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 | |
59f5296b | 188 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 189 | static int rebind_subsystems(struct cgroup_root *dst_root, |
69dfa00c | 190 | unsigned int ss_mask); |
42809dd4 | 191 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
192 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
193 | bool visible); | |
9d755d33 | 194 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 195 | static void kill_css(struct cgroup_subsys_state *css); |
2bb566cb TH |
196 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
197 | bool is_add); | |
b1a21367 | 198 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 199 | |
6fa4918d TH |
200 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
201 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
202 | gfp_t gfp_mask) | |
203 | { | |
204 | int ret; | |
205 | ||
206 | idr_preload(gfp_mask); | |
54504e97 | 207 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 208 | ret = idr_alloc(idr, ptr, start, end, gfp_mask); |
54504e97 | 209 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
210 | idr_preload_end(); |
211 | return ret; | |
212 | } | |
213 | ||
214 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
215 | { | |
216 | void *ret; | |
217 | ||
54504e97 | 218 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 219 | ret = idr_replace(idr, ptr, id); |
54504e97 | 220 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
221 | return ret; |
222 | } | |
223 | ||
224 | static void cgroup_idr_remove(struct idr *idr, int id) | |
225 | { | |
54504e97 | 226 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 227 | idr_remove(idr, id); |
54504e97 | 228 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
229 | } |
230 | ||
d51f39b0 TH |
231 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
232 | { | |
233 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
234 | ||
235 | if (parent_css) | |
236 | return container_of(parent_css, struct cgroup, self); | |
237 | return NULL; | |
238 | } | |
239 | ||
95109b62 TH |
240 | /** |
241 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
242 | * @cgrp: the cgroup of interest | |
9d800df1 | 243 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 244 | * |
ca8bdcaf TH |
245 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
246 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
247 | * the caller is responsible for pinning the returned css if it wants to | |
248 | * keep accessing it outside the said locks. This function may return | |
249 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
250 | */ |
251 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 252 | struct cgroup_subsys *ss) |
95109b62 | 253 | { |
ca8bdcaf | 254 | if (ss) |
aec25020 | 255 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 256 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 257 | else |
9d800df1 | 258 | return &cgrp->self; |
95109b62 | 259 | } |
42809dd4 | 260 | |
aec3dfcb TH |
261 | /** |
262 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
263 | * @cgrp: the cgroup of interest | |
9d800df1 | 264 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb TH |
265 | * |
266 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
267 | * as the matching css of the nearest ancestor including self which has @ss | |
268 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
269 | * function is guaranteed to return non-NULL css. | |
270 | */ | |
271 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
272 | struct cgroup_subsys *ss) | |
273 | { | |
274 | lockdep_assert_held(&cgroup_mutex); | |
275 | ||
276 | if (!ss) | |
9d800df1 | 277 | return &cgrp->self; |
aec3dfcb TH |
278 | |
279 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
280 | return NULL; | |
281 | ||
d51f39b0 TH |
282 | while (cgroup_parent(cgrp) && |
283 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
284 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
285 | |
286 | return cgroup_css(cgrp, ss); | |
95109b62 | 287 | } |
42809dd4 | 288 | |
ddbcc7e8 | 289 | /* convenient tests for these bits */ |
54766d4a | 290 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 291 | { |
184faf32 | 292 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
293 | } |
294 | ||
b4168640 | 295 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 296 | { |
2bd59d48 | 297 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 298 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
299 | |
300 | /* | |
301 | * This is open and unprotected implementation of cgroup_css(). | |
302 | * seq_css() is only called from a kernfs file operation which has | |
303 | * an active reference on the file. Because all the subsystem | |
304 | * files are drained before a css is disassociated with a cgroup, | |
305 | * the matching css from the cgroup's subsys table is guaranteed to | |
306 | * be and stay valid until the enclosing operation is complete. | |
307 | */ | |
308 | if (cft->ss) | |
309 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
310 | else | |
9d800df1 | 311 | return &cgrp->self; |
59f5296b | 312 | } |
b4168640 | 313 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 314 | |
78574cf9 LZ |
315 | /** |
316 | * cgroup_is_descendant - test ancestry | |
317 | * @cgrp: the cgroup to be tested | |
318 | * @ancestor: possible ancestor of @cgrp | |
319 | * | |
320 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
321 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
322 | * and @ancestor are accessible. | |
323 | */ | |
324 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
325 | { | |
326 | while (cgrp) { | |
327 | if (cgrp == ancestor) | |
328 | return true; | |
d51f39b0 | 329 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
330 | } |
331 | return false; | |
332 | } | |
ddbcc7e8 | 333 | |
e9685a03 | 334 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
335 | { |
336 | const int bits = | |
bd89aabc PM |
337 | (1 << CGRP_RELEASABLE) | |
338 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
339 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
340 | } |
341 | ||
e9685a03 | 342 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 343 | { |
bd89aabc | 344 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
345 | } |
346 | ||
1c6727af TH |
347 | /** |
348 | * for_each_css - iterate all css's of a cgroup | |
349 | * @css: the iteration cursor | |
350 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
351 | * @cgrp: the target cgroup to iterate css's of | |
352 | * | |
aec3dfcb | 353 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
354 | */ |
355 | #define for_each_css(css, ssid, cgrp) \ | |
356 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
357 | if (!((css) = rcu_dereference_check( \ | |
358 | (cgrp)->subsys[(ssid)], \ | |
359 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
360 | else | |
361 | ||
aec3dfcb TH |
362 | /** |
363 | * for_each_e_css - iterate all effective css's of a cgroup | |
364 | * @css: the iteration cursor | |
365 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
366 | * @cgrp: the target cgroup to iterate css's of | |
367 | * | |
368 | * Should be called under cgroup_[tree_]mutex. | |
369 | */ | |
370 | #define for_each_e_css(css, ssid, cgrp) \ | |
371 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
372 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
373 | ; \ | |
374 | else | |
375 | ||
30159ec7 | 376 | /** |
3ed80a62 | 377 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 378 | * @ss: the iteration cursor |
780cd8b3 | 379 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 380 | */ |
780cd8b3 | 381 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
382 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
383 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 384 | |
985ed670 TH |
385 | /* iterate across the hierarchies */ |
386 | #define for_each_root(root) \ | |
5549c497 | 387 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 388 | |
f8f22e53 TH |
389 | /* iterate over child cgrps, lock should be held throughout iteration */ |
390 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 391 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 392 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
393 | cgroup_is_dead(child); })) \ |
394 | ; \ | |
395 | else | |
7ae1bad9 | 396 | |
81a6a5cd PM |
397 | /* the list of cgroups eligible for automatic release. Protected by |
398 | * release_list_lock */ | |
399 | static LIST_HEAD(release_list); | |
cdcc136f | 400 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
401 | static void cgroup_release_agent(struct work_struct *work); |
402 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 403 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 404 | |
69d0206c TH |
405 | /* |
406 | * A cgroup can be associated with multiple css_sets as different tasks may | |
407 | * belong to different cgroups on different hierarchies. In the other | |
408 | * direction, a css_set is naturally associated with multiple cgroups. | |
409 | * This M:N relationship is represented by the following link structure | |
410 | * which exists for each association and allows traversing the associations | |
411 | * from both sides. | |
412 | */ | |
413 | struct cgrp_cset_link { | |
414 | /* the cgroup and css_set this link associates */ | |
415 | struct cgroup *cgrp; | |
416 | struct css_set *cset; | |
417 | ||
418 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
419 | struct list_head cset_link; | |
420 | ||
421 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
422 | struct list_head cgrp_link; | |
817929ec PM |
423 | }; |
424 | ||
172a2c06 TH |
425 | /* |
426 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
427 | * hierarchies being mounted. It contains a pointer to the root state |
428 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
429 | * reference-counted, to improve performance when child cgroups | |
430 | * haven't been created. | |
431 | */ | |
5024ae29 | 432 | struct css_set init_css_set = { |
172a2c06 TH |
433 | .refcount = ATOMIC_INIT(1), |
434 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
435 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
436 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
437 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
438 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
439 | }; | |
817929ec | 440 | |
172a2c06 | 441 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 442 | |
842b597e TH |
443 | /** |
444 | * cgroup_update_populated - updated populated count of a cgroup | |
445 | * @cgrp: the target cgroup | |
446 | * @populated: inc or dec populated count | |
447 | * | |
448 | * @cgrp is either getting the first task (css_set) or losing the last. | |
449 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
450 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
451 | * cgroup and all its descendants are empty. | |
452 | * | |
453 | * @cgrp's interface file "cgroup.populated" is zero if | |
454 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
455 | * changes from or to zero, userland is notified that the content of the | |
456 | * interface file has changed. This can be used to detect when @cgrp and | |
457 | * its descendants become populated or empty. | |
458 | */ | |
459 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
460 | { | |
461 | lockdep_assert_held(&css_set_rwsem); | |
462 | ||
463 | do { | |
464 | bool trigger; | |
465 | ||
466 | if (populated) | |
467 | trigger = !cgrp->populated_cnt++; | |
468 | else | |
469 | trigger = !--cgrp->populated_cnt; | |
470 | ||
471 | if (!trigger) | |
472 | break; | |
473 | ||
474 | if (cgrp->populated_kn) | |
475 | kernfs_notify(cgrp->populated_kn); | |
d51f39b0 | 476 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
477 | } while (cgrp); |
478 | } | |
479 | ||
7717f7ba PM |
480 | /* |
481 | * hash table for cgroup groups. This improves the performance to find | |
482 | * an existing css_set. This hash doesn't (currently) take into | |
483 | * account cgroups in empty hierarchies. | |
484 | */ | |
472b1053 | 485 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 486 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 487 | |
0ac801fe | 488 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 489 | { |
0ac801fe | 490 | unsigned long key = 0UL; |
30159ec7 TH |
491 | struct cgroup_subsys *ss; |
492 | int i; | |
472b1053 | 493 | |
30159ec7 | 494 | for_each_subsys(ss, i) |
0ac801fe LZ |
495 | key += (unsigned long)css[i]; |
496 | key = (key >> 16) ^ key; | |
472b1053 | 497 | |
0ac801fe | 498 | return key; |
472b1053 LZ |
499 | } |
500 | ||
89c5509b | 501 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 502 | { |
69d0206c | 503 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
504 | struct cgroup_subsys *ss; |
505 | int ssid; | |
5abb8855 | 506 | |
89c5509b TH |
507 | lockdep_assert_held(&css_set_rwsem); |
508 | ||
509 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 510 | return; |
81a6a5cd | 511 | |
2c6ab6d2 | 512 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
513 | for_each_subsys(ss, ssid) |
514 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 515 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
516 | css_set_count--; |
517 | ||
69d0206c | 518 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 519 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 520 | |
69d0206c TH |
521 | list_del(&link->cset_link); |
522 | list_del(&link->cgrp_link); | |
71b5707e | 523 | |
96d365e0 | 524 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
525 | if (list_empty(&cgrp->cset_links)) { |
526 | cgroup_update_populated(cgrp, false); | |
527 | if (notify_on_release(cgrp)) { | |
528 | if (taskexit) | |
529 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
530 | check_for_release(cgrp); | |
531 | } | |
81a6a5cd | 532 | } |
2c6ab6d2 PM |
533 | |
534 | kfree(link); | |
81a6a5cd | 535 | } |
2c6ab6d2 | 536 | |
5abb8855 | 537 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
538 | } |
539 | ||
89c5509b TH |
540 | static void put_css_set(struct css_set *cset, bool taskexit) |
541 | { | |
542 | /* | |
543 | * Ensure that the refcount doesn't hit zero while any readers | |
544 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
545 | * rwlock | |
546 | */ | |
547 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
548 | return; | |
549 | ||
550 | down_write(&css_set_rwsem); | |
551 | put_css_set_locked(cset, taskexit); | |
552 | up_write(&css_set_rwsem); | |
553 | } | |
554 | ||
817929ec PM |
555 | /* |
556 | * refcounted get/put for css_set objects | |
557 | */ | |
5abb8855 | 558 | static inline void get_css_set(struct css_set *cset) |
817929ec | 559 | { |
5abb8855 | 560 | atomic_inc(&cset->refcount); |
817929ec PM |
561 | } |
562 | ||
b326f9d0 | 563 | /** |
7717f7ba | 564 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
565 | * @cset: candidate css_set being tested |
566 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
567 | * @new_cgrp: cgroup that's being entered by the task |
568 | * @template: desired set of css pointers in css_set (pre-calculated) | |
569 | * | |
6f4b7e63 | 570 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
571 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
572 | */ | |
5abb8855 TH |
573 | static bool compare_css_sets(struct css_set *cset, |
574 | struct css_set *old_cset, | |
7717f7ba PM |
575 | struct cgroup *new_cgrp, |
576 | struct cgroup_subsys_state *template[]) | |
577 | { | |
578 | struct list_head *l1, *l2; | |
579 | ||
aec3dfcb TH |
580 | /* |
581 | * On the default hierarchy, there can be csets which are | |
582 | * associated with the same set of cgroups but different csses. | |
583 | * Let's first ensure that csses match. | |
584 | */ | |
585 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 586 | return false; |
7717f7ba PM |
587 | |
588 | /* | |
589 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
590 | * different cgroups in hierarchies. As different cgroups may |
591 | * share the same effective css, this comparison is always | |
592 | * necessary. | |
7717f7ba | 593 | */ |
69d0206c TH |
594 | l1 = &cset->cgrp_links; |
595 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 596 | while (1) { |
69d0206c | 597 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 598 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
599 | |
600 | l1 = l1->next; | |
601 | l2 = l2->next; | |
602 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
603 | if (l1 == &cset->cgrp_links) { |
604 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
605 | break; |
606 | } else { | |
69d0206c | 607 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
608 | } |
609 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
610 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
611 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
612 | cgrp1 = link1->cgrp; | |
613 | cgrp2 = link2->cgrp; | |
7717f7ba | 614 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 615 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
616 | |
617 | /* | |
618 | * If this hierarchy is the hierarchy of the cgroup | |
619 | * that's changing, then we need to check that this | |
620 | * css_set points to the new cgroup; if it's any other | |
621 | * hierarchy, then this css_set should point to the | |
622 | * same cgroup as the old css_set. | |
623 | */ | |
5abb8855 TH |
624 | if (cgrp1->root == new_cgrp->root) { |
625 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
626 | return false; |
627 | } else { | |
5abb8855 | 628 | if (cgrp1 != cgrp2) |
7717f7ba PM |
629 | return false; |
630 | } | |
631 | } | |
632 | return true; | |
633 | } | |
634 | ||
b326f9d0 TH |
635 | /** |
636 | * find_existing_css_set - init css array and find the matching css_set | |
637 | * @old_cset: the css_set that we're using before the cgroup transition | |
638 | * @cgrp: the cgroup that we're moving into | |
639 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 640 | */ |
5abb8855 TH |
641 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
642 | struct cgroup *cgrp, | |
643 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 644 | { |
3dd06ffa | 645 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 646 | struct cgroup_subsys *ss; |
5abb8855 | 647 | struct css_set *cset; |
0ac801fe | 648 | unsigned long key; |
b326f9d0 | 649 | int i; |
817929ec | 650 | |
aae8aab4 BB |
651 | /* |
652 | * Build the set of subsystem state objects that we want to see in the | |
653 | * new css_set. while subsystems can change globally, the entries here | |
654 | * won't change, so no need for locking. | |
655 | */ | |
30159ec7 | 656 | for_each_subsys(ss, i) { |
f392e51c | 657 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
658 | /* |
659 | * @ss is in this hierarchy, so we want the | |
660 | * effective css from @cgrp. | |
661 | */ | |
662 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 663 | } else { |
aec3dfcb TH |
664 | /* |
665 | * @ss is not in this hierarchy, so we don't want | |
666 | * to change the css. | |
667 | */ | |
5abb8855 | 668 | template[i] = old_cset->subsys[i]; |
817929ec PM |
669 | } |
670 | } | |
671 | ||
0ac801fe | 672 | key = css_set_hash(template); |
5abb8855 TH |
673 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
674 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
675 | continue; |
676 | ||
677 | /* This css_set matches what we need */ | |
5abb8855 | 678 | return cset; |
472b1053 | 679 | } |
817929ec PM |
680 | |
681 | /* No existing cgroup group matched */ | |
682 | return NULL; | |
683 | } | |
684 | ||
69d0206c | 685 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 686 | { |
69d0206c | 687 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 688 | |
69d0206c TH |
689 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
690 | list_del(&link->cset_link); | |
36553434 LZ |
691 | kfree(link); |
692 | } | |
693 | } | |
694 | ||
69d0206c TH |
695 | /** |
696 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
697 | * @count: the number of links to allocate | |
698 | * @tmp_links: list_head the allocated links are put on | |
699 | * | |
700 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
701 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 702 | */ |
69d0206c | 703 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 704 | { |
69d0206c | 705 | struct cgrp_cset_link *link; |
817929ec | 706 | int i; |
69d0206c TH |
707 | |
708 | INIT_LIST_HEAD(tmp_links); | |
709 | ||
817929ec | 710 | for (i = 0; i < count; i++) { |
f4f4be2b | 711 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 712 | if (!link) { |
69d0206c | 713 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
714 | return -ENOMEM; |
715 | } | |
69d0206c | 716 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
717 | } |
718 | return 0; | |
719 | } | |
720 | ||
c12f65d4 LZ |
721 | /** |
722 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 723 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 724 | * @cset: the css_set to be linked |
c12f65d4 LZ |
725 | * @cgrp: the destination cgroup |
726 | */ | |
69d0206c TH |
727 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
728 | struct cgroup *cgrp) | |
c12f65d4 | 729 | { |
69d0206c | 730 | struct cgrp_cset_link *link; |
c12f65d4 | 731 | |
69d0206c | 732 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
733 | |
734 | if (cgroup_on_dfl(cgrp)) | |
735 | cset->dfl_cgrp = cgrp; | |
736 | ||
69d0206c TH |
737 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
738 | link->cset = cset; | |
7717f7ba | 739 | link->cgrp = cgrp; |
842b597e TH |
740 | |
741 | if (list_empty(&cgrp->cset_links)) | |
742 | cgroup_update_populated(cgrp, true); | |
69d0206c | 743 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 744 | |
7717f7ba PM |
745 | /* |
746 | * Always add links to the tail of the list so that the list | |
747 | * is sorted by order of hierarchy creation | |
748 | */ | |
69d0206c | 749 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
750 | } |
751 | ||
b326f9d0 TH |
752 | /** |
753 | * find_css_set - return a new css_set with one cgroup updated | |
754 | * @old_cset: the baseline css_set | |
755 | * @cgrp: the cgroup to be updated | |
756 | * | |
757 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
758 | * substituted into the appropriate hierarchy. | |
817929ec | 759 | */ |
5abb8855 TH |
760 | static struct css_set *find_css_set(struct css_set *old_cset, |
761 | struct cgroup *cgrp) | |
817929ec | 762 | { |
b326f9d0 | 763 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 764 | struct css_set *cset; |
69d0206c TH |
765 | struct list_head tmp_links; |
766 | struct cgrp_cset_link *link; | |
2d8f243a | 767 | struct cgroup_subsys *ss; |
0ac801fe | 768 | unsigned long key; |
2d8f243a | 769 | int ssid; |
472b1053 | 770 | |
b326f9d0 TH |
771 | lockdep_assert_held(&cgroup_mutex); |
772 | ||
817929ec PM |
773 | /* First see if we already have a cgroup group that matches |
774 | * the desired set */ | |
96d365e0 | 775 | down_read(&css_set_rwsem); |
5abb8855 TH |
776 | cset = find_existing_css_set(old_cset, cgrp, template); |
777 | if (cset) | |
778 | get_css_set(cset); | |
96d365e0 | 779 | up_read(&css_set_rwsem); |
817929ec | 780 | |
5abb8855 TH |
781 | if (cset) |
782 | return cset; | |
817929ec | 783 | |
f4f4be2b | 784 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 785 | if (!cset) |
817929ec PM |
786 | return NULL; |
787 | ||
69d0206c | 788 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 789 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 790 | kfree(cset); |
817929ec PM |
791 | return NULL; |
792 | } | |
793 | ||
5abb8855 | 794 | atomic_set(&cset->refcount, 1); |
69d0206c | 795 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 796 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 797 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 798 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 799 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 800 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
801 | |
802 | /* Copy the set of subsystem state objects generated in | |
803 | * find_existing_css_set() */ | |
5abb8855 | 804 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 805 | |
96d365e0 | 806 | down_write(&css_set_rwsem); |
817929ec | 807 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 808 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 809 | struct cgroup *c = link->cgrp; |
69d0206c | 810 | |
7717f7ba PM |
811 | if (c->root == cgrp->root) |
812 | c = cgrp; | |
69d0206c | 813 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 814 | } |
817929ec | 815 | |
69d0206c | 816 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 817 | |
817929ec | 818 | css_set_count++; |
472b1053 | 819 | |
2d8f243a | 820 | /* Add @cset to the hash table */ |
5abb8855 TH |
821 | key = css_set_hash(cset->subsys); |
822 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 823 | |
2d8f243a TH |
824 | for_each_subsys(ss, ssid) |
825 | list_add_tail(&cset->e_cset_node[ssid], | |
826 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
827 | ||
96d365e0 | 828 | up_write(&css_set_rwsem); |
817929ec | 829 | |
5abb8855 | 830 | return cset; |
b4f48b63 PM |
831 | } |
832 | ||
3dd06ffa | 833 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 834 | { |
3dd06ffa | 835 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 836 | |
3dd06ffa | 837 | return root_cgrp->root; |
2bd59d48 TH |
838 | } |
839 | ||
3dd06ffa | 840 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
841 | { |
842 | int id; | |
843 | ||
844 | lockdep_assert_held(&cgroup_mutex); | |
845 | ||
985ed670 | 846 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
847 | if (id < 0) |
848 | return id; | |
849 | ||
850 | root->hierarchy_id = id; | |
851 | return 0; | |
852 | } | |
853 | ||
3dd06ffa | 854 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
855 | { |
856 | lockdep_assert_held(&cgroup_mutex); | |
857 | ||
858 | if (root->hierarchy_id) { | |
