]>
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> |
96d365e0 | 48 | #include <linux/rwsem.h> |
d59cfc09 | 49 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 50 | #include <linux/string.h> |
bbcb81d0 | 51 | #include <linux/sort.h> |
81a6a5cd | 52 | #include <linux/kmod.h> |
846c7bb0 BS |
53 | #include <linux/delayacct.h> |
54 | #include <linux/cgroupstats.h> | |
0ac801fe | 55 | #include <linux/hashtable.h> |
096b7fe0 | 56 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 57 | #include <linux/idr.h> |
d1d9fd33 | 58 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 59 | #include <linux/kthread.h> |
776f02fa | 60 | #include <linux/delay.h> |
846c7bb0 | 61 | |
60063497 | 62 | #include <linux/atomic.h> |
ddbcc7e8 | 63 | |
b1a21367 TH |
64 | /* |
65 | * pidlists linger the following amount before being destroyed. The goal | |
66 | * is avoiding frequent destruction in the middle of consecutive read calls | |
67 | * Expiring in the middle is a performance problem not a correctness one. | |
68 | * 1 sec should be enough. | |
69 | */ | |
70 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
71 | ||
8d7e6fb0 TH |
72 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
73 | MAX_CFTYPE_NAME + 2) | |
74 | ||
e25e2cbb TH |
75 | /* |
76 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
77 | * hierarchy must be performed while holding it. | |
78 | * | |
0e1d768f TH |
79 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
80 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 81 | * |
0e1d768f TH |
82 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
83 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 84 | */ |
2219449a TH |
85 | #ifdef CONFIG_PROVE_RCU |
86 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
87 | DECLARE_RWSEM(css_set_rwsem); |
88 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
89 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 90 | #else |
81a6a5cd | 91 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 92 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
93 | #endif |
94 | ||
6fa4918d | 95 | /* |
15a4c835 TH |
96 | * Protects cgroup_idr and css_idr so that IDs can be released without |
97 | * grabbing cgroup_mutex. | |
6fa4918d TH |
98 | */ |
99 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
100 | ||
69e943b7 TH |
101 | /* |
102 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
103 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 106 | |
d59cfc09 TH |
107 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
108 | ||
8353da1f | 109 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
110 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
111 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 112 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 113 | |
e5fca243 TH |
114 | /* |
115 | * cgroup destruction makes heavy use of work items and there can be a lot | |
116 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
117 | * destruction work items don't end up filling up max_active of system_wq | |
118 | * which may lead to deadlock. | |
119 | */ | |
120 | static struct workqueue_struct *cgroup_destroy_wq; | |
121 | ||
b1a21367 TH |
122 | /* |
123 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
124 | * separate workqueue as flush domain. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
127 | ||
3ed80a62 | 128 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 129 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 130 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
131 | #include <linux/cgroup_subsys.h> |
132 | }; | |
073219e9 TH |
133 | #undef SUBSYS |
134 | ||
135 | /* array of cgroup subsystem names */ | |
136 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
137 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
138 | #include <linux/cgroup_subsys.h> |
139 | }; | |
073219e9 | 140 | #undef SUBSYS |
ddbcc7e8 | 141 | |
ddbcc7e8 | 142 | /* |
3dd06ffa | 143 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
144 | * unattached - it never has more than a single cgroup, and all tasks are |
145 | * part of that cgroup. | |
ddbcc7e8 | 146 | */ |
a2dd4247 | 147 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 148 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 149 | |
a2dd4247 TH |
150 | /* |
151 | * The default hierarchy always exists but is hidden until mounted for the | |
152 | * first time. This is for backward compatibility. | |
153 | */ | |
154 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 | 155 | |
a8ddc821 TH |
156 | /* |
157 | * Set by the boot param of the same name and makes subsystems with NULL | |
158 | * ->dfl_files to use ->legacy_files on the default hierarchy. | |
159 | */ | |
160 | static bool cgroup_legacy_files_on_dfl; | |
161 | ||
5533e011 | 162 | /* some controllers are not supported in the default hierarchy */ |
8ab456ac | 163 | static unsigned long cgrp_dfl_root_inhibit_ss_mask; |
5533e011 | 164 | |
ddbcc7e8 PM |
165 | /* The list of hierarchy roots */ |
166 | ||
9871bf95 TH |
167 | static LIST_HEAD(cgroup_roots); |
168 | static int cgroup_root_count; | |
ddbcc7e8 | 169 | |
3417ae1f | 170 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 171 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 172 | |
794611a1 | 173 | /* |
0cb51d71 TH |
174 | * Assign a monotonically increasing serial number to csses. It guarantees |
175 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
176 | * Also, as csses are always appended to the parent's ->children list, it | |
177 | * guarantees that sibling csses are always sorted in the ascending serial | |
178 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 179 | */ |
0cb51d71 | 180 | static u64 css_serial_nr_next = 1; |
794611a1 | 181 | |
cb4a3167 AS |
182 | /* |
183 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
184 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
185 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 186 | */ |
cb4a3167 AS |
187 | static unsigned long have_fork_callback __read_mostly; |
188 | static unsigned long have_exit_callback __read_mostly; | |
ddbcc7e8 | 189 | |
7e47682e AS |
190 | /* Ditto for the can_fork callback. */ |
191 | static unsigned long have_canfork_callback __read_mostly; | |
192 | ||
a14c6874 TH |
193 | static struct cftype cgroup_dfl_base_files[]; |
194 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 195 | |
3dd06ffa | 196 | static int rebind_subsystems(struct cgroup_root *dst_root, |
8ab456ac | 197 | unsigned long ss_mask); |
42809dd4 | 198 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
199 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
200 | bool visible); | |
9d755d33 | 201 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 202 | static void kill_css(struct cgroup_subsys_state *css); |
2bb566cb TH |
203 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
204 | bool is_add); | |
42809dd4 | 205 | |
6fa4918d TH |
206 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
207 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
208 | gfp_t gfp_mask) | |
209 | { | |
210 | int ret; | |
211 | ||
212 | idr_preload(gfp_mask); | |
54504e97 | 213 | spin_lock_bh(&cgroup_idr_lock); |
cf780b7d | 214 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_WAIT); |
54504e97 | 215 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
216 | idr_preload_end(); |
217 | return ret; | |
218 | } | |
219 | ||
220 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
221 | { | |
222 | void *ret; | |
223 | ||
54504e97 | 224 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 225 | ret = idr_replace(idr, ptr, id); |
54504e97 | 226 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
227 | return ret; |
228 | } | |
229 | ||
230 | static void cgroup_idr_remove(struct idr *idr, int id) | |
231 | { | |
54504e97 | 232 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 233 | idr_remove(idr, id); |
54504e97 | 234 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
235 | } |
236 | ||
d51f39b0 TH |
237 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
238 | { | |
239 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
240 | ||
241 | if (parent_css) | |
242 | return container_of(parent_css, struct cgroup, self); | |
243 | return NULL; | |
244 | } | |
245 | ||
95109b62 TH |
246 | /** |
247 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
248 | * @cgrp: the cgroup of interest | |
9d800df1 | 249 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 250 | * |
ca8bdcaf TH |
251 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
252 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
253 | * the caller is responsible for pinning the returned css if it wants to | |
254 | * keep accessing it outside the said locks. This function may return | |
255 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
256 | */ |
257 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 258 | struct cgroup_subsys *ss) |
95109b62 | 259 | { |
ca8bdcaf | 260 | if (ss) |
aec25020 | 261 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 262 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 263 | else |
9d800df1 | 264 | return &cgrp->self; |
95109b62 | 265 | } |
42809dd4 | 266 | |
aec3dfcb TH |
267 | /** |
268 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
269 | * @cgrp: the cgroup of interest | |
9d800df1 | 270 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 271 | * |
d0f702e6 | 272 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
273 | * as the matching css of the nearest ancestor including self which has @ss |
274 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
275 | * function is guaranteed to return non-NULL css. | |
276 | */ | |
277 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
278 | struct cgroup_subsys *ss) | |
279 | { | |
280 | lockdep_assert_held(&cgroup_mutex); | |
281 | ||
282 | if (!ss) | |
9d800df1 | 283 | return &cgrp->self; |
aec3dfcb TH |
284 | |
285 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
286 | return NULL; | |
287 | ||
eeecbd19 TH |
288 | /* |
289 | * This function is used while updating css associations and thus | |
290 | * can't test the csses directly. Use ->child_subsys_mask. | |
291 | */ | |
d51f39b0 TH |
292 | while (cgroup_parent(cgrp) && |
293 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
294 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
295 | |
296 | return cgroup_css(cgrp, ss); | |
95109b62 | 297 | } |
42809dd4 | 298 | |
eeecbd19 TH |
299 | /** |
300 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
301 | * @cgrp: the cgroup of interest | |
302 | * @ss: the subsystem of interest | |
303 | * | |
304 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
305 | * defined as the matching css of the nearest ancestor including self which | |
306 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
307 | * the root css is returned, so this function always returns a valid css. | |
308 | * The returned css must be put using css_put(). | |
309 | */ | |
310 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
311 | struct cgroup_subsys *ss) | |
312 | { | |
313 | struct cgroup_subsys_state *css; | |
314 | ||
315 | rcu_read_lock(); | |
316 | ||
317 | do { | |
318 | css = cgroup_css(cgrp, ss); | |
319 | ||
320 | if (css && css_tryget_online(css)) | |
321 | goto out_unlock; | |
322 | cgrp = cgroup_parent(cgrp); | |
323 | } while (cgrp); | |
324 | ||
325 | css = init_css_set.subsys[ss->id]; | |
326 | css_get(css); | |
327 | out_unlock: | |
328 | rcu_read_unlock(); | |
329 | return css; | |
330 | } | |
331 | ||
ddbcc7e8 | 332 | /* convenient tests for these bits */ |
54766d4a | 333 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 334 | { |
184faf32 | 335 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
336 | } |
337 | ||
b4168640 | 338 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 339 | { |
2bd59d48 | 340 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 341 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
342 | |
343 | /* | |
344 | * This is open and unprotected implementation of cgroup_css(). | |
345 | * seq_css() is only called from a kernfs file operation which has | |
346 | * an active reference on the file. Because all the subsystem | |
347 | * files are drained before a css is disassociated with a cgroup, | |
348 | * the matching css from the cgroup's subsys table is guaranteed to | |
349 | * be and stay valid until the enclosing operation is complete. | |
350 | */ | |
351 | if (cft->ss) | |
352 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
353 | else | |
9d800df1 | 354 | return &cgrp->self; |
59f5296b | 355 | } |
b4168640 | 356 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 357 | |
78574cf9 LZ |
358 | /** |
359 | * cgroup_is_descendant - test ancestry | |
360 | * @cgrp: the cgroup to be tested | |
361 | * @ancestor: possible ancestor of @cgrp | |
362 | * | |
363 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
364 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
365 | * and @ancestor are accessible. | |
366 | */ | |
367 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
368 | { | |
369 | while (cgrp) { | |
370 | if (cgrp == ancestor) | |
371 | return true; | |
d51f39b0 | 372 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
373 | } |
374 | return false; | |
375 | } | |
ddbcc7e8 | 376 | |
e9685a03 | 377 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 378 | { |
bd89aabc | 379 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
380 | } |
381 | ||
1c6727af TH |
382 | /** |
383 | * for_each_css - iterate all css's of a cgroup | |
384 | * @css: the iteration cursor | |
385 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
386 | * @cgrp: the target cgroup to iterate css's of | |
387 | * | |
aec3dfcb | 388 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
389 | */ |
390 | #define for_each_css(css, ssid, cgrp) \ | |
391 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
392 | if (!((css) = rcu_dereference_check( \ | |
393 | (cgrp)->subsys[(ssid)], \ | |
394 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
395 | else | |
396 | ||
aec3dfcb TH |
397 | /** |
398 | * for_each_e_css - iterate all effective css's of a cgroup | |
399 | * @css: the iteration cursor | |
400 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
401 | * @cgrp: the target cgroup to iterate css's of | |
402 | * | |
403 | * Should be called under cgroup_[tree_]mutex. | |
404 | */ | |
405 | #define for_each_e_css(css, ssid, cgrp) \ | |
406 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
407 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
408 | ; \ | |
409 | else | |
410 | ||
30159ec7 | 411 | /** |
3ed80a62 | 412 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 413 | * @ss: the iteration cursor |
780cd8b3 | 414 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 415 | */ |
780cd8b3 | 416 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
417 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
418 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 419 | |
cb4a3167 AS |
420 | /** |
421 | * for_each_subsys_which - filter for_each_subsys with a bitmask | |
422 | * @ss: the iteration cursor | |
423 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
424 | * @ss_maskp: a pointer to the bitmask | |
425 | * | |
426 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
427 | * mask is set to 1. | |
428 | */ | |
429 | #define for_each_subsys_which(ss, ssid, ss_maskp) \ | |
430 | if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \ | |
4a705c5c | 431 | (ssid) = 0; \ |
cb4a3167 AS |
432 | else \ |
433 | for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \ | |
434 | if (((ss) = cgroup_subsys[ssid]) && false) \ | |
435 | break; \ | |
436 | else | |
437 | ||
985ed670 TH |
438 | /* iterate across the hierarchies */ |
439 | #define for_each_root(root) \ | |
5549c497 | 440 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 441 | |
f8f22e53 TH |
442 | /* iterate over child cgrps, lock should be held throughout iteration */ |
443 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 444 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 445 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
446 | cgroup_is_dead(child); })) \ |
447 | ; \ | |
448 | else | |
7ae1bad9 | 449 | |
81a6a5cd | 450 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 451 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 452 | |
69d0206c TH |
453 | /* |
454 | * A cgroup can be associated with multiple css_sets as different tasks may | |
455 | * belong to different cgroups on different hierarchies. In the other | |
456 | * direction, a css_set is naturally associated with multiple cgroups. | |
457 | * This M:N relationship is represented by the following link structure | |
458 | * which exists for each association and allows traversing the associations | |
459 | * from both sides. | |
460 | */ | |
461 | struct cgrp_cset_link { | |
462 | /* the cgroup and css_set this link associates */ | |
463 | struct cgroup *cgrp; | |
464 | struct css_set *cset; | |
465 | ||
466 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
467 | struct list_head cset_link; | |
468 | ||
469 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
470 | struct list_head cgrp_link; | |
817929ec PM |
471 | }; |
472 | ||
172a2c06 TH |
473 | /* |
474 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
475 | * hierarchies being mounted. It contains a pointer to the root state |
476 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
477 | * reference-counted, to improve performance when child cgroups | |
478 | * haven't been created. | |
479 | */ | |
5024ae29 | 480 | struct css_set init_css_set = { |
172a2c06 TH |
481 | .refcount = ATOMIC_INIT(1), |
482 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
483 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
484 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
485 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
486 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
487 | }; | |
817929ec | 488 | |
172a2c06 | 489 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 490 | |
842b597e TH |
491 | /** |
492 | * cgroup_update_populated - updated populated count of a cgroup | |
493 | * @cgrp: the target cgroup | |
494 | * @populated: inc or dec populated count | |
495 | * | |
496 | * @cgrp is either getting the first task (css_set) or losing the last. | |
497 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
498 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
499 | * cgroup and all its descendants are empty. | |
500 | * | |
501 | * @cgrp's interface file "cgroup.populated" is zero if | |
502 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
503 | * changes from or to zero, userland is notified that the content of the | |
504 | * interface file has changed. This can be used to detect when @cgrp and | |
505 | * its descendants become populated or empty. | |
506 | */ | |
507 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
508 | { | |
509 | lockdep_assert_held(&css_set_rwsem); | |
510 | ||
511 | do { | |
512 | bool trigger; | |
513 | ||
514 | if (populated) | |
515 | trigger = !cgrp->populated_cnt++; | |
516 | else | |
517 | trigger = !--cgrp->populated_cnt; | |
518 | ||
519 | if (!trigger) | |
520 | break; | |
521 | ||
522 | if (cgrp->populated_kn) | |
523 | kernfs_notify(cgrp->populated_kn); | |
d51f39b0 | 524 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
525 | } while (cgrp); |
526 | } | |
527 | ||
7717f7ba PM |
528 | /* |
529 | * hash table for cgroup groups. This improves the performance to find | |
530 | * an existing css_set. This hash doesn't (currently) take into | |
531 | * account cgroups in empty hierarchies. | |
532 | */ | |
472b1053 | 533 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 534 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 535 | |
0ac801fe | 536 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 537 | { |
0ac801fe | 538 | unsigned long key = 0UL; |
30159ec7 TH |
539 | struct cgroup_subsys *ss; |
540 | int i; | |
472b1053 | 541 | |
30159ec7 | 542 | for_each_subsys(ss, i) |
0ac801fe LZ |
543 | key += (unsigned long)css[i]; |
544 | key = (key >> 16) ^ key; | |
472b1053 | 545 | |
0ac801fe | 546 | return key; |
472b1053 LZ |
547 | } |
548 | ||
a25eb52e | 549 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 550 | { |
69d0206c | 551 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
552 | struct cgroup_subsys *ss; |
553 | int ssid; | |
5abb8855 | 554 | |
89c5509b TH |
555 | lockdep_assert_held(&css_set_rwsem); |
556 | ||
557 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 558 | return; |
81a6a5cd | 559 | |
2c6ab6d2 | 560 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
561 | for_each_subsys(ss, ssid) |
562 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 563 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
564 | css_set_count--; |
565 | ||
69d0206c | 566 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 567 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 568 | |
69d0206c TH |
569 | list_del(&link->cset_link); |
570 | list_del(&link->cgrp_link); | |
71b5707e | 571 | |
96d365e0 | 572 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
573 | if (list_empty(&cgrp->cset_links)) { |
574 | cgroup_update_populated(cgrp, false); | |
a25eb52e | 575 | check_for_release(cgrp); |
81a6a5cd | 576 | } |
2c6ab6d2 PM |
577 | |
578 | kfree(link); | |
81a6a5cd | 579 | } |
2c6ab6d2 | 580 | |
5abb8855 | 581 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
582 | } |
583 | ||
a25eb52e | 584 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
585 | { |
586 | /* | |
587 | * Ensure that the refcount doesn't hit zero while any readers | |
588 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
589 | * rwlock | |
590 | */ | |
591 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
592 | return; | |
593 | ||
594 | down_write(&css_set_rwsem); | |
a25eb52e | 595 | put_css_set_locked(cset); |
89c5509b TH |
596 | up_write(&css_set_rwsem); |
597 | } | |
598 | ||
817929ec PM |
599 | /* |
600 | * refcounted get/put for css_set objects | |
601 | */ | |
5abb8855 | 602 | static inline void get_css_set(struct css_set *cset) |
817929ec | 603 | { |
5abb8855 | 604 | atomic_inc(&cset->refcount); |
817929ec PM |
605 | } |
606 | ||
b326f9d0 | 607 | /** |
7717f7ba | 608 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
609 | * @cset: candidate css_set being tested |
610 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
611 | * @new_cgrp: cgroup that's being entered by the task |
612 | * @template: desired set of css pointers in css_set (pre-calculated) | |
613 | * | |
6f4b7e63 | 614 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
615 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
616 | */ | |
5abb8855 TH |
617 | static bool compare_css_sets(struct css_set *cset, |
618 | struct css_set *old_cset, | |
7717f7ba PM |
619 | struct cgroup *new_cgrp, |
620 | struct cgroup_subsys_state *template[]) | |
621 | { | |
622 | struct list_head *l1, *l2; | |
623 | ||
aec3dfcb TH |
624 | /* |
625 | * On the default hierarchy, there can be csets which are | |
626 | * associated with the same set of cgroups but different csses. | |
627 | * Let's first ensure that csses match. | |
628 | */ | |
629 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 630 | return false; |
7717f7ba PM |
631 | |
632 | /* | |
633 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
634 | * different cgroups in hierarchies. As different cgroups may |
635 | * share the same effective css, this comparison is always | |
636 | * necessary. | |
7717f7ba | 637 | */ |
69d0206c TH |
638 | l1 = &cset->cgrp_links; |
639 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 640 | while (1) { |
69d0206c | 641 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 642 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
643 | |
644 | l1 = l1->next; | |
645 | l2 = l2->next; | |
646 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
647 | if (l1 == &cset->cgrp_links) { |
648 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
649 | break; |
650 | } else { | |
69d0206c | 651 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
652 | } |
653 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
654 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
655 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
656 | cgrp1 = link1->cgrp; | |
657 | cgrp2 = link2->cgrp; | |
7717f7ba | 658 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 659 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
660 | |
661 | /* | |
662 | * If this hierarchy is the hierarchy of the cgroup | |
663 | * that's changing, then we need to check that this | |
664 | * css_set points to the new cgroup; if it's any other | |
665 | * hierarchy, then this css_set should point to the | |
666 | * same cgroup as the old css_set. | |
667 | */ | |
5abb8855 TH |
668 | if (cgrp1->root == new_cgrp->root) { |
669 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
670 | return false; |
671 | } else { | |
5abb8855 | 672 | if (cgrp1 != cgrp2) |
7717f7ba PM |
673 | return false; |
674 | } | |
675 | } | |
676 | return true; | |
677 | } | |
678 | ||
b326f9d0 TH |
679 | /** |
680 | * find_existing_css_set - init css array and find the matching css_set | |
681 | * @old_cset: the css_set that we're using before the cgroup transition | |
682 | * @cgrp: the cgroup that we're moving into | |
683 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 684 | */ |
5abb8855 TH |
685 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
686 | struct cgroup *cgrp, | |
687 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 688 | { |
3dd06ffa | 689 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 690 | struct cgroup_subsys *ss; |
5abb8855 | 691 | struct css_set *cset; |
0ac801fe | 692 | unsigned long key; |
b326f9d0 | 693 | int i; |
817929ec | 694 | |
aae8aab4 BB |
695 | /* |
696 | * Build the set of subsystem state objects that we want to see in the | |
697 | * new css_set. while subsystems can change globally, the entries here | |
698 | * won't change, so no need for locking. | |
699 | */ | |
30159ec7 | 700 | for_each_subsys(ss, i) { |
f392e51c | 701 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
702 | /* |
703 | * @ss is in this hierarchy, so we want the | |
704 | * effective css from @cgrp. | |
705 | */ | |
706 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 707 | } else { |
aec3dfcb TH |
708 | /* |
709 | * @ss is not in this hierarchy, so we don't want | |
710 | * to change the css. | |
711 | */ | |
5abb8855 | 712 | template[i] = old_cset->subsys[i]; |
817929ec PM |
713 | } |
714 | } | |
715 | ||
0ac801fe | 716 | key = css_set_hash(template); |
5abb8855 TH |
717 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
718 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
719 | continue; |
720 | ||
721 | /* This css_set matches what we need */ | |
5abb8855 | 722 | return cset; |
472b1053 | 723 | } |
817929ec PM |
724 | |
725 | /* No existing cgroup group matched */ | |
726 | return NULL; | |
727 | } | |
728 | ||
69d0206c | 729 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 730 | { |
69d0206c | 731 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 732 | |
69d0206c TH |
733 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
734 | list_del(&link->cset_link); | |
