]>
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> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
846c7bb0 | 60 | |
60063497 | 61 | #include <linux/atomic.h> |
ddbcc7e8 | 62 | |
b1a21367 TH |
63 | /* |
64 | * pidlists linger the following amount before being destroyed. The goal | |
65 | * is avoiding frequent destruction in the middle of consecutive read calls | |
66 | * Expiring in the middle is a performance problem not a correctness one. | |
67 | * 1 sec should be enough. | |
68 | */ | |
69 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
70 | ||
8d7e6fb0 TH |
71 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
72 | MAX_CFTYPE_NAME + 2) | |
73 | ||
e25e2cbb TH |
74 | /* |
75 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
76 | * hierarchy must be performed while holding it. | |
77 | * | |
0e1d768f TH |
78 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
79 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 80 | * |
0e1d768f TH |
81 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
82 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 83 | */ |
2219449a TH |
84 | #ifdef CONFIG_PROVE_RCU |
85 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
86 | DECLARE_RWSEM(css_set_rwsem); |
87 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
88 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 89 | #else |
81a6a5cd | 90 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 91 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
92 | #endif |
93 | ||
6fa4918d | 94 | /* |
15a4c835 TH |
95 | * Protects cgroup_idr and css_idr so that IDs can be released without |
96 | * grabbing cgroup_mutex. | |
6fa4918d TH |
97 | */ |
98 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
99 | ||
69e943b7 TH |
100 | /* |
101 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
102 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
103 | */ | |
104 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 105 | |
8353da1f | 106 | #define cgroup_assert_mutex_or_rcu_locked() \ |
87fb54f1 TH |
107 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
108 | lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 109 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 110 | |
e5fca243 TH |
111 | /* |
112 | * cgroup destruction makes heavy use of work items and there can be a lot | |
113 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
114 | * destruction work items don't end up filling up max_active of system_wq | |
115 | * which may lead to deadlock. | |
116 | */ | |
117 | static struct workqueue_struct *cgroup_destroy_wq; | |
118 | ||
b1a21367 TH |
119 | /* |
120 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
121 | * separate workqueue as flush domain. | |
122 | */ | |
123 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
124 | ||
3ed80a62 | 125 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 126 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 127 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
128 | #include <linux/cgroup_subsys.h> |
129 | }; | |
073219e9 TH |
130 | #undef SUBSYS |
131 | ||
132 | /* array of cgroup subsystem names */ | |
133 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
134 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
135 | #include <linux/cgroup_subsys.h> |
136 | }; | |
073219e9 | 137 | #undef SUBSYS |
ddbcc7e8 | 138 | |
ddbcc7e8 | 139 | /* |
3dd06ffa | 140 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
141 | * unattached - it never has more than a single cgroup, and all tasks are |
142 | * part of that cgroup. | |
ddbcc7e8 | 143 | */ |
a2dd4247 | 144 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 145 | |
a2dd4247 TH |
146 | /* |
147 | * The default hierarchy always exists but is hidden until mounted for the | |
148 | * first time. This is for backward compatibility. | |
149 | */ | |
150 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 | 151 | |
a8ddc821 TH |
152 | /* |
153 | * Set by the boot param of the same name and makes subsystems with NULL | |
154 | * ->dfl_files to use ->legacy_files on the default hierarchy. | |
155 | */ | |
156 | static bool cgroup_legacy_files_on_dfl; | |
157 | ||
5533e011 | 158 | /* some controllers are not supported in the default hierarchy */ |
5de4fa13 | 159 | static unsigned int cgrp_dfl_root_inhibit_ss_mask; |
5533e011 | 160 | |
ddbcc7e8 PM |
161 | /* The list of hierarchy roots */ |
162 | ||
9871bf95 TH |
163 | static LIST_HEAD(cgroup_roots); |
164 | static int cgroup_root_count; | |
ddbcc7e8 | 165 | |
3417ae1f | 166 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 167 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 168 | |
794611a1 | 169 | /* |
0cb51d71 TH |
170 | * Assign a monotonically increasing serial number to csses. It guarantees |
171 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
172 | * Also, as csses are always appended to the parent's ->children list, it | |
173 | * guarantees that sibling csses are always sorted in the ascending serial | |
174 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 175 | */ |
0cb51d71 | 176 | static u64 css_serial_nr_next = 1; |
794611a1 | 177 | |
ddbcc7e8 | 178 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
179 | * check for fork/exit handlers to call. This avoids us having to do |
180 | * extra work in the fork/exit path if none of the subsystems need to | |
181 | * be called. | |
ddbcc7e8 | 182 | */ |
8947f9d5 | 183 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 184 | |
a14c6874 TH |
185 | static struct cftype cgroup_dfl_base_files[]; |
186 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 187 | |
3dd06ffa | 188 | static int rebind_subsystems(struct cgroup_root *dst_root, |
69dfa00c | 189 | unsigned int ss_mask); |
42809dd4 | 190 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
191 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
192 | bool visible); | |
9d755d33 | 193 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 194 | static void kill_css(struct cgroup_subsys_state *css); |
2bb566cb TH |
195 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
196 | bool is_add); | |
42809dd4 | 197 | |
6fa4918d TH |
198 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
199 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
200 | gfp_t gfp_mask) | |
201 | { | |
202 | int ret; | |
203 | ||
204 | idr_preload(gfp_mask); | |
54504e97 | 205 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 206 | ret = idr_alloc(idr, ptr, start, end, gfp_mask); |
54504e97 | 207 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
208 | idr_preload_end(); |
209 | return ret; | |
210 | } | |
211 | ||
212 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
213 | { | |
214 | void *ret; | |
215 | ||
54504e97 | 216 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 217 | ret = idr_replace(idr, ptr, id); |
54504e97 | 218 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
219 | return ret; |
220 | } | |
221 | ||
222 | static void cgroup_idr_remove(struct idr *idr, int id) | |
223 | { | |
54504e97 | 224 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 225 | idr_remove(idr, id); |
54504e97 | 226 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
227 | } |
228 | ||
d51f39b0 TH |
229 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
230 | { | |
231 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
232 | ||
233 | if (parent_css) | |
234 | return container_of(parent_css, struct cgroup, self); | |
235 | return NULL; | |
236 | } | |
237 | ||
95109b62 TH |
238 | /** |
239 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
240 | * @cgrp: the cgroup of interest | |
9d800df1 | 241 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 242 | * |
ca8bdcaf TH |
243 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
244 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
245 | * the caller is responsible for pinning the returned css if it wants to | |
246 | * keep accessing it outside the said locks. This function may return | |
247 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
248 | */ |
249 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 250 | struct cgroup_subsys *ss) |
95109b62 | 251 | { |
ca8bdcaf | 252 | if (ss) |
aec25020 | 253 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 254 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 255 | else |
9d800df1 | 256 | return &cgrp->self; |
95109b62 | 257 | } |
42809dd4 | 258 | |
aec3dfcb TH |
259 | /** |
260 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
261 | * @cgrp: the cgroup of interest | |
9d800df1 | 262 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb TH |
263 | * |
264 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
265 | * as the matching css of the nearest ancestor including self which has @ss | |
266 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
267 | * function is guaranteed to return non-NULL css. | |
268 | */ | |
269 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
270 | struct cgroup_subsys *ss) | |
271 | { | |
272 | lockdep_assert_held(&cgroup_mutex); | |
273 | ||
274 | if (!ss) | |
9d800df1 | 275 | return &cgrp->self; |
aec3dfcb TH |
276 | |
277 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
278 | return NULL; | |
279 | ||
d51f39b0 TH |
280 | while (cgroup_parent(cgrp) && |
281 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
282 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
283 | |
284 | return cgroup_css(cgrp, ss); | |
95109b62 | 285 | } |
42809dd4 | 286 | |
ddbcc7e8 | 287 | /* convenient tests for these bits */ |
54766d4a | 288 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 289 | { |
184faf32 | 290 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
291 | } |
292 | ||
b4168640 | 293 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 294 | { |
2bd59d48 | 295 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 296 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
297 | |
298 | /* | |
299 | * This is open and unprotected implementation of cgroup_css(). | |
300 | * seq_css() is only called from a kernfs file operation which has | |
301 | * an active reference on the file. Because all the subsystem | |
302 | * files are drained before a css is disassociated with a cgroup, | |
303 | * the matching css from the cgroup's subsys table is guaranteed to | |
304 | * be and stay valid until the enclosing operation is complete. | |
305 | */ | |
306 | if (cft->ss) | |
307 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
308 | else | |
9d800df1 | 309 | return &cgrp->self; |
59f5296b | 310 | } |
b4168640 | 311 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 312 | |
78574cf9 LZ |
313 | /** |
314 | * cgroup_is_descendant - test ancestry | |
315 | * @cgrp: the cgroup to be tested | |
316 | * @ancestor: possible ancestor of @cgrp | |
317 | * | |
318 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
319 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
320 | * and @ancestor are accessible. | |
321 | */ | |
322 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
323 | { | |
324 | while (cgrp) { | |
325 | if (cgrp == ancestor) | |
326 | return true; | |
d51f39b0 | 327 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
328 | } |
329 | return false; | |
330 | } | |
ddbcc7e8 | 331 | |
e9685a03 | 332 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 333 | { |
bd89aabc | 334 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
335 | } |
336 | ||
1c6727af TH |
337 | /** |
338 | * for_each_css - iterate all css's of a cgroup | |
339 | * @css: the iteration cursor | |
340 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
341 | * @cgrp: the target cgroup to iterate css's of | |
342 | * | |
aec3dfcb | 343 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
344 | */ |
345 | #define for_each_css(css, ssid, cgrp) \ | |
346 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
347 | if (!((css) = rcu_dereference_check( \ | |
348 | (cgrp)->subsys[(ssid)], \ | |
349 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
350 | else | |
351 | ||
aec3dfcb TH |
352 | /** |
353 | * for_each_e_css - iterate all effective css's of a cgroup | |
354 | * @css: the iteration cursor | |
355 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
356 | * @cgrp: the target cgroup to iterate css's of | |
357 | * | |
358 | * Should be called under cgroup_[tree_]mutex. | |
359 | */ | |
360 | #define for_each_e_css(css, ssid, cgrp) \ | |
361 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
362 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
363 | ; \ | |
364 | else | |
365 | ||
30159ec7 | 366 | /** |
3ed80a62 | 367 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 368 | * @ss: the iteration cursor |
780cd8b3 | 369 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 370 | */ |
780cd8b3 | 371 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
372 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
373 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 374 | |
985ed670 TH |
375 | /* iterate across the hierarchies */ |
376 | #define for_each_root(root) \ | |
5549c497 | 377 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 378 | |
f8f22e53 TH |
379 | /* iterate over child cgrps, lock should be held throughout iteration */ |
380 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 381 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 382 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
383 | cgroup_is_dead(child); })) \ |
384 | ; \ | |
385 | else | |
7ae1bad9 | 386 | |
81a6a5cd | 387 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 388 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 389 | |
69d0206c TH |
390 | /* |
391 | * A cgroup can be associated with multiple css_sets as different tasks may | |
392 | * belong to different cgroups on different hierarchies. In the other | |
393 | * direction, a css_set is naturally associated with multiple cgroups. | |
394 | * This M:N relationship is represented by the following link structure | |
395 | * which exists for each association and allows traversing the associations | |
396 | * from both sides. | |
397 | */ | |
398 | struct cgrp_cset_link { | |
399 | /* the cgroup and css_set this link associates */ | |
400 | struct cgroup *cgrp; | |
401 | struct css_set *cset; | |
402 | ||
403 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
404 | struct list_head cset_link; | |
405 | ||
406 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
407 | struct list_head cgrp_link; | |
817929ec PM |
408 | }; |
409 | ||
172a2c06 TH |
410 | /* |
411 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
412 | * hierarchies being mounted. It contains a pointer to the root state |
413 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
414 | * reference-counted, to improve performance when child cgroups | |
415 | * haven't been created. | |
416 | */ | |
5024ae29 | 417 | struct css_set init_css_set = { |
172a2c06 TH |
418 | .refcount = ATOMIC_INIT(1), |
419 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
420 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
421 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
422 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
423 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
424 | }; | |
817929ec | 425 | |
172a2c06 | 426 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 427 | |
842b597e TH |
428 | /** |
429 | * cgroup_update_populated - updated populated count of a cgroup | |
430 | * @cgrp: the target cgroup | |
431 | * @populated: inc or dec populated count | |
432 | * | |
433 | * @cgrp is either getting the first task (css_set) or losing the last. | |
434 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
435 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
436 | * cgroup and all its descendants are empty. | |
437 | * | |
438 | * @cgrp's interface file "cgroup.populated" is zero if | |
439 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
440 | * changes from or to zero, userland is notified that the content of the | |
441 | * interface file has changed. This can be used to detect when @cgrp and | |
442 | * its descendants become populated or empty. | |
443 | */ | |
444 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
445 | { | |
446 | lockdep_assert_held(&css_set_rwsem); | |
447 | ||
448 | do { | |
449 | bool trigger; | |
450 | ||
451 | if (populated) | |
452 | trigger = !cgrp->populated_cnt++; | |
453 | else | |
454 | trigger = !--cgrp->populated_cnt; | |
455 | ||
456 | if (!trigger) | |
457 | break; | |
458 | ||
459 | if (cgrp->populated_kn) | |
460 | kernfs_notify(cgrp->populated_kn); | |
d51f39b0 | 461 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
462 | } while (cgrp); |
463 | } | |
464 | ||
7717f7ba PM |
465 | /* |
466 | * hash table for cgroup groups. This improves the performance to find | |
467 | * an existing css_set. This hash doesn't (currently) take into | |
468 | * account cgroups in empty hierarchies. | |
469 | */ | |
472b1053 | 470 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 471 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 472 | |
0ac801fe | 473 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 474 | { |
0ac801fe | 475 | unsigned long key = 0UL; |
30159ec7 TH |
476 | struct cgroup_subsys *ss; |
477 | int i; | |
472b1053 | 478 | |
30159ec7 | 479 | for_each_subsys(ss, i) |
0ac801fe LZ |
480 | key += (unsigned long)css[i]; |
481 | key = (key >> 16) ^ key; | |
472b1053 | 482 | |
0ac801fe | 483 | return key; |
472b1053 LZ |
484 | } |
485 | ||
a25eb52e | 486 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 487 | { |
69d0206c | 488 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
489 | struct cgroup_subsys *ss; |
490 | int ssid; | |
5abb8855 | 491 | |
89c5509b TH |
492 | lockdep_assert_held(&css_set_rwsem); |
493 | ||
494 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 495 | return; |
81a6a5cd | 496 | |
2c6ab6d2 | 497 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
498 | for_each_subsys(ss, ssid) |
499 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 500 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
501 | css_set_count--; |
502 | ||
69d0206c | 503 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 504 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 505 | |
69d0206c TH |
506 | list_del(&link->cset_link); |
507 | list_del(&link->cgrp_link); | |
71b5707e | 508 | |
96d365e0 | 509 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
510 | if (list_empty(&cgrp->cset_links)) { |
511 | cgroup_update_populated(cgrp, false); | |
a25eb52e | 512 | check_for_release(cgrp); |
81a6a5cd | 513 | } |
2c6ab6d2 PM |
514 | |
515 | kfree(link); | |
81a6a5cd | 516 | } |
2c6ab6d2 | 517 | |
5abb8855 | 518 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
519 | } |
520 | ||
a25eb52e | 521 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
522 | { |
523 | /* | |
524 | * Ensure that the refcount doesn't hit zero while any readers | |
525 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
526 | * rwlock | |
527 | */ | |
528 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
529 | return; | |
530 | ||
531 | down_write(&css_set_rwsem); | |
a25eb52e | 532 | put_css_set_locked(cset); |
89c5509b TH |
533 | up_write(&css_set_rwsem); |
534 | } | |
535 | ||
817929ec PM |
536 | /* |
537 | * refcounted get/put for css_set objects | |
538 | */ | |
5abb8855 | 539 | static inline void get_css_set(struct css_set *cset) |
817929ec | 540 | { |
5abb8855 | 541 | atomic_inc(&cset->refcount); |
817929ec PM |
542 | } |
543 | ||
b326f9d0 | 544 | /** |
7717f7ba | 545 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
546 | * @cset: candidate css_set being tested |
547 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
548 | * @new_cgrp: cgroup that's being entered by the task |
549 | * @template: desired set of css pointers in css_set (pre-calculated) | |
550 | * | |
6f4b7e63 | 551 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
552 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
553 | */ | |
5abb8855 TH |
554 | static bool compare_css_sets(struct css_set *cset, |
555 | struct css_set *old_cset, | |
7717f7ba PM |
556 | struct cgroup *new_cgrp, |
557 | struct cgroup_subsys_state *template[]) | |
558 | { | |
559 | struct list_head *l1, *l2; | |
560 | ||
aec3dfcb TH |
561 | /* |
562 | * On the default hierarchy, there can be csets which are | |
563 | * associated with the same set of cgroups but different csses. | |
564 | * Let's first ensure that csses match. | |
565 | */ | |
566 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 567 | return false; |
7717f7ba PM |
568 | |
569 | /* | |
570 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
571 | * different cgroups in hierarchies. As different cgroups may |
572 | * share the same effective css, this comparison is always | |
573 | * necessary. | |
7717f7ba | 574 | */ |
69d0206c TH |
575 | l1 = &cset->cgrp_links; |
576 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 577 | while (1) { |
69d0206c | 578 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 579 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
580 | |
581 | l1 = l1->next; | |
582 | l2 = l2->next; | |
583 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
584 | if (l1 == &cset->cgrp_links) { |
585 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
586 | break; |
587 | } else { | |
69d0206c | 588 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
589 | } |
590 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
591 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
592 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
593 | cgrp1 = link1->cgrp; | |
594 | cgrp2 = link2->cgrp; | |
7717f7ba | 595 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 596 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
597 | |
598 | /* | |
599 | * If this hierarchy is the hierarchy of the cgroup | |
600 | * that's changing, then we need to check that this | |
601 | * css_set points to the new cgroup; if it's any other | |
602 | * hierarchy, then this css_set should point to the | |
603 | * same cgroup as the old css_set. | |
604 | */ | |
5abb8855 TH |
605 | if (cgrp1->root == new_cgrp->root) { |
606 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
607 | return false; |
608 | } else { | |
5abb8855 | 609 | if (cgrp1 != cgrp2) |
7717f7ba PM |
610 | return false; |
611 | } | |
612 | } | |
613 | return true; | |
614 | } | |
615 | ||
b326f9d0 TH |
616 | /** |
617 | * find_existing_css_set - init css array and find the matching css_set | |
618 | * @old_cset: the css_set that we're using before the cgroup transition | |
619 | * @cgrp: the cgroup that we're moving into | |
620 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 621 | */ |
5abb8855 TH |
622 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
623 | struct cgroup *cgrp, | |
624 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 625 | { |
3dd06ffa | 626 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 627 | struct cgroup_subsys *ss; |
5abb8855 | 628 | struct css_set *cset; |
0ac801fe | 629 | unsigned long key; |
b326f9d0 | 630 | int i; |
817929ec | 631 | |
aae8aab4 BB |
632 | /* |
633 | * Build the set of subsystem state objects that we want to see in the | |
634 | * new css_set. while subsystems can change globally, the entries here | |
635 | * won't change, so no need for locking. | |
636 | */ | |
30159ec7 | 637 | for_each_subsys(ss, i) { |
f392e51c | 638 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
639 | /* |
640 | * @ss is in this hierarchy, so we want the | |
641 | * effective css from @cgrp. | |
642 | */ | |
643 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 644 | } else { |
aec3dfcb TH |
645 | /* |
646 | * @ss is not in this hierarchy, so we don't want | |
647 | * to change the css. | |
648 | */ | |
5abb8855 | 649 | template[i] = old_cset->subsys[i]; |
817929ec PM |
650 | } |
651 | } | |
652 | ||
0ac801fe | 653 | key = css_set_hash(template); |
5abb8855 TH |
654 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
655 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
656 | continue; |
657 | ||
658 | /* This css_set matches what we need */ | |
5abb8855 | 659 | return cset; |
472b1053 | 660 | } |
817929ec PM |
661 | |
662 | /* No existing cgroup group matched */ | |
663 | return NULL; | |
664 | } | |
665 | ||
69d0206c | 666 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 667 | { |
69d0206c | 668 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 669 | |
69d0206c TH |
670 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
671 | list_del(&link->cset_link); | |
36553434 LZ |
672 | kfree(link); |
673 | } | |
674 | } | |
675 | ||
69d0206c TH |
676 | /** |
677 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
678 | * @count: the number of links to allocate | |
679 | * @tmp_links: list_head the allocated links are put on | |
680 | * | |
681 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
682 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 683 | */ |
69d0206c | 684 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 685 | { |
69d0206c | 686 | struct cgrp_cset_link *link; |
817929ec | 687 | int i; |
69d0206c TH |
688 | |
689 | INIT_LIST_HEAD(tmp_links); | |
690 | ||
817929ec | 691 | for (i = 0; i < count; i++) { |
f4f4be2b | 692 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 693 | if (!link) { |
69d0206c | 694 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
695 | return -ENOMEM; |
696 | } | |
69d0206c | 697 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
698 | } |
699 | return 0; | |
700 | } | |
701 | ||
c12f65d4 LZ |
702 | /** |
703 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 704 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 705 | * @cset: the css_set to be linked |
c12f65d4 LZ |
706 | * @cgrp: the destination cgroup |
707 | */ | |
69d0206c TH |
708 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
709 | struct cgroup *cgrp) | |
c12f65d4 | 710 | { |
69d0206c | 711 | struct cgrp_cset_link *link; |
c12f65d4 | 712 | |
69d0206c | 713 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
714 | |
715 | if (cgroup_on_dfl(cgrp)) | |
716 | cset->dfl_cgrp = cgrp; | |
717 | ||
69d0206c TH |
718 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
719 | link->cset = cset; | |
7717f7ba | 720 | link->cgrp = cgrp; |
842b597e TH |
721 | |
722 | if (list_empty(&cgrp->cset_links)) | |
723 | cgroup_update_populated(cgrp, true); | |
69d0206c | 724 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 725 | |
7717f7ba PM |
726 | /* |