859 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
860 | root->hierarchy_id = 0; | |
861 | } | |
862 | } | |
863 | ||
3dd06ffa | 864 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
865 | { |
866 | if (root) { | |
867 | /* hierarhcy ID shoulid already have been released */ | |
868 | WARN_ON_ONCE(root->hierarchy_id); | |
869 | ||
870 | idr_destroy(&root->cgroup_idr); | |
871 | kfree(root); | |
872 | } | |
873 | } | |
874 | ||
3dd06ffa | 875 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 876 | { |
3dd06ffa | 877 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 878 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 879 | |
2bd59d48 | 880 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 881 | |
776f02fa | 882 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 883 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 884 | |
f2e85d57 | 885 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 886 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 887 | |
7717f7ba | 888 | /* |
f2e85d57 TH |
889 | * Release all the links from cset_links to this hierarchy's |
890 | * root cgroup | |
7717f7ba | 891 | */ |
96d365e0 | 892 | down_write(&css_set_rwsem); |
f2e85d57 TH |
893 | |
894 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
895 | list_del(&link->cset_link); | |
896 | list_del(&link->cgrp_link); | |
897 | kfree(link); | |
898 | } | |
96d365e0 | 899 | up_write(&css_set_rwsem); |
f2e85d57 TH |
900 | |
901 | if (!list_empty(&root->root_list)) { | |
902 | list_del(&root->root_list); | |
903 | cgroup_root_count--; | |
904 | } | |
905 | ||
906 | cgroup_exit_root_id(root); | |
907 | ||
908 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 909 | |
2bd59d48 | 910 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
911 | cgroup_free_root(root); |
912 | } | |
913 | ||
ceb6a081 TH |
914 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
915 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 916 | struct cgroup_root *root) |
7717f7ba | 917 | { |
7717f7ba PM |
918 | struct cgroup *res = NULL; |
919 | ||
96d365e0 TH |
920 | lockdep_assert_held(&cgroup_mutex); |
921 | lockdep_assert_held(&css_set_rwsem); | |
922 | ||
5abb8855 | 923 | if (cset == &init_css_set) { |
3dd06ffa | 924 | res = &root->cgrp; |
7717f7ba | 925 | } else { |
69d0206c TH |
926 | struct cgrp_cset_link *link; |
927 | ||
928 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 929 | struct cgroup *c = link->cgrp; |
69d0206c | 930 | |
7717f7ba PM |
931 | if (c->root == root) { |
932 | res = c; | |
933 | break; | |
934 | } | |
935 | } | |
936 | } | |
96d365e0 | 937 | |
7717f7ba PM |
938 | BUG_ON(!res); |
939 | return res; | |
940 | } | |
941 | ||
ddbcc7e8 | 942 | /* |
ceb6a081 TH |
943 | * Return the cgroup for "task" from the given hierarchy. Must be |
944 | * called with cgroup_mutex and css_set_rwsem held. | |
945 | */ | |
946 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 947 | struct cgroup_root *root) |
ceb6a081 TH |
948 | { |
949 | /* | |
950 | * No need to lock the task - since we hold cgroup_mutex the | |
951 | * task can't change groups, so the only thing that can happen | |
952 | * is that it exits and its css is set back to init_css_set. | |
953 | */ | |
954 | return cset_cgroup_from_root(task_css_set(task), root); | |
955 | } | |
956 | ||
ddbcc7e8 | 957 | /* |
ddbcc7e8 PM |
958 | * A task must hold cgroup_mutex to modify cgroups. |
959 | * | |
960 | * Any task can increment and decrement the count field without lock. | |
961 | * So in general, code holding cgroup_mutex can't rely on the count | |
962 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 963 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
964 | * means that no tasks are currently attached, therefore there is no |
965 | * way a task attached to that cgroup can fork (the other way to | |
966 | * increment the count). So code holding cgroup_mutex can safely | |
967 | * assume that if the count is zero, it will stay zero. Similarly, if | |
968 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
969 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
970 | * needs that mutex. | |
971 | * | |
ddbcc7e8 PM |
972 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
973 | * (usually) take cgroup_mutex. These are the two most performance | |
974 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
975 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
976 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
977 | * to the release agent with the name of the cgroup (path relative to |
978 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
979 | * |
980 | * A cgroup can only be deleted if both its 'count' of using tasks | |
981 | * is zero, and its list of 'children' cgroups is empty. Since all | |
982 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 983 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 984 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 985 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
986 | * |
987 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 988 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
989 | */ |
990 | ||
69dfa00c | 991 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask); |
2bd59d48 | 992 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 993 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 994 | |
8d7e6fb0 TH |
995 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
996 | char *buf) | |
ddbcc7e8 | 997 | { |
8d7e6fb0 TH |
998 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
999 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1000 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
1001 | cft->ss->name, cft->name); | |
1002 | else | |
1003 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1004 | return buf; | |
ddbcc7e8 PM |
1005 | } |
1006 | ||
f2e85d57 TH |
1007 | /** |
1008 | * cgroup_file_mode - deduce file mode of a control file | |
1009 | * @cft: the control file in question | |
1010 | * | |
1011 | * returns cft->mode if ->mode is not 0 | |
1012 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
1013 | * returns S_IRUGO if it has only a read handler | |
1014 | * returns S_IWUSR if it has only a write hander | |
1015 | */ | |
1016 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1017 | { |
f2e85d57 | 1018 | umode_t mode = 0; |
65dff759 | 1019 | |
f2e85d57 TH |
1020 | if (cft->mode) |
1021 | return cft->mode; | |
1022 | ||
1023 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
1024 | mode |= S_IRUGO; | |
1025 | ||
6770c64e | 1026 | if (cft->write_u64 || cft->write_s64 || cft->write) |
f2e85d57 TH |
1027 | mode |= S_IWUSR; |
1028 | ||
1029 | return mode; | |
65dff759 LZ |
1030 | } |
1031 | ||
59f5296b | 1032 | static void cgroup_get(struct cgroup *cgrp) |
be445626 | 1033 | { |
2bd59d48 | 1034 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
9d755d33 | 1035 | css_get(&cgrp->self); |
be445626 LZ |
1036 | } |
1037 | ||
59f5296b | 1038 | static void cgroup_put(struct cgroup *cgrp) |
be445626 | 1039 | { |
9d755d33 | 1040 | css_put(&cgrp->self); |
be445626 LZ |
1041 | } |
1042 | ||
af0ba678 TH |
1043 | /** |
1044 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1045 | * @cgrp: the target cgroup | |
1046 | * | |
1047 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1048 | * enabled together through its ->depends_on mask. In such cases, more | |
1049 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1050 | * | |
1051 | * This function determines which subsystems need to be enabled given the | |
1052 | * current @cgrp->subtree_control and records it in | |
1053 | * @cgrp->child_subsys_mask. The resulting mask is always a superset of | |
1054 | * @cgrp->subtree_control and follows the usual hierarchy rules. | |
1055 | */ | |
667c2491 TH |
1056 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) |
1057 | { | |
af0ba678 TH |
1058 | struct cgroup *parent = cgroup_parent(cgrp); |
1059 | unsigned int cur_ss_mask = cgrp->subtree_control; | |
1060 | struct cgroup_subsys *ss; | |
1061 | int ssid; | |
1062 | ||
1063 | lockdep_assert_held(&cgroup_mutex); | |
1064 | ||
1065 | if (!cgroup_on_dfl(cgrp)) { | |
1066 | cgrp->child_subsys_mask = cur_ss_mask; | |
1067 | return; | |
1068 | } | |
1069 | ||
1070 | while (true) { | |
1071 | unsigned int new_ss_mask = cur_ss_mask; | |
1072 | ||
1073 | for_each_subsys(ss, ssid) | |
1074 | if (cur_ss_mask & (1 << ssid)) | |
1075 | new_ss_mask |= ss->depends_on; | |
1076 | ||
1077 | /* | |
1078 | * Mask out subsystems which aren't available. This can | |
1079 | * happen only if some depended-upon subsystems were bound | |
1080 | * to non-default hierarchies. | |
1081 | */ | |
1082 | if (parent) | |
1083 | new_ss_mask &= parent->child_subsys_mask; | |
1084 | else | |
1085 | new_ss_mask &= cgrp->root->subsys_mask; | |
1086 | ||
1087 | if (new_ss_mask == cur_ss_mask) | |
1088 | break; | |
1089 | cur_ss_mask = new_ss_mask; | |
1090 | } | |
1091 | ||
1092 | cgrp->child_subsys_mask = cur_ss_mask; | |
667c2491 TH |
1093 | } |
1094 | ||
a9746d8d TH |
1095 | /** |
1096 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1097 | * @kn: the kernfs_node being serviced | |
1098 | * | |
1099 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1100 | * the method finishes if locking succeeded. Note that once this function | |
1101 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1102 | * inaccessible any time. If the caller intends to continue to access the | |
1103 | * cgroup, it should pin it before invoking this function. | |
1104 | */ | |
1105 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1106 | { |
a9746d8d TH |
1107 | struct cgroup *cgrp; |
1108 | ||
1109 | if (kernfs_type(kn) == KERNFS_DIR) | |
1110 | cgrp = kn->priv; | |
1111 | else | |
1112 | cgrp = kn->parent->priv; | |
1113 | ||
1114 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1115 | |
1116 | kernfs_unbreak_active_protection(kn); | |
1117 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1118 | } |
1119 | ||
a9746d8d TH |
1120 | /** |
1121 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1122 | * @kn: the kernfs_node being serviced | |
1123 | * | |
1124 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1125 | * @kn. It breaks the active protection, performs cgroup locking and | |
1126 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1127 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1128 | * matching cgroup_kn_unlock() invocation. | |
1129 | * | |
1130 | * Any cgroup kernfs method implementation which requires locking the | |
1131 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1132 | * locking under kernfs active protection and allows all kernfs operations | |
1133 | * including self-removal. | |
1134 | */ | |
1135 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1136 | { |
a9746d8d TH |
1137 | struct cgroup *cgrp; |
1138 | ||
1139 | if (kernfs_type(kn) == KERNFS_DIR) | |
1140 | cgrp = kn->priv; | |
1141 | else | |
1142 | cgrp = kn->parent->priv; | |
05ef1d7c | 1143 | |
2739d3cc | 1144 | /* |
01f6474c | 1145 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1146 | * active_ref. cgroup liveliness check alone provides enough |
1147 | * protection against removal. Ensure @cgrp stays accessible and | |
1148 | * break the active_ref protection. | |
2739d3cc | 1149 | */ |
a9746d8d TH |
1150 | cgroup_get(cgrp); |
1151 | kernfs_break_active_protection(kn); | |
1152 | ||
2bd59d48 | 1153 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1154 | |
a9746d8d TH |
1155 | if (!cgroup_is_dead(cgrp)) |
1156 | return cgrp; | |
1157 | ||
1158 | cgroup_kn_unlock(kn); | |
1159 | return NULL; | |
ddbcc7e8 | 1160 | } |
05ef1d7c | 1161 | |
2739d3cc | 1162 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1163 | { |
2bd59d48 | 1164 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1165 | |
01f6474c | 1166 | lockdep_assert_held(&cgroup_mutex); |
2bd59d48 | 1167 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1168 | } |
1169 | ||
13af07df | 1170 | /** |
628f7cd4 | 1171 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1172 | * @cgrp: target cgroup |
13af07df AR |
1173 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1174 | */ | |
69dfa00c | 1175 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
05ef1d7c | 1176 | { |
13af07df | 1177 | struct cgroup_subsys *ss; |
b420ba7d | 1178 | int i; |
05ef1d7c | 1179 | |
b420ba7d | 1180 | for_each_subsys(ss, i) { |
0adb0704 | 1181 | struct cftype *cfts; |
b420ba7d | 1182 | |
69dfa00c | 1183 | if (!(subsys_mask & (1 << i))) |
13af07df | 1184 | continue; |
0adb0704 TH |
1185 | list_for_each_entry(cfts, &ss->cfts, node) |
1186 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1187 | } |
ddbcc7e8 PM |
1188 | } |
1189 | ||
69dfa00c | 1190 | static int rebind_subsystems(struct cgroup_root *dst_root, unsigned int ss_mask) |
ddbcc7e8 | 1191 | { |
30159ec7 | 1192 | struct cgroup_subsys *ss; |
5533e011 | 1193 | unsigned int tmp_ss_mask; |
2d8f243a | 1194 | int ssid, i, ret; |
ddbcc7e8 | 1195 | |
ace2bee8 | 1196 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1197 | |
5df36032 TH |
1198 | for_each_subsys(ss, ssid) { |
1199 | if (!(ss_mask & (1 << ssid))) | |
1200 | continue; | |
aae8aab4 | 1201 | |
7fd8c565 TH |
1202 | /* if @ss has non-root csses attached to it, can't move */ |
1203 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1204 | return -EBUSY; |
1d5be6b2 | 1205 | |
5df36032 | 1206 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1207 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1208 | return -EBUSY; |
ddbcc7e8 PM |
1209 | } |
1210 | ||
5533e011 TH |
1211 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1212 | tmp_ss_mask = ss_mask; | |
1213 | if (dst_root == &cgrp_dfl_root) | |
1214 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1215 | ||
1216 | ret = cgroup_populate_dir(&dst_root->cgrp, tmp_ss_mask); | |
a2dd4247 TH |
1217 | if (ret) { |
1218 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1219 | return ret; |
ddbcc7e8 | 1220 | |
a2dd4247 TH |
1221 | /* |
1222 | * Rebinding back to the default root is not allowed to | |
1223 | * fail. Using both default and non-default roots should | |
1224 | * be rare. Moving subsystems back and forth even more so. | |
1225 | * Just warn about it and continue. | |
1226 | */ | |
1227 | if (cgrp_dfl_root_visible) { | |
69dfa00c | 1228 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", |
a2a1f9ea | 1229 | ret, ss_mask); |
ed3d261b | 1230 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); |
a2dd4247 | 1231 | } |
5df36032 | 1232 | } |
3126121f TH |
1233 | |
1234 | /* | |
1235 | * Nothing can fail from this point on. Remove files for the | |
1236 | * removed subsystems and rebind each subsystem. | |
1237 | */ | |
5df36032 | 1238 | for_each_subsys(ss, ssid) |
a2dd4247 | 1239 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1240 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
a8a648c4 | 1241 | |
5df36032 | 1242 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1243 | struct cgroup_root *src_root; |
5df36032 | 1244 | struct cgroup_subsys_state *css; |
2d8f243a | 1245 | struct css_set *cset; |
a8a648c4 | 1246 | |
5df36032 TH |
1247 | if (!(ss_mask & (1 << ssid))) |
1248 | continue; | |
a8a648c4 | 1249 | |
5df36032 | 1250 | src_root = ss->root; |
3dd06ffa | 1251 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1252 | |
3dd06ffa | 1253 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1254 | |
3dd06ffa TH |
1255 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1256 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1257 | ss->root = dst_root; |
3dd06ffa | 1258 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1259 | |
2d8f243a TH |
1260 | down_write(&css_set_rwsem); |
1261 | hash_for_each(css_set_table, i, cset, hlist) | |
1262 | list_move_tail(&cset->e_cset_node[ss->id], | |
1263 | &dst_root->cgrp.e_csets[ss->id]); | |
1264 | up_write(&css_set_rwsem); | |
1265 | ||
f392e51c | 1266 | src_root->subsys_mask &= ~(1 << ssid); |
667c2491 TH |
1267 | src_root->cgrp.subtree_control &= ~(1 << ssid); |
1268 | cgroup_refresh_child_subsys_mask(&src_root->cgrp); | |
f392e51c | 1269 | |
bd53d617 | 1270 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1271 | dst_root->subsys_mask |= 1 << ssid; |
667c2491 TH |
1272 | if (dst_root != &cgrp_dfl_root) { |
1273 | dst_root->cgrp.subtree_control |= 1 << ssid; | |
1274 | cgroup_refresh_child_subsys_mask(&dst_root->cgrp); | |
1275 | } | |
a8a648c4 | 1276 | |
5df36032 TH |
1277 | if (ss->bind) |
1278 | ss->bind(css); | |
ddbcc7e8 | 1279 | } |
ddbcc7e8 | 1280 | |
a2dd4247 | 1281 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1282 | return 0; |
1283 | } | |
1284 | ||
2bd59d48 TH |
1285 | static int cgroup_show_options(struct seq_file *seq, |
1286 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1287 | { |
3dd06ffa | 1288 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1289 | struct cgroup_subsys *ss; |
b85d2040 | 1290 | int ssid; |
ddbcc7e8 | 1291 | |
b85d2040 | 1292 | for_each_subsys(ss, ssid) |
f392e51c | 1293 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1294 | seq_printf(seq, ",%s", ss->name); |
93438629 | 1295 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1296 | seq_puts(seq, ",noprefix"); |
93438629 | 1297 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1298 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1299 | |
1300 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1301 | if (strlen(root->release_agent_path)) |
1302 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1303 | spin_unlock(&release_agent_path_lock); |
1304 | ||
3dd06ffa | 1305 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1306 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1307 | if (strlen(root->name)) |
1308 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1309 | return 0; |
1310 | } | |
1311 | ||
1312 | struct cgroup_sb_opts { | |
69dfa00c TH |
1313 | unsigned int subsys_mask; |
1314 | unsigned int flags; | |
81a6a5cd | 1315 | char *release_agent; |
2260e7fc | 1316 | bool cpuset_clone_children; |
c6d57f33 | 1317 | char *name; |
2c6ab6d2 PM |
1318 | /* User explicitly requested empty subsystem */ |
1319 | bool none; | |
ddbcc7e8 PM |
1320 | }; |
1321 | ||
cf5d5941 | 1322 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1323 | { |
32a8cf23 DL |
1324 | char *token, *o = data; |
1325 | bool all_ss = false, one_ss = false; | |
69dfa00c | 1326 | unsigned int mask = -1U; |
30159ec7 | 1327 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1328 | int nr_opts = 0; |
30159ec7 | 1329 | int i; |
f9ab5b5b LZ |
1330 | |
1331 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1332 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1333 | #endif |
ddbcc7e8 | 1334 | |
c6d57f33 | 1335 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1336 | |
1337 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1338 | nr_opts++; |
1339 | ||
ddbcc7e8 PM |
1340 | if (!*token) |
1341 | return -EINVAL; | |
32a8cf23 | 1342 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1343 | /* Explicitly have no subsystems */ |
1344 | opts->none = true; | |
32a8cf23 DL |
1345 | continue; |
1346 | } | |
1347 | if (!strcmp(token, "all")) { | |
1348 | /* Mutually exclusive option 'all' + subsystem name */ | |
1349 | if (one_ss) | |
1350 | return -EINVAL; | |
1351 | all_ss = true; | |
1352 | continue; | |
1353 | } | |
873fe09e TH |
1354 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1355 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1356 | continue; | |
1357 | } | |
32a8cf23 | 1358 | if (!strcmp(token, "noprefix")) { |
93438629 | 1359 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1360 | continue; |
1361 | } | |
1362 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1363 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1364 | continue; |
1365 | } | |
03b1cde6 | 1366 | if (!strcmp(token, "xattr")) { |
93438629 | 1367 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1368 | continue; |
1369 | } | |
32a8cf23 | 1370 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1371 | /* Specifying two release agents is forbidden */ |
1372 | if (opts->release_agent) | |
1373 | return -EINVAL; | |
c6d57f33 | 1374 | opts->release_agent = |
e400c285 | 1375 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1376 | if (!opts->release_agent) |
1377 | return -ENOMEM; | |
32a8cf23 DL |
1378 | continue; |
1379 | } | |
1380 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1381 | const char *name = token + 5; |
1382 | /* Can't specify an empty name */ | |
1383 | if (!strlen(name)) | |
1384 | return -EINVAL; | |
1385 | /* Must match [\w.-]+ */ | |
1386 | for (i = 0; i < strlen(name); i++) { | |
1387 | char c = name[i]; | |
1388 | if (isalnum(c)) | |
1389 | continue; | |
1390 | if ((c == '.') || (c == '-') || (c == '_')) | |
1391 | continue; | |
1392 | return -EINVAL; | |
1393 | } | |
1394 | /* Specifying two names is forbidden */ | |
1395 | if (opts->name) | |
1396 | return -EINVAL; | |
1397 | opts->name = kstrndup(name, | |
e400c285 | 1398 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1399 | GFP_KERNEL); |
1400 | if (!opts->name) | |
1401 | return -ENOMEM; | |
32a8cf23 DL |
1402 | |
1403 | continue; | |
1404 | } | |
1405 | ||
30159ec7 | 1406 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1407 | if (strcmp(token, ss->name)) |
1408 | continue; | |
1409 | if (ss->disabled) | |
1410 | continue; | |
1411 | ||
1412 | /* Mutually exclusive option 'all' + subsystem name */ | |
1413 | if (all_ss) | |
1414 | return -EINVAL; | |
69dfa00c | 1415 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1416 | one_ss = true; |
1417 | ||
1418 | break; | |
1419 | } | |
1420 | if (i == CGROUP_SUBSYS_COUNT) | |
1421 | return -ENOENT; | |
1422 | } | |
1423 | ||
873fe09e | 1424 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1425 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
7b9a6ba5 TH |
1426 | if (nr_opts != 1) { |
1427 | pr_err("sane_behavior: no other mount options allowed\n"); | |
873fe09e TH |
1428 | return -EINVAL; |
1429 | } | |
7b9a6ba5 | 1430 | return 0; |
873fe09e TH |
1431 | } |
1432 | ||
7b9a6ba5 TH |
1433 | /* |
1434 | * If the 'all' option was specified select all the subsystems, | |
1435 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1436 | * not specified, let's default to 'all' | |
1437 | */ | |
1438 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1439 | for_each_subsys(ss, i) | |
1440 | if (!ss->disabled) | |
1441 | opts->subsys_mask |= (1 << i); | |
1442 | ||
1443 | /* | |
1444 | * We either have to specify by name or by subsystems. (So all | |
1445 | * empty hierarchies must have a name). | |
1446 | */ | |
1447 | if (!opts->subsys_mask && !opts->name) | |
1448 | return -EINVAL; | |
1449 | ||
f9ab5b5b LZ |
1450 | /* |
1451 | * Option noprefix was introduced just for backward compatibility | |
1452 | * with the old cpuset, so we allow noprefix only if mounting just | |
1453 | * the cpuset subsystem. | |
1454 | */ | |
93438629 | 1455 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1456 | return -EINVAL; |
1457 | ||
2c6ab6d2 | 1458 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1459 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1460 | return -EINVAL; |
1461 | ||
ddbcc7e8 PM |
1462 | return 0; |
1463 | } | |
1464 | ||
2bd59d48 | 1465 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1466 | { |
1467 | int ret = 0; | |
3dd06ffa | 1468 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1469 | struct cgroup_sb_opts opts; |
69dfa00c | 1470 | unsigned int added_mask, removed_mask; |
ddbcc7e8 | 1471 | |
aa6ec29b TH |
1472 | if (root == &cgrp_dfl_root) { |
1473 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1474 | return -EINVAL; |
1475 | } | |
1476 | ||
ddbcc7e8 PM |
1477 | mutex_lock(&cgroup_mutex); |
1478 | ||
1479 | /* See what subsystems are wanted */ | |
1480 | ret = parse_cgroupfs_options(data, &opts); | |
1481 | if (ret) | |
1482 | goto out_unlock; | |
1483 | ||
f392e51c | 1484 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1485 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1486 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1487 | |
f392e51c TH |
1488 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1489 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1490 | |
cf5d5941 | 1491 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1492 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1493 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1494 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1495 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1496 | ret = -EINVAL; |
1497 | goto out_unlock; | |
1498 | } | |
1499 | ||
f172e67c | 1500 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1501 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1502 | ret = -EBUSY; |
0670e08b | 1503 | goto out_unlock; |
cf5d5941 | 1504 | } |
ddbcc7e8 | 1505 | |
5df36032 | 1506 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1507 | if (ret) |
0670e08b | 1508 | goto out_unlock; |
ddbcc7e8 | 1509 | |
3dd06ffa | 1510 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1511 | |
69e943b7 TH |
1512 | if (opts.release_agent) { |
1513 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1514 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1515 | spin_unlock(&release_agent_path_lock); |
1516 | } | |
ddbcc7e8 | 1517 | out_unlock: |
66bdc9cf | 1518 | kfree(opts.release_agent); |
c6d57f33 | 1519 | kfree(opts.name); |
ddbcc7e8 | 1520 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1521 | return ret; |
1522 | } | |
1523 | ||
afeb0f9f TH |
1524 | /* |
1525 | * To reduce the fork() overhead for systems that are not actually using | |
1526 | * their cgroups capability, we don't maintain the lists running through | |
1527 | * each css_set to its tasks until we see the list actually used - in other | |
1528 | * words after the first mount. | |
1529 | */ | |
1530 | static bool use_task_css_set_links __read_mostly; | |
1531 | ||
1532 | static void cgroup_enable_task_cg_lists(void) | |
