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 | ||
0a268dbd TH |
31 | #include "cgroup-internal.h" |
32 | ||
aef2feda | 33 | #include <linux/bpf-cgroup.h> |
2ce9738b | 34 | #include <linux/cred.h> |
ddbcc7e8 | 35 | #include <linux/errno.h> |
2ce9738b | 36 | #include <linux/init_task.h> |
ddbcc7e8 | 37 | #include <linux/kernel.h> |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mutex.h> |
40 | #include <linux/mount.h> | |
41 | #include <linux/pagemap.h> | |
a424316c | 42 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
43 | #include <linux/rcupdate.h> |
44 | #include <linux/sched.h> | |
29930025 | 45 | #include <linux/sched/task.h> |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
0ac801fe | 50 | #include <linux/hashtable.h> |
2c6ab6d2 | 51 | #include <linux/idr.h> |
c4c27fbd | 52 | #include <linux/kthread.h> |
60063497 | 53 | #include <linux/atomic.h> |
e93ad19d | 54 | #include <linux/cpuset.h> |
a79a908f AK |
55 | #include <linux/proc_ns.h> |
56 | #include <linux/nsproxy.h> | |
1f3fe7eb | 57 | #include <linux/file.h> |
e34a98d5 | 58 | #include <linux/fs_parser.h> |
d4ff749b | 59 | #include <linux/sched/cputime.h> |
2ce7135a | 60 | #include <linux/psi.h> |
bd1060a1 | 61 | #include <net/sock.h> |
ddbcc7e8 | 62 | |
ed1777de TH |
63 | #define CREATE_TRACE_POINTS |
64 | #include <trace/events/cgroup.h> | |
65 | ||
8d7e6fb0 TH |
66 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
67 | MAX_CFTYPE_NAME + 2) | |
b12e3583 TH |
68 | /* let's not notify more than 100 times per second */ |
69 | #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100) | |
8d7e6fb0 | 70 | |
d20d30eb KC |
71 | /* |
72 | * To avoid confusing the compiler (and generating warnings) with code | |
73 | * that attempts to access what would be a 0-element array (i.e. sized | |
74 | * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this | |
75 | * constant expression can be added. | |
76 | */ | |
77 | #define CGROUP_HAS_SUBSYS_CONFIG (CGROUP_SUBSYS_COUNT > 0) | |
78 | ||
e25e2cbb TH |
79 | /* |
80 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
81 | * hierarchy must be performed while holding it. | |
82 | * | |
f0d9a5f1 | 83 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 84 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 85 | * |
0e1d768f TH |
86 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
87 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 88 | */ |
2219449a | 89 | DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 90 | DEFINE_SPINLOCK(css_set_lock); |
0a268dbd TH |
91 | |
92 | #ifdef CONFIG_PROVE_RCU | |
0e1d768f | 93 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 94 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a TH |
95 | #endif |
96 | ||
e4f8d81c SRV |
97 | DEFINE_SPINLOCK(trace_cgroup_path_lock); |
98 | char trace_cgroup_path[TRACE_CGROUP_PATH_LEN]; | |
29ed1738 | 99 | static bool cgroup_debug __read_mostly; |
e4f8d81c | 100 | |
6fa4918d | 101 | /* |
15a4c835 TH |
102 | * Protects cgroup_idr and css_idr so that IDs can be released without |
103 | * grabbing cgroup_mutex. | |
6fa4918d TH |
104 | */ |
105 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
106 | ||
34c06254 TH |
107 | /* |
108 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
109 | * against file removal/re-creation across css hiding. | |
110 | */ | |
111 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
112 | ||
3f2947b7 | 113 | DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem); |
1ed13287 | 114 | |
8353da1f | 115 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
116 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
117 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 118 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 119 | |
e5fca243 TH |
120 | /* |
121 | * cgroup destruction makes heavy use of work items and there can be a lot | |
122 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
123 | * destruction work items don't end up filling up max_active of system_wq | |
124 | * which may lead to deadlock. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_destroy_wq; | |
127 | ||
3ed80a62 | 128 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 129 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
0a268dbd | 130 | struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
131 | #include <linux/cgroup_subsys.h> |
132 | }; | |
073219e9 TH |
133 | #undef SUBSYS |
134 | ||
135 | /* array of cgroup subsystem names */ | |
136 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
137 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
138 | #include <linux/cgroup_subsys.h> |
139 | }; | |
073219e9 | 140 | #undef SUBSYS |
ddbcc7e8 | 141 | |
49d1dc4b TH |
142 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
143 | #define SUBSYS(_x) \ | |
144 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
145 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
146 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
147 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
148 | #include <linux/cgroup_subsys.h> | |
149 | #undef SUBSYS | |
150 | ||
151 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
152 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
153 | #include <linux/cgroup_subsys.h> | |