36553434 LZ |
735 | kfree(link); |
736 | } | |
737 | } | |
738 | ||
69d0206c TH |
739 | /** |
740 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
741 | * @count: the number of links to allocate | |
742 | * @tmp_links: list_head the allocated links are put on | |
743 | * | |
744 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
745 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 746 | */ |
69d0206c | 747 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 748 | { |
69d0206c | 749 | struct cgrp_cset_link *link; |
817929ec | 750 | int i; |
69d0206c TH |
751 | |
752 | INIT_LIST_HEAD(tmp_links); | |
753 | ||
817929ec | 754 | for (i = 0; i < count; i++) { |
f4f4be2b | 755 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 756 | if (!link) { |
69d0206c | 757 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
758 | return -ENOMEM; |
759 | } | |
69d0206c | 760 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
761 | } |
762 | return 0; | |
763 | } | |
764 | ||
c12f65d4 LZ |
765 | /** |
766 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 767 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 768 | * @cset: the css_set to be linked |
c12f65d4 LZ |
769 | * @cgrp: the destination cgroup |
770 | */ | |
69d0206c TH |
771 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
772 | struct cgroup *cgrp) | |
c12f65d4 | 773 | { |
69d0206c | 774 | struct cgrp_cset_link *link; |
c12f65d4 | 775 | |
69d0206c | 776 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
777 | |
778 | if (cgroup_on_dfl(cgrp)) | |
779 | cset->dfl_cgrp = cgrp; | |
780 | ||
69d0206c TH |
781 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
782 | link->cset = cset; | |
7717f7ba | 783 | link->cgrp = cgrp; |
842b597e TH |
784 | |
785 | if (list_empty(&cgrp->cset_links)) | |
786 | cgroup_update_populated(cgrp, true); | |
69d0206c | 787 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 788 | |
7717f7ba PM |
789 | /* |
790 | * Always add links to the tail of the list so that the list | |
791 | * is sorted by order of hierarchy creation | |
792 | */ | |
69d0206c | 793 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
794 | } |
795 | ||
b326f9d0 TH |
796 | /** |
797 | * find_css_set - return a new css_set with one cgroup updated | |
798 | * @old_cset: the baseline css_set | |
799 | * @cgrp: the cgroup to be updated | |
800 | * | |
801 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
802 | * substituted into the appropriate hierarchy. | |
817929ec | 803 | */ |
5abb8855 TH |
804 | static struct css_set *find_css_set(struct css_set *old_cset, |
805 | struct cgroup *cgrp) | |
817929ec | 806 | { |
b326f9d0 | 807 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 808 | struct css_set *cset; |
69d0206c TH |
809 | struct list_head tmp_links; |
810 | struct cgrp_cset_link *link; | |
2d8f243a | 811 | struct cgroup_subsys *ss; |
0ac801fe | 812 | unsigned long key; |
2d8f243a | 813 | int ssid; |
472b1053 | 814 | |
b326f9d0 TH |
815 | lockdep_assert_held(&cgroup_mutex); |
816 | ||
817929ec PM |
817 | /* First see if we already have a cgroup group that matches |
818 | * the desired set */ | |
96d365e0 | 819 | down_read(&css_set_rwsem); |
5abb8855 TH |
820 | cset = find_existing_css_set(old_cset, cgrp, template); |
821 | if (cset) | |
822 | get_css_set(cset); | |
96d365e0 | 823 | up_read(&css_set_rwsem); |
817929ec | 824 | |
5abb8855 TH |
825 | if (cset) |
826 | return cset; | |
817929ec | 827 | |
f4f4be2b | 828 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 829 | if (!cset) |
817929ec PM |
830 | return NULL; |
831 | ||
69d0206c | 832 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 833 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 834 | kfree(cset); |
817929ec PM |
835 | return NULL; |
836 | } | |
837 | ||
5abb8855 | 838 | atomic_set(&cset->refcount, 1); |
69d0206c | 839 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 840 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 841 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 842 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 843 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 844 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
845 | |
846 | /* Copy the set of subsystem state objects generated in | |
847 | * find_existing_css_set() */ | |
5abb8855 | 848 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 849 | |
96d365e0 | 850 | down_write(&css_set_rwsem); |
817929ec | 851 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 852 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 853 | struct cgroup *c = link->cgrp; |
69d0206c | 854 | |
7717f7ba PM |
855 | if (c->root == cgrp->root) |
856 | c = cgrp; | |
69d0206c | 857 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 858 | } |
817929ec | 859 | |
69d0206c | 860 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 861 | |
817929ec | 862 | css_set_count++; |
472b1053 | 863 | |
2d8f243a | 864 | /* Add @cset to the hash table */ |
5abb8855 TH |
865 | key = css_set_hash(cset->subsys); |
866 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 867 | |
2d8f243a TH |
868 | for_each_subsys(ss, ssid) |
869 | list_add_tail(&cset->e_cset_node[ssid], | |
870 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
871 | ||
96d365e0 | 872 | up_write(&css_set_rwsem); |
817929ec | 873 | |
5abb8855 | 874 | return cset; |
b4f48b63 PM |
875 | } |
876 | ||
3dd06ffa | 877 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 878 | { |
3dd06ffa | 879 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 880 | |
3dd06ffa | 881 | return root_cgrp->root; |
2bd59d48 TH |
882 | } |
883 | ||
3dd06ffa | 884 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
885 | { |
886 | int id; | |
887 | ||
888 | lockdep_assert_held(&cgroup_mutex); | |
889 | ||
985ed670 | 890 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
891 | if (id < 0) |
892 | return id; | |
893 | ||
894 | root->hierarchy_id = id; | |
895 | return 0; | |
896 | } | |
897 | ||
3dd06ffa | 898 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
899 | { |
900 | lockdep_assert_held(&cgroup_mutex); | |
901 | ||
902 | if (root->hierarchy_id) { | |
903 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
904 | root->hierarchy_id = 0; | |
905 | } | |
906 | } | |
907 | ||
3dd06ffa | 908 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
909 | { |
910 | if (root) { | |
d0f702e6 | 911 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
912 | WARN_ON_ONCE(root->hierarchy_id); |
913 | ||
914 | idr_destroy(&root->cgroup_idr); | |
915 | kfree(root); | |
916 | } | |
917 | } | |
918 | ||
3dd06ffa | 919 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 920 | { |
3dd06ffa | 921 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 922 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 923 | |
2bd59d48 | 924 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 925 | |
776f02fa | 926 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 927 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 928 | |
f2e85d57 | 929 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 930 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 931 | |
7717f7ba | 932 | /* |
f2e85d57 TH |
933 | * Release all the links from cset_links to this hierarchy's |
934 | * root cgroup | |
7717f7ba | 935 | */ |
96d365e0 | 936 | down_write(&css_set_rwsem); |
f2e85d57 TH |
937 | |
938 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
939 | list_del(&link->cset_link); | |
940 | list_del(&link->cgrp_link); | |
941 | kfree(link); | |
942 | } | |
96d365e0 | 943 | up_write(&css_set_rwsem); |
f2e85d57 TH |
944 | |
945 | if (!list_empty(&root->root_list)) { | |
946 | list_del(&root->root_list); | |
947 | cgroup_root_count--; | |
948 | } | |
949 | ||
950 | cgroup_exit_root_id(root); | |
951 | ||
952 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 953 | |
2bd59d48 | 954 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
955 | cgroup_free_root(root); |
956 | } | |
957 | ||
ceb6a081 TH |
958 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
959 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 960 | struct cgroup_root *root) |
7717f7ba | 961 | { |
7717f7ba PM |
962 | struct cgroup *res = NULL; |
963 | ||
96d365e0 TH |
964 | lockdep_assert_held(&cgroup_mutex); |
965 | lockdep_assert_held(&css_set_rwsem); | |
966 | ||
5abb8855 | 967 | if (cset == &init_css_set) { |
3dd06ffa | 968 | res = &root->cgrp; |
7717f7ba | 969 | } else { |
69d0206c TH |
970 | struct cgrp_cset_link *link; |
971 | ||
972 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 973 | struct cgroup *c = link->cgrp; |
69d0206c | 974 | |
7717f7ba PM |
975 | if (c->root == root) { |
976 | res = c; | |
977 | break; | |
978 | } | |
979 | } | |
980 | } | |
96d365e0 | 981 | |
7717f7ba PM |
982 | BUG_ON(!res); |
983 | return res; | |
984 | } | |
985 | ||
ddbcc7e8 | 986 | /* |
ceb6a081 TH |
987 | * Return the cgroup for "task" from the given hierarchy. Must be |
988 | * called with cgroup_mutex and css_set_rwsem held. | |
989 | */ | |
990 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 991 | struct cgroup_root *root) |
ceb6a081 TH |
992 | { |
993 | /* | |
994 | * No need to lock the task - since we hold cgroup_mutex the | |
995 | * task can't change groups, so the only thing that can happen | |
996 | * is that it exits and its css is set back to init_css_set. | |
997 | */ | |
998 | return cset_cgroup_from_root(task_css_set(task), root); | |
999 | } | |
1000 | ||
ddbcc7e8 | 1001 | /* |
ddbcc7e8 PM |
1002 | * A task must hold cgroup_mutex to modify cgroups. |
1003 | * | |
1004 | * Any task can increment and decrement the count field without lock. | |
1005 | * So in general, code holding cgroup_mutex can't rely on the count | |
1006 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1007 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1008 | * means that no tasks are currently attached, therefore there is no |
1009 | * way a task attached to that cgroup can fork (the other way to | |
1010 | * increment the count). So code holding cgroup_mutex can safely | |
1011 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1012 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1013 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1014 | * needs that mutex. | |
1015 | * | |
ddbcc7e8 PM |
1016 | * A cgroup can only be deleted if both its 'count' of using tasks |
1017 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1018 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1019 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1020 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1021 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1022 | * |
1023 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1024 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1025 | */ |
1026 | ||
8ab456ac | 1027 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 1028 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1029 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1030 | |
8d7e6fb0 TH |
1031 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1032 | char *buf) | |
ddbcc7e8 | 1033 | { |
3e1d2eed TH |
1034 | struct cgroup_subsys *ss = cft->ss; |
1035 | ||
8d7e6fb0 TH |
1036 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1037 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1038 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1039 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1040 | cft->name); | |
8d7e6fb0 TH |
1041 | else |
1042 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1043 | return buf; | |
ddbcc7e8 PM |
1044 | } |
1045 | ||
f2e85d57 TH |
1046 | /** |
1047 | * cgroup_file_mode - deduce file mode of a control file | |
1048 | * @cft: the control file in question | |
1049 | * | |
1050 | * returns cft->mode if ->mode is not 0 | |
1051 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
1052 | * returns S_IRUGO if it has only a read handler | |
1053 | * returns S_IWUSR if it has only a write hander | |
1054 | */ | |
1055 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1056 | { |
f2e85d57 | 1057 | umode_t mode = 0; |
65dff759 | 1058 | |
f2e85d57 TH |
1059 | if (cft->mode) |
1060 | return cft->mode; | |
1061 | ||
1062 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
1063 | mode |= S_IRUGO; | |
1064 | ||
6770c64e | 1065 | if (cft->write_u64 || cft->write_s64 || cft->write) |
f2e85d57 TH |
1066 | mode |= S_IWUSR; |
1067 | ||
1068 | return mode; | |
65dff759 LZ |
1069 | } |
1070 | ||
59f5296b | 1071 | static void cgroup_get(struct cgroup *cgrp) |
be445626 | 1072 | { |
2bd59d48 | 1073 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
9d755d33 | 1074 | css_get(&cgrp->self); |
be445626 LZ |
1075 | } |
1076 | ||
aa32362f LZ |
1077 | static bool cgroup_tryget(struct cgroup *cgrp) |
1078 | { | |
1079 | return css_tryget(&cgrp->self); | |
1080 | } | |
1081 | ||
59f5296b | 1082 | static void cgroup_put(struct cgroup *cgrp) |
be445626 | 1083 | { |
9d755d33 | 1084 | css_put(&cgrp->self); |
be445626 LZ |
1085 | } |
1086 | ||
af0ba678 | 1087 | /** |
0f060deb | 1088 | * cgroup_calc_child_subsys_mask - calculate child_subsys_mask |
af0ba678 | 1089 | * @cgrp: the target cgroup |
0f060deb | 1090 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1091 | * |
1092 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1093 | * enabled together through its ->depends_on mask. In such cases, more | |
1094 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1095 | * | |
0f060deb TH |
1096 | * This function calculates which subsystems need to be enabled if |
1097 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1098 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1099 | */ |
8ab456ac AS |
1100 | static unsigned long cgroup_calc_child_subsys_mask(struct cgroup *cgrp, |
1101 | unsigned long subtree_control) | |
667c2491 | 1102 | { |
af0ba678 | 1103 | struct cgroup *parent = cgroup_parent(cgrp); |
8ab456ac | 1104 | unsigned long cur_ss_mask = subtree_control; |
af0ba678 TH |
1105 | struct cgroup_subsys *ss; |
1106 | int ssid; | |
1107 | ||
1108 | lockdep_assert_held(&cgroup_mutex); | |
1109 | ||
0f060deb TH |
1110 | if (!cgroup_on_dfl(cgrp)) |
1111 | return cur_ss_mask; | |
af0ba678 TH |
1112 | |
1113 | while (true) { | |
8ab456ac | 1114 | unsigned long new_ss_mask = cur_ss_mask; |
af0ba678 | 1115 | |
a966a4ed AS |
1116 | for_each_subsys_which(ss, ssid, &cur_ss_mask) |
1117 | new_ss_mask |= ss->depends_on; | |
af0ba678 TH |
1118 | |
1119 | /* | |
1120 | * Mask out subsystems which aren't available. This can | |
1121 | * happen only if some depended-upon subsystems were bound | |
1122 | * to non-default hierarchies. | |
1123 | */ | |
1124 | if (parent) | |
1125 | new_ss_mask &= parent->child_subsys_mask; | |
1126 | else | |
1127 | new_ss_mask &= cgrp->root->subsys_mask; | |
1128 | ||
1129 | if (new_ss_mask == cur_ss_mask) | |
1130 | break; | |
1131 | cur_ss_mask = new_ss_mask; | |
1132 | } | |
1133 | ||
0f060deb TH |
1134 | return cur_ss_mask; |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1139 | * @cgrp: the target cgroup | |
1140 | * | |
1141 | * Update @cgrp->child_subsys_mask according to the current | |
1142 | * @cgrp->subtree_control using cgroup_calc_child_subsys_mask(). | |
1143 | */ | |
1144 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) | |
1145 | { | |
1146 | cgrp->child_subsys_mask = | |
1147 | cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1148 | } |
1149 | ||
a9746d8d TH |
1150 | /** |
1151 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1152 | * @kn: the kernfs_node being serviced | |
1153 | * | |
1154 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1155 | * the method finishes if locking succeeded. Note that once this function | |
1156 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1157 | * inaccessible any time. If the caller intends to continue to access the | |
1158 | * cgroup, it should pin it before invoking this function. | |
1159 | */ | |
1160 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1161 | { |
a9746d8d TH |
1162 | struct cgroup *cgrp; |
1163 | ||
1164 | if (kernfs_type(kn) == KERNFS_DIR) | |
1165 | cgrp = kn->priv; | |
1166 | else | |
1167 | cgrp = kn->parent->priv; | |
1168 | ||
1169 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1170 | |
1171 | kernfs_unbreak_active_protection(kn); | |
1172 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1173 | } |
1174 | ||
a9746d8d TH |
1175 | /** |
1176 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1177 | * @kn: the kernfs_node being serviced | |
1178 | * | |
1179 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1180 | * @kn. It breaks the active protection, performs cgroup locking and | |
1181 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1182 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1183 | * matching cgroup_kn_unlock() invocation. | |
1184 | * | |
1185 | * Any cgroup kernfs method implementation which requires locking the | |
1186 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1187 | * locking under kernfs active protection and allows all kernfs operations | |
1188 | * including self-removal. | |
1189 | */ | |
1190 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1191 | { |
a9746d8d TH |
1192 | struct cgroup *cgrp; |
1193 | ||
1194 | if (kernfs_type(kn) == KERNFS_DIR) | |
1195 | cgrp = kn->priv; | |
1196 | else | |
1197 | cgrp = kn->parent->priv; | |
05ef1d7c | 1198 | |
2739d3cc | 1199 | /* |
01f6474c | 1200 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1201 | * active_ref. cgroup liveliness check alone provides enough |
1202 | * protection against removal. Ensure @cgrp stays accessible and | |
1203 | * break the active_ref protection. | |
2739d3cc | 1204 | */ |
aa32362f LZ |
1205 | if (!cgroup_tryget(cgrp)) |
1206 | return NULL; | |
a9746d8d TH |
1207 | kernfs_break_active_protection(kn); |
1208 | ||
2bd59d48 | 1209 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1210 | |
a9746d8d TH |
1211 | if (!cgroup_is_dead(cgrp)) |
1212 | return cgrp; | |
1213 | ||
1214 | cgroup_kn_unlock(kn); | |
1215 | return NULL; | |
ddbcc7e8 | 1216 | } |
05ef1d7c | 1217 | |
2739d3cc | 1218 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1219 | { |
2bd59d48 | 1220 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1221 | |
01f6474c | 1222 | lockdep_assert_held(&cgroup_mutex); |
2bd59d48 | 1223 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1224 | } |
1225 | ||
13af07df | 1226 | /** |
628f7cd4 | 1227 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1228 | * @cgrp: target cgroup |
13af07df AR |
1229 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1230 | */ | |
8ab456ac | 1231 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1232 | { |
13af07df | 1233 | struct cgroup_subsys *ss; |
b420ba7d | 1234 | int i; |
05ef1d7c | 1235 | |
b420ba7d | 1236 | for_each_subsys(ss, i) { |
0adb0704 | 1237 | struct cftype *cfts; |
b420ba7d | 1238 | |
69dfa00c | 1239 | if (!(subsys_mask & (1 << i))) |
13af07df | 1240 | continue; |
0adb0704 TH |
1241 | list_for_each_entry(cfts, &ss->cfts, node) |
1242 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1243 | } |
ddbcc7e8 PM |
1244 | } |
1245 | ||
8ab456ac AS |
1246 | static int rebind_subsystems(struct cgroup_root *dst_root, |
1247 | unsigned long ss_mask) | |
ddbcc7e8 | 1248 | { |
30159ec7 | 1249 | struct cgroup_subsys *ss; |
8ab456ac | 1250 | unsigned long tmp_ss_mask; |
2d8f243a | 1251 | int ssid, i, ret; |
ddbcc7e8 | 1252 | |
ace2bee8 | 1253 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1254 | |
a966a4ed | 1255 | for_each_subsys_which(ss, ssid, &ss_mask) { |
7fd8c565 TH |
1256 | /* if @ss has non-root csses attached to it, can't move */ |
1257 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1258 | return -EBUSY; |
1d5be6b2 | 1259 | |
5df36032 | 1260 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1261 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1262 | return -EBUSY; |
ddbcc7e8 PM |
1263 | } |
1264 | ||
5533e011 TH |
1265 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1266 | tmp_ss_mask = ss_mask; | |
1267 | if (dst_root == &cgrp_dfl_root) | |
1268 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1269 | ||
1270 | ret = cgroup_populate_dir(&dst_root->cgrp, tmp_ss_mask); | |
a2dd4247 TH |
1271 | if (ret) { |
1272 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1273 | return ret; |
ddbcc7e8 | 1274 | |
a2dd4247 TH |
1275 | /* |
1276 | * Rebinding back to the default root is not allowed to | |
1277 | * fail. Using both default and non-default roots should | |
1278 | * be rare. Moving subsystems back and forth even more so. | |
1279 | * Just warn about it and continue. | |
1280 | */ | |
1281 | if (cgrp_dfl_root_visible) { | |
8ab456ac | 1282 | pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n", |
a2a1f9ea | 1283 | ret, ss_mask); |
ed3d261b | 1284 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); |
a2dd4247 | 1285 | } |
5df36032 | 1286 | } |
3126121f TH |
1287 | |
1288 | /* | |
1289 | * Nothing can fail from this point on. Remove files for the | |
1290 | * removed subsystems and rebind each subsystem. | |
1291 | */ | |
a966a4ed AS |
1292 | for_each_subsys_which(ss, ssid, &ss_mask) |
1293 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); | |
a8a648c4 | 1294 | |
a966a4ed | 1295 | for_each_subsys_which(ss, ssid, &ss_mask) { |
3dd06ffa | 1296 | struct cgroup_root *src_root; |
5df36032 | 1297 | struct cgroup_subsys_state *css; |
2d8f243a | 1298 | struct css_set *cset; |
a8a648c4 | 1299 | |
5df36032 | 1300 | src_root = ss->root; |
3dd06ffa | 1301 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1302 | |
3dd06ffa | 1303 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1304 | |
3dd06ffa TH |
1305 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1306 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1307 | ss->root = dst_root; |
3dd06ffa | 1308 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1309 | |
2d8f243a TH |
1310 | down_write(&css_set_rwsem); |
1311 | hash_for_each(css_set_table, i, cset, hlist) | |
1312 | list_move_tail(&cset->e_cset_node[ss->id], | |
1313 | &dst_root->cgrp.e_csets[ss->id]); | |
1314 | up_write(&css_set_rwsem); | |
1315 | ||
f392e51c | 1316 | src_root->subsys_mask &= ~(1 << ssid); |
667c2491 TH |
1317 | src_root->cgrp.subtree_control &= ~(1 << ssid); |
1318 | cgroup_refresh_child_subsys_mask(&src_root->cgrp); | |
f392e51c | 1319 | |
bd53d617 | 1320 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1321 | dst_root->subsys_mask |= 1 << ssid; |
667c2491 TH |
1322 | if (dst_root != &cgrp_dfl_root) { |
1323 | dst_root->cgrp.subtree_control |= 1 << ssid; | |
1324 | cgroup_refresh_child_subsys_mask(&dst_root->cgrp); | |
1325 | } | |
a8a648c4 | 1326 | |
5df36032 TH |
1327 | if (ss->bind) |
1328 | ss->bind(css); | |
ddbcc7e8 | 1329 | } |
ddbcc7e8 | 1330 | |
a2dd4247 | 1331 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1332 | return 0; |
1333 | } | |
1334 | ||
2bd59d48 TH |
1335 | static int cgroup_show_options(struct seq_file *seq, |
1336 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1337 | { |
3dd06ffa | 1338 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1339 | struct cgroup_subsys *ss; |
b85d2040 | 1340 | int ssid; |
ddbcc7e8 | 1341 | |
d98817d4 TH |
1342 | if (root != &cgrp_dfl_root) |
1343 | for_each_subsys(ss, ssid) | |
1344 | if (root->subsys_mask & (1 << ssid)) | |
3e1d2eed | 1345 | seq_printf(seq, ",%s", ss->legacy_name); |
93438629 | 1346 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1347 | seq_puts(seq, ",noprefix"); |
93438629 | 1348 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1349 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1350 | |
1351 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1352 | if (strlen(root->release_agent_path)) |
1353 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1354 | spin_unlock(&release_agent_path_lock); |
1355 | ||
3dd06ffa | 1356 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1357 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1358 | if (strlen(root->name)) |
1359 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1360 | return 0; |
1361 | } | |
1362 | ||
1363 | struct cgroup_sb_opts { | |
8ab456ac | 1364 | unsigned long subsys_mask; |
69dfa00c | 1365 | unsigned int flags; |
81a6a5cd | 1366 | char *release_agent; |
2260e7fc | 1367 | bool cpuset_clone_children; |
c6d57f33 | 1368 | char *name; |
2c6ab6d2 PM |
1369 | /* User explicitly requested empty subsystem */ |
1370 | bool none; | |
ddbcc7e8 PM |
1371 | }; |
1372 | ||
cf5d5941 | 1373 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1374 | { |
32a8cf23 DL |
1375 | char *token, *o = data; |
1376 | bool all_ss = false, one_ss = false; | |
8ab456ac | 1377 | unsigned long mask = -1UL; |
30159ec7 | 1378 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1379 | int nr_opts = 0; |
30159ec7 | 1380 | int i; |
f9ab5b5b LZ |
1381 | |
1382 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1383 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1384 | #endif |
ddbcc7e8 | 1385 | |
c6d57f33 | 1386 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1387 | |
1388 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1389 | nr_opts++; |
1390 | ||
ddbcc7e8 PM |
1391 | if (!*token) |
1392 | return -EINVAL; | |
32a8cf23 | 1393 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1394 | /* Explicitly have no subsystems */ |
1395 | opts->none = true; | |
32a8cf23 DL |
1396 | continue; |
1397 | } | |
1398 | if (!strcmp(token, "all")) { | |
1399 | /* Mutually exclusive option 'all' + subsystem name */ | |
1400 | if (one_ss) | |
1401 | return -EINVAL; | |
1402 | all_ss = true; | |
1403 | continue; | |
1404 | } | |
873fe09e TH |
1405 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1406 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1407 | continue; | |
1408 | } | |
32a8cf23 | 1409 | if (!strcmp(token, "noprefix")) { |
93438629 | 1410 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1411 | continue; |
1412 | } | |
1413 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1414 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1415 | continue; |
1416 | } | |
03b1cde6 | 1417 | if (!strcmp(token, "xattr")) { |
93438629 | 1418 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1419 | continue; |
1420 | } | |
32a8cf23 | 1421 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1422 | /* Specifying two release agents is forbidden */ |
1423 | if (opts->release_agent) | |
1424 | return -EINVAL; | |
c6d57f33 | 1425 | opts->release_agent = |
e400c285 | 1426 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1427 | if (!opts->release_agent) |
1428 | return -ENOMEM; | |
32a8cf23 DL |
1429 | continue; |
1430 | } | |
1431 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1432 | const char *name = token + 5; |
1433 | /* Can't specify an empty name */ | |
1434 | if (!strlen(name)) | |
1435 | return -EINVAL; | |
1436 | /* Must match [\w.-]+ */ | |
1437 | for (i = 0; i < strlen(name); i++) { | |
1438 | char c = name[i]; | |
1439 | if (isalnum(c)) | |
1440 | continue; | |
1441 | if ((c == '.') || (c == '-') || (c == '_')) | |
1442 | continue; | |
1443 | return -EINVAL; | |
1444 | } | |
1445 | /* Specifying two names is forbidden */ | |
1446 | if (opts->name) | |
1447 | return -EINVAL; | |
1448 | opts->name = kstrndup(name, | |
e400c285 | 1449 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1450 | GFP_KERNEL); |