727 | * Always add links to the tail of the list so that the list | |
728 | * is sorted by order of hierarchy creation | |
729 | */ | |
69d0206c | 730 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
731 | } |
732 | ||
b326f9d0 TH |
733 | /** |
734 | * find_css_set - return a new css_set with one cgroup updated | |
735 | * @old_cset: the baseline css_set | |
736 | * @cgrp: the cgroup to be updated | |
737 | * | |
738 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
739 | * substituted into the appropriate hierarchy. | |
817929ec | 740 | */ |
5abb8855 TH |
741 | static struct css_set *find_css_set(struct css_set *old_cset, |
742 | struct cgroup *cgrp) | |
817929ec | 743 | { |
b326f9d0 | 744 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 745 | struct css_set *cset; |
69d0206c TH |
746 | struct list_head tmp_links; |
747 | struct cgrp_cset_link *link; | |
2d8f243a | 748 | struct cgroup_subsys *ss; |
0ac801fe | 749 | unsigned long key; |
2d8f243a | 750 | int ssid; |
472b1053 | 751 | |
b326f9d0 TH |
752 | lockdep_assert_held(&cgroup_mutex); |
753 | ||
817929ec PM |
754 | /* First see if we already have a cgroup group that matches |
755 | * the desired set */ | |
96d365e0 | 756 | down_read(&css_set_rwsem); |
5abb8855 TH |
757 | cset = find_existing_css_set(old_cset, cgrp, template); |
758 | if (cset) | |
759 | get_css_set(cset); | |
96d365e0 | 760 | up_read(&css_set_rwsem); |
817929ec | 761 | |
5abb8855 TH |
762 | if (cset) |
763 | return cset; | |
817929ec | 764 | |
f4f4be2b | 765 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 766 | if (!cset) |
817929ec PM |
767 | return NULL; |
768 | ||
69d0206c | 769 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 770 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 771 | kfree(cset); |
817929ec PM |
772 | return NULL; |
773 | } | |
774 | ||
5abb8855 | 775 | atomic_set(&cset->refcount, 1); |
69d0206c | 776 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 777 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 778 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 779 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 780 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 781 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
782 | |
783 | /* Copy the set of subsystem state objects generated in | |
784 | * find_existing_css_set() */ | |
5abb8855 | 785 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 786 | |
96d365e0 | 787 | down_write(&css_set_rwsem); |
817929ec | 788 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 789 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 790 | struct cgroup *c = link->cgrp; |
69d0206c | 791 | |
7717f7ba PM |
792 | if (c->root == cgrp->root) |
793 | c = cgrp; | |
69d0206c | 794 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 795 | } |
817929ec | 796 | |
69d0206c | 797 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 798 | |
817929ec | 799 | css_set_count++; |
472b1053 | 800 | |
2d8f243a | 801 | /* Add @cset to the hash table */ |
5abb8855 TH |
802 | key = css_set_hash(cset->subsys); |
803 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 804 | |
2d8f243a TH |
805 | for_each_subsys(ss, ssid) |
806 | list_add_tail(&cset->e_cset_node[ssid], | |
807 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
808 | ||
96d365e0 | 809 | up_write(&css_set_rwsem); |
817929ec | 810 | |
5abb8855 | 811 | return cset; |
b4f48b63 PM |
812 | } |
813 | ||
3dd06ffa | 814 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 815 | { |
3dd06ffa | 816 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 817 | |
3dd06ffa | 818 | return root_cgrp->root; |
2bd59d48 TH |
819 | } |
820 | ||
3dd06ffa | 821 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
822 | { |
823 | int id; | |
824 | ||
825 | lockdep_assert_held(&cgroup_mutex); | |
826 | ||
985ed670 | 827 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
828 | if (id < 0) |
829 | return id; | |
830 | ||
831 | root->hierarchy_id = id; | |
832 | return 0; | |
833 | } | |
834 | ||
3dd06ffa | 835 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
836 | { |
837 | lockdep_assert_held(&cgroup_mutex); | |
838 | ||
839 | if (root->hierarchy_id) { | |
840 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
841 | root->hierarchy_id = 0; | |
842 | } | |
843 | } | |
844 | ||
3dd06ffa | 845 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
846 | { |
847 | if (root) { | |
848 | /* hierarhcy ID shoulid already have been released */ | |
849 | WARN_ON_ONCE(root->hierarchy_id); | |
850 | ||
851 | idr_destroy(&root->cgroup_idr); | |
852 | kfree(root); | |
853 | } | |
854 | } | |
855 | ||
3dd06ffa | 856 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 857 | { |
3dd06ffa | 858 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 859 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 860 | |
2bd59d48 | 861 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 862 | |
776f02fa | 863 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 864 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 865 | |
f2e85d57 | 866 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 867 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 868 | |
7717f7ba | 869 | /* |
f2e85d57 TH |
870 | * Release all the links from cset_links to this hierarchy's |
871 | * root cgroup | |
7717f7ba | 872 | */ |
96d365e0 | 873 | down_write(&css_set_rwsem); |
f2e85d57 TH |
874 | |
875 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
876 | list_del(&link->cset_link); | |
877 | list_del(&link->cgrp_link); | |
878 | kfree(link); | |
879 | } | |
96d365e0 | 880 | up_write(&css_set_rwsem); |
f2e85d57 TH |
881 | |
882 | if (!list_empty(&root->root_list)) { | |
883 | list_del(&root->root_list); | |
884 | cgroup_root_count--; | |
885 | } | |
886 | ||
887 | cgroup_exit_root_id(root); | |
888 | ||
889 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 890 | |
2bd59d48 | 891 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
892 | cgroup_free_root(root); |
893 | } | |
894 | ||
ceb6a081 TH |
895 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
896 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 897 | struct cgroup_root *root) |
7717f7ba | 898 | { |
7717f7ba PM |
899 | struct cgroup *res = NULL; |
900 | ||
96d365e0 TH |
901 | lockdep_assert_held(&cgroup_mutex); |
902 | lockdep_assert_held(&css_set_rwsem); | |
903 | ||
5abb8855 | 904 | if (cset == &init_css_set) { |
3dd06ffa | 905 | res = &root->cgrp; |
7717f7ba | 906 | } else { |
69d0206c TH |
907 | struct cgrp_cset_link *link; |
908 | ||
909 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 910 | struct cgroup *c = link->cgrp; |
69d0206c | 911 | |
7717f7ba PM |
912 | if (c->root == root) { |
913 | res = c; | |
914 | break; | |
915 | } | |
916 | } | |
917 | } | |
96d365e0 | 918 | |
7717f7ba PM |
919 | BUG_ON(!res); |
920 | return res; | |
921 | } | |
922 | ||
ddbcc7e8 | 923 | /* |
ceb6a081 TH |
924 | * Return the cgroup for "task" from the given hierarchy. Must be |
925 | * called with cgroup_mutex and css_set_rwsem held. | |
926 | */ | |
927 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 928 | struct cgroup_root *root) |
ceb6a081 TH |
929 | { |
930 | /* | |
931 | * No need to lock the task - since we hold cgroup_mutex the | |
932 | * task can't change groups, so the only thing that can happen | |
933 | * is that it exits and its css is set back to init_css_set. | |
934 | */ | |
935 | return cset_cgroup_from_root(task_css_set(task), root); | |
936 | } | |
937 | ||
ddbcc7e8 | 938 | /* |
ddbcc7e8 PM |
939 | * A task must hold cgroup_mutex to modify cgroups. |
940 | * | |
941 | * Any task can increment and decrement the count field without lock. | |
942 | * So in general, code holding cgroup_mutex can't rely on the count | |
943 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 944 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
945 | * means that no tasks are currently attached, therefore there is no |
946 | * way a task attached to that cgroup can fork (the other way to | |
947 | * increment the count). So code holding cgroup_mutex can safely | |
948 | * assume that if the count is zero, it will stay zero. Similarly, if | |
949 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
950 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
951 | * needs that mutex. | |
952 | * | |
ddbcc7e8 PM |
953 | * A cgroup can only be deleted if both its 'count' of using tasks |
954 | * is zero, and its list of 'children' cgroups is empty. Since all | |
955 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 956 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 957 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 958 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
959 | * |
960 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 961 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
962 | */ |
963 | ||
69dfa00c | 964 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask); |
2bd59d48 | 965 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 966 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 967 | |
8d7e6fb0 TH |
968 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
969 | char *buf) | |
ddbcc7e8 | 970 | { |
8d7e6fb0 TH |
971 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
972 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
973 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
974 | cft->ss->name, cft->name); | |
975 | else | |
976 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
977 | return buf; | |
ddbcc7e8 PM |
978 | } |
979 | ||
f2e85d57 TH |
980 | /** |
981 | * cgroup_file_mode - deduce file mode of a control file | |
982 | * @cft: the control file in question | |
983 | * | |
984 | * returns cft->mode if ->mode is not 0 | |
985 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
986 | * returns S_IRUGO if it has only a read handler | |
987 | * returns S_IWUSR if it has only a write hander | |
988 | */ | |
989 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 990 | { |
f2e85d57 | 991 | umode_t mode = 0; |
65dff759 | 992 | |
f2e85d57 TH |
993 | if (cft->mode) |
994 | return cft->mode; | |
995 | ||
996 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
997 | mode |= S_IRUGO; | |
998 | ||
6770c64e | 999 | if (cft->write_u64 || cft->write_s64 || cft->write) |
f2e85d57 TH |
1000 | mode |= S_IWUSR; |
1001 | ||
1002 | return mode; | |
65dff759 LZ |
1003 | } |
1004 | ||
59f5296b | 1005 | static void cgroup_get(struct cgroup *cgrp) |
be445626 | 1006 | { |
2bd59d48 | 1007 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
9d755d33 | 1008 | css_get(&cgrp->self); |
be445626 LZ |
1009 | } |
1010 | ||
aa32362f LZ |
1011 | static bool cgroup_tryget(struct cgroup *cgrp) |
1012 | { | |
1013 | return css_tryget(&cgrp->self); | |
1014 | } | |
1015 | ||
59f5296b | 1016 | static void cgroup_put(struct cgroup *cgrp) |
be445626 | 1017 | { |
9d755d33 | 1018 | css_put(&cgrp->self); |
be445626 LZ |
1019 | } |
1020 | ||
af0ba678 | 1021 | /** |
0f060deb | 1022 | * cgroup_calc_child_subsys_mask - calculate child_subsys_mask |
af0ba678 | 1023 | * @cgrp: the target cgroup |
0f060deb | 1024 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1025 | * |
1026 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1027 | * enabled together through its ->depends_on mask. In such cases, more | |
1028 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1029 | * | |
0f060deb TH |
1030 | * This function calculates which subsystems need to be enabled if |
1031 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1032 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1033 | */ |
0f060deb TH |
1034 | static unsigned int cgroup_calc_child_subsys_mask(struct cgroup *cgrp, |
1035 | unsigned int subtree_control) | |
667c2491 | 1036 | { |
af0ba678 | 1037 | struct cgroup *parent = cgroup_parent(cgrp); |
0f060deb | 1038 | unsigned int cur_ss_mask = subtree_control; |
af0ba678 TH |
1039 | struct cgroup_subsys *ss; |
1040 | int ssid; | |
1041 | ||
1042 | lockdep_assert_held(&cgroup_mutex); | |
1043 | ||
0f060deb TH |
1044 | if (!cgroup_on_dfl(cgrp)) |
1045 | return cur_ss_mask; | |
af0ba678 TH |
1046 | |
1047 | while (true) { | |
1048 | unsigned int new_ss_mask = cur_ss_mask; | |
1049 | ||
1050 | for_each_subsys(ss, ssid) | |
1051 | if (cur_ss_mask & (1 << ssid)) | |
1052 | new_ss_mask |= ss->depends_on; | |
1053 | ||
1054 | /* | |
1055 | * Mask out subsystems which aren't available. This can | |
1056 | * happen only if some depended-upon subsystems were bound | |
1057 | * to non-default hierarchies. | |
1058 | */ | |
1059 | if (parent) | |
1060 | new_ss_mask &= parent->child_subsys_mask; | |
1061 | else | |
1062 | new_ss_mask &= cgrp->root->subsys_mask; | |
1063 | ||
1064 | if (new_ss_mask == cur_ss_mask) | |
1065 | break; | |
1066 | cur_ss_mask = new_ss_mask; | |
1067 | } | |
1068 | ||
0f060deb TH |
1069 | return cur_ss_mask; |
1070 | } | |
1071 | ||
1072 | /** | |
1073 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1074 | * @cgrp: the target cgroup | |
1075 | * | |
1076 | * Update @cgrp->child_subsys_mask according to the current | |
1077 | * @cgrp->subtree_control using cgroup_calc_child_subsys_mask(). | |
1078 | */ | |
1079 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) | |
1080 | { | |
1081 | cgrp->child_subsys_mask = | |
1082 | cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1083 | } |
1084 | ||
a9746d8d TH |
1085 | /** |
1086 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1087 | * @kn: the kernfs_node being serviced | |
1088 | * | |
1089 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1090 | * the method finishes if locking succeeded. Note that once this function | |
1091 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1092 | * inaccessible any time. If the caller intends to continue to access the | |
1093 | * cgroup, it should pin it before invoking this function. | |
1094 | */ | |
1095 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1096 | { |
a9746d8d TH |
1097 | struct cgroup *cgrp; |
1098 | ||
1099 | if (kernfs_type(kn) == KERNFS_DIR) | |
1100 | cgrp = kn->priv; | |
1101 | else | |
1102 | cgrp = kn->parent->priv; | |
1103 | ||
1104 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1105 | |
1106 | kernfs_unbreak_active_protection(kn); | |
1107 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1108 | } |
1109 | ||
a9746d8d TH |
1110 | /** |
1111 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1112 | * @kn: the kernfs_node being serviced | |
1113 | * | |
1114 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1115 | * @kn. It breaks the active protection, performs cgroup locking and | |
1116 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1117 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1118 | * matching cgroup_kn_unlock() invocation. | |
1119 | * | |
1120 | * Any cgroup kernfs method implementation which requires locking the | |
1121 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1122 | * locking under kernfs active protection and allows all kernfs operations | |
1123 | * including self-removal. | |
1124 | */ | |
1125 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1126 | { |
a9746d8d TH |
1127 | struct cgroup *cgrp; |
1128 | ||
1129 | if (kernfs_type(kn) == KERNFS_DIR) | |
1130 | cgrp = kn->priv; | |
1131 | else | |
1132 | cgrp = kn->parent->priv; | |
05ef1d7c | 1133 | |
2739d3cc | 1134 | /* |
01f6474c | 1135 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1136 | * active_ref. cgroup liveliness check alone provides enough |
1137 | * protection against removal. Ensure @cgrp stays accessible and | |
1138 | * break the active_ref protection. | |
2739d3cc | 1139 | */ |
aa32362f LZ |
1140 | if (!cgroup_tryget(cgrp)) |
1141 | return NULL; | |
a9746d8d TH |
1142 | kernfs_break_active_protection(kn); |
1143 | ||
2bd59d48 | 1144 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1145 | |
a9746d8d TH |
1146 | if (!cgroup_is_dead(cgrp)) |
1147 | return cgrp; | |
1148 | ||
1149 | cgroup_kn_unlock(kn); | |
1150 | return NULL; | |
ddbcc7e8 | 1151 | } |
05ef1d7c | 1152 | |
2739d3cc | 1153 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1154 | { |
2bd59d48 | 1155 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1156 | |
01f6474c | 1157 | lockdep_assert_held(&cgroup_mutex); |
2bd59d48 | 1158 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1159 | } |
1160 | ||
13af07df | 1161 | /** |
628f7cd4 | 1162 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1163 | * @cgrp: target cgroup |
13af07df AR |
1164 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1165 | */ | |
69dfa00c | 1166 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
05ef1d7c | 1167 | { |
13af07df | 1168 | struct cgroup_subsys *ss; |
b420ba7d | 1169 | int i; |
05ef1d7c | 1170 | |
b420ba7d | 1171 | for_each_subsys(ss, i) { |
0adb0704 | 1172 | struct cftype *cfts; |
b420ba7d | 1173 | |
69dfa00c | 1174 | if (!(subsys_mask & (1 << i))) |
13af07df | 1175 | continue; |
0adb0704 TH |
1176 | list_for_each_entry(cfts, &ss->cfts, node) |
1177 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1178 | } |
ddbcc7e8 PM |
1179 | } |
1180 | ||
69dfa00c | 1181 | static int rebind_subsystems(struct cgroup_root *dst_root, unsigned int ss_mask) |
ddbcc7e8 | 1182 | { |
30159ec7 | 1183 | struct cgroup_subsys *ss; |
5533e011 | 1184 | unsigned int tmp_ss_mask; |
2d8f243a | 1185 | int ssid, i, ret; |
ddbcc7e8 | 1186 | |
ace2bee8 | 1187 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1188 | |
5df36032 TH |
1189 | for_each_subsys(ss, ssid) { |
1190 | if (!(ss_mask & (1 << ssid))) | |
1191 | continue; | |
aae8aab4 | 1192 | |
7fd8c565 TH |
1193 | /* if @ss has non-root csses attached to it, can't move */ |
1194 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1195 | return -EBUSY; |
1d5be6b2 | 1196 | |
5df36032 | 1197 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1198 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1199 | return -EBUSY; |
ddbcc7e8 PM |
1200 | } |
1201 | ||
5533e011 TH |
1202 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1203 | tmp_ss_mask = ss_mask; | |
1204 | if (dst_root == &cgrp_dfl_root) | |
1205 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1206 | ||
1207 | ret = cgroup_populate_dir(&dst_root->cgrp, tmp_ss_mask); | |
a2dd4247 TH |
1208 | if (ret) { |
1209 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1210 | return ret; |
ddbcc7e8 | 1211 | |
a2dd4247 TH |
1212 | /* |
1213 | * Rebinding back to the default root is not allowed to | |
1214 | * fail. Using both default and non-default roots should | |
1215 | * be rare. Moving subsystems back and forth even more so. | |
1216 | * Just warn about it and continue. | |
1217 | */ | |
1218 | if (cgrp_dfl_root_visible) { | |
69dfa00c | 1219 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", |
a2a1f9ea | 1220 | ret, ss_mask); |
ed3d261b | 1221 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); |
a2dd4247 | 1222 | } |
5df36032 | 1223 | } |
3126121f TH |
1224 | |
1225 | /* | |
1226 | * Nothing can fail from this point on. Remove files for the | |
1227 | * removed subsystems and rebind each subsystem. | |
1228 | */ | |
5df36032 | 1229 | for_each_subsys(ss, ssid) |
a2dd4247 | 1230 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1231 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
a8a648c4 | 1232 | |
5df36032 | 1233 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1234 | struct cgroup_root *src_root; |
5df36032 | 1235 | struct cgroup_subsys_state *css; |
2d8f243a | 1236 | struct css_set *cset; |
a8a648c4 | 1237 | |
5df36032 TH |
1238 | if (!(ss_mask & (1 << ssid))) |
1239 | continue; | |
a8a648c4 | 1240 | |
5df36032 | 1241 | src_root = ss->root; |
3dd06ffa | 1242 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1243 | |
3dd06ffa | 1244 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1245 | |
3dd06ffa TH |
1246 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1247 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1248 | ss->root = dst_root; |
3dd06ffa | 1249 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1250 | |
2d8f243a TH |
1251 | down_write(&css_set_rwsem); |
1252 | hash_for_each(css_set_table, i, cset, hlist) | |
1253 | list_move_tail(&cset->e_cset_node[ss->id], | |
1254 | &dst_root->cgrp.e_csets[ss->id]); | |
1255 | up_write(&css_set_rwsem); | |
1256 | ||
f392e51c | 1257 | src_root->subsys_mask &= ~(1 << ssid); |
667c2491 TH |
1258 | src_root->cgrp.subtree_control &= ~(1 << ssid); |
1259 | cgroup_refresh_child_subsys_mask(&src_root->cgrp); | |
f392e51c | 1260 | |
bd53d617 | 1261 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1262 | dst_root->subsys_mask |= 1 << ssid; |
667c2491 TH |
1263 | if (dst_root != &cgrp_dfl_root) { |
1264 | dst_root->cgrp.subtree_control |= 1 << ssid; | |
1265 | cgroup_refresh_child_subsys_mask(&dst_root->cgrp); | |
1266 | } | |
a8a648c4 | 1267 | |
5df36032 TH |
1268 | if (ss->bind) |
1269 | ss->bind(css); | |
ddbcc7e8 | 1270 | } |
ddbcc7e8 | 1271 | |
a2dd4247 | 1272 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1273 | return 0; |
1274 | } | |
1275 | ||
2bd59d48 TH |
1276 | static int cgroup_show_options(struct seq_file *seq, |
1277 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1278 | { |
3dd06ffa | 1279 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1280 | struct cgroup_subsys *ss; |
b85d2040 | 1281 | int ssid; |
ddbcc7e8 | 1282 | |
b85d2040 | 1283 | for_each_subsys(ss, ssid) |
f392e51c | 1284 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1285 | seq_printf(seq, ",%s", ss->name); |
93438629 | 1286 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1287 | seq_puts(seq, ",noprefix"); |
93438629 | 1288 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1289 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1290 | |
1291 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1292 | if (strlen(root->release_agent_path)) |
1293 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1294 | spin_unlock(&release_agent_path_lock); |
1295 | ||
3dd06ffa | 1296 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1297 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1298 | if (strlen(root->name)) |
1299 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1300 | return 0; |
1301 | } | |
1302 | ||
1303 | struct cgroup_sb_opts { | |
69dfa00c TH |
1304 | unsigned int subsys_mask; |
1305 | unsigned int flags; | |
81a6a5cd | 1306 | char *release_agent; |
2260e7fc | 1307 | bool cpuset_clone_children; |
c6d57f33 | 1308 | char *name; |
2c6ab6d2 PM |
1309 | /* User explicitly requested empty subsystem */ |
1310 | bool none; | |
ddbcc7e8 PM |
1311 | }; |
1312 | ||
cf5d5941 | 1313 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1314 | { |
32a8cf23 DL |
1315 | char *token, *o = data; |
1316 | bool all_ss = false, one_ss = false; | |
69dfa00c | 1317 | unsigned int mask = -1U; |
30159ec7 | 1318 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1319 | int nr_opts = 0; |
30159ec7 | 1320 | int i; |
f9ab5b5b LZ |
1321 | |
1322 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1323 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1324 | #endif |
ddbcc7e8 | 1325 | |
c6d57f33 | 1326 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1327 | |
1328 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1329 | nr_opts++; |
1330 | ||
ddbcc7e8 PM |
1331 | if (!*token) |
1332 | return -EINVAL; | |
32a8cf23 | 1333 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1334 | /* Explicitly have no subsystems */ |
1335 | opts->none = true; | |
32a8cf23 DL |
1336 | continue; |
1337 | } | |
1338 | if (!strcmp(token, "all")) { | |
1339 | /* Mutually exclusive option 'all' + subsystem name */ | |
1340 | if (one_ss) | |
1341 | return -EINVAL; | |
1342 | all_ss = true; | |
1343 | continue; | |
1344 | } | |
873fe09e TH |
1345 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1346 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1347 | continue; | |
1348 | } | |
32a8cf23 | 1349 | if (!strcmp(token, "noprefix")) { |
93438629 | 1350 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1351 | continue; |
1352 | } | |
1353 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1354 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1355 | continue; |
1356 | } | |
03b1cde6 | 1357 | if (!strcmp(token, "xattr")) { |
93438629 | 1358 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1359 | continue; |
1360 | } | |
32a8cf23 | 1361 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1362 | /* Specifying two release agents is forbidden */ |
1363 | if (opts->release_agent) | |
1364 | return -EINVAL; | |
c6d57f33 | 1365 | opts->release_agent = |
e400c285 | 1366 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1367 | if (!opts->release_agent) |
1368 | return -ENOMEM; | |
32a8cf23 DL |
1369 | continue; |
1370 | } | |
1371 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1372 | const char *name = token + 5; |
1373 | /* Can't specify an empty name */ | |
1374 | if (!strlen(name)) | |
1375 | return -EINVAL; | |
1376 | /* Must match [\w.-]+ */ | |
1377 | for (i = 0; i < strlen(name); i++) { | |
1378 | char c = name[i]; | |
1379 | if (isalnum(c)) | |
1380 | continue; | |
1381 | if ((c == '.') || (c == '-') || (c == '_')) | |
1382 | continue; | |
1383 | return -EINVAL; | |
1384 | } | |
1385 | /* Specifying two names is forbidden */ | |
1386 | if (opts->name) | |
1387 | return -EINVAL; | |
1388 | opts->name = kstrndup(name, | |
e400c285 | 1389 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1390 | GFP_KERNEL); |
1391 | if (!opts->name) | |
1392 | return -ENOMEM; | |
32a8cf23 DL |
1393 | |
1394 | continue; | |
1395 | } | |
1396 | ||
30159ec7 | 1397 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1398 | if (strcmp(token, ss->name)) |
1399 | continue; | |
1400 | if (ss->disabled) | |
1401 | continue; | |
1402 | ||
1403 | /* Mutually exclusive option 'all' + subsystem name */ | |
1404 | if (all_ss) | |
1405 | return -EINVAL; | |
69dfa00c | 1406 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1407 | one_ss = true; |
1408 | ||
1409 | break; | |
1410 | } | |
1411 | if (i == CGROUP_SUBSYS_COUNT) | |
1412 | return -ENOENT; | |
1413 | } | |
1414 | ||
873fe09e | 1415 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1416 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
7b9a6ba5 TH |
1417 | if (nr_opts != 1) { |
1418 | pr_err("sane_behavior: no other mount options allowed\n"); | |
873fe09e TH |
1419 | return -EINVAL; |
1420 | } | |
7b9a6ba5 | 1421 | return 0; |
873fe09e TH |
1422 | } |
1423 | ||
7b9a6ba5 TH |
1424 | /* |
1425 | * If the 'all' option was specified select all the subsystems, | |