154 | }; | |
155 | #undef SUBSYS | |
156 | ||
157 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
158 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
159 | #include <linux/cgroup_subsys.h> | |
160 | }; | |
161 | #undef SUBSYS | |
162 | ||
c58632b3 | 163 | static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu); |
041cd640 | 164 | |
6b6ebb34 | 165 | /* the default hierarchy */ |
c58632b3 | 166 | struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu }; |
d0ec4230 | 167 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 168 | |
a2dd4247 TH |
169 | /* |
170 | * The default hierarchy always exists but is hidden until mounted for the | |
171 | * first time. This is for backward compatibility. | |
172 | */ | |
a7165264 | 173 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 174 | |
5533e011 | 175 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 176 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 177 | |
f6d635ad | 178 | /* some controllers are implicitly enabled on the default hierarchy */ |
b807421a | 179 | static u16 cgrp_dfl_implicit_ss_mask; |
f6d635ad | 180 | |
8cfd8147 TH |
181 | /* some controllers can be threaded on the default hierarchy */ |
182 | static u16 cgrp_dfl_threaded_ss_mask; | |
183 | ||
ddbcc7e8 | 184 | /* The list of hierarchy roots */ |
0a268dbd | 185 | LIST_HEAD(cgroup_roots); |
9871bf95 | 186 | static int cgroup_root_count; |
ddbcc7e8 | 187 | |
3417ae1f | 188 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 189 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 190 | |
794611a1 | 191 | /* |
0cb51d71 TH |
192 | * Assign a monotonically increasing serial number to csses. It guarantees |
193 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
194 | * Also, as csses are always appended to the parent's ->children list, it | |
195 | * guarantees that sibling csses are always sorted in the ascending serial | |
196 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 197 | */ |
0cb51d71 | 198 | static u64 css_serial_nr_next = 1; |
794611a1 | 199 | |
cb4a3167 | 200 | /* |
b807421a TH |
201 | * These bitmasks identify subsystems with specific features to avoid |
202 | * having to do iterative checks repeatedly. | |
ddbcc7e8 | 203 | */ |
6e5c8307 TH |
204 | static u16 have_fork_callback __read_mostly; |
205 | static u16 have_exit_callback __read_mostly; | |
51bee5ab | 206 | static u16 have_release_callback __read_mostly; |
b807421a | 207 | static u16 have_canfork_callback __read_mostly; |
ddbcc7e8 | 208 | |
a79a908f AK |
209 | /* cgroup namespace for init task */ |
210 | struct cgroup_namespace init_cgroup_ns = { | |
f387882d | 211 | .ns.count = REFCOUNT_INIT(2), |
a79a908f AK |
212 | .user_ns = &init_user_ns, |
213 | .ns.ops = &cgroupns_operations, | |
214 | .ns.inum = PROC_CGROUP_INIT_INO, | |
215 | .root_cset = &init_css_set, | |
216 | }; | |
217 | ||
67e9c74b | 218 | static struct file_system_type cgroup2_fs_type; |
d62beb7f | 219 | static struct cftype cgroup_base_files[]; |
8a693f77 | 220 | static struct cftype cgroup_psi_files[]; |
628f7cd4 | 221 | |
3958e2d0 SB |
222 | /* cgroup optional features */ |
223 | enum cgroup_opt_features { | |
224 | #ifdef CONFIG_PSI | |
225 | OPT_FEATURE_PRESSURE, | |
226 | #endif | |
227 | OPT_FEATURE_COUNT | |
228 | }; | |
229 | ||
230 | static const char *cgroup_opt_feature_names[OPT_FEATURE_COUNT] = { | |
231 | #ifdef CONFIG_PSI | |
232 | "pressure", | |
233 | #endif | |
234 | }; | |
235 | ||
236 | static u16 cgroup_feature_disable_mask __read_mostly; | |
237 | ||
334c3679 TH |
238 | static int cgroup_apply_control(struct cgroup *cgrp); |
239 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
b636fd38 TH |
240 | static void css_task_iter_skip(struct css_task_iter *it, |
241 | struct task_struct *task); | |
42809dd4 | 242 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
243 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
244 | struct cgroup_subsys *ss); | |
9d755d33 | 245 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 246 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
247 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
248 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 249 | bool is_add); |
42809dd4 | 250 | |
6ab42860 TH |
251 | #ifdef CONFIG_DEBUG_CGROUP_REF |
252 | #define CGROUP_REF_FN_ATTRS noinline | |
79a7f41f | 253 | #define CGROUP_REF_EXPORT(fn) EXPORT_SYMBOL_GPL(fn); |
6ab42860 TH |
254 | #include <linux/cgroup_refcnt.h> |
255 | #endif | |
256 | ||
fc5ed1e9 TH |
257 | /** |
258 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
259 | * @ssid: subsys ID of interest | |
260 | * | |
261 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
262 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
263 | * is slower static_key_enabled() based test indexed by @ssid. | |
264 | */ | |
0a268dbd | 265 | bool cgroup_ssid_enabled(int ssid) |
fc5ed1e9 | 266 | { |
d20d30eb | 267 | if (!CGROUP_HAS_SUBSYS_CONFIG) |
cfe02a8a AB |
268 | return false; |
269 | ||
fc5ed1e9 TH |
270 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
271 | } | |
272 | ||
9e10a130 TH |
273 | /** |
274 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
275 | * @cgrp: the cgroup of interest | |
276 | * | |
277 | * The default hierarchy is the v2 interface of cgroup and this function | |