1451 | if (!opts->name) | |
1452 | return -ENOMEM; | |
32a8cf23 DL |
1453 | |
1454 | continue; | |
1455 | } | |
1456 | ||
30159ec7 | 1457 | for_each_subsys(ss, i) { |
3e1d2eed | 1458 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 DL |
1459 | continue; |
1460 | if (ss->disabled) | |
1461 | continue; | |
1462 | ||
1463 | /* Mutually exclusive option 'all' + subsystem name */ | |
1464 | if (all_ss) | |
1465 | return -EINVAL; | |
69dfa00c | 1466 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1467 | one_ss = true; |
1468 | ||
1469 | break; | |
1470 | } | |
1471 | if (i == CGROUP_SUBSYS_COUNT) | |
1472 | return -ENOENT; | |
1473 | } | |
1474 | ||
873fe09e | 1475 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1476 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
7b9a6ba5 TH |
1477 | if (nr_opts != 1) { |
1478 | pr_err("sane_behavior: no other mount options allowed\n"); | |
873fe09e TH |
1479 | return -EINVAL; |
1480 | } | |
7b9a6ba5 | 1481 | return 0; |
873fe09e TH |
1482 | } |
1483 | ||
7b9a6ba5 TH |
1484 | /* |
1485 | * If the 'all' option was specified select all the subsystems, | |
1486 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1487 | * not specified, let's default to 'all' | |
1488 | */ | |
1489 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1490 | for_each_subsys(ss, i) | |
1491 | if (!ss->disabled) | |
1492 | opts->subsys_mask |= (1 << i); | |
1493 | ||
1494 | /* | |
1495 | * We either have to specify by name or by subsystems. (So all | |
1496 | * empty hierarchies must have a name). | |
1497 | */ | |
1498 | if (!opts->subsys_mask && !opts->name) | |
1499 | return -EINVAL; | |
1500 | ||
f9ab5b5b LZ |
1501 | /* |
1502 | * Option noprefix was introduced just for backward compatibility | |
1503 | * with the old cpuset, so we allow noprefix only if mounting just | |
1504 | * the cpuset subsystem. | |
1505 | */ | |
93438629 | 1506 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1507 | return -EINVAL; |
1508 | ||
2c6ab6d2 | 1509 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1510 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1511 | return -EINVAL; |
1512 | ||
ddbcc7e8 PM |
1513 | return 0; |
1514 | } | |
1515 | ||
2bd59d48 | 1516 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1517 | { |
1518 | int ret = 0; | |
3dd06ffa | 1519 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1520 | struct cgroup_sb_opts opts; |
8ab456ac | 1521 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1522 | |
aa6ec29b TH |
1523 | if (root == &cgrp_dfl_root) { |
1524 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1525 | return -EINVAL; |
1526 | } | |
1527 | ||
ddbcc7e8 PM |
1528 | mutex_lock(&cgroup_mutex); |
1529 | ||
1530 | /* See what subsystems are wanted */ | |
1531 | ret = parse_cgroupfs_options(data, &opts); | |
1532 | if (ret) | |
1533 | goto out_unlock; | |
1534 | ||
f392e51c | 1535 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1536 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1537 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1538 | |
f392e51c TH |
1539 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1540 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1541 | |
cf5d5941 | 1542 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1543 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1544 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1545 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1546 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1547 | ret = -EINVAL; |
1548 | goto out_unlock; | |
1549 | } | |
1550 | ||
f172e67c | 1551 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1552 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1553 | ret = -EBUSY; |
0670e08b | 1554 | goto out_unlock; |
cf5d5941 | 1555 | } |
ddbcc7e8 | 1556 | |
5df36032 | 1557 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1558 | if (ret) |
0670e08b | 1559 | goto out_unlock; |
ddbcc7e8 | 1560 | |
3dd06ffa | 1561 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1562 | |
69e943b7 TH |
1563 | if (opts.release_agent) { |
1564 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1565 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1566 | spin_unlock(&release_agent_path_lock); |
1567 | } | |
ddbcc7e8 | 1568 | out_unlock: |
66bdc9cf | 1569 | kfree(opts.release_agent); |
c6d57f33 | 1570 | kfree(opts.name); |
ddbcc7e8 | 1571 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1572 | return ret; |
1573 | } | |
1574 | ||
afeb0f9f TH |
1575 | /* |
1576 | * To reduce the fork() overhead for systems that are not actually using | |
1577 | * their cgroups capability, we don't maintain the lists running through | |
1578 | * each css_set to its tasks until we see the list actually used - in other | |
1579 | * words after the first mount. | |
1580 | */ | |
1581 | static bool use_task_css_set_links __read_mostly; | |
1582 | ||
1583 | static void cgroup_enable_task_cg_lists(void) | |
1584 | { | |
1585 | struct task_struct *p, *g; | |
1586 | ||
96d365e0 | 1587 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1588 | |
1589 | if (use_task_css_set_links) | |
1590 | goto out_unlock; | |
1591 | ||
1592 | use_task_css_set_links = true; | |
1593 | ||
1594 | /* | |
1595 | * We need tasklist_lock because RCU is not safe against | |
1596 | * while_each_thread(). Besides, a forking task that has passed | |
1597 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1598 | * is not guaranteed to have its child immediately visible in the | |
1599 | * tasklist if we walk through it with RCU. | |
1600 | */ | |
1601 | read_lock(&tasklist_lock); | |
1602 | do_each_thread(g, p) { | |
afeb0f9f TH |
1603 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1604 | task_css_set(p) != &init_css_set); | |
1605 | ||
1606 | /* | |
1607 | * We should check if the process is exiting, otherwise | |
1608 | * it will race with cgroup_exit() in that the list | |
1609 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1610 | * Do it while holding siglock so that we don't end up |
1611 | * racing against cgroup_exit(). | |
afeb0f9f | 1612 | */ |
f153ad11 | 1613 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1614 | if (!(p->flags & PF_EXITING)) { |
1615 | struct css_set *cset = task_css_set(p); | |
1616 | ||
1617 | list_add(&p->cg_list, &cset->tasks); | |
1618 | get_css_set(cset); | |
1619 | } | |
f153ad11 | 1620 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1621 | } while_each_thread(g, p); |
1622 | read_unlock(&tasklist_lock); | |
1623 | out_unlock: | |
96d365e0 | 1624 | up_write(&css_set_rwsem); |
afeb0f9f | 1625 | } |
ddbcc7e8 | 1626 | |
cc31edce PM |
1627 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1628 | { | |
2d8f243a TH |
1629 | struct cgroup_subsys *ss; |
1630 | int ssid; | |
1631 | ||
d5c419b6 TH |
1632 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1633 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1634 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1635 | INIT_LIST_HEAD(&cgrp->pidlists); |
1636 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1637 | cgrp->self.cgroup = cgrp; |
184faf32 | 1638 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1639 | |
1640 | for_each_subsys(ss, ssid) | |
1641 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1642 | |
1643 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1644 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1645 | } |
c6d57f33 | 1646 | |
3dd06ffa | 1647 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1648 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1649 | { |
3dd06ffa | 1650 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1651 | |
ddbcc7e8 | 1652 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1653 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1654 | cgrp->root = root; |
cc31edce | 1655 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1656 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1657 | |
c6d57f33 PM |
1658 | root->flags = opts->flags; |
1659 | if (opts->release_agent) | |
1660 | strcpy(root->release_agent_path, opts->release_agent); | |
1661 | if (opts->name) | |
1662 | strcpy(root->name, opts->name); | |
2260e7fc | 1663 | if (opts->cpuset_clone_children) |
3dd06ffa | 1664 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1665 | } |
1666 | ||
8ab456ac | 1667 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1668 | { |
d427dfeb | 1669 | LIST_HEAD(tmp_links); |
3dd06ffa | 1670 | struct cgroup *root_cgrp = &root->cgrp; |
a14c6874 | 1671 | struct cftype *base_files; |
d427dfeb | 1672 | struct css_set *cset; |
d427dfeb | 1673 | int i, ret; |
2c6ab6d2 | 1674 | |
d427dfeb | 1675 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1676 | |
cf780b7d | 1677 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1678 | if (ret < 0) |
2bd59d48 | 1679 | goto out; |
d427dfeb | 1680 | root_cgrp->id = ret; |
c6d57f33 | 1681 | |
2aad2a86 TH |
1682 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1683 | GFP_KERNEL); | |
9d755d33 TH |
1684 | if (ret) |
1685 | goto out; | |
1686 | ||
d427dfeb | 1687 | /* |
96d365e0 | 1688 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1689 | * but that's OK - it can only be increased by someone holding |
1690 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1691 | * have some link structures left over | |
1692 | */ | |
1693 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1694 | if (ret) | |
9d755d33 | 1695 | goto cancel_ref; |
ddbcc7e8 | 1696 | |
985ed670 | 1697 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1698 | if (ret) |
9d755d33 | 1699 | goto cancel_ref; |
ddbcc7e8 | 1700 | |
2bd59d48 TH |
1701 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1702 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1703 | root_cgrp); | |
1704 | if (IS_ERR(root->kf_root)) { | |
1705 | ret = PTR_ERR(root->kf_root); | |
1706 | goto exit_root_id; | |
1707 | } | |
1708 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1709 | |
a14c6874 TH |
1710 | if (root == &cgrp_dfl_root) |
1711 | base_files = cgroup_dfl_base_files; | |
1712 | else | |
1713 | base_files = cgroup_legacy_base_files; | |
1714 | ||
1715 | ret = cgroup_addrm_files(root_cgrp, base_files, true); | |
d427dfeb | 1716 | if (ret) |
2bd59d48 | 1717 | goto destroy_root; |
ddbcc7e8 | 1718 | |
5df36032 | 1719 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1720 | if (ret) |
2bd59d48 | 1721 | goto destroy_root; |
ddbcc7e8 | 1722 | |
d427dfeb TH |
1723 | /* |
1724 | * There must be no failure case after here, since rebinding takes | |
1725 | * care of subsystems' refcounts, which are explicitly dropped in | |
1726 | * the failure exit path. | |
1727 | */ | |
1728 | list_add(&root->root_list, &cgroup_roots); | |
1729 | cgroup_root_count++; | |
0df6a63f | 1730 | |
d427dfeb | 1731 | /* |
3dd06ffa | 1732 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1733 | * objects. |
1734 | */ | |
96d365e0 | 1735 | down_write(&css_set_rwsem); |
d427dfeb TH |
1736 | hash_for_each(css_set_table, i, cset, hlist) |
1737 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1738 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1739 | |
d5c419b6 | 1740 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1741 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1742 | |
2bd59d48 | 1743 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1744 | ret = 0; |
2bd59d48 | 1745 | goto out; |
d427dfeb | 1746 | |
2bd59d48 TH |
1747 | destroy_root: |
1748 | kernfs_destroy_root(root->kf_root); | |
1749 | root->kf_root = NULL; | |
1750 | exit_root_id: | |
d427dfeb | 1751 | cgroup_exit_root_id(root); |
9d755d33 | 1752 | cancel_ref: |
9a1049da | 1753 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1754 | out: |
d427dfeb TH |
1755 | free_cgrp_cset_links(&tmp_links); |
1756 | return ret; | |
ddbcc7e8 PM |
1757 | } |
1758 | ||
f7e83571 | 1759 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1760 | int flags, const char *unused_dev_name, |
f7e83571 | 1761 | void *data) |
ddbcc7e8 | 1762 | { |
3a32bd72 | 1763 | struct super_block *pinned_sb = NULL; |
970317aa | 1764 | struct cgroup_subsys *ss; |
3dd06ffa | 1765 | struct cgroup_root *root; |
ddbcc7e8 | 1766 | struct cgroup_sb_opts opts; |
2bd59d48 | 1767 | struct dentry *dentry; |
8e30e2b8 | 1768 | int ret; |
970317aa | 1769 | int i; |
c6b3d5bc | 1770 | bool new_sb; |
ddbcc7e8 | 1771 | |
56fde9e0 TH |
1772 | /* |
1773 | * The first time anyone tries to mount a cgroup, enable the list | |
1774 | * linking each css_set to its tasks and fix up all existing tasks. | |
1775 | */ | |
1776 | if (!use_task_css_set_links) | |
1777 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1778 | |
aae8aab4 | 1779 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1780 | |
1781 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1782 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1783 | if (ret) |
8e30e2b8 | 1784 | goto out_unlock; |
a015edd2 | 1785 | |
2bd59d48 | 1786 | /* look for a matching existing root */ |
7b9a6ba5 | 1787 | if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2dd4247 TH |
1788 | cgrp_dfl_root_visible = true; |
1789 | root = &cgrp_dfl_root; | |
1790 | cgroup_get(&root->cgrp); | |
1791 | ret = 0; | |
1792 | goto out_unlock; | |
ddbcc7e8 PM |
1793 | } |
1794 | ||
970317aa LZ |
1795 | /* |
1796 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1797 | * still be dying after the previous unmount. Let's drain the | |
1798 | * dying subsystems. We just need to ensure that the ones | |
1799 | * unmounted previously finish dying and don't care about new ones | |
1800 | * starting. Testing ref liveliness is good enough. | |
1801 | */ | |
1802 | for_each_subsys(ss, i) { | |
1803 | if (!(opts.subsys_mask & (1 << i)) || | |
1804 | ss->root == &cgrp_dfl_root) | |
1805 | continue; | |
1806 | ||
1807 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1808 | mutex_unlock(&cgroup_mutex); | |
1809 | msleep(10); | |
1810 | ret = restart_syscall(); | |
1811 | goto out_free; | |
1812 | } | |
1813 | cgroup_put(&ss->root->cgrp); | |
1814 | } | |
1815 | ||
985ed670 | 1816 | for_each_root(root) { |
2bd59d48 | 1817 | bool name_match = false; |
3126121f | 1818 | |
3dd06ffa | 1819 | if (root == &cgrp_dfl_root) |
985ed670 | 1820 | continue; |
3126121f | 1821 | |
cf5d5941 | 1822 | /* |
2bd59d48 TH |
1823 | * If we asked for a name then it must match. Also, if |
1824 | * name matches but sybsys_mask doesn't, we should fail. | |
1825 | * Remember whether name matched. | |
cf5d5941 | 1826 | */ |
2bd59d48 TH |
1827 | if (opts.name) { |
1828 | if (strcmp(opts.name, root->name)) | |
1829 | continue; | |
1830 | name_match = true; | |
1831 | } | |
ddbcc7e8 | 1832 | |
c6d57f33 | 1833 | /* |
2bd59d48 TH |
1834 | * If we asked for subsystems (or explicitly for no |
1835 | * subsystems) then they must match. | |
c6d57f33 | 1836 | */ |
2bd59d48 | 1837 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1838 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1839 | if (!name_match) |
1840 | continue; | |
1841 | ret = -EBUSY; | |
1842 | goto out_unlock; | |
1843 | } | |
873fe09e | 1844 | |
7b9a6ba5 TH |
1845 | if (root->flags ^ opts.flags) |
1846 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 1847 | |
776f02fa | 1848 | /* |
3a32bd72 LZ |
1849 | * We want to reuse @root whose lifetime is governed by its |
1850 | * ->cgrp. Let's check whether @root is alive and keep it | |
1851 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1852 | * want to block it by pinning the sb so that @root doesn't | |
1853 | * get killed before mount is complete. | |
1854 | * | |
1855 | * With the sb pinned, tryget_live can reliably indicate | |
1856 | * whether @root can be reused. If it's being killed, | |
1857 | * drain it. We can use wait_queue for the wait but this | |
1858 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 1859 | */ |
3a32bd72 LZ |
1860 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
1861 | if (IS_ERR(pinned_sb) || | |
1862 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 1863 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
1864 | if (!IS_ERR_OR_NULL(pinned_sb)) |
1865 | deactivate_super(pinned_sb); | |
776f02fa | 1866 | msleep(10); |
a015edd2 TH |
1867 | ret = restart_syscall(); |
1868 | goto out_free; | |
776f02fa | 1869 | } |
ddbcc7e8 | 1870 | |
776f02fa | 1871 | ret = 0; |
2bd59d48 | 1872 | goto out_unlock; |
ddbcc7e8 | 1873 | } |
ddbcc7e8 | 1874 | |
817929ec | 1875 | /* |
172a2c06 TH |
1876 | * No such thing, create a new one. name= matching without subsys |
1877 | * specification is allowed for already existing hierarchies but we | |
1878 | * can't create new one without subsys specification. | |
817929ec | 1879 | */ |
172a2c06 TH |
1880 | if (!opts.subsys_mask && !opts.none) { |
1881 | ret = -EINVAL; | |
1882 | goto out_unlock; | |
817929ec | 1883 | } |
817929ec | 1884 | |
172a2c06 TH |
1885 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1886 | if (!root) { | |
1887 | ret = -ENOMEM; | |
2bd59d48 | 1888 | goto out_unlock; |
839ec545 | 1889 | } |
e5f6a860 | 1890 | |
172a2c06 TH |
1891 | init_cgroup_root(root, &opts); |
1892 | ||
35585573 | 1893 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1894 | if (ret) |
1895 | cgroup_free_root(root); | |
fa3ca07e | 1896 | |
8e30e2b8 | 1897 | out_unlock: |
ddbcc7e8 | 1898 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 1899 | out_free: |
c6d57f33 PM |
1900 | kfree(opts.release_agent); |
1901 | kfree(opts.name); | |
03b1cde6 | 1902 | |
2bd59d48 | 1903 | if (ret) |
8e30e2b8 | 1904 | return ERR_PTR(ret); |
2bd59d48 | 1905 | |
c9482a5b JZ |
1906 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
1907 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 1908 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 1909 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
1910 | |
1911 | /* | |
1912 | * If @pinned_sb, we're reusing an existing root and holding an | |
1913 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1914 | */ | |
1915 | if (pinned_sb) { | |
1916 | WARN_ON(new_sb); | |
1917 | deactivate_super(pinned_sb); | |
1918 | } | |
1919 | ||
2bd59d48 TH |
1920 | return dentry; |
1921 | } | |
1922 | ||
1923 | static void cgroup_kill_sb(struct super_block *sb) | |
1924 | { | |
1925 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1926 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1927 | |
9d755d33 TH |
1928 | /* |
1929 | * If @root doesn't have any mounts or children, start killing it. | |
1930 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
1931 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
1932 | * |
1933 | * And don't kill the default root. | |
9d755d33 | 1934 | */ |
3c606d35 | 1935 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 1936 | root == &cgrp_dfl_root) |
9d755d33 TH |
1937 | cgroup_put(&root->cgrp); |
1938 | else | |
1939 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
1940 | ||
2bd59d48 | 1941 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1942 | } |
1943 | ||
1944 | static struct file_system_type cgroup_fs_type = { | |
1945 | .name = "cgroup", | |
f7e83571 | 1946 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1947 | .kill_sb = cgroup_kill_sb, |
1948 | }; | |
1949 | ||
857a2beb | 1950 | /** |
913ffdb5 | 1951 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1952 | * @task: target task |
857a2beb TH |
1953 | * @buf: the buffer to write the path into |
1954 | * @buflen: the length of the buffer | |
1955 | * | |
913ffdb5 TH |
1956 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1957 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1958 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1959 | * cgroup controller callbacks. | |
1960 | * | |
e61734c5 | 1961 | * Return value is the same as kernfs_path(). |
857a2beb | 1962 | */ |
e61734c5 | 1963 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1964 | { |
3dd06ffa | 1965 | struct cgroup_root *root; |
913ffdb5 | 1966 | struct cgroup *cgrp; |
e61734c5 TH |
1967 | int hierarchy_id = 1; |
1968 | char *path = NULL; | |
857a2beb TH |
1969 | |
1970 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1971 | down_read(&css_set_rwsem); |
857a2beb | 1972 | |
913ffdb5 TH |
1973 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1974 | ||
857a2beb TH |
1975 | if (root) { |
1976 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1977 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1978 | } else { |
1979 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1980 | if (strlcpy(buf, "/", buflen) < buflen) |
1981 | path = buf; | |
857a2beb TH |
1982 | } |
1983 | ||
96d365e0 | 1984 | up_read(&css_set_rwsem); |
857a2beb | 1985 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1986 | return path; |
857a2beb | 1987 | } |
913ffdb5 | 1988 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1989 | |
b3dc094e | 1990 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1991 | struct cgroup_taskset { |
b3dc094e TH |
1992 | /* the src and dst cset list running through cset->mg_node */ |
1993 | struct list_head src_csets; | |
1994 | struct list_head dst_csets; | |
1995 | ||
1996 | /* | |
1997 | * Fields for cgroup_taskset_*() iteration. | |
1998 | * | |
1999 | * Before migration is committed, the target migration tasks are on | |
2000 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2001 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2002 | * or ->dst_csets depending on whether migration is committed. | |
2003 | * | |
2004 | * ->cur_csets and ->cur_task point to the current task position | |
2005 | * during iteration. | |
2006 | */ | |
2007 | struct list_head *csets; | |
2008 | struct css_set *cur_cset; | |
2009 | struct task_struct *cur_task; | |
2f7ee569 TH |
2010 | }; |
2011 | ||
2012 | /** | |
2013 | * cgroup_taskset_first - reset taskset and return the first task | |
2014 | * @tset: taskset of interest | |
2015 | * | |
2016 | * @tset iteration is initialized and the first task is returned. | |
2017 | */ | |
2018 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
2019 | { | |
b3dc094e TH |
2020 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2021 | tset->cur_task = NULL; | |
2022 | ||
2023 | return cgroup_taskset_next(tset); | |
2f7ee569 | 2024 | } |
2f7ee569 TH |
2025 | |
2026 | /** | |
2027 | * cgroup_taskset_next - iterate to the next task in taskset | |
2028 | * @tset: taskset of interest | |
2029 | * | |
2030 | * Return the next task in @tset. Iteration must have been initialized | |
2031 | * with cgroup_taskset_first(). | |
2032 | */ | |
2033 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
2034 | { | |
b3dc094e TH |
2035 | struct css_set *cset = tset->cur_cset; |
2036 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2037 | |
b3dc094e TH |
2038 | while (&cset->mg_node != tset->csets) { |
2039 | if (!task) | |
2040 | task = list_first_entry(&cset->mg_tasks, | |
2041 | struct task_struct, cg_list); | |
2042 | else | |
2043 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2044 | |
b3dc094e TH |
2045 | if (&task->cg_list != &cset->mg_tasks) { |
2046 | tset->cur_cset = cset; | |
2047 | tset->cur_task = task; | |
2048 | return task; | |
2049 | } | |
2f7ee569 | 2050 | |
b3dc094e TH |
2051 | cset = list_next_entry(cset, mg_node); |
2052 | task = NULL; | |
2053 | } | |
2f7ee569 | 2054 | |
b3dc094e | 2055 | return NULL; |
2f7ee569 | 2056 | } |
2f7ee569 | 2057 | |
cb0f1fe9 | 2058 | /** |
74a1166d | 2059 | * cgroup_task_migrate - move a task from one cgroup to another. |
60106946 | 2060 | * @old_cgrp: the cgroup @tsk is being migrated from |
cb0f1fe9 TH |
2061 | * @tsk: the task being migrated |
2062 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 2063 | * |
cb0f1fe9 | 2064 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 2065 | */ |
5abb8855 TH |
2066 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
2067 | struct task_struct *tsk, | |
2068 | struct css_set *new_cset) | |
74a1166d | 2069 | { |
5abb8855 | 2070 | struct css_set *old_cset; |
74a1166d | 2071 | |
cb0f1fe9 TH |
2072 | lockdep_assert_held(&cgroup_mutex); |
2073 | lockdep_assert_held(&css_set_rwsem); | |
2074 | ||
74a1166d | 2075 | /* |
d59cfc09 TH |
2076 | * We are synchronized through cgroup_threadgroup_rwsem against |
2077 | * PF_EXITING setting such that we can't race against cgroup_exit() | |
2078 | * changing the css_set to init_css_set and dropping the old one. | |
74a1166d | 2079 | */ |
c84cdf75 | 2080 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 2081 | old_cset = task_css_set(tsk); |
74a1166d | 2082 | |
b3dc094e | 2083 | get_css_set(new_cset); |
5abb8855 | 2084 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 2085 | |
1b9aba49 TH |
2086 | /* |
2087 | * Use move_tail so that cgroup_taskset_first() still returns the | |
2088 | * leader after migration. This works because cgroup_migrate() | |
2089 | * ensures that the dst_cset of the leader is the first on the | |
2090 | * tset's dst_csets list. | |
2091 | */ | |
2092 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
2093 | |
2094 | /* | |
5abb8855 TH |
2095 | * We just gained a reference on old_cset by taking it from the |
2096 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
2097 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 2098 | */ |
a25eb52e | 2099 | put_css_set_locked(old_cset); |
74a1166d BB |
2100 | } |
2101 | ||
a043e3b2 | 2102 | /** |
1958d2d5 TH |
2103 | * cgroup_migrate_finish - cleanup after attach |
2104 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2105 | * |
1958d2d5 TH |
2106 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2107 | * those functions for details. | |
74a1166d | 2108 | */ |
1958d2d5 | 2109 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2110 | { |
1958d2d5 | 2111 | struct css_set *cset, *tmp_cset; |
74a1166d | 2112 | |
1958d2d5 TH |
2113 | lockdep_assert_held(&cgroup_mutex); |
2114 | ||
2115 | down_write(&css_set_rwsem); | |
2116 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2117 | cset->mg_src_cgrp = NULL; | |
2118 | cset->mg_dst_cset = NULL; | |
2119 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2120 | put_css_set_locked(cset); |
1958d2d5 TH |
2121 | } |
2122 | up_write(&css_set_rwsem); | |
2123 | } | |
2124 | ||
2125 | /** | |
2126 | * cgroup_migrate_add_src - add a migration source css_set | |
2127 | * @src_cset: the source css_set to add | |
2128 | * @dst_cgrp: the destination cgroup | |
2129 | * @preloaded_csets: list of preloaded css_sets | |
2130 | * | |
2131 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2132 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2133 | * up by cgroup_migrate_finish(). | |
2134 | * | |
d59cfc09 TH |
2135 | * This function may be called without holding cgroup_threadgroup_rwsem |
2136 | * even if the target is a process. Threads may be created and destroyed | |
2137 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2138 | * into play and the preloaded css_sets are guaranteed to cover all | |
2139 | * migrations. | |
1958d2d5 TH |
2140 | */ |
2141 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2142 | struct cgroup *dst_cgrp, | |
2143 | struct list_head *preloaded_csets) | |
2144 | { | |
2145 | struct cgroup *src_cgrp; | |
2146 | ||
2147 | lockdep_assert_held(&cgroup_mutex); | |
2148 | lockdep_assert_held(&css_set_rwsem); | |
2149 | ||
2150 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2151 | ||
1958d2d5 TH |
