | return -ENOENT; |
4723 | } | |
bad34660 | 4724 | rcu_read_unlock(); |
846c7bb0 | 4725 | |
9d800df1 | 4726 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4727 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4728 | switch (tsk->state) { |
4729 | case TASK_RUNNING: | |
4730 | stats->nr_running++; | |
4731 | break; | |
4732 | case TASK_INTERRUPTIBLE: | |
4733 | stats->nr_sleeping++; | |
4734 | break; | |
4735 | case TASK_UNINTERRUPTIBLE: | |
4736 | stats->nr_uninterruptible++; | |
4737 | break; | |
4738 | case TASK_STOPPED: | |
4739 | stats->nr_stopped++; | |
4740 | break; | |
4741 | default: | |
4742 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4743 | stats->nr_io_wait++; | |
4744 | break; | |
4745 | } | |
4746 | } | |
72ec7029 | 4747 | css_task_iter_end(&it); |
846c7bb0 | 4748 | |
bad34660 | 4749 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4750 | return 0; |
846c7bb0 BS |
4751 | } |
4752 | ||
8f3ff208 | 4753 | |
bbcb81d0 | 4754 | /* |
102a775e | 4755 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4756 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4757 | * in the cgroup->l->list array. |
bbcb81d0 | 4758 | */ |
cc31edce | 4759 | |
102a775e | 4760 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4761 | { |
cc31edce PM |
4762 | /* |
4763 | * Initially we receive a position value that corresponds to | |
4764 | * one more than the last pid shown (or 0 on the first call or | |
4765 | * after a seek to the start). Use a binary-search to find the | |
4766 | * next pid to display, if any | |
4767 | */ | |
2bd59d48 | 4768 | struct kernfs_open_file *of = s->private; |
7da11279 | 4769 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4770 | struct cgroup_pidlist *l; |
7da11279 | 4771 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4772 | int index = 0, pid = *pos; |
4bac00d1 TH |
4773 | int *iter, ret; |
4774 | ||
4775 | mutex_lock(&cgrp->pidlist_mutex); | |
4776 | ||
4777 | /* | |
5d22444f | 4778 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4779 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4780 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4781 | * could already have been destroyed. |
4782 | */ | |
5d22444f TH |
4783 | if (of->priv) |
4784 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4785 | |
4786 | /* | |
4787 | * Either this is the first start() after open or the matching | |
4788 | * pidlist has been destroyed inbetween. Create a new one. | |
4789 | */ | |
5d22444f TH |
4790 | if (!of->priv) { |
4791 | ret = pidlist_array_load(cgrp, type, | |
4792 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4793 | if (ret) |
4794 | return ERR_PTR(ret); | |
4795 | } | |
5d22444f | 4796 | l = of->priv; |
cc31edce | 4797 | |
cc31edce | 4798 | if (pid) { |
102a775e | 4799 | int end = l->length; |
20777766 | 4800 | |
cc31edce PM |
4801 | while (index < end) { |
4802 | int mid = (index + end) / 2; | |
afb2bc14 | 4803 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4804 | index = mid; |
4805 | break; | |
afb2bc14 | 4806 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4807 | index = mid + 1; |
4808 | else | |
4809 | end = mid; | |
4810 | } | |
4811 | } | |
4812 | /* If we're off the end of the array, we're done */ | |
102a775e | 4813 | if (index >= l->length) |
cc31edce PM |
4814 | return NULL; |
4815 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4816 | iter = l->list + index; |
afb2bc14 | 4817 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4818 | return iter; |
4819 | } | |
4820 | ||
102a775e | 4821 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4822 | { |
2bd59d48 | 4823 | struct kernfs_open_file *of = s->private; |
5d22444f | 4824 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4825 | |
5d22444f TH |
4826 | if (l) |
4827 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4828 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4829 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4830 | } |
4831 | ||
102a775e | 4832 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4833 | { |
2bd59d48 | 4834 | struct kernfs_open_file *of = s->private; |
5d22444f | 4835 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4836 | pid_t *p = v; |
4837 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4838 | /* |
4839 | * Advance to the next pid in the array. If this goes off the | |
4840 | * end, we're done | |
4841 | */ | |
4842 | p++; | |
4843 | if (p >= end) { | |
4844 | return NULL; | |
4845 | } else { | |
7da11279 | 4846 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4847 | return p; |
4848 | } | |
4849 | } | |
4850 | ||
102a775e | 4851 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4852 | { |
94ff212d JP |
4853 | seq_printf(s, "%d\n", *(int *)v); |
4854 | ||
4855 | return 0; | |
cc31edce | 4856 | } |
bbcb81d0 | 4857 | |
182446d0 TH |
4858 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4859 | struct cftype *cft) | |
81a6a5cd | 4860 | { |
182446d0 | 4861 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4862 | } |
4863 | ||
182446d0 TH |
4864 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4865 | struct cftype *cft, u64 val) | |
6379c106 | 4866 | { |
6379c106 | 4867 | if (val) |
182446d0 | 4868 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4869 | else |
182446d0 | 4870 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4871 | return 0; |
4872 | } | |
4873 | ||
182446d0 TH |
4874 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4875 | struct cftype *cft) | |
97978e6d | 4876 | { |
182446d0 | 4877 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4878 | } |
4879 | ||
182446d0 TH |
4880 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4881 | struct cftype *cft, u64 val) | |
97978e6d DL |
4882 | { |
4883 | if (val) | |
182446d0 | 4884 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4885 | else |
182446d0 | 4886 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4887 | return 0; |
4888 | } | |
4889 | ||
a14c6874 TH |
4890 | /* cgroup core interface files for the default hierarchy */ |
4891 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4892 | { |
d5c56ced | 4893 | .name = "cgroup.procs", |
6f60eade | 4894 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4895 | .seq_start = cgroup_pidlist_start, |
4896 | .seq_next = cgroup_pidlist_next, | |
4897 | .seq_stop = cgroup_pidlist_stop, | |
4898 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4899 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4900 | .write = cgroup_procs_write, |
102a775e | 4901 | }, |
f8f22e53 TH |
4902 | { |
4903 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4904 | .seq_show = cgroup_controllers_show, |
4905 | }, | |
4906 | { | |
4907 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4908 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4909 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4910 | }, |
842b597e | 4911 | { |
4a07c222 | 4912 | .name = "cgroup.events", |
a14c6874 | 4913 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4914 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4915 | .seq_show = cgroup_events_show, |
842b597e | 4916 | }, |
a14c6874 TH |
4917 | { } /* terminate */ |
4918 | }; | |
d5c56ced | 4919 | |
a14c6874 TH |
4920 | /* cgroup core interface files for the legacy hierarchies */ |
4921 | static struct cftype cgroup_legacy_base_files[] = { | |
4922 | { | |
4923 | .name = "cgroup.procs", | |
4924 | .seq_start = cgroup_pidlist_start, | |
4925 | .seq_next = cgroup_pidlist_next, | |
4926 | .seq_stop = cgroup_pidlist_stop, | |
4927 | .seq_show = cgroup_pidlist_show, | |
4928 | .private = CGROUP_FILE_PROCS, | |
4929 | .write = cgroup_procs_write, | |
a14c6874 TH |
4930 | }, |
4931 | { | |
4932 | .name = "cgroup.clone_children", | |
4933 | .read_u64 = cgroup_clone_children_read, | |
4934 | .write_u64 = cgroup_clone_children_write, | |
4935 | }, | |
4936 | { | |
4937 | .name = "cgroup.sane_behavior", | |
4938 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4939 | .seq_show = cgroup_sane_behavior_show, | |
4940 | }, | |
d5c56ced TH |
4941 | { |
4942 | .name = "tasks", | |
6612f05b TH |
4943 | .seq_start = cgroup_pidlist_start, |
4944 | .seq_next = cgroup_pidlist_next, | |
4945 | .seq_stop = cgroup_pidlist_stop, | |
4946 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4947 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4948 | .write = cgroup_tasks_write, |
d5c56ced TH |
4949 | }, |
4950 | { | |
4951 | .name = "notify_on_release", | |
d5c56ced TH |
4952 | .read_u64 = cgroup_read_notify_on_release, |
4953 | .write_u64 = cgroup_write_notify_on_release, | |
4954 | }, | |
6e6ff25b TH |
4955 | { |
4956 | .name = "release_agent", | |
a14c6874 | 4957 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4958 | .seq_show = cgroup_release_agent_show, |
451af504 | 4959 | .write = cgroup_release_agent_write, |
5f469907 | 4960 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4961 | }, |
db0416b6 | 4962 | { } /* terminate */ |
bbcb81d0 PM |
4963 | }; |
4964 | ||
0c21ead1 TH |
4965 | /* |
4966 | * css destruction is four-stage process. | |
4967 | * | |