278 | * can be used to test whether a cgroup is on the default hierarchy for | |
58315c96 | 279 | * cases where a subsystem should behave differently depending on the |
9e10a130 TH |
280 | * interface version. |
281 | * | |
9e10a130 TH |
282 | * List of changed behaviors: |
283 | * | |
284 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
285 | * and "name" are disallowed. | |
286 | * | |
287 | * - When mounting an existing superblock, mount options should match. | |
288 | * | |
9e10a130 TH |
289 | * - rename(2) is disallowed. |
290 | * | |
291 | * - "tasks" is removed. Everything should be at process granularity. Use | |
292 | * "cgroup.procs" instead. | |
293 | * | |
294 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
58315c96 | 295 | * recycled in-between reads. |
9e10a130 TH |
296 | * |
297 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
298 | * notification mechanism will be implemented. | |
299 | * | |
300 | * - "cgroup.clone_children" is removed. | |
301 | * | |
302 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
303 | * and its descendants contain no task; otherwise, 1. The file also | |
304 | * generates kernfs notification which can be monitored through poll and | |
305 | * [di]notify when the value of the file changes. | |
306 | * | |
307 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
308 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
309 | * moved to an ancestor. | |
310 | * | |
311 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
312 | * masks of ancestors. | |
313 | * | |
9e10a130 TH |
314 | * - blkcg: blk-throttle becomes properly hierarchical. |
315 | * | |
316 | * - debug: disallowed on the default hierarchy. | |
317 | */ | |
0a268dbd | 318 | bool cgroup_on_dfl(const struct cgroup *cgrp) |
9e10a130 TH |
319 | { |
320 | return cgrp->root == &cgrp_dfl_root; | |
321 | } | |
322 | ||
6fa4918d TH |
323 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
324 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
325 | gfp_t gfp_mask) | |
326 | { | |
327 | int ret; | |
328 | ||
329 | idr_preload(gfp_mask); | |
54504e97 | 330 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 331 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 332 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
333 | idr_preload_end(); |
334 | return ret; | |
335 | } | |
336 | ||
337 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
338 | { | |
339 | void *ret; | |
340 | ||
54504e97 | 341 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 342 | ret = idr_replace(idr, ptr, id); |
54504e97 | 343 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
344 | return ret; |
345 | } | |
346 | ||
347 | static void cgroup_idr_remove(struct idr *idr, int id) | |
348 | { | |
54504e97 | 349 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 350 | idr_remove(idr, id); |
54504e97 | 351 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
352 | } |
353 | ||
27f26753 | 354 | static bool cgroup_has_tasks(struct cgroup *cgrp) |
d51f39b0 | 355 | { |
27f26753 TH |
356 | return cgrp->nr_populated_csets; |
357 | } | |
d51f39b0 | 358 | |
7a0cf0e7 | 359 | bool cgroup_is_threaded(struct cgroup *cgrp) |
454000ad TH |
360 | { |
361 | return cgrp->dom_cgrp != cgrp; | |
362 | } | |
363 | ||
8cfd8147 TH |
364 | /* can @cgrp host both domain and threaded children? */ |
365 | static bool cgroup_is_mixable(struct cgroup *cgrp) | |
366 | { | |
367 | /* | |
368 | * Root isn't under domain level resource control exempting it from | |
369 | * the no-internal-process constraint, so it can serve as a thread | |
370 | * root and a parent of resource domains at the same time. | |
371 | */ | |
372 | return !cgroup_parent(cgrp); | |
373 | } | |
374 | ||
58315c96 | 375 | /* can @cgrp become a thread root? Should always be true for a thread root */ |
8cfd8147 TH |
376 | static bool cgroup_can_be_thread_root(struct cgroup *cgrp) |
377 | { | |
378 | /* mixables don't care */ | |
379 | if (cgroup_is_mixable(cgrp)) | |
380 | return true; | |
381 | ||
382 | /* domain roots can't be nested under threaded */ | |
383 | if (cgroup_is_threaded(cgrp)) | |
384 | return false; | |
385 | ||
386 | /* can only have either domain or threaded children */ | |
387 | if (cgrp->nr_populated_domain_children) | |
388 | return false; | |
389 | ||
390 | /* and no domain controllers can be enabled */ | |
391 | if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
392 | return false; | |
393 | ||
394 | return true; | |
395 | } | |
396 | ||
397 | /* is @cgrp root of a threaded subtree? */ | |
7a0cf0e7 | 398 | bool cgroup_is_thread_root(struct cgroup *cgrp) |
8cfd8147 TH |
399 | { |
400 | /* thread root should be a domain */ | |
401 | if (cgroup_is_threaded(cgrp)) | |
402 | return false; | |
403 | ||
404 | /* a domain w/ threaded children is a thread root */ | |
405 | if (cgrp->nr_threaded_children) | |
406 | return true; | |
407 | ||
408 | /* | |
409 | * A domain which has tasks and explicit threaded controllers | |
410 | * enabled is a thread root. | |
411 | */ | |
412 | if (cgroup_has_tasks(cgrp) && | |
413 | (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask)) | |
414 | return true; | |
415 | ||
416 | return false; | |
417 | } | |
418 | ||