1426 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1427 | * not specified, let's default to 'all' | |
1428 | */ | |
1429 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1430 | for_each_subsys(ss, i) | |
1431 | if (!ss->disabled) | |
1432 | opts->subsys_mask |= (1 << i); | |
1433 | ||
1434 | /* | |
1435 | * We either have to specify by name or by subsystems. (So all | |
1436 | * empty hierarchies must have a name). | |
1437 | */ | |
1438 | if (!opts->subsys_mask && !opts->name) | |
1439 | return -EINVAL; | |
1440 | ||
f9ab5b5b LZ |
1441 | /* |
1442 | * Option noprefix was introduced just for backward compatibility | |
1443 | * with the old cpuset, so we allow noprefix only if mounting just | |
1444 | * the cpuset subsystem. | |
1445 | */ | |
93438629 | 1446 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1447 | return -EINVAL; |
1448 | ||
2c6ab6d2 | 1449 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1450 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1451 | return -EINVAL; |
1452 | ||
ddbcc7e8 PM |
1453 | return 0; |
1454 | } | |
1455 | ||
2bd59d48 | 1456 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1457 | { |
1458 | int ret = 0; | |
3dd06ffa | 1459 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1460 | struct cgroup_sb_opts opts; |
69dfa00c | 1461 | unsigned int added_mask, removed_mask; |
ddbcc7e8 | 1462 | |
aa6ec29b TH |
1463 | if (root == &cgrp_dfl_root) { |
1464 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1465 | return -EINVAL; |
1466 | } | |
1467 | ||
ddbcc7e8 PM |
1468 | mutex_lock(&cgroup_mutex); |
1469 | ||
1470 | /* See what subsystems are wanted */ | |
1471 | ret = parse_cgroupfs_options(data, &opts); | |
1472 | if (ret) | |
1473 | goto out_unlock; | |
1474 | ||
f392e51c | 1475 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1476 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1477 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1478 | |
f392e51c TH |
1479 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1480 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1481 | |
cf5d5941 | 1482 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1483 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1484 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1485 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1486 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1487 | ret = -EINVAL; |
1488 | goto out_unlock; | |
1489 | } | |
1490 | ||
f172e67c | 1491 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1492 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1493 | ret = -EBUSY; |
0670e08b | 1494 | goto out_unlock; |
cf5d5941 | 1495 | } |
ddbcc7e8 | 1496 | |
5df36032 | 1497 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1498 | if (ret) |
0670e08b | 1499 | goto out_unlock; |
ddbcc7e8 | 1500 | |
3dd06ffa | 1501 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1502 | |
69e943b7 TH |
1503 | if (opts.release_agent) { |
1504 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1505 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1506 | spin_unlock(&release_agent_path_lock); |
1507 | } | |
ddbcc7e8 | 1508 | out_unlock: |
66bdc9cf | 1509 | kfree(opts.release_agent); |
c6d57f33 | 1510 | kfree(opts.name); |
ddbcc7e8 | 1511 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1512 | return ret; |
1513 | } | |
1514 | ||
afeb0f9f TH |
1515 | /* |
1516 | * To reduce the fork() overhead for systems that are not actually using | |
1517 | * their cgroups capability, we don't maintain the lists running through | |
1518 | * each css_set to its tasks until we see the list actually used - in other | |
1519 | * words after the first mount. | |
1520 | */ | |
1521 | static bool use_task_css_set_links __read_mostly; | |
1522 | ||
1523 | static void cgroup_enable_task_cg_lists(void) | |
1524 | { | |
1525 | struct task_struct *p, *g; | |
1526 | ||
96d365e0 | 1527 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1528 | |
1529 | if (use_task_css_set_links) | |
1530 | goto out_unlock; | |
1531 | ||
1532 | use_task_css_set_links = true; | |
1533 | ||
1534 | /* | |
1535 | * We need tasklist_lock because RCU is not safe against | |
1536 | * while_each_thread(). Besides, a forking task that has passed | |
1537 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1538 | * is not guaranteed to have its child immediately visible in the | |
1539 | * tasklist if we walk through it with RCU. | |
1540 | */ | |
1541 | read_lock(&tasklist_lock); | |
1542 | do_each_thread(g, p) { | |
afeb0f9f TH |
1543 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1544 | task_css_set(p) != &init_css_set); | |
1545 | ||
1546 | /* | |
1547 | * We should check if the process is exiting, otherwise | |
1548 | * it will race with cgroup_exit() in that the list | |
1549 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1550 | * Do it while holding siglock so that we don't end up |
1551 | * racing against cgroup_exit(). | |
afeb0f9f | 1552 | */ |
f153ad11 | 1553 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1554 | if (!(p->flags & PF_EXITING)) { |
1555 | struct css_set *cset = task_css_set(p); | |
1556 | ||
1557 | list_add(&p->cg_list, &cset->tasks); | |
1558 | get_css_set(cset); | |
1559 | } | |
f153ad11 | 1560 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1561 | } while_each_thread(g, p); |
1562 | read_unlock(&tasklist_lock); | |
1563 | out_unlock: | |
96d365e0 | 1564 | up_write(&css_set_rwsem); |
afeb0f9f | 1565 | } |
ddbcc7e8 | 1566 | |
cc31edce PM |
1567 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1568 | { | |
2d8f243a TH |
1569 | struct cgroup_subsys *ss; |
1570 | int ssid; | |
1571 | ||
d5c419b6 TH |
1572 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1573 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1574 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1575 | INIT_LIST_HEAD(&cgrp->pidlists); |
1576 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1577 | cgrp->self.cgroup = cgrp; |
184faf32 | 1578 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1579 | |
1580 | for_each_subsys(ss, ssid) | |
1581 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1582 | |
1583 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1584 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1585 | } |
c6d57f33 | 1586 | |
3dd06ffa | 1587 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1588 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1589 | { |
3dd06ffa | 1590 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1591 | |
ddbcc7e8 | 1592 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1593 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1594 | cgrp->root = root; |
cc31edce | 1595 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1596 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1597 | |
c6d57f33 PM |
1598 | root->flags = opts->flags; |
1599 | if (opts->release_agent) | |
1600 | strcpy(root->release_agent_path, opts->release_agent); | |
1601 | if (opts->name) | |
1602 | strcpy(root->name, opts->name); | |
2260e7fc | 1603 | if (opts->cpuset_clone_children) |
3dd06ffa | 1604 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1605 | } |
1606 | ||
69dfa00c | 1607 | static int cgroup_setup_root(struct cgroup_root *root, unsigned int ss_mask) |
2c6ab6d2 | 1608 | { |
d427dfeb | 1609 | LIST_HEAD(tmp_links); |
3dd06ffa | 1610 | struct cgroup *root_cgrp = &root->cgrp; |
a14c6874 | 1611 | struct cftype *base_files; |
d427dfeb | 1612 | struct css_set *cset; |
d427dfeb | 1613 | int i, ret; |
2c6ab6d2 | 1614 | |
d427dfeb | 1615 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1616 | |
6fa4918d | 1617 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_NOWAIT); |
d427dfeb | 1618 | if (ret < 0) |
2bd59d48 | 1619 | goto out; |
d427dfeb | 1620 | root_cgrp->id = ret; |
c6d57f33 | 1621 | |
2aad2a86 TH |
1622 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1623 | GFP_KERNEL); | |
9d755d33 TH |
1624 | if (ret) |
1625 | goto out; | |
1626 | ||
d427dfeb | 1627 | /* |
96d365e0 | 1628 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1629 | * but that's OK - it can only be increased by someone holding |
1630 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1631 | * have some link structures left over | |
1632 | */ | |
1633 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1634 | if (ret) | |
9d755d33 | 1635 | goto cancel_ref; |
ddbcc7e8 | 1636 | |
985ed670 | 1637 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1638 | if (ret) |
9d755d33 | 1639 | goto cancel_ref; |
ddbcc7e8 | 1640 | |
2bd59d48 TH |
1641 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1642 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1643 | root_cgrp); | |
1644 | if (IS_ERR(root->kf_root)) { | |
1645 | ret = PTR_ERR(root->kf_root); | |
1646 | goto exit_root_id; | |
1647 | } | |
1648 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1649 | |
a14c6874 TH |
1650 | if (root == &cgrp_dfl_root) |
1651 | base_files = cgroup_dfl_base_files; | |
1652 | else | |
1653 | base_files = cgroup_legacy_base_files; | |
1654 | ||
1655 | ret = cgroup_addrm_files(root_cgrp, base_files, true); | |
d427dfeb | 1656 | if (ret) |
2bd59d48 | 1657 | goto destroy_root; |
ddbcc7e8 | 1658 | |
5df36032 | 1659 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1660 | if (ret) |
2bd59d48 | 1661 | goto destroy_root; |
ddbcc7e8 | 1662 | |
d427dfeb TH |
1663 | /* |
1664 | * There must be no failure case after here, since rebinding takes | |
1665 | * care of subsystems' refcounts, which are explicitly dropped in | |
1666 | * the failure exit path. | |
1667 | */ | |
1668 | list_add(&root->root_list, &cgroup_roots); | |
1669 | cgroup_root_count++; | |
0df6a63f | 1670 | |
d427dfeb | 1671 | /* |
3dd06ffa | 1672 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1673 | * objects. |
1674 | */ | |
96d365e0 | 1675 | down_write(&css_set_rwsem); |
d427dfeb TH |
1676 | hash_for_each(css_set_table, i, cset, hlist) |
1677 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1678 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1679 | |
d5c419b6 | 1680 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1681 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1682 | |
2bd59d48 | 1683 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1684 | ret = 0; |
2bd59d48 | 1685 | goto out; |
d427dfeb | 1686 | |
2bd59d48 TH |
1687 | destroy_root: |
1688 | kernfs_destroy_root(root->kf_root); | |
1689 | root->kf_root = NULL; | |
1690 | exit_root_id: | |
d427dfeb | 1691 | cgroup_exit_root_id(root); |
9d755d33 | 1692 | cancel_ref: |
9a1049da | 1693 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1694 | out: |
d427dfeb TH |
1695 | free_cgrp_cset_links(&tmp_links); |
1696 | return ret; | |
ddbcc7e8 PM |
1697 | } |
1698 | ||
f7e83571 | 1699 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1700 | int flags, const char *unused_dev_name, |
f7e83571 | 1701 | void *data) |
ddbcc7e8 | 1702 | { |
3a32bd72 | 1703 | struct super_block *pinned_sb = NULL; |
970317aa | 1704 | struct cgroup_subsys *ss; |
3dd06ffa | 1705 | struct cgroup_root *root; |
ddbcc7e8 | 1706 | struct cgroup_sb_opts opts; |
2bd59d48 | 1707 | struct dentry *dentry; |
8e30e2b8 | 1708 | int ret; |
970317aa | 1709 | int i; |
c6b3d5bc | 1710 | bool new_sb; |
ddbcc7e8 | 1711 | |
56fde9e0 TH |
1712 | /* |
1713 | * The first time anyone tries to mount a cgroup, enable the list | |
1714 | * linking each css_set to its tasks and fix up all existing tasks. | |
1715 | */ | |
1716 | if (!use_task_css_set_links) | |
1717 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1718 | |
aae8aab4 | 1719 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1720 | |
1721 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1722 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1723 | if (ret) |
8e30e2b8 | 1724 | goto out_unlock; |
a015edd2 | 1725 | |
2bd59d48 | 1726 | /* look for a matching existing root */ |
7b9a6ba5 | 1727 | if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2dd4247 TH |
1728 | cgrp_dfl_root_visible = true; |
1729 | root = &cgrp_dfl_root; | |
1730 | cgroup_get(&root->cgrp); | |
1731 | ret = 0; | |
1732 | goto out_unlock; | |
ddbcc7e8 PM |
1733 | } |
1734 | ||
970317aa LZ |
1735 | /* |
1736 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1737 | * still be dying after the previous unmount. Let's drain the | |
1738 | * dying subsystems. We just need to ensure that the ones | |
1739 | * unmounted previously finish dying and don't care about new ones | |
1740 | * starting. Testing ref liveliness is good enough. | |
1741 | */ | |
1742 | for_each_subsys(ss, i) { | |
1743 | if (!(opts.subsys_mask & (1 << i)) || | |
1744 | ss->root == &cgrp_dfl_root) | |
1745 | continue; | |
1746 | ||
1747 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1748 | mutex_unlock(&cgroup_mutex); | |
1749 | msleep(10); | |
1750 | ret = restart_syscall(); | |
1751 | goto out_free; | |
1752 | } | |
1753 | cgroup_put(&ss->root->cgrp); | |
1754 | } | |
1755 | ||
985ed670 | 1756 | for_each_root(root) { |
2bd59d48 | 1757 | bool name_match = false; |
3126121f | 1758 | |
3dd06ffa | 1759 | if (root == &cgrp_dfl_root) |
985ed670 | 1760 | continue; |
3126121f | 1761 | |
cf5d5941 | 1762 | /* |
2bd59d48 TH |
1763 | * If we asked for a name then it must match. Also, if |
1764 | * name matches but sybsys_mask doesn't, we should fail. | |
1765 | * Remember whether name matched. | |
cf5d5941 | 1766 | */ |
2bd59d48 TH |
1767 | if (opts.name) { |
1768 | if (strcmp(opts.name, root->name)) | |
1769 | continue; | |
1770 | name_match = true; | |
1771 | } | |
ddbcc7e8 | 1772 | |
c6d57f33 | 1773 | /* |
2bd59d48 TH |
1774 | * If we asked for subsystems (or explicitly for no |
1775 | * subsystems) then they must match. | |
c6d57f33 | 1776 | */ |
2bd59d48 | 1777 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1778 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1779 | if (!name_match) |
1780 | continue; | |
1781 | ret = -EBUSY; | |
1782 | goto out_unlock; | |
1783 | } | |
873fe09e | 1784 | |
7b9a6ba5 TH |
1785 | if (root->flags ^ opts.flags) |
1786 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 1787 | |
776f02fa | 1788 | /* |
3a32bd72 LZ |
1789 | * We want to reuse @root whose lifetime is governed by its |
1790 | * ->cgrp. Let's check whether @root is alive and keep it | |
1791 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1792 | * want to block it by pinning the sb so that @root doesn't | |
1793 | * get killed before mount is complete. | |
1794 | * | |
1795 | * With the sb pinned, tryget_live can reliably indicate | |
1796 | * whether @root can be reused. If it's being killed, | |
1797 | * drain it. We can use wait_queue for the wait but this | |
1798 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 1799 | */ |
3a32bd72 LZ |
1800 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
1801 | if (IS_ERR(pinned_sb) || | |
1802 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 1803 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
1804 | if (!IS_ERR_OR_NULL(pinned_sb)) |
1805 | deactivate_super(pinned_sb); | |
776f02fa | 1806 | msleep(10); |
a015edd2 TH |
1807 | ret = restart_syscall(); |
1808 | goto out_free; | |
776f02fa | 1809 | } |
ddbcc7e8 | 1810 | |
776f02fa | 1811 | ret = 0; |
2bd59d48 | 1812 | goto out_unlock; |
ddbcc7e8 | 1813 | } |
ddbcc7e8 | 1814 | |
817929ec | 1815 | /* |
172a2c06 TH |
1816 | * No such thing, create a new one. name= matching without subsys |
1817 | * specification is allowed for already existing hierarchies but we | |
1818 | * can't create new one without subsys specification. | |
817929ec | 1819 | */ |
172a2c06 TH |
1820 | if (!opts.subsys_mask && !opts.none) { |
1821 | ret = -EINVAL; | |
1822 | goto out_unlock; | |
817929ec | 1823 | } |
817929ec | 1824 | |
172a2c06 TH |
1825 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1826 | if (!root) { | |
1827 | ret = -ENOMEM; | |
2bd59d48 | 1828 | goto out_unlock; |
839ec545 | 1829 | } |
e5f6a860 | 1830 | |
172a2c06 TH |
1831 | init_cgroup_root(root, &opts); |
1832 | ||
35585573 | 1833 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1834 | if (ret) |
1835 | cgroup_free_root(root); | |
fa3ca07e | 1836 | |
8e30e2b8 | 1837 | out_unlock: |
ddbcc7e8 | 1838 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 1839 | out_free: |
c6d57f33 PM |
1840 | kfree(opts.release_agent); |
1841 | kfree(opts.name); | |
03b1cde6 | 1842 | |
2bd59d48 | 1843 | if (ret) |
8e30e2b8 | 1844 | return ERR_PTR(ret); |
2bd59d48 | 1845 | |
c9482a5b JZ |
1846 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
1847 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 1848 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 1849 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
1850 | |
1851 | /* | |
1852 | * If @pinned_sb, we're reusing an existing root and holding an | |
1853 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1854 | */ | |
1855 | if (pinned_sb) { | |
1856 | WARN_ON(new_sb); | |
1857 | deactivate_super(pinned_sb); | |
1858 | } | |
1859 | ||
2bd59d48 TH |
1860 | return dentry; |
1861 | } | |
1862 | ||
1863 | static void cgroup_kill_sb(struct super_block *sb) | |
1864 | { | |
1865 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1866 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1867 | |
9d755d33 TH |
1868 | /* |
1869 | * If @root doesn't have any mounts or children, start killing it. | |
1870 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
1871 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
1872 | * |
1873 | * And don't kill the default root. | |
9d755d33 | 1874 | */ |
1f779fb2 LZ |
1875 | if (css_has_online_children(&root->cgrp.self) || |
1876 | root == &cgrp_dfl_root) | |
9d755d33 TH |
1877 | cgroup_put(&root->cgrp); |
1878 | else | |
1879 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
1880 | ||
2bd59d48 | 1881 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1882 | } |
1883 | ||
1884 | static struct file_system_type cgroup_fs_type = { | |
1885 | .name = "cgroup", | |
f7e83571 | 1886 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1887 | .kill_sb = cgroup_kill_sb, |
1888 | }; | |
1889 | ||
676db4af GKH |
1890 | static struct kobject *cgroup_kobj; |
1891 | ||
857a2beb | 1892 | /** |
913ffdb5 | 1893 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1894 | * @task: target task |
857a2beb TH |
1895 | * @buf: the buffer to write the path into |
1896 | * @buflen: the length of the buffer | |
1897 | * | |
913ffdb5 TH |
1898 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1899 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1900 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1901 | * cgroup controller callbacks. | |
1902 | * | |
e61734c5 | 1903 | * Return value is the same as kernfs_path(). |
857a2beb | 1904 | */ |
e61734c5 | 1905 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1906 | { |
3dd06ffa | 1907 | struct cgroup_root *root; |
913ffdb5 | 1908 | struct cgroup *cgrp; |
e61734c5 TH |
1909 | int hierarchy_id = 1; |
1910 | char *path = NULL; | |
857a2beb TH |
1911 | |
1912 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1913 | down_read(&css_set_rwsem); |
857a2beb | 1914 | |
913ffdb5 TH |
1915 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1916 | ||
857a2beb TH |
1917 | if (root) { |
1918 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1919 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1920 | } else { |
1921 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1922 | if (strlcpy(buf, "/", buflen) < buflen) |
1923 | path = buf; | |
857a2beb TH |
1924 | } |
1925 | ||
96d365e0 | 1926 | up_read(&css_set_rwsem); |
857a2beb | 1927 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1928 | return path; |
857a2beb | 1929 | } |
913ffdb5 | 1930 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1931 | |
b3dc094e | 1932 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1933 | struct cgroup_taskset { |
b3dc094e TH |
1934 | /* the src and dst cset list running through cset->mg_node */ |
1935 | struct list_head src_csets; | |
1936 | struct list_head dst_csets; | |
1937 | ||
1938 | /* | |
1939 | * Fields for cgroup_taskset_*() iteration. | |
1940 | * | |
1941 | * Before migration is committed, the target migration tasks are on | |
1942 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1943 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1944 | * or ->dst_csets depending on whether migration is committed. | |
1945 | * | |
1946 | * ->cur_csets and ->cur_task point to the current task position | |
1947 | * during iteration. | |
1948 | */ | |
1949 | struct list_head *csets; | |
1950 | struct css_set *cur_cset; | |
1951 | struct task_struct *cur_task; | |
2f7ee569 TH |
1952 | }; |
1953 | ||
1954 | /** | |
1955 | * cgroup_taskset_first - reset taskset and return the first task | |
1956 | * @tset: taskset of interest | |
1957 | * | |
1958 | * @tset iteration is initialized and the first task is returned. | |
1959 | */ | |
1960 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1961 | { | |
b3dc094e TH |
1962 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1963 | tset->cur_task = NULL; | |
1964 | ||
1965 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1966 | } |
2f7ee569 TH |
1967 | |
1968 | /** | |
1969 | * cgroup_taskset_next - iterate to the next task in taskset | |
1970 | * @tset: taskset of interest | |
1971 | * | |
1972 | * Return the next task in @tset. Iteration must have been initialized | |
1973 | * with cgroup_taskset_first(). | |
1974 | */ | |
1975 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1976 | { | |
b3dc094e TH |
1977 | struct css_set *cset = tset->cur_cset; |
1978 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1979 | |
b3dc094e TH |
1980 | while (&cset->mg_node != tset->csets) { |
1981 | if (!task) | |
1982 | task = list_first_entry(&cset->mg_tasks, | |
1983 | struct task_struct, cg_list); | |
1984 | else | |
1985 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1986 | |
b3dc094e TH |
1987 | if (&task->cg_list != &cset->mg_tasks) { |
1988 | tset->cur_cset = cset; | |
1989 | tset->cur_task = task; | |
1990 | return task; | |
1991 | } | |
2f7ee569 | 1992 | |
b3dc094e TH |
1993 | cset = list_next_entry(cset, mg_node); |
1994 | task = NULL; | |
1995 | } | |
2f7ee569 | 1996 | |
b3dc094e | 1997 | return NULL; |
2f7ee569 | 1998 | } |
2f7ee569 | 1999 | |
cb0f1fe9 | 2000 | /** |
74a1166d | 2001 | * cgroup_task_migrate - move a task from one cgroup to another. |
60106946 | 2002 | * @old_cgrp: the cgroup @tsk is being migrated from |
cb0f1fe9 TH |
2003 | * @tsk: the task being migrated |
2004 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 2005 | * |
cb0f1fe9 | 2006 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 2007 | */ |
5abb8855 TH |
2008 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
2009 | struct task_struct *tsk, | |
2010 | struct css_set *new_cset) | |
74a1166d | 2011 | { |
5abb8855 | 2012 | struct css_set *old_cset; |
74a1166d | 2013 | |
cb0f1fe9 TH |
2014 | lockdep_assert_held(&cgroup_mutex); |
2015 | lockdep_assert_held(&css_set_rwsem); | |
2016 | ||
74a1166d | 2017 | /* |
026085ef MSB |
2018 | * We are synchronized through threadgroup_lock() against PF_EXITING |
2019 | * setting such that we can't race against cgroup_exit() changing the | |
2020 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 2021 | */ |
c84cdf75 | 2022 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 2023 | old_cset = task_css_set(tsk); |
74a1166d | 2024 | |
b3dc094e | 2025 | get_css_set(new_cset); |
5abb8855 | 2026 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 2027 | |
1b9aba49 TH |
2028 | /* |
2029 | * Use move_tail so that cgroup_taskset_first() still returns the | |
2030 | * leader after migration. This works because cgroup_migrate() | |
2031 | * ensures that the dst_cset of the leader is the first on the | |
2032 | * tset's dst_csets list. | |
2033 | */ | |
2034 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
2035 | |
2036 | /* | |
5abb8855 TH |
2037 | * We just gained a reference on old_cset by taking it from the |
2038 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
2039 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 2040 | */ |
a25eb52e | 2041 | put_css_set_locked(old_cset); |
74a1166d BB |
2042 | } |
2043 | ||
a043e3b2 | 2044 | /** |
1958d2d5 TH |
2045 | * cgroup_migrate_finish - cleanup after attach |
2046 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2047 | * |
1958d2d5 TH |
2048 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2049 | * those functions for details. | |
74a1166d | 2050 | */ |
1958d2d5 | 2051 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2052 | { |
1958d2d5 | 2053 | struct css_set *cset, *tmp_cset; |
74a1166d | 2054 | |
1958d2d5 TH |
2055 | lockdep_assert_held(&cgroup_mutex); |
2056 | ||
2057 | down_write(&css_set_rwsem); | |
2058 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2059 | cset->mg_src_cgrp = NULL; | |
2060 | cset->mg_dst_cset = NULL; | |
2061 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2062 | put_css_set_locked(cset); |
1958d2d5 TH |
2063 | } |
2064 | up_write(&css_set_rwsem); | |
2065 | } | |
2066 | ||
2067 | /** | |
2068 | * cgroup_migrate_add_src - add a migration source css_set | |
2069 | * @src_cset: the source css_set to add | |
2070 | * @dst_cgrp: the destination cgroup | |
2071 | * @preloaded_csets: list of preloaded css_sets | |
2072 | * | |
2073 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2074 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2075 | * up by cgroup_migrate_finish(). | |
2076 | * | |
2077 | * This function may be called without holding threadgroup_lock even if the | |
2078 | * target is a process. Threads may be created and destroyed but as long | |
2079 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
2080 | * the preloaded css_sets are guaranteed to cover all migrations. | |
2081 | */ | |
2082 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2083 | struct cgroup *dst_cgrp, | |
2084 | struct list_head *preloaded_csets) | |
2085 | { | |
2086 | struct cgroup *src_cgrp; | |
2087 | ||
2088 | lockdep_assert_held(&cgroup_mutex); | |
2089 | lockdep_assert_held(&css_set_rwsem); | |
2090 | ||
2091 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2092 | ||
1958d2d5 TH |
2093 | if (!list_empty(&src_cset->mg_preload_node)) |
2094 | return; | |
2095 | ||
2096 | WARN_ON(src_cset->mg_src_cgrp); | |
2097 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2098 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2099 | ||
2100 | src_cset->mg_src_cgrp = src_cgrp; | |
2101 | get_css_set(src_cset); | |