2152 | if (!list_empty(&src_cset->mg_preload_node)) |
2153 | return; | |
2154 | ||
2155 | WARN_ON(src_cset->mg_src_cgrp); | |
2156 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2157 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2158 | ||
2159 | src_cset->mg_src_cgrp = src_cgrp; | |
2160 | get_css_set(src_cset); | |
2161 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2162 | } | |
2163 | ||
2164 | /** | |
2165 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2166 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2167 | * @preloaded_csets: list of preloaded source css_sets |
2168 | * | |
2169 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2170 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2171 | * pins all destination css_sets, links each to its source, and append them |
2172 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2173 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2174 | * |
2175 | * This function must be called after cgroup_migrate_add_src() has been | |
2176 | * called on each migration source css_set. After migration is performed | |
2177 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2178 | * @preloaded_csets. | |
2179 | */ | |
2180 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2181 | struct list_head *preloaded_csets) | |
2182 | { | |
2183 | LIST_HEAD(csets); | |
f817de98 | 2184 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2185 | |
2186 | lockdep_assert_held(&cgroup_mutex); | |
2187 | ||
f8f22e53 TH |
2188 | /* |
2189 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2190 | * with tasks so that child cgroups don't compete against tasks. | |
2191 | */ | |
d51f39b0 | 2192 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2193 | dst_cgrp->child_subsys_mask) |
2194 | return -EBUSY; | |
2195 | ||
1958d2d5 | 2196 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2197 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2198 | struct css_set *dst_cset; |
2199 | ||
f817de98 TH |
2200 | dst_cset = find_css_set(src_cset, |
2201 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2202 | if (!dst_cset) |
2203 | goto err; | |
2204 | ||
2205 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2206 | |
2207 | /* | |
2208 | * If src cset equals dst, it's noop. Drop the src. | |
2209 | * cgroup_migrate() will skip the cset too. Note that we | |
2210 | * can't handle src == dst as some nodes are used by both. | |
2211 | */ | |
2212 | if (src_cset == dst_cset) { | |
2213 | src_cset->mg_src_cgrp = NULL; | |
2214 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2215 | put_css_set(src_cset); |
2216 | put_css_set(dst_cset); | |
f817de98 TH |
2217 | continue; |
2218 | } | |
2219 | ||
1958d2d5 TH |
2220 | src_cset->mg_dst_cset = dst_cset; |
2221 | ||
2222 | if (list_empty(&dst_cset->mg_preload_node)) | |
2223 | list_add(&dst_cset->mg_preload_node, &csets); | |
2224 | else | |
a25eb52e | 2225 | put_css_set(dst_cset); |
1958d2d5 TH |
2226 | } |
2227 | ||
f817de98 | 2228 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2229 | return 0; |
2230 | err: | |
2231 | cgroup_migrate_finish(&csets); | |
2232 | return -ENOMEM; | |
2233 | } | |
2234 | ||
2235 | /** | |
2236 | * cgroup_migrate - migrate a process or task to a cgroup | |
2237 | * @cgrp: the destination cgroup | |
2238 | * @leader: the leader of the process or the task to migrate | |
2239 | * @threadgroup: whether @leader points to the whole process or a single task | |
2240 | * | |
2241 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
d59cfc09 | 2242 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2243 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2244 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2245 | * function and following up with cgroup_migrate_finish(). | |
2246 | * | |
2247 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2248 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2249 | * failure, when migrating multiple targets, the success or failure can be | |
2250 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2251 | * actually starting migrating. | |
2252 | */ | |
2253 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2254 | bool threadgroup) | |
74a1166d | 2255 | { |
b3dc094e TH |
2256 | struct cgroup_taskset tset = { |
2257 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2258 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2259 | .csets = &tset.src_csets, | |
2260 | }; | |
1c6727af | 2261 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2262 | struct css_set *cset, *tmp_cset; |
2263 | struct task_struct *task, *tmp_task; | |
2264 | int i, ret; | |
74a1166d | 2265 | |
fb5d2b4c MSB |
2266 | /* |
2267 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2268 | * already PF_EXITING could be freed from underneath us unless we | |
2269 | * take an rcu_read_lock. | |
2270 | */ | |
b3dc094e | 2271 | down_write(&css_set_rwsem); |
fb5d2b4c | 2272 | rcu_read_lock(); |
9db8de37 | 2273 | task = leader; |
74a1166d | 2274 | do { |
9db8de37 TH |
2275 | /* @task either already exited or can't exit until the end */ |
2276 | if (task->flags & PF_EXITING) | |
ea84753c | 2277 | goto next; |
134d3373 | 2278 | |
eaf797ab TH |
2279 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2280 | if (list_empty(&task->cg_list)) | |
ea84753c | 2281 | goto next; |
cd3d0952 | 2282 | |
b3dc094e | 2283 | cset = task_css_set(task); |
1958d2d5 | 2284 | if (!cset->mg_src_cgrp) |
ea84753c | 2285 | goto next; |
b3dc094e | 2286 | |
61d1d219 | 2287 | /* |
1b9aba49 TH |
2288 | * cgroup_taskset_first() must always return the leader. |
2289 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2290 | */ |
1b9aba49 TH |
2291 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2292 | if (list_empty(&cset->mg_node)) | |
2293 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2294 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2295 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2296 | &tset.dst_csets); | |
ea84753c | 2297 | next: |
081aa458 LZ |
2298 | if (!threadgroup) |
2299 | break; | |
9db8de37 | 2300 | } while_each_thread(leader, task); |
fb5d2b4c | 2301 | rcu_read_unlock(); |
b3dc094e | 2302 | up_write(&css_set_rwsem); |
74a1166d | 2303 | |
134d3373 | 2304 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2305 | if (list_empty(&tset.src_csets)) |
2306 | return 0; | |
134d3373 | 2307 | |
1958d2d5 | 2308 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2309 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2310 | if (css->ss->can_attach) { |
9db8de37 TH |
2311 | ret = css->ss->can_attach(css, &tset); |
2312 | if (ret) { | |
1c6727af | 2313 | failed_css = css; |
74a1166d BB |
2314 | goto out_cancel_attach; |
2315 | } | |
2316 | } | |
74a1166d BB |
2317 | } |
2318 | ||
2319 | /* | |
1958d2d5 TH |
2320 | * Now that we're guaranteed success, proceed to move all tasks to |
2321 | * the new cgroup. There are no failure cases after here, so this | |
2322 | * is the commit point. | |
74a1166d | 2323 | */ |
cb0f1fe9 | 2324 | down_write(&css_set_rwsem); |
b3dc094e TH |
2325 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2326 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2327 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2328 | cset->mg_dst_cset); | |
74a1166d | 2329 | } |
cb0f1fe9 | 2330 | up_write(&css_set_rwsem); |
74a1166d BB |
2331 | |
2332 | /* | |
1958d2d5 TH |
2333 | * Migration is committed, all target tasks are now on dst_csets. |
2334 | * Nothing is sensitive to fork() after this point. Notify | |
2335 | * controllers that migration is complete. | |
74a1166d | 2336 | */ |
1958d2d5 | 2337 | tset.csets = &tset.dst_csets; |
74a1166d | 2338 | |
aec3dfcb | 2339 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2340 | if (css->ss->attach) |
2341 | css->ss->attach(css, &tset); | |
74a1166d | 2342 | |
9db8de37 | 2343 | ret = 0; |
b3dc094e TH |
2344 | goto out_release_tset; |
2345 | ||
74a1166d | 2346 | out_cancel_attach: |
aec3dfcb | 2347 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2348 | if (css == failed_css) |
2349 | break; | |
2350 | if (css->ss->cancel_attach) | |
2351 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2352 | } |
b3dc094e TH |
2353 | out_release_tset: |
2354 | down_write(&css_set_rwsem); | |
2355 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2356 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2357 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2358 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2359 | } |
2360 | up_write(&css_set_rwsem); | |
9db8de37 | 2361 | return ret; |
74a1166d BB |
2362 | } |
2363 | ||
1958d2d5 TH |
2364 | /** |
2365 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2366 | * @dst_cgrp: the cgroup to attach to | |
2367 | * @leader: the task or the leader of the threadgroup to be attached | |
2368 | * @threadgroup: attach the whole threadgroup? | |
2369 | * | |
d59cfc09 | 2370 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2371 | */ |
2372 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2373 | struct task_struct *leader, bool threadgroup) | |
2374 | { | |
2375 | LIST_HEAD(preloaded_csets); | |
2376 | struct task_struct *task; | |
2377 | int ret; | |
2378 | ||
2379 | /* look up all src csets */ | |
2380 | down_read(&css_set_rwsem); | |
2381 | rcu_read_lock(); | |
2382 | task = leader; | |
2383 | do { | |
2384 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2385 | &preloaded_csets); | |
2386 | if (!threadgroup) | |
2387 | break; | |
2388 | } while_each_thread(leader, task); | |
2389 | rcu_read_unlock(); | |
2390 | up_read(&css_set_rwsem); | |
2391 | ||
2392 | /* prepare dst csets and commit */ | |
2393 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2394 | if (!ret) | |
2395 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2396 | ||
2397 | cgroup_migrate_finish(&preloaded_csets); | |
2398 | return ret; | |
74a1166d BB |
2399 | } |
2400 | ||
187fe840 TH |
2401 | static int cgroup_procs_write_permission(struct task_struct *task, |
2402 | struct cgroup *dst_cgrp, | |
2403 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2404 | { |
2405 | const struct cred *cred = current_cred(); | |
2406 | const struct cred *tcred = get_task_cred(task); | |
2407 | int ret = 0; | |
2408 | ||
2409 | /* | |
2410 | * even if we're attaching all tasks in the thread group, we only | |
2411 | * need to check permissions on one of them. | |
2412 | */ | |
2413 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2414 | !uid_eq(cred->euid, tcred->uid) && | |
2415 | !uid_eq(cred->euid, tcred->suid)) | |
2416 | ret = -EACCES; | |
2417 | ||
187fe840 TH |
2418 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2419 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2420 | struct cgroup *cgrp; | |
2421 | struct inode *inode; | |
2422 | ||
2423 | down_read(&css_set_rwsem); | |
2424 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
2425 | up_read(&css_set_rwsem); | |
2426 | ||
2427 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2428 | cgrp = cgroup_parent(cgrp); | |
2429 | ||
2430 | ret = -ENOMEM; | |
2431 | inode = kernfs_get_inode(sb, cgrp->procs_kn); | |
2432 | if (inode) { | |
2433 | ret = inode_permission(inode, MAY_WRITE); | |
2434 | iput(inode); | |
2435 | } | |
2436 | } | |
2437 | ||
dedf22e9 TH |
2438 | put_cred(tcred); |
2439 | return ret; | |
2440 | } | |
2441 | ||
74a1166d BB |
2442 | /* |
2443 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2444 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2445 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2446 | */ |
acbef755 TH |
2447 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2448 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2449 | { |
bbcb81d0 | 2450 | struct task_struct *tsk; |
e76ecaee | 2451 | struct cgroup *cgrp; |
acbef755 | 2452 | pid_t pid; |
bbcb81d0 PM |
2453 | int ret; |
2454 | ||
acbef755 TH |
2455 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2456 | return -EINVAL; | |
2457 | ||
e76ecaee TH |
2458 | cgrp = cgroup_kn_lock_live(of->kn); |
2459 | if (!cgrp) | |
74a1166d BB |
2460 | return -ENODEV; |
2461 | ||
b5ba75b5 | 2462 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2463 | rcu_read_lock(); |
bbcb81d0 | 2464 | if (pid) { |
73507f33 | 2465 | tsk = find_task_by_vpid(pid); |
74a1166d | 2466 | if (!tsk) { |
dd4b0a46 | 2467 | ret = -ESRCH; |
b5ba75b5 | 2468 | goto out_unlock_rcu; |
bbcb81d0 | 2469 | } |
dedf22e9 | 2470 | } else { |
b78949eb | 2471 | tsk = current; |
dedf22e9 | 2472 | } |
cd3d0952 TH |
2473 | |
2474 | if (threadgroup) | |
b78949eb | 2475 | tsk = tsk->group_leader; |
c4c27fbd MG |
2476 | |
2477 | /* | |
14a40ffc | 2478 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2479 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2480 | * with no rt_runtime allocated. Just say no. | |
2481 | */ | |
14a40ffc | 2482 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2483 | ret = -EINVAL; |
b5ba75b5 | 2484 | goto out_unlock_rcu; |
c4c27fbd MG |
2485 | } |
2486 | ||
b78949eb MSB |
2487 | get_task_struct(tsk); |
2488 | rcu_read_unlock(); | |
2489 | ||
187fe840 | 2490 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2491 | if (!ret) |
2492 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2493 | |
bbcb81d0 | 2494 | put_task_struct(tsk); |
b5ba75b5 TH |
2495 | goto out_unlock_threadgroup; |
2496 | ||
2497 | out_unlock_rcu: | |
2498 | rcu_read_unlock(); | |
2499 | out_unlock_threadgroup: | |
2500 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2501 | cgroup_kn_unlock(of->kn); |
acbef755 | 2502 | return ret ?: nbytes; |
bbcb81d0 PM |
2503 | } |
2504 | ||
7ae1bad9 TH |
2505 | /** |
2506 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2507 | * @from: attach to all cgroups of a given task | |
2508 | * @tsk: the task to be attached | |
2509 | */ | |
2510 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2511 | { | |
3dd06ffa | 2512 | struct cgroup_root *root; |
7ae1bad9 TH |
2513 | int retval = 0; |
2514 | ||
47cfcd09 | 2515 | mutex_lock(&cgroup_mutex); |
985ed670 | 2516 | for_each_root(root) { |
96d365e0 TH |
2517 | struct cgroup *from_cgrp; |
2518 | ||
3dd06ffa | 2519 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2520 | continue; |
2521 | ||
96d365e0 TH |
2522 | down_read(&css_set_rwsem); |
2523 | from_cgrp = task_cgroup_from_root(from, root); | |
2524 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2525 | |
6f4b7e63 | 2526 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2527 | if (retval) |
2528 | break; | |
2529 | } | |
47cfcd09 | 2530 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2531 | |
2532 | return retval; | |
2533 | } | |
2534 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2535 | ||
acbef755 TH |
2536 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2537 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2538 | { |
acbef755 | 2539 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2540 | } |
2541 | ||
acbef755 TH |
2542 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2543 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2544 | { |
acbef755 | 2545 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2546 | } |
2547 | ||
451af504 TH |
2548 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2549 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2550 | { |
e76ecaee | 2551 | struct cgroup *cgrp; |
5f469907 | 2552 | |
e76ecaee | 2553 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2554 | |
e76ecaee TH |
2555 | cgrp = cgroup_kn_lock_live(of->kn); |
2556 | if (!cgrp) | |
e788e066 | 2557 | return -ENODEV; |
69e943b7 | 2558 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2559 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2560 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2561 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2562 | cgroup_kn_unlock(of->kn); |
451af504 | 2563 | return nbytes; |
e788e066 PM |
2564 | } |
2565 | ||
2da8ca82 | 2566 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2567 | { |
2da8ca82 | 2568 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2569 | |
46cfeb04 | 2570 | spin_lock(&release_agent_path_lock); |
e788e066 | 2571 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2572 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2573 | seq_putc(seq, '\n'); |
e788e066 PM |
2574 | return 0; |
2575 | } | |
2576 | ||
2da8ca82 | 2577 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2578 | { |
c1d5d42e | 2579 | seq_puts(seq, "0\n"); |
e788e066 PM |
2580 | return 0; |
2581 | } | |
2582 | ||
8ab456ac | 2583 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned long ss_mask) |
355e0c48 | 2584 | { |
f8f22e53 TH |
2585 | struct cgroup_subsys *ss; |
2586 | bool printed = false; | |
2587 | int ssid; | |
a742c59d | 2588 | |
a966a4ed AS |
2589 | for_each_subsys_which(ss, ssid, &ss_mask) { |
2590 | if (printed) | |
2591 | seq_putc(seq, ' '); | |
2592 | seq_printf(seq, "%s", ss->name); | |
2593 | printed = true; | |
e73d2c61 | 2594 | } |
f8f22e53 TH |
2595 | if (printed) |
2596 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2597 | } |
2598 | ||
f8f22e53 TH |
2599 | /* show controllers which are currently attached to the default hierarchy */ |
2600 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2601 | { |
f8f22e53 TH |
2602 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2603 | ||
5533e011 TH |
2604 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2605 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2606 | return 0; |
db3b1497 PM |
2607 | } |
2608 | ||
f8f22e53 TH |
2609 | /* show controllers which are enabled from the parent */ |
2610 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2611 | { |
f8f22e53 TH |
2612 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2613 | ||
667c2491 | 2614 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2615 | return 0; |
ddbcc7e8 PM |
2616 | } |
2617 | ||
f8f22e53 TH |
2618 | /* show controllers which are enabled for a given cgroup's children */ |
2619 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2620 | { |
f8f22e53 TH |
2621 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2622 | ||
667c2491 | 2623 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2624 | return 0; |
2625 | } | |
2626 | ||
2627 | /** | |
2628 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2629 | * @cgrp: root of the subtree to update csses for | |
2630 | * | |
2631 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2632 | * css associations need to be updated accordingly. This function looks up | |
2633 | * all css_sets which are attached to the subtree, creates the matching | |
2634 | * updated css_sets and migrates the tasks to the new ones. | |
2635 | */ | |
2636 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2637 | { | |
2638 | LIST_HEAD(preloaded_csets); | |
2639 | struct cgroup_subsys_state *css; | |
2640 | struct css_set *src_cset; | |
2641 | int ret; | |
2642 | ||
f8f22e53 TH |
2643 | lockdep_assert_held(&cgroup_mutex); |
2644 | ||
b5ba75b5 TH |
2645 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2646 | ||
f8f22e53 TH |
2647 | /* look up all csses currently attached to @cgrp's subtree */ |
2648 | down_read(&css_set_rwsem); | |
2649 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2650 | struct cgrp_cset_link *link; | |
2651 | ||
2652 | /* self is not affected by child_subsys_mask change */ | |
2653 | if (css->cgroup == cgrp) | |
2654 | continue; | |
2655 | ||
2656 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2657 | cgroup_migrate_add_src(link->cset, cgrp, | |
2658 | &preloaded_csets); | |
2659 | } | |
2660 | up_read(&css_set_rwsem); | |
2661 | ||
2662 | /* NULL dst indicates self on default hierarchy */ | |
2663 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2664 | if (ret) | |
2665 | goto out_finish; | |
2666 | ||
2667 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2668 | struct task_struct *last_task = NULL, *task; | |
2669 | ||
2670 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2671 | if (!src_cset->mg_src_cgrp) | |
2672 | break; | |
2673 | ||
2674 | /* | |
2675 | * All tasks in src_cset need to be migrated to the | |
2676 | * matching dst_cset. Empty it process by process. We | |
2677 | * walk tasks but migrate processes. The leader might even | |
2678 | * belong to a different cset but such src_cset would also | |
2679 | * be among the target src_csets because the default | |
2680 | * hierarchy enforces per-process membership. | |
2681 | */ | |
2682 | while (true) { | |
2683 | down_read(&css_set_rwsem); | |
2684 | task = list_first_entry_or_null(&src_cset->tasks, | |
2685 | struct task_struct, cg_list); | |
2686 | if (task) { | |
2687 | task = task->group_leader; | |
2688 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2689 | get_task_struct(task); | |
2690 | } | |
2691 | up_read(&css_set_rwsem); | |
2692 | ||
2693 | if (!task) | |
2694 | break; | |
2695 | ||
2696 | /* guard against possible infinite loop */ | |
2697 | if (WARN(last_task == task, | |
2698 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2699 | goto out_finish; | |
2700 | last_task = task; | |
2701 | ||
f8f22e53 TH |
2702 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); |
2703 | ||
f8f22e53 TH |
2704 | put_task_struct(task); |
2705 | ||
2706 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2707 | goto out_finish; | |
2708 | } | |
2709 | } | |
2710 | ||
2711 | out_finish: | |
2712 | cgroup_migrate_finish(&preloaded_csets); | |
b5ba75b5 | 2713 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2714 | return ret; |
2715 | } | |
2716 | ||
2717 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2718 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2719 | char *buf, size_t nbytes, | |
2720 | loff_t off) | |
f8f22e53 | 2721 | { |
8ab456ac AS |
2722 | unsigned long enable = 0, disable = 0; |
2723 | unsigned long css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
a9746d8d | 2724 | struct cgroup *cgrp, *child; |
f8f22e53 | 2725 | struct cgroup_subsys *ss; |
451af504 | 2726 | char *tok; |
f8f22e53 TH |
2727 | int ssid, ret; |
2728 | ||
2729 | /* | |
d37167ab TH |
2730 | * Parse input - space separated list of subsystem names prefixed |
2731 | * with either + or -. | |
f8f22e53 | 2732 | */ |
451af504 TH |
2733 | buf = strstrip(buf); |
2734 | while ((tok = strsep(&buf, " "))) { | |
a966a4ed AS |
2735 | unsigned long tmp_ss_mask = ~cgrp_dfl_root_inhibit_ss_mask; |
2736 | ||
d37167ab TH |
2737 | if (tok[0] == '\0') |
2738 | continue; | |
a966a4ed AS |
2739 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
2740 | if (ss->disabled || strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
2741 | continue; |
2742 | ||
2743 | if (*tok == '+') { | |
7d331fa9 TH |
2744 | enable |= 1 << ssid; |
2745 | disable &= ~(1 << ssid); | |
f8f22e53 | 2746 | } else if (*tok == '-') { |
7d331fa9 TH |
2747 | disable |= 1 << ssid; |
2748 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2749 | } else { |
2750 | return -EINVAL; | |
2751 | } | |
2752 | break; | |
2753 | } | |
2754 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2755 | return -EINVAL; | |
2756 | } | |
2757 | ||
a9746d8d TH |
2758 | cgrp = cgroup_kn_lock_live(of->kn); |
2759 | if (!cgrp) | |
2760 | return -ENODEV; | |
f8f22e53 TH |
2761 | |
2762 | for_each_subsys(ss, ssid) { | |
2763 | if (enable & (1 << ssid)) { | |
667c2491 | 2764 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2765 | enable &= ~(1 << ssid); |
2766 | continue; | |
2767 | } | |
2768 | ||
c29adf24 TH |
2769 | /* unavailable or not enabled on the parent? */ |
2770 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2771 | (cgroup_parent(cgrp) && | |
667c2491 | 2772 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
2773 | ret = -ENOENT; |
2774 | goto out_unlock; | |
2775 | } | |
f8f22e53 | 2776 | } else if (disable & (1 << ssid)) { |
667c2491 | 2777 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2778 | disable &= ~(1 << ssid); |
2779 | continue; | |
2780 | } | |
2781 | ||
2782 | /* a child has it enabled? */ | |
2783 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2784 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2785 | ret = -EBUSY; |
ddab2b6e | 2786 | goto out_unlock; |
f8f22e53 TH |
2787 | } |
2788 | } | |
2789 | } | |
2790 | } | |
2791 | ||
2792 | if (!enable && !disable) { | |
2793 | ret = 0; | |
ddab2b6e | 2794 | goto out_unlock; |
f8f22e53 TH |
2795 | } |
2796 | ||
2797 | /* | |
667c2491 | 2798 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2799 | * with tasks so that child cgroups don't compete against tasks. |
2800 | */ | |
d51f39b0 | 2801 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
2802 | ret = -EBUSY; |
2803 | goto out_unlock; | |
2804 | } | |
2805 | ||
2806 | /* | |
f63070d3 TH |
2807 | * Update subsys masks and calculate what needs to be done. More |
2808 | * subsystems than specified may need to be enabled or disabled | |
2809 | * depending on subsystem dependencies. | |
2810 | */ | |
755bf5ee TH |
2811 | old_sc = cgrp->subtree_control; |
2812 | old_ss = cgrp->child_subsys_mask; | |
2813 | new_sc = (old_sc | enable) & ~disable; | |
2814 | new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc); | |
f63070d3 | 2815 | |
755bf5ee TH |
2816 | css_enable = ~old_ss & new_ss; |
2817 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
2818 | enable |= css_enable; |
2819 | disable |= css_disable; | |
c29adf24 | 2820 | |
db6e3053 TH |
2821 | /* |
2822 | * Because css offlining is asynchronous, userland might try to | |
2823 | * re-enable the same controller while the previous instance is | |
2824 | * still around. In such cases, wait till it's gone using | |
2825 | * offline_waitq. | |
2826 | */ | |
a966a4ed | 2827 | for_each_subsys_which(ss, ssid, &css_enable) { |
db6e3053 TH |
2828 | cgroup_for_each_live_child(child, cgrp) { |
2829 | DEFINE_WAIT(wait); | |
2830 | ||
2831 | if (!cgroup_css(child, ss)) | |
2832 | continue; | |
2833 | ||
2834 | cgroup_get(child); | |
2835 | prepare_to_wait(&child->offline_waitq, &wait, | |
2836 | TASK_UNINTERRUPTIBLE); | |
2837 | cgroup_kn_unlock(of->kn); | |
2838 | schedule(); | |
2839 | finish_wait(&child->offline_waitq, &wait); | |
2840 | cgroup_put(child); | |
2841 | ||
2842 | return restart_syscall(); | |
2843 | } | |
2844 | } | |
2845 | ||
755bf5ee TH |
2846 | cgrp->subtree_control = new_sc; |
2847 | cgrp->child_subsys_mask = new_ss; | |
2848 | ||
f63070d3 TH |
2849 | /* |
2850 | * Create new csses or make the existing ones visible. A css is | |
2851 | * created invisible if it's being implicitly enabled through | |
2852 | * dependency. An invisible css is made visible when the userland | |
2853 | * explicitly enables it. | |
f8f22e53 TH |
2854 | */ |
2855 | for_each_subsys(ss, ssid) { | |
2856 | if (!(enable & (1 << ssid))) | |
2857 | continue; | |
2858 | ||
2859 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
2860 | if (css_enable & (1 << ssid)) |
2861 | ret = create_css(child, ss, | |
2862 | cgrp->subtree_control & (1 << ssid)); | |
2863 | else | |
2864 | ret = cgroup_populate_dir(child, 1 << ssid); | |
f8f22e53 TH |
2865 | if (ret) |
2866 | goto err_undo_css; | |
2867 | } | |
2868 | } | |
2869 | ||
c29adf24 TH |
2870 | /* |
2871 | * At this point, cgroup_e_css() results reflect the new csses | |
2872 | * making the following cgroup_update_dfl_csses() properly update | |
2873 | * css associations of all tasks in the subtree. | |
2874 | */ | |
f8f22e53 TH |