419 | /* a domain which isn't connected to the root w/o brekage can't be used */ | |
420 | static bool cgroup_is_valid_domain(struct cgroup *cgrp) | |
421 | { | |
422 | /* the cgroup itself can be a thread root */ | |
423 | if (cgroup_is_threaded(cgrp)) | |
424 | return false; | |
425 | ||
426 | /* but the ancestors can't be unless mixable */ | |
427 | while ((cgrp = cgroup_parent(cgrp))) { | |
428 | if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp)) | |
429 | return false; | |
430 | if (cgroup_is_threaded(cgrp)) | |
431 | return false; | |
432 | } | |
433 | ||
434 | return true; | |
d51f39b0 TH |
435 | } |
436 | ||
5531dc91 TH |
437 | /* subsystems visibly enabled on a cgroup */ |
438 | static u16 cgroup_control(struct cgroup *cgrp) | |
439 | { | |
440 | struct cgroup *parent = cgroup_parent(cgrp); | |
441 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
442 | ||
8cfd8147 TH |
443 | if (parent) { |
444 | u16 ss_mask = parent->subtree_control; | |
445 | ||
446 | /* threaded cgroups can only have threaded controllers */ | |
447 | if (cgroup_is_threaded(cgrp)) | |
448 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
449 | return ss_mask; | |
450 | } | |
5531dc91 TH |
451 | |
452 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
453 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
454 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
455 | return root_ss_mask; |
456 | } | |
457 | ||
458 | /* subsystems enabled on a cgroup */ | |
459 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
460 | { | |
461 | struct cgroup *parent = cgroup_parent(cgrp); | |
462 | ||
8cfd8147 TH |
463 | if (parent) { |
464 | u16 ss_mask = parent->subtree_ss_mask; | |
465 | ||
466 | /* threaded cgroups can only have threaded controllers */ | |
467 | if (cgroup_is_threaded(cgrp)) | |
468 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
469 | return ss_mask; | |
470 | } | |
5531dc91 TH |
471 | |
472 | return cgrp->root->subsys_mask; | |
473 | } | |
474 | ||
95109b62 TH |
475 | /** |
476 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
477 | * @cgrp: the cgroup of interest | |
9d800df1 | 478 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 479 | * |
ca8bdcaf TH |
480 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
481 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
482 | * the caller is responsible for pinning the returned css if it wants to | |
483 | * keep accessing it outside the said locks. This function may return | |
484 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
485 | */ |
486 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 487 | struct cgroup_subsys *ss) |
95109b62 | 488 | { |
d20d30eb | 489 | if (CGROUP_HAS_SUBSYS_CONFIG && ss) |
aec25020 | 490 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 491 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 492 | else |
9d800df1 | 493 | return &cgrp->self; |
95109b62 | 494 | } |
42809dd4 | 495 | |
d41bf8c9 TH |
496 | /** |
497 | * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem | |
498 | * @cgrp: the cgroup of interest | |
499 | * @ss: the subsystem of interest | |
500 | * | |
08b2b6fd | 501 | * Find and get @cgrp's css associated with @ss. If the css doesn't exist |
d41bf8c9 TH |
502 | * or is offline, %NULL is returned. |
503 | */ | |
504 | static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp, | |
505 | struct cgroup_subsys *ss) | |
506 | { | |
507 | struct cgroup_subsys_state *css; | |
508 | ||
509 | rcu_read_lock(); | |
510 | css = cgroup_css(cgrp, ss); | |
a581563f | 511 | if (css && !css_tryget_online(css)) |
d41bf8c9 TH |
512 | css = NULL; |
513 | rcu_read_unlock(); | |
514 | ||
515 | return css; | |
516 | } | |
517 | ||
aec3dfcb | 518 | /** |
fc5a828b | 519 | * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss |
aec3dfcb | 520 | * @cgrp: the cgroup of interest |
9d800df1 | 521 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 522 | * |
d0f702e6 | 523 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
524 | * as the matching css of the nearest ancestor including self which has @ss |
525 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
526 | * function is guaranteed to return non-NULL css. | |
527 | */ | |
fc5a828b DZ |
528 | static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp, |
529 | struct cgroup_subsys *ss) | |
aec3dfcb TH |
530 | { |
531 | lockdep_assert_held(&cgroup_mutex); | |
532 | ||
533 | if (!ss) | |
9d800df1 | 534 | return &cgrp->self; |
aec3dfcb | 535 | |
eeecbd19 TH |
536 | /* |
537 | * This function is used while updating css associations and thus | |
5531dc91 | 538 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 539 | */ |
5531dc91 | 540 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 541 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
542 | if (!cgrp) |
543 | return NULL; | |
544 | } | |
aec3dfcb TH |
545 | |
546 | return cgroup_css(cgrp, ss); | |
95109b62 | 547 | } |
42809dd4 | 548 | |
fc5a828b DZ |
549 | /** |
550 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
551 | * @cgrp: the cgroup of interest | |
552 | * @ss: the subsystem of interest | |
553 | * | |
554 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
555 | * defined as the matching css of the nearest ancestor including self which | |