2102 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2103 | } | |
2104 | ||
2105 | /** | |
2106 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2107 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2108 | * @preloaded_csets: list of preloaded source css_sets |
2109 | * | |
2110 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2111 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2112 | * pins all destination css_sets, links each to its source, and append them |
2113 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2114 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2115 | * |
2116 | * This function must be called after cgroup_migrate_add_src() has been | |
2117 | * called on each migration source css_set. After migration is performed | |
2118 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2119 | * @preloaded_csets. | |
2120 | */ | |
2121 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2122 | struct list_head *preloaded_csets) | |
2123 | { | |
2124 | LIST_HEAD(csets); | |
f817de98 | 2125 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2126 | |
2127 | lockdep_assert_held(&cgroup_mutex); | |
2128 | ||
f8f22e53 TH |
2129 | /* |
2130 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2131 | * with tasks so that child cgroups don't compete against tasks. | |
2132 | */ | |
d51f39b0 | 2133 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2134 | dst_cgrp->child_subsys_mask) |
2135 | return -EBUSY; | |
2136 | ||
1958d2d5 | 2137 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2138 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2139 | struct css_set *dst_cset; |
2140 | ||
f817de98 TH |
2141 | dst_cset = find_css_set(src_cset, |
2142 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2143 | if (!dst_cset) |
2144 | goto err; | |
2145 | ||
2146 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2147 | |
2148 | /* | |
2149 | * If src cset equals dst, it's noop. Drop the src. | |
2150 | * cgroup_migrate() will skip the cset too. Note that we | |
2151 | * can't handle src == dst as some nodes are used by both. | |
2152 | */ | |
2153 | if (src_cset == dst_cset) { | |
2154 | src_cset->mg_src_cgrp = NULL; | |
2155 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2156 | put_css_set(src_cset); |
2157 | put_css_set(dst_cset); | |
f817de98 TH |
2158 | continue; |
2159 | } | |
2160 | ||
1958d2d5 TH |
2161 | src_cset->mg_dst_cset = dst_cset; |
2162 | ||
2163 | if (list_empty(&dst_cset->mg_preload_node)) | |
2164 | list_add(&dst_cset->mg_preload_node, &csets); | |
2165 | else | |
a25eb52e | 2166 | put_css_set(dst_cset); |
1958d2d5 TH |
2167 | } |
2168 | ||
f817de98 | 2169 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2170 | return 0; |
2171 | err: | |
2172 | cgroup_migrate_finish(&csets); | |
2173 | return -ENOMEM; | |
2174 | } | |
2175 | ||
2176 | /** | |
2177 | * cgroup_migrate - migrate a process or task to a cgroup | |
2178 | * @cgrp: the destination cgroup | |
2179 | * @leader: the leader of the process or the task to migrate | |
2180 | * @threadgroup: whether @leader points to the whole process or a single task | |
2181 | * | |
2182 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2183 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2184 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2185 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2186 | * function and following up with cgroup_migrate_finish(). | |
2187 | * | |
2188 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2189 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2190 | * failure, when migrating multiple targets, the success or failure can be | |
2191 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2192 | * actually starting migrating. | |
2193 | */ | |
2194 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2195 | bool threadgroup) | |
74a1166d | 2196 | { |
b3dc094e TH |
2197 | struct cgroup_taskset tset = { |
2198 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2199 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2200 | .csets = &tset.src_csets, | |
2201 | }; | |
1c6727af | 2202 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2203 | struct css_set *cset, *tmp_cset; |
2204 | struct task_struct *task, *tmp_task; | |
2205 | int i, ret; | |
74a1166d | 2206 | |
fb5d2b4c MSB |
2207 | /* |
2208 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2209 | * already PF_EXITING could be freed from underneath us unless we | |
2210 | * take an rcu_read_lock. | |
2211 | */ | |
b3dc094e | 2212 | down_write(&css_set_rwsem); |
fb5d2b4c | 2213 | rcu_read_lock(); |
9db8de37 | 2214 | task = leader; |
74a1166d | 2215 | do { |
9db8de37 TH |
2216 | /* @task either already exited or can't exit until the end */ |
2217 | if (task->flags & PF_EXITING) | |
ea84753c | 2218 | goto next; |
134d3373 | 2219 | |
eaf797ab TH |
2220 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2221 | if (list_empty(&task->cg_list)) | |
ea84753c | 2222 | goto next; |
cd3d0952 | 2223 | |
b3dc094e | 2224 | cset = task_css_set(task); |
1958d2d5 | 2225 | if (!cset->mg_src_cgrp) |
ea84753c | 2226 | goto next; |
b3dc094e | 2227 | |
61d1d219 | 2228 | /* |
1b9aba49 TH |
2229 | * cgroup_taskset_first() must always return the leader. |
2230 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2231 | */ |
1b9aba49 TH |
2232 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2233 | if (list_empty(&cset->mg_node)) | |
2234 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2235 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2236 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2237 | &tset.dst_csets); | |
ea84753c | 2238 | next: |
081aa458 LZ |
2239 | if (!threadgroup) |
2240 | break; | |
9db8de37 | 2241 | } while_each_thread(leader, task); |
fb5d2b4c | 2242 | rcu_read_unlock(); |
b3dc094e | 2243 | up_write(&css_set_rwsem); |
74a1166d | 2244 | |
134d3373 | 2245 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2246 | if (list_empty(&tset.src_csets)) |
2247 | return 0; | |
134d3373 | 2248 | |
1958d2d5 | 2249 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2250 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2251 | if (css->ss->can_attach) { |
9db8de37 TH |
2252 | ret = css->ss->can_attach(css, &tset); |
2253 | if (ret) { | |
1c6727af | 2254 | failed_css = css; |
74a1166d BB |
2255 | goto out_cancel_attach; |
2256 | } | |
2257 | } | |
74a1166d BB |
2258 | } |
2259 | ||
2260 | /* | |
1958d2d5 TH |
2261 | * Now that we're guaranteed success, proceed to move all tasks to |
2262 | * the new cgroup. There are no failure cases after here, so this | |
2263 | * is the commit point. | |
74a1166d | 2264 | */ |
cb0f1fe9 | 2265 | down_write(&css_set_rwsem); |
b3dc094e TH |
2266 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2267 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2268 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2269 | cset->mg_dst_cset); | |
74a1166d | 2270 | } |
cb0f1fe9 | 2271 | up_write(&css_set_rwsem); |
74a1166d BB |
2272 | |
2273 | /* | |
1958d2d5 TH |
2274 | * Migration is committed, all target tasks are now on dst_csets. |
2275 | * Nothing is sensitive to fork() after this point. Notify | |
2276 | * controllers that migration is complete. | |
74a1166d | 2277 | */ |
1958d2d5 | 2278 | tset.csets = &tset.dst_csets; |
74a1166d | 2279 | |
aec3dfcb | 2280 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2281 | if (css->ss->attach) |
2282 | css->ss->attach(css, &tset); | |
74a1166d | 2283 | |
9db8de37 | 2284 | ret = 0; |
b3dc094e TH |
2285 | goto out_release_tset; |
2286 | ||
74a1166d | 2287 | out_cancel_attach: |
aec3dfcb | 2288 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2289 | if (css == failed_css) |
2290 | break; | |
2291 | if (css->ss->cancel_attach) | |
2292 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2293 | } |
b3dc094e TH |
2294 | out_release_tset: |
2295 | down_write(&css_set_rwsem); | |
2296 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2297 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2298 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2299 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2300 | } |
2301 | up_write(&css_set_rwsem); | |
9db8de37 | 2302 | return ret; |
74a1166d BB |
2303 | } |
2304 | ||
1958d2d5 TH |
2305 | /** |
2306 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2307 | * @dst_cgrp: the cgroup to attach to | |
2308 | * @leader: the task or the leader of the threadgroup to be attached | |
2309 | * @threadgroup: attach the whole threadgroup? | |
2310 | * | |
0e1d768f | 2311 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2312 | */ |
2313 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2314 | struct task_struct *leader, bool threadgroup) | |
2315 | { | |
2316 | LIST_HEAD(preloaded_csets); | |
2317 | struct task_struct *task; | |
2318 | int ret; | |
2319 | ||
2320 | /* look up all src csets */ | |
2321 | down_read(&css_set_rwsem); | |
2322 | rcu_read_lock(); | |
2323 | task = leader; | |
2324 | do { | |
2325 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2326 | &preloaded_csets); | |
2327 | if (!threadgroup) | |
2328 | break; | |
2329 | } while_each_thread(leader, task); | |
2330 | rcu_read_unlock(); | |
2331 | up_read(&css_set_rwsem); | |
2332 | ||
2333 | /* prepare dst csets and commit */ | |
2334 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2335 | if (!ret) | |
2336 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2337 | ||
2338 | cgroup_migrate_finish(&preloaded_csets); | |
2339 | return ret; | |
74a1166d BB |
2340 | } |
2341 | ||
2342 | /* | |
2343 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2344 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2345 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2346 | */ |
acbef755 TH |
2347 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2348 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2349 | { |
bbcb81d0 | 2350 | struct task_struct *tsk; |
c69e8d9c | 2351 | const struct cred *cred = current_cred(), *tcred; |
e76ecaee | 2352 | struct cgroup *cgrp; |
acbef755 | 2353 | pid_t pid; |
bbcb81d0 PM |
2354 | int ret; |
2355 | ||
acbef755 TH |
2356 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2357 | return -EINVAL; | |
2358 | ||
e76ecaee TH |
2359 | cgrp = cgroup_kn_lock_live(of->kn); |
2360 | if (!cgrp) | |
74a1166d BB |
2361 | return -ENODEV; |
2362 | ||
b78949eb MSB |
2363 | retry_find_task: |
2364 | rcu_read_lock(); | |
bbcb81d0 | 2365 | if (pid) { |
73507f33 | 2366 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2367 | if (!tsk) { |
2368 | rcu_read_unlock(); | |
dd4b0a46 | 2369 | ret = -ESRCH; |
b78949eb | 2370 | goto out_unlock_cgroup; |
bbcb81d0 | 2371 | } |
74a1166d BB |
2372 | /* |
2373 | * even if we're attaching all tasks in the thread group, we | |
2374 | * only need to check permissions on one of them. | |
2375 | */ | |
c69e8d9c | 2376 | tcred = __task_cred(tsk); |
14a590c3 EB |
2377 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2378 | !uid_eq(cred->euid, tcred->uid) && | |
2379 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2380 | rcu_read_unlock(); |
b78949eb MSB |
2381 | ret = -EACCES; |
2382 | goto out_unlock_cgroup; | |
bbcb81d0 | 2383 | } |
b78949eb MSB |
2384 | } else |
2385 | tsk = current; | |
cd3d0952 TH |
2386 | |
2387 | if (threadgroup) | |
b78949eb | 2388 | tsk = tsk->group_leader; |
c4c27fbd MG |
2389 | |
2390 | /* | |
14a40ffc | 2391 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2392 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2393 | * with no rt_runtime allocated. Just say no. | |
2394 | */ | |
14a40ffc | 2395 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2396 | ret = -EINVAL; |
2397 | rcu_read_unlock(); | |
2398 | goto out_unlock_cgroup; | |
2399 | } | |
2400 | ||
b78949eb MSB |
2401 | get_task_struct(tsk); |
2402 | rcu_read_unlock(); | |
2403 | ||
2404 | threadgroup_lock(tsk); | |
2405 | if (threadgroup) { | |
2406 | if (!thread_group_leader(tsk)) { | |
2407 | /* | |
2408 | * a race with de_thread from another thread's exec() | |
2409 | * may strip us of our leadership, if this happens, | |
2410 | * there is no choice but to throw this task away and | |
2411 | * try again; this is | |
2412 | * "double-double-toil-and-trouble-check locking". | |
2413 | */ | |
2414 | threadgroup_unlock(tsk); | |
2415 | put_task_struct(tsk); | |
2416 | goto retry_find_task; | |
2417 | } | |
081aa458 LZ |
2418 | } |
2419 | ||
2420 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2421 | ||
cd3d0952 TH |
2422 | threadgroup_unlock(tsk); |
2423 | ||
bbcb81d0 | 2424 | put_task_struct(tsk); |
b78949eb | 2425 | out_unlock_cgroup: |
e76ecaee | 2426 | cgroup_kn_unlock(of->kn); |
acbef755 | 2427 | return ret ?: nbytes; |
bbcb81d0 PM |
2428 | } |
2429 | ||
7ae1bad9 TH |
2430 | /** |
2431 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2432 | * @from: attach to all cgroups of a given task | |
2433 | * @tsk: the task to be attached | |
2434 | */ | |
2435 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2436 | { | |
3dd06ffa | 2437 | struct cgroup_root *root; |
7ae1bad9 TH |
2438 | int retval = 0; |
2439 | ||
47cfcd09 | 2440 | mutex_lock(&cgroup_mutex); |
985ed670 | 2441 | for_each_root(root) { |
96d365e0 TH |
2442 | struct cgroup *from_cgrp; |
2443 | ||
3dd06ffa | 2444 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2445 | continue; |
2446 | ||
96d365e0 TH |
2447 | down_read(&css_set_rwsem); |
2448 | from_cgrp = task_cgroup_from_root(from, root); | |
2449 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2450 | |
6f4b7e63 | 2451 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2452 | if (retval) |
2453 | break; | |
2454 | } | |
47cfcd09 | 2455 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2456 | |
2457 | return retval; | |
2458 | } | |
2459 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2460 | ||
acbef755 TH |
2461 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2462 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2463 | { |
acbef755 | 2464 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2465 | } |
2466 | ||
acbef755 TH |
2467 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2468 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2469 | { |
acbef755 | 2470 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2471 | } |
2472 | ||
451af504 TH |
2473 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2474 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2475 | { |
e76ecaee | 2476 | struct cgroup *cgrp; |
5f469907 | 2477 | |
e76ecaee | 2478 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2479 | |
e76ecaee TH |
2480 | cgrp = cgroup_kn_lock_live(of->kn); |
2481 | if (!cgrp) | |
e788e066 | 2482 | return -ENODEV; |
69e943b7 | 2483 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2484 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2485 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2486 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2487 | cgroup_kn_unlock(of->kn); |
451af504 | 2488 | return nbytes; |
e788e066 PM |
2489 | } |
2490 | ||
2da8ca82 | 2491 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2492 | { |
2da8ca82 | 2493 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2494 | |
46cfeb04 | 2495 | spin_lock(&release_agent_path_lock); |
e788e066 | 2496 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2497 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2498 | seq_putc(seq, '\n'); |
e788e066 PM |
2499 | return 0; |
2500 | } | |
2501 | ||
2da8ca82 | 2502 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2503 | { |
c1d5d42e | 2504 | seq_puts(seq, "0\n"); |
e788e066 PM |
2505 | return 0; |
2506 | } | |
2507 | ||
f8f22e53 | 2508 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
355e0c48 | 2509 | { |
f8f22e53 TH |
2510 | struct cgroup_subsys *ss; |
2511 | bool printed = false; | |
2512 | int ssid; | |
a742c59d | 2513 | |
f8f22e53 TH |
2514 | for_each_subsys(ss, ssid) { |
2515 | if (ss_mask & (1 << ssid)) { | |
2516 | if (printed) | |
2517 | seq_putc(seq, ' '); | |
2518 | seq_printf(seq, "%s", ss->name); | |
2519 | printed = true; | |
2520 | } | |
e73d2c61 | 2521 | } |
f8f22e53 TH |
2522 | if (printed) |
2523 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2524 | } |
2525 | ||
f8f22e53 TH |
2526 | /* show controllers which are currently attached to the default hierarchy */ |
2527 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2528 | { |
f8f22e53 TH |
2529 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2530 | ||
5533e011 TH |
2531 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2532 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2533 | return 0; |
db3b1497 PM |
2534 | } |
2535 | ||
f8f22e53 TH |
2536 | /* show controllers which are enabled from the parent */ |
2537 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2538 | { |
f8f22e53 TH |
2539 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2540 | ||
667c2491 | 2541 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2542 | return 0; |
ddbcc7e8 PM |
2543 | } |
2544 | ||
f8f22e53 TH |
2545 | /* show controllers which are enabled for a given cgroup's children */ |
2546 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2547 | { |
f8f22e53 TH |
2548 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2549 | ||
667c2491 | 2550 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2551 | return 0; |
2552 | } | |
2553 | ||
2554 | /** | |
2555 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2556 | * @cgrp: root of the subtree to update csses for | |
2557 | * | |
2558 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2559 | * css associations need to be updated accordingly. This function looks up | |
2560 | * all css_sets which are attached to the subtree, creates the matching | |
2561 | * updated css_sets and migrates the tasks to the new ones. | |
2562 | */ | |
2563 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2564 | { | |
2565 | LIST_HEAD(preloaded_csets); | |
2566 | struct cgroup_subsys_state *css; | |
2567 | struct css_set *src_cset; | |
2568 | int ret; | |
2569 | ||
f8f22e53 TH |
2570 | lockdep_assert_held(&cgroup_mutex); |
2571 | ||
2572 | /* look up all csses currently attached to @cgrp's subtree */ | |
2573 | down_read(&css_set_rwsem); | |
2574 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2575 | struct cgrp_cset_link *link; | |
2576 | ||
2577 | /* self is not affected by child_subsys_mask change */ | |
2578 | if (css->cgroup == cgrp) | |
2579 | continue; | |
2580 | ||
2581 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2582 | cgroup_migrate_add_src(link->cset, cgrp, | |
2583 | &preloaded_csets); | |
2584 | } | |
2585 | up_read(&css_set_rwsem); | |
2586 | ||
2587 | /* NULL dst indicates self on default hierarchy */ | |
2588 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2589 | if (ret) | |
2590 | goto out_finish; | |
2591 | ||
2592 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2593 | struct task_struct *last_task = NULL, *task; | |
2594 | ||
2595 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2596 | if (!src_cset->mg_src_cgrp) | |
2597 | break; | |
2598 | ||
2599 | /* | |
2600 | * All tasks in src_cset need to be migrated to the | |
2601 | * matching dst_cset. Empty it process by process. We | |
2602 | * walk tasks but migrate processes. The leader might even | |
2603 | * belong to a different cset but such src_cset would also | |
2604 | * be among the target src_csets because the default | |
2605 | * hierarchy enforces per-process membership. | |
2606 | */ | |
2607 | while (true) { | |
2608 | down_read(&css_set_rwsem); | |
2609 | task = list_first_entry_or_null(&src_cset->tasks, | |
2610 | struct task_struct, cg_list); | |
2611 | if (task) { | |
2612 | task = task->group_leader; | |
2613 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2614 | get_task_struct(task); | |
2615 | } | |
2616 | up_read(&css_set_rwsem); | |
2617 | ||
2618 | if (!task) | |
2619 | break; | |
2620 | ||
2621 | /* guard against possible infinite loop */ | |
2622 | if (WARN(last_task == task, | |
2623 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2624 | goto out_finish; | |
2625 | last_task = task; | |
2626 | ||
2627 | threadgroup_lock(task); | |
2628 | /* raced against de_thread() from another thread? */ | |
2629 | if (!thread_group_leader(task)) { | |
2630 | threadgroup_unlock(task); | |
2631 | put_task_struct(task); | |
2632 | continue; | |
2633 | } | |
2634 | ||
2635 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2636 | ||
2637 | threadgroup_unlock(task); | |
2638 | put_task_struct(task); | |
2639 | ||
2640 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2641 | goto out_finish; | |
2642 | } | |
2643 | } | |
2644 | ||
2645 | out_finish: | |
2646 | cgroup_migrate_finish(&preloaded_csets); | |
2647 | return ret; | |
2648 | } | |
2649 | ||
2650 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2651 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2652 | char *buf, size_t nbytes, | |
2653 | loff_t off) | |
f8f22e53 | 2654 | { |
7d331fa9 | 2655 | unsigned int enable = 0, disable = 0; |
755bf5ee | 2656 | unsigned int css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; |
a9746d8d | 2657 | struct cgroup *cgrp, *child; |
f8f22e53 | 2658 | struct cgroup_subsys *ss; |
451af504 | 2659 | char *tok; |
f8f22e53 TH |
2660 | int ssid, ret; |
2661 | ||
2662 | /* | |
d37167ab TH |
2663 | * Parse input - space separated list of subsystem names prefixed |
2664 | * with either + or -. | |
f8f22e53 | 2665 | */ |
451af504 TH |
2666 | buf = strstrip(buf); |
2667 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2668 | if (tok[0] == '\0') |
2669 | continue; | |
f8f22e53 | 2670 | for_each_subsys(ss, ssid) { |
5533e011 TH |
2671 | if (ss->disabled || strcmp(tok + 1, ss->name) || |
2672 | ((1 << ss->id) & cgrp_dfl_root_inhibit_ss_mask)) | |
f8f22e53 TH |
2673 | continue; |
2674 | ||
2675 | if (*tok == '+') { | |
7d331fa9 TH |
2676 | enable |= 1 << ssid; |
2677 | disable &= ~(1 << ssid); | |
f8f22e53 | 2678 | } else if (*tok == '-') { |
7d331fa9 TH |
2679 | disable |= 1 << ssid; |
2680 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2681 | } else { |
2682 | return -EINVAL; | |
2683 | } | |
2684 | break; | |
2685 | } | |
2686 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2687 | return -EINVAL; | |
2688 | } | |
2689 | ||
a9746d8d TH |
2690 | cgrp = cgroup_kn_lock_live(of->kn); |
2691 | if (!cgrp) | |
2692 | return -ENODEV; | |
f8f22e53 TH |
2693 | |
2694 | for_each_subsys(ss, ssid) { | |
2695 | if (enable & (1 << ssid)) { | |
667c2491 | 2696 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2697 | enable &= ~(1 << ssid); |
2698 | continue; | |
2699 | } | |
2700 | ||
c29adf24 TH |
2701 | /* unavailable or not enabled on the parent? */ |
2702 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2703 | (cgroup_parent(cgrp) && | |
667c2491 | 2704 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
2705 | ret = -ENOENT; |
2706 | goto out_unlock; | |
2707 | } | |
f8f22e53 | 2708 | } else if (disable & (1 << ssid)) { |
667c2491 | 2709 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2710 | disable &= ~(1 << ssid); |
2711 | continue; | |
2712 | } | |
2713 | ||
2714 | /* a child has it enabled? */ | |
2715 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2716 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2717 | ret = -EBUSY; |
ddab2b6e | 2718 | goto out_unlock; |
f8f22e53 TH |
2719 | } |
2720 | } | |
2721 | } | |
2722 | } | |
2723 | ||
2724 | if (!enable && !disable) { | |
2725 | ret = 0; | |
ddab2b6e | 2726 | goto out_unlock; |
f8f22e53 TH |
2727 | } |
2728 | ||
2729 | /* | |
667c2491 | 2730 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2731 | * with tasks so that child cgroups don't compete against tasks. |
2732 | */ | |
d51f39b0 | 2733 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
2734 | ret = -EBUSY; |
2735 | goto out_unlock; | |
2736 | } | |
2737 | ||
2738 | /* | |
f63070d3 TH |
2739 | * Update subsys masks and calculate what needs to be done. More |
2740 | * subsystems than specified may need to be enabled or disabled | |
2741 | * depending on subsystem dependencies. | |
2742 | */ | |
755bf5ee TH |
2743 | old_sc = cgrp->subtree_control; |
2744 | old_ss = cgrp->child_subsys_mask; | |
2745 | new_sc = (old_sc | enable) & ~disable; | |
2746 | new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc); | |
f63070d3 | 2747 | |
755bf5ee TH |
2748 | css_enable = ~old_ss & new_ss; |
2749 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
2750 | enable |= css_enable; |
2751 | disable |= css_disable; | |
c29adf24 | 2752 | |
db6e3053 TH |
2753 | /* |
2754 | * Because css offlining is asynchronous, userland might try to | |
2755 | * re-enable the same controller while the previous instance is | |
2756 | * still around. In such cases, wait till it's gone using | |
2757 | * offline_waitq. | |
2758 | */ | |
2759 | for_each_subsys(ss, ssid) { | |
2760 | if (!(css_enable & (1 << ssid))) | |
2761 | continue; | |
2762 | ||
2763 | cgroup_for_each_live_child(child, cgrp) { | |
2764 | DEFINE_WAIT(wait); | |
2765 | ||
2766 | if (!cgroup_css(child, ss)) | |
2767 | continue; | |
2768 | ||
2769 | cgroup_get(child); | |
2770 | prepare_to_wait(&child->offline_waitq, &wait, | |
2771 | TASK_UNINTERRUPTIBLE); | |
2772 | cgroup_kn_unlock(of->kn); | |
2773 | schedule(); | |
2774 | finish_wait(&child->offline_waitq, &wait); | |
2775 | cgroup_put(child); | |
2776 | ||
2777 | return restart_syscall(); | |
2778 | } | |
2779 | } | |
2780 | ||
755bf5ee TH |
2781 | cgrp->subtree_control = new_sc; |
2782 | cgrp->child_subsys_mask = new_ss; | |
2783 | ||
f63070d3 TH |
2784 | /* |
2785 | * Create new csses or make the existing ones visible. A css is | |
2786 | * created invisible if it's being implicitly enabled through | |
2787 | * dependency. An invisible css is made visible when the userland | |
2788 | * explicitly enables it. | |
f8f22e53 TH |
2789 | */ |
2790 | for_each_subsys(ss, ssid) { | |
2791 | if (!(enable & (1 << ssid))) | |
2792 | continue; | |
2793 | ||
2794 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
2795 | if (css_enable & (1 << ssid)) |
2796 | ret = create_css(child, ss, | |
2797 | cgrp->subtree_control & (1 << ssid)); | |
2798 | else | |
2799 | ret = cgroup_populate_dir(child, 1 << ssid); | |
f8f22e53 TH |
2800 | if (ret) |
2801 | goto err_undo_css; | |
2802 | } | |
2803 | } | |
2804 | ||
c29adf24 TH |
2805 | /* |
2806 | * At this point, cgroup_e_css() results reflect the new csses | |
2807 | * making the following cgroup_update_dfl_csses() properly update | |