1533 | { | |
1534 | struct task_struct *p, *g; | |
1535 | ||
96d365e0 | 1536 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1537 | |
1538 | if (use_task_css_set_links) | |
1539 | goto out_unlock; | |
1540 | ||
1541 | use_task_css_set_links = true; | |
1542 | ||
1543 | /* | |
1544 | * We need tasklist_lock because RCU is not safe against | |
1545 | * while_each_thread(). Besides, a forking task that has passed | |
1546 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1547 | * is not guaranteed to have its child immediately visible in the | |
1548 | * tasklist if we walk through it with RCU. | |
1549 | */ | |
1550 | read_lock(&tasklist_lock); | |
1551 | do_each_thread(g, p) { | |
afeb0f9f TH |
1552 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1553 | task_css_set(p) != &init_css_set); | |
1554 | ||
1555 | /* | |
1556 | * We should check if the process is exiting, otherwise | |
1557 | * it will race with cgroup_exit() in that the list | |
1558 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1559 | * Do it while holding siglock so that we don't end up |
1560 | * racing against cgroup_exit(). | |
afeb0f9f | 1561 | */ |
f153ad11 | 1562 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1563 | if (!(p->flags & PF_EXITING)) { |
1564 | struct css_set *cset = task_css_set(p); | |
1565 | ||
1566 | list_add(&p->cg_list, &cset->tasks); | |
1567 | get_css_set(cset); | |
1568 | } | |
f153ad11 | 1569 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1570 | } while_each_thread(g, p); |
1571 | read_unlock(&tasklist_lock); | |
1572 | out_unlock: | |
96d365e0 | 1573 | up_write(&css_set_rwsem); |
afeb0f9f | 1574 | } |
ddbcc7e8 | 1575 | |
cc31edce PM |
1576 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1577 | { | |
2d8f243a TH |
1578 | struct cgroup_subsys *ss; |
1579 | int ssid; | |
1580 | ||
d5c419b6 TH |
1581 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1582 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1583 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1584 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1585 | INIT_LIST_HEAD(&cgrp->pidlists); |
1586 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1587 | cgrp->self.cgroup = cgrp; |
184faf32 | 1588 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1589 | |
1590 | for_each_subsys(ss, ssid) | |
1591 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1592 | |
1593 | init_waitqueue_head(&cgrp->offline_waitq); | |
cc31edce | 1594 | } |
c6d57f33 | 1595 | |
3dd06ffa | 1596 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1597 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1598 | { |
3dd06ffa | 1599 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1600 | |
ddbcc7e8 | 1601 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1602 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1603 | cgrp->root = root; |
cc31edce | 1604 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1605 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1606 | |
c6d57f33 PM |
1607 | root->flags = opts->flags; |
1608 | if (opts->release_agent) | |
1609 | strcpy(root->release_agent_path, opts->release_agent); | |
1610 | if (opts->name) | |
1611 | strcpy(root->name, opts->name); | |
2260e7fc | 1612 | if (opts->cpuset_clone_children) |
3dd06ffa | 1613 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1614 | } |
1615 | ||
69dfa00c | 1616 | static int cgroup_setup_root(struct cgroup_root *root, unsigned int ss_mask) |
2c6ab6d2 | 1617 | { |
d427dfeb | 1618 | LIST_HEAD(tmp_links); |
3dd06ffa | 1619 | struct cgroup *root_cgrp = &root->cgrp; |
a14c6874 | 1620 | struct cftype *base_files; |
d427dfeb | 1621 | struct css_set *cset; |
d427dfeb | 1622 | int i, ret; |
2c6ab6d2 | 1623 | |
d427dfeb | 1624 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1625 | |
6fa4918d | 1626 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_NOWAIT); |
d427dfeb | 1627 | if (ret < 0) |
2bd59d48 | 1628 | goto out; |
d427dfeb | 1629 | root_cgrp->id = ret; |
c6d57f33 | 1630 | |
9d755d33 TH |
1631 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release); |
1632 | if (ret) | |
1633 | goto out; | |
1634 | ||
d427dfeb | 1635 | /* |
96d365e0 | 1636 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1637 | * but that's OK - it can only be increased by someone holding |
1638 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1639 | * have some link structures left over | |
1640 | */ | |
1641 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1642 | if (ret) | |
9d755d33 | 1643 | goto cancel_ref; |
ddbcc7e8 | 1644 | |
985ed670 | 1645 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1646 | if (ret) |
9d755d33 | 1647 | goto cancel_ref; |
ddbcc7e8 | 1648 | |
2bd59d48 TH |
1649 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1650 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1651 | root_cgrp); | |
1652 | if (IS_ERR(root->kf_root)) { | |
1653 | ret = PTR_ERR(root->kf_root); | |
1654 | goto exit_root_id; | |
1655 | } | |
1656 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1657 | |
a14c6874 TH |
1658 | if (root == &cgrp_dfl_root) |
1659 | base_files = cgroup_dfl_base_files; | |
1660 | else | |
1661 | base_files = cgroup_legacy_base_files; | |
1662 | ||
1663 | ret = cgroup_addrm_files(root_cgrp, base_files, true); | |
d427dfeb | 1664 | if (ret) |
2bd59d48 | 1665 | goto destroy_root; |
ddbcc7e8 | 1666 | |
5df36032 | 1667 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1668 | if (ret) |
2bd59d48 | 1669 | goto destroy_root; |
ddbcc7e8 | 1670 | |
d427dfeb TH |
1671 | /* |
1672 | * There must be no failure case after here, since rebinding takes | |
1673 | * care of subsystems' refcounts, which are explicitly dropped in | |
1674 | * the failure exit path. | |
1675 | */ | |
1676 | list_add(&root->root_list, &cgroup_roots); | |
1677 | cgroup_root_count++; | |
0df6a63f | 1678 | |
d427dfeb | 1679 | /* |
3dd06ffa | 1680 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1681 | * objects. |
1682 | */ | |
96d365e0 | 1683 | down_write(&css_set_rwsem); |
d427dfeb TH |
1684 | hash_for_each(css_set_table, i, cset, hlist) |
1685 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1686 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1687 | |
d5c419b6 | 1688 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1689 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1690 | |
2bd59d48 | 1691 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1692 | ret = 0; |
2bd59d48 | 1693 | goto out; |
d427dfeb | 1694 | |
2bd59d48 TH |
1695 | destroy_root: |
1696 | kernfs_destroy_root(root->kf_root); | |
1697 | root->kf_root = NULL; | |
1698 | exit_root_id: | |
d427dfeb | 1699 | cgroup_exit_root_id(root); |
9d755d33 | 1700 | cancel_ref: |
9a1049da | 1701 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1702 | out: |
d427dfeb TH |
1703 | free_cgrp_cset_links(&tmp_links); |
1704 | return ret; | |
ddbcc7e8 PM |
1705 | } |
1706 | ||
f7e83571 | 1707 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1708 | int flags, const char *unused_dev_name, |
f7e83571 | 1709 | void *data) |
ddbcc7e8 | 1710 | { |
3a32bd72 | 1711 | struct super_block *pinned_sb = NULL; |
970317aa | 1712 | struct cgroup_subsys *ss; |
3dd06ffa | 1713 | struct cgroup_root *root; |
ddbcc7e8 | 1714 | struct cgroup_sb_opts opts; |
2bd59d48 | 1715 | struct dentry *dentry; |
8e30e2b8 | 1716 | int ret; |
970317aa | 1717 | int i; |
c6b3d5bc | 1718 | bool new_sb; |
ddbcc7e8 | 1719 | |
56fde9e0 TH |
1720 | /* |
1721 | * The first time anyone tries to mount a cgroup, enable the list | |
1722 | * linking each css_set to its tasks and fix up all existing tasks. | |
1723 | */ | |
1724 | if (!use_task_css_set_links) | |
1725 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1726 | |
aae8aab4 | 1727 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1728 | |
1729 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1730 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1731 | if (ret) |
8e30e2b8 | 1732 | goto out_unlock; |
a015edd2 | 1733 | |
2bd59d48 | 1734 | /* look for a matching existing root */ |
7b9a6ba5 | 1735 | if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2dd4247 TH |
1736 | cgrp_dfl_root_visible = true; |
1737 | root = &cgrp_dfl_root; | |
1738 | cgroup_get(&root->cgrp); | |
1739 | ret = 0; | |
1740 | goto out_unlock; | |
ddbcc7e8 PM |
1741 | } |
1742 | ||
970317aa LZ |
1743 | /* |
1744 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1745 | * still be dying after the previous unmount. Let's drain the | |
1746 | * dying subsystems. We just need to ensure that the ones | |
1747 | * unmounted previously finish dying and don't care about new ones | |
1748 | * starting. Testing ref liveliness is good enough. | |
1749 | */ | |
1750 | for_each_subsys(ss, i) { | |
1751 | if (!(opts.subsys_mask & (1 << i)) || | |
1752 | ss->root == &cgrp_dfl_root) | |
1753 | continue; | |
1754 | ||
1755 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1756 | mutex_unlock(&cgroup_mutex); | |
1757 | msleep(10); | |
1758 | ret = restart_syscall(); | |
1759 | goto out_free; | |
1760 | } | |
1761 | cgroup_put(&ss->root->cgrp); | |
1762 | } | |
1763 | ||
985ed670 | 1764 | for_each_root(root) { |
2bd59d48 | 1765 | bool name_match = false; |
3126121f | 1766 | |
3dd06ffa | 1767 | if (root == &cgrp_dfl_root) |
985ed670 | 1768 | continue; |
3126121f | 1769 | |
cf5d5941 | 1770 | /* |
2bd59d48 TH |
1771 | * If we asked for a name then it must match. Also, if |
1772 | * name matches but sybsys_mask doesn't, we should fail. | |
1773 | * Remember whether name matched. | |
cf5d5941 | 1774 | */ |
2bd59d48 TH |
1775 | if (opts.name) { |
1776 | if (strcmp(opts.name, root->name)) | |
1777 | continue; | |
1778 | name_match = true; | |
1779 | } | |
ddbcc7e8 | 1780 | |
c6d57f33 | 1781 | /* |
2bd59d48 TH |
1782 | * If we asked for subsystems (or explicitly for no |
1783 | * subsystems) then they must match. | |
c6d57f33 | 1784 | */ |
2bd59d48 | 1785 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1786 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1787 | if (!name_match) |
1788 | continue; | |
1789 | ret = -EBUSY; | |
1790 | goto out_unlock; | |
1791 | } | |
873fe09e | 1792 | |
7b9a6ba5 TH |
1793 | if (root->flags ^ opts.flags) |
1794 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 1795 | |
776f02fa | 1796 | /* |
3a32bd72 LZ |
1797 | * We want to reuse @root whose lifetime is governed by its |
1798 | * ->cgrp. Let's check whether @root is alive and keep it | |
1799 | * that way. As cgroup_kill_sb() can happen anytime, we | |
1800 | * want to block it by pinning the sb so that @root doesn't | |
1801 | * get killed before mount is complete. | |
1802 | * | |
1803 | * With the sb pinned, tryget_live can reliably indicate | |
1804 | * whether @root can be reused. If it's being killed, | |
1805 | * drain it. We can use wait_queue for the wait but this | |
1806 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 1807 | */ |
3a32bd72 LZ |
1808 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
1809 | if (IS_ERR(pinned_sb) || | |
1810 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 1811 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
1812 | if (!IS_ERR_OR_NULL(pinned_sb)) |
1813 | deactivate_super(pinned_sb); | |
776f02fa | 1814 | msleep(10); |
a015edd2 TH |
1815 | ret = restart_syscall(); |
1816 | goto out_free; | |
776f02fa | 1817 | } |
ddbcc7e8 | 1818 | |
776f02fa | 1819 | ret = 0; |
2bd59d48 | 1820 | goto out_unlock; |
ddbcc7e8 | 1821 | } |
ddbcc7e8 | 1822 | |
817929ec | 1823 | /* |
172a2c06 TH |
1824 | * No such thing, create a new one. name= matching without subsys |
1825 | * specification is allowed for already existing hierarchies but we | |
1826 | * can't create new one without subsys specification. | |
817929ec | 1827 | */ |
172a2c06 TH |
1828 | if (!opts.subsys_mask && !opts.none) { |
1829 | ret = -EINVAL; | |
1830 | goto out_unlock; | |
817929ec | 1831 | } |
817929ec | 1832 | |
172a2c06 TH |
1833 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1834 | if (!root) { | |
1835 | ret = -ENOMEM; | |
2bd59d48 | 1836 | goto out_unlock; |
839ec545 | 1837 | } |
e5f6a860 | 1838 | |
172a2c06 TH |
1839 | init_cgroup_root(root, &opts); |
1840 | ||
35585573 | 1841 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1842 | if (ret) |
1843 | cgroup_free_root(root); | |
fa3ca07e | 1844 | |
8e30e2b8 | 1845 | out_unlock: |
ddbcc7e8 | 1846 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 1847 | out_free: |
c6d57f33 PM |
1848 | kfree(opts.release_agent); |
1849 | kfree(opts.name); | |
03b1cde6 | 1850 | |
2bd59d48 | 1851 | if (ret) |
8e30e2b8 | 1852 | return ERR_PTR(ret); |
2bd59d48 | 1853 | |
c9482a5b JZ |
1854 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
1855 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 1856 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 1857 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
1858 | |
1859 | /* | |
1860 | * If @pinned_sb, we're reusing an existing root and holding an | |
1861 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
1862 | */ | |
1863 | if (pinned_sb) { | |
1864 | WARN_ON(new_sb); | |
1865 | deactivate_super(pinned_sb); | |
1866 | } | |
1867 | ||
2bd59d48 TH |
1868 | return dentry; |
1869 | } | |
1870 | ||
1871 | static void cgroup_kill_sb(struct super_block *sb) | |
1872 | { | |
1873 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1874 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1875 | |
9d755d33 TH |
1876 | /* |
1877 | * If @root doesn't have any mounts or children, start killing it. | |
1878 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
1879 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
1880 | * |
1881 | * And don't kill the default root. | |
9d755d33 | 1882 | */ |
1f779fb2 LZ |
1883 | if (css_has_online_children(&root->cgrp.self) || |
1884 | root == &cgrp_dfl_root) | |
9d755d33 TH |
1885 | cgroup_put(&root->cgrp); |
1886 | else | |
1887 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
1888 | ||
2bd59d48 | 1889 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1890 | } |
1891 | ||
1892 | static struct file_system_type cgroup_fs_type = { | |
1893 | .name = "cgroup", | |
f7e83571 | 1894 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1895 | .kill_sb = cgroup_kill_sb, |
1896 | }; | |
1897 | ||
676db4af GK |
1898 | static struct kobject *cgroup_kobj; |
1899 | ||
857a2beb | 1900 | /** |
913ffdb5 | 1901 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1902 | * @task: target task |
857a2beb TH |
1903 | * @buf: the buffer to write the path into |
1904 | * @buflen: the length of the buffer | |
1905 | * | |
913ffdb5 TH |
1906 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1907 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1908 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1909 | * cgroup controller callbacks. | |
1910 | * | |
e61734c5 | 1911 | * Return value is the same as kernfs_path(). |
857a2beb | 1912 | */ |
e61734c5 | 1913 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1914 | { |
3dd06ffa | 1915 | struct cgroup_root *root; |
913ffdb5 | 1916 | struct cgroup *cgrp; |
e61734c5 TH |
1917 | int hierarchy_id = 1; |
1918 | char *path = NULL; | |
857a2beb TH |
1919 | |
1920 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1921 | down_read(&css_set_rwsem); |
857a2beb | 1922 | |
913ffdb5 TH |
1923 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1924 | ||
857a2beb TH |
1925 | if (root) { |
1926 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1927 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1928 | } else { |
1929 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1930 | if (strlcpy(buf, "/", buflen) < buflen) |
1931 | path = buf; | |
857a2beb TH |
1932 | } |
1933 | ||
96d365e0 | 1934 | up_read(&css_set_rwsem); |
857a2beb | 1935 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1936 | return path; |
857a2beb | 1937 | } |
913ffdb5 | 1938 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1939 | |
b3dc094e | 1940 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1941 | struct cgroup_taskset { |
b3dc094e TH |
1942 | /* the src and dst cset list running through cset->mg_node */ |
1943 | struct list_head src_csets; | |
1944 | struct list_head dst_csets; | |
1945 | ||
1946 | /* | |
1947 | * Fields for cgroup_taskset_*() iteration. | |
1948 | * | |
1949 | * Before migration is committed, the target migration tasks are on | |
1950 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1951 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1952 | * or ->dst_csets depending on whether migration is committed. | |
1953 | * | |
1954 | * ->cur_csets and ->cur_task point to the current task position | |
1955 | * during iteration. | |
1956 | */ | |
1957 | struct list_head *csets; | |
1958 | struct css_set *cur_cset; | |
1959 | struct task_struct *cur_task; | |
2f7ee569 TH |
1960 | }; |
1961 | ||
1962 | /** | |
1963 | * cgroup_taskset_first - reset taskset and return the first task | |
1964 | * @tset: taskset of interest | |
1965 | * | |
1966 | * @tset iteration is initialized and the first task is returned. | |
1967 | */ | |
1968 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1969 | { | |
b3dc094e TH |
1970 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1971 | tset->cur_task = NULL; | |
1972 | ||
1973 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1974 | } |
2f7ee569 TH |
1975 | |
1976 | /** | |
1977 | * cgroup_taskset_next - iterate to the next task in taskset | |
1978 | * @tset: taskset of interest | |
1979 | * | |
1980 | * Return the next task in @tset. Iteration must have been initialized | |
1981 | * with cgroup_taskset_first(). | |
1982 | */ | |
1983 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1984 | { | |
b3dc094e TH |
1985 | struct css_set *cset = tset->cur_cset; |
1986 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1987 | |
b3dc094e TH |
1988 | while (&cset->mg_node != tset->csets) { |
1989 | if (!task) | |
1990 | task = list_first_entry(&cset->mg_tasks, | |
1991 | struct task_struct, cg_list); | |
1992 | else | |
1993 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1994 | |
b3dc094e TH |
1995 | if (&task->cg_list != &cset->mg_tasks) { |
1996 | tset->cur_cset = cset; | |
1997 | tset->cur_task = task; | |
1998 | return task; | |
1999 | } | |
2f7ee569 | 2000 | |
b3dc094e TH |
2001 | cset = list_next_entry(cset, mg_node); |
2002 | task = NULL; | |
2003 | } | |
2f7ee569 | 2004 | |
b3dc094e | 2005 | return NULL; |
2f7ee569 | 2006 | } |
2f7ee569 | 2007 | |
cb0f1fe9 | 2008 | /** |
74a1166d | 2009 | * cgroup_task_migrate - move a task from one cgroup to another. |
60106946 | 2010 | * @old_cgrp: the cgroup @tsk is being migrated from |
cb0f1fe9 TH |
2011 | * @tsk: the task being migrated |
2012 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 2013 | * |
cb0f1fe9 | 2014 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 2015 | */ |
5abb8855 TH |
2016 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
2017 | struct task_struct *tsk, | |
2018 | struct css_set *new_cset) | |
74a1166d | 2019 | { |
5abb8855 | 2020 | struct css_set *old_cset; |
74a1166d | 2021 | |
cb0f1fe9 TH |
2022 | lockdep_assert_held(&cgroup_mutex); |
2023 | lockdep_assert_held(&css_set_rwsem); | |
2024 | ||
74a1166d | 2025 | /* |
026085ef MSB |
2026 | * We are synchronized through threadgroup_lock() against PF_EXITING |
2027 | * setting such that we can't race against cgroup_exit() changing the | |
2028 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 2029 | */ |
c84cdf75 | 2030 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 2031 | old_cset = task_css_set(tsk); |
74a1166d | 2032 | |
b3dc094e | 2033 | get_css_set(new_cset); |
5abb8855 | 2034 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 2035 | |
1b9aba49 TH |
2036 | /* |
2037 | * Use move_tail so that cgroup_taskset_first() still returns the | |
2038 | * leader after migration. This works because cgroup_migrate() | |
2039 | * ensures that the dst_cset of the leader is the first on the | |
2040 | * tset's dst_csets list. | |
2041 | */ | |
2042 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
2043 | |
2044 | /* | |
5abb8855 TH |
2045 | * We just gained a reference on old_cset by taking it from the |
2046 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
2047 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 2048 | */ |
5abb8855 | 2049 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 2050 | put_css_set_locked(old_cset, false); |
74a1166d BB |
2051 | } |
2052 | ||
a043e3b2 | 2053 | /** |
1958d2d5 TH |
2054 | * cgroup_migrate_finish - cleanup after attach |
2055 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2056 | * |
1958d2d5 TH |
2057 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2058 | * those functions for details. | |
74a1166d | 2059 | */ |
1958d2d5 | 2060 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2061 | { |
1958d2d5 | 2062 | struct css_set *cset, *tmp_cset; |
74a1166d | 2063 | |
1958d2d5 TH |
2064 | lockdep_assert_held(&cgroup_mutex); |
2065 | ||
2066 | down_write(&css_set_rwsem); | |
2067 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2068 | cset->mg_src_cgrp = NULL; | |
2069 | cset->mg_dst_cset = NULL; | |
2070 | list_del_init(&cset->mg_preload_node); | |
2071 | put_css_set_locked(cset, false); | |
2072 | } | |
2073 | up_write(&css_set_rwsem); | |
2074 | } | |
2075 | ||
2076 | /** | |
2077 | * cgroup_migrate_add_src - add a migration source css_set | |
2078 | * @src_cset: the source css_set to add | |
2079 | * @dst_cgrp: the destination cgroup | |
2080 | * @preloaded_csets: list of preloaded css_sets | |
2081 | * | |
2082 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2083 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2084 | * up by cgroup_migrate_finish(). | |
2085 | * | |
2086 | * This function may be called without holding threadgroup_lock even if the | |
2087 | * target is a process. Threads may be created and destroyed but as long | |
2088 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
2089 | * the preloaded css_sets are guaranteed to cover all migrations. | |
2090 | */ | |
2091 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2092 | struct cgroup *dst_cgrp, | |
2093 | struct list_head *preloaded_csets) | |
2094 | { | |
2095 | struct cgroup *src_cgrp; | |
2096 | ||
2097 | lockdep_assert_held(&cgroup_mutex); | |
2098 | lockdep_assert_held(&css_set_rwsem); | |
2099 | ||
2100 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2101 | ||
1958d2d5 TH |
2102 | if (!list_empty(&src_cset->mg_preload_node)) |
2103 | return; | |
2104 | ||
2105 | WARN_ON(src_cset->mg_src_cgrp); | |
2106 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2107 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2108 | ||
2109 | src_cset->mg_src_cgrp = src_cgrp; | |
2110 | get_css_set(src_cset); | |
2111 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2112 | } | |
2113 | ||
2114 | /** | |
2115 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2116 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2117 | * @preloaded_csets: list of preloaded source css_sets |
2118 | * | |
2119 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2120 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2121 | * pins all destination css_sets, links each to its source, and append them |
2122 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2123 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2124 | * |
2125 | * This function must be called after cgroup_migrate_add_src() has been | |
2126 | * called on each migration source css_set. After migration is performed | |
2127 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2128 | * @preloaded_csets. | |
2129 | */ | |
2130 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2131 | struct list_head *preloaded_csets) | |
2132 | { | |
2133 | LIST_HEAD(csets); | |
f817de98 | 2134 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2135 | |
2136 | lockdep_assert_held(&cgroup_mutex); | |
2137 | ||
f8f22e53 TH |
2138 | /* |
2139 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2140 | * with tasks so that child cgroups don't compete against tasks. | |
2141 | */ | |
d51f39b0 | 2142 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2143 | dst_cgrp->child_subsys_mask) |
2144 | return -EBUSY; | |
2145 | ||
1958d2d5 | 2146 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2147 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2148 | struct css_set *dst_cset; |
2149 | ||
f817de98 TH |
2150 | dst_cset = find_css_set(src_cset, |
2151 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2152 | if (!dst_cset) |
2153 | goto err; | |
2154 | ||
2155 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2156 | |
2157 | /* | |
2158 | * If src cset equals dst, it's noop. Drop the src. | |
2159 | * cgroup_migrate() will skip the cset too. Note that we | |
2160 | * can't handle src == dst as some nodes are used by both. | |
2161 | */ | |
2162 | if (src_cset == dst_cset) { | |
2163 | src_cset->mg_src_cgrp = NULL; | |
2164 | list_del_init(&src_cset->mg_preload_node); | |
2165 | put_css_set(src_cset, false); | |
2166 | put_css_set(dst_cset, false); | |
2167 | continue; | |
2168 | } | |
2169 | ||
1958d2d5 TH |
2170 | src_cset->mg_dst_cset = dst_cset; |
2171 | ||
2172 | if (list_empty(&dst_cset->mg_preload_node)) | |
2173 | list_add(&dst_cset->mg_preload_node, &csets); | |
2174 | else | |
2175 | put_css_set(dst_cset, false); | |
2176 | } | |
2177 | ||
f817de98 | 2178 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2179 | return 0; |
2180 | err: | |
2181 | cgroup_migrate_finish(&csets); | |
2182 | return -ENOMEM; | |
2183 | } | |
2184 | ||
2185 | /** | |
2186 | * cgroup_migrate - migrate a process or task to a cgroup | |
2187 | * @cgrp: the destination cgroup | |
2188 | * @leader: the leader of the process or the task to migrate | |
2189 | * @threadgroup: whether @leader points to the whole process or a single task | |
2190 | * | |