4968 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4969 | * Implemented in kill_css(). | |
4970 | * | |
4971 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4972 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4973 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4974 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4975 | * |
4976 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4977 | * accessors are inside RCU read sections. css_release() schedules the | |
4978 | * RCU callback. | |
4979 | * | |
4980 | * 4. After the grace period, the css can be freed. Implemented in | |
4981 | * css_free_work_fn(). | |
4982 | * | |
4983 | * It is actually hairier because both step 2 and 4 require process context | |
4984 | * and thus involve punting to css->destroy_work adding two additional | |
4985 | * steps to the already complex sequence. | |
4986 | */ | |
35ef10da | 4987 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4988 | { |
4989 | struct cgroup_subsys_state *css = | |
35ef10da | 4990 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4991 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4992 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4993 | |
9a1049da TH |
4994 | percpu_ref_exit(&css->refcnt); |
4995 | ||
01e58659 | 4996 | if (ss) { |
9d755d33 | 4997 | /* css free path */ |
8bb5ef79 | 4998 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4999 | int id = css->id; |
5000 | ||
01e58659 VD |
5001 | ss->css_free(css); |
5002 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 5003 | cgroup_put(cgrp); |
8bb5ef79 TH |
5004 | |
5005 | if (parent) | |
5006 | css_put(parent); | |
9d755d33 TH |
5007 | } else { |
5008 | /* cgroup free path */ | |
5009 | atomic_dec(&cgrp->root->nr_cgrps); | |
5010 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 5011 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 5012 | |
d51f39b0 | 5013 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
5014 | /* |
5015 | * We get a ref to the parent, and put the ref when | |
5016 | * this cgroup is being freed, so it's guaranteed | |
5017 | * that the parent won't be destroyed before its | |
5018 | * children. | |
5019 | */ | |
d51f39b0 | 5020 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
5021 | kernfs_put(cgrp->kn); |
5022 | kfree(cgrp); | |
5023 | } else { | |
5024 | /* | |
5025 | * This is root cgroup's refcnt reaching zero, | |
5026 | * which indicates that the root should be | |
5027 | * released. | |
5028 | */ | |
5029 | cgroup_destroy_root(cgrp->root); | |
5030 | } | |
5031 | } | |
48ddbe19 TH |
5032 | } |
5033 | ||
0c21ead1 | 5034 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
5035 | { |
5036 | struct cgroup_subsys_state *css = | |
0c21ead1 | 5037 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 5038 | |
35ef10da | 5039 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 5040 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
5041 | } |
5042 | ||
25e15d83 | 5043 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
5044 | { |
5045 | struct cgroup_subsys_state *css = | |
25e15d83 | 5046 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 5047 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 5048 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 5049 | |
1fed1b2e TH |
5050 | mutex_lock(&cgroup_mutex); |
5051 | ||
de3f0341 | 5052 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
5053 | list_del_rcu(&css->sibling); |
5054 | ||
9d755d33 TH |
5055 | if (ss) { |
5056 | /* css release path */ | |
01e58659 | 5057 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
5058 | if (ss->css_released) |
5059 | ss->css_released(css); | |
9d755d33 TH |
5060 | } else { |
5061 | /* cgroup release path */ | |
ed1777de TH |
5062 | trace_cgroup_release(cgrp); |
5063 | ||
9d755d33 TH |
5064 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
5065 | cgrp->id = -1; | |
a4189487 LZ |
5066 | |
5067 | /* | |
5068 | * There are two control paths which try to determine | |
5069 | * cgroup from dentry without going through kernfs - | |
5070 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
5071 | * Those are supported by RCU protecting clearing of | |
5072 | * cgrp->kn->priv backpointer. | |
5073 | */ | |
6cd0f5bb TH |
5074 | if (cgrp->kn) |
5075 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
5076 | NULL); | |
30070984 DM |
5077 | |
5078 | cgroup_bpf_put(cgrp); | |
9d755d33 | 5079 | } |
d3daf28d | 5080 | |
1fed1b2e TH |
5081 | mutex_unlock(&cgroup_mutex); |
5082 | ||
0c21ead1 | 5083 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
5084 | } |
5085 | ||
d3daf28d TH |
5086 | static void css_release(struct percpu_ref *ref) |
5087 | { | |
5088 | struct cgroup_subsys_state *css = | |
5089 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5090 | ||
25e15d83 TH |
5091 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5092 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5093 | } |
5094 | ||
ddfcadab TH |
5095 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5096 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5097 | { |
0cb51d71 TH |
5098 | lockdep_assert_held(&cgroup_mutex); |
5099 | ||
ddfcadab TH |
5100 | cgroup_get(cgrp); |
5101 | ||
d5c419b6 | 5102 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5103 | css->cgroup = cgrp; |
72c97e54 | 5104 | css->ss = ss; |
8fa3b8d6 | 5105 | css->id = -1; |
d5c419b6 TH |
5106 | INIT_LIST_HEAD(&css->sibling); |
5107 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 5108 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5109 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5110 | |
d51f39b0 TH |
5111 | if (cgroup_parent(cgrp)) { |
5112 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5113 | css_get(css->parent); |
ddfcadab | 5114 | } |
48ddbe19 | 5115 | |
ca8bdcaf | 5116 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5117 | } |
5118 | ||
2a4ac633 | 5119 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5120 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5121 | { |
623f926b | 5122 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5123 | int ret = 0; |
5124 | ||
a31f2d3f TH |
5125 | lockdep_assert_held(&cgroup_mutex); |
5126 | ||
92fb9748 | 5127 | if (ss->css_online) |
eb95419b | 5128 | ret = ss->css_online(css); |
ae7f164a | 5129 | if (!ret) { |
eb95419b | 5130 | css->flags |= CSS_ONLINE; |
aec25020 | 5131 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5132 | |
5133 | atomic_inc(&css->online_cnt); | |
5134 | if (css->parent) | |
5135 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5136 | } |
b1929db4 | 5137 | return ret; |
a31f2d3f TH |
5138 | } |
5139 | ||
2a4ac633 | 5140 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5141 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5142 | { |
623f926b | 5143 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5144 | |
5145 | lockdep_assert_held(&cgroup_mutex); | |
5146 | ||
5147 | if (!(css->flags & CSS_ONLINE)) | |
5148 | return; | |
5149 | ||
fa06235b VD |
5150 | if (ss->css_reset) |
5151 | ss->css_reset(css); | |
5152 | ||
d7eeac19 | 5153 | if (ss->css_offline) |
eb95419b | 5154 | ss->css_offline(css); |
a31f2d3f | 5155 | |
eb95419b | 5156 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5157 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5158 | |
5159 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5160 | } |
5161 | ||
c81c925a | 5162 | /** |
6cd0f5bb | 5163 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5164 | * @cgrp: the cgroup new css will be associated with |
5165 | * @ss: the subsys of new css | |
5166 | * | |
5167 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5168 | * css is online and installed in @cgrp. This function doesn't create the |
5169 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5170 | */ |
6cd0f5bb TH |
5171 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5172 | struct cgroup_subsys *ss) | |
c81c925a | 5173 | { |
d51f39b0 | 5174 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5175 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5176 | struct cgroup_subsys_state *css; |
5177 | int err; | |
5178 | ||
c81c925a TH |
5179 | lockdep_assert_held(&cgroup_mutex); |
5180 | ||
1fed1b2e | 5181 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
5182 | if (!css) |
5183 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 5184 | if (IS_ERR(css)) |
6cd0f5bb | 5185 | return css; |
c81c925a | 5186 | |
ddfcadab | 5187 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5188 | |
2aad2a86 | 5189 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5190 | if (err) |
3eb59ec6 | 5191 | goto err_free_css; |
c81c925a | 5192 | |
cf780b7d | 5193 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 5194 | if (err < 0) |
b00c52da | 5195 | goto err_free_css; |
15a4c835 | 5196 | css->id = err; |
c81c925a | 5197 | |
15a4c835 | 5198 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5199 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5200 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5201 | |
5202 | err = online_css(css); | |
5203 | if (err) | |
1fed1b2e | 5204 | goto err_list_del; |