2808 | * css associations of all tasks in the subtree. | |
2809 | */ | |
f8f22e53 TH |
2810 | ret = cgroup_update_dfl_csses(cgrp); |
2811 | if (ret) | |
2812 | goto err_undo_css; | |
2813 | ||
f63070d3 TH |
2814 | /* |
2815 | * All tasks are migrated out of disabled csses. Kill or hide | |
2816 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
2817 | * disabled while other subsystems are still depending on it. The |
2818 | * css must not actively control resources and be in the vanilla | |
2819 | * state if it's made visible again later. Controllers which may | |
2820 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 2821 | */ |
f8f22e53 TH |
2822 | for_each_subsys(ss, ssid) { |
2823 | if (!(disable & (1 << ssid))) | |
2824 | continue; | |
2825 | ||
f63070d3 | 2826 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
2827 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
2828 | ||
2829 | if (css_disable & (1 << ssid)) { | |
2830 | kill_css(css); | |
2831 | } else { | |
f63070d3 | 2832 | cgroup_clear_dir(child, 1 << ssid); |
b4536f0c TH |
2833 | if (ss->css_reset) |
2834 | ss->css_reset(css); | |
2835 | } | |
f63070d3 | 2836 | } |
f8f22e53 TH |
2837 | } |
2838 | ||
56c807ba TH |
2839 | /* |
2840 | * The effective csses of all the descendants (excluding @cgrp) may | |
2841 | * have changed. Subsystems can optionally subscribe to this event | |
2842 | * by implementing ->css_e_css_changed() which is invoked if any of | |
2843 | * the effective csses seen from the css's cgroup may have changed. | |
2844 | */ | |
2845 | for_each_subsys(ss, ssid) { | |
2846 | struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss); | |
2847 | struct cgroup_subsys_state *css; | |
2848 | ||
2849 | if (!ss->css_e_css_changed || !this_css) | |
2850 | continue; | |
2851 | ||
2852 | css_for_each_descendant_pre(css, this_css) | |
2853 | if (css != this_css) | |
2854 | ss->css_e_css_changed(css); | |
2855 | } | |
2856 | ||
f8f22e53 TH |
2857 | kernfs_activate(cgrp->kn); |
2858 | ret = 0; | |
2859 | out_unlock: | |
a9746d8d | 2860 | cgroup_kn_unlock(of->kn); |
451af504 | 2861 | return ret ?: nbytes; |
f8f22e53 TH |
2862 | |
2863 | err_undo_css: | |
755bf5ee TH |
2864 | cgrp->subtree_control = old_sc; |
2865 | cgrp->child_subsys_mask = old_ss; | |
f8f22e53 TH |
2866 | |
2867 | for_each_subsys(ss, ssid) { | |
2868 | if (!(enable & (1 << ssid))) | |
2869 | continue; | |
2870 | ||
2871 | cgroup_for_each_live_child(child, cgrp) { | |
2872 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
2873 | |
2874 | if (!css) | |
2875 | continue; | |
2876 | ||
2877 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 2878 | kill_css(css); |
f63070d3 TH |
2879 | else |
2880 | cgroup_clear_dir(child, 1 << ssid); | |
f8f22e53 TH |
2881 | } |
2882 | } | |
2883 | goto out_unlock; | |
2884 | } | |
2885 | ||
842b597e TH |
2886 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2887 | { | |
2888 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2889 | return 0; | |
2890 | } | |
2891 | ||
2bd59d48 TH |
2892 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2893 | size_t nbytes, loff_t off) | |
355e0c48 | 2894 | { |
2bd59d48 TH |
2895 | struct cgroup *cgrp = of->kn->parent->priv; |
2896 | struct cftype *cft = of->kn->priv; | |
2897 | struct cgroup_subsys_state *css; | |
a742c59d | 2898 | int ret; |
355e0c48 | 2899 | |
b4168640 TH |
2900 | if (cft->write) |
2901 | return cft->write(of, buf, nbytes, off); | |
2902 | ||
2bd59d48 TH |
2903 | /* |
2904 | * kernfs guarantees that a file isn't deleted with operations in | |
2905 | * flight, which means that the matching css is and stays alive and | |
2906 | * doesn't need to be pinned. The RCU locking is not necessary | |
2907 | * either. It's just for the convenience of using cgroup_css(). | |
2908 | */ | |
2909 | rcu_read_lock(); | |
2910 | css = cgroup_css(cgrp, cft->ss); | |
2911 | rcu_read_unlock(); | |
a742c59d | 2912 | |
451af504 | 2913 | if (cft->write_u64) { |
a742c59d TH |
2914 | unsigned long long v; |
2915 | ret = kstrtoull(buf, 0, &v); | |
2916 | if (!ret) | |
2917 | ret = cft->write_u64(css, cft, v); | |
2918 | } else if (cft->write_s64) { | |
2919 | long long v; | |
2920 | ret = kstrtoll(buf, 0, &v); | |
2921 | if (!ret) | |
2922 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2923 | } else { |
a742c59d | 2924 | ret = -EINVAL; |
e73d2c61 | 2925 | } |
2bd59d48 | 2926 | |
a742c59d | 2927 | return ret ?: nbytes; |
355e0c48 PM |
2928 | } |
2929 | ||
6612f05b | 2930 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2931 | { |
2bd59d48 | 2932 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2933 | } |
2934 | ||
6612f05b | 2935 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2936 | { |
2bd59d48 | 2937 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2938 | } |
2939 | ||
6612f05b | 2940 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2941 | { |
2bd59d48 | 2942 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2943 | } |
2944 | ||
91796569 | 2945 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2946 | { |
7da11279 TH |
2947 | struct cftype *cft = seq_cft(m); |
2948 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2949 | |
2da8ca82 TH |
2950 | if (cft->seq_show) |
2951 | return cft->seq_show(m, arg); | |
e73d2c61 | 2952 | |
f4c753b7 | 2953 | if (cft->read_u64) |
896f5199 TH |
2954 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2955 | else if (cft->read_s64) | |
2956 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2957 | else | |
2958 | return -EINVAL; | |
2959 | return 0; | |
91796569 PM |
2960 | } |
2961 | ||
2bd59d48 TH |
2962 | static struct kernfs_ops cgroup_kf_single_ops = { |
2963 | .atomic_write_len = PAGE_SIZE, | |
2964 | .write = cgroup_file_write, | |
2965 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2966 | }; |
2967 | ||
2bd59d48 TH |
2968 | static struct kernfs_ops cgroup_kf_ops = { |
2969 | .atomic_write_len = PAGE_SIZE, | |
2970 | .write = cgroup_file_write, | |
2971 | .seq_start = cgroup_seqfile_start, | |
2972 | .seq_next = cgroup_seqfile_next, | |
2973 | .seq_stop = cgroup_seqfile_stop, | |
2974 | .seq_show = cgroup_seqfile_show, | |
2975 | }; | |
ddbcc7e8 PM |
2976 | |
2977 | /* | |
2978 | * cgroup_rename - Only allow simple rename of directories in place. | |
2979 | */ | |
2bd59d48 TH |
2980 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2981 | const char *new_name_str) | |
ddbcc7e8 | 2982 | { |
2bd59d48 | 2983 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2984 | int ret; |
65dff759 | 2985 | |
2bd59d48 | 2986 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2987 | return -ENOTDIR; |
2bd59d48 | 2988 | if (kn->parent != new_parent) |
ddbcc7e8 | 2989 | return -EIO; |
65dff759 | 2990 | |
6db8e85c TH |
2991 | /* |
2992 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 2993 | * limited. Disallow on the default hierarchy. |
6db8e85c | 2994 | */ |
aa6ec29b | 2995 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 2996 | return -EPERM; |
099fca32 | 2997 | |
e1b2dc17 | 2998 | /* |
8353da1f | 2999 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3000 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3001 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3002 | */ |
3003 | kernfs_break_active_protection(new_parent); | |
3004 | kernfs_break_active_protection(kn); | |
099fca32 | 3005 | |
2bd59d48 | 3006 | mutex_lock(&cgroup_mutex); |
099fca32 | 3007 | |
2bd59d48 | 3008 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3009 | |
2bd59d48 | 3010 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3011 | |
3012 | kernfs_unbreak_active_protection(kn); | |
3013 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3014 | return ret; |
099fca32 LZ |
3015 | } |
3016 | ||
49957f8e TH |
3017 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3018 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3019 | { | |
3020 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3021 | .ia_uid = current_fsuid(), | |
3022 | .ia_gid = current_fsgid(), }; | |
3023 | ||
3024 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3025 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3026 | return 0; | |
3027 | ||
3028 | return kernfs_setattr(kn, &iattr); | |
3029 | } | |
3030 | ||
2bb566cb | 3031 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 3032 | { |
8d7e6fb0 | 3033 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3034 | struct kernfs_node *kn; |
3035 | struct lock_class_key *key = NULL; | |
49957f8e | 3036 | int ret; |
05ef1d7c | 3037 | |
2bd59d48 TH |
3038 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3039 | key = &cft->lockdep_key; | |
3040 | #endif | |
3041 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3042 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
3043 | NULL, false, key); | |
49957f8e TH |
3044 | if (IS_ERR(kn)) |
3045 | return PTR_ERR(kn); | |
3046 | ||
3047 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3048 | if (ret) { |
49957f8e | 3049 | kernfs_remove(kn); |
f8f22e53 TH |
3050 | return ret; |
3051 | } | |
3052 | ||
b7fc5ad2 | 3053 | if (cft->seq_show == cgroup_populated_show) |
842b597e | 3054 | cgrp->populated_kn = kn; |
f8f22e53 | 3055 | return 0; |
ddbcc7e8 PM |
3056 | } |
3057 | ||
b1f28d31 TH |
3058 | /** |
3059 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
3060 | * @cgrp: the target cgroup | |
b1f28d31 TH |
3061 | * @cfts: array of cftypes to be added |
3062 | * @is_add: whether to add or remove | |
3063 | * | |
3064 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
3065 | * For removals, this function never fails. If addition fails, this |
3066 | * function doesn't remove files already added. The caller is responsible | |
3067 | * for cleaning up. | |
b1f28d31 | 3068 | */ |
2bb566cb TH |
3069 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
3070 | bool is_add) | |
ddbcc7e8 | 3071 | { |
03b1cde6 | 3072 | struct cftype *cft; |
b1f28d31 TH |
3073 | int ret; |
3074 | ||
01f6474c | 3075 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 TH |
3076 | |
3077 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3078 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3079 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3080 | continue; |
05ebb6e6 | 3081 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3082 | continue; |
d51f39b0 | 3083 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3084 | continue; |
d51f39b0 | 3085 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3086 | continue; |
3087 | ||
2739d3cc | 3088 | if (is_add) { |
2bb566cb | 3089 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 3090 | if (ret) { |
ed3d261b JP |
3091 | pr_warn("%s: failed to add %s, err=%d\n", |
3092 | __func__, cft->name, ret); | |
b1f28d31 TH |
3093 | return ret; |
3094 | } | |
2739d3cc LZ |
3095 | } else { |
3096 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3097 | } |
ddbcc7e8 | 3098 | } |
b1f28d31 | 3099 | return 0; |
ddbcc7e8 PM |
3100 | } |
3101 | ||
21a2d343 | 3102 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3103 | { |
3104 | LIST_HEAD(pending); | |
2bb566cb | 3105 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3106 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3107 | struct cgroup_subsys_state *css; |
9ccece80 | 3108 | int ret = 0; |
8e3f6541 | 3109 | |
01f6474c | 3110 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3111 | |
e8c82d20 | 3112 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3113 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3114 | struct cgroup *cgrp = css->cgroup; |
3115 | ||
e8c82d20 LZ |
3116 | if (cgroup_is_dead(cgrp)) |
3117 | continue; | |
3118 | ||
21a2d343 | 3119 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
3120 | if (ret) |
3121 | break; | |
8e3f6541 | 3122 | } |
21a2d343 TH |
3123 | |
3124 | if (is_add && !ret) | |
3125 | kernfs_activate(root->kn); | |
9ccece80 | 3126 | return ret; |
8e3f6541 TH |
3127 | } |
3128 | ||
2da440a2 | 3129 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3130 | { |
2bb566cb | 3131 | struct cftype *cft; |
8e3f6541 | 3132 | |
2bd59d48 TH |
3133 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3134 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3135 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3136 | kfree(cft->kf_ops); | |
3137 | cft->kf_ops = NULL; | |
2da440a2 | 3138 | cft->ss = NULL; |
a8ddc821 TH |
3139 | |
3140 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3141 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3142 | } |
2da440a2 TH |
3143 | } |
3144 | ||
2bd59d48 | 3145 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3146 | { |
3147 | struct cftype *cft; | |
3148 | ||
2bd59d48 TH |
3149 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3150 | struct kernfs_ops *kf_ops; | |
3151 | ||
0adb0704 TH |
3152 | WARN_ON(cft->ss || cft->kf_ops); |
3153 | ||
2bd59d48 TH |
3154 | if (cft->seq_start) |
3155 | kf_ops = &cgroup_kf_ops; | |
3156 | else | |
3157 | kf_ops = &cgroup_kf_single_ops; | |
3158 | ||
3159 | /* | |
3160 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3161 | * make a copy of kf_ops to set its atomic_write_len. | |
3162 | */ | |
3163 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3164 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3165 | if (!kf_ops) { | |
3166 | cgroup_exit_cftypes(cfts); | |
3167 | return -ENOMEM; | |
3168 | } | |
3169 | kf_ops->atomic_write_len = cft->max_write_len; | |
3170 | } | |
8e3f6541 | 3171 | |
2bd59d48 | 3172 | cft->kf_ops = kf_ops; |
2bb566cb | 3173 | cft->ss = ss; |
2bd59d48 | 3174 | } |
2bb566cb | 3175 | |
2bd59d48 | 3176 | return 0; |
2da440a2 TH |
3177 | } |
3178 | ||
21a2d343 TH |
3179 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3180 | { | |
01f6474c | 3181 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3182 | |
3183 | if (!cfts || !cfts[0].ss) | |
3184 | return -ENOENT; | |
3185 | ||
3186 | list_del(&cfts->node); | |
3187 | cgroup_apply_cftypes(cfts, false); | |
3188 | cgroup_exit_cftypes(cfts); | |
3189 | return 0; | |
8e3f6541 | 3190 | } |
8e3f6541 | 3191 | |
79578621 TH |
3192 | /** |
3193 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3194 | * @cfts: zero-length name terminated array of cftypes |
3195 | * | |
2bb566cb TH |
3196 | * Unregister @cfts. Files described by @cfts are removed from all |
3197 | * existing cgroups and all future cgroups won't have them either. This | |
3198 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3199 | * |
3200 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3201 | * registered. |
79578621 | 3202 | */ |
2bb566cb | 3203 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3204 | { |
21a2d343 | 3205 | int ret; |
79578621 | 3206 | |
01f6474c | 3207 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3208 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3209 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3210 | return ret; |
80b13586 TH |
3211 | } |
3212 | ||
8e3f6541 TH |
3213 | /** |
3214 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3215 | * @ss: target cgroup subsystem | |
3216 | * @cfts: zero-length name terminated array of cftypes | |
3217 | * | |
3218 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3219 | * existing cgroups to which @ss is attached and all future cgroups will | |
3220 | * have them too. This function can be called anytime whether @ss is | |
3221 | * attached or not. | |
3222 | * | |
3223 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3224 | * function currently returns 0 as long as @cfts registration is successful | |
3225 | * even if some file creation attempts on existing cgroups fail. | |
3226 | */ | |
2cf669a5 | 3227 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3228 | { |
9ccece80 | 3229 | int ret; |
8e3f6541 | 3230 | |
c731ae1d LZ |
3231 | if (ss->disabled) |
3232 | return 0; | |
3233 | ||
dc5736ed LZ |
3234 | if (!cfts || cfts[0].name[0] == '\0') |
3235 | return 0; | |
2bb566cb | 3236 | |
2bd59d48 TH |
3237 | ret = cgroup_init_cftypes(ss, cfts); |
3238 | if (ret) | |
3239 | return ret; | |
79578621 | 3240 | |
01f6474c | 3241 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3242 | |
0adb0704 | 3243 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3244 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3245 | if (ret) |
21a2d343 | 3246 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3247 | |
01f6474c | 3248 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3249 | return ret; |
79578621 TH |
3250 | } |
3251 | ||
a8ddc821 TH |
3252 | /** |
3253 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3254 | * @ss: target cgroup subsystem | |
3255 | * @cfts: zero-length name terminated array of cftypes | |
3256 | * | |
3257 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3258 | * the default hierarchy. | |
3259 | */ | |
3260 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3261 | { | |
3262 | struct cftype *cft; | |
3263 | ||
3264 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3265 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3266 | return cgroup_add_cftypes(ss, cfts); |
3267 | } | |
3268 | ||
3269 | /** | |
3270 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3271 | * @ss: target cgroup subsystem | |
3272 | * @cfts: zero-length name terminated array of cftypes | |
3273 | * | |
3274 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3275 | * the legacy hierarchies. | |
3276 | */ | |
2cf669a5 TH |
3277 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3278 | { | |
a8ddc821 TH |
3279 | struct cftype *cft; |
3280 | ||
fa8137be VG |
3281 | /* |
3282 | * If legacy_flies_on_dfl, we want to show the legacy files on the | |
3283 | * dfl hierarchy but iff the target subsystem hasn't been updated | |
3284 | * for the dfl hierarchy yet. | |
3285 | */ | |
3286 | if (!cgroup_legacy_files_on_dfl || | |
3287 | ss->dfl_cftypes != ss->legacy_cftypes) { | |
3288 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
3289 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
3290 | } | |
3291 | ||
2cf669a5 TH |
3292 | return cgroup_add_cftypes(ss, cfts); |
3293 | } | |
3294 | ||
a043e3b2 LZ |
3295 | /** |
3296 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3297 | * @cgrp: the cgroup in question | |
3298 | * | |
3299 | * Return the number of tasks in the cgroup. | |
3300 | */ | |
07bc356e | 3301 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3302 | { |
3303 | int count = 0; | |
69d0206c | 3304 | struct cgrp_cset_link *link; |
817929ec | 3305 | |
96d365e0 | 3306 | down_read(&css_set_rwsem); |
69d0206c TH |
3307 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3308 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3309 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3310 | return count; |
3311 | } | |
3312 | ||
53fa5261 | 3313 | /** |
492eb21b | 3314 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3315 | * @pos: the current position (%NULL to initiate traversal) |
3316 | * @parent: css whose children to walk | |
53fa5261 | 3317 | * |
c2931b70 | 3318 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3319 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3320 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3321 | * be returned regardless of their states. | |
3322 | * | |
3323 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3324 | * css which finished ->css_online() is guaranteed to be visible in the | |
3325 | * future iterations and will stay visible until the last reference is put. | |
3326 | * A css which hasn't finished ->css_online() or already finished | |
3327 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3328 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3329 | */ |
c2931b70 TH |
3330 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3331 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3332 | { |
c2931b70 | 3333 | struct cgroup_subsys_state *next; |
53fa5261 | 3334 | |
8353da1f | 3335 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3336 | |
3337 | /* | |
de3f0341 TH |
3338 | * @pos could already have been unlinked from the sibling list. |
3339 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3340 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3341 | * @pos is taken off list, at which time its next pointer is valid, | |
3342 | * and, as releases are serialized, the one pointed to by the next | |
3343 | * pointer is guaranteed to not have started release yet. This | |
3344 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3345 | * critical section, the one pointed to by its next pointer is | |
3346 | * guaranteed to not have finished its RCU grace period even if we | |
3347 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3348 | * |
de3f0341 TH |
3349 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3350 | * dereferenced; however, as each css is given a monotonically | |
3351 | * increasing unique serial number and always appended to the | |
3352 | * sibling list, the next one can be found by walking the parent's | |
3353 | * children until the first css with higher serial number than | |
3354 | * @pos's. While this path can be slower, it happens iff iteration | |
3355 | * races against release and the race window is very small. | |
53fa5261 | 3356 | */ |
3b287a50 | 3357 | if (!pos) { |
c2931b70 TH |
3358 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3359 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3360 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3361 | } else { |
c2931b70 | 3362 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3363 | if (next->serial_nr > pos->serial_nr) |
3364 | break; | |
53fa5261 TH |
3365 | } |
3366 | ||
3b281afb TH |
3367 | /* |
3368 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3369 | * the next sibling. |
3b281afb | 3370 | */ |
c2931b70 TH |
3371 | if (&next->sibling != &parent->children) |
3372 | return next; | |
3b281afb | 3373 | return NULL; |
53fa5261 | 3374 | } |
53fa5261 | 3375 | |
574bd9f7 | 3376 | /** |
492eb21b | 3377 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3378 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3379 | * @root: css whose descendants to walk |
574bd9f7 | 3380 | * |
492eb21b | 3381 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3382 | * to visit for pre-order traversal of @root's descendants. @root is |
3383 | * included in the iteration and the first node to be visited. | |
75501a6d | 3384 | * |
87fb54f1 TH |
3385 | * While this function requires cgroup_mutex or RCU read locking, it |
3386 | * doesn't require the whole traversal to be contained in a single critical | |
3387 | * section. This function will return the correct next descendant as long | |
3388 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
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. | |
574bd9f7 | 3396 | */ |
492eb21b TH |
3397 | struct cgroup_subsys_state * |
3398 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3399 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3400 | { |
492eb21b | 3401 | struct cgroup_subsys_state *next; |
574bd9f7 | 3402 | |
8353da1f | 3403 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3404 | |
bd8815a6 | 3405 | /* if first iteration, visit @root */ |
7805d000 | 3406 | if (!pos) |
bd8815a6 | 3407 | return root; |
574bd9f7 TH |
3408 | |
3409 | /* visit the first child if exists */ | |
492eb21b | 3410 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3411 | if (next) |
3412 | return next; | |
3413 | ||
3414 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3415 | while (pos != root) { |
5c9d535b | 3416 | next = css_next_child(pos, pos->parent); |
75501a6d | 3417 | if (next) |
574bd9f7 | 3418 | return next; |
5c9d535b | 3419 | pos = pos->parent; |
7805d000 | 3420 | } |
574bd9f7 TH |
3421 | |
3422 | return NULL; | |
3423 | } | |
574bd9f7 | 3424 | |
12a9d2fe | 3425 | /** |
492eb21b TH |
3426 | * css_rightmost_descendant - return the rightmost descendant of a css |
3427 | * @pos: css of interest | |
12a9d2fe | 3428 | * |
492eb21b TH |
3429 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3430 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3431 | * subtree of @pos. |
75501a6d | 3432 | * |
87fb54f1 TH |
3433 | * While this function requires cgroup_mutex or RCU read locking, it |
3434 | * doesn't require the whole traversal to be contained in a single critical | |
3435 | * section. This function will return the correct rightmost descendant as | |
3436 | * long as @pos is accessible. | |
12a9d2fe | 3437 | */ |
492eb21b TH |
3438 | struct cgroup_subsys_state * |
3439 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3440 | { |
492eb21b | 3441 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3442 | |
8353da1f | 3443 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3444 | |
3445 | do { | |
3446 | last = pos; | |
3447 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3448 | pos = NULL; | |
492eb21b | 3449 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3450 | pos = tmp; |
3451 | } while (pos); | |
3452 | ||
3453 | return last; | |
3454 | } | |
12a9d2fe | 3455 | |
492eb21b TH |
3456 | static struct cgroup_subsys_state * |
3457 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3458 | { |
492eb21b | 3459 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3460 | |
3461 | do { | |
3462 | last = pos; | |
492eb21b | 3463 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3464 | } while (pos); |
3465 | ||
3466 | return last; | |
3467 | } | |
3468 | ||
3469 | /** | |
492eb21b | 3470 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3471 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3472 | * @root: css whose descendants to walk |
574bd9f7 | 3473 | * |
492eb21b | 3474 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3475 | * to visit for post-order traversal of @root's descendants. @root is |
3476 | * included in the iteration and the last node to be visited. | |
75501a6d | 3477 | * |
87fb54f1 TH |
3478 | * While this function requires cgroup_mutex or RCU read locking, it |
3479 | * doesn't require the whole traversal to be contained in a single critical | |
3480 | * section. This function will return the correct next descendant as long | |
3481 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3482 | * @cgroup. | |
c2931b70 TH |
3483 | * |
3484 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3485 | * css which finished ->css_online() is guaranteed to be visible in the | |
3486 | * future iterations and will stay visible until the last reference is put. | |
3487 | * A css which hasn't finished ->css_online() or already finished | |
3488 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3489 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3490 | */ |