2875 | ret = cgroup_update_dfl_csses(cgrp); |
2876 | if (ret) | |
2877 | goto err_undo_css; | |
2878 | ||
f63070d3 TH |
2879 | /* |
2880 | * All tasks are migrated out of disabled csses. Kill or hide | |
2881 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
2882 | * disabled while other subsystems are still depending on it. The |
2883 | * css must not actively control resources and be in the vanilla | |
2884 | * state if it's made visible again later. Controllers which may | |
2885 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 2886 | */ |
f8f22e53 TH |
2887 | for_each_subsys(ss, ssid) { |
2888 | if (!(disable & (1 << ssid))) | |
2889 | continue; | |
2890 | ||
f63070d3 | 2891 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
2892 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
2893 | ||
2894 | if (css_disable & (1 << ssid)) { | |
2895 | kill_css(css); | |
2896 | } else { | |
f63070d3 | 2897 | cgroup_clear_dir(child, 1 << ssid); |
b4536f0c TH |
2898 | if (ss->css_reset) |
2899 | ss->css_reset(css); | |
2900 | } | |
f63070d3 | 2901 | } |
f8f22e53 TH |
2902 | } |
2903 | ||
56c807ba TH |
2904 | /* |
2905 | * The effective csses of all the descendants (excluding @cgrp) may | |
2906 | * have changed. Subsystems can optionally subscribe to this event | |
2907 | * by implementing ->css_e_css_changed() which is invoked if any of | |
2908 | * the effective csses seen from the css's cgroup may have changed. | |
2909 | */ | |
2910 | for_each_subsys(ss, ssid) { | |
2911 | struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss); | |
2912 | struct cgroup_subsys_state *css; | |
2913 | ||
2914 | if (!ss->css_e_css_changed || !this_css) | |
2915 | continue; | |
2916 | ||
2917 | css_for_each_descendant_pre(css, this_css) | |
2918 | if (css != this_css) | |
2919 | ss->css_e_css_changed(css); | |
2920 | } | |
2921 | ||
f8f22e53 TH |
2922 | kernfs_activate(cgrp->kn); |
2923 | ret = 0; | |
2924 | out_unlock: | |
a9746d8d | 2925 | cgroup_kn_unlock(of->kn); |
451af504 | 2926 | return ret ?: nbytes; |
f8f22e53 TH |
2927 | |
2928 | err_undo_css: | |
755bf5ee TH |
2929 | cgrp->subtree_control = old_sc; |
2930 | cgrp->child_subsys_mask = old_ss; | |
f8f22e53 TH |
2931 | |
2932 | for_each_subsys(ss, ssid) { | |
2933 | if (!(enable & (1 << ssid))) | |
2934 | continue; | |
2935 | ||
2936 | cgroup_for_each_live_child(child, cgrp) { | |
2937 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
2938 | |
2939 | if (!css) | |
2940 | continue; | |
2941 | ||
2942 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 2943 | kill_css(css); |
f63070d3 TH |
2944 | else |
2945 | cgroup_clear_dir(child, 1 << ssid); | |
f8f22e53 TH |
2946 | } |
2947 | } | |
2948 | goto out_unlock; | |
2949 | } | |
2950 | ||
842b597e TH |
2951 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2952 | { | |
2953 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2954 | return 0; | |
2955 | } | |
2956 | ||
2bd59d48 TH |
2957 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2958 | size_t nbytes, loff_t off) | |
355e0c48 | 2959 | { |
2bd59d48 TH |
2960 | struct cgroup *cgrp = of->kn->parent->priv; |
2961 | struct cftype *cft = of->kn->priv; | |
2962 | struct cgroup_subsys_state *css; | |
a742c59d | 2963 | int ret; |
355e0c48 | 2964 | |
b4168640 TH |
2965 | if (cft->write) |
2966 | return cft->write(of, buf, nbytes, off); | |
2967 | ||
2bd59d48 TH |
2968 | /* |
2969 | * kernfs guarantees that a file isn't deleted with operations in | |
2970 | * flight, which means that the matching css is and stays alive and | |
2971 | * doesn't need to be pinned. The RCU locking is not necessary | |
2972 | * either. It's just for the convenience of using cgroup_css(). | |
2973 | */ | |
2974 | rcu_read_lock(); | |
2975 | css = cgroup_css(cgrp, cft->ss); | |
2976 | rcu_read_unlock(); | |
a742c59d | 2977 | |
451af504 | 2978 | if (cft->write_u64) { |
a742c59d TH |
2979 | unsigned long long v; |
2980 | ret = kstrtoull(buf, 0, &v); | |
2981 | if (!ret) | |
2982 | ret = cft->write_u64(css, cft, v); | |
2983 | } else if (cft->write_s64) { | |
2984 | long long v; | |
2985 | ret = kstrtoll(buf, 0, &v); | |
2986 | if (!ret) | |
2987 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2988 | } else { |
a742c59d | 2989 | ret = -EINVAL; |
e73d2c61 | 2990 | } |
2bd59d48 | 2991 | |
a742c59d | 2992 | return ret ?: nbytes; |
355e0c48 PM |
2993 | } |
2994 | ||
6612f05b | 2995 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2996 | { |
2bd59d48 | 2997 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2998 | } |
2999 | ||
6612f05b | 3000 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3001 | { |
2bd59d48 | 3002 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3003 | } |
3004 | ||
6612f05b | 3005 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3006 | { |
2bd59d48 | 3007 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3008 | } |
3009 | ||
91796569 | 3010 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3011 | { |
7da11279 TH |
3012 | struct cftype *cft = seq_cft(m); |
3013 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3014 | |
2da8ca82 TH |
3015 | if (cft->seq_show) |
3016 | return cft->seq_show(m, arg); | |
e73d2c61 | 3017 | |
f4c753b7 | 3018 | if (cft->read_u64) |
896f5199 TH |
3019 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3020 | else if (cft->read_s64) | |
3021 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3022 | else | |
3023 | return -EINVAL; | |
3024 | return 0; | |
91796569 PM |
3025 | } |
3026 | ||
2bd59d48 TH |
3027 | static struct kernfs_ops cgroup_kf_single_ops = { |
3028 | .atomic_write_len = PAGE_SIZE, | |
3029 | .write = cgroup_file_write, | |
3030 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3031 | }; |
3032 | ||
2bd59d48 TH |
3033 | static struct kernfs_ops cgroup_kf_ops = { |
3034 | .atomic_write_len = PAGE_SIZE, | |
3035 | .write = cgroup_file_write, | |
3036 | .seq_start = cgroup_seqfile_start, | |
3037 | .seq_next = cgroup_seqfile_next, | |
3038 | .seq_stop = cgroup_seqfile_stop, | |
3039 | .seq_show = cgroup_seqfile_show, | |
3040 | }; | |
ddbcc7e8 PM |
3041 | |
3042 | /* | |
3043 | * cgroup_rename - Only allow simple rename of directories in place. | |
3044 | */ | |
2bd59d48 TH |
3045 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3046 | const char *new_name_str) | |
ddbcc7e8 | 3047 | { |
2bd59d48 | 3048 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3049 | int ret; |
65dff759 | 3050 | |
2bd59d48 | 3051 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3052 | return -ENOTDIR; |
2bd59d48 | 3053 | if (kn->parent != new_parent) |
ddbcc7e8 | 3054 | return -EIO; |
65dff759 | 3055 | |
6db8e85c TH |
3056 | /* |
3057 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3058 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3059 | */ |
aa6ec29b | 3060 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3061 | return -EPERM; |
099fca32 | 3062 | |
e1b2dc17 | 3063 | /* |
8353da1f | 3064 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3065 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3066 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3067 | */ |
3068 | kernfs_break_active_protection(new_parent); | |
3069 | kernfs_break_active_protection(kn); | |
099fca32 | 3070 | |
2bd59d48 | 3071 | mutex_lock(&cgroup_mutex); |
099fca32 | 3072 | |
2bd59d48 | 3073 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3074 | |
2bd59d48 | 3075 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3076 | |
3077 | kernfs_unbreak_active_protection(kn); | |
3078 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3079 | return ret; |
099fca32 LZ |
3080 | } |
3081 | ||
49957f8e TH |
3082 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3083 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3084 | { | |
3085 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3086 | .ia_uid = current_fsuid(), | |
3087 | .ia_gid = current_fsgid(), }; | |
3088 | ||
3089 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3090 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3091 | return 0; | |
3092 | ||
3093 | return kernfs_setattr(kn, &iattr); | |
3094 | } | |
3095 | ||
2bb566cb | 3096 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 3097 | { |
8d7e6fb0 | 3098 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3099 | struct kernfs_node *kn; |
3100 | struct lock_class_key *key = NULL; | |
49957f8e | 3101 | int ret; |
05ef1d7c | 3102 | |
2bd59d48 TH |
3103 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3104 | key = &cft->lockdep_key; | |
3105 | #endif | |
3106 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3107 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3108 | NULL, key); |
49957f8e TH |
3109 | if (IS_ERR(kn)) |
3110 | return PTR_ERR(kn); | |
3111 | ||
3112 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3113 | if (ret) { |
49957f8e | 3114 | kernfs_remove(kn); |
f8f22e53 TH |
3115 | return ret; |
3116 | } | |
3117 | ||
187fe840 TH |
3118 | if (cft->write == cgroup_procs_write) |
3119 | cgrp->procs_kn = kn; | |
3120 | else if (cft->seq_show == cgroup_populated_show) | |
842b597e | 3121 | cgrp->populated_kn = kn; |
f8f22e53 | 3122 | return 0; |
ddbcc7e8 PM |
3123 | } |
3124 | ||
b1f28d31 TH |
3125 | /** |
3126 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
3127 | * @cgrp: the target cgroup | |
b1f28d31 TH |
3128 | * @cfts: array of cftypes to be added |
3129 | * @is_add: whether to add or remove | |
3130 | * | |
3131 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
3132 | * For removals, this function never fails. If addition fails, this |
3133 | * function doesn't remove files already added. The caller is responsible | |
3134 | * for cleaning up. | |
b1f28d31 | 3135 | */ |
2bb566cb TH |
3136 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
3137 | bool is_add) | |
ddbcc7e8 | 3138 | { |
03b1cde6 | 3139 | struct cftype *cft; |
b1f28d31 TH |
3140 | int ret; |
3141 | ||
01f6474c | 3142 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 TH |
3143 | |
3144 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3145 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3146 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3147 | continue; |
05ebb6e6 | 3148 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3149 | continue; |
d51f39b0 | 3150 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3151 | continue; |
d51f39b0 | 3152 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3153 | continue; |
3154 | ||
2739d3cc | 3155 | if (is_add) { |
2bb566cb | 3156 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 3157 | if (ret) { |
ed3d261b JP |
3158 | pr_warn("%s: failed to add %s, err=%d\n", |
3159 | __func__, cft->name, ret); | |
b1f28d31 TH |
3160 | return ret; |
3161 | } | |
2739d3cc LZ |
3162 | } else { |
3163 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3164 | } |
ddbcc7e8 | 3165 | } |
b1f28d31 | 3166 | return 0; |
ddbcc7e8 PM |
3167 | } |
3168 | ||
21a2d343 | 3169 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3170 | { |
3171 | LIST_HEAD(pending); | |
2bb566cb | 3172 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3173 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3174 | struct cgroup_subsys_state *css; |
9ccece80 | 3175 | int ret = 0; |
8e3f6541 | 3176 | |
01f6474c | 3177 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3178 | |
e8c82d20 | 3179 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3180 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3181 | struct cgroup *cgrp = css->cgroup; |
3182 | ||
e8c82d20 LZ |
3183 | if (cgroup_is_dead(cgrp)) |
3184 | continue; | |
3185 | ||
21a2d343 | 3186 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
3187 | if (ret) |
3188 | break; | |
8e3f6541 | 3189 | } |
21a2d343 TH |
3190 | |
3191 | if (is_add && !ret) | |
3192 | kernfs_activate(root->kn); | |
9ccece80 | 3193 | return ret; |
8e3f6541 TH |
3194 | } |
3195 | ||
2da440a2 | 3196 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3197 | { |
2bb566cb | 3198 | struct cftype *cft; |
8e3f6541 | 3199 | |
2bd59d48 TH |
3200 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3201 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3202 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3203 | kfree(cft->kf_ops); | |
3204 | cft->kf_ops = NULL; | |
2da440a2 | 3205 | cft->ss = NULL; |
a8ddc821 TH |
3206 | |
3207 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3208 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3209 | } |
2da440a2 TH |
3210 | } |
3211 | ||
2bd59d48 | 3212 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3213 | { |
3214 | struct cftype *cft; | |
3215 | ||
2bd59d48 TH |
3216 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3217 | struct kernfs_ops *kf_ops; | |
3218 | ||
0adb0704 TH |
3219 | WARN_ON(cft->ss || cft->kf_ops); |
3220 | ||
2bd59d48 TH |
3221 | if (cft->seq_start) |
3222 | kf_ops = &cgroup_kf_ops; | |
3223 | else | |
3224 | kf_ops = &cgroup_kf_single_ops; | |
3225 | ||
3226 | /* | |
3227 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3228 | * make a copy of kf_ops to set its atomic_write_len. | |
3229 | */ | |
3230 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3231 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3232 | if (!kf_ops) { | |
3233 | cgroup_exit_cftypes(cfts); | |
3234 | return -ENOMEM; | |
3235 | } | |
3236 | kf_ops->atomic_write_len = cft->max_write_len; | |
3237 | } | |
8e3f6541 | 3238 | |
2bd59d48 | 3239 | cft->kf_ops = kf_ops; |
2bb566cb | 3240 | cft->ss = ss; |
2bd59d48 | 3241 | } |
2bb566cb | 3242 | |
2bd59d48 | 3243 | return 0; |
2da440a2 TH |
3244 | } |
3245 | ||
21a2d343 TH |
3246 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3247 | { | |
01f6474c | 3248 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3249 | |
3250 | if (!cfts || !cfts[0].ss) | |
3251 | return -ENOENT; | |
3252 | ||
3253 | list_del(&cfts->node); | |
3254 | cgroup_apply_cftypes(cfts, false); | |
3255 | cgroup_exit_cftypes(cfts); | |
3256 | return 0; | |
8e3f6541 | 3257 | } |
8e3f6541 | 3258 | |
79578621 TH |
3259 | /** |
3260 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3261 | * @cfts: zero-length name terminated array of cftypes |
3262 | * | |
2bb566cb TH |
3263 | * Unregister @cfts. Files described by @cfts are removed from all |
3264 | * existing cgroups and all future cgroups won't have them either. This | |
3265 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3266 | * |
3267 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3268 | * registered. |
79578621 | 3269 | */ |
2bb566cb | 3270 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3271 | { |
21a2d343 | 3272 | int ret; |
79578621 | 3273 | |
01f6474c | 3274 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3275 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3276 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3277 | return ret; |
80b13586 TH |
3278 | } |
3279 | ||
8e3f6541 TH |
3280 | /** |
3281 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3282 | * @ss: target cgroup subsystem | |
3283 | * @cfts: zero-length name terminated array of cftypes | |
3284 | * | |
3285 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3286 | * existing cgroups to which @ss is attached and all future cgroups will | |
3287 | * have them too. This function can be called anytime whether @ss is | |
3288 | * attached or not. | |
3289 | * | |
3290 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3291 | * function currently returns 0 as long as @cfts registration is successful | |
3292 | * even if some file creation attempts on existing cgroups fail. | |
3293 | */ | |
2cf669a5 | 3294 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3295 | { |
9ccece80 | 3296 | int ret; |
8e3f6541 | 3297 | |
c731ae1d LZ |
3298 | if (ss->disabled) |
3299 | return 0; | |
3300 | ||
dc5736ed LZ |
3301 | if (!cfts || cfts[0].name[0] == '\0') |
3302 | return 0; | |
2bb566cb | 3303 | |
2bd59d48 TH |
3304 | ret = cgroup_init_cftypes(ss, cfts); |
3305 | if (ret) | |
3306 | return ret; | |
79578621 | 3307 | |
01f6474c | 3308 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3309 | |
0adb0704 | 3310 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3311 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3312 | if (ret) |
21a2d343 | 3313 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3314 | |
01f6474c | 3315 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3316 | return ret; |
79578621 TH |
3317 | } |
3318 | ||
a8ddc821 TH |
3319 | /** |
3320 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3321 | * @ss: target cgroup subsystem | |
3322 | * @cfts: zero-length name terminated array of cftypes | |
3323 | * | |
3324 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3325 | * the default hierarchy. | |
3326 | */ | |
3327 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3328 | { | |
3329 | struct cftype *cft; | |
3330 | ||
3331 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3332 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3333 | return cgroup_add_cftypes(ss, cfts); |
3334 | } | |
3335 | ||
3336 | /** | |
3337 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3338 | * @ss: target cgroup subsystem | |
3339 | * @cfts: zero-length name terminated array of cftypes | |
3340 | * | |
3341 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3342 | * the legacy hierarchies. | |
3343 | */ | |
2cf669a5 TH |
3344 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3345 | { | |
a8ddc821 TH |
3346 | struct cftype *cft; |
3347 | ||
fa8137be VG |
3348 | /* |
3349 | * If legacy_flies_on_dfl, we want to show the legacy files on the | |
3350 | * dfl hierarchy but iff the target subsystem hasn't been updated | |
3351 | * for the dfl hierarchy yet. | |
3352 | */ | |
3353 | if (!cgroup_legacy_files_on_dfl || | |
3354 | ss->dfl_cftypes != ss->legacy_cftypes) { | |
3355 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
3356 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
3357 | } | |
3358 | ||
2cf669a5 TH |
3359 | return cgroup_add_cftypes(ss, cfts); |
3360 | } | |
3361 | ||
a043e3b2 LZ |
3362 | /** |
3363 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3364 | * @cgrp: the cgroup in question | |
3365 | * | |
3366 | * Return the number of tasks in the cgroup. | |
3367 | */ | |
07bc356e | 3368 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3369 | { |
3370 | int count = 0; | |
69d0206c | 3371 | struct cgrp_cset_link *link; |
817929ec | 3372 | |
96d365e0 | 3373 | down_read(&css_set_rwsem); |
69d0206c TH |
3374 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3375 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3376 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3377 | return count; |
3378 | } | |
3379 | ||
53fa5261 | 3380 | /** |
492eb21b | 3381 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3382 | * @pos: the current position (%NULL to initiate traversal) |
3383 | * @parent: css whose children to walk | |
53fa5261 | 3384 | * |
c2931b70 | 3385 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3386 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3387 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3388 | * be returned regardless of their states. | |
3389 | * | |
3390 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3391 | * css which finished ->css_online() is guaranteed to be visible in the | |
3392 | * future iterations and will stay visible until the last reference is put. | |
3393 | * A css which hasn't finished ->css_online() or already finished | |
3394 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3395 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3396 | */ |
c2931b70 TH |
3397 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3398 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3399 | { |
c2931b70 | 3400 | struct cgroup_subsys_state *next; |
53fa5261 | 3401 | |
8353da1f | 3402 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3403 | |
3404 | /* | |
de3f0341 TH |
3405 | * @pos could already have been unlinked from the sibling list. |
3406 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3407 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3408 | * @pos is taken off list, at which time its next pointer is valid, | |
3409 | * and, as releases are serialized, the one pointed to by the next | |
3410 | * pointer is guaranteed to not have started release yet. This | |
3411 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3412 | * critical section, the one pointed to by its next pointer is | |
3413 | * guaranteed to not have finished its RCU grace period even if we | |
3414 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3415 | * |
de3f0341 TH |
3416 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3417 | * dereferenced; however, as each css is given a monotonically | |
3418 | * increasing unique serial number and always appended to the | |
3419 | * sibling list, the next one can be found by walking the parent's | |
3420 | * children until the first css with higher serial number than | |
3421 | * @pos's. While this path can be slower, it happens iff iteration | |
3422 | * races against release and the race window is very small. | |
53fa5261 | 3423 | */ |
3b287a50 | 3424 | if (!pos) { |
c2931b70 TH |
3425 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3426 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3427 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3428 | } else { |
c2931b70 | 3429 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3430 | if (next->serial_nr > pos->serial_nr) |
3431 | break; | |
53fa5261 TH |
3432 | } |
3433 | ||
3b281afb TH |
3434 | /* |
3435 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3436 | * the next sibling. |
3b281afb | 3437 | */ |
c2931b70 TH |
3438 | if (&next->sibling != &parent->children) |
3439 | return next; | |
3b281afb | 3440 | return NULL; |
53fa5261 | 3441 | } |
53fa5261 | 3442 | |
574bd9f7 | 3443 | /** |
492eb21b | 3444 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3445 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3446 | * @root: css whose descendants to walk |
574bd9f7 | 3447 | * |
492eb21b | 3448 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3449 | * to visit for pre-order traversal of @root's descendants. @root is |
3450 | * included in the iteration and the first node to be visited. | |
75501a6d | 3451 | * |
87fb54f1 TH |
3452 | * While this function requires cgroup_mutex or RCU read locking, it |
3453 | * doesn't require the whole traversal to be contained in a single critical | |
3454 | * section. This function will return the correct next descendant as long | |
3455 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3456 | * |
3457 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3458 | * css which finished ->css_online() is guaranteed to be visible in the | |
3459 | * future iterations and will stay visible until the last reference is put. | |
3460 | * A css which hasn't finished ->css_online() or already finished | |
3461 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3462 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3463 | */ |
492eb21b TH |
3464 | struct cgroup_subsys_state * |
3465 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3466 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3467 | { |
492eb21b | 3468 | struct cgroup_subsys_state *next; |
574bd9f7 | 3469 | |
8353da1f | 3470 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3471 | |
bd8815a6 | 3472 | /* if first iteration, visit @root */ |
7805d000 | 3473 | if (!pos) |
bd8815a6 | 3474 | return root; |
574bd9f7 TH |
3475 | |
3476 | /* visit the first child if exists */ | |
492eb21b | 3477 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3478 | if (next) |
3479 | return next; | |
3480 | ||
3481 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3482 | while (pos != root) { |
5c9d535b | 3483 | next = css_next_child(pos, pos->parent); |
75501a6d | 3484 | if (next) |
574bd9f7 | 3485 | return next; |
5c9d535b | 3486 | pos = pos->parent; |
7805d000 | 3487 | } |
574bd9f7 TH |
3488 | |
3489 | return NULL; | |
3490 | } | |
574bd9f7 | 3491 | |
12a9d2fe | 3492 | /** |
492eb21b TH |
3493 | * css_rightmost_descendant - return the rightmost descendant of a css |
3494 | * @pos: css of interest | |
12a9d2fe | 3495 | * |
492eb21b TH |
3496 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3497 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3498 | * subtree of @pos. |
75501a6d | 3499 | * |
87fb54f1 TH |
3500 | * While this function requires cgroup_mutex or RCU read locking, it |
3501 | * doesn't require the whole traversal to be contained in a single critical | |
3502 | * section. This function will return the correct rightmost descendant as | |
3503 | * long as @pos is accessible. | |
12a9d2fe | 3504 | */ |
492eb21b TH |
3505 | struct cgroup_subsys_state * |
3506 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3507 | { |
492eb21b | 3508 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3509 | |
8353da1f | 3510 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3511 | |
3512 | do { | |
3513 | last = pos; | |
3514 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3515 | pos = NULL; | |
492eb21b | 3516 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3517 | pos = tmp; |
3518 | } while (pos); | |
3519 | ||
3520 | return last; | |
3521 | } | |
12a9d2fe | 3522 | |
492eb21b TH |
3523 | static struct cgroup_subsys_state * |
3524 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3525 | { |
492eb21b | 3526 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3527 | |
3528 | do { | |
3529 | last = pos; | |
492eb21b | 3530 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3531 | } while (pos); |
3532 | ||
3533 | return last; | |
3534 | } | |
3535 | ||
3536 | /** | |
492eb21b | 3537 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3538 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3539 | * @root: css whose descendants to walk |
574bd9f7 | 3540 | * |
492eb21b | 3541 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3542 | * to visit for post-order traversal of @root's descendants. @root is |
3543 | * included in the iteration and the last node to be visited. | |
75501a6d | 3544 | * |
87fb54f1 TH |
3545 | * While this function requires cgroup_mutex or RCU read locking, it |
3546 | * doesn't require the whole traversal to be contained in a single critical | |
3547 | * section. This function will return the correct next descendant as long | |
3548 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3549 | * @cgroup. | |
c2931b70 TH |
3550 | * |
3551 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3552 | * css which finished ->css_online() is guaranteed to be visible in the | |
3553 | * future iterations and will stay visible until the last reference is put. | |
3554 | * A css which hasn't finished ->css_online() or already finished | |
3555 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3556 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3557 | */ |
492eb21b TH |
3558 | struct cgroup_subsys_state * |
3559 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3560 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3561 | { |
492eb21b | 3562 | struct cgroup_subsys_state *next; |
574bd9f7 | 3563 | |
8353da1f | 3564 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3565 | |
58b79a91 TH |
3566 | /* if first iteration, visit leftmost descendant which may be @root */ |