94419627 | 5205 | |
c81c925a | 5206 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5207 | cgroup_parent(parent)) { |
ed3d261b | 5208 | 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 | 5209 | current->comm, current->pid, ss->name); |
c81c925a | 5210 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5211 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5212 | ss->warned_broken_hierarchy = true; |
5213 | } | |
5214 | ||
6cd0f5bb | 5215 | return css; |
c81c925a | 5216 | |
1fed1b2e TH |
5217 | err_list_del: |
5218 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 5219 | err_free_css: |
a2bed820 | 5220 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 5221 | return ERR_PTR(err); |
c81c925a TH |
5222 | } |
5223 | ||
a5bca215 | 5224 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 5225 | { |
a5bca215 | 5226 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
5227 | struct cgroup *cgrp, *tcgrp; |
5228 | int level = parent->level + 1; | |
03970d3c | 5229 | int ret; |
ddbcc7e8 | 5230 | |
0a950f65 | 5231 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
5232 | cgrp = kzalloc(sizeof(*cgrp) + |
5233 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
5234 | if (!cgrp) |
5235 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5236 | |
2aad2a86 | 5237 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5238 | if (ret) |
5239 | goto out_free_cgrp; | |
5240 | ||
0ab02ca8 LZ |
5241 | /* |
5242 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5243 | * a half-baked cgroup. | |
5244 | */ | |
cf780b7d | 5245 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5246 | if (cgrp->id < 0) { |
ba0f4d76 | 5247 | ret = -ENOMEM; |
9d755d33 | 5248 | goto out_cancel_ref; |
976c06bc TH |
5249 | } |
5250 | ||
cc31edce | 5251 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5252 | |
9d800df1 | 5253 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5254 | cgrp->root = root; |
b11cfb58 TH |
5255 | cgrp->level = level; |
5256 | ||
5257 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5258 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5259 | |
b6abdb0e LZ |
5260 | if (notify_on_release(parent)) |
5261 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5262 | ||
2260e7fc TH |
5263 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5264 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5265 | |
0cb51d71 | 5266 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5267 | |
4e139afc | 5268 | /* allocation complete, commit to creation */ |
d5c419b6 | 5269 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5270 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5271 | cgroup_get(parent); |
415cf07a | 5272 | |
0d80255e TH |
5273 | /* |
5274 | * @cgrp is now fully operational. If something fails after this | |
5275 | * point, it'll be released via the normal destruction path. | |
5276 | */ | |
6fa4918d | 5277 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5278 | |
bd53d617 TH |
5279 | /* |
5280 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5281 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5282 | */ |
03970d3c | 5283 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5284 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c | 5285 | |
30070984 DM |
5286 | if (parent) |
5287 | cgroup_bpf_inherit(cgrp, parent); | |
5288 | ||
03970d3c TH |
5289 | cgroup_propagate_control(cgrp); |
5290 | ||
5291 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5292 | ret = cgroup_apply_control_enable(cgrp); | |
5293 | if (ret) | |
5294 | goto out_destroy; | |
2bd59d48 | 5295 | |
a5bca215 TH |
5296 | return cgrp; |
5297 | ||
5298 | out_cancel_ref: | |
5299 | percpu_ref_exit(&cgrp->self.refcnt); | |
5300 | out_free_cgrp: | |
5301 | kfree(cgrp); | |
5302 | return ERR_PTR(ret); | |
5303 | out_destroy: | |
5304 | cgroup_destroy_locked(cgrp); | |
5305 | return ERR_PTR(ret); | |
5306 | } | |
5307 | ||
5308 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5309 | umode_t mode) | |
5310 | { | |
5311 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5312 | struct kernfs_node *kn; |
03970d3c | 5313 | int ret; |
a5bca215 TH |
5314 | |
5315 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5316 | if (strchr(name, '\n')) | |
5317 | return -EINVAL; | |
5318 | ||
945ba199 | 5319 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5320 | if (!parent) |
5321 | return -ENODEV; | |
5322 | ||
5323 | cgrp = cgroup_create(parent); | |
5324 | if (IS_ERR(cgrp)) { | |
5325 | ret = PTR_ERR(cgrp); | |
5326 | goto out_unlock; | |
5327 | } | |
5328 | ||
195e9b6c TH |
5329 | /* create the directory */ |
5330 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5331 | if (IS_ERR(kn)) { | |
5332 | ret = PTR_ERR(kn); | |
5333 | goto out_destroy; | |
5334 | } | |
5335 | cgrp->kn = kn; | |
5336 | ||
5337 | /* | |
5338 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5339 | * that @cgrp->kn is always accessible. | |
5340 | */ | |
5341 | kernfs_get(kn); | |
5342 | ||
5343 | ret = cgroup_kn_set_ugid(kn); | |
5344 | if (ret) | |
5345 | goto out_destroy; | |
5346 | ||
334c3679 | 5347 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5348 | if (ret) |
5349 | goto out_destroy; | |
5350 | ||
03970d3c TH |
5351 | ret = cgroup_apply_control_enable(cgrp); |
5352 | if (ret) | |
5353 | goto out_destroy; | |
195e9b6c | 5354 | |
ed1777de TH |
5355 | trace_cgroup_mkdir(cgrp); |
5356 | ||
195e9b6c | 5357 | /* let's create and online css's */ |
2bd59d48 | 5358 | kernfs_activate(kn); |
ddbcc7e8 | 5359 | |
ba0f4d76 TH |
5360 | ret = 0; |
5361 | goto out_unlock; | |
ddbcc7e8 | 5362 | |
a5bca215 TH |
5363 | out_destroy: |
5364 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5365 | out_unlock: |
a9746d8d | 5366 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5367 | return ret; |
ddbcc7e8 PM |
5368 | } |
5369 | ||
223dbc38 TH |
5370 | /* |
5371 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5372 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5373 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5374 | */ |
5375 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5376 | { |
223dbc38 TH |
5377 | struct cgroup_subsys_state *css = |
5378 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5379 | |
f20104de | 5380 | mutex_lock(&cgroup_mutex); |
09a503ea | 5381 | |
aa226ff4 TH |
5382 | do { |
5383 | offline_css(css); | |
5384 | css_put(css); | |
5385 | /* @css can't go away while we're holding cgroup_mutex */ | |
5386 | css = css->parent; | |
5387 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5388 | ||
5389 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5390 | } |
5391 | ||
223dbc38 TH |
5392 | /* css kill confirmation processing requires process context, bounce */ |
5393 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5394 | { |
5395 | struct cgroup_subsys_state *css = | |
5396 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5397 | ||
aa226ff4 TH |
5398 | if (atomic_dec_and_test(&css->online_cnt)) { |
5399 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5400 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5401 | } | |
d3daf28d TH |
5402 | } |
5403 | ||
f392e51c TH |
5404 | /** |
5405 | * kill_css - destroy a css | |
5406 | * @css: css to destroy | |
5407 | * | |
5408 | * This function initiates destruction of @css by removing cgroup interface | |
5409 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5410 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5411 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5412 | */ |
5413 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5414 | { |
01f6474c | 5415 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5416 | |
2bd59d48 TH |
5417 | /* |
5418 | * This must happen before css is disassociated with its cgroup. | |
5419 | * See seq_css() for details. | |
5420 | */ | |
334c3679 | 5421 | css_clear_dir(css); |
3c14f8b4 | 5422 | |
edae0c33 TH |
5423 | /* |
5424 | * Killing would put the base ref, but we need to keep it alive | |
5425 | * until after ->css_offline(). | |
5426 | */ | |
5427 | css_get(css); | |
5428 | ||
5429 | /* | |
5430 | * cgroup core guarantees that, by the time ->css_offline() is | |
5431 | * invoked, no new css reference will be given out via | |
ec903c0c | 5432 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5433 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5434 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5435 | * | |
5436 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5437 | * css is confirmed to be seen as killed on all CPUs. | |
5438 | */ | |
5439 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5440 | } |
5441 | ||
5442 | /** | |
5443 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5444 | * @cgrp: cgroup to be destroyed | |
5445 | * | |
5446 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5447 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5448 | * guarantee that css_tryget_online() won't succeed by the time |
5449 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5450 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5451 | * |