556 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
557 | * the root css is returned, so this function always returns a valid css. | |
558 | * | |
559 | * The returned css is not guaranteed to be online, and therefore it is the | |
58315c96 | 560 | * callers responsibility to try get a reference for it. |
fc5a828b DZ |
561 | */ |
562 | struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
563 | struct cgroup_subsys *ss) | |
564 | { | |
565 | struct cgroup_subsys_state *css; | |
566 | ||
d20d30eb KC |
567 | if (!CGROUP_HAS_SUBSYS_CONFIG) |
568 | return NULL; | |
569 | ||
fc5a828b DZ |
570 | do { |
571 | css = cgroup_css(cgrp, ss); | |
572 | ||
573 | if (css) | |
574 | return css; | |
575 | cgrp = cgroup_parent(cgrp); | |
576 | } while (cgrp); | |
577 | ||
578 | return init_css_set.subsys[ss->id]; | |
579 | } | |
580 | ||
eeecbd19 TH |
581 | /** |
582 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
583 | * @cgrp: the cgroup of interest | |
584 | * @ss: the subsystem of interest | |
585 | * | |
586 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
587 | * defined as the matching css of the nearest ancestor including self which | |
588 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
589 | * the root css is returned, so this function always returns a valid css. | |
590 | * The returned css must be put using css_put(). | |
591 | */ | |
592 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
593 | struct cgroup_subsys *ss) | |
594 | { | |
595 | struct cgroup_subsys_state *css; | |
596 | ||
d20d30eb KC |
597 | if (!CGROUP_HAS_SUBSYS_CONFIG) |
598 | return NULL; | |
599 | ||
eeecbd19 TH |
600 | rcu_read_lock(); |
601 | ||
602 | do { | |
603 | css = cgroup_css(cgrp, ss); | |
604 | ||
605 | if (css && css_tryget_online(css)) | |
606 | goto out_unlock; | |
607 | cgrp = cgroup_parent(cgrp); | |
608 | } while (cgrp); | |
609 | ||
610 | css = init_css_set.subsys[ss->id]; | |
611 | css_get(css); | |
612 | out_unlock: | |
613 | rcu_read_unlock(); | |
614 | return css; | |
615 | } | |
c74d40e8 | 616 | EXPORT_SYMBOL_GPL(cgroup_get_e_css); |
eeecbd19 | 617 | |
a590b90d | 618 | static void cgroup_get_live(struct cgroup *cgrp) |
052c3f3a TH |
619 | { |
620 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
621 | css_get(&cgrp->self); | |
622 | } | |
623 | ||
aade7f9e RG |
624 | /** |
625 | * __cgroup_task_count - count the number of tasks in a cgroup. The caller | |
626 | * is responsible for taking the css_set_lock. | |
627 | * @cgrp: the cgroup in question | |
628 | */ | |
629 | int __cgroup_task_count(const struct cgroup *cgrp) | |
630 | { | |
631 | int count = 0; | |
632 | struct cgrp_cset_link *link; | |
633 | ||
634 | lockdep_assert_held(&css_set_lock); | |
635 | ||
636 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
637 | count += link->cset->nr_tasks; | |
638 | ||
639 | return count; | |
640 | } | |
641 | ||
642 | /** | |
643 | * cgroup_task_count - count the number of tasks in a cgroup. | |
644 | * @cgrp: the cgroup in question | |
645 | */ | |
646 | int cgroup_task_count(const struct cgroup *cgrp) | |
647 | { | |
648 | int count; | |
649 | ||
650 | spin_lock_irq(&css_set_lock); | |
651 | count = __cgroup_task_count(cgrp); | |
652 | spin_unlock_irq(&css_set_lock); | |
653 | ||
654 | return count; | |
655 | } | |
656 | ||
b4168640 | 657 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 658 | { |
2bd59d48 | 659 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 660 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
661 | |
662 | /* | |
663 | * This is open and unprotected implementation of cgroup_css(). | |
664 | * seq_css() is only called from a kernfs file operation which has | |
665 | * an active reference on the file. Because all the subsystem | |
666 | * files are drained before a css is disassociated with a cgroup, | |
667 | * the matching css from the cgroup's subsys table is guaranteed to | |
668 | * be and stay valid until the enclosing operation is complete. | |
669 | */ | |
d20d30eb | 670 | if (CGROUP_HAS_SUBSYS_CONFIG && cft->ss) |
2bd59d48 TH |
671 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); |
672 | else | |
9d800df1 | 673 | return &cgrp->self; |
59f5296b | 674 | } |
b4168640 | 675 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 676 | |
1c6727af TH |
677 | /** |
678 | * for_each_css - iterate all css's of a cgroup | |
679 | * @css: the iteration cursor | |
680 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
681 | * @cgrp: the target cgroup to iterate css's of | |
682 | * | |
aec3dfcb | 683 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
684 | */ |
685 | #define for_each_css(css, ssid, cgrp) \ | |
686 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
687 | if (!((css) = rcu_dereference_check( \ | |
688 | (cgrp)->subsys[(ssid)], \ | |
689 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
690 | else | |
691 | ||
aec3dfcb TH |
692 | /** |
693 | * for_each_e_css - iterate all effective css's of a cgroup | |
694 | * @css: the iteration cursor | |
695 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
696 | * @cgrp: the target cgroup to iterate css's of | |
697 | * | |
698 | * Should be called under cgroup_[tree_]mutex. | |
699 | */ | |
fc5a828b DZ |