2191 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2192 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2193 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2194 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2195 | * function and following up with cgroup_migrate_finish(). | |
2196 | * | |
2197 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2198 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2199 | * failure, when migrating multiple targets, the success or failure can be | |
2200 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2201 | * actually starting migrating. | |
2202 | */ | |
2203 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2204 | bool threadgroup) | |
74a1166d | 2205 | { |
b3dc094e TH |
2206 | struct cgroup_taskset tset = { |
2207 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2208 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2209 | .csets = &tset.src_csets, | |
2210 | }; | |
1c6727af | 2211 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2212 | struct css_set *cset, *tmp_cset; |
2213 | struct task_struct *task, *tmp_task; | |
2214 | int i, ret; | |
74a1166d | 2215 | |
fb5d2b4c MSB |
2216 | /* |
2217 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2218 | * already PF_EXITING could be freed from underneath us unless we | |
2219 | * take an rcu_read_lock. | |
2220 | */ | |
b3dc094e | 2221 | down_write(&css_set_rwsem); |
fb5d2b4c | 2222 | rcu_read_lock(); |
9db8de37 | 2223 | task = leader; |
74a1166d | 2224 | do { |
9db8de37 TH |
2225 | /* @task either already exited or can't exit until the end */ |
2226 | if (task->flags & PF_EXITING) | |
ea84753c | 2227 | goto next; |
134d3373 | 2228 | |
eaf797ab TH |
2229 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2230 | if (list_empty(&task->cg_list)) | |
ea84753c | 2231 | goto next; |
cd3d0952 | 2232 | |
b3dc094e | 2233 | cset = task_css_set(task); |
1958d2d5 | 2234 | if (!cset->mg_src_cgrp) |
ea84753c | 2235 | goto next; |
b3dc094e | 2236 | |
61d1d219 | 2237 | /* |
1b9aba49 TH |
2238 | * cgroup_taskset_first() must always return the leader. |
2239 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2240 | */ |
1b9aba49 TH |
2241 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2242 | if (list_empty(&cset->mg_node)) | |
2243 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2244 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2245 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2246 | &tset.dst_csets); | |
ea84753c | 2247 | next: |
081aa458 LZ |
2248 | if (!threadgroup) |
2249 | break; | |
9db8de37 | 2250 | } while_each_thread(leader, task); |
fb5d2b4c | 2251 | rcu_read_unlock(); |
b3dc094e | 2252 | up_write(&css_set_rwsem); |
74a1166d | 2253 | |
134d3373 | 2254 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2255 | if (list_empty(&tset.src_csets)) |
2256 | return 0; | |
134d3373 | 2257 | |
1958d2d5 | 2258 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2259 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2260 | if (css->ss->can_attach) { |
9db8de37 TH |
2261 | ret = css->ss->can_attach(css, &tset); |
2262 | if (ret) { | |
1c6727af | 2263 | failed_css = css; |
74a1166d BB |
2264 | goto out_cancel_attach; |
2265 | } | |
2266 | } | |
74a1166d BB |
2267 | } |
2268 | ||
2269 | /* | |
1958d2d5 TH |
2270 | * Now that we're guaranteed success, proceed to move all tasks to |
2271 | * the new cgroup. There are no failure cases after here, so this | |
2272 | * is the commit point. | |
74a1166d | 2273 | */ |
cb0f1fe9 | 2274 | down_write(&css_set_rwsem); |
b3dc094e TH |
2275 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2276 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2277 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2278 | cset->mg_dst_cset); | |
74a1166d | 2279 | } |
cb0f1fe9 | 2280 | up_write(&css_set_rwsem); |
74a1166d BB |
2281 | |
2282 | /* | |
1958d2d5 TH |
2283 | * Migration is committed, all target tasks are now on dst_csets. |
2284 | * Nothing is sensitive to fork() after this point. Notify | |
2285 | * controllers that migration is complete. | |
74a1166d | 2286 | */ |
1958d2d5 | 2287 | tset.csets = &tset.dst_csets; |
74a1166d | 2288 | |
aec3dfcb | 2289 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2290 | if (css->ss->attach) |
2291 | css->ss->attach(css, &tset); | |
74a1166d | 2292 | |
9db8de37 | 2293 | ret = 0; |
b3dc094e TH |
2294 | goto out_release_tset; |
2295 | ||
74a1166d | 2296 | out_cancel_attach: |
aec3dfcb | 2297 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2298 | if (css == failed_css) |
2299 | break; | |
2300 | if (css->ss->cancel_attach) | |
2301 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2302 | } |
b3dc094e TH |
2303 | out_release_tset: |
2304 | down_write(&css_set_rwsem); | |
2305 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2306 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2307 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2308 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2309 | } |
2310 | up_write(&css_set_rwsem); | |
9db8de37 | 2311 | return ret; |
74a1166d BB |
2312 | } |
2313 | ||
1958d2d5 TH |
2314 | /** |
2315 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2316 | * @dst_cgrp: the cgroup to attach to | |
2317 | * @leader: the task or the leader of the threadgroup to be attached | |
2318 | * @threadgroup: attach the whole threadgroup? | |
2319 | * | |
0e1d768f | 2320 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2321 | */ |
2322 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2323 | struct task_struct *leader, bool threadgroup) | |
2324 | { | |
2325 | LIST_HEAD(preloaded_csets); | |
2326 | struct task_struct *task; | |
2327 | int ret; | |
2328 | ||
2329 | /* look up all src csets */ | |
2330 | down_read(&css_set_rwsem); | |
2331 | rcu_read_lock(); | |
2332 | task = leader; | |
2333 | do { | |
2334 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2335 | &preloaded_csets); | |
2336 | if (!threadgroup) | |
2337 | break; | |
2338 | } while_each_thread(leader, task); | |
2339 | rcu_read_unlock(); | |
2340 | up_read(&css_set_rwsem); | |
2341 | ||
2342 | /* prepare dst csets and commit */ | |
2343 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2344 | if (!ret) | |
2345 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2346 | ||
2347 | cgroup_migrate_finish(&preloaded_csets); | |
2348 | return ret; | |
74a1166d BB |
2349 | } |
2350 | ||
2351 | /* | |
2352 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2353 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2354 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2355 | */ |
acbef755 TH |
2356 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2357 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2358 | { |
bbcb81d0 | 2359 | struct task_struct *tsk; |
c69e8d9c | 2360 | const struct cred *cred = current_cred(), *tcred; |
e76ecaee | 2361 | struct cgroup *cgrp; |
acbef755 | 2362 | pid_t pid; |
bbcb81d0 PM |
2363 | int ret; |
2364 | ||
acbef755 TH |
2365 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2366 | return -EINVAL; | |
2367 | ||
e76ecaee TH |
2368 | cgrp = cgroup_kn_lock_live(of->kn); |
2369 | if (!cgrp) | |
74a1166d BB |
2370 | return -ENODEV; |
2371 | ||
b78949eb MSB |
2372 | retry_find_task: |
2373 | rcu_read_lock(); | |
bbcb81d0 | 2374 | if (pid) { |
73507f33 | 2375 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2376 | if (!tsk) { |
2377 | rcu_read_unlock(); | |
dd4b0a46 | 2378 | ret = -ESRCH; |
b78949eb | 2379 | goto out_unlock_cgroup; |
bbcb81d0 | 2380 | } |
74a1166d BB |
2381 | /* |
2382 | * even if we're attaching all tasks in the thread group, we | |
2383 | * only need to check permissions on one of them. | |
2384 | */ | |
c69e8d9c | 2385 | tcred = __task_cred(tsk); |
14a590c3 EB |
2386 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2387 | !uid_eq(cred->euid, tcred->uid) && | |
2388 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2389 | rcu_read_unlock(); |
b78949eb MSB |
2390 | ret = -EACCES; |
2391 | goto out_unlock_cgroup; | |
bbcb81d0 | 2392 | } |
b78949eb MSB |
2393 | } else |
2394 | tsk = current; | |
cd3d0952 TH |
2395 | |
2396 | if (threadgroup) | |
b78949eb | 2397 | tsk = tsk->group_leader; |
c4c27fbd MG |
2398 | |
2399 | /* | |
14a40ffc | 2400 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2401 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2402 | * with no rt_runtime allocated. Just say no. | |
2403 | */ | |
14a40ffc | 2404 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2405 | ret = -EINVAL; |
2406 | rcu_read_unlock(); | |
2407 | goto out_unlock_cgroup; | |
2408 | } | |
2409 | ||
b78949eb MSB |
2410 | get_task_struct(tsk); |
2411 | rcu_read_unlock(); | |
2412 | ||
2413 | threadgroup_lock(tsk); | |
2414 | if (threadgroup) { | |
2415 | if (!thread_group_leader(tsk)) { | |
2416 | /* | |
2417 | * a race with de_thread from another thread's exec() | |
2418 | * may strip us of our leadership, if this happens, | |
2419 | * there is no choice but to throw this task away and | |
2420 | * try again; this is | |
2421 | * "double-double-toil-and-trouble-check locking". | |
2422 | */ | |
2423 | threadgroup_unlock(tsk); | |
2424 | put_task_struct(tsk); | |
2425 | goto retry_find_task; | |
2426 | } | |
081aa458 LZ |
2427 | } |
2428 | ||
2429 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2430 | ||
cd3d0952 TH |
2431 | threadgroup_unlock(tsk); |
2432 | ||
bbcb81d0 | 2433 | put_task_struct(tsk); |
b78949eb | 2434 | out_unlock_cgroup: |
e76ecaee | 2435 | cgroup_kn_unlock(of->kn); |
acbef755 | 2436 | return ret ?: nbytes; |
bbcb81d0 PM |
2437 | } |
2438 | ||
7ae1bad9 TH |
2439 | /** |
2440 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2441 | * @from: attach to all cgroups of a given task | |
2442 | * @tsk: the task to be attached | |
2443 | */ | |
2444 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2445 | { | |
3dd06ffa | 2446 | struct cgroup_root *root; |
7ae1bad9 TH |
2447 | int retval = 0; |
2448 | ||
47cfcd09 | 2449 | mutex_lock(&cgroup_mutex); |
985ed670 | 2450 | for_each_root(root) { |
96d365e0 TH |
2451 | struct cgroup *from_cgrp; |
2452 | ||
3dd06ffa | 2453 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2454 | continue; |
2455 | ||
96d365e0 TH |
2456 | down_read(&css_set_rwsem); |
2457 | from_cgrp = task_cgroup_from_root(from, root); | |
2458 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2459 | |
6f4b7e63 | 2460 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2461 | if (retval) |
2462 | break; | |
2463 | } | |
47cfcd09 | 2464 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2465 | |
2466 | return retval; | |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2469 | ||
acbef755 TH |
2470 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2471 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2472 | { |
acbef755 | 2473 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2474 | } |
2475 | ||
acbef755 TH |
2476 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2477 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2478 | { |
acbef755 | 2479 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2480 | } |
2481 | ||
451af504 TH |
2482 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2483 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2484 | { |
e76ecaee | 2485 | struct cgroup *cgrp; |
5f469907 | 2486 | |
e76ecaee | 2487 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2488 | |
e76ecaee TH |
2489 | cgrp = cgroup_kn_lock_live(of->kn); |
2490 | if (!cgrp) | |
e788e066 | 2491 | return -ENODEV; |
69e943b7 | 2492 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2493 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2494 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2495 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2496 | cgroup_kn_unlock(of->kn); |
451af504 | 2497 | return nbytes; |
e788e066 PM |
2498 | } |
2499 | ||
2da8ca82 | 2500 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2501 | { |
2da8ca82 | 2502 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2503 | |
46cfeb04 | 2504 | spin_lock(&release_agent_path_lock); |
e788e066 | 2505 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2506 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2507 | seq_putc(seq, '\n'); |
e788e066 PM |
2508 | return 0; |
2509 | } | |
2510 | ||
2da8ca82 | 2511 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2512 | { |
c1d5d42e | 2513 | seq_puts(seq, "0\n"); |
e788e066 PM |
2514 | return 0; |
2515 | } | |
2516 | ||
f8f22e53 | 2517 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
355e0c48 | 2518 | { |
f8f22e53 TH |
2519 | struct cgroup_subsys *ss; |
2520 | bool printed = false; | |
2521 | int ssid; | |
a742c59d | 2522 | |
f8f22e53 TH |
2523 | for_each_subsys(ss, ssid) { |
2524 | if (ss_mask & (1 << ssid)) { | |
2525 | if (printed) | |
2526 | seq_putc(seq, ' '); | |
2527 | seq_printf(seq, "%s", ss->name); | |
2528 | printed = true; | |
2529 | } | |
e73d2c61 | 2530 | } |
f8f22e53 TH |
2531 | if (printed) |
2532 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2533 | } |
2534 | ||
f8f22e53 TH |
2535 | /* show controllers which are currently attached to the default hierarchy */ |
2536 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2537 | { |
f8f22e53 TH |
2538 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2539 | ||
5533e011 TH |
2540 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2541 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2542 | return 0; |
db3b1497 PM |
2543 | } |
2544 | ||
f8f22e53 TH |
2545 | /* show controllers which are enabled from the parent */ |
2546 | static int cgroup_controllers_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, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2551 | return 0; |
ddbcc7e8 PM |
2552 | } |
2553 | ||
f8f22e53 TH |
2554 | /* show controllers which are enabled for a given cgroup's children */ |
2555 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2556 | { |
f8f22e53 TH |
2557 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2558 | ||
667c2491 | 2559 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2560 | return 0; |
2561 | } | |
2562 | ||
2563 | /** | |
2564 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2565 | * @cgrp: root of the subtree to update csses for | |
2566 | * | |
2567 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2568 | * css associations need to be updated accordingly. This function looks up | |
2569 | * all css_sets which are attached to the subtree, creates the matching | |
2570 | * updated css_sets and migrates the tasks to the new ones. | |
2571 | */ | |
2572 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2573 | { | |
2574 | LIST_HEAD(preloaded_csets); | |
2575 | struct cgroup_subsys_state *css; | |
2576 | struct css_set *src_cset; | |
2577 | int ret; | |
2578 | ||
f8f22e53 TH |
2579 | lockdep_assert_held(&cgroup_mutex); |
2580 | ||
2581 | /* look up all csses currently attached to @cgrp's subtree */ | |
2582 | down_read(&css_set_rwsem); | |
2583 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2584 | struct cgrp_cset_link *link; | |
2585 | ||
2586 | /* self is not affected by child_subsys_mask change */ | |
2587 | if (css->cgroup == cgrp) | |
2588 | continue; | |
2589 | ||
2590 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2591 | cgroup_migrate_add_src(link->cset, cgrp, | |
2592 | &preloaded_csets); | |
2593 | } | |
2594 | up_read(&css_set_rwsem); | |
2595 | ||
2596 | /* NULL dst indicates self on default hierarchy */ | |
2597 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2598 | if (ret) | |
2599 | goto out_finish; | |
2600 | ||
2601 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2602 | struct task_struct *last_task = NULL, *task; | |
2603 | ||
2604 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2605 | if (!src_cset->mg_src_cgrp) | |
2606 | break; | |
2607 | ||
2608 | /* | |
2609 | * All tasks in src_cset need to be migrated to the | |
2610 | * matching dst_cset. Empty it process by process. We | |
2611 | * walk tasks but migrate processes. The leader might even | |
2612 | * belong to a different cset but such src_cset would also | |
2613 | * be among the target src_csets because the default | |
2614 | * hierarchy enforces per-process membership. | |
2615 | */ | |
2616 | while (true) { | |
2617 | down_read(&css_set_rwsem); | |
2618 | task = list_first_entry_or_null(&src_cset->tasks, | |
2619 | struct task_struct, cg_list); | |
2620 | if (task) { | |
2621 | task = task->group_leader; | |
2622 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2623 | get_task_struct(task); | |
2624 | } | |
2625 | up_read(&css_set_rwsem); | |
2626 | ||
2627 | if (!task) | |
2628 | break; | |
2629 | ||
2630 | /* guard against possible infinite loop */ | |
2631 | if (WARN(last_task == task, | |
2632 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2633 | goto out_finish; | |
2634 | last_task = task; | |
2635 | ||
2636 | threadgroup_lock(task); | |
2637 | /* raced against de_thread() from another thread? */ | |
2638 | if (!thread_group_leader(task)) { | |
2639 | threadgroup_unlock(task); | |
2640 | put_task_struct(task); | |
2641 | continue; | |
2642 | } | |
2643 | ||
2644 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2645 | ||
2646 | threadgroup_unlock(task); | |
2647 | put_task_struct(task); | |
2648 | ||
2649 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2650 | goto out_finish; | |
2651 | } | |
2652 | } | |
2653 | ||
2654 | out_finish: | |
2655 | cgroup_migrate_finish(&preloaded_csets); | |
2656 | return ret; | |
2657 | } | |
2658 | ||
2659 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2660 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2661 | char *buf, size_t nbytes, | |
2662 | loff_t off) | |
f8f22e53 | 2663 | { |
7d331fa9 | 2664 | unsigned int enable = 0, disable = 0; |
f63070d3 | 2665 | unsigned int css_enable, css_disable, old_ctrl, new_ctrl; |
a9746d8d | 2666 | struct cgroup *cgrp, *child; |
f8f22e53 | 2667 | struct cgroup_subsys *ss; |
451af504 | 2668 | char *tok; |
f8f22e53 TH |
2669 | int ssid, ret; |
2670 | ||
2671 | /* | |
d37167ab TH |
2672 | * Parse input - space separated list of subsystem names prefixed |
2673 | * with either + or -. | |
f8f22e53 | 2674 | */ |
451af504 TH |
2675 | buf = strstrip(buf); |
2676 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2677 | if (tok[0] == '\0') |
2678 | continue; | |
f8f22e53 | 2679 | for_each_subsys(ss, ssid) { |
5533e011 TH |
2680 | if (ss->disabled || strcmp(tok + 1, ss->name) || |
2681 | ((1 << ss->id) & cgrp_dfl_root_inhibit_ss_mask)) | |
f8f22e53 TH |
2682 | continue; |
2683 | ||
2684 | if (*tok == '+') { | |
7d331fa9 TH |
2685 | enable |= 1 << ssid; |
2686 | disable &= ~(1 << ssid); | |
f8f22e53 | 2687 | } else if (*tok == '-') { |
7d331fa9 TH |
2688 | disable |= 1 << ssid; |
2689 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2690 | } else { |
2691 | return -EINVAL; | |
2692 | } | |
2693 | break; | |
2694 | } | |
2695 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2696 | return -EINVAL; | |
2697 | } | |
2698 | ||
a9746d8d TH |
2699 | cgrp = cgroup_kn_lock_live(of->kn); |
2700 | if (!cgrp) | |
2701 | return -ENODEV; | |
f8f22e53 TH |
2702 | |
2703 | for_each_subsys(ss, ssid) { | |
2704 | if (enable & (1 << ssid)) { | |
667c2491 | 2705 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2706 | enable &= ~(1 << ssid); |
2707 | continue; | |
2708 | } | |
2709 | ||
c29adf24 TH |
2710 | /* unavailable or not enabled on the parent? */ |
2711 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2712 | (cgroup_parent(cgrp) && | |
667c2491 | 2713 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
2714 | ret = -ENOENT; |
2715 | goto out_unlock; | |
2716 | } | |
2717 | ||
f63070d3 TH |
2718 | /* |
2719 | * @ss is already enabled through dependency and | |
2720 | * we'll just make it visible. Skip draining. | |
2721 | */ | |
2722 | if (cgrp->child_subsys_mask & (1 << ssid)) | |
2723 | continue; | |
2724 | ||
f8f22e53 TH |
2725 | /* |
2726 | * Because css offlining is asynchronous, userland | |
2727 | * might try to re-enable the same controller while | |
2728 | * the previous instance is still around. In such | |
2729 | * cases, wait till it's gone using offline_waitq. | |
2730 | */ | |
2731 | cgroup_for_each_live_child(child, cgrp) { | |
0cee8b77 | 2732 | DEFINE_WAIT(wait); |
f8f22e53 TH |
2733 | |
2734 | if (!cgroup_css(child, ss)) | |
2735 | continue; | |
2736 | ||
0cee8b77 | 2737 | cgroup_get(child); |
f8f22e53 TH |
2738 | prepare_to_wait(&child->offline_waitq, &wait, |
2739 | TASK_UNINTERRUPTIBLE); | |
a9746d8d | 2740 | cgroup_kn_unlock(of->kn); |
f8f22e53 TH |
2741 | schedule(); |
2742 | finish_wait(&child->offline_waitq, &wait); | |
0cee8b77 | 2743 | cgroup_put(child); |
7d331fa9 | 2744 | |
a9746d8d | 2745 | return restart_syscall(); |
f8f22e53 | 2746 | } |
f8f22e53 | 2747 | } else if (disable & (1 << ssid)) { |
667c2491 | 2748 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2749 | disable &= ~(1 << ssid); |
2750 | continue; | |
2751 | } | |
2752 | ||
2753 | /* a child has it enabled? */ | |
2754 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2755 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2756 | ret = -EBUSY; |
ddab2b6e | 2757 | goto out_unlock; |
f8f22e53 TH |
2758 | } |
2759 | } | |
2760 | } | |
2761 | } | |
2762 | ||
2763 | if (!enable && !disable) { | |
2764 | ret = 0; | |
ddab2b6e | 2765 | goto out_unlock; |
f8f22e53 TH |
2766 | } |
2767 | ||
2768 | /* | |
667c2491 | 2769 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2770 | * with tasks so that child cgroups don't compete against tasks. |
2771 | */ | |
d51f39b0 | 2772 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
2773 | ret = -EBUSY; |
2774 | goto out_unlock; | |
2775 | } | |
2776 | ||
2777 | /* | |
f63070d3 TH |
2778 | * Update subsys masks and calculate what needs to be done. More |
2779 | * subsystems than specified may need to be enabled or disabled | |
2780 | * depending on subsystem dependencies. | |
2781 | */ | |
667c2491 TH |
2782 | cgrp->subtree_control |= enable; |
2783 | cgrp->subtree_control &= ~disable; | |
f63070d3 TH |
2784 | |
2785 | old_ctrl = cgrp->child_subsys_mask; | |
667c2491 | 2786 | cgroup_refresh_child_subsys_mask(cgrp); |
f63070d3 TH |
2787 | new_ctrl = cgrp->child_subsys_mask; |
2788 | ||
2789 | css_enable = ~old_ctrl & new_ctrl; | |
2790 | css_disable = old_ctrl & ~new_ctrl; | |
2791 | enable |= css_enable; | |
2792 | disable |= css_disable; | |
c29adf24 | 2793 | |
f63070d3 TH |
2794 | /* |
2795 | * Create new csses or make the existing ones visible. A css is | |
2796 | * created invisible if it's being implicitly enabled through | |
2797 | * dependency. An invisible css is made visible when the userland | |
2798 | * explicitly enables it. | |
f8f22e53 TH |
2799 | */ |
2800 | for_each_subsys(ss, ssid) { | |
2801 | if (!(enable & (1 << ssid))) | |
2802 | continue; | |
2803 | ||
2804 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
2805 | if (css_enable & (1 << ssid)) |
2806 | ret = create_css(child, ss, | |
2807 | cgrp->subtree_control & (1 << ssid)); | |
2808 | else | |
2809 | ret = cgroup_populate_dir(child, 1 << ssid); | |
f8f22e53 TH |
2810 | if (ret) |
2811 | goto err_undo_css; | |
2812 | } | |
2813 | } | |
2814 | ||
c29adf24 TH |
2815 | /* |
2816 | * At this point, cgroup_e_css() results reflect the new csses | |
2817 | * making the following cgroup_update_dfl_csses() properly update | |
2818 | * css associations of all tasks in the subtree. | |
2819 | */ | |
f8f22e53 TH |
2820 | ret = cgroup_update_dfl_csses(cgrp); |
2821 | if (ret) | |
2822 | goto err_undo_css; | |
2823 | ||
f63070d3 TH |
2824 | /* |
2825 | * All tasks are migrated out of disabled csses. Kill or hide | |
2826 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
2827 | * disabled while other subsystems are still depending on it. The |
2828 | * css must not actively control resources and be in the vanilla | |
2829 | * state if it's made visible again later. Controllers which may | |
2830 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 2831 | */ |
f8f22e53 TH |
2832 | for_each_subsys(ss, ssid) { |
2833 | if (!(disable & (1 << ssid))) | |
2834 | continue; | |
2835 | ||
f63070d3 | 2836 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
2837 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
2838 | ||
2839 | if (css_disable & (1 << ssid)) { | |
2840 | kill_css(css); | |
2841 | } else { | |
f63070d3 | 2842 | cgroup_clear_dir(child, 1 << ssid); |
b4536f0c TH |
2843 | if (ss->css_reset) |
2844 | ss->css_reset(css); | |
2845 | } | |
f63070d3 | 2846 | } |
f8f22e53 TH |
2847 | } |
2848 | ||
2849 | kernfs_activate(cgrp->kn); | |
2850 | ret = 0; | |
2851 | out_unlock: | |
a9746d8d | 2852 | cgroup_kn_unlock(of->kn); |
451af504 | 2853 | return ret ?: nbytes; |
f8f22e53 TH |
2854 | |
2855 | err_undo_css: | |
667c2491 TH |
2856 | cgrp->subtree_control &= ~enable; |
2857 | cgrp->subtree_control |= disable; | |
2858 | cgroup_refresh_child_subsys_mask(cgrp); | |
f8f22e53 TH |
2859 | |
2860 | for_each_subsys(ss, ssid) { | |
2861 | if (!(enable & (1 << ssid))) | |
2862 | continue; | |
2863 | ||
2864 | cgroup_for_each_live_child(child, cgrp) { | |
2865 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
2866 | |
2867 | if (!css) | |
2868 | continue; | |
2869 | ||
2870 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 2871 | kill_css(css); |
f63070d3 TH |
2872 | else |
2873 | cgroup_clear_dir(child, 1 << ssid); | |
f8f22e53 TH |
2874 | } |
2875 | } | |
2876 | goto out_unlock; | |
2877 | } | |
2878 | ||
842b597e TH |
2879 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2880 | { | |
2881 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2882 | return 0; | |
2883 | } | |
2884 | ||
2bd59d48 TH |
2885 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2886 | size_t nbytes, loff_t off) | |
355e0c48 | 2887 | { |
2bd59d48 TH |
2888 | struct cgroup *cgrp = of->kn->parent->priv; |
2889 | struct cftype *cft = of->kn->priv; | |
2890 | struct cgroup_subsys_state *css; | |