5452 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5453 | * userland visible parts and start killing the percpu refcnts of | |
5454 | * css's. Set up so that the next stage will be kicked off once all | |
5455 | * the percpu refcnts are confirmed to be killed. | |
5456 | * | |
5457 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5458 | * rest of destruction. Once all cgroup references are gone, the | |
5459 | * cgroup is RCU-freed. | |
5460 | * | |
5461 | * This function implements s1. After this step, @cgrp is gone as far as | |
5462 | * the userland is concerned and a new cgroup with the same name may be | |
5463 | * created. As cgroup doesn't care about the names internally, this | |
5464 | * doesn't cause any problem. | |
5465 | */ | |
42809dd4 TH |
5466 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5467 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5468 | { |
2bd59d48 | 5469 | struct cgroup_subsys_state *css; |
2b021cbf | 5470 | struct cgrp_cset_link *link; |
1c6727af | 5471 | int ssid; |
ddbcc7e8 | 5472 | |
42809dd4 TH |
5473 | lockdep_assert_held(&cgroup_mutex); |
5474 | ||
91486f61 TH |
5475 | /* |
5476 | * Only migration can raise populated from zero and we're already | |
5477 | * holding cgroup_mutex. | |
5478 | */ | |
5479 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5480 | return -EBUSY; |
a043e3b2 | 5481 | |
bb78a92f | 5482 | /* |
d5c419b6 TH |
5483 | * Make sure there's no live children. We can't test emptiness of |
5484 | * ->self.children as dead children linger on it while being | |
5485 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5486 | */ |
f3d46500 | 5487 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5488 | return -EBUSY; |
5489 | ||
455050d2 | 5490 | /* |
2b021cbf TH |
5491 | * Mark @cgrp and the associated csets dead. The former prevents |
5492 | * further task migration and child creation by disabling | |
5493 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5494 | * the migration path. | |
455050d2 | 5495 | */ |
184faf32 | 5496 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5497 | |
82d6489d | 5498 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5499 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5500 | link->cset->dead = true; | |
82d6489d | 5501 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5502 | |
249f3468 | 5503 | /* initiate massacre of all css's */ |
1c6727af TH |
5504 | for_each_css(css, ssid, cgrp) |
5505 | kill_css(css); | |
455050d2 | 5506 | |
455050d2 | 5507 | /* |
01f6474c TH |
5508 | * Remove @cgrp directory along with the base files. @cgrp has an |
5509 | * extra ref on its kn. | |
f20104de | 5510 | */ |
01f6474c | 5511 | kernfs_remove(cgrp->kn); |
f20104de | 5512 | |
d51f39b0 | 5513 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5514 | |
249f3468 | 5515 | /* put the base reference */ |
9d755d33 | 5516 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5517 | |
ea15f8cc TH |
5518 | return 0; |
5519 | }; | |
5520 | ||
2bd59d48 | 5521 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5522 | { |
a9746d8d | 5523 | struct cgroup *cgrp; |
2bd59d48 | 5524 | int ret = 0; |
42809dd4 | 5525 | |
945ba199 | 5526 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5527 | if (!cgrp) |
5528 | return 0; | |
42809dd4 | 5529 | |
a9746d8d | 5530 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5531 | |
ed1777de TH |
5532 | if (!ret) |
5533 | trace_cgroup_rmdir(cgrp); | |
5534 | ||
a9746d8d | 5535 | cgroup_kn_unlock(kn); |
42809dd4 | 5536 | return ret; |
8e3f6541 TH |
5537 | } |
5538 | ||
2bd59d48 TH |
5539 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5540 | .remount_fs = cgroup_remount, | |
5541 | .show_options = cgroup_show_options, | |
5542 | .mkdir = cgroup_mkdir, | |
5543 | .rmdir = cgroup_rmdir, | |
5544 | .rename = cgroup_rename, | |
4f41fc59 | 5545 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5546 | }; |
5547 | ||
15a4c835 | 5548 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5549 | { |
ddbcc7e8 | 5550 | struct cgroup_subsys_state *css; |
cfe36bde | 5551 | |
a5ae9899 | 5552 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5553 | |
648bb56d TH |
5554 | mutex_lock(&cgroup_mutex); |
5555 | ||
15a4c835 | 5556 | idr_init(&ss->css_idr); |
0adb0704 | 5557 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5558 | |
3dd06ffa TH |
5559 | /* Create the root cgroup state for this subsystem */ |
5560 | ss->root = &cgrp_dfl_root; | |
5561 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5562 | /* We don't handle early failures gracefully */ |
5563 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5564 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5565 | |
5566 | /* | |
5567 | * Root csses are never destroyed and we can't initialize | |
5568 | * percpu_ref during early init. Disable refcnting. | |
5569 | */ | |
5570 | css->flags |= CSS_NO_REF; | |
5571 | ||
15a4c835 | 5572 | if (early) { |
9395a450 | 5573 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5574 | css->id = 1; |
5575 | } else { | |
5576 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5577 | BUG_ON(css->id < 0); | |
5578 | } | |
ddbcc7e8 | 5579 | |
e8d55fde | 5580 | /* Update the init_css_set to contain a subsys |
817929ec | 5581 | * pointer to this state - since the subsystem is |
e8d55fde | 5582 | * newly registered, all tasks and hence the |
3dd06ffa | 5583 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5584 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5585 | |
cb4a3167 AS |
5586 | have_fork_callback |= (bool)ss->fork << ss->id; |
5587 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5588 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5589 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5590 | |
e8d55fde LZ |
5591 | /* At system boot, before all subsystems have been |
5592 | * registered, no tasks have been forked, so we don't | |
5593 | * need to invoke fork callbacks here. */ | |
5594 | BUG_ON(!list_empty(&init_task.tasks)); | |
5595 | ||
ae7f164a | 5596 | BUG_ON(online_css(css)); |
a8638030 | 5597 | |
cf5d5941 BB |
5598 | mutex_unlock(&cgroup_mutex); |
5599 | } | |
cf5d5941 | 5600 | |
ddbcc7e8 | 5601 | /** |
a043e3b2 LZ |
5602 | * cgroup_init_early - cgroup initialization at system boot |
5603 | * | |
5604 | * Initialize cgroups at system boot, and initialize any | |
5605 | * subsystems that request early init. | |
ddbcc7e8 PM |
5606 | */ |
5607 | int __init cgroup_init_early(void) | |
5608 | { | |
7b9a6ba5 | 5609 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5610 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5611 | int i; |
30159ec7 | 5612 | |
3dd06ffa | 5613 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5614 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5615 | ||
a4ea1cc9 | 5616 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5617 | |
3ed80a62 | 5618 | for_each_subsys(ss, i) { |
aec25020 | 5619 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5620 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5621 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5622 | ss->id, ss->name); |
073219e9 TH |
5623 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5624 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5625 | ||
aec25020 | 5626 | ss->id = i; |
073219e9 | 5627 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5628 | if (!ss->legacy_name) |
5629 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5630 | |
5631 | if (ss->early_init) | |
15a4c835 | 5632 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5633 | } |
5634 | return 0; | |
5635 | } | |
5636 | ||
6e5c8307 | 5637 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5638 | |
ddbcc7e8 | 5639 | /** |
a043e3b2 LZ |
5640 | * cgroup_init - cgroup initialization |
5641 | * | |
5642 | * Register cgroup filesystem and /proc file, and initialize | |
5643 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5644 | */ |
5645 | int __init cgroup_init(void) | |
5646 | { | |
30159ec7 | 5647 | struct cgroup_subsys *ss; |
035f4f51 | 5648 | int ssid; |
ddbcc7e8 | 5649 | |
6e5c8307 | 5650 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5651 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5652 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5653 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5654 | |
3942a9bd PZ |
5655 | /* |
5656 | * The latency of the synchronize_sched() is too high for cgroups, | |
5657 | * avoid it at the cost of forcing all readers into the slow path. | |
5658 | */ | |
5659 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
5660 | ||
a79a908f AK |
5661 | get_user_ns(init_cgroup_ns.user_ns); |
5662 | ||
54e7b4eb | 5663 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5664 | |
2378d8b8 TH |
5665 | /* |
5666 | * Add init_css_set to the hash table so that dfl_root can link to | |
5667 | * it during init. | |
5668 | */ | |
5669 | hash_add(css_set_table, &init_css_set.hlist, | |
5670 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5671 | |
3dd06ffa | 5672 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5673 | |
54e7b4eb TH |
5674 | mutex_unlock(&cgroup_mutex); |
5675 | ||