3567 | if (!pos) | |
3568 | return css_leftmost_descendant(root); | |
574bd9f7 | 3569 | |
bd8815a6 TH |
3570 | /* if we visited @root, we're done */ |
3571 | if (pos == root) | |
3572 | return NULL; | |
3573 | ||
574bd9f7 | 3574 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3575 | next = css_next_child(pos, pos->parent); |
75501a6d | 3576 | if (next) |
492eb21b | 3577 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3578 | |
3579 | /* no sibling left, visit parent */ | |
5c9d535b | 3580 | return pos->parent; |
574bd9f7 | 3581 | } |
574bd9f7 | 3582 | |
f3d46500 TH |
3583 | /** |
3584 | * css_has_online_children - does a css have online children | |
3585 | * @css: the target css | |
3586 | * | |
3587 | * Returns %true if @css has any online children; otherwise, %false. This | |
3588 | * function can be called from any context but the caller is responsible | |
3589 | * for synchronizing against on/offlining as necessary. | |
3590 | */ | |
3591 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3592 | { |
f3d46500 TH |
3593 | struct cgroup_subsys_state *child; |
3594 | bool ret = false; | |
cbc125ef TH |
3595 | |
3596 | rcu_read_lock(); | |
f3d46500 | 3597 | css_for_each_child(child, css) { |
99bae5f9 | 3598 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3599 | ret = true; |
3600 | break; | |
cbc125ef TH |
3601 | } |
3602 | } | |
3603 | rcu_read_unlock(); | |
f3d46500 | 3604 | return ret; |
574bd9f7 | 3605 | } |
574bd9f7 | 3606 | |
0942eeee | 3607 | /** |
72ec7029 | 3608 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3609 | * @it: the iterator to advance |
3610 | * | |
3611 | * Advance @it to the next css_set to walk. | |
d515876e | 3612 | */ |
72ec7029 | 3613 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3614 | { |
0f0a2b4f | 3615 | struct list_head *l = it->cset_pos; |
d515876e TH |
3616 | struct cgrp_cset_link *link; |
3617 | struct css_set *cset; | |
3618 | ||
3619 | /* Advance to the next non-empty css_set */ | |
3620 | do { | |
3621 | l = l->next; | |
0f0a2b4f TH |
3622 | if (l == it->cset_head) { |
3623 | it->cset_pos = NULL; | |
d515876e TH |
3624 | return; |
3625 | } | |
3ebb2b6e TH |
3626 | |
3627 | if (it->ss) { | |
3628 | cset = container_of(l, struct css_set, | |
3629 | e_cset_node[it->ss->id]); | |
3630 | } else { | |
3631 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3632 | cset = link->cset; | |
3633 | } | |
c7561128 TH |
3634 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3635 | ||
0f0a2b4f | 3636 | it->cset_pos = l; |
c7561128 TH |
3637 | |
3638 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3639 | it->task_pos = cset->tasks.next; |
c7561128 | 3640 | else |
0f0a2b4f TH |
3641 | it->task_pos = cset->mg_tasks.next; |
3642 | ||
3643 | it->tasks_head = &cset->tasks; | |
3644 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3645 | } |
3646 | ||
0942eeee | 3647 | /** |
72ec7029 TH |
3648 | * css_task_iter_start - initiate task iteration |
3649 | * @css: the css to walk tasks of | |
0942eeee TH |
3650 | * @it: the task iterator to use |
3651 | * | |
72ec7029 TH |
3652 | * Initiate iteration through the tasks of @css. The caller can call |
3653 | * css_task_iter_next() to walk through the tasks until the function | |
3654 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3655 | * called. | |
0942eeee TH |
3656 | * |
3657 | * Note that this function acquires a lock which is released when the | |
3658 | * iteration finishes. The caller can't sleep while iteration is in | |
3659 | * progress. | |
3660 | */ | |
72ec7029 TH |
3661 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3662 | struct css_task_iter *it) | |
96d365e0 | 3663 | __acquires(css_set_rwsem) |
817929ec | 3664 | { |
56fde9e0 TH |
3665 | /* no one should try to iterate before mounting cgroups */ |
3666 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3667 | |
96d365e0 | 3668 | down_read(&css_set_rwsem); |
c59cd3d8 | 3669 | |
3ebb2b6e TH |
3670 | it->ss = css->ss; |
3671 | ||
3672 | if (it->ss) | |
3673 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3674 | else | |
3675 | it->cset_pos = &css->cgroup->cset_links; | |
3676 | ||
0f0a2b4f | 3677 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3678 | |
72ec7029 | 3679 | css_advance_task_iter(it); |
817929ec PM |
3680 | } |
3681 | ||
0942eeee | 3682 | /** |
72ec7029 | 3683 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3684 | * @it: the task iterator being iterated |
3685 | * | |
3686 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3687 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3688 | * reaches the end. | |
0942eeee | 3689 | */ |
72ec7029 | 3690 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3691 | { |
3692 | struct task_struct *res; | |
0f0a2b4f | 3693 | struct list_head *l = it->task_pos; |
817929ec PM |
3694 | |
3695 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3696 | if (!it->cset_pos) |
817929ec PM |
3697 | return NULL; |
3698 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3699 | |
3700 | /* | |
3701 | * Advance iterator to find next entry. cset->tasks is consumed | |
3702 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3703 | * next cset. | |
3704 | */ | |
817929ec | 3705 | l = l->next; |
c7561128 | 3706 | |
0f0a2b4f TH |
3707 | if (l == it->tasks_head) |
3708 | l = it->mg_tasks_head->next; | |
c7561128 | 3709 | |
0f0a2b4f | 3710 | if (l == it->mg_tasks_head) |
72ec7029 | 3711 | css_advance_task_iter(it); |
c7561128 | 3712 | else |
0f0a2b4f | 3713 | it->task_pos = l; |
c7561128 | 3714 | |
817929ec PM |
3715 | return res; |
3716 | } | |
3717 | ||
0942eeee | 3718 | /** |
72ec7029 | 3719 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3720 | * @it: the task iterator to finish |
3721 | * | |
72ec7029 | 3722 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3723 | */ |
72ec7029 | 3724 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3725 | __releases(css_set_rwsem) |
31a7df01 | 3726 | { |
96d365e0 | 3727 | up_read(&css_set_rwsem); |
31a7df01 CW |
3728 | } |
3729 | ||
3730 | /** | |
8cc99345 TH |
3731 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3732 | * @to: cgroup to which the tasks will be moved | |
3733 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3734 | * |
eaf797ab TH |
3735 | * Locking rules between cgroup_post_fork() and the migration path |
3736 | * guarantee that, if a task is forking while being migrated, the new child | |
3737 | * is guaranteed to be either visible in the source cgroup after the | |
3738 | * parent's migration is complete or put into the target cgroup. No task | |
3739 | * can slip out of migration through forking. | |
31a7df01 | 3740 | */ |
8cc99345 | 3741 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3742 | { |
952aaa12 TH |
3743 | LIST_HEAD(preloaded_csets); |
3744 | struct cgrp_cset_link *link; | |
72ec7029 | 3745 | struct css_task_iter it; |
e406d1cf | 3746 | struct task_struct *task; |
952aaa12 | 3747 | int ret; |
31a7df01 | 3748 | |
952aaa12 | 3749 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3750 | |
952aaa12 TH |
3751 | /* all tasks in @from are being moved, all csets are source */ |
3752 | down_read(&css_set_rwsem); | |
3753 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3754 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3755 | up_read(&css_set_rwsem); | |
31a7df01 | 3756 | |
952aaa12 TH |
3757 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3758 | if (ret) | |
3759 | goto out_err; | |
8cc99345 | 3760 | |
952aaa12 TH |
3761 | /* |
3762 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3763 | * ->can_attach() fails. | |
3764 | */ | |
e406d1cf | 3765 | do { |
9d800df1 | 3766 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
3767 | task = css_task_iter_next(&it); |
3768 | if (task) | |
3769 | get_task_struct(task); | |
3770 | css_task_iter_end(&it); | |
3771 | ||
3772 | if (task) { | |
952aaa12 | 3773 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3774 | put_task_struct(task); |
3775 | } | |
3776 | } while (task && !ret); | |
952aaa12 TH |
3777 | out_err: |
3778 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3779 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3780 | return ret; |
8cc99345 TH |
3781 | } |
3782 | ||
bbcb81d0 | 3783 | /* |
102a775e | 3784 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3785 | * |
3786 | * Reading this file can return large amounts of data if a cgroup has | |
3787 | * *lots* of attached tasks. So it may need several calls to read(), | |
3788 | * but we cannot guarantee that the information we produce is correct | |
3789 | * unless we produce it entirely atomically. | |
3790 | * | |
bbcb81d0 | 3791 | */ |
bbcb81d0 | 3792 | |
24528255 LZ |
3793 | /* which pidlist file are we talking about? */ |
3794 | enum cgroup_filetype { | |
3795 | CGROUP_FILE_PROCS, | |
3796 | CGROUP_FILE_TASKS, | |
3797 | }; | |
3798 | ||
3799 | /* | |
3800 | * A pidlist is a list of pids that virtually represents the contents of one | |
3801 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3802 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3803 | * to the cgroup. | |
3804 | */ | |
3805 | struct cgroup_pidlist { | |
3806 | /* | |
3807 | * used to find which pidlist is wanted. doesn't change as long as | |
3808 | * this particular list stays in the list. | |
3809 | */ | |
3810 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3811 | /* array of xids */ | |
3812 | pid_t *list; | |
3813 | /* how many elements the above list has */ | |
3814 | int length; | |
24528255 LZ |
3815 | /* each of these stored in a list by its cgroup */ |
3816 | struct list_head links; | |
3817 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3818 | struct cgroup *owner; | |
b1a21367 TH |
3819 | /* for delayed destruction */ |
3820 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3821 | }; |
3822 | ||
d1d9fd33 BB |
3823 | /* |
3824 | * The following two functions "fix" the issue where there are more pids | |
3825 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3826 | * TODO: replace with a kernel-wide solution to this problem | |
3827 | */ | |
3828 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3829 | static void *pidlist_allocate(int count) | |
3830 | { | |
3831 | if (PIDLIST_TOO_LARGE(count)) | |
3832 | return vmalloc(count * sizeof(pid_t)); | |
3833 | else | |
3834 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3835 | } | |
b1a21367 | 3836 | |
d1d9fd33 BB |
3837 | static void pidlist_free(void *p) |
3838 | { | |
58794514 | 3839 | kvfree(p); |
d1d9fd33 | 3840 | } |
d1d9fd33 | 3841 | |
b1a21367 TH |
3842 | /* |
3843 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3844 | * should be left afterwards. | |
3845 | */ | |
3846 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3847 | { | |
3848 | struct cgroup_pidlist *l, *tmp_l; | |
3849 | ||
3850 | mutex_lock(&cgrp->pidlist_mutex); | |
3851 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3852 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3853 | mutex_unlock(&cgrp->pidlist_mutex); | |
3854 | ||
3855 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3856 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3857 | } | |
3858 | ||
3859 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3860 | { | |
3861 | struct delayed_work *dwork = to_delayed_work(work); | |
3862 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3863 | destroy_dwork); | |
3864 | struct cgroup_pidlist *tofree = NULL; | |
3865 | ||
3866 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3867 | |
3868 | /* | |
04502365 TH |
3869 | * Destroy iff we didn't get queued again. The state won't change |
3870 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3871 | */ |
04502365 | 3872 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3873 | list_del(&l->links); |
3874 | pidlist_free(l->list); | |
3875 | put_pid_ns(l->key.ns); | |
3876 | tofree = l; | |
3877 | } | |
3878 | ||
b1a21367 TH |
3879 | mutex_unlock(&l->owner->pidlist_mutex); |
3880 | kfree(tofree); | |
3881 | } | |
3882 | ||
bbcb81d0 | 3883 | /* |
102a775e | 3884 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3885 | * Returns the number of unique elements. |
bbcb81d0 | 3886 | */ |
6ee211ad | 3887 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3888 | { |
102a775e | 3889 | int src, dest = 1; |
102a775e BB |
3890 | |
3891 | /* | |
3892 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3893 | * edge cases first; no work needs to be done for either | |
3894 | */ | |
3895 | if (length == 0 || length == 1) | |
3896 | return length; | |
3897 | /* src and dest walk down the list; dest counts unique elements */ | |
3898 | for (src = 1; src < length; src++) { | |
3899 | /* find next unique element */ | |
3900 | while (list[src] == list[src-1]) { | |
3901 | src++; | |
3902 | if (src == length) | |
3903 | goto after; | |
3904 | } | |
3905 | /* dest always points to where the next unique element goes */ | |
3906 | list[dest] = list[src]; | |
3907 | dest++; | |
3908 | } | |
3909 | after: | |
102a775e BB |
3910 | return dest; |
3911 | } | |
3912 | ||
afb2bc14 TH |
3913 | /* |
3914 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3915 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3916 | * different per namespace, each namespace needs differently sorted list, | |
3917 | * making it impossible to use, for example, single rbtree of member tasks | |
3918 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3919 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3920 | * pidlists keyed by cgroup and namespace. | |
3921 | * | |
3922 | * All this extra complexity was caused by the original implementation | |
3923 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
3924 | * want to do away with it. Explicitly scramble sort order if on the |
3925 | * default hierarchy so that no such expectation exists in the new | |
3926 | * interface. | |
afb2bc14 TH |
3927 | * |
3928 | * Scrambling is done by swapping every two consecutive bits, which is | |
3929 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3930 | */ | |
3931 | static pid_t pid_fry(pid_t pid) | |
3932 | { | |
3933 | unsigned a = pid & 0x55555555; | |
3934 | unsigned b = pid & 0xAAAAAAAA; | |
3935 | ||
3936 | return (a << 1) | (b >> 1); | |
3937 | } | |
3938 | ||
3939 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3940 | { | |
aa6ec29b | 3941 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
3942 | return pid_fry(pid); |
3943 | else | |
3944 | return pid; | |
3945 | } | |
3946 | ||
102a775e BB |
3947 | static int cmppid(const void *a, const void *b) |
3948 | { | |
3949 | return *(pid_t *)a - *(pid_t *)b; | |
3950 | } | |
3951 | ||
afb2bc14 TH |
3952 | static int fried_cmppid(const void *a, const void *b) |
3953 | { | |
3954 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3955 | } | |
3956 | ||
e6b81710 TH |
3957 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3958 | enum cgroup_filetype type) | |
3959 | { | |
3960 | struct cgroup_pidlist *l; | |
3961 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3962 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3963 | ||
3964 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3965 | ||
3966 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3967 | if (l->key.type == type && l->key.ns == ns) | |
3968 | return l; | |
3969 | return NULL; | |
3970 | } | |
3971 | ||
72a8cb30 BB |
3972 | /* |
3973 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3974 | * returns with the lock on that pidlist already held, and takes care | |
3975 | * of the use count, or returns NULL with no locks held if we're out of | |
3976 | * memory. | |
3977 | */ | |
e6b81710 TH |
3978 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3979 | enum cgroup_filetype type) | |
72a8cb30 BB |
3980 | { |
3981 | struct cgroup_pidlist *l; | |
b70cc5fd | 3982 | |
e6b81710 TH |
3983 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3984 | ||
3985 | l = cgroup_pidlist_find(cgrp, type); | |
3986 | if (l) | |
3987 | return l; | |
3988 | ||
72a8cb30 | 3989 | /* entry not found; create a new one */ |
f4f4be2b | 3990 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3991 | if (!l) |
72a8cb30 | 3992 | return l; |
e6b81710 | 3993 | |
b1a21367 | 3994 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3995 | l->key.type = type; |
e6b81710 TH |
3996 | /* don't need task_nsproxy() if we're looking at ourself */ |
3997 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3998 | l->owner = cgrp; |
3999 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4000 | return l; |
4001 | } | |
4002 | ||
102a775e BB |
4003 | /* |
4004 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4005 | */ | |
72a8cb30 BB |
4006 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4007 | struct cgroup_pidlist **lp) | |
102a775e BB |
4008 | { |
4009 | pid_t *array; | |
4010 | int length; | |
4011 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4012 | struct css_task_iter it; |
817929ec | 4013 | struct task_struct *tsk; |
102a775e BB |
4014 | struct cgroup_pidlist *l; |
4015 | ||
4bac00d1 TH |
4016 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4017 | ||
102a775e BB |
4018 | /* |
4019 | * If cgroup gets more users after we read count, we won't have | |
4020 | * enough space - tough. This race is indistinguishable to the | |
4021 | * caller from the case that the additional cgroup users didn't | |
4022 | * show up until sometime later on. | |
4023 | */ | |
4024 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4025 | array = pidlist_allocate(length); |
102a775e BB |
4026 | if (!array) |
4027 | return -ENOMEM; | |
4028 | /* now, populate the array */ | |
9d800df1 | 4029 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4030 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4031 | if (unlikely(n == length)) |
817929ec | 4032 | break; |
102a775e | 4033 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4034 | if (type == CGROUP_FILE_PROCS) |
4035 | pid = task_tgid_vnr(tsk); | |
4036 | else | |
4037 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4038 | if (pid > 0) /* make sure to only use valid results */ |
4039 | array[n++] = pid; | |
817929ec | 4040 | } |
72ec7029 | 4041 | css_task_iter_end(&it); |
102a775e BB |
4042 | length = n; |
4043 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4044 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4045 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4046 | else | |
4047 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4048 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4049 | length = pidlist_uniq(array, length); |
e6b81710 | 4050 | |
e6b81710 | 4051 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4052 | if (!l) { |
d1d9fd33 | 4053 | pidlist_free(array); |
72a8cb30 | 4054 | return -ENOMEM; |
102a775e | 4055 | } |
e6b81710 TH |
4056 | |
4057 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4058 | pidlist_free(l->list); |
102a775e BB |
4059 | l->list = array; |
4060 | l->length = length; | |
72a8cb30 | 4061 | *lp = l; |
102a775e | 4062 | return 0; |
bbcb81d0 PM |
4063 | } |
4064 | ||
846c7bb0 | 4065 | /** |
a043e3b2 | 4066 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4067 | * @stats: cgroupstats to fill information into |
4068 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4069 | * been requested. | |
a043e3b2 LZ |
4070 | * |
4071 | * Build and fill cgroupstats so that taskstats can export it to user | |
4072 | * space. | |
846c7bb0 BS |
4073 | */ |
4074 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4075 | { | |
2bd59d48 | 4076 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4077 | struct cgroup *cgrp; |
72ec7029 | 4078 | struct css_task_iter it; |
846c7bb0 | 4079 | struct task_struct *tsk; |
33d283be | 4080 | |
2bd59d48 TH |
4081 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4082 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4083 | kernfs_type(kn) != KERNFS_DIR) | |
4084 | return -EINVAL; | |
4085 | ||
bad34660 LZ |
4086 | mutex_lock(&cgroup_mutex); |
4087 | ||
846c7bb0 | 4088 | /* |
2bd59d48 | 4089 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4090 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4091 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4092 | */ |
2bd59d48 TH |
4093 | rcu_read_lock(); |
4094 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4095 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4096 | rcu_read_unlock(); |
bad34660 | 4097 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4098 | return -ENOENT; |
4099 | } | |
bad34660 | 4100 | rcu_read_unlock(); |
846c7bb0 | 4101 | |
9d800df1 | 4102 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4103 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4104 | switch (tsk->state) { |
4105 | case TASK_RUNNING: | |
4106 | stats->nr_running++; | |
4107 | break; | |
4108 | case TASK_INTERRUPTIBLE: | |
4109 | stats->nr_sleeping++; | |
4110 | break; | |
4111 | case TASK_UNINTERRUPTIBLE: | |
4112 | stats->nr_uninterruptible++; | |
4113 | break; | |
4114 | case TASK_STOPPED: | |
4115 | stats->nr_stopped++; | |
4116 | break; | |
4117 | default: | |
4118 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4119 | stats->nr_io_wait++; | |
4120 | break; | |
4121 | } | |
4122 | } | |
72ec7029 | 4123 | css_task_iter_end(&it); |
846c7bb0 | 4124 | |
bad34660 | 4125 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4126 | return 0; |
846c7bb0 BS |
4127 | } |
4128 | ||
8f3ff208 | 4129 | |
bbcb81d0 | 4130 | /* |
102a775e | 4131 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4132 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4133 | * in the cgroup->l->list array. |
bbcb81d0 | 4134 | */ |
cc31edce | 4135 | |
102a775e | 4136 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4137 | { |
cc31edce PM |
4138 | /* |
4139 | * Initially we receive a position value that corresponds to | |
4140 | * one more than the last pid shown (or 0 on the first call or | |
4141 | * after a seek to the start). Use a binary-search to find the | |
4142 | * next pid to display, if any | |
4143 | */ | |
2bd59d48 | 4144 | struct kernfs_open_file *of = s->private; |
7da11279 | 4145 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4146 | struct cgroup_pidlist *l; |
7da11279 | 4147 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4148 | int index = 0, pid = *pos; |
4bac00d1 TH |
4149 | int *iter, ret; |
4150 | ||
4151 | mutex_lock(&cgrp->pidlist_mutex); | |
4152 | ||
4153 | /* | |
5d22444f | 4154 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4155 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4156 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4157 | * could already have been destroyed. |
4158 | */ | |
5d22444f TH |
4159 | if (of->priv) |
4160 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4161 | |
4162 | /* | |
4163 | * Either this is the first start() after open or the matching | |
4164 | * pidlist has been destroyed inbetween. Create a new one. | |
4165 | */ | |
5d22444f TH |
4166 | if (!of->priv) { |
4167 | ret = pidlist_array_load(cgrp, type, | |
4168 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4169 | if (ret) |
4170 | return ERR_PTR(ret); | |
4171 | } | |
5d22444f | 4172 | l = of->priv; |
cc31edce | 4173 | |
cc31edce | 4174 | if (pid) { |
102a775e | 4175 | int end = l->length; |
20777766 | 4176 | |
cc31edce PM |
4177 | while (index < end) { |
4178 | int mid = (index + end) / 2; | |
afb2bc14 | 4179 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4180 | index = mid; |
4181 | break; | |
afb2bc14 | 4182 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4183 | index = mid + 1; |
4184 | else | |
4185 | end = mid; | |
4186 | } | |
4187 | } | |
4188 | /* If we're off the end of the array, we're done */ | |
102a775e | 4189 | if (index >= l->length) |
cc31edce PM |
4190 | return NULL; |
4191 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4192 | iter = l->list + index; |
afb2bc14 | 4193 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4194 | return iter; |
4195 | } | |
4196 | ||
102a775e | 4197 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4198 | { |
2bd59d48 | 4199 | struct kernfs_open_file *of = s->private; |
5d22444f | 4200 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4201 | |
5d22444f TH |
4202 | if (l) |
4203 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4204 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4205 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4206 | } |
4207 | ||
102a775e | 4208 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4209 | { |
2bd59d48 | 4210 | struct kernfs_open_file *of = s->private; |
5d22444f | 4211 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4212 | pid_t *p = v; |
4213 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4214 | /* |
4215 | * Advance to the next pid in the array. If this goes off the | |
4216 | * end, we're done | |
4217 | */ | |
4218 | p++; | |
4219 | if (p >= end) { | |
4220 | return NULL; | |
4221 | } else { | |
7da11279 | 4222 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4223 | return p; |
4224 | } | |
4225 | } | |
4226 | ||
102a775e | 4227 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4228 | { |
94ff212d JP |
4229 | seq_printf(s, "%d\n", *(int *)v); |
4230 | ||
4231 | return 0; | |
cc31edce | 4232 | } |
bbcb81d0 | 4233 | |
182446d0 TH |
4234 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4235 | struct cftype *cft) | |
81a6a5cd | 4236 | { |
182446d0 | 4237 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4238 | } |
4239 | ||
182446d0 TH |
4240 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4241 | struct cftype *cft, u64 val) | |
6379c106 | 4242 | { |
6379c106 | 4243 | if (val) |
182446d0 | 4244 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4245 | else |
182446d0 | 4246 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4247 | return 0; |
4248 | } | |
4249 | ||
182446d0 TH |
4250 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4251 | struct cftype *cft) | |
97978e6d | 4252 | { |
182446d0 | 4253 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4254 | } |
4255 | ||
182446d0 TH |
4256 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4257 | struct cftype *cft, u64 val) | |
97978e6d DL |
4258 | { |
4259 | if (val) | |
182446d0 | 4260 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4261 | else |
182446d0 | 4262 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4263 | return 0; |
4264 | } | |
4265 | ||
a14c6874 TH |
4266 | /* cgroup core interface files for the default hierarchy */ |
4267 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4268 | { |
d5c56ced | 4269 | .name = "cgroup.procs", |
6612f05b TH |
4270 | .seq_start = cgroup_pidlist_start, |
4271 | .seq_next = cgroup_pidlist_next, | |
4272 | .seq_stop = cgroup_pidlist_stop, | |
4273 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4274 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4275 | .write = cgroup_procs_write, |
74a1166d | 4276 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 4277 | }, |
f8f22e53 TH |
4278 | { |
4279 | .name = "cgroup.controllers", | |
a14c6874 | 4280 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4281 | .seq_show = cgroup_root_controllers_show, |
4282 | }, | |
4283 | { | |
4284 | .name = "cgroup.controllers", | |
a14c6874 | 4285 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4286 | .seq_show = cgroup_controllers_show, |