a742c59d | 2891 | int ret; |
355e0c48 | 2892 | |
b4168640 TH |
2893 | if (cft->write) |
2894 | return cft->write(of, buf, nbytes, off); | |
2895 | ||
2bd59d48 TH |
2896 | /* |
2897 | * kernfs guarantees that a file isn't deleted with operations in | |
2898 | * flight, which means that the matching css is and stays alive and | |
2899 | * doesn't need to be pinned. The RCU locking is not necessary | |
2900 | * either. It's just for the convenience of using cgroup_css(). | |
2901 | */ | |
2902 | rcu_read_lock(); | |
2903 | css = cgroup_css(cgrp, cft->ss); | |
2904 | rcu_read_unlock(); | |
a742c59d | 2905 | |
451af504 | 2906 | if (cft->write_u64) { |
a742c59d TH |
2907 | unsigned long long v; |
2908 | ret = kstrtoull(buf, 0, &v); | |
2909 | if (!ret) | |
2910 | ret = cft->write_u64(css, cft, v); | |
2911 | } else if (cft->write_s64) { | |
2912 | long long v; | |
2913 | ret = kstrtoll(buf, 0, &v); | |
2914 | if (!ret) | |
2915 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2916 | } else { |
a742c59d | 2917 | ret = -EINVAL; |
e73d2c61 | 2918 | } |
2bd59d48 | 2919 | |
a742c59d | 2920 | return ret ?: nbytes; |
355e0c48 PM |
2921 | } |
2922 | ||
6612f05b | 2923 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2924 | { |
2bd59d48 | 2925 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2926 | } |
2927 | ||
6612f05b | 2928 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2929 | { |
2bd59d48 | 2930 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2931 | } |
2932 | ||
6612f05b | 2933 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2934 | { |
2bd59d48 | 2935 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2936 | } |
2937 | ||
91796569 | 2938 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2939 | { |
7da11279 TH |
2940 | struct cftype *cft = seq_cft(m); |
2941 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2942 | |
2da8ca82 TH |
2943 | if (cft->seq_show) |
2944 | return cft->seq_show(m, arg); | |
e73d2c61 | 2945 | |
f4c753b7 | 2946 | if (cft->read_u64) |
896f5199 TH |
2947 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2948 | else if (cft->read_s64) | |
2949 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2950 | else | |
2951 | return -EINVAL; | |
2952 | return 0; | |
91796569 PM |
2953 | } |
2954 | ||
2bd59d48 TH |
2955 | static struct kernfs_ops cgroup_kf_single_ops = { |
2956 | .atomic_write_len = PAGE_SIZE, | |
2957 | .write = cgroup_file_write, | |
2958 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2959 | }; |
2960 | ||
2bd59d48 TH |
2961 | static struct kernfs_ops cgroup_kf_ops = { |
2962 | .atomic_write_len = PAGE_SIZE, | |
2963 | .write = cgroup_file_write, | |
2964 | .seq_start = cgroup_seqfile_start, | |
2965 | .seq_next = cgroup_seqfile_next, | |
2966 | .seq_stop = cgroup_seqfile_stop, | |
2967 | .seq_show = cgroup_seqfile_show, | |
2968 | }; | |
ddbcc7e8 PM |
2969 | |
2970 | /* | |
2971 | * cgroup_rename - Only allow simple rename of directories in place. | |
2972 | */ | |
2bd59d48 TH |
2973 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2974 | const char *new_name_str) | |
ddbcc7e8 | 2975 | { |
2bd59d48 | 2976 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2977 | int ret; |
65dff759 | 2978 | |
2bd59d48 | 2979 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2980 | return -ENOTDIR; |
2bd59d48 | 2981 | if (kn->parent != new_parent) |
ddbcc7e8 | 2982 | return -EIO; |
65dff759 | 2983 | |
6db8e85c TH |
2984 | /* |
2985 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 2986 | * limited. Disallow on the default hierarchy. |
6db8e85c | 2987 | */ |
aa6ec29b | 2988 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 2989 | return -EPERM; |
099fca32 | 2990 | |
e1b2dc17 | 2991 | /* |
8353da1f | 2992 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 2993 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 2994 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
2995 | */ |
2996 | kernfs_break_active_protection(new_parent); | |
2997 | kernfs_break_active_protection(kn); | |
099fca32 | 2998 | |
2bd59d48 | 2999 | mutex_lock(&cgroup_mutex); |
099fca32 | 3000 | |
2bd59d48 | 3001 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3002 | |
2bd59d48 | 3003 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3004 | |
3005 | kernfs_unbreak_active_protection(kn); | |
3006 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3007 | return ret; |
099fca32 LZ |
3008 | } |
3009 | ||
49957f8e TH |
3010 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3011 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3012 | { | |
3013 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3014 | .ia_uid = current_fsuid(), | |
3015 | .ia_gid = current_fsgid(), }; | |
3016 | ||
3017 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3018 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3019 | return 0; | |
3020 | ||
3021 | return kernfs_setattr(kn, &iattr); | |
3022 | } | |
3023 | ||
2bb566cb | 3024 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 3025 | { |
8d7e6fb0 | 3026 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3027 | struct kernfs_node *kn; |
3028 | struct lock_class_key *key = NULL; | |
49957f8e | 3029 | int ret; |
05ef1d7c | 3030 | |
2bd59d48 TH |
3031 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3032 | key = &cft->lockdep_key; | |
3033 | #endif | |
3034 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3035 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
3036 | NULL, false, key); | |
49957f8e TH |
3037 | if (IS_ERR(kn)) |
3038 | return PTR_ERR(kn); | |
3039 | ||
3040 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3041 | if (ret) { |
49957f8e | 3042 | kernfs_remove(kn); |
f8f22e53 TH |
3043 | return ret; |
3044 | } | |
3045 | ||
b7fc5ad2 | 3046 | if (cft->seq_show == cgroup_populated_show) |
842b597e | 3047 | cgrp->populated_kn = kn; |
f8f22e53 | 3048 | return 0; |
ddbcc7e8 PM |
3049 | } |
3050 | ||
b1f28d31 TH |
3051 | /** |
3052 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
3053 | * @cgrp: the target cgroup | |
b1f28d31 TH |
3054 | * @cfts: array of cftypes to be added |
3055 | * @is_add: whether to add or remove | |
3056 | * | |
3057 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
3058 | * For removals, this function never fails. If addition fails, this |
3059 | * function doesn't remove files already added. The caller is responsible | |
3060 | * for cleaning up. | |
b1f28d31 | 3061 | */ |
2bb566cb TH |
3062 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
3063 | bool is_add) | |
ddbcc7e8 | 3064 | { |
03b1cde6 | 3065 | struct cftype *cft; |
b1f28d31 TH |
3066 | int ret; |
3067 | ||
01f6474c | 3068 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 TH |
3069 | |
3070 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3071 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3072 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3073 | continue; |
05ebb6e6 | 3074 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3075 | continue; |
d51f39b0 | 3076 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3077 | continue; |
d51f39b0 | 3078 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3079 | continue; |
3080 | ||
2739d3cc | 3081 | if (is_add) { |
2bb566cb | 3082 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 3083 | if (ret) { |
ed3d261b JP |
3084 | pr_warn("%s: failed to add %s, err=%d\n", |
3085 | __func__, cft->name, ret); | |
b1f28d31 TH |
3086 | return ret; |
3087 | } | |
2739d3cc LZ |
3088 | } else { |
3089 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3090 | } |
ddbcc7e8 | 3091 | } |
b1f28d31 | 3092 | return 0; |
ddbcc7e8 PM |
3093 | } |
3094 | ||
21a2d343 | 3095 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3096 | { |
3097 | LIST_HEAD(pending); | |
2bb566cb | 3098 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3099 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3100 | struct cgroup_subsys_state *css; |
9ccece80 | 3101 | int ret = 0; |
8e3f6541 | 3102 | |
01f6474c | 3103 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3104 | |
e8c82d20 | 3105 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3106 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3107 | struct cgroup *cgrp = css->cgroup; |
3108 | ||
e8c82d20 LZ |
3109 | if (cgroup_is_dead(cgrp)) |
3110 | continue; | |
3111 | ||
21a2d343 | 3112 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
3113 | if (ret) |
3114 | break; | |
8e3f6541 | 3115 | } |
21a2d343 TH |
3116 | |
3117 | if (is_add && !ret) | |
3118 | kernfs_activate(root->kn); | |
9ccece80 | 3119 | return ret; |
8e3f6541 TH |
3120 | } |
3121 | ||
2da440a2 | 3122 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3123 | { |
2bb566cb | 3124 | struct cftype *cft; |
8e3f6541 | 3125 | |
2bd59d48 TH |
3126 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3127 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3128 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3129 | kfree(cft->kf_ops); | |
3130 | cft->kf_ops = NULL; | |
2da440a2 | 3131 | cft->ss = NULL; |
a8ddc821 TH |
3132 | |
3133 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3134 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3135 | } |
2da440a2 TH |
3136 | } |
3137 | ||
2bd59d48 | 3138 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3139 | { |
3140 | struct cftype *cft; | |
3141 | ||
2bd59d48 TH |
3142 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3143 | struct kernfs_ops *kf_ops; | |
3144 | ||
0adb0704 TH |
3145 | WARN_ON(cft->ss || cft->kf_ops); |
3146 | ||
2bd59d48 TH |
3147 | if (cft->seq_start) |
3148 | kf_ops = &cgroup_kf_ops; | |
3149 | else | |
3150 | kf_ops = &cgroup_kf_single_ops; | |
3151 | ||
3152 | /* | |
3153 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3154 | * make a copy of kf_ops to set its atomic_write_len. | |
3155 | */ | |
3156 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3157 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3158 | if (!kf_ops) { | |
3159 | cgroup_exit_cftypes(cfts); | |
3160 | return -ENOMEM; | |
3161 | } | |
3162 | kf_ops->atomic_write_len = cft->max_write_len; | |
3163 | } | |
8e3f6541 | 3164 | |
2bd59d48 | 3165 | cft->kf_ops = kf_ops; |
2bb566cb | 3166 | cft->ss = ss; |
2bd59d48 | 3167 | } |
2bb566cb | 3168 | |
2bd59d48 | 3169 | return 0; |
2da440a2 TH |
3170 | } |
3171 | ||
21a2d343 TH |
3172 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3173 | { | |
01f6474c | 3174 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3175 | |
3176 | if (!cfts || !cfts[0].ss) | |
3177 | return -ENOENT; | |
3178 | ||
3179 | list_del(&cfts->node); | |
3180 | cgroup_apply_cftypes(cfts, false); | |
3181 | cgroup_exit_cftypes(cfts); | |
3182 | return 0; | |
8e3f6541 | 3183 | } |
8e3f6541 | 3184 | |
79578621 TH |
3185 | /** |
3186 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3187 | * @cfts: zero-length name terminated array of cftypes |
3188 | * | |
2bb566cb TH |
3189 | * Unregister @cfts. Files described by @cfts are removed from all |
3190 | * existing cgroups and all future cgroups won't have them either. This | |
3191 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3192 | * |
3193 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3194 | * registered. |
79578621 | 3195 | */ |
2bb566cb | 3196 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3197 | { |
21a2d343 | 3198 | int ret; |
79578621 | 3199 | |
01f6474c | 3200 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3201 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3202 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3203 | return ret; |
80b13586 TH |
3204 | } |
3205 | ||
8e3f6541 TH |
3206 | /** |
3207 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3208 | * @ss: target cgroup subsystem | |
3209 | * @cfts: zero-length name terminated array of cftypes | |
3210 | * | |
3211 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3212 | * existing cgroups to which @ss is attached and all future cgroups will | |
3213 | * have them too. This function can be called anytime whether @ss is | |
3214 | * attached or not. | |
3215 | * | |
3216 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3217 | * function currently returns 0 as long as @cfts registration is successful | |
3218 | * even if some file creation attempts on existing cgroups fail. | |
3219 | */ | |
2cf669a5 | 3220 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3221 | { |
9ccece80 | 3222 | int ret; |
8e3f6541 | 3223 | |
c731ae1d LZ |
3224 | if (ss->disabled) |
3225 | return 0; | |
3226 | ||
dc5736ed LZ |
3227 | if (!cfts || cfts[0].name[0] == '\0') |
3228 | return 0; | |
2bb566cb | 3229 | |
2bd59d48 TH |
3230 | ret = cgroup_init_cftypes(ss, cfts); |
3231 | if (ret) | |
3232 | return ret; | |
79578621 | 3233 | |
01f6474c | 3234 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3235 | |
0adb0704 | 3236 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3237 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3238 | if (ret) |
21a2d343 | 3239 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3240 | |
01f6474c | 3241 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3242 | return ret; |
79578621 TH |
3243 | } |
3244 | ||
a8ddc821 TH |
3245 | /** |
3246 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3247 | * @ss: target cgroup subsystem | |
3248 | * @cfts: zero-length name terminated array of cftypes | |
3249 | * | |
3250 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3251 | * the default hierarchy. | |
3252 | */ | |
3253 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3254 | { | |
3255 | struct cftype *cft; | |
3256 | ||
3257 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3258 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3259 | return cgroup_add_cftypes(ss, cfts); |
3260 | } | |
3261 | ||
3262 | /** | |
3263 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3264 | * @ss: target cgroup subsystem | |
3265 | * @cfts: zero-length name terminated array of cftypes | |
3266 | * | |
3267 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3268 | * the legacy hierarchies. | |
3269 | */ | |
2cf669a5 TH |
3270 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3271 | { | |
a8ddc821 TH |
3272 | struct cftype *cft; |
3273 | ||
3274 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3275 | cft->flags |= __CFTYPE_NOT_ON_DFL; |
2cf669a5 TH |
3276 | return cgroup_add_cftypes(ss, cfts); |
3277 | } | |
3278 | ||
a043e3b2 LZ |
3279 | /** |
3280 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3281 | * @cgrp: the cgroup in question | |
3282 | * | |
3283 | * Return the number of tasks in the cgroup. | |
3284 | */ | |
07bc356e | 3285 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3286 | { |
3287 | int count = 0; | |
69d0206c | 3288 | struct cgrp_cset_link *link; |
817929ec | 3289 | |
96d365e0 | 3290 | down_read(&css_set_rwsem); |
69d0206c TH |
3291 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3292 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3293 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3294 | return count; |
3295 | } | |
3296 | ||
53fa5261 | 3297 | /** |
492eb21b | 3298 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3299 | * @pos: the current position (%NULL to initiate traversal) |
3300 | * @parent: css whose children to walk | |
53fa5261 | 3301 | * |
c2931b70 | 3302 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3303 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3304 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3305 | * be returned regardless of their states. | |
3306 | * | |
3307 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3308 | * css which finished ->css_online() is guaranteed to be visible in the | |
3309 | * future iterations and will stay visible until the last reference is put. | |
3310 | * A css which hasn't finished ->css_online() or already finished | |
3311 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3312 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3313 | */ |
c2931b70 TH |
3314 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3315 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3316 | { |
c2931b70 | 3317 | struct cgroup_subsys_state *next; |
53fa5261 | 3318 | |
8353da1f | 3319 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3320 | |
3321 | /* | |
de3f0341 TH |
3322 | * @pos could already have been unlinked from the sibling list. |
3323 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3324 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3325 | * @pos is taken off list, at which time its next pointer is valid, | |
3326 | * and, as releases are serialized, the one pointed to by the next | |
3327 | * pointer is guaranteed to not have started release yet. This | |
3328 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3329 | * critical section, the one pointed to by its next pointer is | |
3330 | * guaranteed to not have finished its RCU grace period even if we | |
3331 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3332 | * |
de3f0341 TH |
3333 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3334 | * dereferenced; however, as each css is given a monotonically | |
3335 | * increasing unique serial number and always appended to the | |
3336 | * sibling list, the next one can be found by walking the parent's | |
3337 | * children until the first css with higher serial number than | |
3338 | * @pos's. While this path can be slower, it happens iff iteration | |
3339 | * races against release and the race window is very small. | |
53fa5261 | 3340 | */ |
3b287a50 | 3341 | if (!pos) { |
c2931b70 TH |
3342 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3343 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3344 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3345 | } else { |
c2931b70 | 3346 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3347 | if (next->serial_nr > pos->serial_nr) |
3348 | break; | |
53fa5261 TH |
3349 | } |
3350 | ||
3b281afb TH |
3351 | /* |
3352 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3353 | * the next sibling. |
3b281afb | 3354 | */ |
c2931b70 TH |
3355 | if (&next->sibling != &parent->children) |
3356 | return next; | |
3b281afb | 3357 | return NULL; |
53fa5261 | 3358 | } |
53fa5261 | 3359 | |
574bd9f7 | 3360 | /** |
492eb21b | 3361 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3362 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3363 | * @root: css whose descendants to walk |
574bd9f7 | 3364 | * |
492eb21b | 3365 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3366 | * to visit for pre-order traversal of @root's descendants. @root is |
3367 | * included in the iteration and the first node to be visited. | |
75501a6d | 3368 | * |
87fb54f1 TH |
3369 | * While this function requires cgroup_mutex or RCU read locking, it |
3370 | * doesn't require the whole traversal to be contained in a single critical | |
3371 | * section. This function will return the correct next descendant as long | |
3372 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3373 | * |
3374 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3375 | * css which finished ->css_online() is guaranteed to be visible in the | |
3376 | * future iterations and will stay visible until the last reference is put. | |
3377 | * A css which hasn't finished ->css_online() or already finished | |
3378 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3379 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3380 | */ |
492eb21b TH |
3381 | struct cgroup_subsys_state * |
3382 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3383 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3384 | { |
492eb21b | 3385 | struct cgroup_subsys_state *next; |
574bd9f7 | 3386 | |
8353da1f | 3387 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3388 | |
bd8815a6 | 3389 | /* if first iteration, visit @root */ |
7805d000 | 3390 | if (!pos) |
bd8815a6 | 3391 | return root; |
574bd9f7 TH |
3392 | |
3393 | /* visit the first child if exists */ | |
492eb21b | 3394 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3395 | if (next) |
3396 | return next; | |
3397 | ||
3398 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3399 | while (pos != root) { |
5c9d535b | 3400 | next = css_next_child(pos, pos->parent); |
75501a6d | 3401 | if (next) |
574bd9f7 | 3402 | return next; |
5c9d535b | 3403 | pos = pos->parent; |
7805d000 | 3404 | } |
574bd9f7 TH |
3405 | |
3406 | return NULL; | |
3407 | } | |
574bd9f7 | 3408 | |
12a9d2fe | 3409 | /** |
492eb21b TH |
3410 | * css_rightmost_descendant - return the rightmost descendant of a css |
3411 | * @pos: css of interest | |
12a9d2fe | 3412 | * |
492eb21b TH |
3413 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3414 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3415 | * subtree of @pos. |
75501a6d | 3416 | * |
87fb54f1 TH |
3417 | * While this function requires cgroup_mutex or RCU read locking, it |
3418 | * doesn't require the whole traversal to be contained in a single critical | |
3419 | * section. This function will return the correct rightmost descendant as | |
3420 | * long as @pos is accessible. | |
12a9d2fe | 3421 | */ |
492eb21b TH |
3422 | struct cgroup_subsys_state * |
3423 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3424 | { |
492eb21b | 3425 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3426 | |
8353da1f | 3427 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3428 | |
3429 | do { | |
3430 | last = pos; | |
3431 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3432 | pos = NULL; | |
492eb21b | 3433 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3434 | pos = tmp; |
3435 | } while (pos); | |
3436 | ||
3437 | return last; | |
3438 | } | |
12a9d2fe | 3439 | |
492eb21b TH |
3440 | static struct cgroup_subsys_state * |
3441 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3442 | { |
492eb21b | 3443 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3444 | |
3445 | do { | |
3446 | last = pos; | |
492eb21b | 3447 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3448 | } while (pos); |
3449 | ||
3450 | return last; | |
3451 | } | |
3452 | ||
3453 | /** | |
492eb21b | 3454 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3455 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3456 | * @root: css whose descendants to walk |
574bd9f7 | 3457 | * |
492eb21b | 3458 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3459 | * to visit for post-order traversal of @root's descendants. @root is |
3460 | * included in the iteration and the last node to be visited. | |
75501a6d | 3461 | * |
87fb54f1 TH |
3462 | * While this function requires cgroup_mutex or RCU read locking, it |
3463 | * doesn't require the whole traversal to be contained in a single critical | |
3464 | * section. This function will return the correct next descendant as long | |
3465 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3466 | * @cgroup. | |
c2931b70 TH |
3467 | * |
3468 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3469 | * css which finished ->css_online() is guaranteed to be visible in the | |
3470 | * future iterations and will stay visible until the last reference is put. | |
3471 | * A css which hasn't finished ->css_online() or already finished | |
3472 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3473 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3474 | */ |
492eb21b TH |
3475 | struct cgroup_subsys_state * |
3476 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3477 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3478 | { |
492eb21b | 3479 | struct cgroup_subsys_state *next; |
574bd9f7 | 3480 | |
8353da1f | 3481 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3482 | |
58b79a91 TH |
3483 | /* if first iteration, visit leftmost descendant which may be @root */ |
3484 | if (!pos) | |
3485 | return css_leftmost_descendant(root); | |
574bd9f7 | 3486 | |
bd8815a6 TH |
3487 | /* if we visited @root, we're done */ |
3488 | if (pos == root) | |
3489 | return NULL; | |
3490 | ||
574bd9f7 | 3491 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3492 | next = css_next_child(pos, pos->parent); |
75501a6d | 3493 | if (next) |
492eb21b | 3494 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3495 | |
3496 | /* no sibling left, visit parent */ | |
5c9d535b | 3497 | return pos->parent; |
574bd9f7 | 3498 | } |
574bd9f7 | 3499 | |
f3d46500 TH |
3500 | /** |
3501 | * css_has_online_children - does a css have online children | |
3502 | * @css: the target css | |
3503 | * | |
3504 | * Returns %true if @css has any online children; otherwise, %false. This | |
3505 | * function can be called from any context but the caller is responsible | |
3506 | * for synchronizing against on/offlining as necessary. | |
3507 | */ | |
3508 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3509 | { |
f3d46500 TH |
3510 | struct cgroup_subsys_state *child; |
3511 | bool ret = false; | |
cbc125ef TH |
3512 | |
3513 | rcu_read_lock(); | |
f3d46500 | 3514 | css_for_each_child(child, css) { |
99bae5f9 | 3515 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3516 | ret = true; |
3517 | break; | |
cbc125ef TH |
3518 | } |
3519 | } | |
3520 | rcu_read_unlock(); | |
f3d46500 | 3521 | return ret; |
574bd9f7 | 3522 | } |
574bd9f7 | 3523 | |
0942eeee | 3524 | /** |
72ec7029 | 3525 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3526 | * @it: the iterator to advance |
3527 | * | |
3528 | * Advance @it to the next css_set to walk. | |
d515876e | 3529 | */ |
72ec7029 | 3530 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3531 | { |
0f0a2b4f | 3532 | struct list_head *l = it->cset_pos; |
d515876e TH |
3533 | struct cgrp_cset_link *link; |
3534 | struct css_set *cset; | |
3535 | ||
3536 | /* Advance to the next non-empty css_set */ | |
3537 | do { | |
3538 | l = l->next; | |
0f0a2b4f TH |
3539 | if (l == it->cset_head) { |
3540 | it->cset_pos = NULL; | |
d515876e TH |
3541 | return; |
3542 | } | |
3ebb2b6e TH |
3543 | |
3544 | if (it->ss) { | |
3545 | cset = container_of(l, struct css_set, | |