172a2c06 | 5676 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5677 | if (ss->early_init) { |
5678 | struct cgroup_subsys_state *css = | |
5679 | init_css_set.subsys[ss->id]; | |
5680 | ||
5681 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5682 | GFP_KERNEL); | |
5683 | BUG_ON(css->id < 0); | |
5684 | } else { | |
5685 | cgroup_init_subsys(ss, false); | |
5686 | } | |
172a2c06 | 5687 | |
2d8f243a TH |
5688 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5689 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5690 | |
5691 | /* | |
c731ae1d LZ |
5692 | * Setting dfl_root subsys_mask needs to consider the |
5693 | * disabled flag and cftype registration needs kmalloc, | |
5694 | * both of which aren't available during early_init. | |
172a2c06 | 5695 | */ |
a3e72739 TH |
5696 | if (cgroup_disable_mask & (1 << ssid)) { |
5697 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5698 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5699 | ss->name); | |
a8ddc821 | 5700 | continue; |
a3e72739 | 5701 | } |
a8ddc821 | 5702 | |
223ffb29 JW |
5703 | if (cgroup_ssid_no_v1(ssid)) |
5704 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5705 | ss->name); | |
5706 | ||
a8ddc821 TH |
5707 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5708 | ||
f6d635ad TH |
5709 | if (ss->implicit_on_dfl) |
5710 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5711 | else if (!ss->dfl_cftypes) | |
a7165264 | 5712 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5713 | |
a8ddc821 TH |
5714 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5715 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5716 | } else { | |
5717 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5718 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5719 | } |
295458e6 VD |
5720 | |
5721 | if (ss->bind) | |
5722 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5723 | } |
5724 | ||
2378d8b8 TH |
5725 | /* init_css_set.subsys[] has been updated, re-hash */ |
5726 | hash_del(&init_css_set.hlist); | |
5727 | hash_add(css_set_table, &init_css_set.hlist, | |
5728 | css_set_hash(init_css_set.subsys)); | |
5729 | ||
035f4f51 TH |
5730 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5731 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5732 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5733 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5734 | |
2bd59d48 | 5735 | return 0; |
ddbcc7e8 | 5736 | } |
b4f48b63 | 5737 | |
e5fca243 TH |
5738 | static int __init cgroup_wq_init(void) |
5739 | { | |
5740 | /* | |
5741 | * There isn't much point in executing destruction path in | |
5742 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5743 | * Use 1 for @max_active. |
e5fca243 TH |
5744 | * |
5745 | * We would prefer to do this in cgroup_init() above, but that | |
5746 | * is called before init_workqueues(): so leave this until after. | |
5747 | */ | |
1a11533f | 5748 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5749 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5750 | |
5751 | /* | |
5752 | * Used to destroy pidlists and separate to serve as flush domain. | |
5753 | * Cap @max_active to 1 too. | |
5754 | */ | |
5755 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5756 | 0, 1); | |
5757 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5758 | ||
e5fca243 TH |
5759 | return 0; |
5760 | } | |
5761 | core_initcall(cgroup_wq_init); | |
5762 | ||
a424316c PM |
5763 | /* |
5764 | * proc_cgroup_show() | |
5765 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5766 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5767 | */ |
006f4ac4 ZL |
5768 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5769 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5770 | { |
4c737b41 | 5771 | char *buf; |
a424316c | 5772 | int retval; |
3dd06ffa | 5773 | struct cgroup_root *root; |
a424316c PM |
5774 | |
5775 | retval = -ENOMEM; | |
e61734c5 | 5776 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5777 | if (!buf) |
5778 | goto out; | |
5779 | ||
a424316c | 5780 | mutex_lock(&cgroup_mutex); |
82d6489d | 5781 | spin_lock_irq(&css_set_lock); |
a424316c | 5782 | |
985ed670 | 5783 | for_each_root(root) { |
a424316c | 5784 | struct cgroup_subsys *ss; |
bd89aabc | 5785 | struct cgroup *cgrp; |
b85d2040 | 5786 | int ssid, count = 0; |
a424316c | 5787 | |
a7165264 | 5788 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5789 | continue; |
5790 | ||
2c6ab6d2 | 5791 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5792 | if (root != &cgrp_dfl_root) |
5793 | for_each_subsys(ss, ssid) | |
5794 | if (root->subsys_mask & (1 << ssid)) | |
5795 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5796 | ss->legacy_name); |
c6d57f33 PM |
5797 | if (strlen(root->name)) |
5798 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5799 | root->name); | |
a424316c | 5800 | seq_putc(m, ':'); |
2e91fa7f | 5801 | |
7717f7ba | 5802 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5803 | |
5804 | /* | |
5805 | * On traditional hierarchies, all zombie tasks show up as | |
5806 | * belonging to the root cgroup. On the default hierarchy, | |
5807 | * while a zombie doesn't show up in "cgroup.procs" and | |
5808 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5809 | * reporting the cgroup it belonged to before exiting. If | |
5810 | * the cgroup is removed before the zombie is reaped, | |
5811 | * " (deleted)" is appended to the cgroup path. | |
5812 | */ | |
5813 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 5814 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 5815 | current->nsproxy->cgroup_ns); |
e0223003 | 5816 | if (retval >= PATH_MAX) |
2e91fa7f | 5817 | retval = -ENAMETOOLONG; |
e0223003 | 5818 | if (retval < 0) |
2e91fa7f | 5819 | goto out_unlock; |
4c737b41 TH |
5820 | |
5821 | seq_puts(m, buf); | |
2e91fa7f | 5822 | } else { |
4c737b41 | 5823 | seq_puts(m, "/"); |
e61734c5 | 5824 | } |
2e91fa7f | 5825 | |
2e91fa7f TH |
5826 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
5827 | seq_puts(m, " (deleted)\n"); | |
5828 | else | |
5829 | seq_putc(m, '\n'); | |
a424316c PM |
5830 | } |
5831 | ||
006f4ac4 | 5832 | retval = 0; |
a424316c | 5833 | out_unlock: |
82d6489d | 5834 | spin_unlock_irq(&css_set_lock); |
a424316c | 5835 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5836 | kfree(buf); |
5837 | out: | |
5838 | return retval; | |
5839 | } | |
5840 | ||
a424316c PM |
5841 | /* Display information about each subsystem and each hierarchy */ |
5842 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5843 | { | |
30159ec7 | 5844 | struct cgroup_subsys *ss; |
a424316c | 5845 | int i; |
a424316c | 5846 | |
8bab8dde | 5847 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5848 | /* |
5849 | * ideally we don't want subsystems moving around while we do this. | |
5850 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5851 | * subsys/hierarchy state. | |
5852 | */ | |
a424316c | 5853 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5854 | |
5855 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5856 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5857 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5858 | atomic_read(&ss->root->nr_cgrps), |
5859 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5860 | |
a424316c PM |
5861 | mutex_unlock(&cgroup_mutex); |
5862 | return 0; | |
5863 | } | |
5864 | ||
5865 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5866 | { | |
9dce07f1 | 5867 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5868 | } |
5869 | ||
828c0950 | 5870 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5871 | .open = cgroupstats_open, |
5872 | .read = seq_read, | |
5873 | .llseek = seq_lseek, | |
5874 | .release = single_release, | |
5875 | }; | |
5876 | ||
b4f48b63 | 5877 | /** |
eaf797ab | 5878 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5879 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5880 | * |
eaf797ab TH |
5881 | * A task is associated with the init_css_set until cgroup_post_fork() |
5882 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5883 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5884 | */ |
5885 | void cgroup_fork(struct task_struct *child) | |
5886 | { | |
eaf797ab | 5887 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5888 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5889 | } |
5890 | ||
7e47682e AS |
5891 | /** |
5892 | * cgroup_can_fork - called on a new task before the process is exposed | |
5893 | * @child: the task in question. | |
5894 | * | |
5895 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5896 | * returns an error, the fork aborts with that error code. This allows for | |
5897 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5898 | */ | |
b53202e6 | 5899 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5900 | { |
5901 | struct cgroup_subsys *ss; | |
5902 | int i, j, ret; | |
5903 | ||
b4e0eeaf | 5904 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5905 | ret = ss->can_fork(child); |
7e47682e AS |
5906 | if (ret) |
5907 | goto out_revert; | |
b4e0eeaf | 5908 | } while_each_subsys_mask(); |