4287 | }, | |
4288 | { | |
4289 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4290 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4291 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4292 | }, |
842b597e TH |
4293 | { |
4294 | .name = "cgroup.populated", | |
a14c6874 | 4295 | .flags = CFTYPE_NOT_ON_ROOT, |
842b597e TH |
4296 | .seq_show = cgroup_populated_show, |
4297 | }, | |
a14c6874 TH |
4298 | { } /* terminate */ |
4299 | }; | |
d5c56ced | 4300 | |
a14c6874 TH |
4301 | /* cgroup core interface files for the legacy hierarchies */ |
4302 | static struct cftype cgroup_legacy_base_files[] = { | |
4303 | { | |
4304 | .name = "cgroup.procs", | |
4305 | .seq_start = cgroup_pidlist_start, | |
4306 | .seq_next = cgroup_pidlist_next, | |
4307 | .seq_stop = cgroup_pidlist_stop, | |
4308 | .seq_show = cgroup_pidlist_show, | |
4309 | .private = CGROUP_FILE_PROCS, | |
4310 | .write = cgroup_procs_write, | |
4311 | .mode = S_IRUGO | S_IWUSR, | |
4312 | }, | |
4313 | { | |
4314 | .name = "cgroup.clone_children", | |
4315 | .read_u64 = cgroup_clone_children_read, | |
4316 | .write_u64 = cgroup_clone_children_write, | |
4317 | }, | |
4318 | { | |
4319 | .name = "cgroup.sane_behavior", | |
4320 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4321 | .seq_show = cgroup_sane_behavior_show, | |
4322 | }, | |
d5c56ced TH |
4323 | { |
4324 | .name = "tasks", | |
6612f05b TH |
4325 | .seq_start = cgroup_pidlist_start, |
4326 | .seq_next = cgroup_pidlist_next, | |
4327 | .seq_stop = cgroup_pidlist_stop, | |
4328 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4329 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4330 | .write = cgroup_tasks_write, |
d5c56ced TH |
4331 | .mode = S_IRUGO | S_IWUSR, |
4332 | }, | |
4333 | { | |
4334 | .name = "notify_on_release", | |
d5c56ced TH |
4335 | .read_u64 = cgroup_read_notify_on_release, |
4336 | .write_u64 = cgroup_write_notify_on_release, | |
4337 | }, | |
6e6ff25b TH |
4338 | { |
4339 | .name = "release_agent", | |
a14c6874 | 4340 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4341 | .seq_show = cgroup_release_agent_show, |
451af504 | 4342 | .write = cgroup_release_agent_write, |
5f469907 | 4343 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4344 | }, |
db0416b6 | 4345 | { } /* terminate */ |
bbcb81d0 PM |
4346 | }; |
4347 | ||
13af07df | 4348 | /** |
628f7cd4 | 4349 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4350 | * @cgrp: target cgroup |
13af07df | 4351 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4352 | * |
4353 | * On failure, no file is added. | |
13af07df | 4354 | */ |
8ab456ac | 4355 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 4356 | { |
ddbcc7e8 | 4357 | struct cgroup_subsys *ss; |
b420ba7d | 4358 | int i, ret = 0; |
bbcb81d0 | 4359 | |
8e3f6541 | 4360 | /* process cftsets of each subsystem */ |
b420ba7d | 4361 | for_each_subsys(ss, i) { |
0adb0704 | 4362 | struct cftype *cfts; |
b420ba7d | 4363 | |
69dfa00c | 4364 | if (!(subsys_mask & (1 << i))) |
13af07df | 4365 | continue; |
8e3f6541 | 4366 | |
0adb0704 TH |
4367 | list_for_each_entry(cfts, &ss->cfts, node) { |
4368 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4369 | if (ret < 0) |
4370 | goto err; | |
4371 | } | |
ddbcc7e8 | 4372 | } |
ddbcc7e8 | 4373 | return 0; |
bee55099 TH |
4374 | err: |
4375 | cgroup_clear_dir(cgrp, subsys_mask); | |
4376 | return ret; | |
ddbcc7e8 PM |
4377 | } |
4378 | ||
0c21ead1 TH |
4379 | /* |
4380 | * css destruction is four-stage process. | |
4381 | * | |
4382 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4383 | * Implemented in kill_css(). | |
4384 | * | |
4385 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4386 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4387 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4388 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4389 | * |
4390 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4391 | * accessors are inside RCU read sections. css_release() schedules the | |
4392 | * RCU callback. | |
4393 | * | |
4394 | * 4. After the grace period, the css can be freed. Implemented in | |
4395 | * css_free_work_fn(). | |
4396 | * | |
4397 | * It is actually hairier because both step 2 and 4 require process context | |
4398 | * and thus involve punting to css->destroy_work adding two additional | |
4399 | * steps to the already complex sequence. | |
4400 | */ | |
35ef10da | 4401 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4402 | { |
4403 | struct cgroup_subsys_state *css = | |
35ef10da | 4404 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4405 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4406 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4407 | |
9a1049da TH |
4408 | percpu_ref_exit(&css->refcnt); |
4409 | ||
01e58659 | 4410 | if (ss) { |
9d755d33 | 4411 | /* css free path */ |
01e58659 VD |
4412 | int id = css->id; |
4413 | ||
9d755d33 TH |
4414 | if (css->parent) |
4415 | css_put(css->parent); | |
0ae78e0b | 4416 | |
01e58659 VD |
4417 | ss->css_free(css); |
4418 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 TH |
4419 | cgroup_put(cgrp); |
4420 | } else { | |
4421 | /* cgroup free path */ | |
4422 | atomic_dec(&cgrp->root->nr_cgrps); | |
4423 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4424 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4425 | |
d51f39b0 | 4426 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4427 | /* |
4428 | * We get a ref to the parent, and put the ref when | |
4429 | * this cgroup is being freed, so it's guaranteed | |
4430 | * that the parent won't be destroyed before its | |
4431 | * children. | |
4432 | */ | |
d51f39b0 | 4433 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4434 | kernfs_put(cgrp->kn); |
4435 | kfree(cgrp); | |
4436 | } else { | |
4437 | /* | |
4438 | * This is root cgroup's refcnt reaching zero, | |
4439 | * which indicates that the root should be | |
4440 | * released. | |
4441 | */ | |
4442 | cgroup_destroy_root(cgrp->root); | |
4443 | } | |
4444 | } | |
48ddbe19 TH |
4445 | } |
4446 | ||
0c21ead1 | 4447 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4448 | { |
4449 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4450 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4451 | |
35ef10da | 4452 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4453 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4454 | } |
4455 | ||
25e15d83 | 4456 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4457 | { |
4458 | struct cgroup_subsys_state *css = | |
25e15d83 | 4459 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4460 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4461 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4462 | |
1fed1b2e TH |
4463 | mutex_lock(&cgroup_mutex); |
4464 | ||
de3f0341 | 4465 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4466 | list_del_rcu(&css->sibling); |
4467 | ||
9d755d33 TH |
4468 | if (ss) { |
4469 | /* css release path */ | |
01e58659 | 4470 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4471 | if (ss->css_released) |
4472 | ss->css_released(css); | |
9d755d33 TH |
4473 | } else { |
4474 | /* cgroup release path */ | |
9d755d33 TH |
4475 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4476 | cgrp->id = -1; | |
a4189487 LZ |
4477 | |
4478 | /* | |
4479 | * There are two control paths which try to determine | |
4480 | * cgroup from dentry without going through kernfs - | |
4481 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4482 | * Those are supported by RCU protecting clearing of | |
4483 | * cgrp->kn->priv backpointer. | |
4484 | */ | |
4485 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
9d755d33 | 4486 | } |
d3daf28d | 4487 | |
1fed1b2e TH |
4488 | mutex_unlock(&cgroup_mutex); |
4489 | ||
0c21ead1 | 4490 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4491 | } |
4492 | ||
d3daf28d TH |
4493 | static void css_release(struct percpu_ref *ref) |
4494 | { | |
4495 | struct cgroup_subsys_state *css = | |
4496 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4497 | ||
25e15d83 TH |
4498 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4499 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4500 | } |
4501 | ||
ddfcadab TH |
4502 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4503 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4504 | { |
0cb51d71 TH |
4505 | lockdep_assert_held(&cgroup_mutex); |
4506 | ||
ddfcadab TH |
4507 | cgroup_get(cgrp); |
4508 | ||
d5c419b6 | 4509 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4510 | css->cgroup = cgrp; |
72c97e54 | 4511 | css->ss = ss; |
d5c419b6 TH |
4512 | INIT_LIST_HEAD(&css->sibling); |
4513 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4514 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4515 | |
d51f39b0 TH |
4516 | if (cgroup_parent(cgrp)) { |
4517 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4518 | css_get(css->parent); |
ddfcadab | 4519 | } |
48ddbe19 | 4520 | |
ca8bdcaf | 4521 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4522 | } |
4523 | ||
2a4ac633 | 4524 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4525 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4526 | { |
623f926b | 4527 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4528 | int ret = 0; |
4529 | ||
a31f2d3f TH |
4530 | lockdep_assert_held(&cgroup_mutex); |
4531 | ||
92fb9748 | 4532 | if (ss->css_online) |
eb95419b | 4533 | ret = ss->css_online(css); |
ae7f164a | 4534 | if (!ret) { |
eb95419b | 4535 | css->flags |= CSS_ONLINE; |
aec25020 | 4536 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4537 | } |
b1929db4 | 4538 | return ret; |
a31f2d3f TH |
4539 | } |
4540 | ||
2a4ac633 | 4541 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4542 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4543 | { |
623f926b | 4544 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4545 | |
4546 | lockdep_assert_held(&cgroup_mutex); | |
4547 | ||
4548 | if (!(css->flags & CSS_ONLINE)) | |
4549 | return; | |
4550 | ||
d7eeac19 | 4551 | if (ss->css_offline) |
eb95419b | 4552 | ss->css_offline(css); |
a31f2d3f | 4553 | |
eb95419b | 4554 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4555 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4556 | |
4557 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4558 | } |
4559 | ||
c81c925a TH |
4560 | /** |
4561 | * create_css - create a cgroup_subsys_state | |
4562 | * @cgrp: the cgroup new css will be associated with | |
4563 | * @ss: the subsys of new css | |
f63070d3 | 4564 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4565 | * |
4566 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4567 | * css is online and installed in @cgrp with all interface files created if |
4568 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4569 | */ |
f63070d3 TH |
4570 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4571 | bool visible) | |
c81c925a | 4572 | { |
d51f39b0 | 4573 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4574 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4575 | struct cgroup_subsys_state *css; |
4576 | int err; | |
4577 | ||
c81c925a TH |
4578 | lockdep_assert_held(&cgroup_mutex); |
4579 | ||
1fed1b2e | 4580 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4581 | if (IS_ERR(css)) |
4582 | return PTR_ERR(css); | |
4583 | ||
ddfcadab | 4584 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4585 | |
2aad2a86 | 4586 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4587 | if (err) |
3eb59ec6 | 4588 | goto err_free_css; |
c81c925a | 4589 | |
cf780b7d | 4590 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4591 | if (err < 0) |
4592 | goto err_free_percpu_ref; | |
4593 | css->id = err; | |
c81c925a | 4594 | |
f63070d3 TH |
4595 | if (visible) { |
4596 | err = cgroup_populate_dir(cgrp, 1 << ss->id); | |
4597 | if (err) | |
4598 | goto err_free_id; | |
4599 | } | |
15a4c835 TH |
4600 | |
4601 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4602 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4603 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4604 | |
4605 | err = online_css(css); | |
4606 | if (err) | |
1fed1b2e | 4607 | goto err_list_del; |
94419627 | 4608 | |
c81c925a | 4609 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4610 | cgroup_parent(parent)) { |
ed3d261b | 4611 | 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 | 4612 | current->comm, current->pid, ss->name); |
c81c925a | 4613 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4614 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4615 | ss->warned_broken_hierarchy = true; |
4616 | } | |
4617 | ||
4618 | return 0; | |
4619 | ||
1fed1b2e TH |
4620 | err_list_del: |
4621 | list_del_rcu(&css->sibling); | |
32d01dc7 | 4622 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
15a4c835 TH |
4623 | err_free_id: |
4624 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4625 | err_free_percpu_ref: |
9a1049da | 4626 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4627 | err_free_css: |
a2bed820 | 4628 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4629 | return err; |
4630 | } | |
4631 | ||
b3bfd983 TH |
4632 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4633 | umode_t mode) | |
ddbcc7e8 | 4634 | { |
a9746d8d TH |
4635 | struct cgroup *parent, *cgrp; |
4636 | struct cgroup_root *root; | |
ddbcc7e8 | 4637 | struct cgroup_subsys *ss; |
2bd59d48 | 4638 | struct kernfs_node *kn; |
a14c6874 | 4639 | struct cftype *base_files; |
b3bfd983 | 4640 | int ssid, ret; |
ddbcc7e8 | 4641 | |
71b1fb5c AC |
4642 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4643 | */ | |
4644 | if (strchr(name, '\n')) | |
4645 | return -EINVAL; | |
4646 | ||
a9746d8d TH |
4647 | parent = cgroup_kn_lock_live(parent_kn); |
4648 | if (!parent) | |
4649 | return -ENODEV; | |
4650 | root = parent->root; | |
ddbcc7e8 | 4651 | |
0a950f65 | 4652 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc | 4653 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
ba0f4d76 TH |
4654 | if (!cgrp) { |
4655 | ret = -ENOMEM; | |
4656 | goto out_unlock; | |
0ab02ca8 LZ |
4657 | } |
4658 | ||
2aad2a86 | 4659 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4660 | if (ret) |
4661 | goto out_free_cgrp; | |
4662 | ||
0ab02ca8 LZ |
4663 | /* |
4664 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4665 | * a half-baked cgroup. | |
4666 | */ | |
cf780b7d | 4667 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4668 | if (cgrp->id < 0) { |
ba0f4d76 | 4669 | ret = -ENOMEM; |
9d755d33 | 4670 | goto out_cancel_ref; |
976c06bc TH |
4671 | } |
4672 | ||
cc31edce | 4673 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4674 | |
9d800df1 | 4675 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4676 | cgrp->root = root; |
ddbcc7e8 | 4677 | |
b6abdb0e LZ |
4678 | if (notify_on_release(parent)) |
4679 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4680 | ||
2260e7fc TH |
4681 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4682 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4683 | |
2bd59d48 | 4684 | /* create the directory */ |
e61734c5 | 4685 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4686 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4687 | ret = PTR_ERR(kn); |
4688 | goto out_free_id; | |
2bd59d48 TH |
4689 | } |
4690 | cgrp->kn = kn; | |
ddbcc7e8 | 4691 | |
4e139afc | 4692 | /* |
6f30558f TH |
4693 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4694 | * that @cgrp->kn is always accessible. | |
4e139afc | 4695 | */ |
6f30558f | 4696 | kernfs_get(kn); |
ddbcc7e8 | 4697 | |
0cb51d71 | 4698 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4699 | |
4e139afc | 4700 | /* allocation complete, commit to creation */ |
d5c419b6 | 4701 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4702 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4703 | cgroup_get(parent); |
415cf07a | 4704 | |
0d80255e TH |
4705 | /* |
4706 | * @cgrp is now fully operational. If something fails after this | |
4707 | * point, it'll be released via the normal destruction path. | |
4708 | */ | |
6fa4918d | 4709 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4710 | |
ba0f4d76 TH |
4711 | ret = cgroup_kn_set_ugid(kn); |
4712 | if (ret) | |
4713 | goto out_destroy; | |
49957f8e | 4714 | |
a14c6874 TH |
4715 | if (cgroup_on_dfl(cgrp)) |
4716 | base_files = cgroup_dfl_base_files; | |
4717 | else | |
4718 | base_files = cgroup_legacy_base_files; | |
4719 | ||
4720 | ret = cgroup_addrm_files(cgrp, base_files, true); | |
ba0f4d76 TH |
4721 | if (ret) |
4722 | goto out_destroy; | |
628f7cd4 | 4723 | |
9d403e99 | 4724 | /* let's create and online css's */ |
b85d2040 | 4725 | for_each_subsys(ss, ssid) { |
f392e51c | 4726 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
4727 | ret = create_css(cgrp, ss, |
4728 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
4729 | if (ret) |
4730 | goto out_destroy; | |
b85d2040 | 4731 | } |
a8638030 | 4732 | } |
ddbcc7e8 | 4733 | |
bd53d617 TH |
4734 | /* |
4735 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4736 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4737 | */ |
667c2491 TH |
4738 | if (!cgroup_on_dfl(cgrp)) { |
4739 | cgrp->subtree_control = parent->subtree_control; | |
4740 | cgroup_refresh_child_subsys_mask(cgrp); | |
4741 | } | |
2bd59d48 | 4742 | |
2bd59d48 | 4743 | kernfs_activate(kn); |
ddbcc7e8 | 4744 | |
ba0f4d76 TH |
4745 | ret = 0; |
4746 | goto out_unlock; | |
ddbcc7e8 | 4747 | |
ba0f4d76 | 4748 | out_free_id: |
6fa4918d | 4749 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 4750 | out_cancel_ref: |
9a1049da | 4751 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 4752 | out_free_cgrp: |
bd89aabc | 4753 | kfree(cgrp); |
ba0f4d76 | 4754 | out_unlock: |
a9746d8d | 4755 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4756 | return ret; |
4b8b47eb | 4757 | |
ba0f4d76 | 4758 | out_destroy: |
4b8b47eb | 4759 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4760 | goto out_unlock; |
ddbcc7e8 PM |
4761 | } |
4762 | ||
223dbc38 TH |
4763 | /* |
4764 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4765 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4766 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4767 | */ |
4768 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4769 | { |
223dbc38 TH |
4770 | struct cgroup_subsys_state *css = |
4771 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4772 | |
f20104de | 4773 | mutex_lock(&cgroup_mutex); |
09a503ea | 4774 | offline_css(css); |
f20104de | 4775 | mutex_unlock(&cgroup_mutex); |
09a503ea | 4776 | |
09a503ea | 4777 | css_put(css); |
d3daf28d TH |
4778 | } |
4779 | ||
223dbc38 TH |
4780 | /* css kill confirmation processing requires process context, bounce */ |
4781 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4782 | { |
4783 | struct cgroup_subsys_state *css = | |
4784 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4785 | ||
223dbc38 | 4786 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4787 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4788 | } |
4789 | ||
f392e51c TH |
4790 | /** |
4791 | * kill_css - destroy a css | |
4792 | * @css: css to destroy | |
4793 | * | |
4794 | * This function initiates destruction of @css by removing cgroup interface | |
4795 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4796 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4797 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4798 | */ |
4799 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4800 | { |
01f6474c | 4801 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4802 | |
2bd59d48 TH |
4803 | /* |
4804 | * This must happen before css is disassociated with its cgroup. | |
4805 | * See seq_css() for details. | |
4806 | */ | |
aec25020 | 4807 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4808 | |
edae0c33 TH |
4809 | /* |
4810 | * Killing would put the base ref, but we need to keep it alive | |
4811 | * until after ->css_offline(). | |
4812 | */ | |
4813 | css_get(css); | |
4814 | ||
4815 | /* | |
4816 | * cgroup core guarantees that, by the time ->css_offline() is | |
4817 | * invoked, no new css reference will be given out via | |
ec903c0c | 4818 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4819 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4820 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4821 | * | |
4822 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4823 | * css is confirmed to be seen as killed on all CPUs. | |
4824 | */ | |
4825 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4826 | } |
4827 | ||
4828 | /** | |
4829 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4830 | * @cgrp: cgroup to be destroyed | |
4831 | * | |
4832 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4833 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4834 | * guarantee that css_tryget_online() won't succeed by the time |
4835 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4836 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4837 | * |
4838 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4839 | * userland visible parts and start killing the percpu refcnts of | |
4840 | * css's. Set up so that the next stage will be kicked off once all | |
4841 | * the percpu refcnts are confirmed to be killed. | |
4842 | * | |
4843 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4844 | * rest of destruction. Once all cgroup references are gone, the | |
4845 | * cgroup is RCU-freed. | |
4846 | * | |
4847 | * This function implements s1. After this step, @cgrp is gone as far as | |
4848 | * the userland is concerned and a new cgroup with the same name may be | |
4849 | * created. As cgroup doesn't care about the names internally, this | |
4850 | * doesn't cause any problem. | |
4851 | */ | |
42809dd4 TH |
4852 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4853 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4854 | { |
2bd59d48 | 4855 | struct cgroup_subsys_state *css; |
ddd69148 | 4856 | bool empty; |
1c6727af | 4857 | int ssid; |
ddbcc7e8 | 4858 | |
42809dd4 TH |
4859 | lockdep_assert_held(&cgroup_mutex); |
4860 | ||
ddd69148 | 4861 | /* |
96d365e0 | 4862 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4863 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4864 | */ |
96d365e0 | 4865 | down_read(&css_set_rwsem); |
bb78a92f | 4866 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4867 | up_read(&css_set_rwsem); |
ddd69148 | 4868 | if (!empty) |
ddbcc7e8 | 4869 | return -EBUSY; |
a043e3b2 | 4870 | |
bb78a92f | 4871 | /* |
d5c419b6 TH |
4872 | * Make sure there's no live children. We can't test emptiness of |
4873 | * ->self.children as dead children linger on it while being | |
4874 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 4875 | */ |
f3d46500 | 4876 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
4877 | return -EBUSY; |
4878 | ||
455050d2 TH |
4879 | /* |
4880 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 4881 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 4882 | */ |
184faf32 | 4883 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 4884 | |
249f3468 | 4885 | /* initiate massacre of all css's */ |
1c6727af TH |
4886 | for_each_css(css, ssid, cgrp) |
4887 | kill_css(css); | |
455050d2 | 4888 | |
455050d2 | 4889 | /* |
01f6474c TH |
4890 | * Remove @cgrp directory along with the base files. @cgrp has an |
4891 | * extra ref on its kn. | |
f20104de | 4892 | */ |
01f6474c | 4893 | kernfs_remove(cgrp->kn); |
f20104de | 4894 | |
d51f39b0 | 4895 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 4896 | |
249f3468 | 4897 | /* put the base reference */ |
9d755d33 | 4898 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 4899 | |
ea15f8cc TH |
4900 | return 0; |
4901 | }; | |
4902 | ||
2bd59d48 | 4903 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4904 | { |
a9746d8d | 4905 | struct cgroup *cgrp; |
2bd59d48 | 4906 | int ret = 0; |
42809dd4 | 4907 | |
a9746d8d TH |
4908 | cgrp = cgroup_kn_lock_live(kn); |
4909 | if (!cgrp) | |
4910 | return 0; | |
42809dd4 | 4911 | |
a9746d8d | 4912 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 4913 | |
a9746d8d | 4914 | cgroup_kn_unlock(kn); |
42809dd4 | 4915 | return ret; |
8e3f6541 TH |
4916 | } |
4917 | ||
2bd59d48 TH |
4918 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4919 | .remount_fs = cgroup_remount, | |
4920 | .show_options = cgroup_show_options, | |
4921 | .mkdir = cgroup_mkdir, | |
4922 | .rmdir = cgroup_rmdir, | |
4923 | .rename = cgroup_rename, | |
4924 | }; | |
4925 | ||
15a4c835 | 4926 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4927 | { |
ddbcc7e8 | 4928 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4929 | |
4930 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4931 | |
648bb56d TH |
4932 | mutex_lock(&cgroup_mutex); |
4933 | ||
15a4c835 | 4934 | idr_init(&ss->css_idr); |
0adb0704 | 4935 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4936 | |
3dd06ffa TH |
4937 | /* Create the root cgroup state for this subsystem */ |
4938 | ss->root = &cgrp_dfl_root; | |
4939 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4940 | /* We don't handle early failures gracefully */ |
4941 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4942 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
4943 | |
4944 | /* | |
4945 | * Root csses are never destroyed and we can't initialize | |
4946 | * percpu_ref during early init. Disable refcnting. | |
4947 | */ | |
4948 | css->flags |= CSS_NO_REF; | |
4949 | ||
15a4c835 | 4950 | if (early) { |
9395a450 | 4951 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
4952 | css->id = 1; |
4953 | } else { | |
4954 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4955 | BUG_ON(css->id < 0); | |
4956 | } | |
ddbcc7e8 | 4957 | |
e8d55fde | 4958 | /* Update the init_css_set to contain a subsys |
817929ec | 4959 | * pointer to this state - since the subsystem is |
e8d55fde | 4960 | * newly registered, all tasks and hence the |
3dd06ffa | 4961 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4962 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 4963 | |
cb4a3167 AS |
4964 | have_fork_callback |= (bool)ss->fork << ss->id; |
4965 | have_exit_callback |= (bool)ss->exit << ss->id; | |
7e47682e | 4966 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 4967 | |
e8d55fde LZ |
4968 | /* At system boot, before all subsystems have been |
4969 | * registered, no tasks have been forked, so we don't | |
4970 | * need to invoke fork callbacks here. */ | |
4971 | BUG_ON(!list_empty(&init_task.tasks)); | |
4972 | ||
ae7f164a | 4973 | BUG_ON(online_css(css)); |
a8638030 | 4974 | |
cf5d5941 BB |
4975 | mutex_unlock(&cgroup_mutex); |
4976 | } | |
cf5d5941 | 4977 | |
ddbcc7e8 | 4978 | /** |
a043e3b2 LZ |
4979 | * cgroup_init_early - cgroup initialization at system boot |
4980 | * | |
4981 | * Initialize cgroups at system boot, and initialize any | |
4982 | * subsystems that request early init. | |
ddbcc7e8 PM |
4983 | */ |
4984 | int __init cgroup_init_early(void) | |
4985 | { | |
7b9a6ba5 | 4986 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 4987 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4988 | int i; |
30159ec7 | 4989 | |
3dd06ffa | 4990 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
4991 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
4992 | ||