492eb21b TH |
3491 | struct cgroup_subsys_state * |
3492 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3493 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3494 | { |
492eb21b | 3495 | struct cgroup_subsys_state *next; |
574bd9f7 | 3496 | |
8353da1f | 3497 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3498 | |
58b79a91 TH |
3499 | /* if first iteration, visit leftmost descendant which may be @root */ |
3500 | if (!pos) | |
3501 | return css_leftmost_descendant(root); | |
574bd9f7 | 3502 | |
bd8815a6 TH |
3503 | /* if we visited @root, we're done */ |
3504 | if (pos == root) | |
3505 | return NULL; | |
3506 | ||
574bd9f7 | 3507 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3508 | next = css_next_child(pos, pos->parent); |
75501a6d | 3509 | if (next) |
492eb21b | 3510 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3511 | |
3512 | /* no sibling left, visit parent */ | |
5c9d535b | 3513 | return pos->parent; |
574bd9f7 | 3514 | } |
574bd9f7 | 3515 | |
f3d46500 TH |
3516 | /** |
3517 | * css_has_online_children - does a css have online children | |
3518 | * @css: the target css | |
3519 | * | |
3520 | * Returns %true if @css has any online children; otherwise, %false. This | |
3521 | * function can be called from any context but the caller is responsible | |
3522 | * for synchronizing against on/offlining as necessary. | |
3523 | */ | |
3524 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3525 | { |
f3d46500 TH |
3526 | struct cgroup_subsys_state *child; |
3527 | bool ret = false; | |
cbc125ef TH |
3528 | |
3529 | rcu_read_lock(); | |
f3d46500 | 3530 | css_for_each_child(child, css) { |
99bae5f9 | 3531 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3532 | ret = true; |
3533 | break; | |
cbc125ef TH |
3534 | } |
3535 | } | |
3536 | rcu_read_unlock(); | |
f3d46500 | 3537 | return ret; |
574bd9f7 | 3538 | } |
574bd9f7 | 3539 | |
0942eeee | 3540 | /** |
72ec7029 | 3541 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3542 | * @it: the iterator to advance |
3543 | * | |
3544 | * Advance @it to the next css_set to walk. | |
d515876e | 3545 | */ |
72ec7029 | 3546 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3547 | { |
0f0a2b4f | 3548 | struct list_head *l = it->cset_pos; |
d515876e TH |
3549 | struct cgrp_cset_link *link; |
3550 | struct css_set *cset; | |
3551 | ||
3552 | /* Advance to the next non-empty css_set */ | |
3553 | do { | |
3554 | l = l->next; | |
0f0a2b4f TH |
3555 | if (l == it->cset_head) { |
3556 | it->cset_pos = NULL; | |
d515876e TH |
3557 | return; |
3558 | } | |
3ebb2b6e TH |
3559 | |
3560 | if (it->ss) { | |
3561 | cset = container_of(l, struct css_set, | |
3562 | e_cset_node[it->ss->id]); | |
3563 | } else { | |
3564 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3565 | cset = link->cset; | |
3566 | } | |
c7561128 TH |
3567 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3568 | ||
0f0a2b4f | 3569 | it->cset_pos = l; |
c7561128 TH |
3570 | |
3571 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3572 | it->task_pos = cset->tasks.next; |
c7561128 | 3573 | else |
0f0a2b4f TH |
3574 | it->task_pos = cset->mg_tasks.next; |
3575 | ||
3576 | it->tasks_head = &cset->tasks; | |
3577 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3578 | } |
3579 | ||
0942eeee | 3580 | /** |
72ec7029 TH |
3581 | * css_task_iter_start - initiate task iteration |
3582 | * @css: the css to walk tasks of | |
0942eeee TH |
3583 | * @it: the task iterator to use |
3584 | * | |
72ec7029 TH |
3585 | * Initiate iteration through the tasks of @css. The caller can call |
3586 | * css_task_iter_next() to walk through the tasks until the function | |
3587 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3588 | * called. | |
0942eeee TH |
3589 | * |
3590 | * Note that this function acquires a lock which is released when the | |
3591 | * iteration finishes. The caller can't sleep while iteration is in | |
3592 | * progress. | |
3593 | */ | |
72ec7029 TH |
3594 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3595 | struct css_task_iter *it) | |
96d365e0 | 3596 | __acquires(css_set_rwsem) |
817929ec | 3597 | { |
56fde9e0 TH |
3598 | /* no one should try to iterate before mounting cgroups */ |
3599 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3600 | |
96d365e0 | 3601 | down_read(&css_set_rwsem); |
c59cd3d8 | 3602 | |
3ebb2b6e TH |
3603 | it->ss = css->ss; |
3604 | ||
3605 | if (it->ss) | |
3606 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3607 | else | |
3608 | it->cset_pos = &css->cgroup->cset_links; | |
3609 | ||
0f0a2b4f | 3610 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3611 | |
72ec7029 | 3612 | css_advance_task_iter(it); |
817929ec PM |
3613 | } |
3614 | ||
0942eeee | 3615 | /** |
72ec7029 | 3616 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3617 | * @it: the task iterator being iterated |
3618 | * | |
3619 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3620 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3621 | * reaches the end. | |
0942eeee | 3622 | */ |
72ec7029 | 3623 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3624 | { |
3625 | struct task_struct *res; | |
0f0a2b4f | 3626 | struct list_head *l = it->task_pos; |
817929ec PM |
3627 | |
3628 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3629 | if (!it->cset_pos) |
817929ec PM |
3630 | return NULL; |
3631 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3632 | |
3633 | /* | |
3634 | * Advance iterator to find next entry. cset->tasks is consumed | |
3635 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3636 | * next cset. | |
3637 | */ | |
817929ec | 3638 | l = l->next; |
c7561128 | 3639 | |
0f0a2b4f TH |
3640 | if (l == it->tasks_head) |
3641 | l = it->mg_tasks_head->next; | |
c7561128 | 3642 | |
0f0a2b4f | 3643 | if (l == it->mg_tasks_head) |
72ec7029 | 3644 | css_advance_task_iter(it); |
c7561128 | 3645 | else |
0f0a2b4f | 3646 | it->task_pos = l; |
c7561128 | 3647 | |
817929ec PM |
3648 | return res; |
3649 | } | |
3650 | ||
0942eeee | 3651 | /** |
72ec7029 | 3652 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3653 | * @it: the task iterator to finish |
3654 | * | |
72ec7029 | 3655 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3656 | */ |
72ec7029 | 3657 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3658 | __releases(css_set_rwsem) |
31a7df01 | 3659 | { |
96d365e0 | 3660 | up_read(&css_set_rwsem); |
31a7df01 CW |
3661 | } |
3662 | ||
3663 | /** | |
8cc99345 TH |
3664 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3665 | * @to: cgroup to which the tasks will be moved | |
3666 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3667 | * |
eaf797ab TH |
3668 | * Locking rules between cgroup_post_fork() and the migration path |
3669 | * guarantee that, if a task is forking while being migrated, the new child | |
3670 | * is guaranteed to be either visible in the source cgroup after the | |
3671 | * parent's migration is complete or put into the target cgroup. No task | |
3672 | * can slip out of migration through forking. | |
31a7df01 | 3673 | */ |
8cc99345 | 3674 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3675 | { |
952aaa12 TH |
3676 | LIST_HEAD(preloaded_csets); |
3677 | struct cgrp_cset_link *link; | |
72ec7029 | 3678 | struct css_task_iter it; |
e406d1cf | 3679 | struct task_struct *task; |
952aaa12 | 3680 | int ret; |
31a7df01 | 3681 | |
952aaa12 | 3682 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3683 | |
952aaa12 TH |
3684 | /* all tasks in @from are being moved, all csets are source */ |
3685 | down_read(&css_set_rwsem); | |
3686 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3687 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3688 | up_read(&css_set_rwsem); | |
31a7df01 | 3689 | |
952aaa12 TH |
3690 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3691 | if (ret) | |
3692 | goto out_err; | |
8cc99345 | 3693 | |
952aaa12 TH |
3694 | /* |
3695 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3696 | * ->can_attach() fails. | |
3697 | */ | |
e406d1cf | 3698 | do { |
9d800df1 | 3699 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
3700 | task = css_task_iter_next(&it); |
3701 | if (task) | |
3702 | get_task_struct(task); | |
3703 | css_task_iter_end(&it); | |
3704 | ||
3705 | if (task) { | |
952aaa12 | 3706 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3707 | put_task_struct(task); |
3708 | } | |
3709 | } while (task && !ret); | |
952aaa12 TH |
3710 | out_err: |
3711 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3712 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3713 | return ret; |
8cc99345 TH |
3714 | } |
3715 | ||
bbcb81d0 | 3716 | /* |
102a775e | 3717 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3718 | * |
3719 | * Reading this file can return large amounts of data if a cgroup has | |
3720 | * *lots* of attached tasks. So it may need several calls to read(), | |
3721 | * but we cannot guarantee that the information we produce is correct | |
3722 | * unless we produce it entirely atomically. | |
3723 | * | |
bbcb81d0 | 3724 | */ |
bbcb81d0 | 3725 | |
24528255 LZ |
3726 | /* which pidlist file are we talking about? */ |
3727 | enum cgroup_filetype { | |
3728 | CGROUP_FILE_PROCS, | |
3729 | CGROUP_FILE_TASKS, | |
3730 | }; | |
3731 | ||
3732 | /* | |
3733 | * A pidlist is a list of pids that virtually represents the contents of one | |
3734 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3735 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3736 | * to the cgroup. | |
3737 | */ | |
3738 | struct cgroup_pidlist { | |
3739 | /* | |
3740 | * used to find which pidlist is wanted. doesn't change as long as | |
3741 | * this particular list stays in the list. | |
3742 | */ | |
3743 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3744 | /* array of xids */ | |
3745 | pid_t *list; | |
3746 | /* how many elements the above list has */ | |
3747 | int length; | |
24528255 LZ |
3748 | /* each of these stored in a list by its cgroup */ |
3749 | struct list_head links; | |
3750 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3751 | struct cgroup *owner; | |
b1a21367 TH |
3752 | /* for delayed destruction */ |
3753 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3754 | }; |
3755 | ||
d1d9fd33 BB |
3756 | /* |
3757 | * The following two functions "fix" the issue where there are more pids | |
3758 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3759 | * TODO: replace with a kernel-wide solution to this problem | |
3760 | */ | |
3761 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3762 | static void *pidlist_allocate(int count) | |
3763 | { | |
3764 | if (PIDLIST_TOO_LARGE(count)) | |
3765 | return vmalloc(count * sizeof(pid_t)); | |
3766 | else | |
3767 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3768 | } | |
b1a21367 | 3769 | |
d1d9fd33 BB |
3770 | static void pidlist_free(void *p) |
3771 | { | |
3772 | if (is_vmalloc_addr(p)) | |
3773 | vfree(p); | |
3774 | else | |
3775 | kfree(p); | |
3776 | } | |
d1d9fd33 | 3777 | |
b1a21367 TH |
3778 | /* |
3779 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3780 | * should be left afterwards. | |
3781 | */ | |
3782 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3783 | { | |
3784 | struct cgroup_pidlist *l, *tmp_l; | |
3785 | ||
3786 | mutex_lock(&cgrp->pidlist_mutex); | |
3787 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3788 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3789 | mutex_unlock(&cgrp->pidlist_mutex); | |
3790 | ||
3791 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3792 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3793 | } | |
3794 | ||
3795 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3796 | { | |
3797 | struct delayed_work *dwork = to_delayed_work(work); | |
3798 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3799 | destroy_dwork); | |
3800 | struct cgroup_pidlist *tofree = NULL; | |
3801 | ||
3802 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3803 | |
3804 | /* | |
04502365 TH |
3805 | * Destroy iff we didn't get queued again. The state won't change |
3806 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3807 | */ |
04502365 | 3808 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3809 | list_del(&l->links); |
3810 | pidlist_free(l->list); | |
3811 | put_pid_ns(l->key.ns); | |
3812 | tofree = l; | |
3813 | } | |
3814 | ||
b1a21367 TH |
3815 | mutex_unlock(&l->owner->pidlist_mutex); |
3816 | kfree(tofree); | |
3817 | } | |
3818 | ||
bbcb81d0 | 3819 | /* |
102a775e | 3820 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3821 | * Returns the number of unique elements. |
bbcb81d0 | 3822 | */ |
6ee211ad | 3823 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3824 | { |
102a775e | 3825 | int src, dest = 1; |
102a775e BB |
3826 | |
3827 | /* | |
3828 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3829 | * edge cases first; no work needs to be done for either | |
3830 | */ | |
3831 | if (length == 0 || length == 1) | |
3832 | return length; | |
3833 | /* src and dest walk down the list; dest counts unique elements */ | |
3834 | for (src = 1; src < length; src++) { | |
3835 | /* find next unique element */ | |
3836 | while (list[src] == list[src-1]) { | |
3837 | src++; | |
3838 | if (src == length) | |
3839 | goto after; | |
3840 | } | |
3841 | /* dest always points to where the next unique element goes */ | |
3842 | list[dest] = list[src]; | |
3843 | dest++; | |
3844 | } | |
3845 | after: | |
102a775e BB |
3846 | return dest; |
3847 | } | |
3848 | ||
afb2bc14 TH |
3849 | /* |
3850 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3851 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3852 | * different per namespace, each namespace needs differently sorted list, | |
3853 | * making it impossible to use, for example, single rbtree of member tasks | |
3854 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3855 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3856 | * pidlists keyed by cgroup and namespace. | |
3857 | * | |
3858 | * All this extra complexity was caused by the original implementation | |
3859 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
3860 | * want to do away with it. Explicitly scramble sort order if on the |
3861 | * default hierarchy so that no such expectation exists in the new | |
3862 | * interface. | |
afb2bc14 TH |
3863 | * |
3864 | * Scrambling is done by swapping every two consecutive bits, which is | |
3865 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3866 | */ | |
3867 | static pid_t pid_fry(pid_t pid) | |
3868 | { | |
3869 | unsigned a = pid & 0x55555555; | |
3870 | unsigned b = pid & 0xAAAAAAAA; | |
3871 | ||
3872 | return (a << 1) | (b >> 1); | |
3873 | } | |
3874 | ||
3875 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3876 | { | |
aa6ec29b | 3877 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
3878 | return pid_fry(pid); |
3879 | else | |
3880 | return pid; | |
3881 | } | |
3882 | ||
102a775e BB |
3883 | static int cmppid(const void *a, const void *b) |
3884 | { | |
3885 | return *(pid_t *)a - *(pid_t *)b; | |
3886 | } | |
3887 | ||
afb2bc14 TH |
3888 | static int fried_cmppid(const void *a, const void *b) |
3889 | { | |
3890 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3891 | } | |
3892 | ||
e6b81710 TH |
3893 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3894 | enum cgroup_filetype type) | |
3895 | { | |
3896 | struct cgroup_pidlist *l; | |
3897 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3898 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3899 | ||
3900 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3901 | ||
3902 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3903 | if (l->key.type == type && l->key.ns == ns) | |
3904 | return l; | |
3905 | return NULL; | |
3906 | } | |
3907 | ||
72a8cb30 BB |
3908 | /* |
3909 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3910 | * returns with the lock on that pidlist already held, and takes care | |
3911 | * of the use count, or returns NULL with no locks held if we're out of | |
3912 | * memory. | |
3913 | */ | |
e6b81710 TH |
3914 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3915 | enum cgroup_filetype type) | |
72a8cb30 BB |
3916 | { |
3917 | struct cgroup_pidlist *l; | |
b70cc5fd | 3918 | |
e6b81710 TH |
3919 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3920 | ||
3921 | l = cgroup_pidlist_find(cgrp, type); | |
3922 | if (l) | |
3923 | return l; | |
3924 | ||
72a8cb30 | 3925 | /* entry not found; create a new one */ |
f4f4be2b | 3926 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3927 | if (!l) |
72a8cb30 | 3928 | return l; |
e6b81710 | 3929 | |
b1a21367 | 3930 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3931 | l->key.type = type; |
e6b81710 TH |
3932 | /* don't need task_nsproxy() if we're looking at ourself */ |
3933 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3934 | l->owner = cgrp; |
3935 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3936 | return l; |
3937 | } | |
3938 | ||
102a775e BB |
3939 | /* |
3940 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3941 | */ | |
72a8cb30 BB |
3942 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3943 | struct cgroup_pidlist **lp) | |
102a775e BB |
3944 | { |
3945 | pid_t *array; | |
3946 | int length; | |
3947 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3948 | struct css_task_iter it; |
817929ec | 3949 | struct task_struct *tsk; |
102a775e BB |
3950 | struct cgroup_pidlist *l; |
3951 | ||
4bac00d1 TH |
3952 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3953 | ||
102a775e BB |
3954 | /* |
3955 | * If cgroup gets more users after we read count, we won't have | |
3956 | * enough space - tough. This race is indistinguishable to the | |
3957 | * caller from the case that the additional cgroup users didn't | |
3958 | * show up until sometime later on. | |
3959 | */ | |
3960 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3961 | array = pidlist_allocate(length); |
102a775e BB |
3962 | if (!array) |
3963 | return -ENOMEM; | |
3964 | /* now, populate the array */ | |
9d800df1 | 3965 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 3966 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 3967 | if (unlikely(n == length)) |
817929ec | 3968 | break; |
102a775e | 3969 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3970 | if (type == CGROUP_FILE_PROCS) |
3971 | pid = task_tgid_vnr(tsk); | |
3972 | else | |
3973 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3974 | if (pid > 0) /* make sure to only use valid results */ |
3975 | array[n++] = pid; | |
817929ec | 3976 | } |
72ec7029 | 3977 | css_task_iter_end(&it); |
102a775e BB |
3978 | length = n; |
3979 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 3980 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
3981 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
3982 | else | |
3983 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3984 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3985 | length = pidlist_uniq(array, length); |
e6b81710 | 3986 | |
e6b81710 | 3987 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3988 | if (!l) { |
d1d9fd33 | 3989 | pidlist_free(array); |
72a8cb30 | 3990 | return -ENOMEM; |
102a775e | 3991 | } |
e6b81710 TH |
3992 | |
3993 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3994 | pidlist_free(l->list); |
102a775e BB |
3995 | l->list = array; |
3996 | l->length = length; | |
72a8cb30 | 3997 | *lp = l; |
102a775e | 3998 | return 0; |
bbcb81d0 PM |
3999 | } |
4000 | ||
846c7bb0 | 4001 | /** |
a043e3b2 | 4002 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4003 | * @stats: cgroupstats to fill information into |
4004 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4005 | * been requested. | |
a043e3b2 LZ |
4006 | * |
4007 | * Build and fill cgroupstats so that taskstats can export it to user | |
4008 | * space. | |
846c7bb0 BS |
4009 | */ |
4010 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4011 | { | |
2bd59d48 | 4012 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4013 | struct cgroup *cgrp; |
72ec7029 | 4014 | struct css_task_iter it; |
846c7bb0 | 4015 | struct task_struct *tsk; |
33d283be | 4016 | |
2bd59d48 TH |
4017 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4018 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4019 | kernfs_type(kn) != KERNFS_DIR) | |
4020 | return -EINVAL; | |
4021 | ||
bad34660 LZ |
4022 | mutex_lock(&cgroup_mutex); |
4023 | ||
846c7bb0 | 4024 | /* |
2bd59d48 | 4025 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4026 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4027 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4028 | */ |
2bd59d48 TH |
4029 | rcu_read_lock(); |
4030 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4031 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4032 | rcu_read_unlock(); |
bad34660 | 4033 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4034 | return -ENOENT; |
4035 | } | |
bad34660 | 4036 | rcu_read_unlock(); |
846c7bb0 | 4037 | |
9d800df1 | 4038 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4039 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4040 | switch (tsk->state) { |
4041 | case TASK_RUNNING: | |
4042 | stats->nr_running++; | |
4043 | break; | |
4044 | case TASK_INTERRUPTIBLE: | |
4045 | stats->nr_sleeping++; | |
4046 | break; | |
4047 | case TASK_UNINTERRUPTIBLE: | |
4048 | stats->nr_uninterruptible++; | |
4049 | break; | |
4050 | case TASK_STOPPED: | |
4051 | stats->nr_stopped++; | |
4052 | break; | |
4053 | default: | |
4054 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4055 | stats->nr_io_wait++; | |
4056 | break; | |
4057 | } | |
4058 | } | |
72ec7029 | 4059 | css_task_iter_end(&it); |
846c7bb0 | 4060 | |
bad34660 | 4061 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4062 | return 0; |
846c7bb0 BS |
4063 | } |
4064 | ||
8f3ff208 | 4065 | |
bbcb81d0 | 4066 | /* |
102a775e | 4067 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4068 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4069 | * in the cgroup->l->list array. |
bbcb81d0 | 4070 | */ |
cc31edce | 4071 | |
102a775e | 4072 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4073 | { |
cc31edce PM |
4074 | /* |
4075 | * Initially we receive a position value that corresponds to | |
4076 | * one more than the last pid shown (or 0 on the first call or | |
4077 | * after a seek to the start). Use a binary-search to find the | |
4078 | * next pid to display, if any | |
4079 | */ | |
2bd59d48 | 4080 | struct kernfs_open_file *of = s->private; |
7da11279 | 4081 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4082 | struct cgroup_pidlist *l; |
7da11279 | 4083 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4084 | int index = 0, pid = *pos; |
4bac00d1 TH |
4085 | int *iter, ret; |
4086 | ||
4087 | mutex_lock(&cgrp->pidlist_mutex); | |
4088 | ||
4089 | /* | |
5d22444f | 4090 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4091 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4092 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4093 | * could already have been destroyed. |
4094 | */ | |
5d22444f TH |
4095 | if (of->priv) |
4096 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4097 | |
4098 | /* | |
4099 | * Either this is the first start() after open or the matching | |
4100 | * pidlist has been destroyed inbetween. Create a new one. | |
4101 | */ | |
5d22444f TH |
4102 | if (!of->priv) { |
4103 | ret = pidlist_array_load(cgrp, type, | |
4104 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4105 | if (ret) |
4106 | return ERR_PTR(ret); | |
4107 | } | |
5d22444f | 4108 | l = of->priv; |
cc31edce | 4109 | |
cc31edce | 4110 | if (pid) { |
102a775e | 4111 | int end = l->length; |
20777766 | 4112 | |
cc31edce PM |
4113 | while (index < end) { |
4114 | int mid = (index + end) / 2; | |
afb2bc14 | 4115 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4116 | index = mid; |
4117 | break; | |
afb2bc14 | 4118 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4119 | index = mid + 1; |
4120 | else | |
4121 | end = mid; | |
4122 | } | |
4123 | } | |
4124 | /* If we're off the end of the array, we're done */ | |
102a775e | 4125 | if (index >= l->length) |
cc31edce PM |
4126 | return NULL; |
4127 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4128 | iter = l->list + index; |
afb2bc14 | 4129 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4130 | return iter; |
4131 | } | |
4132 | ||
102a775e | 4133 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4134 | { |
2bd59d48 | 4135 | struct kernfs_open_file *of = s->private; |
5d22444f | 4136 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4137 | |
5d22444f TH |
4138 | if (l) |
4139 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4140 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4141 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4142 | } |
4143 | ||
102a775e | 4144 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4145 | { |
2bd59d48 | 4146 | struct kernfs_open_file *of = s->private; |
5d22444f | 4147 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4148 | pid_t *p = v; |
4149 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4150 | /* |
4151 | * Advance to the next pid in the array. If this goes off the | |
4152 | * end, we're done | |
4153 | */ | |
4154 | p++; | |
4155 | if (p >= end) { | |
4156 | return NULL; | |
4157 | } else { | |
7da11279 | 4158 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4159 | return p; |
4160 | } | |
4161 | } | |
4162 | ||
102a775e | 4163 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
4164 | { |
4165 | return seq_printf(s, "%d\n", *(int *)v); | |
4166 | } | |
bbcb81d0 | 4167 | |
182446d0 TH |
4168 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4169 | struct cftype *cft) | |
81a6a5cd | 4170 | { |
182446d0 | 4171 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4172 | } |
4173 | ||
182446d0 TH |
4174 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4175 | struct cftype *cft, u64 val) | |
6379c106 | 4176 | { |
6379c106 | 4177 | if (val) |
182446d0 | 4178 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4179 | else |
182446d0 | 4180 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4181 | return 0; |
4182 | } | |
4183 | ||
182446d0 TH |
4184 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4185 | struct cftype *cft) | |
97978e6d | 4186 | { |
182446d0 | 4187 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4188 | } |
4189 | ||
182446d0 TH |
4190 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4191 | struct cftype *cft, u64 val) | |