7e47682e AS |
5909 | |
5910 | return 0; | |
5911 | ||
5912 | out_revert: | |
5913 | for_each_subsys(ss, j) { | |
5914 | if (j >= i) | |
5915 | break; | |
5916 | if (ss->cancel_fork) | |
b53202e6 | 5917 | ss->cancel_fork(child); |
7e47682e AS |
5918 | } |
5919 | ||
5920 | return ret; | |
5921 | } | |
5922 | ||
5923 | /** | |
5924 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5925 | * @child: the task in question | |
5926 | * | |
5927 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5928 | * cgroup_can_fork() succeded. | |
5929 | */ | |
b53202e6 | 5930 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5931 | { |
5932 | struct cgroup_subsys *ss; | |
5933 | int i; | |
5934 | ||
5935 | for_each_subsys(ss, i) | |
5936 | if (ss->cancel_fork) | |
b53202e6 | 5937 | ss->cancel_fork(child); |
7e47682e AS |
5938 | } |
5939 | ||
817929ec | 5940 | /** |
a043e3b2 LZ |
5941 | * cgroup_post_fork - called on a new task after adding it to the task list |
5942 | * @child: the task in question | |
5943 | * | |
5edee61e TH |
5944 | * Adds the task to the list running through its css_set if necessary and |
5945 | * call the subsystem fork() callbacks. Has to be after the task is | |
5946 | * visible on the task list in case we race with the first call to | |
0942eeee | 5947 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5948 | * list. |
a043e3b2 | 5949 | */ |
b53202e6 | 5950 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5951 | { |
30159ec7 | 5952 | struct cgroup_subsys *ss; |
5edee61e TH |
5953 | int i; |
5954 | ||
3ce3230a | 5955 | /* |
251f8c03 | 5956 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5957 | * function sets use_task_css_set_links before grabbing |
5958 | * tasklist_lock and we just went through tasklist_lock to add | |
5959 | * @child, it's guaranteed that either we see the set | |
5960 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5961 | * @child during its iteration. | |
5962 | * | |
5963 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5964 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5965 | * association is stable, and, on completion of the parent's |
5966 | * migration, @child is visible in the source of migration or | |
5967 | * already in the destination cgroup. This guarantee is necessary | |
5968 | * when implementing operations which need to migrate all tasks of | |
5969 | * a cgroup to another. | |
5970 | * | |
251f8c03 | 5971 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5972 | * will remain in init_css_set. This is safe because all tasks are |
5973 | * in the init_css_set before cg_links is enabled and there's no | |
5974 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5975 | */ |
817929ec | 5976 | if (use_task_css_set_links) { |
eaf797ab TH |
5977 | struct css_set *cset; |
5978 | ||
82d6489d | 5979 | spin_lock_irq(&css_set_lock); |
0e1d768f | 5980 | cset = task_css_set(current); |
eaf797ab | 5981 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5982 | get_css_set(cset); |
f6d7d049 | 5983 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5984 | } |
82d6489d | 5985 | spin_unlock_irq(&css_set_lock); |
817929ec | 5986 | } |
5edee61e TH |
5987 | |
5988 | /* | |
5989 | * Call ss->fork(). This must happen after @child is linked on | |
5990 | * css_set; otherwise, @child might change state between ->fork() | |
5991 | * and addition to css_set. | |
5992 | */ | |
b4e0eeaf | 5993 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5994 | ss->fork(child); |
b4e0eeaf | 5995 | } while_each_subsys_mask(); |
817929ec | 5996 | } |
5edee61e | 5997 | |
b4f48b63 PM |
5998 | /** |
5999 | * cgroup_exit - detach cgroup from exiting task | |
6000 | * @tsk: pointer to task_struct of exiting process | |
6001 | * | |
6002 | * Description: Detach cgroup from @tsk and release it. | |
6003 | * | |
6004 | * Note that cgroups marked notify_on_release force every task in | |
6005 | * them to take the global cgroup_mutex mutex when exiting. | |
6006 | * This could impact scaling on very large systems. Be reluctant to | |
6007 | * use notify_on_release cgroups where very high task exit scaling | |
6008 | * is required on large systems. | |
6009 | * | |
0e1d768f TH |
6010 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
6011 | * call cgroup_exit() while the task is still competent to handle | |
6012 | * notify_on_release(), then leave the task attached to the root cgroup in | |
6013 | * each hierarchy for the remainder of its exit. No need to bother with | |
6014 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 6015 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 6016 | */ |
1ec41830 | 6017 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 6018 | { |
30159ec7 | 6019 | struct cgroup_subsys *ss; |
5abb8855 | 6020 | struct css_set *cset; |
d41d5a01 | 6021 | int i; |
817929ec PM |
6022 | |
6023 | /* | |
0e1d768f | 6024 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 6025 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 6026 | */ |
0de0942d TH |
6027 | cset = task_css_set(tsk); |
6028 | ||
817929ec | 6029 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 6030 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 6031 | css_set_move_task(tsk, cset, NULL, false); |
82d6489d | 6032 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
6033 | } else { |
6034 | get_css_set(cset); | |
817929ec PM |
6035 | } |
6036 | ||
cb4a3167 | 6037 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 6038 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 6039 | ss->exit(tsk); |
b4e0eeaf | 6040 | } while_each_subsys_mask(); |
2e91fa7f | 6041 | } |
30159ec7 | 6042 | |
2e91fa7f TH |
6043 | void cgroup_free(struct task_struct *task) |
6044 | { | |
6045 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
6046 | struct cgroup_subsys *ss; |
6047 | int ssid; | |
6048 | ||
b4e0eeaf | 6049 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 6050 | ss->free(task); |
b4e0eeaf | 6051 | } while_each_subsys_mask(); |
d41d5a01 | 6052 | |
2e91fa7f | 6053 | put_css_set(cset); |
b4f48b63 | 6054 | } |
697f4161 | 6055 | |
bd89aabc | 6056 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 6057 | { |
27bd4dbb | 6058 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
6059 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
6060 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
6061 | } |
6062 | ||
81a6a5cd PM |
6063 | /* |
6064 | * Notify userspace when a cgroup is released, by running the | |
6065 | * configured release agent with the name of the cgroup (path | |
6066 | * relative to the root of cgroup file system) as the argument. | |
6067 | * | |
6068 | * Most likely, this user command will try to rmdir this cgroup. | |
6069 | * | |
6070 | * This races with the possibility that some other task will be | |
6071 | * attached to this cgroup before it is removed, or that some other | |
6072 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
6073 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
6074 | * unused, and this cgroup will be reprieved from its death sentence, | |
6075 | * to continue to serve a useful existence. Next time it's released, | |
6076 | * we will get notified again, if it still has 'notify_on_release' set. | |
6077 | * | |
6078 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
6079 | * means only wait until the task is successfully execve()'d. The | |
6080 | * separate release agent task is forked by call_usermodehelper(), | |
6081 | * then control in this thread returns here, without waiting for the | |
6082 | * release agent task. We don't bother to wait because the caller of | |
6083 | * this routine has no use for the exit status of the release agent | |
6084 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 6085 | */ |
81a6a5cd PM |
6086 | static void cgroup_release_agent(struct work_struct *work) |
6087 | { | |
971ff493 ZL |
6088 | struct cgroup *cgrp = |
6089 | container_of(work, struct cgroup, release_agent_work); | |
4c737b41 | 6090 | char *pathbuf = NULL, *agentbuf = NULL; |
971ff493 | 6091 | char *argv[3], *envp[3]; |
4c737b41 | 6092 | int ret; |
971ff493 | 6093 | |
81a6a5cd | 6094 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
6095 | |
6096 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
6097 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
6098 | if (!pathbuf || !agentbuf) | |
6099 | goto out; | |
6100 | ||
82d6489d | 6101 | spin_lock_irq(&css_set_lock); |
4c737b41 | 6102 | ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); |
82d6489d | 6103 | spin_unlock_irq(&css_set_lock); |
e0223003 | 6104 | if (ret < 0 || ret >= PATH_MAX) |
971ff493 ZL |
6105 | goto out; |
6106 | ||
6107 | argv[0] = agentbuf; | |
4c737b41 | 6108 | argv[1] = pathbuf; |
971ff493 ZL |
6109 | argv[2] = NULL; |
6110 | ||
6111 | /* minimal command environment */ | |
6112 | envp[0] = "HOME=/"; | |
6113 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
6114 | envp[2] = NULL; | |
6115 | ||
81a6a5cd | 6116 | mutex_unlock(&cgroup_mutex); |
971ff493 | 6117 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 6118 | goto out_free; |
971ff493 | 6119 | out: |