3546 | e_cset_node[it->ss->id]); | |
3547 | } else { | |
3548 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3549 | cset = link->cset; | |
3550 | } | |
c7561128 TH |
3551 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3552 | ||
0f0a2b4f | 3553 | it->cset_pos = l; |
c7561128 TH |
3554 | |
3555 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3556 | it->task_pos = cset->tasks.next; |
c7561128 | 3557 | else |
0f0a2b4f TH |
3558 | it->task_pos = cset->mg_tasks.next; |
3559 | ||
3560 | it->tasks_head = &cset->tasks; | |
3561 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3562 | } |
3563 | ||
0942eeee | 3564 | /** |
72ec7029 TH |
3565 | * css_task_iter_start - initiate task iteration |
3566 | * @css: the css to walk tasks of | |
0942eeee TH |
3567 | * @it: the task iterator to use |
3568 | * | |
72ec7029 TH |
3569 | * Initiate iteration through the tasks of @css. The caller can call |
3570 | * css_task_iter_next() to walk through the tasks until the function | |
3571 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3572 | * called. | |
0942eeee TH |
3573 | * |
3574 | * Note that this function acquires a lock which is released when the | |
3575 | * iteration finishes. The caller can't sleep while iteration is in | |
3576 | * progress. | |
3577 | */ | |
72ec7029 TH |
3578 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3579 | struct css_task_iter *it) | |
96d365e0 | 3580 | __acquires(css_set_rwsem) |
817929ec | 3581 | { |
56fde9e0 TH |
3582 | /* no one should try to iterate before mounting cgroups */ |
3583 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3584 | |
96d365e0 | 3585 | down_read(&css_set_rwsem); |
c59cd3d8 | 3586 | |
3ebb2b6e TH |
3587 | it->ss = css->ss; |
3588 | ||
3589 | if (it->ss) | |
3590 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3591 | else | |
3592 | it->cset_pos = &css->cgroup->cset_links; | |
3593 | ||
0f0a2b4f | 3594 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3595 | |
72ec7029 | 3596 | css_advance_task_iter(it); |
817929ec PM |
3597 | } |
3598 | ||
0942eeee | 3599 | /** |
72ec7029 | 3600 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3601 | * @it: the task iterator being iterated |
3602 | * | |
3603 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3604 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3605 | * reaches the end. | |
0942eeee | 3606 | */ |
72ec7029 | 3607 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3608 | { |
3609 | struct task_struct *res; | |
0f0a2b4f | 3610 | struct list_head *l = it->task_pos; |
817929ec PM |
3611 | |
3612 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3613 | if (!it->cset_pos) |
817929ec PM |
3614 | return NULL; |
3615 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3616 | |
3617 | /* | |
3618 | * Advance iterator to find next entry. cset->tasks is consumed | |
3619 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3620 | * next cset. | |
3621 | */ | |
817929ec | 3622 | l = l->next; |
c7561128 | 3623 | |
0f0a2b4f TH |
3624 | if (l == it->tasks_head) |
3625 | l = it->mg_tasks_head->next; | |
c7561128 | 3626 | |
0f0a2b4f | 3627 | if (l == it->mg_tasks_head) |
72ec7029 | 3628 | css_advance_task_iter(it); |
c7561128 | 3629 | else |
0f0a2b4f | 3630 | it->task_pos = l; |
c7561128 | 3631 | |
817929ec PM |
3632 | return res; |
3633 | } | |
3634 | ||
0942eeee | 3635 | /** |
72ec7029 | 3636 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3637 | * @it: the task iterator to finish |
3638 | * | |
72ec7029 | 3639 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3640 | */ |
72ec7029 | 3641 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3642 | __releases(css_set_rwsem) |
31a7df01 | 3643 | { |
96d365e0 | 3644 | up_read(&css_set_rwsem); |
31a7df01 CW |
3645 | } |
3646 | ||
3647 | /** | |
8cc99345 TH |
3648 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3649 | * @to: cgroup to which the tasks will be moved | |
3650 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3651 | * |
eaf797ab TH |
3652 | * Locking rules between cgroup_post_fork() and the migration path |
3653 | * guarantee that, if a task is forking while being migrated, the new child | |
3654 | * is guaranteed to be either visible in the source cgroup after the | |
3655 | * parent's migration is complete or put into the target cgroup. No task | |
3656 | * can slip out of migration through forking. | |
31a7df01 | 3657 | */ |
8cc99345 | 3658 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3659 | { |
952aaa12 TH |
3660 | LIST_HEAD(preloaded_csets); |
3661 | struct cgrp_cset_link *link; | |
72ec7029 | 3662 | struct css_task_iter it; |
e406d1cf | 3663 | struct task_struct *task; |
952aaa12 | 3664 | int ret; |
31a7df01 | 3665 | |
952aaa12 | 3666 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3667 | |
952aaa12 TH |
3668 | /* all tasks in @from are being moved, all csets are source */ |
3669 | down_read(&css_set_rwsem); | |
3670 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3671 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3672 | up_read(&css_set_rwsem); | |
31a7df01 | 3673 | |
952aaa12 TH |
3674 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3675 | if (ret) | |
3676 | goto out_err; | |
8cc99345 | 3677 | |
952aaa12 TH |
3678 | /* |
3679 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3680 | * ->can_attach() fails. | |
3681 | */ | |
e406d1cf | 3682 | do { |
9d800df1 | 3683 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
3684 | task = css_task_iter_next(&it); |
3685 | if (task) | |
3686 | get_task_struct(task); | |
3687 | css_task_iter_end(&it); | |
3688 | ||
3689 | if (task) { | |
952aaa12 | 3690 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3691 | put_task_struct(task); |
3692 | } | |
3693 | } while (task && !ret); | |
952aaa12 TH |
3694 | out_err: |
3695 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3696 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3697 | return ret; |
8cc99345 TH |
3698 | } |
3699 | ||
bbcb81d0 | 3700 | /* |
102a775e | 3701 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3702 | * |
3703 | * Reading this file can return large amounts of data if a cgroup has | |
3704 | * *lots* of attached tasks. So it may need several calls to read(), | |
3705 | * but we cannot guarantee that the information we produce is correct | |
3706 | * unless we produce it entirely atomically. | |
3707 | * | |
bbcb81d0 | 3708 | */ |
bbcb81d0 | 3709 | |
24528255 LZ |
3710 | /* which pidlist file are we talking about? */ |
3711 | enum cgroup_filetype { | |
3712 | CGROUP_FILE_PROCS, | |
3713 | CGROUP_FILE_TASKS, | |
3714 | }; | |
3715 | ||
3716 | /* | |
3717 | * A pidlist is a list of pids that virtually represents the contents of one | |
3718 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3719 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3720 | * to the cgroup. | |
3721 | */ | |
3722 | struct cgroup_pidlist { | |
3723 | /* | |
3724 | * used to find which pidlist is wanted. doesn't change as long as | |
3725 | * this particular list stays in the list. | |
3726 | */ | |
3727 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3728 | /* array of xids */ | |
3729 | pid_t *list; | |
3730 | /* how many elements the above list has */ | |
3731 | int length; | |
24528255 LZ |
3732 | /* each of these stored in a list by its cgroup */ |
3733 | struct list_head links; | |
3734 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3735 | struct cgroup *owner; | |
b1a21367 TH |
3736 | /* for delayed destruction */ |
3737 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3738 | }; |
3739 | ||
d1d9fd33 BB |
3740 | /* |
3741 | * The following two functions "fix" the issue where there are more pids | |
3742 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3743 | * TODO: replace with a kernel-wide solution to this problem | |
3744 | */ | |
3745 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3746 | static void *pidlist_allocate(int count) | |
3747 | { | |
3748 | if (PIDLIST_TOO_LARGE(count)) | |
3749 | return vmalloc(count * sizeof(pid_t)); | |
3750 | else | |
3751 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3752 | } | |
b1a21367 | 3753 | |
d1d9fd33 BB |
3754 | static void pidlist_free(void *p) |
3755 | { | |
3756 | if (is_vmalloc_addr(p)) | |
3757 | vfree(p); | |
3758 | else | |
3759 | kfree(p); | |
3760 | } | |
d1d9fd33 | 3761 | |
b1a21367 TH |
3762 | /* |
3763 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3764 | * should be left afterwards. | |
3765 | */ | |
3766 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3767 | { | |
3768 | struct cgroup_pidlist *l, *tmp_l; | |
3769 | ||
3770 | mutex_lock(&cgrp->pidlist_mutex); | |
3771 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3772 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3773 | mutex_unlock(&cgrp->pidlist_mutex); | |
3774 | ||
3775 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3776 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3777 | } | |
3778 | ||
3779 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3780 | { | |
3781 | struct delayed_work *dwork = to_delayed_work(work); | |
3782 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3783 | destroy_dwork); | |
3784 | struct cgroup_pidlist *tofree = NULL; | |
3785 | ||
3786 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3787 | |
3788 | /* | |
04502365 TH |
3789 | * Destroy iff we didn't get queued again. The state won't change |
3790 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3791 | */ |
04502365 | 3792 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3793 | list_del(&l->links); |
3794 | pidlist_free(l->list); | |
3795 | put_pid_ns(l->key.ns); | |
3796 | tofree = l; | |
3797 | } | |
3798 | ||
b1a21367 TH |
3799 | mutex_unlock(&l->owner->pidlist_mutex); |
3800 | kfree(tofree); | |
3801 | } | |
3802 | ||
bbcb81d0 | 3803 | /* |
102a775e | 3804 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3805 | * Returns the number of unique elements. |
bbcb81d0 | 3806 | */ |
6ee211ad | 3807 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3808 | { |
102a775e | 3809 | int src, dest = 1; |
102a775e BB |
3810 | |
3811 | /* | |
3812 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3813 | * edge cases first; no work needs to be done for either | |
3814 | */ | |
3815 | if (length == 0 || length == 1) | |
3816 | return length; | |
3817 | /* src and dest walk down the list; dest counts unique elements */ | |
3818 | for (src = 1; src < length; src++) { | |
3819 | /* find next unique element */ | |
3820 | while (list[src] == list[src-1]) { | |
3821 | src++; | |
3822 | if (src == length) | |
3823 | goto after; | |
3824 | } | |
3825 | /* dest always points to where the next unique element goes */ | |
3826 | list[dest] = list[src]; | |
3827 | dest++; | |
3828 | } | |
3829 | after: | |
102a775e BB |
3830 | return dest; |
3831 | } | |
3832 | ||
afb2bc14 TH |
3833 | /* |
3834 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3835 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3836 | * different per namespace, each namespace needs differently sorted list, | |
3837 | * making it impossible to use, for example, single rbtree of member tasks | |
3838 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3839 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3840 | * pidlists keyed by cgroup and namespace. | |
3841 | * | |
3842 | * All this extra complexity was caused by the original implementation | |
3843 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
3844 | * want to do away with it. Explicitly scramble sort order if on the |
3845 | * default hierarchy so that no such expectation exists in the new | |
3846 | * interface. | |
afb2bc14 TH |
3847 | * |
3848 | * Scrambling is done by swapping every two consecutive bits, which is | |
3849 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3850 | */ | |
3851 | static pid_t pid_fry(pid_t pid) | |
3852 | { | |
3853 | unsigned a = pid & 0x55555555; | |
3854 | unsigned b = pid & 0xAAAAAAAA; | |
3855 | ||
3856 | return (a << 1) | (b >> 1); | |
3857 | } | |
3858 | ||
3859 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3860 | { | |
aa6ec29b | 3861 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
3862 | return pid_fry(pid); |
3863 | else | |
3864 | return pid; | |
3865 | } | |
3866 | ||
102a775e BB |
3867 | static int cmppid(const void *a, const void *b) |
3868 | { | |
3869 | return *(pid_t *)a - *(pid_t *)b; | |
3870 | } | |
3871 | ||
afb2bc14 TH |
3872 | static int fried_cmppid(const void *a, const void *b) |
3873 | { | |
3874 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3875 | } | |
3876 | ||
e6b81710 TH |
3877 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3878 | enum cgroup_filetype type) | |
3879 | { | |
3880 | struct cgroup_pidlist *l; | |
3881 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3882 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3883 | ||
3884 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3885 | ||
3886 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3887 | if (l->key.type == type && l->key.ns == ns) | |
3888 | return l; | |
3889 | return NULL; | |
3890 | } | |
3891 | ||
72a8cb30 BB |
3892 | /* |
3893 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3894 | * returns with the lock on that pidlist already held, and takes care | |
3895 | * of the use count, or returns NULL with no locks held if we're out of | |
3896 | * memory. | |
3897 | */ | |
e6b81710 TH |
3898 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3899 | enum cgroup_filetype type) | |
72a8cb30 BB |
3900 | { |
3901 | struct cgroup_pidlist *l; | |
b70cc5fd | 3902 | |
e6b81710 TH |
3903 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3904 | ||
3905 | l = cgroup_pidlist_find(cgrp, type); | |
3906 | if (l) | |
3907 | return l; | |
3908 | ||
72a8cb30 | 3909 | /* entry not found; create a new one */ |
f4f4be2b | 3910 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3911 | if (!l) |
72a8cb30 | 3912 | return l; |
e6b81710 | 3913 | |
b1a21367 | 3914 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3915 | l->key.type = type; |
e6b81710 TH |
3916 | /* don't need task_nsproxy() if we're looking at ourself */ |
3917 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3918 | l->owner = cgrp; |
3919 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3920 | return l; |
3921 | } | |
3922 | ||
102a775e BB |
3923 | /* |
3924 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3925 | */ | |
72a8cb30 BB |
3926 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3927 | struct cgroup_pidlist **lp) | |
102a775e BB |
3928 | { |
3929 | pid_t *array; | |
3930 | int length; | |
3931 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3932 | struct css_task_iter it; |
817929ec | 3933 | struct task_struct *tsk; |
102a775e BB |
3934 | struct cgroup_pidlist *l; |
3935 | ||
4bac00d1 TH |
3936 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3937 | ||
102a775e BB |
3938 | /* |
3939 | * If cgroup gets more users after we read count, we won't have | |
3940 | * enough space - tough. This race is indistinguishable to the | |
3941 | * caller from the case that the additional cgroup users didn't | |
3942 | * show up until sometime later on. | |
3943 | */ | |
3944 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3945 | array = pidlist_allocate(length); |
102a775e BB |
3946 | if (!array) |
3947 | return -ENOMEM; | |
3948 | /* now, populate the array */ | |
9d800df1 | 3949 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 3950 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 3951 | if (unlikely(n == length)) |
817929ec | 3952 | break; |
102a775e | 3953 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3954 | if (type == CGROUP_FILE_PROCS) |
3955 | pid = task_tgid_vnr(tsk); | |
3956 | else | |
3957 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3958 | if (pid > 0) /* make sure to only use valid results */ |
3959 | array[n++] = pid; | |
817929ec | 3960 | } |
72ec7029 | 3961 | css_task_iter_end(&it); |
102a775e BB |
3962 | length = n; |
3963 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 3964 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
3965 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
3966 | else | |
3967 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3968 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3969 | length = pidlist_uniq(array, length); |
e6b81710 | 3970 | |
e6b81710 | 3971 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3972 | if (!l) { |
e6b81710 | 3973 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3974 | pidlist_free(array); |
72a8cb30 | 3975 | return -ENOMEM; |
102a775e | 3976 | } |
e6b81710 TH |
3977 | |
3978 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3979 | pidlist_free(l->list); |
102a775e BB |
3980 | l->list = array; |
3981 | l->length = length; | |
72a8cb30 | 3982 | *lp = l; |
102a775e | 3983 | return 0; |
bbcb81d0 PM |
3984 | } |
3985 | ||
846c7bb0 | 3986 | /** |
a043e3b2 | 3987 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3988 | * @stats: cgroupstats to fill information into |
3989 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3990 | * been requested. | |
a043e3b2 LZ |
3991 | * |
3992 | * Build and fill cgroupstats so that taskstats can export it to user | |
3993 | * space. | |
846c7bb0 BS |
3994 | */ |
3995 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3996 | { | |
2bd59d48 | 3997 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3998 | struct cgroup *cgrp; |
72ec7029 | 3999 | struct css_task_iter it; |
846c7bb0 | 4000 | struct task_struct *tsk; |
33d283be | 4001 | |
2bd59d48 TH |
4002 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4003 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4004 | kernfs_type(kn) != KERNFS_DIR) | |
4005 | return -EINVAL; | |
4006 | ||
bad34660 LZ |
4007 | mutex_lock(&cgroup_mutex); |
4008 | ||
846c7bb0 | 4009 | /* |
2bd59d48 | 4010 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4011 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4012 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4013 | */ |
2bd59d48 TH |
4014 | rcu_read_lock(); |
4015 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4016 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4017 | rcu_read_unlock(); |
bad34660 | 4018 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4019 | return -ENOENT; |
4020 | } | |
bad34660 | 4021 | rcu_read_unlock(); |
846c7bb0 | 4022 | |
9d800df1 | 4023 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4024 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4025 | switch (tsk->state) { |
4026 | case TASK_RUNNING: | |
4027 | stats->nr_running++; | |
4028 | break; | |
4029 | case TASK_INTERRUPTIBLE: | |
4030 | stats->nr_sleeping++; | |
4031 | break; | |
4032 | case TASK_UNINTERRUPTIBLE: | |
4033 | stats->nr_uninterruptible++; | |
4034 | break; | |
4035 | case TASK_STOPPED: | |
4036 | stats->nr_stopped++; | |
4037 | break; | |
4038 | default: | |
4039 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4040 | stats->nr_io_wait++; | |
4041 | break; | |
4042 | } | |
4043 | } | |
72ec7029 | 4044 | css_task_iter_end(&it); |
846c7bb0 | 4045 | |
bad34660 | 4046 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4047 | return 0; |
846c7bb0 BS |
4048 | } |
4049 | ||
8f3ff208 | 4050 | |
bbcb81d0 | 4051 | /* |
102a775e | 4052 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4053 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4054 | * in the cgroup->l->list array. |
bbcb81d0 | 4055 | */ |
cc31edce | 4056 | |
102a775e | 4057 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4058 | { |
cc31edce PM |
4059 | /* |
4060 | * Initially we receive a position value that corresponds to | |
4061 | * one more than the last pid shown (or 0 on the first call or | |
4062 | * after a seek to the start). Use a binary-search to find the | |
4063 | * next pid to display, if any | |
4064 | */ | |
2bd59d48 | 4065 | struct kernfs_open_file *of = s->private; |
7da11279 | 4066 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4067 | struct cgroup_pidlist *l; |
7da11279 | 4068 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4069 | int index = 0, pid = *pos; |
4bac00d1 TH |
4070 | int *iter, ret; |
4071 | ||
4072 | mutex_lock(&cgrp->pidlist_mutex); | |
4073 | ||
4074 | /* | |
5d22444f | 4075 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4076 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4077 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4078 | * could already have been destroyed. |
4079 | */ | |
5d22444f TH |
4080 | if (of->priv) |
4081 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4082 | |
4083 | /* | |
4084 | * Either this is the first start() after open or the matching | |
4085 | * pidlist has been destroyed inbetween. Create a new one. | |
4086 | */ | |
5d22444f TH |
4087 | if (!of->priv) { |
4088 | ret = pidlist_array_load(cgrp, type, | |
4089 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4090 | if (ret) |
4091 | return ERR_PTR(ret); | |
4092 | } | |
5d22444f | 4093 | l = of->priv; |
cc31edce | 4094 | |
cc31edce | 4095 | if (pid) { |
102a775e | 4096 | int end = l->length; |
20777766 | 4097 | |
cc31edce PM |
4098 | while (index < end) { |
4099 | int mid = (index + end) / 2; | |
afb2bc14 | 4100 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4101 | index = mid; |
4102 | break; | |
afb2bc14 | 4103 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4104 | index = mid + 1; |
4105 | else | |
4106 | end = mid; | |
4107 | } | |
4108 | } | |
4109 | /* If we're off the end of the array, we're done */ | |
102a775e | 4110 | if (index >= l->length) |
cc31edce PM |
4111 | return NULL; |
4112 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4113 | iter = l->list + index; |
afb2bc14 | 4114 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4115 | return iter; |
4116 | } | |
4117 | ||
102a775e | 4118 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4119 | { |
2bd59d48 | 4120 | struct kernfs_open_file *of = s->private; |
5d22444f | 4121 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4122 | |
5d22444f TH |
4123 | if (l) |
4124 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4125 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4126 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4127 | } |
4128 | ||
102a775e | 4129 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4130 | { |
2bd59d48 | 4131 | struct kernfs_open_file *of = s->private; |
5d22444f | 4132 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4133 | pid_t *p = v; |
4134 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4135 | /* |
4136 | * Advance to the next pid in the array. If this goes off the | |
4137 | * end, we're done | |
4138 | */ | |
4139 | p++; | |
4140 | if (p >= end) { | |
4141 | return NULL; | |
4142 | } else { | |
7da11279 | 4143 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4144 | return p; |
4145 | } | |
4146 | } | |
4147 | ||
102a775e | 4148 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
4149 | { |
4150 | return seq_printf(s, "%d\n", *(int *)v); | |
4151 | } | |
bbcb81d0 | 4152 | |
182446d0 TH |
4153 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4154 | struct cftype *cft) | |
81a6a5cd | 4155 | { |
182446d0 | 4156 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4157 | } |
4158 | ||
182446d0 TH |
4159 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4160 | struct cftype *cft, u64 val) | |
6379c106 | 4161 | { |
182446d0 | 4162 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 4163 | if (val) |
182446d0 | 4164 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4165 | else |
182446d0 | 4166 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4167 | return 0; |
4168 | } | |
4169 | ||
182446d0 TH |
4170 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4171 | struct cftype *cft) | |
97978e6d | 4172 | { |
182446d0 | 4173 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4174 | } |
4175 | ||
182446d0 TH |
4176 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4177 | struct cftype *cft, u64 val) | |
97978e6d DL |
4178 | { |
4179 | if (val) | |
182446d0 | 4180 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4181 | else |
182446d0 | 4182 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4183 | return 0; |
4184 | } | |
4185 | ||
a14c6874 TH |
4186 | /* cgroup core interface files for the default hierarchy */ |
4187 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4188 | { |
d5c56ced | 4189 | .name = "cgroup.procs", |
6612f05b TH |
4190 | .seq_start = cgroup_pidlist_start, |
4191 | .seq_next = cgroup_pidlist_next, | |
4192 | .seq_stop = cgroup_pidlist_stop, | |
4193 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4194 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4195 | .write = cgroup_procs_write, |
74a1166d | 4196 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 4197 | }, |
f8f22e53 TH |
4198 | { |
4199 | .name = "cgroup.controllers", | |
a14c6874 | 4200 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4201 | .seq_show = cgroup_root_controllers_show, |
4202 | }, | |
4203 | { | |
4204 | .name = "cgroup.controllers", | |
a14c6874 | 4205 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4206 | .seq_show = cgroup_controllers_show, |
4207 | }, | |
4208 | { | |
4209 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4210 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4211 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4212 | }, |
842b597e TH |
4213 | { |
4214 | .name = "cgroup.populated", | |
a14c6874 | 4215 | .flags = CFTYPE_NOT_ON_ROOT, |
842b597e TH |
4216 | .seq_show = cgroup_populated_show, |
4217 | }, | |
a14c6874 TH |
4218 | { } /* terminate */ |
4219 | }; | |
d5c56ced | 4220 | |
a14c6874 TH |
4221 | /* cgroup core interface files for the legacy hierarchies */ |
4222 | static struct cftype cgroup_legacy_base_files[] = { | |
4223 | { | |
4224 | .name = "cgroup.procs", | |
4225 | .seq_start = cgroup_pidlist_start, | |
4226 | .seq_next = cgroup_pidlist_next, | |