700 | #define for_each_e_css(css, ssid, cgrp) \ |
701 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
702 | if (!((css) = cgroup_e_css_by_mask(cgrp, \ | |
703 | cgroup_subsys[(ssid)]))) \ | |
704 | ; \ | |
aec3dfcb TH |
705 | else |
706 | ||
cb4a3167 | 707 | /** |
b4e0eeaf | 708 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
709 | * @ss: the iteration cursor |
710 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 711 | * @ss_mask: the bitmask |
cb4a3167 AS |
712 | * |
713 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 714 | * @ss_mask is set. |
cb4a3167 | 715 | */ |
b4e0eeaf TH |
716 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
717 | unsigned long __ss_mask = (ss_mask); \ | |
d20d30eb | 718 | if (!CGROUP_HAS_SUBSYS_CONFIG) { \ |
4a705c5c | 719 | (ssid) = 0; \ |
b4e0eeaf TH |
720 | break; \ |
721 | } \ | |
722 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
723 | (ss) = cgroup_subsys[ssid]; \ | |
724 | { | |
725 | ||
726 | #define while_each_subsys_mask() \ | |
727 | } \ | |
728 | } \ | |
729 | } while (false) | |
cb4a3167 | 730 | |
f8f22e53 TH |
731 | /* iterate over child cgrps, lock should be held throughout iteration */ |
732 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 733 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 734 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
735 | cgroup_is_dead(child); })) \ |
736 | ; \ | |
737 | else | |
7ae1bad9 | 738 | |
58315c96 | 739 | /* walk live descendants in pre order */ |
ce3f1d9d TH |
740 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ |
741 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
742 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
743 | (dsct) = (d_css)->cgroup; \ | |
744 | cgroup_is_dead(dsct); })) \ | |
745 | ; \ | |
746 | else | |
747 | ||
748 | /* walk live descendants in postorder */ | |
749 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
750 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
751 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
752 | (dsct) = (d_css)->cgroup; \ | |
753 | cgroup_is_dead(dsct); })) \ | |
754 | ; \ | |
755 | else | |
756 | ||
172a2c06 TH |
757 | /* |
758 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
759 | * hierarchies being mounted. It contains a pointer to the root state |
760 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
761 | * reference-counted, to improve performance when child cgroups | |
762 | * haven't been created. | |
763 | */ | |
5024ae29 | 764 | struct css_set init_css_set = { |
4b9502e6 | 765 | .refcount = REFCOUNT_INIT(1), |
454000ad | 766 | .dom_cset = &init_css_set, |
172a2c06 TH |
767 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
768 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
c03cd773 | 769 | .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks), |
5f617ebb | 770 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
454000ad | 771 | .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets), |
5f617ebb | 772 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), |
07fd5b6c TH |
773 | .mg_src_preload_node = LIST_HEAD_INIT(init_css_set.mg_src_preload_node), |
774 | .mg_dst_preload_node = LIST_HEAD_INIT(init_css_set.mg_dst_preload_node), | |
172a2c06 | 775 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), |
38683148 TH |
776 | |
777 | /* | |
778 | * The following field is re-initialized when this cset gets linked | |
779 | * in cgroup_init(). However, let's initialize the field | |
780 | * statically too so that the default cgroup can be accessed safely | |
781 | * early during boot. | |
782 | */ | |
783 | .dfl_cgrp = &cgrp_dfl_root.cgrp, | |
172a2c06 | 784 | }; |
817929ec | 785 | |
172a2c06 | 786 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 787 | |
454000ad TH |
788 | static bool css_set_threaded(struct css_set *cset) |
789 | { | |
790 | return cset->dom_cset != cset; | |
791 | } | |
792 | ||
0de0942d TH |
793 | /** |
794 | * css_set_populated - does a css_set contain any tasks? | |
795 | * @cset: target css_set | |
73a7242a WL |
796 | * |
797 | * css_set_populated() should be the same as !!cset->nr_tasks at steady | |
798 | * state. However, css_set_populated() can be called while a task is being | |
799 | * added to or removed from the linked list before the nr_tasks is | |
800 | * properly updated. Hence, we can't just look at ->nr_tasks here. | |
0de0942d TH |
801 | */ |
802 | static bool css_set_populated(struct css_set *cset) | |
803 | { | |
f0d9a5f1 | 804 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
805 | |
806 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
807 | } | |
808 | ||
842b597e | 809 | /** |
788b950c | 810 | * cgroup_update_populated - update the populated count of a cgroup |
842b597e TH |
811 | * @cgrp: the target cgroup |
812 | * @populated: inc or dec populated count | |
813 | * | |
0de0942d | 814 | * One of the css_sets associated with @cgrp is either getting its first |
788b950c TH |
815 | * task or losing the last. Update @cgrp->nr_populated_* accordingly. The |
816 | * count is propagated towards root so that a given cgroup's | |
817 | * nr_populated_children is zero iff none of its descendants contain any | |
818 | * tasks. | |
842b597e | 819 | * |
788b950c TH |