97978e6d DL |
4192 | { |
4193 | if (val) | |
182446d0 | 4194 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4195 | else |
182446d0 | 4196 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4197 | return 0; |
4198 | } | |
4199 | ||
a14c6874 TH |
4200 | /* cgroup core interface files for the default hierarchy */ |
4201 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4202 | { |
d5c56ced | 4203 | .name = "cgroup.procs", |
6612f05b TH |
4204 | .seq_start = cgroup_pidlist_start, |
4205 | .seq_next = cgroup_pidlist_next, | |
4206 | .seq_stop = cgroup_pidlist_stop, | |
4207 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4208 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4209 | .write = cgroup_procs_write, |
74a1166d | 4210 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 4211 | }, |
f8f22e53 TH |
4212 | { |
4213 | .name = "cgroup.controllers", | |
a14c6874 | 4214 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4215 | .seq_show = cgroup_root_controllers_show, |
4216 | }, | |
4217 | { | |
4218 | .name = "cgroup.controllers", | |
a14c6874 | 4219 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4220 | .seq_show = cgroup_controllers_show, |
4221 | }, | |
4222 | { | |
4223 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4224 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4225 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4226 | }, |
842b597e TH |
4227 | { |
4228 | .name = "cgroup.populated", | |
a14c6874 | 4229 | .flags = CFTYPE_NOT_ON_ROOT, |
842b597e TH |
4230 | .seq_show = cgroup_populated_show, |
4231 | }, | |
a14c6874 TH |
4232 | { } /* terminate */ |
4233 | }; | |
d5c56ced | 4234 | |
a14c6874 TH |
4235 | /* cgroup core interface files for the legacy hierarchies */ |
4236 | static struct cftype cgroup_legacy_base_files[] = { | |
4237 | { | |
4238 | .name = "cgroup.procs", | |
4239 | .seq_start = cgroup_pidlist_start, | |
4240 | .seq_next = cgroup_pidlist_next, | |
4241 | .seq_stop = cgroup_pidlist_stop, | |
4242 | .seq_show = cgroup_pidlist_show, | |
4243 | .private = CGROUP_FILE_PROCS, | |
4244 | .write = cgroup_procs_write, | |
4245 | .mode = S_IRUGO | S_IWUSR, | |
4246 | }, | |
4247 | { | |
4248 | .name = "cgroup.clone_children", | |
4249 | .read_u64 = cgroup_clone_children_read, | |
4250 | .write_u64 = cgroup_clone_children_write, | |
4251 | }, | |
4252 | { | |
4253 | .name = "cgroup.sane_behavior", | |
4254 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4255 | .seq_show = cgroup_sane_behavior_show, | |
4256 | }, | |
d5c56ced TH |
4257 | { |
4258 | .name = "tasks", | |
6612f05b TH |
4259 | .seq_start = cgroup_pidlist_start, |
4260 | .seq_next = cgroup_pidlist_next, | |
4261 | .seq_stop = cgroup_pidlist_stop, | |
4262 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4263 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4264 | .write = cgroup_tasks_write, |
d5c56ced TH |
4265 | .mode = S_IRUGO | S_IWUSR, |
4266 | }, | |
4267 | { | |
4268 | .name = "notify_on_release", | |
d5c56ced TH |
4269 | .read_u64 = cgroup_read_notify_on_release, |
4270 | .write_u64 = cgroup_write_notify_on_release, | |
4271 | }, | |
6e6ff25b TH |
4272 | { |
4273 | .name = "release_agent", | |
a14c6874 | 4274 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4275 | .seq_show = cgroup_release_agent_show, |
451af504 | 4276 | .write = cgroup_release_agent_write, |
5f469907 | 4277 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4278 | }, |
db0416b6 | 4279 | { } /* terminate */ |
bbcb81d0 PM |
4280 | }; |
4281 | ||
13af07df | 4282 | /** |
628f7cd4 | 4283 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4284 | * @cgrp: target cgroup |
13af07df | 4285 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4286 | * |
4287 | * On failure, no file is added. | |
13af07df | 4288 | */ |
69dfa00c | 4289 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
ddbcc7e8 | 4290 | { |
ddbcc7e8 | 4291 | struct cgroup_subsys *ss; |
b420ba7d | 4292 | int i, ret = 0; |
bbcb81d0 | 4293 | |
8e3f6541 | 4294 | /* process cftsets of each subsystem */ |
b420ba7d | 4295 | for_each_subsys(ss, i) { |
0adb0704 | 4296 | struct cftype *cfts; |
b420ba7d | 4297 | |
69dfa00c | 4298 | if (!(subsys_mask & (1 << i))) |
13af07df | 4299 | continue; |
8e3f6541 | 4300 | |
0adb0704 TH |
4301 | list_for_each_entry(cfts, &ss->cfts, node) { |
4302 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4303 | if (ret < 0) |
4304 | goto err; | |
4305 | } | |
ddbcc7e8 | 4306 | } |
ddbcc7e8 | 4307 | return 0; |
bee55099 TH |
4308 | err: |
4309 | cgroup_clear_dir(cgrp, subsys_mask); | |
4310 | return ret; | |
ddbcc7e8 PM |
4311 | } |
4312 | ||
0c21ead1 TH |
4313 | /* |
4314 | * css destruction is four-stage process. | |
4315 | * | |
4316 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4317 | * Implemented in kill_css(). | |
4318 | * | |
4319 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4320 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4321 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4322 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4323 | * |
4324 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4325 | * accessors are inside RCU read sections. css_release() schedules the | |
4326 | * RCU callback. | |
4327 | * | |
4328 | * 4. After the grace period, the css can be freed. Implemented in | |
4329 | * css_free_work_fn(). | |
4330 | * | |
4331 | * It is actually hairier because both step 2 and 4 require process context | |
4332 | * and thus involve punting to css->destroy_work adding two additional | |
4333 | * steps to the already complex sequence. | |
4334 | */ | |
35ef10da | 4335 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4336 | { |
4337 | struct cgroup_subsys_state *css = | |
35ef10da | 4338 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4339 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4340 | |
9a1049da TH |
4341 | percpu_ref_exit(&css->refcnt); |
4342 | ||
9d755d33 TH |
4343 | if (css->ss) { |
4344 | /* css free path */ | |
4345 | if (css->parent) | |
4346 | css_put(css->parent); | |
0ae78e0b | 4347 | |
9d755d33 TH |
4348 | css->ss->css_free(css); |
4349 | cgroup_put(cgrp); | |
4350 | } else { | |
4351 | /* cgroup free path */ | |
4352 | atomic_dec(&cgrp->root->nr_cgrps); | |
4353 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4354 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4355 | |
d51f39b0 | 4356 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4357 | /* |
4358 | * We get a ref to the parent, and put the ref when | |
4359 | * this cgroup is being freed, so it's guaranteed | |
4360 | * that the parent won't be destroyed before its | |
4361 | * children. | |
4362 | */ | |
d51f39b0 | 4363 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4364 | kernfs_put(cgrp->kn); |
4365 | kfree(cgrp); | |
4366 | } else { | |
4367 | /* | |
4368 | * This is root cgroup's refcnt reaching zero, | |
4369 | * which indicates that the root should be | |
4370 | * released. | |
4371 | */ | |
4372 | cgroup_destroy_root(cgrp->root); | |
4373 | } | |
4374 | } | |
48ddbe19 TH |
4375 | } |
4376 | ||
0c21ead1 | 4377 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4378 | { |
4379 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4380 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4381 | |
35ef10da | 4382 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4383 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4384 | } |
4385 | ||
25e15d83 | 4386 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4387 | { |
4388 | struct cgroup_subsys_state *css = | |
25e15d83 | 4389 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4390 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4391 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4392 | |
1fed1b2e TH |
4393 | mutex_lock(&cgroup_mutex); |
4394 | ||
de3f0341 | 4395 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4396 | list_del_rcu(&css->sibling); |
4397 | ||
9d755d33 TH |
4398 | if (ss) { |
4399 | /* css release path */ | |
4400 | cgroup_idr_remove(&ss->css_idr, css->id); | |
7d172cc8 TH |
4401 | if (ss->css_released) |
4402 | ss->css_released(css); | |
9d755d33 TH |
4403 | } else { |
4404 | /* cgroup release path */ | |
9d755d33 TH |
4405 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4406 | cgrp->id = -1; | |
a4189487 LZ |
4407 | |
4408 | /* | |
4409 | * There are two control paths which try to determine | |
4410 | * cgroup from dentry without going through kernfs - | |
4411 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4412 | * Those are supported by RCU protecting clearing of | |
4413 | * cgrp->kn->priv backpointer. | |
4414 | */ | |
4415 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
9d755d33 | 4416 | } |
d3daf28d | 4417 | |
1fed1b2e TH |
4418 | mutex_unlock(&cgroup_mutex); |
4419 | ||
0c21ead1 | 4420 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4421 | } |
4422 | ||
d3daf28d TH |
4423 | static void css_release(struct percpu_ref *ref) |
4424 | { | |
4425 | struct cgroup_subsys_state *css = | |
4426 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4427 | ||
25e15d83 TH |
4428 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4429 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4430 | } |
4431 | ||
ddfcadab TH |
4432 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4433 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4434 | { |
0cb51d71 TH |
4435 | lockdep_assert_held(&cgroup_mutex); |
4436 | ||
ddfcadab TH |
4437 | cgroup_get(cgrp); |
4438 | ||
d5c419b6 | 4439 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4440 | css->cgroup = cgrp; |
72c97e54 | 4441 | css->ss = ss; |
d5c419b6 TH |
4442 | INIT_LIST_HEAD(&css->sibling); |
4443 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4444 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4445 | |
d51f39b0 TH |
4446 | if (cgroup_parent(cgrp)) { |
4447 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4448 | css_get(css->parent); |
ddfcadab | 4449 | } |
48ddbe19 | 4450 | |
ca8bdcaf | 4451 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4452 | } |
4453 | ||
2a4ac633 | 4454 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4455 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4456 | { |
623f926b | 4457 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4458 | int ret = 0; |
4459 | ||
a31f2d3f TH |
4460 | lockdep_assert_held(&cgroup_mutex); |
4461 | ||
92fb9748 | 4462 | if (ss->css_online) |
eb95419b | 4463 | ret = ss->css_online(css); |
ae7f164a | 4464 | if (!ret) { |
eb95419b | 4465 | css->flags |= CSS_ONLINE; |
aec25020 | 4466 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4467 | } |
b1929db4 | 4468 | return ret; |
a31f2d3f TH |
4469 | } |
4470 | ||
2a4ac633 | 4471 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4472 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4473 | { |
623f926b | 4474 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4475 | |
4476 | lockdep_assert_held(&cgroup_mutex); | |
4477 | ||
4478 | if (!(css->flags & CSS_ONLINE)) | |
4479 | return; | |
4480 | ||
d7eeac19 | 4481 | if (ss->css_offline) |
eb95419b | 4482 | ss->css_offline(css); |
a31f2d3f | 4483 | |
eb95419b | 4484 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4485 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4486 | |
4487 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4488 | } |
4489 | ||
c81c925a TH |
4490 | /** |
4491 | * create_css - create a cgroup_subsys_state | |
4492 | * @cgrp: the cgroup new css will be associated with | |
4493 | * @ss: the subsys of new css | |
f63070d3 | 4494 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4495 | * |
4496 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4497 | * css is online and installed in @cgrp with all interface files created if |
4498 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4499 | */ |
f63070d3 TH |
4500 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4501 | bool visible) | |
c81c925a | 4502 | { |
d51f39b0 | 4503 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4504 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4505 | struct cgroup_subsys_state *css; |
4506 | int err; | |
4507 | ||
c81c925a TH |
4508 | lockdep_assert_held(&cgroup_mutex); |
4509 | ||
1fed1b2e | 4510 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4511 | if (IS_ERR(css)) |
4512 | return PTR_ERR(css); | |
4513 | ||
ddfcadab | 4514 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4515 | |
2aad2a86 | 4516 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4517 | if (err) |
3eb59ec6 | 4518 | goto err_free_css; |
c81c925a | 4519 | |
15a4c835 TH |
4520 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_NOWAIT); |
4521 | if (err < 0) | |
4522 | goto err_free_percpu_ref; | |
4523 | css->id = err; | |
c81c925a | 4524 | |
f63070d3 TH |
4525 | if (visible) { |
4526 | err = cgroup_populate_dir(cgrp, 1 << ss->id); | |
4527 | if (err) | |
4528 | goto err_free_id; | |
4529 | } | |
15a4c835 TH |
4530 | |
4531 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4532 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4533 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4534 | |
4535 | err = online_css(css); | |
4536 | if (err) | |
1fed1b2e | 4537 | goto err_list_del; |
94419627 | 4538 | |
c81c925a | 4539 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4540 | cgroup_parent(parent)) { |
ed3d261b | 4541 | 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 | 4542 | current->comm, current->pid, ss->name); |
c81c925a | 4543 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4544 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4545 | ss->warned_broken_hierarchy = true; |
4546 | } | |
4547 | ||
4548 | return 0; | |
4549 | ||
1fed1b2e TH |
4550 | err_list_del: |
4551 | list_del_rcu(&css->sibling); | |
32d01dc7 | 4552 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
15a4c835 TH |
4553 | err_free_id: |
4554 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4555 | err_free_percpu_ref: |
9a1049da | 4556 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4557 | err_free_css: |
a2bed820 | 4558 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4559 | return err; |
4560 | } | |
4561 | ||
b3bfd983 TH |
4562 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4563 | umode_t mode) | |
ddbcc7e8 | 4564 | { |
a9746d8d TH |
4565 | struct cgroup *parent, *cgrp; |
4566 | struct cgroup_root *root; | |
ddbcc7e8 | 4567 | struct cgroup_subsys *ss; |
2bd59d48 | 4568 | struct kernfs_node *kn; |
a14c6874 | 4569 | struct cftype *base_files; |
b3bfd983 | 4570 | int ssid, ret; |
ddbcc7e8 | 4571 | |
71b1fb5c AC |
4572 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4573 | */ | |
4574 | if (strchr(name, '\n')) | |
4575 | return -EINVAL; | |
4576 | ||
a9746d8d TH |
4577 | parent = cgroup_kn_lock_live(parent_kn); |
4578 | if (!parent) | |
4579 | return -ENODEV; | |
4580 | root = parent->root; | |
ddbcc7e8 | 4581 | |
0a950f65 | 4582 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc | 4583 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
ba0f4d76 TH |
4584 | if (!cgrp) { |
4585 | ret = -ENOMEM; | |
4586 | goto out_unlock; | |
0ab02ca8 LZ |
4587 | } |
4588 | ||
2aad2a86 | 4589 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4590 | if (ret) |
4591 | goto out_free_cgrp; | |
4592 | ||
0ab02ca8 LZ |
4593 | /* |
4594 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4595 | * a half-baked cgroup. | |
4596 | */ | |
6fa4918d | 4597 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_NOWAIT); |
0ab02ca8 | 4598 | if (cgrp->id < 0) { |
ba0f4d76 | 4599 | ret = -ENOMEM; |
9d755d33 | 4600 | goto out_cancel_ref; |
976c06bc TH |
4601 | } |
4602 | ||
cc31edce | 4603 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4604 | |
9d800df1 | 4605 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4606 | cgrp->root = root; |
ddbcc7e8 | 4607 | |
b6abdb0e LZ |
4608 | if (notify_on_release(parent)) |
4609 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4610 | ||
2260e7fc TH |
4611 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4612 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4613 | |
2bd59d48 | 4614 | /* create the directory */ |
e61734c5 | 4615 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4616 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4617 | ret = PTR_ERR(kn); |
4618 | goto out_free_id; | |
2bd59d48 TH |
4619 | } |
4620 | cgrp->kn = kn; | |
ddbcc7e8 | 4621 | |
4e139afc | 4622 | /* |
6f30558f TH |
4623 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4624 | * that @cgrp->kn is always accessible. | |
4e139afc | 4625 | */ |
6f30558f | 4626 | kernfs_get(kn); |
ddbcc7e8 | 4627 | |
0cb51d71 | 4628 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4629 | |
4e139afc | 4630 | /* allocation complete, commit to creation */ |
d5c419b6 | 4631 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4632 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4633 | cgroup_get(parent); |
415cf07a | 4634 | |
0d80255e TH |
4635 | /* |
4636 | * @cgrp is now fully operational. If something fails after this | |
4637 | * point, it'll be released via the normal destruction path. | |
4638 | */ | |
6fa4918d | 4639 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4640 | |
ba0f4d76 TH |
4641 | ret = cgroup_kn_set_ugid(kn); |
4642 | if (ret) | |
4643 | goto out_destroy; | |
49957f8e | 4644 | |
a14c6874 TH |
4645 | if (cgroup_on_dfl(cgrp)) |
4646 | base_files = cgroup_dfl_base_files; | |
4647 | else | |
4648 | base_files = cgroup_legacy_base_files; | |
4649 | ||
4650 | ret = cgroup_addrm_files(cgrp, base_files, true); | |
ba0f4d76 TH |
4651 | if (ret) |
4652 | goto out_destroy; | |
628f7cd4 | 4653 | |
9d403e99 | 4654 | /* let's create and online css's */ |
b85d2040 | 4655 | for_each_subsys(ss, ssid) { |
f392e51c | 4656 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
4657 | ret = create_css(cgrp, ss, |
4658 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
4659 | if (ret) |
4660 | goto out_destroy; | |
b85d2040 | 4661 | } |
a8638030 | 4662 | } |
ddbcc7e8 | 4663 | |
bd53d617 TH |
4664 | /* |
4665 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4666 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4667 | */ |
667c2491 TH |
4668 | if (!cgroup_on_dfl(cgrp)) { |
4669 | cgrp->subtree_control = parent->subtree_control; | |
4670 | cgroup_refresh_child_subsys_mask(cgrp); | |
4671 | } | |
2bd59d48 | 4672 | |
2bd59d48 | 4673 | kernfs_activate(kn); |
ddbcc7e8 | 4674 | |
ba0f4d76 TH |
4675 | ret = 0; |
4676 | goto out_unlock; | |
ddbcc7e8 | 4677 | |
ba0f4d76 | 4678 | out_free_id: |
6fa4918d | 4679 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 4680 | out_cancel_ref: |
9a1049da | 4681 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 4682 | out_free_cgrp: |
bd89aabc | 4683 | kfree(cgrp); |
ba0f4d76 | 4684 | out_unlock: |
a9746d8d | 4685 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4686 | return ret; |
4b8b47eb | 4687 | |
ba0f4d76 | 4688 | out_destroy: |
4b8b47eb | 4689 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4690 | goto out_unlock; |
ddbcc7e8 PM |
4691 | } |
4692 | ||
223dbc38 TH |
4693 | /* |
4694 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4695 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4696 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4697 | */ |
4698 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4699 | { |
223dbc38 TH |
4700 | struct cgroup_subsys_state *css = |
4701 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4702 | |
f20104de | 4703 | mutex_lock(&cgroup_mutex); |
09a503ea | 4704 | offline_css(css); |
f20104de | 4705 | mutex_unlock(&cgroup_mutex); |
09a503ea | 4706 | |
09a503ea | 4707 | css_put(css); |
d3daf28d TH |
4708 | } |
4709 | ||
223dbc38 TH |
4710 | /* css kill confirmation processing requires process context, bounce */ |
4711 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4712 | { |
4713 | struct cgroup_subsys_state *css = | |
4714 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4715 | ||
223dbc38 | 4716 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4717 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4718 | } |
4719 | ||
f392e51c TH |
4720 | /** |
4721 | * kill_css - destroy a css | |
4722 | * @css: css to destroy | |
4723 | * | |
4724 | * This function initiates destruction of @css by removing cgroup interface | |
4725 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4726 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4727 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4728 | */ |
4729 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4730 | { |
01f6474c | 4731 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4732 | |
2bd59d48 TH |
4733 | /* |
4734 | * This must happen before css is disassociated with its cgroup. | |
4735 | * See seq_css() for details. | |
4736 | */ | |
aec25020 | 4737 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4738 | |
edae0c33 TH |
4739 | /* |
4740 | * Killing would put the base ref, but we need to keep it alive | |
4741 | * until after ->css_offline(). | |
4742 | */ | |
4743 | css_get(css); | |
4744 | ||
4745 | /* | |
4746 | * cgroup core guarantees that, by the time ->css_offline() is | |
4747 | * invoked, no new css reference will be given out via | |
ec903c0c | 4748 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4749 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4750 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4751 | * | |
4752 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4753 | * css is confirmed to be seen as killed on all CPUs. | |
4754 | */ | |
4755 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4756 | } |
4757 | ||
4758 | /** | |
4759 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4760 | * @cgrp: cgroup to be destroyed | |
4761 | * | |
4762 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4763 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4764 | * guarantee that css_tryget_online() won't succeed by the time |
4765 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4766 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4767 | * |
4768 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4769 | * userland visible parts and start killing the percpu refcnts of | |
4770 | * css's. Set up so that the next stage will be kicked off once all | |
4771 | * the percpu refcnts are confirmed to be killed. | |
4772 | * | |
4773 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4774 | * rest of destruction. Once all cgroup references are gone, the | |
4775 | * cgroup is RCU-freed. | |
4776 | * | |
4777 | * This function implements s1. After this step, @cgrp is gone as far as | |
4778 | * the userland is concerned and a new cgroup with the same name may be | |
4779 | * created. As cgroup doesn't care about the names internally, this | |
4780 | * doesn't cause any problem. | |
4781 | */ | |
42809dd4 TH |
4782 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4783 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4784 | { |
2bd59d48 | 4785 | struct cgroup_subsys_state *css; |
ddd69148 | 4786 | bool empty; |
1c6727af | 4787 | int ssid; |
ddbcc7e8 | 4788 | |
42809dd4 TH |
4789 | lockdep_assert_held(&cgroup_mutex); |
4790 | ||
ddd69148 | 4791 | /* |
96d365e0 | 4792 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4793 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4794 | */ |
96d365e0 | 4795 | down_read(&css_set_rwsem); |
bb78a92f | 4796 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4797 | up_read(&css_set_rwsem); |
ddd69148 | 4798 | if (!empty) |
ddbcc7e8 | 4799 | return -EBUSY; |
a043e3b2 | 4800 | |
bb78a92f | 4801 | /* |
d5c419b6 TH |
4802 | * Make sure there's no live children. We can't test emptiness of |
4803 | * ->self.children as dead children linger on it while being | |
4804 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 4805 | */ |
f3d46500 | 4806 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
4807 | return -EBUSY; |
4808 | ||
455050d2 TH |
4809 | /* |
4810 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 4811 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 4812 | */ |
184faf32 | 4813 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 4814 | |
249f3468 | 4815 | /* initiate massacre of all css's */ |
1c6727af TH |
4816 | for_each_css(css, ssid, cgrp) |
4817 | kill_css(css); | |
455050d2 | 4818 | |
455050d2 | 4819 | /* |
01f6474c TH |
4820 | * Remove @cgrp directory along with the base files. @cgrp has an |
4821 | * extra ref on its kn. | |
f20104de | 4822 | */ |
01f6474c | 4823 | kernfs_remove(cgrp->kn); |
f20104de | 4824 | |
d51f39b0 | 4825 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 4826 | |
249f3468 | 4827 | /* put the base reference */ |
9d755d33 | 4828 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 4829 | |
ea15f8cc TH |
4830 | return 0; |
4831 | }; | |
4832 | ||
2bd59d48 | 4833 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4834 | { |
a9746d8d | 4835 | struct cgroup *cgrp; |
2bd59d48 | 4836 | int ret = 0; |
42809dd4 | 4837 | |
a9746d8d TH |
4838 | cgrp = cgroup_kn_lock_live(kn); |
4839 | if (!cgrp) | |
4840 | return 0; | |
42809dd4 | 4841 | |
a9746d8d | 4842 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 4843 | |
a9746d8d | 4844 | cgroup_kn_unlock(kn); |
42809dd4 | 4845 | return ret; |
8e3f6541 TH |
4846 | } |
4847 | ||
2bd59d48 TH |
4848 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4849 | .remount_fs = cgroup_remount, | |
4850 | .show_options = cgroup_show_options, | |
4851 | .mkdir = cgroup_mkdir, | |
4852 | .rmdir = cgroup_rmdir, | |
4853 | .rename = cgroup_rename, | |
4854 | }; | |
4855 | ||
15a4c835 | 4856 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4857 | { |
ddbcc7e8 | 4858 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4859 | |
4860 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4861 | |
648bb56d TH |
4862 | mutex_lock(&cgroup_mutex); |
4863 | ||
15a4c835 | 4864 | idr_init(&ss->css_idr); |
0adb0704 | 4865 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4866 | |
3dd06ffa TH |
4867 | /* Create the root cgroup state for this subsystem */ |
4868 | ss->root = &cgrp_dfl_root; | |
4869 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4870 | /* We don't handle early failures gracefully */ |
4871 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4872 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
4873 | |
4874 | /* | |
4875 | * Root csses are never destroyed and we can't initialize | |
4876 | * percpu_ref during early init. Disable refcnting. | |
4877 | */ | |
4878 | css->flags |= CSS_NO_REF; | |