4227 | .seq_stop = cgroup_pidlist_stop, | |
4228 | .seq_show = cgroup_pidlist_show, | |
4229 | .private = CGROUP_FILE_PROCS, | |
4230 | .write = cgroup_procs_write, | |
4231 | .mode = S_IRUGO | S_IWUSR, | |
4232 | }, | |
4233 | { | |
4234 | .name = "cgroup.clone_children", | |
4235 | .read_u64 = cgroup_clone_children_read, | |
4236 | .write_u64 = cgroup_clone_children_write, | |
4237 | }, | |
4238 | { | |
4239 | .name = "cgroup.sane_behavior", | |
4240 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4241 | .seq_show = cgroup_sane_behavior_show, | |
4242 | }, | |
d5c56ced TH |
4243 | { |
4244 | .name = "tasks", | |
6612f05b TH |
4245 | .seq_start = cgroup_pidlist_start, |
4246 | .seq_next = cgroup_pidlist_next, | |
4247 | .seq_stop = cgroup_pidlist_stop, | |
4248 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4249 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4250 | .write = cgroup_tasks_write, |
d5c56ced TH |
4251 | .mode = S_IRUGO | S_IWUSR, |
4252 | }, | |
4253 | { | |
4254 | .name = "notify_on_release", | |
d5c56ced TH |
4255 | .read_u64 = cgroup_read_notify_on_release, |
4256 | .write_u64 = cgroup_write_notify_on_release, | |
4257 | }, | |
6e6ff25b TH |
4258 | { |
4259 | .name = "release_agent", | |
a14c6874 | 4260 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4261 | .seq_show = cgroup_release_agent_show, |
451af504 | 4262 | .write = cgroup_release_agent_write, |
5f469907 | 4263 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4264 | }, |
db0416b6 | 4265 | { } /* terminate */ |
bbcb81d0 PM |
4266 | }; |
4267 | ||
13af07df | 4268 | /** |
628f7cd4 | 4269 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4270 | * @cgrp: target cgroup |
13af07df | 4271 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4272 | * |
4273 | * On failure, no file is added. | |
13af07df | 4274 | */ |
69dfa00c | 4275 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
ddbcc7e8 | 4276 | { |
ddbcc7e8 | 4277 | struct cgroup_subsys *ss; |
b420ba7d | 4278 | int i, ret = 0; |
bbcb81d0 | 4279 | |
8e3f6541 | 4280 | /* process cftsets of each subsystem */ |
b420ba7d | 4281 | for_each_subsys(ss, i) { |
0adb0704 | 4282 | struct cftype *cfts; |
b420ba7d | 4283 | |
69dfa00c | 4284 | if (!(subsys_mask & (1 << i))) |
13af07df | 4285 | continue; |
8e3f6541 | 4286 | |
0adb0704 TH |
4287 | list_for_each_entry(cfts, &ss->cfts, node) { |
4288 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4289 | if (ret < 0) |
4290 | goto err; | |
4291 | } | |
ddbcc7e8 | 4292 | } |
ddbcc7e8 | 4293 | return 0; |
bee55099 TH |
4294 | err: |
4295 | cgroup_clear_dir(cgrp, subsys_mask); | |
4296 | return ret; | |
ddbcc7e8 PM |
4297 | } |
4298 | ||
0c21ead1 TH |
4299 | /* |
4300 | * css destruction is four-stage process. | |
4301 | * | |
4302 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4303 | * Implemented in kill_css(). | |
4304 | * | |
4305 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4306 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4307 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4308 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4309 | * |
4310 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4311 | * accessors are inside RCU read sections. css_release() schedules the | |
4312 | * RCU callback. | |
4313 | * | |
4314 | * 4. After the grace period, the css can be freed. Implemented in | |
4315 | * css_free_work_fn(). | |
4316 | * | |
4317 | * It is actually hairier because both step 2 and 4 require process context | |
4318 | * and thus involve punting to css->destroy_work adding two additional | |
4319 | * steps to the already complex sequence. | |
4320 | */ | |
35ef10da | 4321 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4322 | { |
4323 | struct cgroup_subsys_state *css = | |
35ef10da | 4324 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4325 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4326 | |
9a1049da TH |
4327 | percpu_ref_exit(&css->refcnt); |
4328 | ||
9d755d33 TH |
4329 | if (css->ss) { |
4330 | /* css free path */ | |
4331 | if (css->parent) | |
4332 | css_put(css->parent); | |
0ae78e0b | 4333 | |
9d755d33 TH |
4334 | css->ss->css_free(css); |
4335 | cgroup_put(cgrp); | |
4336 | } else { | |
4337 | /* cgroup free path */ | |
4338 | atomic_dec(&cgrp->root->nr_cgrps); | |
4339 | cgroup_pidlist_destroy_all(cgrp); | |
4340 | ||
d51f39b0 | 4341 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4342 | /* |
4343 | * We get a ref to the parent, and put the ref when | |
4344 | * this cgroup is being freed, so it's guaranteed | |
4345 | * that the parent won't be destroyed before its | |
4346 | * children. | |
4347 | */ | |
d51f39b0 | 4348 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4349 | kernfs_put(cgrp->kn); |
4350 | kfree(cgrp); | |
4351 | } else { | |
4352 | /* | |
4353 | * This is root cgroup's refcnt reaching zero, | |
4354 | * which indicates that the root should be | |
4355 | * released. | |
4356 | */ | |
4357 | cgroup_destroy_root(cgrp->root); | |
4358 | } | |
4359 | } | |
48ddbe19 TH |
4360 | } |
4361 | ||
0c21ead1 | 4362 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4363 | { |
4364 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4365 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4366 | |
35ef10da | 4367 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4368 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4369 | } |
4370 | ||
25e15d83 | 4371 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4372 | { |
4373 | struct cgroup_subsys_state *css = | |
25e15d83 | 4374 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4375 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4376 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4377 | |
1fed1b2e TH |
4378 | mutex_lock(&cgroup_mutex); |
4379 | ||
de3f0341 | 4380 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4381 | list_del_rcu(&css->sibling); |
4382 | ||
9d755d33 TH |
4383 | if (ss) { |
4384 | /* css release path */ | |
4385 | cgroup_idr_remove(&ss->css_idr, css->id); | |
4386 | } else { | |
4387 | /* cgroup release path */ | |
9d755d33 TH |
4388 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4389 | cgrp->id = -1; | |
4390 | } | |
d3daf28d | 4391 | |
1fed1b2e TH |
4392 | mutex_unlock(&cgroup_mutex); |
4393 | ||
0c21ead1 | 4394 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4395 | } |
4396 | ||
d3daf28d TH |
4397 | static void css_release(struct percpu_ref *ref) |
4398 | { | |
4399 | struct cgroup_subsys_state *css = | |
4400 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4401 | ||
25e15d83 TH |
4402 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4403 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4404 | } |
4405 | ||
ddfcadab TH |
4406 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4407 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4408 | { |
0cb51d71 TH |
4409 | lockdep_assert_held(&cgroup_mutex); |
4410 | ||
ddfcadab TH |
4411 | cgroup_get(cgrp); |
4412 | ||
d5c419b6 | 4413 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4414 | css->cgroup = cgrp; |
72c97e54 | 4415 | css->ss = ss; |
d5c419b6 TH |
4416 | INIT_LIST_HEAD(&css->sibling); |
4417 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4418 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4419 | |
d51f39b0 TH |
4420 | if (cgroup_parent(cgrp)) { |
4421 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4422 | css_get(css->parent); |
ddfcadab | 4423 | } |
48ddbe19 | 4424 | |
ca8bdcaf | 4425 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4426 | } |
4427 | ||
2a4ac633 | 4428 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4429 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4430 | { |
623f926b | 4431 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4432 | int ret = 0; |
4433 | ||
a31f2d3f TH |
4434 | lockdep_assert_held(&cgroup_mutex); |
4435 | ||
92fb9748 | 4436 | if (ss->css_online) |
eb95419b | 4437 | ret = ss->css_online(css); |
ae7f164a | 4438 | if (!ret) { |
eb95419b | 4439 | css->flags |= CSS_ONLINE; |
aec25020 | 4440 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4441 | } |
b1929db4 | 4442 | return ret; |
a31f2d3f TH |
4443 | } |
4444 | ||
2a4ac633 | 4445 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4446 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4447 | { |
623f926b | 4448 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4449 | |
4450 | lockdep_assert_held(&cgroup_mutex); | |
4451 | ||
4452 | if (!(css->flags & CSS_ONLINE)) | |
4453 | return; | |
4454 | ||
d7eeac19 | 4455 | if (ss->css_offline) |
eb95419b | 4456 | ss->css_offline(css); |
a31f2d3f | 4457 | |
eb95419b | 4458 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4459 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4460 | |
4461 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4462 | } |
4463 | ||
c81c925a TH |
4464 | /** |
4465 | * create_css - create a cgroup_subsys_state | |
4466 | * @cgrp: the cgroup new css will be associated with | |
4467 | * @ss: the subsys of new css | |
f63070d3 | 4468 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4469 | * |
4470 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4471 | * css is online and installed in @cgrp with all interface files created if |
4472 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4473 | */ |
f63070d3 TH |
4474 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4475 | bool visible) | |
c81c925a | 4476 | { |
d51f39b0 | 4477 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4478 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4479 | struct cgroup_subsys_state *css; |
4480 | int err; | |
4481 | ||
c81c925a TH |
4482 | lockdep_assert_held(&cgroup_mutex); |
4483 | ||
1fed1b2e | 4484 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4485 | if (IS_ERR(css)) |
4486 | return PTR_ERR(css); | |
4487 | ||
ddfcadab | 4488 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4489 | |
c81c925a TH |
4490 | err = percpu_ref_init(&css->refcnt, css_release); |
4491 | if (err) | |
3eb59ec6 | 4492 | goto err_free_css; |
c81c925a | 4493 | |
15a4c835 TH |
4494 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_NOWAIT); |
4495 | if (err < 0) | |
4496 | goto err_free_percpu_ref; | |
4497 | css->id = err; | |
c81c925a | 4498 | |
f63070d3 TH |
4499 | if (visible) { |
4500 | err = cgroup_populate_dir(cgrp, 1 << ss->id); | |
4501 | if (err) | |
4502 | goto err_free_id; | |
4503 | } | |
15a4c835 TH |
4504 | |
4505 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4506 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4507 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4508 | |
4509 | err = online_css(css); | |
4510 | if (err) | |
1fed1b2e | 4511 | goto err_list_del; |
94419627 | 4512 | |
c81c925a | 4513 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4514 | cgroup_parent(parent)) { |
ed3d261b | 4515 | 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 | 4516 | current->comm, current->pid, ss->name); |
c81c925a | 4517 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4518 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4519 | ss->warned_broken_hierarchy = true; |
4520 | } | |
4521 | ||
4522 | return 0; | |
4523 | ||
1fed1b2e TH |
4524 | err_list_del: |
4525 | list_del_rcu(&css->sibling); | |
32d01dc7 | 4526 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
15a4c835 TH |
4527 | err_free_id: |
4528 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4529 | err_free_percpu_ref: |
9a1049da | 4530 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4531 | err_free_css: |
a2bed820 | 4532 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4533 | return err; |
4534 | } | |
4535 | ||
b3bfd983 TH |
4536 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4537 | umode_t mode) | |
ddbcc7e8 | 4538 | { |
a9746d8d TH |
4539 | struct cgroup *parent, *cgrp; |
4540 | struct cgroup_root *root; | |
ddbcc7e8 | 4541 | struct cgroup_subsys *ss; |
2bd59d48 | 4542 | struct kernfs_node *kn; |
a14c6874 | 4543 | struct cftype *base_files; |
b3bfd983 | 4544 | int ssid, ret; |
ddbcc7e8 | 4545 | |
a9746d8d TH |
4546 | parent = cgroup_kn_lock_live(parent_kn); |
4547 | if (!parent) | |
4548 | return -ENODEV; | |
4549 | root = parent->root; | |
ddbcc7e8 | 4550 | |
0a950f65 | 4551 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc | 4552 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
ba0f4d76 TH |
4553 | if (!cgrp) { |
4554 | ret = -ENOMEM; | |
4555 | goto out_unlock; | |
0ab02ca8 LZ |
4556 | } |
4557 | ||
9d755d33 TH |
4558 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release); |
4559 | if (ret) | |
4560 | goto out_free_cgrp; | |
4561 | ||
0ab02ca8 LZ |
4562 | /* |
4563 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4564 | * a half-baked cgroup. | |
4565 | */ | |
6fa4918d | 4566 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_NOWAIT); |
0ab02ca8 | 4567 | if (cgrp->id < 0) { |
ba0f4d76 | 4568 | ret = -ENOMEM; |
9d755d33 | 4569 | goto out_cancel_ref; |
976c06bc TH |
4570 | } |
4571 | ||
cc31edce | 4572 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4573 | |
9d800df1 | 4574 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4575 | cgrp->root = root; |
ddbcc7e8 | 4576 | |
b6abdb0e LZ |
4577 | if (notify_on_release(parent)) |
4578 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4579 | ||
2260e7fc TH |
4580 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4581 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4582 | |
2bd59d48 | 4583 | /* create the directory */ |
e61734c5 | 4584 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4585 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4586 | ret = PTR_ERR(kn); |
4587 | goto out_free_id; | |
2bd59d48 TH |
4588 | } |
4589 | cgrp->kn = kn; | |
ddbcc7e8 | 4590 | |
4e139afc | 4591 | /* |
6f30558f TH |
4592 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4593 | * that @cgrp->kn is always accessible. | |
4e139afc | 4594 | */ |
6f30558f | 4595 | kernfs_get(kn); |
ddbcc7e8 | 4596 | |
0cb51d71 | 4597 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4598 | |
4e139afc | 4599 | /* allocation complete, commit to creation */ |
d5c419b6 | 4600 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4601 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4602 | cgroup_get(parent); |
415cf07a | 4603 | |
0d80255e TH |
4604 | /* |
4605 | * @cgrp is now fully operational. If something fails after this | |
4606 | * point, it'll be released via the normal destruction path. | |
4607 | */ | |
6fa4918d | 4608 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4609 | |
ba0f4d76 TH |
4610 | ret = cgroup_kn_set_ugid(kn); |
4611 | if (ret) | |
4612 | goto out_destroy; | |
49957f8e | 4613 | |
a14c6874 TH |
4614 | if (cgroup_on_dfl(cgrp)) |
4615 | base_files = cgroup_dfl_base_files; | |
4616 | else | |
4617 | base_files = cgroup_legacy_base_files; | |
4618 | ||
4619 | ret = cgroup_addrm_files(cgrp, base_files, true); | |
ba0f4d76 TH |
4620 | if (ret) |
4621 | goto out_destroy; | |
628f7cd4 | 4622 | |
9d403e99 | 4623 | /* let's create and online css's */ |
b85d2040 | 4624 | for_each_subsys(ss, ssid) { |
f392e51c | 4625 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
4626 | ret = create_css(cgrp, ss, |
4627 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
4628 | if (ret) |
4629 | goto out_destroy; | |
b85d2040 | 4630 | } |
a8638030 | 4631 | } |
ddbcc7e8 | 4632 | |
bd53d617 TH |
4633 | /* |
4634 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4635 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4636 | */ |
667c2491 TH |
4637 | if (!cgroup_on_dfl(cgrp)) { |
4638 | cgrp->subtree_control = parent->subtree_control; | |
4639 | cgroup_refresh_child_subsys_mask(cgrp); | |
4640 | } | |
2bd59d48 | 4641 | |
2bd59d48 | 4642 | kernfs_activate(kn); |
ddbcc7e8 | 4643 | |
ba0f4d76 TH |
4644 | ret = 0; |
4645 | goto out_unlock; | |
ddbcc7e8 | 4646 | |
ba0f4d76 | 4647 | out_free_id: |
6fa4918d | 4648 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 4649 | out_cancel_ref: |
9a1049da | 4650 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 4651 | out_free_cgrp: |
bd89aabc | 4652 | kfree(cgrp); |
ba0f4d76 | 4653 | out_unlock: |
a9746d8d | 4654 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4655 | return ret; |
4b8b47eb | 4656 | |
ba0f4d76 | 4657 | out_destroy: |
4b8b47eb | 4658 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4659 | goto out_unlock; |
ddbcc7e8 PM |
4660 | } |
4661 | ||
223dbc38 TH |
4662 | /* |
4663 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4664 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4665 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4666 | */ |
4667 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4668 | { |
223dbc38 TH |
4669 | struct cgroup_subsys_state *css = |
4670 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4671 | |
f20104de | 4672 | mutex_lock(&cgroup_mutex); |
09a503ea | 4673 | offline_css(css); |
f20104de | 4674 | mutex_unlock(&cgroup_mutex); |
09a503ea | 4675 | |
09a503ea | 4676 | css_put(css); |
d3daf28d TH |
4677 | } |
4678 | ||
223dbc38 TH |
4679 | /* css kill confirmation processing requires process context, bounce */ |
4680 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4681 | { |
4682 | struct cgroup_subsys_state *css = | |
4683 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4684 | ||
223dbc38 | 4685 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4686 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4687 | } |
4688 | ||
f392e51c TH |
4689 | /** |
4690 | * kill_css - destroy a css | |
4691 | * @css: css to destroy | |
4692 | * | |
4693 | * This function initiates destruction of @css by removing cgroup interface | |
4694 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4695 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4696 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4697 | */ |
4698 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4699 | { |
01f6474c | 4700 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4701 | |
2bd59d48 TH |
4702 | /* |
4703 | * This must happen before css is disassociated with its cgroup. | |
4704 | * See seq_css() for details. | |
4705 | */ | |
aec25020 | 4706 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4707 | |
edae0c33 TH |
4708 | /* |
4709 | * Killing would put the base ref, but we need to keep it alive | |
4710 | * until after ->css_offline(). | |
4711 | */ | |
4712 | css_get(css); | |
4713 | ||
4714 | /* | |
4715 | * cgroup core guarantees that, by the time ->css_offline() is | |
4716 | * invoked, no new css reference will be given out via | |
ec903c0c | 4717 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4718 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4719 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4720 | * | |
4721 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4722 | * css is confirmed to be seen as killed on all CPUs. | |
4723 | */ | |
4724 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4725 | } |
4726 | ||
4727 | /** | |
4728 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4729 | * @cgrp: cgroup to be destroyed | |
4730 | * | |
4731 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4732 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4733 | * guarantee that css_tryget_online() won't succeed by the time |
4734 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4735 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4736 | * |
4737 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4738 | * userland visible parts and start killing the percpu refcnts of | |
4739 | * css's. Set up so that the next stage will be kicked off once all | |
4740 | * the percpu refcnts are confirmed to be killed. | |
4741 | * | |
4742 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4743 | * rest of destruction. Once all cgroup references are gone, the | |
4744 | * cgroup is RCU-freed. | |
4745 | * | |
4746 | * This function implements s1. After this step, @cgrp is gone as far as | |
4747 | * the userland is concerned and a new cgroup with the same name may be | |
4748 | * created. As cgroup doesn't care about the names internally, this | |
4749 | * doesn't cause any problem. | |
4750 | */ | |
42809dd4 TH |
4751 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4752 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4753 | { |
2bd59d48 | 4754 | struct cgroup_subsys_state *css; |
ddd69148 | 4755 | bool empty; |
1c6727af | 4756 | int ssid; |
ddbcc7e8 | 4757 | |
42809dd4 TH |
4758 | lockdep_assert_held(&cgroup_mutex); |
4759 | ||
ddd69148 | 4760 | /* |
96d365e0 | 4761 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4762 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4763 | */ |
96d365e0 | 4764 | down_read(&css_set_rwsem); |
bb78a92f | 4765 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4766 | up_read(&css_set_rwsem); |
ddd69148 | 4767 | if (!empty) |
ddbcc7e8 | 4768 | return -EBUSY; |
a043e3b2 | 4769 | |
bb78a92f | 4770 | /* |
d5c419b6 TH |
4771 | * Make sure there's no live children. We can't test emptiness of |
4772 | * ->self.children as dead children linger on it while being | |
4773 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 4774 | */ |
f3d46500 | 4775 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
4776 | return -EBUSY; |
4777 | ||
455050d2 TH |
4778 | /* |
4779 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 4780 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 4781 | */ |
184faf32 | 4782 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 4783 | |
249f3468 | 4784 | /* initiate massacre of all css's */ |
1c6727af TH |
4785 | for_each_css(css, ssid, cgrp) |
4786 | kill_css(css); | |
455050d2 | 4787 | |
184faf32 | 4788 | /* CSS_ONLINE is clear, remove from ->release_list for the last time */ |
455050d2 TH |
4789 | raw_spin_lock(&release_list_lock); |
4790 | if (!list_empty(&cgrp->release_list)) | |
4791 | list_del_init(&cgrp->release_list); | |
4792 | raw_spin_unlock(&release_list_lock); | |
4793 | ||
4794 | /* | |
01f6474c TH |
4795 | * Remove @cgrp directory along with the base files. @cgrp has an |
4796 | * extra ref on its kn. | |
f20104de | 4797 | */ |
01f6474c | 4798 | kernfs_remove(cgrp->kn); |
f20104de | 4799 | |
d51f39b0 TH |
4800 | set_bit(CGRP_RELEASABLE, &cgroup_parent(cgrp)->flags); |
4801 | check_for_release(cgroup_parent(cgrp)); | |
2bd59d48 | 4802 | |
249f3468 | 4803 | /* put the base reference */ |
9d755d33 | 4804 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 4805 | |
ea15f8cc TH |
4806 | return 0; |
4807 | }; | |
4808 | ||
2bd59d48 | 4809 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4810 | { |
a9746d8d | 4811 | struct cgroup *cgrp; |
2bd59d48 | 4812 | int ret = 0; |
42809dd4 | 4813 | |
a9746d8d TH |
4814 | cgrp = cgroup_kn_lock_live(kn); |
4815 | if (!cgrp) | |
4816 | return 0; | |
4817 | cgroup_get(cgrp); /* for @kn->priv clearing */ | |
42809dd4 | 4818 | |
a9746d8d | 4819 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 4820 | |
a9746d8d | 4821 | cgroup_kn_unlock(kn); |
8e3f6541 TH |
4822 | |
4823 | /* | |
cfc79d5b TH |
4824 | * There are two control paths which try to determine cgroup from |
4825 | * dentry without going through kernfs - cgroupstats_build() and | |
4826 | * css_tryget_online_from_dir(). Those are supported by RCU | |
4827 | * protecting clearing of cgrp->kn->priv backpointer, which should | |
4828 | * happen after all files under it have been removed. | |
8e3f6541 | 4829 | */ |
cfc79d5b TH |
4830 | if (!ret) |
4831 | RCU_INIT_POINTER(*(void __rcu __force **)&kn->priv, NULL); | |
2bb566cb | 4832 | |
2bd59d48 | 4833 | cgroup_put(cgrp); |
42809dd4 | 4834 | return ret; |
8e3f6541 TH |
4835 | } |
4836 | ||
2bd59d48 TH |
4837 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4838 | .remount_fs = cgroup_remount, | |
4839 | .show_options = cgroup_show_options, | |
4840 | .mkdir = cgroup_mkdir, | |
4841 | .rmdir = cgroup_rmdir, | |
4842 | .rename = cgroup_rename, | |
4843 | }; | |
4844 | ||
15a4c835 | 4845 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4846 | { |
ddbcc7e8 | 4847 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4848 | |
4849 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4850 | |
648bb56d TH |
4851 | mutex_lock(&cgroup_mutex); |
4852 | ||
15a4c835 | 4853 | idr_init(&ss->css_idr); |
0adb0704 | 4854 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4855 | |
3dd06ffa TH |
4856 | /* Create the root cgroup state for this subsystem */ |
4857 | ss->root = &cgrp_dfl_root; | |
4858 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4859 | /* We don't handle early failures gracefully */ |
4860 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4861 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
4862 | |
4863 | /* | |
4864 | * Root csses are never destroyed and we can't initialize | |
4865 | * percpu_ref during early init. Disable refcnting. | |
4866 | */ | |
4867 | css->flags |= CSS_NO_REF; | |
4868 | ||
15a4c835 | 4869 | if (early) { |
9395a450 | 4870 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
4871 | css->id = 1; |
4872 | } else { | |
4873 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4874 | BUG_ON(css->id < 0); | |
4875 | } | |
ddbcc7e8 | 4876 | |
e8d55fde | 4877 | /* Update the init_css_set to contain a subsys |
817929ec | 4878 | * pointer to this state - since the subsystem is |
e8d55fde | 4879 | * newly registered, all tasks and hence the |
3dd06ffa | 4880 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4881 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4882 | |
4883 | need_forkexit_callback |= ss->fork || ss->exit; | |
4884 | ||
e8d55fde LZ |
4885 | /* At system boot, before all subsystems have been |
4886 | * registered, no tasks have been forked, so we don't | |
4887 | * need to invoke fork callbacks here. */ | |
4888 | BUG_ON(!list_empty(&init_task.tasks)); | |
4889 | ||
ae7f164a | 4890 | BUG_ON(online_css(css)); |
a8638030 | 4891 | |
cf5d5941 BB |
4892 | mutex_unlock(&cgroup_mutex); |
4893 | } | |
cf5d5941 | 4894 | |