820 | * @cgrp's interface file "cgroup.populated" is zero if both |
821 | * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and | |
822 | * 1 otherwise. When the sum changes from or to zero, userland is notified | |
823 | * that the content of the interface file has changed. This can be used to | |
824 | * detect when @cgrp and its descendants become populated or empty. | |
842b597e TH |
825 | */ |
826 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
827 | { | |
788b950c TH |
828 | struct cgroup *child = NULL; |
829 | int adj = populated ? 1 : -1; | |
830 | ||
f0d9a5f1 | 831 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
832 | |
833 | do { | |
788b950c | 834 | bool was_populated = cgroup_is_populated(cgrp); |
842b597e | 835 | |
454000ad | 836 | if (!child) { |
788b950c | 837 | cgrp->nr_populated_csets += adj; |
454000ad TH |
838 | } else { |
839 | if (cgroup_is_threaded(child)) | |
840 | cgrp->nr_populated_threaded_children += adj; | |
841 | else | |
842 | cgrp->nr_populated_domain_children += adj; | |
843 | } | |
842b597e | 844 | |
788b950c | 845 | if (was_populated == cgroup_is_populated(cgrp)) |
842b597e TH |
846 | break; |
847 | ||
d62beb7f | 848 | cgroup1_check_for_release(cgrp); |
4c476d8c RG |
849 | TRACE_CGROUP_PATH(notify_populated, cgrp, |
850 | cgroup_is_populated(cgrp)); | |
6f60eade TH |
851 | cgroup_file_notify(&cgrp->events_file); |
852 | ||
788b950c | 853 | child = cgrp; |
d51f39b0 | 854 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
855 | } while (cgrp); |
856 | } | |
857 | ||
0de0942d TH |
858 | /** |
859 | * css_set_update_populated - update populated state of a css_set | |
860 | * @cset: target css_set | |
861 | * @populated: whether @cset is populated or depopulated | |
862 | * | |
863 | * @cset is either getting the first task or losing the last. Update the | |
788b950c | 864 | * populated counters of all associated cgroups accordingly. |
0de0942d TH |
865 | */ |
866 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
867 | { | |
868 | struct cgrp_cset_link *link; | |
869 | ||
f0d9a5f1 | 870 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
871 | |
872 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
873 | cgroup_update_populated(link->cgrp, populated); | |
874 | } | |
875 | ||
b636fd38 TH |
876 | /* |
877 | * @task is leaving, advance task iterators which are pointing to it so | |
878 | * that they can resume at the next position. Advancing an iterator might | |
879 | * remove it from the list, use safe walk. See css_task_iter_skip() for | |
880 | * details. | |
881 | */ | |
882 | static void css_set_skip_task_iters(struct css_set *cset, | |
883 | struct task_struct *task) | |
884 | { | |
885 | struct css_task_iter *it, *pos; | |
886 | ||
887 | list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node) | |
888 | css_task_iter_skip(it, task); | |
889 | } | |
890 | ||
f6d7d049 TH |
891 | /** |
892 | * css_set_move_task - move a task from one css_set to another | |
893 | * @task: task being moved | |
894 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
895 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
896 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
897 | * | |
898 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
899 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
900 | * instead of moved, @to_cset can be NULL. | |
901 | * | |
788b950c | 902 | * This function automatically handles populated counter updates and |
ed27b9f7 TH |
903 | * css_task_iter adjustments but the caller is responsible for managing |
904 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
905 | */ |
906 | static void css_set_move_task(struct task_struct *task, | |
907 | struct css_set *from_cset, struct css_set *to_cset, | |
908 | bool use_mg_tasks) | |
909 | { | |
f0d9a5f1 | 910 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 911 | |
20b454a6 TH |
912 | if (to_cset && !css_set_populated(to_cset)) |
913 | css_set_update_populated(to_cset, true); | |
914 | ||
f6d7d049 TH |
915 | if (from_cset) { |
916 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
ed27b9f7 | 917 | |
b636fd38 | 918 | css_set_skip_task_iters(from_cset, task); |
f6d7d049 TH |
919 | list_del_init(&task->cg_list); |
920 | if (!css_set_populated(from_cset)) | |
921 | css_set_update_populated(from_cset, false); | |
922 | } else { | |
923 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
924 | } | |
925 | ||
926 | if (to_cset) { | |
927 | /* | |
928 | * We are synchronized through cgroup_threadgroup_rwsem | |
929 | * against PF_EXITING setting such that we can't race | |
e7c7b1d8 | 930 | * against cgroup_exit()/cgroup_free() dropping the css_set. |
f6d7d049 TH |
931 | */ |
932 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
933 | ||
2ce7135a | 934 | cgroup_move_task(task, to_cset); |
f6d7d049 TH |
935 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : |
936 | &to_cset->tasks); | |
937 | } | |
938 | } | |
939 | ||
7717f7ba PM |
940 | /* |
941 | * hash table for cgroup groups. This improves the performance to find | |
942 | * an existing css_set. This hash doesn't (currently) take into | |
943 | * account cgroups in empty hierarchies. | |
944 | */ | |
472b1053 | 945 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 946 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 947 | |