81a6a5cd | 6120 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 6121 | out_free: |
971ff493 ZL |
6122 | kfree(agentbuf); |
6123 | kfree(pathbuf); | |
81a6a5cd | 6124 | } |
8bab8dde PM |
6125 | |
6126 | static int __init cgroup_disable(char *str) | |
6127 | { | |
30159ec7 | 6128 | struct cgroup_subsys *ss; |
8bab8dde | 6129 | char *token; |
30159ec7 | 6130 | int i; |
8bab8dde PM |
6131 | |
6132 | while ((token = strsep(&str, ",")) != NULL) { | |
6133 | if (!*token) | |
6134 | continue; | |
be45c900 | 6135 | |
3ed80a62 | 6136 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6137 | if (strcmp(token, ss->name) && |
6138 | strcmp(token, ss->legacy_name)) | |
6139 | continue; | |
a3e72739 | 6140 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
6141 | } |
6142 | } | |
6143 | return 1; | |
6144 | } | |
6145 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6146 | |
223ffb29 JW |
6147 | static int __init cgroup_no_v1(char *str) |
6148 | { | |
6149 | struct cgroup_subsys *ss; | |
6150 | char *token; | |
6151 | int i; | |
6152 | ||
6153 | while ((token = strsep(&str, ",")) != NULL) { | |
6154 | if (!*token) | |
6155 | continue; | |
6156 | ||
6157 | if (!strcmp(token, "all")) { | |
6e5c8307 | 6158 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
6159 | break; |
6160 | } | |
6161 | ||
6162 | for_each_subsys(ss, i) { | |
6163 | if (strcmp(token, ss->name) && | |
6164 | strcmp(token, ss->legacy_name)) | |
6165 | continue; | |
6166 | ||
6167 | cgroup_no_v1_mask |= 1 << i; | |
6168 | } | |
6169 | } | |
6170 | return 1; | |
6171 | } | |
6172 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
6173 | ||
b77d7b60 | 6174 | /** |
ec903c0c | 6175 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6176 | * @dentry: directory dentry of interest |
6177 | * @ss: subsystem of interest | |
b77d7b60 | 6178 | * |
5a17f543 TH |
6179 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6180 | * to get the corresponding css and return it. If such css doesn't exist | |
6181 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6182 | */ |
ec903c0c TH |
6183 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6184 | struct cgroup_subsys *ss) | |
e5d1367f | 6185 | { |
2bd59d48 | 6186 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6187 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6188 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6189 | struct cgroup *cgrp; |
e5d1367f | 6190 | |
35cf0836 | 6191 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6192 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6193 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6194 | return ERR_PTR(-EBADF); |
6195 | ||
5a17f543 TH |
6196 | rcu_read_lock(); |
6197 | ||
2bd59d48 TH |
6198 | /* |
6199 | * This path doesn't originate from kernfs and @kn could already | |
6200 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6201 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
6202 | */ |
6203 | cgrp = rcu_dereference(kn->priv); | |
6204 | if (cgrp) | |
6205 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6206 | |
ec903c0c | 6207 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6208 | css = ERR_PTR(-ENOENT); |
6209 | ||
6210 | rcu_read_unlock(); | |
6211 | return css; | |
e5d1367f | 6212 | } |
e5d1367f | 6213 | |
1cb650b9 LZ |
6214 | /** |
6215 | * css_from_id - lookup css by id | |
6216 | * @id: the cgroup id | |
6217 | * @ss: cgroup subsys to be looked into | |
6218 | * | |
6219 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6220 | * Should be called under rcu_read_lock(). | |
6221 | */ | |
6222 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6223 | { | |
6fa4918d | 6224 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 6225 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
6226 | } |
6227 | ||
16af4396 TH |
6228 | /** |
6229 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6230 | * @path: path on the default hierarchy | |
6231 | * | |
6232 | * Find the cgroup at @path on the default hierarchy, increment its | |
6233 | * reference count and return it. Returns pointer to the found cgroup on | |
6234 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6235 | * if @path points to a non-directory. | |
6236 | */ | |
6237 | struct cgroup *cgroup_get_from_path(const char *path) | |
6238 | { | |
6239 | struct kernfs_node *kn; | |
6240 | struct cgroup *cgrp; | |
6241 | ||
6242 | mutex_lock(&cgroup_mutex); | |
6243 | ||
6244 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6245 | if (kn) { | |
6246 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6247 | cgrp = kn->priv; | |
6248 | cgroup_get(cgrp); | |
6249 | } else { | |
6250 | cgrp = ERR_PTR(-ENOTDIR); | |
6251 | } | |
6252 | kernfs_put(kn); | |
6253 | } else { | |
6254 | cgrp = ERR_PTR(-ENOENT); | |
6255 | } | |
6256 | ||
6257 | mutex_unlock(&cgroup_mutex); | |
6258 | return cgrp; | |
6259 | } | |
6260 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6261 | ||
1f3fe7eb MKL |
6262 | /** |
6263 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
6264 | * @fd: fd obtained by open(cgroup2_dir) | |
6265 | * | |
6266 | * Find the cgroup from a fd which should be obtained | |
6267 | * by opening a cgroup directory. Returns a pointer to the | |
6268 | * cgroup on success. ERR_PTR is returned if the cgroup | |
6269 | * cannot be found. | |
6270 | */ | |
6271 | struct cgroup *cgroup_get_from_fd(int fd) | |
6272 | { | |
6273 | struct cgroup_subsys_state *css; | |
6274 | struct cgroup *cgrp; | |
6275 | struct file *f; | |
6276 | ||
6277 | f = fget_raw(fd); | |
6278 | if (!f) | |
6279 | return ERR_PTR(-EBADF); | |
6280 | ||
6281 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
6282 | fput(f); | |
6283 | if (IS_ERR(css)) | |
6284 | return ERR_CAST(css); | |
6285 | ||
6286 | cgrp = css->cgroup; | |
6287 | if (!cgroup_on_dfl(cgrp)) { | |
6288 | cgroup_put(cgrp); | |
6289 | return ERR_PTR(-EBADF); | |
6290 | } | |
6291 | ||
6292 | return cgrp; | |
6293 | } | |
6294 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
6295 | ||
bd1060a1 TH |
6296 | /* |
6297 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6298 | * definition in cgroup-defs.h. | |
6299 | */ | |
6300 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6301 | ||
6302 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6303 | ||
3fa4cc9c | 6304 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6305 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6306 | ||
6307 | void cgroup_sk_alloc_disable(void) | |
6308 | { | |
6309 | if (cgroup_sk_alloc_disabled) | |
6310 | return; | |
6311 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6312 | cgroup_sk_alloc_disabled = true; | |
6313 | } | |
6314 | ||
6315 | #else | |
6316 | ||
6317 | #define cgroup_sk_alloc_disabled false | |
6318 | ||
6319 | #endif | |
6320 | ||
6321 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6322 | { | |
6323 | if (cgroup_sk_alloc_disabled) | |
6324 | return; | |
6325 | ||
d979a39d JW |
6326 | /* Socket clone path */ |
6327 | if (skcd->val) { | |
6328 | cgroup_get(sock_cgroup_ptr(skcd)); | |
6329 | return; | |
6330 | } | |
6331 | ||
bd1060a1 TH |
6332 | rcu_read_lock(); |
6333 | ||
6334 | while (true) { | |
6335 | struct css_set *cset; | |
6336 | ||
6337 | cset = task_css_set(current); | |
6338 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6339 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6340 | break; | |
6341 | } | |
6342 | cpu_relax(); | |
6343 | } | |
6344 | ||
6345 | rcu_read_unlock(); | |
6346 | } | |
6347 | ||
6348 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6349 | { | |
6350 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6351 | } | |
6352 | ||
6353 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6354 | ||
a79a908f AK |
6355 | /* cgroup namespaces */ |
6356 | ||
d08311dd EB |
6357 | static struct ucounts *inc_cgroup_namespaces(struct user_namespace *ns) |
6358 | { | |
6359 | return inc_ucount(ns, current_euid(), UCOUNT_CGROUP_NAMESPACES); | |
6360 | } | |
6361 | ||
6362 | static void dec_cgroup_namespaces(struct ucounts *ucounts) | |
6363 | { | |
6364 | dec_ucount(ucounts, UCOUNT_CGROUP_NAMESPACES); | |
6365 | } | |
6366 | ||
a79a908f AK |
6367 | static struct cgroup_namespace *alloc_cgroup_ns(void) |
6368 | { | |
6369 | struct cgroup_namespace *new_ns; | |
6370 | int ret; | |
6371 | ||
6372 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); | |
6373 | if (!new_ns) | |
6374 | return ERR_PTR(-ENOMEM); | |
6375 | ret = ns_alloc_inum(&new_ns->ns); | |
6376 | if (ret) { | |
6377 | kfree(new_ns); | |
6378 | return ERR_PTR(ret); | |
6379 | } | |
6380 | atomic_set(&new_ns->count, 1); | |
6381 | new_ns->ns.ops = &cgroupns_operations; | |
6382 | return new_ns; | |
6383 | } | |
6384 | ||
6385 | void free_cgroup_ns(struct cgroup_namespace *ns) | |
6386 | { | |
6387 | put_css_set(ns->root_cset); | |
d08311dd | 6388 | dec_cgroup_namespaces(ns->ucounts); |
a79a908f AK |
6389 | put_user_ns(ns->user_ns); |
6390 | ns_free_inum(&ns->ns); | |
6391 | kfree(ns); | |
6392 | } | |
6393 | EXPORT_SYMBOL(free_cgroup_ns); | |
6394 | ||
6395 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, | |