4879 | ||
15a4c835 | 4880 | if (early) { |
9395a450 | 4881 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
4882 | css->id = 1; |
4883 | } else { | |
4884 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4885 | BUG_ON(css->id < 0); | |
4886 | } | |
ddbcc7e8 | 4887 | |
e8d55fde | 4888 | /* Update the init_css_set to contain a subsys |
817929ec | 4889 | * pointer to this state - since the subsystem is |
e8d55fde | 4890 | * newly registered, all tasks and hence the |
3dd06ffa | 4891 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4892 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4893 | |
4894 | need_forkexit_callback |= ss->fork || ss->exit; | |
4895 | ||
e8d55fde LZ |
4896 | /* At system boot, before all subsystems have been |
4897 | * registered, no tasks have been forked, so we don't | |
4898 | * need to invoke fork callbacks here. */ | |
4899 | BUG_ON(!list_empty(&init_task.tasks)); | |
4900 | ||
ae7f164a | 4901 | BUG_ON(online_css(css)); |
a8638030 | 4902 | |
cf5d5941 BB |
4903 | mutex_unlock(&cgroup_mutex); |
4904 | } | |
cf5d5941 | 4905 | |
ddbcc7e8 | 4906 | /** |
a043e3b2 LZ |
4907 | * cgroup_init_early - cgroup initialization at system boot |
4908 | * | |
4909 | * Initialize cgroups at system boot, and initialize any | |
4910 | * subsystems that request early init. | |
ddbcc7e8 PM |
4911 | */ |
4912 | int __init cgroup_init_early(void) | |
4913 | { | |
7b9a6ba5 | 4914 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 4915 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4916 | int i; |
30159ec7 | 4917 | |
3dd06ffa | 4918 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
4919 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
4920 | ||
a4ea1cc9 | 4921 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4922 | |
3ed80a62 | 4923 | for_each_subsys(ss, i) { |
aec25020 | 4924 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4925 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4926 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4927 | ss->id, ss->name); |
073219e9 TH |
4928 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4929 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4930 | ||
aec25020 | 4931 | ss->id = i; |
073219e9 | 4932 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4933 | |
4934 | if (ss->early_init) | |
15a4c835 | 4935 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
4936 | } |
4937 | return 0; | |
4938 | } | |
4939 | ||
4940 | /** | |
a043e3b2 LZ |
4941 | * cgroup_init - cgroup initialization |
4942 | * | |
4943 | * Register cgroup filesystem and /proc file, and initialize | |
4944 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4945 | */ |
4946 | int __init cgroup_init(void) | |
4947 | { | |
30159ec7 | 4948 | struct cgroup_subsys *ss; |
0ac801fe | 4949 | unsigned long key; |
172a2c06 | 4950 | int ssid, err; |
ddbcc7e8 | 4951 | |
a14c6874 TH |
4952 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
4953 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 4954 | |
54e7b4eb | 4955 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4956 | |
82fe9b0d TH |
4957 | /* Add init_css_set to the hash table */ |
4958 | key = css_set_hash(init_css_set.subsys); | |
4959 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4960 | ||
3dd06ffa | 4961 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4962 | |
54e7b4eb TH |
4963 | mutex_unlock(&cgroup_mutex); |
4964 | ||
172a2c06 | 4965 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
4966 | if (ss->early_init) { |
4967 | struct cgroup_subsys_state *css = | |
4968 | init_css_set.subsys[ss->id]; | |
4969 | ||
4970 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
4971 | GFP_KERNEL); | |
4972 | BUG_ON(css->id < 0); | |
4973 | } else { | |
4974 | cgroup_init_subsys(ss, false); | |
4975 | } | |
172a2c06 | 4976 | |
2d8f243a TH |
4977 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4978 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
4979 | |
4980 | /* | |
c731ae1d LZ |
4981 | * Setting dfl_root subsys_mask needs to consider the |
4982 | * disabled flag and cftype registration needs kmalloc, | |
4983 | * both of which aren't available during early_init. | |
172a2c06 | 4984 | */ |
a8ddc821 TH |
4985 | if (ss->disabled) |
4986 | continue; | |
4987 | ||
4988 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
4989 | ||
4990 | if (cgroup_legacy_files_on_dfl && !ss->dfl_cftypes) | |
4991 | ss->dfl_cftypes = ss->legacy_cftypes; | |
4992 | ||
5de4fa13 TH |
4993 | if (!ss->dfl_cftypes) |
4994 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
4995 | ||
a8ddc821 TH |
4996 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
4997 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
4998 | } else { | |
4999 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5000 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5001 | } |
676db4af GKH |
5002 | } |
5003 | ||
676db4af | 5004 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
5005 | if (!cgroup_kobj) |
5006 | return -ENOMEM; | |
676db4af | 5007 | |
ddbcc7e8 | 5008 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
5009 | if (err < 0) { |
5010 | kobject_put(cgroup_kobj); | |
2bd59d48 | 5011 | return err; |
676db4af | 5012 | } |
ddbcc7e8 | 5013 | |
46ae220b | 5014 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 5015 | return 0; |
ddbcc7e8 | 5016 | } |
b4f48b63 | 5017 | |
e5fca243 TH |
5018 | static int __init cgroup_wq_init(void) |
5019 | { | |
5020 | /* | |
5021 | * There isn't much point in executing destruction path in | |
5022 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5023 | * Use 1 for @max_active. |
e5fca243 TH |
5024 | * |
5025 | * We would prefer to do this in cgroup_init() above, but that | |
5026 | * is called before init_workqueues(): so leave this until after. | |
5027 | */ | |
1a11533f | 5028 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5029 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5030 | |
5031 | /* | |
5032 | * Used to destroy pidlists and separate to serve as flush domain. | |
5033 | * Cap @max_active to 1 too. | |
5034 | */ | |
5035 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5036 | 0, 1); | |
5037 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5038 | ||
e5fca243 TH |
5039 | return 0; |
5040 | } | |
5041 | core_initcall(cgroup_wq_init); | |
5042 | ||
a424316c PM |
5043 | /* |
5044 | * proc_cgroup_show() | |
5045 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5046 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5047 | */ |
006f4ac4 ZL |
5048 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5049 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5050 | { |
e61734c5 | 5051 | char *buf, *path; |
a424316c | 5052 | int retval; |
3dd06ffa | 5053 | struct cgroup_root *root; |
a424316c PM |
5054 | |
5055 | retval = -ENOMEM; | |
e61734c5 | 5056 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5057 | if (!buf) |
5058 | goto out; | |
5059 | ||
a424316c | 5060 | mutex_lock(&cgroup_mutex); |
96d365e0 | 5061 | down_read(&css_set_rwsem); |
a424316c | 5062 | |
985ed670 | 5063 | for_each_root(root) { |
a424316c | 5064 | struct cgroup_subsys *ss; |
bd89aabc | 5065 | struct cgroup *cgrp; |
b85d2040 | 5066 | int ssid, count = 0; |
a424316c | 5067 | |
a2dd4247 | 5068 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
5069 | continue; |
5070 | ||
2c6ab6d2 | 5071 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 5072 | for_each_subsys(ss, ssid) |
f392e51c | 5073 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 5074 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
5075 | if (strlen(root->name)) |
5076 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5077 | root->name); | |
a424316c | 5078 | seq_putc(m, ':'); |
7717f7ba | 5079 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
5080 | path = cgroup_path(cgrp, buf, PATH_MAX); |
5081 | if (!path) { | |
5082 | retval = -ENAMETOOLONG; | |
a424316c | 5083 | goto out_unlock; |
e61734c5 TH |
5084 | } |
5085 | seq_puts(m, path); | |
a424316c PM |
5086 | seq_putc(m, '\n'); |
5087 | } | |
5088 | ||
006f4ac4 | 5089 | retval = 0; |
a424316c | 5090 | out_unlock: |
96d365e0 | 5091 | up_read(&css_set_rwsem); |
a424316c | 5092 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5093 | kfree(buf); |
5094 | out: | |
5095 | return retval; | |
5096 | } | |
5097 | ||
a424316c PM |
5098 | /* Display information about each subsystem and each hierarchy */ |
5099 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5100 | { | |
30159ec7 | 5101 | struct cgroup_subsys *ss; |
a424316c | 5102 | int i; |
a424316c | 5103 | |
8bab8dde | 5104 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5105 | /* |
5106 | * ideally we don't want subsystems moving around while we do this. | |
5107 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5108 | * subsys/hierarchy state. | |
5109 | */ | |
a424316c | 5110 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5111 | |
5112 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
5113 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
5114 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 5115 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 5116 | |
a424316c PM |
5117 | mutex_unlock(&cgroup_mutex); |
5118 | return 0; | |
5119 | } | |
5120 | ||
5121 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5122 | { | |
9dce07f1 | 5123 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5124 | } |
5125 | ||
828c0950 | 5126 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5127 | .open = cgroupstats_open, |
5128 | .read = seq_read, | |
5129 | .llseek = seq_lseek, | |
5130 | .release = single_release, | |
5131 | }; | |
5132 | ||
b4f48b63 | 5133 | /** |
eaf797ab | 5134 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5135 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5136 | * |
eaf797ab TH |
5137 | * A task is associated with the init_css_set until cgroup_post_fork() |
5138 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5139 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5140 | */ |
5141 | void cgroup_fork(struct task_struct *child) | |
5142 | { | |
eaf797ab | 5143 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5144 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5145 | } |
5146 | ||
817929ec | 5147 | /** |
a043e3b2 LZ |
5148 | * cgroup_post_fork - called on a new task after adding it to the task list |
5149 | * @child: the task in question | |
5150 | * | |
5edee61e TH |
5151 | * Adds the task to the list running through its css_set if necessary and |
5152 | * call the subsystem fork() callbacks. Has to be after the task is | |
5153 | * visible on the task list in case we race with the first call to | |
0942eeee | 5154 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5155 | * list. |
a043e3b2 | 5156 | */ |
817929ec PM |
5157 | void cgroup_post_fork(struct task_struct *child) |
5158 | { | |
30159ec7 | 5159 | struct cgroup_subsys *ss; |
5edee61e TH |
5160 | int i; |
5161 | ||
3ce3230a | 5162 | /* |
251f8c03 | 5163 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5164 | * function sets use_task_css_set_links before grabbing |
5165 | * tasklist_lock and we just went through tasklist_lock to add | |
5166 | * @child, it's guaranteed that either we see the set | |
5167 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5168 | * @child during its iteration. | |
5169 | * | |
5170 | * If we won the race, @child is associated with %current's | |
5171 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
5172 | * association is stable, and, on completion of the parent's | |
5173 | * migration, @child is visible in the source of migration or | |
5174 | * already in the destination cgroup. This guarantee is necessary | |
5175 | * when implementing operations which need to migrate all tasks of | |
5176 | * a cgroup to another. | |
5177 | * | |
251f8c03 | 5178 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5179 | * will remain in init_css_set. This is safe because all tasks are |
5180 | * in the init_css_set before cg_links is enabled and there's no | |
5181 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5182 | */ |
817929ec | 5183 | if (use_task_css_set_links) { |
eaf797ab TH |
5184 | struct css_set *cset; |
5185 | ||
96d365e0 | 5186 | down_write(&css_set_rwsem); |
0e1d768f | 5187 | cset = task_css_set(current); |
eaf797ab TH |
5188 | if (list_empty(&child->cg_list)) { |
5189 | rcu_assign_pointer(child->cgroups, cset); | |
5190 | list_add(&child->cg_list, &cset->tasks); | |
5191 | get_css_set(cset); | |
5192 | } | |
96d365e0 | 5193 | up_write(&css_set_rwsem); |
817929ec | 5194 | } |
5edee61e TH |
5195 | |
5196 | /* | |
5197 | * Call ss->fork(). This must happen after @child is linked on | |
5198 | * css_set; otherwise, @child might change state between ->fork() | |
5199 | * and addition to css_set. | |
5200 | */ | |
5201 | if (need_forkexit_callback) { | |
3ed80a62 | 5202 | for_each_subsys(ss, i) |
5edee61e TH |
5203 | if (ss->fork) |
5204 | ss->fork(child); | |
5edee61e | 5205 | } |
817929ec | 5206 | } |
5edee61e | 5207 | |
b4f48b63 PM |
5208 | /** |
5209 | * cgroup_exit - detach cgroup from exiting task | |
5210 | * @tsk: pointer to task_struct of exiting process | |
5211 | * | |
5212 | * Description: Detach cgroup from @tsk and release it. | |
5213 | * | |
5214 | * Note that cgroups marked notify_on_release force every task in | |
5215 | * them to take the global cgroup_mutex mutex when exiting. | |
5216 | * This could impact scaling on very large systems. Be reluctant to | |
5217 | * use notify_on_release cgroups where very high task exit scaling | |
5218 | * is required on large systems. | |
5219 | * | |
0e1d768f TH |
5220 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5221 | * call cgroup_exit() while the task is still competent to handle | |
5222 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5223 | * each hierarchy for the remainder of its exit. No need to bother with | |
5224 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5225 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5226 | */ |
1ec41830 | 5227 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5228 | { |
30159ec7 | 5229 | struct cgroup_subsys *ss; |
5abb8855 | 5230 | struct css_set *cset; |
eaf797ab | 5231 | bool put_cset = false; |
d41d5a01 | 5232 | int i; |
817929ec PM |
5233 | |
5234 | /* | |
0e1d768f TH |
5235 | * Unlink from @tsk from its css_set. As migration path can't race |
5236 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
5237 | */ |
5238 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 5239 | down_write(&css_set_rwsem); |
0e1d768f | 5240 | list_del_init(&tsk->cg_list); |
96d365e0 | 5241 | up_write(&css_set_rwsem); |
0e1d768f | 5242 | put_cset = true; |
817929ec PM |
5243 | } |
5244 | ||
b4f48b63 | 5245 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
5246 | cset = task_css_set(tsk); |
5247 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 5248 | |
1ec41830 | 5249 | if (need_forkexit_callback) { |
3ed80a62 TH |
5250 | /* see cgroup_post_fork() for details */ |
5251 | for_each_subsys(ss, i) { | |
d41d5a01 | 5252 | if (ss->exit) { |
eb95419b TH |
5253 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
5254 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5255 | |
eb95419b | 5256 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
5257 | } |
5258 | } | |
5259 | } | |
d41d5a01 | 5260 | |
eaf797ab | 5261 | if (put_cset) |
a25eb52e | 5262 | put_css_set(cset); |
b4f48b63 | 5263 | } |
697f4161 | 5264 | |
bd89aabc | 5265 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5266 | { |
a25eb52e | 5267 | if (notify_on_release(cgrp) && !cgroup_has_tasks(cgrp) && |
971ff493 ZL |
5268 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5269 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5270 | } |
5271 | ||
81a6a5cd PM |
5272 | /* |
5273 | * Notify userspace when a cgroup is released, by running the | |
5274 | * configured release agent with the name of the cgroup (path | |
5275 | * relative to the root of cgroup file system) as the argument. | |
5276 | * | |
5277 | * Most likely, this user command will try to rmdir this cgroup. | |
5278 | * | |
5279 | * This races with the possibility that some other task will be | |
5280 | * attached to this cgroup before it is removed, or that some other | |
5281 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5282 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5283 | * unused, and this cgroup will be reprieved from its death sentence, | |
5284 | * to continue to serve a useful existence. Next time it's released, | |
5285 | * we will get notified again, if it still has 'notify_on_release' set. | |
5286 | * | |
5287 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5288 | * means only wait until the task is successfully execve()'d. The | |
5289 | * separate release agent task is forked by call_usermodehelper(), | |
5290 | * then control in this thread returns here, without waiting for the | |
5291 | * release agent task. We don't bother to wait because the caller of | |
5292 | * this routine has no use for the exit status of the release agent | |
5293 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5294 | */ |
81a6a5cd PM |
5295 | static void cgroup_release_agent(struct work_struct *work) |
5296 | { | |
971ff493 ZL |
5297 | struct cgroup *cgrp = |
5298 | container_of(work, struct cgroup, release_agent_work); | |
5299 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5300 | char *argv[3], *envp[3]; | |
5301 | ||
81a6a5cd | 5302 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5303 | |
5304 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5305 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5306 | if (!pathbuf || !agentbuf) | |
5307 | goto out; | |
5308 | ||
5309 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5310 | if (!path) | |
5311 | goto out; | |
5312 | ||
5313 | argv[0] = agentbuf; | |
5314 | argv[1] = path; | |
5315 | argv[2] = NULL; | |
5316 | ||
5317 | /* minimal command environment */ | |
5318 | envp[0] = "HOME=/"; | |
5319 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5320 | envp[2] = NULL; | |
5321 | ||
81a6a5cd | 5322 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5323 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5324 | goto out_free; |
971ff493 | 5325 | out: |
81a6a5cd | 5326 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5327 | out_free: |
971ff493 ZL |
5328 | kfree(agentbuf); |
5329 | kfree(pathbuf); | |
81a6a5cd | 5330 | } |
8bab8dde PM |
5331 | |
5332 | static int __init cgroup_disable(char *str) | |
5333 | { | |
30159ec7 | 5334 | struct cgroup_subsys *ss; |
8bab8dde | 5335 | char *token; |
30159ec7 | 5336 | int i; |
8bab8dde PM |
5337 | |
5338 | while ((token = strsep(&str, ",")) != NULL) { | |
5339 | if (!*token) | |
5340 | continue; | |
be45c900 | 5341 | |
3ed80a62 | 5342 | for_each_subsys(ss, i) { |
8bab8dde PM |
5343 | if (!strcmp(token, ss->name)) { |
5344 | ss->disabled = 1; | |
5345 | printk(KERN_INFO "Disabling %s control group" | |
5346 | " subsystem\n", ss->name); | |
5347 | break; | |
5348 | } | |
5349 | } | |
5350 | } | |
5351 | return 1; | |
5352 | } | |
5353 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5354 | |
a8ddc821 TH |
5355 | static int __init cgroup_set_legacy_files_on_dfl(char *str) |
5356 | { | |
5357 | printk("cgroup: using legacy files on the default hierarchy\n"); | |
5358 | cgroup_legacy_files_on_dfl = true; | |
5359 | return 0; | |
5360 | } | |
5361 | __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl); | |
5362 | ||
b77d7b60 | 5363 | /** |
ec903c0c | 5364 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5365 | * @dentry: directory dentry of interest |
5366 | * @ss: subsystem of interest | |
b77d7b60 | 5367 | * |
5a17f543 TH |
5368 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5369 | * to get the corresponding css and return it. If such css doesn't exist | |
5370 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5371 | */ |
ec903c0c TH |
5372 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5373 | struct cgroup_subsys *ss) | |
e5d1367f | 5374 | { |
2bd59d48 TH |
5375 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5376 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5377 | struct cgroup *cgrp; |
e5d1367f | 5378 | |
35cf0836 | 5379 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5380 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5381 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5382 | return ERR_PTR(-EBADF); |
5383 | ||
5a17f543 TH |
5384 | rcu_read_lock(); |
5385 | ||
2bd59d48 TH |
5386 | /* |
5387 | * This path doesn't originate from kernfs and @kn could already | |
5388 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5389 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5390 | */ |
5391 | cgrp = rcu_dereference(kn->priv); | |
5392 | if (cgrp) | |
5393 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5394 | |
ec903c0c | 5395 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5396 | css = ERR_PTR(-ENOENT); |
5397 | ||
5398 | rcu_read_unlock(); | |
5399 | return css; | |
e5d1367f | 5400 | } |
e5d1367f | 5401 | |
1cb650b9 LZ |
5402 | /** |
5403 | * css_from_id - lookup css by id | |
5404 | * @id: the cgroup id | |
5405 | * @ss: cgroup subsys to be looked into | |
5406 | * | |
5407 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5408 | * Should be called under rcu_read_lock(). | |
5409 | */ | |
5410 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5411 | { | |
6fa4918d | 5412 | WARN_ON_ONCE(!rcu_read_lock_held()); |
15a4c835 | 5413 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5414 | } |
5415 | ||
fe693435 | 5416 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5417 | static struct cgroup_subsys_state * |
5418 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5419 | { |
5420 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5421 | ||
5422 | if (!css) | |
5423 | return ERR_PTR(-ENOMEM); | |
5424 | ||
5425 | return css; | |
5426 | } | |
5427 | ||
eb95419b | 5428 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5429 | { |
eb95419b | 5430 | kfree(css); |
fe693435 PM |
5431 | } |
5432 | ||
182446d0 TH |
5433 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5434 | struct cftype *cft) | |
fe693435 | 5435 | { |
182446d0 | 5436 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5437 | } |
5438 | ||
182446d0 TH |
5439 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5440 | struct cftype *cft) | |
fe693435 PM |
5441 | { |
5442 | return (u64)(unsigned long)current->cgroups; | |
5443 | } | |
5444 | ||
182446d0 | 5445 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5446 | struct cftype *cft) |
fe693435 PM |
5447 | { |
5448 | u64 count; | |
5449 | ||
5450 | rcu_read_lock(); | |
a8ad805c | 5451 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5452 | rcu_read_unlock(); |
5453 | return count; | |
5454 | } | |
5455 | ||
2da8ca82 | 5456 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5457 | { |
69d0206c | 5458 | struct cgrp_cset_link *link; |
5abb8855 | 5459 | struct css_set *cset; |
e61734c5 TH |
5460 | char *name_buf; |
5461 | ||
5462 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5463 | if (!name_buf) | |
5464 | return -ENOMEM; | |
7717f7ba | 5465 | |
96d365e0 | 5466 | down_read(&css_set_rwsem); |
7717f7ba | 5467 | rcu_read_lock(); |
5abb8855 | 5468 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5469 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5470 | struct cgroup *c = link->cgrp; |
7717f7ba | 5471 | |
a2dd4247 | 5472 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5473 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5474 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5475 | } |
5476 | rcu_read_unlock(); | |
96d365e0 | 5477 | up_read(&css_set_rwsem); |
e61734c5 | 5478 | kfree(name_buf); |
7717f7ba PM |
5479 | return 0; |
5480 | } | |
5481 | ||
5482 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5483 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5484 | { |
2da8ca82 | 5485 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5486 | struct cgrp_cset_link *link; |
7717f7ba | 5487 | |
96d365e0 | 5488 | down_read(&css_set_rwsem); |
182446d0 | 5489 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5490 | struct css_set *cset = link->cset; |
7717f7ba PM |
5491 | struct task_struct *task; |
5492 | int count = 0; | |
c7561128 | 5493 | |
5abb8855 | 5494 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5495 | |
5abb8855 | 5496 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5497 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5498 | goto overflow; | |
5499 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5500 | } | |
5501 | ||
5502 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5503 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5504 | goto overflow; | |
5505 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5506 | } |
c7561128 TH |
5507 | continue; |
5508 | overflow: | |
5509 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5510 | } |
96d365e0 | 5511 | up_read(&css_set_rwsem); |
7717f7ba PM |
5512 | return 0; |
5513 | } | |
5514 | ||
182446d0 | 5515 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5516 | { |
a25eb52e ZL |
5517 | return (!cgroup_has_tasks(css->cgroup) && |
5518 | !css_has_online_children(&css->cgroup->self)); | |
fe693435 PM |
5519 | } |
5520 | ||
5521 | static struct cftype debug_files[] = { | |
fe693435 PM |
5522 | { |
5523 | .name = "taskcount", | |
5524 | .read_u64 = debug_taskcount_read, | |
5525 | }, | |
5526 | ||
5527 | { | |
5528 | .name = "current_css_set", | |
5529 | .read_u64 = current_css_set_read, | |
5530 | }, | |
5531 | ||
5532 | { | |
5533 | .name = "current_css_set_refcount", | |
5534 | .read_u64 = current_css_set_refcount_read, | |
5535 | }, | |
5536 | ||
7717f7ba PM |
5537 | { |
5538 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5539 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5540 | }, |
5541 | ||
5542 | { | |
5543 | .name = "cgroup_css_links", | |
2da8ca82 | 5544 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5545 | }, |
5546 | ||
fe693435 PM |
5547 | { |
5548 | .name = "releasable", | |
5549 | .read_u64 = releasable_read, | |
5550 | }, | |
fe693435 | 5551 | |
4baf6e33 TH |
5552 | { } /* terminate */ |
5553 | }; | |
fe693435 | 5554 | |
073219e9 | 5555 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5556 | .css_alloc = debug_css_alloc, |
5557 | .css_free = debug_css_free, | |
5577964e | 5558 | .legacy_cftypes = debug_files, |
fe693435 PM |
5559 | }; |
5560 | #endif /* CONFIG_CGROUP_DEBUG */ |