ddbcc7e8 | 4895 | /** |
a043e3b2 LZ |
4896 | * cgroup_init_early - cgroup initialization at system boot |
4897 | * | |
4898 | * Initialize cgroups at system boot, and initialize any | |
4899 | * subsystems that request early init. | |
ddbcc7e8 PM |
4900 | */ |
4901 | int __init cgroup_init_early(void) | |
4902 | { | |
7b9a6ba5 | 4903 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 4904 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4905 | int i; |
30159ec7 | 4906 | |
3dd06ffa | 4907 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
4908 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
4909 | ||
a4ea1cc9 | 4910 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4911 | |
3ed80a62 | 4912 | for_each_subsys(ss, i) { |
aec25020 | 4913 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4914 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4915 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4916 | ss->id, ss->name); |
073219e9 TH |
4917 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4918 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4919 | ||
aec25020 | 4920 | ss->id = i; |
073219e9 | 4921 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4922 | |
4923 | if (ss->early_init) | |
15a4c835 | 4924 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
4925 | } |
4926 | return 0; | |
4927 | } | |
4928 | ||
4929 | /** | |
a043e3b2 LZ |
4930 | * cgroup_init - cgroup initialization |
4931 | * | |
4932 | * Register cgroup filesystem and /proc file, and initialize | |
4933 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4934 | */ |
4935 | int __init cgroup_init(void) | |
4936 | { | |
30159ec7 | 4937 | struct cgroup_subsys *ss; |
0ac801fe | 4938 | unsigned long key; |
172a2c06 | 4939 | int ssid, err; |
ddbcc7e8 | 4940 | |
a14c6874 TH |
4941 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
4942 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 4943 | |
54e7b4eb | 4944 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4945 | |
82fe9b0d TH |
4946 | /* Add init_css_set to the hash table */ |
4947 | key = css_set_hash(init_css_set.subsys); | |
4948 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4949 | ||
3dd06ffa | 4950 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4951 | |
54e7b4eb TH |
4952 | mutex_unlock(&cgroup_mutex); |
4953 | ||
172a2c06 | 4954 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
4955 | if (ss->early_init) { |
4956 | struct cgroup_subsys_state *css = | |
4957 | init_css_set.subsys[ss->id]; | |
4958 | ||
4959 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
4960 | GFP_KERNEL); | |
4961 | BUG_ON(css->id < 0); | |
4962 | } else { | |
4963 | cgroup_init_subsys(ss, false); | |
4964 | } | |
172a2c06 | 4965 | |
2d8f243a TH |
4966 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4967 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
4968 | |
4969 | /* | |
c731ae1d LZ |
4970 | * Setting dfl_root subsys_mask needs to consider the |
4971 | * disabled flag and cftype registration needs kmalloc, | |
4972 | * both of which aren't available during early_init. | |
172a2c06 | 4973 | */ |
a8ddc821 TH |
4974 | if (ss->disabled) |
4975 | continue; | |
4976 | ||
4977 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
4978 | ||
4979 | if (cgroup_legacy_files_on_dfl && !ss->dfl_cftypes) | |
4980 | ss->dfl_cftypes = ss->legacy_cftypes; | |
4981 | ||
5de4fa13 TH |
4982 | if (!ss->dfl_cftypes) |
4983 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
4984 | ||
a8ddc821 TH |
4985 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
4986 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
4987 | } else { | |
4988 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
4989 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 4990 | } |
676db4af GK |
4991 | } |
4992 | ||
676db4af | 4993 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4994 | if (!cgroup_kobj) |
4995 | return -ENOMEM; | |
676db4af | 4996 | |
ddbcc7e8 | 4997 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4998 | if (err < 0) { |
4999 | kobject_put(cgroup_kobj); | |
2bd59d48 | 5000 | return err; |
676db4af | 5001 | } |
ddbcc7e8 | 5002 | |
46ae220b | 5003 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 5004 | return 0; |
ddbcc7e8 | 5005 | } |
b4f48b63 | 5006 | |
e5fca243 TH |
5007 | static int __init cgroup_wq_init(void) |
5008 | { | |
5009 | /* | |
5010 | * There isn't much point in executing destruction path in | |
5011 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5012 | * Use 1 for @max_active. |
e5fca243 TH |
5013 | * |
5014 | * We would prefer to do this in cgroup_init() above, but that | |
5015 | * is called before init_workqueues(): so leave this until after. | |
5016 | */ | |
1a11533f | 5017 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5018 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5019 | |
5020 | /* | |
5021 | * Used to destroy pidlists and separate to serve as flush domain. | |
5022 | * Cap @max_active to 1 too. | |
5023 | */ | |
5024 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5025 | 0, 1); | |
5026 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5027 | ||
e5fca243 TH |
5028 | return 0; |
5029 | } | |
5030 | core_initcall(cgroup_wq_init); | |
5031 | ||
a424316c PM |
5032 | /* |
5033 | * proc_cgroup_show() | |
5034 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5035 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
5036 | */ |
5037 | ||
5038 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 5039 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
5040 | { |
5041 | struct pid *pid; | |
5042 | struct task_struct *tsk; | |
e61734c5 | 5043 | char *buf, *path; |
a424316c | 5044 | int retval; |
3dd06ffa | 5045 | struct cgroup_root *root; |
a424316c PM |
5046 | |
5047 | retval = -ENOMEM; | |
e61734c5 | 5048 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5049 | if (!buf) |
5050 | goto out; | |
5051 | ||
5052 | retval = -ESRCH; | |
5053 | pid = m->private; | |
5054 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
5055 | if (!tsk) | |
5056 | goto out_free; | |
5057 | ||
5058 | retval = 0; | |
5059 | ||
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 | ||
5089 | out_unlock: | |
96d365e0 | 5090 | up_read(&css_set_rwsem); |
a424316c PM |
5091 | mutex_unlock(&cgroup_mutex); |
5092 | put_task_struct(tsk); | |
5093 | out_free: | |
5094 | kfree(buf); | |
5095 | out: | |
5096 | return retval; | |
5097 | } | |
5098 | ||
a424316c PM |
5099 | /* Display information about each subsystem and each hierarchy */ |
5100 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5101 | { | |
30159ec7 | 5102 | struct cgroup_subsys *ss; |
a424316c | 5103 | int i; |
a424316c | 5104 | |
8bab8dde | 5105 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5106 | /* |
5107 | * ideally we don't want subsystems moving around while we do this. | |
5108 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5109 | * subsys/hierarchy state. | |
5110 | */ | |
a424316c | 5111 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5112 | |
5113 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
5114 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
5115 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 5116 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 5117 | |
a424316c PM |
5118 | mutex_unlock(&cgroup_mutex); |
5119 | return 0; | |
5120 | } | |
5121 | ||
5122 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5123 | { | |
9dce07f1 | 5124 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5125 | } |
5126 | ||
828c0950 | 5127 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5128 | .open = cgroupstats_open, |
5129 | .read = seq_read, | |
5130 | .llseek = seq_lseek, | |
5131 | .release = single_release, | |
5132 | }; | |
5133 | ||
b4f48b63 | 5134 | /** |
eaf797ab | 5135 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5136 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5137 | * |
eaf797ab TH |
5138 | * A task is associated with the init_css_set until cgroup_post_fork() |
5139 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5140 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5141 | */ |
5142 | void cgroup_fork(struct task_struct *child) | |
5143 | { | |
eaf797ab | 5144 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5145 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5146 | } |
5147 | ||
817929ec | 5148 | /** |
a043e3b2 LZ |
5149 | * cgroup_post_fork - called on a new task after adding it to the task list |
5150 | * @child: the task in question | |
5151 | * | |
5edee61e TH |
5152 | * Adds the task to the list running through its css_set if necessary and |
5153 | * call the subsystem fork() callbacks. Has to be after the task is | |
5154 | * visible on the task list in case we race with the first call to | |
0942eeee | 5155 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5156 | * list. |
a043e3b2 | 5157 | */ |
817929ec PM |
5158 | void cgroup_post_fork(struct task_struct *child) |
5159 | { | |
30159ec7 | 5160 | struct cgroup_subsys *ss; |
5edee61e TH |
5161 | int i; |
5162 | ||
3ce3230a | 5163 | /* |
eaf797ab TH |
5164 | * This may race against cgroup_enable_task_cg_links(). As that |
5165 | * function sets use_task_css_set_links before grabbing | |
5166 | * tasklist_lock and we just went through tasklist_lock to add | |
5167 | * @child, it's guaranteed that either we see the set | |
5168 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5169 | * @child during its iteration. | |
5170 | * | |
5171 | * If we won the race, @child is associated with %current's | |
5172 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
5173 | * association is stable, and, on completion of the parent's | |
5174 | * migration, @child is visible in the source of migration or | |
5175 | * already in the destination cgroup. This guarantee is necessary | |
5176 | * when implementing operations which need to migrate all tasks of | |
5177 | * a cgroup to another. | |
5178 | * | |
5179 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
5180 | * will remain in init_css_set. This is safe because all tasks are | |
5181 | * in the init_css_set before cg_links is enabled and there's no | |
5182 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5183 | */ |
817929ec | 5184 | if (use_task_css_set_links) { |
eaf797ab TH |
5185 | struct css_set *cset; |
5186 | ||
96d365e0 | 5187 | down_write(&css_set_rwsem); |
0e1d768f | 5188 | cset = task_css_set(current); |
eaf797ab TH |
5189 | if (list_empty(&child->cg_list)) { |
5190 | rcu_assign_pointer(child->cgroups, cset); | |
5191 | list_add(&child->cg_list, &cset->tasks); | |
5192 | get_css_set(cset); | |
5193 | } | |
96d365e0 | 5194 | up_write(&css_set_rwsem); |
817929ec | 5195 | } |
5edee61e TH |
5196 | |
5197 | /* | |
5198 | * Call ss->fork(). This must happen after @child is linked on | |
5199 | * css_set; otherwise, @child might change state between ->fork() | |
5200 | * and addition to css_set. | |
5201 | */ | |
5202 | if (need_forkexit_callback) { | |
3ed80a62 | 5203 | for_each_subsys(ss, i) |
5edee61e TH |
5204 | if (ss->fork) |
5205 | ss->fork(child); | |
5edee61e | 5206 | } |
817929ec | 5207 | } |
5edee61e | 5208 | |
b4f48b63 PM |
5209 | /** |
5210 | * cgroup_exit - detach cgroup from exiting task | |
5211 | * @tsk: pointer to task_struct of exiting process | |
5212 | * | |
5213 | * Description: Detach cgroup from @tsk and release it. | |
5214 | * | |
5215 | * Note that cgroups marked notify_on_release force every task in | |
5216 | * them to take the global cgroup_mutex mutex when exiting. | |
5217 | * This could impact scaling on very large systems. Be reluctant to | |
5218 | * use notify_on_release cgroups where very high task exit scaling | |
5219 | * is required on large systems. | |
5220 | * | |
0e1d768f TH |
5221 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5222 | * call cgroup_exit() while the task is still competent to handle | |
5223 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5224 | * each hierarchy for the remainder of its exit. No need to bother with | |
5225 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5226 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5227 | */ |
1ec41830 | 5228 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5229 | { |
30159ec7 | 5230 | struct cgroup_subsys *ss; |
5abb8855 | 5231 | struct css_set *cset; |
eaf797ab | 5232 | bool put_cset = false; |
d41d5a01 | 5233 | int i; |
817929ec PM |
5234 | |
5235 | /* | |
0e1d768f TH |
5236 | * Unlink from @tsk from its css_set. As migration path can't race |
5237 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
5238 | */ |
5239 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 5240 | down_write(&css_set_rwsem); |
0e1d768f | 5241 | list_del_init(&tsk->cg_list); |
96d365e0 | 5242 | up_write(&css_set_rwsem); |
0e1d768f | 5243 | put_cset = true; |
817929ec PM |
5244 | } |
5245 | ||
b4f48b63 | 5246 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
5247 | cset = task_css_set(tsk); |
5248 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 5249 | |
1ec41830 | 5250 | if (need_forkexit_callback) { |
3ed80a62 TH |
5251 | /* see cgroup_post_fork() for details */ |
5252 | for_each_subsys(ss, i) { | |
d41d5a01 | 5253 | if (ss->exit) { |
eb95419b TH |
5254 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
5255 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5256 | |
eb95419b | 5257 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
5258 | } |
5259 | } | |
5260 | } | |
d41d5a01 | 5261 | |
eaf797ab TH |
5262 | if (put_cset) |
5263 | put_css_set(cset, true); | |
b4f48b63 | 5264 | } |
697f4161 | 5265 | |
bd89aabc | 5266 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5267 | { |
f3d46500 TH |
5268 | if (cgroup_is_releasable(cgrp) && list_empty(&cgrp->cset_links) && |
5269 | !css_has_online_children(&cgrp->self)) { | |
f50daa70 LZ |
5270 | /* |
5271 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 5272 | * already queued for a userspace notification, queue |
f50daa70 LZ |
5273 | * it now |
5274 | */ | |
81a6a5cd | 5275 | int need_schedule_work = 0; |
f50daa70 | 5276 | |
cdcc136f | 5277 | raw_spin_lock(&release_list_lock); |
54766d4a | 5278 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
5279 | list_empty(&cgrp->release_list)) { |
5280 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5281 | need_schedule_work = 1; |
5282 | } | |
cdcc136f | 5283 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5284 | if (need_schedule_work) |
5285 | schedule_work(&release_agent_work); | |
5286 | } | |
5287 | } | |
5288 | ||
81a6a5cd PM |
5289 | /* |
5290 | * Notify userspace when a cgroup is released, by running the | |
5291 | * configured release agent with the name of the cgroup (path | |
5292 | * relative to the root of cgroup file system) as the argument. | |
5293 | * | |
5294 | * Most likely, this user command will try to rmdir this cgroup. | |
5295 | * | |
5296 | * This races with the possibility that some other task will be | |
5297 | * attached to this cgroup before it is removed, or that some other | |
5298 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5299 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5300 | * unused, and this cgroup will be reprieved from its death sentence, | |
5301 | * to continue to serve a useful existence. Next time it's released, | |
5302 | * we will get notified again, if it still has 'notify_on_release' set. | |
5303 | * | |
5304 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5305 | * means only wait until the task is successfully execve()'d. The | |
5306 | * separate release agent task is forked by call_usermodehelper(), | |
5307 | * then control in this thread returns here, without waiting for the | |
5308 | * release agent task. We don't bother to wait because the caller of | |
5309 | * this routine has no use for the exit status of the release agent | |
5310 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5311 | */ |
81a6a5cd PM |
5312 | static void cgroup_release_agent(struct work_struct *work) |
5313 | { | |
5314 | BUG_ON(work != &release_agent_work); | |
5315 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5316 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5317 | while (!list_empty(&release_list)) { |
5318 | char *argv[3], *envp[3]; | |
5319 | int i; | |
e61734c5 | 5320 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 5321 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5322 | struct cgroup, |
5323 | release_list); | |
bd89aabc | 5324 | list_del_init(&cgrp->release_list); |
cdcc136f | 5325 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 5326 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
5327 | if (!pathbuf) |
5328 | goto continue_free; | |
e61734c5 TH |
5329 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
5330 | if (!path) | |
e788e066 PM |
5331 | goto continue_free; |
5332 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5333 | if (!agentbuf) | |
5334 | goto continue_free; | |
81a6a5cd PM |
5335 | |
5336 | i = 0; | |
e788e066 | 5337 | argv[i++] = agentbuf; |
e61734c5 | 5338 | argv[i++] = path; |
81a6a5cd PM |
5339 | argv[i] = NULL; |
5340 | ||
5341 | i = 0; | |
5342 | /* minimal command environment */ | |
5343 | envp[i++] = "HOME=/"; | |
5344 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5345 | envp[i] = NULL; | |
5346 | ||
5347 | /* Drop the lock while we invoke the usermode helper, | |
5348 | * since the exec could involve hitting disk and hence | |
5349 | * be a slow process */ | |
5350 | mutex_unlock(&cgroup_mutex); | |
5351 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5352 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5353 | continue_free: |
5354 | kfree(pathbuf); | |
5355 | kfree(agentbuf); | |
cdcc136f | 5356 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5357 | } |
cdcc136f | 5358 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5359 | mutex_unlock(&cgroup_mutex); |
5360 | } | |
8bab8dde PM |
5361 | |
5362 | static int __init cgroup_disable(char *str) | |
5363 | { | |
30159ec7 | 5364 | struct cgroup_subsys *ss; |
8bab8dde | 5365 | char *token; |
30159ec7 | 5366 | int i; |
8bab8dde PM |
5367 | |
5368 | while ((token = strsep(&str, ",")) != NULL) { | |
5369 | if (!*token) | |
5370 | continue; | |
be45c900 | 5371 | |
3ed80a62 | 5372 | for_each_subsys(ss, i) { |
8bab8dde PM |
5373 | if (!strcmp(token, ss->name)) { |
5374 | ss->disabled = 1; | |
5375 | printk(KERN_INFO "Disabling %s control group" | |
5376 | " subsystem\n", ss->name); | |
5377 | break; | |
5378 | } | |
5379 | } | |
5380 | } | |
5381 | return 1; | |
5382 | } | |
5383 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5384 | |
a8ddc821 TH |
5385 | static int __init cgroup_set_legacy_files_on_dfl(char *str) |
5386 | { | |
5387 | printk("cgroup: using legacy files on the default hierarchy\n"); | |
5388 | cgroup_legacy_files_on_dfl = true; | |
5389 | return 0; | |
5390 | } | |
5391 | __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl); | |
5392 | ||
b77d7b60 | 5393 | /** |
ec903c0c | 5394 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5395 | * @dentry: directory dentry of interest |
5396 | * @ss: subsystem of interest | |
b77d7b60 | 5397 | * |
5a17f543 TH |
5398 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5399 | * to get the corresponding css and return it. If such css doesn't exist | |
5400 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5401 | */ |
ec903c0c TH |
5402 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5403 | struct cgroup_subsys *ss) | |
e5d1367f | 5404 | { |
2bd59d48 TH |
5405 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5406 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5407 | struct cgroup *cgrp; |
e5d1367f | 5408 | |
35cf0836 | 5409 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5410 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5411 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5412 | return ERR_PTR(-EBADF); |
5413 | ||
5a17f543 TH |
5414 | rcu_read_lock(); |
5415 | ||
2bd59d48 TH |
5416 | /* |
5417 | * This path doesn't originate from kernfs and @kn could already | |
5418 | * have been or be removed at any point. @kn->priv is RCU | |
cfc79d5b | 5419 | * protected for this access. See cgroup_rmdir() for details. |
2bd59d48 TH |
5420 | */ |
5421 | cgrp = rcu_dereference(kn->priv); | |
5422 | if (cgrp) | |
5423 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5424 | |
ec903c0c | 5425 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5426 | css = ERR_PTR(-ENOENT); |
5427 | ||
5428 | rcu_read_unlock(); | |
5429 | return css; | |
e5d1367f | 5430 | } |
e5d1367f | 5431 | |
1cb650b9 LZ |
5432 | /** |
5433 | * css_from_id - lookup css by id | |
5434 | * @id: the cgroup id | |
5435 | * @ss: cgroup subsys to be looked into | |
5436 | * | |
5437 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5438 | * Should be called under rcu_read_lock(). | |
5439 | */ | |
5440 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5441 | { | |
6fa4918d | 5442 | WARN_ON_ONCE(!rcu_read_lock_held()); |
15a4c835 | 5443 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5444 | } |
5445 | ||
fe693435 | 5446 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5447 | static struct cgroup_subsys_state * |
5448 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5449 | { |
5450 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5451 | ||
5452 | if (!css) | |
5453 | return ERR_PTR(-ENOMEM); | |
5454 | ||
5455 | return css; | |
5456 | } | |
5457 | ||
eb95419b | 5458 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5459 | { |
eb95419b | 5460 | kfree(css); |
fe693435 PM |
5461 | } |
5462 | ||
182446d0 TH |
5463 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5464 | struct cftype *cft) | |
fe693435 | 5465 | { |
182446d0 | 5466 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5467 | } |
5468 | ||
182446d0 TH |
5469 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5470 | struct cftype *cft) | |
fe693435 PM |
5471 | { |
5472 | return (u64)(unsigned long)current->cgroups; | |
5473 | } | |
5474 | ||
182446d0 | 5475 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5476 | struct cftype *cft) |
fe693435 PM |
5477 | { |
5478 | u64 count; | |
5479 | ||
5480 | rcu_read_lock(); | |
a8ad805c | 5481 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5482 | rcu_read_unlock(); |
5483 | return count; | |
5484 | } | |
5485 | ||
2da8ca82 | 5486 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5487 | { |
69d0206c | 5488 | struct cgrp_cset_link *link; |
5abb8855 | 5489 | struct css_set *cset; |
e61734c5 TH |
5490 | char *name_buf; |
5491 | ||
5492 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5493 | if (!name_buf) | |
5494 | return -ENOMEM; | |
7717f7ba | 5495 | |
96d365e0 | 5496 | down_read(&css_set_rwsem); |
7717f7ba | 5497 | rcu_read_lock(); |
5abb8855 | 5498 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5499 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5500 | struct cgroup *c = link->cgrp; |
7717f7ba | 5501 | |
a2dd4247 | 5502 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5503 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5504 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5505 | } |
5506 | rcu_read_unlock(); | |
96d365e0 | 5507 | up_read(&css_set_rwsem); |
e61734c5 | 5508 | kfree(name_buf); |
7717f7ba PM |
5509 | return 0; |
5510 | } | |
5511 | ||
5512 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5513 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5514 | { |
2da8ca82 | 5515 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5516 | struct cgrp_cset_link *link; |
7717f7ba | 5517 | |
96d365e0 | 5518 | down_read(&css_set_rwsem); |
182446d0 | 5519 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5520 | struct css_set *cset = link->cset; |
7717f7ba PM |
5521 | struct task_struct *task; |
5522 | int count = 0; | |
c7561128 | 5523 | |
5abb8855 | 5524 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5525 | |
5abb8855 | 5526 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5527 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5528 | goto overflow; | |
5529 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5530 | } | |
5531 | ||
5532 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5533 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5534 | goto overflow; | |
5535 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5536 | } |
c7561128 TH |
5537 | continue; |
5538 | overflow: | |
5539 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5540 | } |
96d365e0 | 5541 | up_read(&css_set_rwsem); |
7717f7ba PM |
5542 | return 0; |
5543 | } | |
5544 | ||
182446d0 | 5545 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5546 | { |
182446d0 | 5547 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5548 | } |
5549 | ||
5550 | static struct cftype debug_files[] = { | |
fe693435 PM |
5551 | { |
5552 | .name = "taskcount", | |
5553 | .read_u64 = debug_taskcount_read, | |
5554 | }, | |
5555 | ||
5556 | { | |
5557 | .name = "current_css_set", | |
5558 | .read_u64 = current_css_set_read, | |
5559 | }, | |
5560 | ||
5561 | { | |
5562 | .name = "current_css_set_refcount", | |
5563 | .read_u64 = current_css_set_refcount_read, | |
5564 | }, | |
5565 | ||
7717f7ba PM |
5566 | { |
5567 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5568 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5569 | }, |
5570 | ||
5571 | { | |
5572 | .name = "cgroup_css_links", | |
2da8ca82 | 5573 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5574 | }, |
5575 | ||
fe693435 PM |
5576 | { |
5577 | .name = "releasable", | |
5578 | .read_u64 = releasable_read, | |
5579 | }, | |
fe693435 | 5580 | |
4baf6e33 TH |
5581 | { } /* terminate */ |
5582 | }; | |
fe693435 | 5583 | |
073219e9 | 5584 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5585 | .css_alloc = debug_css_alloc, |
5586 | .css_free = debug_css_free, | |
5577964e | 5587 | .legacy_cftypes = debug_files, |
fe693435 PM |
5588 | }; |
5589 | #endif /* CONFIG_CGROUP_DEBUG */ |