6396 | struct user_namespace *user_ns, | |
6397 | struct cgroup_namespace *old_ns) | |
6398 | { | |
fa5ff8a1 | 6399 | struct cgroup_namespace *new_ns; |
d08311dd | 6400 | struct ucounts *ucounts; |
fa5ff8a1 | 6401 | struct css_set *cset; |
a79a908f AK |
6402 | |
6403 | BUG_ON(!old_ns); | |
6404 | ||
6405 | if (!(flags & CLONE_NEWCGROUP)) { | |
6406 | get_cgroup_ns(old_ns); | |
6407 | return old_ns; | |
6408 | } | |
6409 | ||
6410 | /* Allow only sysadmin to create cgroup namespace. */ | |
a79a908f | 6411 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) |
fa5ff8a1 | 6412 | return ERR_PTR(-EPERM); |
a79a908f | 6413 | |
d08311dd EB |
6414 | ucounts = inc_cgroup_namespaces(user_ns); |
6415 | if (!ucounts) | |
df75e774 | 6416 | return ERR_PTR(-ENOSPC); |
d08311dd | 6417 | |
7bd88308 | 6418 | /* It is not safe to take cgroup_mutex here */ |
82d6489d | 6419 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
6420 | cset = task_css_set(current); |
6421 | get_css_set(cset); | |
82d6489d | 6422 | spin_unlock_irq(&css_set_lock); |
a79a908f | 6423 | |
a79a908f | 6424 | new_ns = alloc_cgroup_ns(); |
d2202557 | 6425 | if (IS_ERR(new_ns)) { |
fa5ff8a1 | 6426 | put_css_set(cset); |
d08311dd | 6427 | dec_cgroup_namespaces(ucounts); |
fa5ff8a1 | 6428 | return new_ns; |
d2202557 | 6429 | } |
a79a908f AK |
6430 | |
6431 | new_ns->user_ns = get_user_ns(user_ns); | |
d08311dd | 6432 | new_ns->ucounts = ucounts; |
a79a908f AK |
6433 | new_ns->root_cset = cset; |
6434 | ||
6435 | return new_ns; | |
a79a908f AK |
6436 | } |
6437 | ||
6438 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) | |
6439 | { | |
6440 | return container_of(ns, struct cgroup_namespace, ns); | |
6441 | } | |
6442 | ||
a0530e08 | 6443 | static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
a79a908f | 6444 | { |
a0530e08 AK |
6445 | struct cgroup_namespace *cgroup_ns = to_cg_ns(ns); |
6446 | ||
6447 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) || | |
6448 | !ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN)) | |
6449 | return -EPERM; | |
6450 | ||
6451 | /* Don't need to do anything if we are attaching to our own cgroupns. */ | |
6452 | if (cgroup_ns == nsproxy->cgroup_ns) | |
6453 | return 0; | |
6454 | ||
6455 | get_cgroup_ns(cgroup_ns); | |
6456 | put_cgroup_ns(nsproxy->cgroup_ns); | |
6457 | nsproxy->cgroup_ns = cgroup_ns; | |
6458 | ||
6459 | return 0; | |
a79a908f AK |
6460 | } |
6461 | ||
6462 | static struct ns_common *cgroupns_get(struct task_struct *task) | |
6463 | { | |
6464 | struct cgroup_namespace *ns = NULL; | |
6465 | struct nsproxy *nsproxy; | |
6466 | ||
6467 | task_lock(task); | |
6468 | nsproxy = task->nsproxy; | |
6469 | if (nsproxy) { | |
6470 | ns = nsproxy->cgroup_ns; | |
6471 | get_cgroup_ns(ns); | |
6472 | } | |
6473 | task_unlock(task); | |
6474 | ||
6475 | return ns ? &ns->ns : NULL; | |
6476 | } | |
6477 | ||
6478 | static void cgroupns_put(struct ns_common *ns) | |
6479 | { | |
6480 | put_cgroup_ns(to_cg_ns(ns)); | |
6481 | } | |
6482 | ||
bcac25a5 AV |
6483 | static struct user_namespace *cgroupns_owner(struct ns_common *ns) |
6484 | { | |
6485 | return to_cg_ns(ns)->user_ns; | |
6486 | } | |
6487 | ||
a79a908f AK |
6488 | const struct proc_ns_operations cgroupns_operations = { |
6489 | .name = "cgroup", | |
6490 | .type = CLONE_NEWCGROUP, | |
6491 | .get = cgroupns_get, | |
6492 | .put = cgroupns_put, | |
6493 | .install = cgroupns_install, | |
bcac25a5 | 6494 | .owner = cgroupns_owner, |
a79a908f AK |
6495 | }; |
6496 | ||
6497 | static __init int cgroup_namespaces_init(void) | |
6498 | { | |
6499 | return 0; | |
6500 | } | |
6501 | subsys_initcall(cgroup_namespaces_init); | |
6502 | ||
30070984 DM |
6503 | #ifdef CONFIG_CGROUP_BPF |
6504 | void cgroup_bpf_update(struct cgroup *cgrp, | |
6505 | struct bpf_prog *prog, | |
6506 | enum bpf_attach_type type) | |
6507 | { | |
6508 | struct cgroup *parent = cgroup_parent(cgrp); | |
6509 | ||
6510 | mutex_lock(&cgroup_mutex); | |
6511 | __cgroup_bpf_update(cgrp, parent, prog, type); | |
6512 | mutex_unlock(&cgroup_mutex); | |
6513 | } | |
6514 | #endif /* CONFIG_CGROUP_BPF */ | |
6515 | ||
fe693435 | 6516 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6517 | static struct cgroup_subsys_state * |
6518 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6519 | { |
6520 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6521 | ||
6522 | if (!css) | |
6523 | return ERR_PTR(-ENOMEM); | |
6524 | ||
6525 | return css; | |
6526 | } | |
6527 | ||
eb95419b | 6528 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6529 | { |
eb95419b | 6530 | kfree(css); |
fe693435 PM |
6531 | } |
6532 | ||
182446d0 TH |
6533 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6534 | struct cftype *cft) | |
fe693435 | 6535 | { |
182446d0 | 6536 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6537 | } |
6538 | ||
182446d0 TH |
6539 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6540 | struct cftype *cft) | |
fe693435 PM |
6541 | { |
6542 | return (u64)(unsigned long)current->cgroups; | |
6543 | } | |
6544 | ||
182446d0 | 6545 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6546 | struct cftype *cft) |
fe693435 PM |
6547 | { |
6548 | u64 count; | |
6549 | ||
6550 | rcu_read_lock(); | |
a8ad805c | 6551 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6552 | rcu_read_unlock(); |
6553 | return count; | |
6554 | } | |
6555 | ||
2da8ca82 | 6556 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6557 | { |
69d0206c | 6558 | struct cgrp_cset_link *link; |
5abb8855 | 6559 | struct css_set *cset; |
e61734c5 TH |
6560 | char *name_buf; |
6561 | ||
6562 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6563 | if (!name_buf) | |
6564 | return -ENOMEM; | |
7717f7ba | 6565 | |
82d6489d | 6566 | spin_lock_irq(&css_set_lock); |
7717f7ba | 6567 | rcu_read_lock(); |
5abb8855 | 6568 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6569 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6570 | struct cgroup *c = link->cgrp; |
7717f7ba | 6571 | |
a2dd4247 | 6572 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6573 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6574 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6575 | } |
6576 | rcu_read_unlock(); | |
82d6489d | 6577 | spin_unlock_irq(&css_set_lock); |
e61734c5 | 6578 | kfree(name_buf); |
7717f7ba PM |
6579 | return 0; |
6580 | } | |
6581 | ||
6582 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6583 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6584 | { |
2da8ca82 | 6585 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6586 | struct cgrp_cset_link *link; |
7717f7ba | 6587 | |
82d6489d | 6588 | spin_lock_irq(&css_set_lock); |
182446d0 | 6589 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6590 | struct css_set *cset = link->cset; |
7717f7ba PM |
6591 | struct task_struct *task; |
6592 | int count = 0; | |
c7561128 | 6593 | |
5abb8855 | 6594 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6595 | |
5abb8855 | 6596 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6597 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6598 | goto overflow; | |
6599 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6600 | } | |
6601 | ||
6602 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6603 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6604 | goto overflow; | |
6605 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6606 | } |
c7561128 TH |
6607 | continue; |
6608 | overflow: | |
6609 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6610 | } |
82d6489d | 6611 | spin_unlock_irq(&css_set_lock); |
7717f7ba PM |
6612 | return 0; |
6613 | } | |
6614 | ||
182446d0 | 6615 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6616 | { |
27bd4dbb | 6617 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6618 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6619 | } |
6620 | ||
6621 | static struct cftype debug_files[] = { | |
fe693435 PM |
6622 | { |
6623 | .name = "taskcount", | |
6624 | .read_u64 = debug_taskcount_read, | |
6625 | }, | |
6626 | ||
6627 | { | |
6628 | .name = "current_css_set", | |
6629 | .read_u64 = current_css_set_read, | |
6630 | }, | |
6631 | ||
6632 | { | |
6633 | .name = "current_css_set_refcount", | |
6634 | .read_u64 = current_css_set_refcount_read, | |
6635 | }, | |
6636 | ||
7717f7ba PM |
6637 | { |
6638 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6639 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6640 | }, |
6641 | ||
6642 | { | |
6643 | .name = "cgroup_css_links", | |
2da8ca82 | 6644 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6645 | }, |
6646 | ||
fe693435 PM |
6647 | { |
6648 | .name = "releasable", | |
6649 | .read_u64 = releasable_read, | |
6650 | }, | |
fe693435 | 6651 | |
4baf6e33 TH |
6652 | { } /* terminate */ |
6653 | }; | |
fe693435 | 6654 | |
073219e9 | 6655 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6656 | .css_alloc = debug_css_alloc, |
6657 | .css_free = debug_css_free, | |
5577964e | 6658 | .legacy_cftypes = debug_files, |
fe693435 PM |
6659 | }; |
6660 | #endif /* CONFIG_CGROUP_DEBUG */ |