a4ea1cc9 | 4993 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4994 | |
3ed80a62 | 4995 | for_each_subsys(ss, i) { |
aec25020 | 4996 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4997 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4998 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4999 | ss->id, ss->name); |
073219e9 TH |
5000 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5001 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5002 | ||
aec25020 | 5003 | ss->id = i; |
073219e9 | 5004 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5005 | if (!ss->legacy_name) |
5006 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5007 | |
5008 | if (ss->early_init) | |
15a4c835 | 5009 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5010 | } |
5011 | return 0; | |
5012 | } | |
5013 | ||
5014 | /** | |
a043e3b2 LZ |
5015 | * cgroup_init - cgroup initialization |
5016 | * | |
5017 | * Register cgroup filesystem and /proc file, and initialize | |
5018 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5019 | */ |
5020 | int __init cgroup_init(void) | |
5021 | { | |
30159ec7 | 5022 | struct cgroup_subsys *ss; |
0ac801fe | 5023 | unsigned long key; |
172a2c06 | 5024 | int ssid, err; |
ddbcc7e8 | 5025 | |
d59cfc09 | 5026 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5027 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5028 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5029 | |
54e7b4eb | 5030 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5031 | |
82fe9b0d TH |
5032 | /* Add init_css_set to the hash table */ |
5033 | key = css_set_hash(init_css_set.subsys); | |
5034 | hash_add(css_set_table, &init_css_set.hlist, key); | |
5035 | ||
3dd06ffa | 5036 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5037 | |
54e7b4eb TH |
5038 | mutex_unlock(&cgroup_mutex); |
5039 | ||
172a2c06 | 5040 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5041 | if (ss->early_init) { |
5042 | struct cgroup_subsys_state *css = | |
5043 | init_css_set.subsys[ss->id]; | |
5044 | ||
5045 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5046 | GFP_KERNEL); | |
5047 | BUG_ON(css->id < 0); | |
5048 | } else { | |
5049 | cgroup_init_subsys(ss, false); | |
5050 | } | |
172a2c06 | 5051 | |
2d8f243a TH |
5052 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5053 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5054 | |
5055 | /* | |
c731ae1d LZ |
5056 | * Setting dfl_root subsys_mask needs to consider the |
5057 | * disabled flag and cftype registration needs kmalloc, | |
5058 | * both of which aren't available during early_init. | |
172a2c06 | 5059 | */ |
a8ddc821 TH |
5060 | if (ss->disabled) |
5061 | continue; | |
5062 | ||
5063 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
5064 | ||
5065 | if (cgroup_legacy_files_on_dfl && !ss->dfl_cftypes) | |
5066 | ss->dfl_cftypes = ss->legacy_cftypes; | |
5067 | ||
5de4fa13 TH |
5068 | if (!ss->dfl_cftypes) |
5069 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
5070 | ||
a8ddc821 TH |
5071 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5072 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5073 | } else { | |
5074 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5075 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5076 | } |
295458e6 VD |
5077 | |
5078 | if (ss->bind) | |
5079 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5080 | } |
5081 | ||
f9bb4882 EB |
5082 | err = sysfs_create_mount_point(fs_kobj, "cgroup"); |
5083 | if (err) | |
5084 | return err; | |
676db4af | 5085 | |
ddbcc7e8 | 5086 | err = register_filesystem(&cgroup_fs_type); |
676db4af | 5087 | if (err < 0) { |
f9bb4882 | 5088 | sysfs_remove_mount_point(fs_kobj, "cgroup"); |
2bd59d48 | 5089 | return err; |
676db4af | 5090 | } |
ddbcc7e8 | 5091 | |
46ae220b | 5092 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 5093 | return 0; |
ddbcc7e8 | 5094 | } |
b4f48b63 | 5095 | |
e5fca243 TH |
5096 | static int __init cgroup_wq_init(void) |
5097 | { | |
5098 | /* | |
5099 | * There isn't much point in executing destruction path in | |
5100 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5101 | * Use 1 for @max_active. |
e5fca243 TH |
5102 | * |
5103 | * We would prefer to do this in cgroup_init() above, but that | |
5104 | * is called before init_workqueues(): so leave this until after. | |
5105 | */ | |
1a11533f | 5106 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5107 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5108 | |
5109 | /* | |
5110 | * Used to destroy pidlists and separate to serve as flush domain. | |
5111 | * Cap @max_active to 1 too. | |
5112 | */ | |
5113 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5114 | 0, 1); | |
5115 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5116 | ||
e5fca243 TH |
5117 | return 0; |
5118 | } | |
5119 | core_initcall(cgroup_wq_init); | |
5120 | ||
a424316c PM |
5121 | /* |
5122 | * proc_cgroup_show() | |
5123 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5124 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5125 | */ |
006f4ac4 ZL |
5126 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5127 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5128 | { |
e61734c5 | 5129 | char *buf, *path; |
a424316c | 5130 | int retval; |
3dd06ffa | 5131 | struct cgroup_root *root; |
a424316c PM |
5132 | |
5133 | retval = -ENOMEM; | |
e61734c5 | 5134 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5135 | if (!buf) |
5136 | goto out; | |
5137 | ||
a424316c | 5138 | mutex_lock(&cgroup_mutex); |
96d365e0 | 5139 | down_read(&css_set_rwsem); |
a424316c | 5140 | |
985ed670 | 5141 | for_each_root(root) { |
a424316c | 5142 | struct cgroup_subsys *ss; |
bd89aabc | 5143 | struct cgroup *cgrp; |
b85d2040 | 5144 | int ssid, count = 0; |
a424316c | 5145 | |
a2dd4247 | 5146 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
5147 | continue; |
5148 | ||
2c6ab6d2 | 5149 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5150 | if (root != &cgrp_dfl_root) |
5151 | for_each_subsys(ss, ssid) | |
5152 | if (root->subsys_mask & (1 << ssid)) | |
5153 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5154 | ss->legacy_name); |
c6d57f33 PM |
5155 | if (strlen(root->name)) |
5156 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5157 | root->name); | |
a424316c | 5158 | seq_putc(m, ':'); |
7717f7ba | 5159 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
5160 | path = cgroup_path(cgrp, buf, PATH_MAX); |
5161 | if (!path) { | |
5162 | retval = -ENAMETOOLONG; | |
a424316c | 5163 | goto out_unlock; |
e61734c5 TH |
5164 | } |
5165 | seq_puts(m, path); | |
a424316c PM |
5166 | seq_putc(m, '\n'); |
5167 | } | |
5168 | ||
006f4ac4 | 5169 | retval = 0; |
a424316c | 5170 | out_unlock: |
96d365e0 | 5171 | up_read(&css_set_rwsem); |
a424316c | 5172 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5173 | kfree(buf); |
5174 | out: | |
5175 | return retval; | |
5176 | } | |
5177 | ||
a424316c PM |
5178 | /* Display information about each subsystem and each hierarchy */ |
5179 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5180 | { | |
30159ec7 | 5181 | struct cgroup_subsys *ss; |
a424316c | 5182 | int i; |
a424316c | 5183 | |
8bab8dde | 5184 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5185 | /* |
5186 | * ideally we don't want subsystems moving around while we do this. | |
5187 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5188 | * subsys/hierarchy state. | |
5189 | */ | |
a424316c | 5190 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5191 | |
5192 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5193 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5194 | ss->legacy_name, ss->root->hierarchy_id, |
3c9c825b | 5195 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 5196 | |
a424316c PM |
5197 | mutex_unlock(&cgroup_mutex); |
5198 | return 0; | |
5199 | } | |
5200 | ||
5201 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5202 | { | |
9dce07f1 | 5203 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5204 | } |
5205 | ||
828c0950 | 5206 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5207 | .open = cgroupstats_open, |
5208 | .read = seq_read, | |
5209 | .llseek = seq_lseek, | |
5210 | .release = single_release, | |
5211 | }; | |
5212 | ||
7e47682e AS |
5213 | static void **subsys_canfork_priv_p(void *ss_priv[CGROUP_CANFORK_COUNT], int i) |
5214 | { | |
5215 | if (CGROUP_CANFORK_START <= i && i < CGROUP_CANFORK_END) | |
5216 | return &ss_priv[i - CGROUP_CANFORK_START]; | |
5217 | return NULL; | |
5218 | } | |
5219 | ||
5220 | static void *subsys_canfork_priv(void *ss_priv[CGROUP_CANFORK_COUNT], int i) | |
5221 | { | |
5222 | void **private = subsys_canfork_priv_p(ss_priv, i); | |
5223 | return private ? *private : NULL; | |
5224 | } | |
5225 | ||
b4f48b63 | 5226 | /** |
eaf797ab | 5227 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5228 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5229 | * |
eaf797ab TH |
5230 | * A task is associated with the init_css_set until cgroup_post_fork() |
5231 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5232 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5233 | */ |
5234 | void cgroup_fork(struct task_struct *child) | |
5235 | { | |
eaf797ab | 5236 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5237 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5238 | } |
5239 | ||
7e47682e AS |
5240 | /** |
5241 | * cgroup_can_fork - called on a new task before the process is exposed | |
5242 | * @child: the task in question. | |
5243 | * | |
5244 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5245 | * returns an error, the fork aborts with that error code. This allows for | |
5246 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5247 | */ | |
5248 | int cgroup_can_fork(struct task_struct *child, | |
5249 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5250 | { | |
5251 | struct cgroup_subsys *ss; | |
5252 | int i, j, ret; | |
5253 | ||
5254 | for_each_subsys_which(ss, i, &have_canfork_callback) { | |
5255 | ret = ss->can_fork(child, subsys_canfork_priv_p(ss_priv, i)); | |
5256 | if (ret) | |
5257 | goto out_revert; | |
5258 | } | |
5259 | ||
5260 | return 0; | |
5261 | ||
5262 | out_revert: | |
5263 | for_each_subsys(ss, j) { | |
5264 | if (j >= i) | |
5265 | break; | |
5266 | if (ss->cancel_fork) | |
5267 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, j)); | |
5268 | } | |
5269 | ||
5270 | return ret; | |
5271 | } | |
5272 | ||
5273 | /** | |
5274 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5275 | * @child: the task in question | |
5276 | * | |
5277 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5278 | * cgroup_can_fork() succeded. | |
5279 | */ | |
5280 | void cgroup_cancel_fork(struct task_struct *child, | |
5281 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5282 | { | |
5283 | struct cgroup_subsys *ss; | |
5284 | int i; | |
5285 | ||
5286 | for_each_subsys(ss, i) | |
5287 | if (ss->cancel_fork) | |
5288 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, i)); | |
5289 | } | |
5290 | ||
817929ec | 5291 | /** |
a043e3b2 LZ |
5292 | * cgroup_post_fork - called on a new task after adding it to the task list |
5293 | * @child: the task in question | |
5294 | * | |
5edee61e TH |
5295 | * Adds the task to the list running through its css_set if necessary and |
5296 | * call the subsystem fork() callbacks. Has to be after the task is | |
5297 | * visible on the task list in case we race with the first call to | |
0942eeee | 5298 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5299 | * list. |
a043e3b2 | 5300 | */ |
7e47682e AS |
5301 | void cgroup_post_fork(struct task_struct *child, |
5302 | void *old_ss_priv[CGROUP_CANFORK_COUNT]) | |
817929ec | 5303 | { |
30159ec7 | 5304 | struct cgroup_subsys *ss; |
5edee61e TH |
5305 | int i; |
5306 | ||
3ce3230a | 5307 | /* |
251f8c03 | 5308 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5309 | * function sets use_task_css_set_links before grabbing |
5310 | * tasklist_lock and we just went through tasklist_lock to add | |
5311 | * @child, it's guaranteed that either we see the set | |
5312 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5313 | * @child during its iteration. | |
5314 | * | |
5315 | * If we won the race, @child is associated with %current's | |
5316 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
5317 | * association is stable, and, on completion of the parent's | |
5318 | * migration, @child is visible in the source of migration or | |
5319 | * already in the destination cgroup. This guarantee is necessary | |
5320 | * when implementing operations which need to migrate all tasks of | |
5321 | * a cgroup to another. | |
5322 | * | |
251f8c03 | 5323 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5324 | * will remain in init_css_set. This is safe because all tasks are |
5325 | * in the init_css_set before cg_links is enabled and there's no | |
5326 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5327 | */ |
817929ec | 5328 | if (use_task_css_set_links) { |
eaf797ab TH |
5329 | struct css_set *cset; |
5330 | ||
96d365e0 | 5331 | down_write(&css_set_rwsem); |
0e1d768f | 5332 | cset = task_css_set(current); |
eaf797ab TH |
5333 | if (list_empty(&child->cg_list)) { |
5334 | rcu_assign_pointer(child->cgroups, cset); | |
5335 | list_add(&child->cg_list, &cset->tasks); | |
5336 | get_css_set(cset); | |
5337 | } | |
96d365e0 | 5338 | up_write(&css_set_rwsem); |
817929ec | 5339 | } |
5edee61e TH |
5340 | |
5341 | /* | |
5342 | * Call ss->fork(). This must happen after @child is linked on | |
5343 | * css_set; otherwise, @child might change state between ->fork() | |
5344 | * and addition to css_set. | |
5345 | */ | |
cb4a3167 | 5346 | for_each_subsys_which(ss, i, &have_fork_callback) |
7e47682e | 5347 | ss->fork(child, subsys_canfork_priv(old_ss_priv, i)); |
817929ec | 5348 | } |
5edee61e | 5349 | |
b4f48b63 PM |
5350 | /** |
5351 | * cgroup_exit - detach cgroup from exiting task | |
5352 | * @tsk: pointer to task_struct of exiting process | |
5353 | * | |
5354 | * Description: Detach cgroup from @tsk and release it. | |
5355 | * | |
5356 | * Note that cgroups marked notify_on_release force every task in | |
5357 | * them to take the global cgroup_mutex mutex when exiting. | |
5358 | * This could impact scaling on very large systems. Be reluctant to | |
5359 | * use notify_on_release cgroups where very high task exit scaling | |
5360 | * is required on large systems. | |
5361 | * | |
0e1d768f TH |
5362 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5363 | * call cgroup_exit() while the task is still competent to handle | |
5364 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5365 | * each hierarchy for the remainder of its exit. No need to bother with | |
5366 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5367 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5368 | */ |
1ec41830 | 5369 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5370 | { |
30159ec7 | 5371 | struct cgroup_subsys *ss; |
5abb8855 | 5372 | struct css_set *cset; |
eaf797ab | 5373 | bool put_cset = false; |
d41d5a01 | 5374 | int i; |
817929ec PM |
5375 | |
5376 | /* | |
0e1d768f TH |
5377 | * Unlink from @tsk from its css_set. As migration path can't race |
5378 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
5379 | */ |
5380 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 5381 | down_write(&css_set_rwsem); |
0e1d768f | 5382 | list_del_init(&tsk->cg_list); |
96d365e0 | 5383 | up_write(&css_set_rwsem); |
0e1d768f | 5384 | put_cset = true; |
817929ec PM |
5385 | } |
5386 | ||
b4f48b63 | 5387 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
5388 | cset = task_css_set(tsk); |
5389 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 5390 | |
cb4a3167 AS |
5391 | /* see cgroup_post_fork() for details */ |
5392 | for_each_subsys_which(ss, i, &have_exit_callback) { | |
5393 | struct cgroup_subsys_state *old_css = cset->subsys[i]; | |
5394 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5395 | |
cb4a3167 | 5396 | ss->exit(css, old_css, tsk); |
d41d5a01 | 5397 | } |
d41d5a01 | 5398 | |
eaf797ab | 5399 | if (put_cset) |
a25eb52e | 5400 | put_css_set(cset); |
b4f48b63 | 5401 | } |
697f4161 | 5402 | |
bd89aabc | 5403 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5404 | { |
a25eb52e | 5405 | if (notify_on_release(cgrp) && !cgroup_has_tasks(cgrp) && |
971ff493 ZL |
5406 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5407 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5408 | } |
5409 | ||
81a6a5cd PM |
5410 | /* |
5411 | * Notify userspace when a cgroup is released, by running the | |
5412 | * configured release agent with the name of the cgroup (path | |
5413 | * relative to the root of cgroup file system) as the argument. | |
5414 | * | |
5415 | * Most likely, this user command will try to rmdir this cgroup. | |
5416 | * | |
5417 | * This races with the possibility that some other task will be | |
5418 | * attached to this cgroup before it is removed, or that some other | |
5419 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5420 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5421 | * unused, and this cgroup will be reprieved from its death sentence, | |
5422 | * to continue to serve a useful existence. Next time it's released, | |
5423 | * we will get notified again, if it still has 'notify_on_release' set. | |
5424 | * | |
5425 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5426 | * means only wait until the task is successfully execve()'d. The | |
5427 | * separate release agent task is forked by call_usermodehelper(), | |
5428 | * then control in this thread returns here, without waiting for the | |
5429 | * release agent task. We don't bother to wait because the caller of | |
5430 | * this routine has no use for the exit status of the release agent | |
5431 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5432 | */ |
81a6a5cd PM |
5433 | static void cgroup_release_agent(struct work_struct *work) |
5434 | { | |
971ff493 ZL |
5435 | struct cgroup *cgrp = |
5436 | container_of(work, struct cgroup, release_agent_work); | |
5437 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5438 | char *argv[3], *envp[3]; | |
5439 | ||
81a6a5cd | 5440 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5441 | |
5442 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5443 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5444 | if (!pathbuf || !agentbuf) | |
5445 | goto out; | |
5446 | ||
5447 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5448 | if (!path) | |
5449 | goto out; | |
5450 | ||
5451 | argv[0] = agentbuf; | |
5452 | argv[1] = path; | |
5453 | argv[2] = NULL; | |
5454 | ||
5455 | /* minimal command environment */ | |
5456 | envp[0] = "HOME=/"; | |
5457 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5458 | envp[2] = NULL; | |
5459 | ||
81a6a5cd | 5460 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5461 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5462 | goto out_free; |
971ff493 | 5463 | out: |
81a6a5cd | 5464 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5465 | out_free: |
971ff493 ZL |
5466 | kfree(agentbuf); |
5467 | kfree(pathbuf); | |
81a6a5cd | 5468 | } |
8bab8dde PM |
5469 | |
5470 | static int __init cgroup_disable(char *str) | |
5471 | { | |
30159ec7 | 5472 | struct cgroup_subsys *ss; |
8bab8dde | 5473 | char *token; |
30159ec7 | 5474 | int i; |
8bab8dde PM |
5475 | |
5476 | while ((token = strsep(&str, ",")) != NULL) { | |
5477 | if (!*token) | |
5478 | continue; | |
be45c900 | 5479 | |
3ed80a62 | 5480 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5481 | if (strcmp(token, ss->name) && |
5482 | strcmp(token, ss->legacy_name)) | |
5483 | continue; | |
5484 | ||
5485 | ss->disabled = 1; | |
5486 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5487 | ss->name); | |
5488 | break; | |
8bab8dde PM |
5489 | } |
5490 | } | |
5491 | return 1; | |
5492 | } | |
5493 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5494 | |
a8ddc821 TH |
5495 | static int __init cgroup_set_legacy_files_on_dfl(char *str) |
5496 | { | |
5497 | printk("cgroup: using legacy files on the default hierarchy\n"); | |
5498 | cgroup_legacy_files_on_dfl = true; | |
5499 | return 0; | |
5500 | } | |
5501 | __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl); | |
5502 | ||
b77d7b60 | 5503 | /** |
ec903c0c | 5504 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5505 | * @dentry: directory dentry of interest |
5506 | * @ss: subsystem of interest | |
b77d7b60 | 5507 | * |
5a17f543 TH |
5508 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5509 | * to get the corresponding css and return it. If such css doesn't exist | |
5510 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5511 | */ |
ec903c0c TH |
5512 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5513 | struct cgroup_subsys *ss) | |
e5d1367f | 5514 | { |
2bd59d48 TH |
5515 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5516 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5517 | struct cgroup *cgrp; |
e5d1367f | 5518 | |
35cf0836 | 5519 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5520 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5521 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5522 | return ERR_PTR(-EBADF); |
5523 | ||
5a17f543 TH |
5524 | rcu_read_lock(); |
5525 | ||
2bd59d48 TH |
5526 | /* |
5527 | * This path doesn't originate from kernfs and @kn could already | |
5528 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5529 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5530 | */ |
5531 | cgrp = rcu_dereference(kn->priv); | |
5532 | if (cgrp) | |
5533 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5534 | |
ec903c0c | 5535 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5536 | css = ERR_PTR(-ENOENT); |
5537 | ||
5538 | rcu_read_unlock(); | |
5539 | return css; | |
e5d1367f | 5540 | } |
e5d1367f | 5541 | |
1cb650b9 LZ |
5542 | /** |
5543 | * css_from_id - lookup css by id | |
5544 | * @id: the cgroup id | |
5545 | * @ss: cgroup subsys to be looked into | |
5546 | * | |
5547 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5548 | * Should be called under rcu_read_lock(). | |
5549 | */ | |
5550 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5551 | { | |
6fa4918d | 5552 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5553 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5554 | } |
5555 | ||
fe693435 | 5556 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5557 | static struct cgroup_subsys_state * |
5558 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5559 | { |
5560 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5561 | ||
5562 | if (!css) | |
5563 | return ERR_PTR(-ENOMEM); | |
5564 | ||
5565 | return css; | |
5566 | } | |
5567 | ||
eb95419b | 5568 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5569 | { |
eb95419b | 5570 | kfree(css); |
fe693435 PM |
5571 | } |
5572 | ||
182446d0 TH |
5573 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5574 | struct cftype *cft) | |
fe693435 | 5575 | { |
182446d0 | 5576 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5577 | } |
5578 | ||
182446d0 TH |
5579 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5580 | struct cftype *cft) | |
fe693435 PM |
5581 | { |
5582 | return (u64)(unsigned long)current->cgroups; | |
5583 | } | |
5584 | ||
182446d0 | 5585 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5586 | struct cftype *cft) |
fe693435 PM |
5587 | { |
5588 | u64 count; | |
5589 | ||
5590 | rcu_read_lock(); | |
a8ad805c | 5591 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5592 | rcu_read_unlock(); |
5593 | return count; | |
5594 | } | |
5595 | ||
2da8ca82 | 5596 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5597 | { |
69d0206c | 5598 | struct cgrp_cset_link *link; |
5abb8855 | 5599 | struct css_set *cset; |
e61734c5 TH |
5600 | char *name_buf; |
5601 | ||
5602 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5603 | if (!name_buf) | |
5604 | return -ENOMEM; | |
7717f7ba | 5605 | |
96d365e0 | 5606 | down_read(&css_set_rwsem); |
7717f7ba | 5607 | rcu_read_lock(); |
5abb8855 | 5608 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5609 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5610 | struct cgroup *c = link->cgrp; |
7717f7ba | 5611 | |
a2dd4247 | 5612 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5613 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5614 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5615 | } |
5616 | rcu_read_unlock(); | |
96d365e0 | 5617 | up_read(&css_set_rwsem); |
e61734c5 | 5618 | kfree(name_buf); |
7717f7ba PM |
5619 | return 0; |
5620 | } | |
5621 | ||
5622 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5623 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5624 | { |
2da8ca82 | 5625 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5626 | struct cgrp_cset_link *link; |
7717f7ba | 5627 | |
96d365e0 | 5628 | down_read(&css_set_rwsem); |
182446d0 | 5629 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5630 | struct css_set *cset = link->cset; |
7717f7ba PM |
5631 | struct task_struct *task; |
5632 | int count = 0; | |
c7561128 | 5633 | |
5abb8855 | 5634 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5635 | |
5abb8855 | 5636 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5637 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5638 | goto overflow; | |
5639 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5640 | } | |
5641 | ||
5642 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5643 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5644 | goto overflow; | |
5645 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5646 | } |
c7561128 TH |
5647 | continue; |
5648 | overflow: | |
5649 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5650 | } |
96d365e0 | 5651 | up_read(&css_set_rwsem); |
7717f7ba PM |
5652 | return 0; |
5653 | } | |
5654 | ||
182446d0 | 5655 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5656 | { |
a25eb52e ZL |
5657 | return (!cgroup_has_tasks(css->cgroup) && |
5658 | !css_has_online_children(&css->cgroup->self)); | |
fe693435 PM |
5659 | } |
5660 | ||
5661 | static struct cftype debug_files[] = { | |
fe693435 PM |
5662 | { |
5663 | .name = "taskcount", | |
5664 | .read_u64 = debug_taskcount_read, | |
5665 | }, | |
5666 | ||
5667 | { | |
5668 | .name = "current_css_set", | |
5669 | .read_u64 = current_css_set_read, | |
5670 | }, | |
5671 | ||
5672 | { | |
5673 | .name = "current_css_set_refcount", | |
5674 | .read_u64 = current_css_set_refcount_read, | |
5675 | }, | |
5676 | ||
7717f7ba PM |
5677 | { |
5678 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5679 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5680 | }, |
5681 | ||
5682 | { | |
5683 | .name = "cgroup_css_links", | |
2da8ca82 | 5684 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5685 | }, |
5686 | ||
fe693435 PM |
5687 | { |
5688 | .name = "releasable", | |
5689 | .read_u64 = releasable_read, | |
5690 | }, | |
fe693435 | 5691 | |
4baf6e33 TH |
5692 | { } /* terminate */ |
5693 | }; | |
fe693435 | 5694 | |
073219e9 | 5695 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5696 | .css_alloc = debug_css_alloc, |
5697 | .css_free = debug_css_free, | |
5577964e | 5698 | .legacy_cftypes = debug_files, |
fe693435 PM |
5699 | }; |
5700 | #endif /* CONFIG_CGROUP_DEBUG */ |