0ac801fe | 948 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 949 | { |
0ac801fe | 950 | unsigned long key = 0UL; |
30159ec7 TH |
951 | struct cgroup_subsys *ss; |
952 | int i; | |
472b1053 | 953 | |
30159ec7 | 954 | for_each_subsys(ss, i) |
0ac801fe LZ |
955 | key += (unsigned long)css[i]; |
956 | key = (key >> 16) ^ key; | |
472b1053 | 957 | |
0ac801fe | 958 | return key; |
472b1053 LZ |
959 | } |
960 | ||
dcfe149b | 961 | void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 962 | { |
69d0206c | 963 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
964 | struct cgroup_subsys *ss; |
965 | int ssid; | |
5abb8855 | 966 | |
f0d9a5f1 | 967 | lockdep_assert_held(&css_set_lock); |
89c5509b | 968 | |
4b9502e6 | 969 | if (!refcount_dec_and_test(&cset->refcount)) |
146aa1bd | 970 | return; |
81a6a5cd | 971 | |
454000ad TH |
972 | WARN_ON_ONCE(!list_empty(&cset->threaded_csets)); |
973 | ||
58315c96 | 974 | /* This css_set is dead. Unlink it and release cgroup and css refs */ |
53254f90 | 975 | for_each_subsys(ss, ssid) { |
2d8f243a | 976 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
977 | css_put(cset->subsys[ssid]); |
978 | } | |
5abb8855 | 979 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
980 | css_set_count--; |
981 | ||
69d0206c | 982 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
983 | list_del(&link->cset_link); |
984 | list_del(&link->cgrp_link); | |
2ceb231b TH |
985 | if (cgroup_parent(link->cgrp)) |
986 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 987 | kfree(link); |
81a6a5cd | 988 | } |
2c6ab6d2 | 989 | |
454000ad TH |
990 | if (css_set_threaded(cset)) { |
991 | list_del(&cset->threaded_csets_node); | |
992 | put_css_set_locked(cset->dom_cset); | |
993 | } | |
994 | ||
5abb8855 | 995 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
996 | } |
997 | ||
b326f9d0 | 998 | /** |
7717f7ba | 999 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
1000 | * @cset: candidate css_set being tested |
1001 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
1002 | * @new_cgrp: cgroup that's being entered by the task |
1003 | * @template: desired set of css pointers in css_set (pre-calculated) | |
1004 | * | |
6f4b7e63 | 1005 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
1006 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
1007 | */ | |
5abb8855 TH |
1008 | static bool compare_css_sets(struct css_set *cset, |
1009 | struct css_set *old_cset, | |
7717f7ba PM |
1010 | struct cgroup *new_cgrp, |
1011 | struct cgroup_subsys_state *template[]) | |
1012 | { | |
454000ad | 1013 | struct cgroup *new_dfl_cgrp; |
7717f7ba PM |
1014 | struct list_head *l1, *l2; |
1015 | ||
aec3dfcb TH |
1016 | /* |
1017 | * On the default hierarchy, there can be csets which are | |
1018 | * associated with the same set of cgroups but different csses. | |
1019 | * Let's first ensure that csses match. | |
1020 | */ | |
1021 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 1022 | return false; |
7717f7ba | 1023 | |
454000ad TH |
1024 | |
1025 | /* @cset's domain should match the default cgroup's */ | |
1026 | if (cgroup_on_dfl(new_cgrp)) | |
1027 | new_dfl_cgrp = new_cgrp; | |
1028 | else | |
1029 | new_dfl_cgrp = old_cset->dfl_cgrp; | |
1030 | ||
1031 | if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp) | |
1032 | return false; | |
1033 | ||
7717f7ba PM |
1034 | /* |
1035 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
1036 | * different cgroups in hierarchies. As different cgroups may |
1037 | * share the same effective css, this comparison is always | |
1038 | * necessary. | |
7717f7ba | 1039 | */ |
69d0206c TH |
1040 | l1 = &cset->cgrp_links; |
1041 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 1042 | while (1) { |
69d0206c | 1043 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 1044 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
1045 | |
1046 | l1 = l1->next; | |
1047 | l2 = l2->next; | |
1048 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
1049 | if (l1 == &cset->cgrp_links) { |
1050 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
1051 | break; |
1052 | } else { | |
69d0206c | 1053 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
1054 | } |
1055 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
1056 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
1057 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
1058 | cgrp1 = link1->cgrp; | |
1059 | cgrp2 = link2->cgrp; | |
7717f7ba | 1060 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 1061 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
1062 | |
1063 | /* | |
1064 | * If this hierarchy is the hierarchy of the cgroup | |
1065 | * that's changing, then we need to check that this | |
1066 | * css_set points to the new cgroup; if it's any other | |
1067 | * hierarchy, then this css_set should point to the | |
1068 | * same cgroup as the old css_set. | |
1069 | */ | |
5abb8855 TH |
1070 | if (cgrp1->root == new_cgrp->root) { |
1071 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
1072 | return false; |
1073 | } else { | |
5abb8855 | 1074 | if (cgrp1 != cgrp2) |
7717f7ba PM |
1075 | return false; |
1076 | } | |
1077 | } | |
1078 | return true; | |
1079 | } | |
1080 | ||
b326f9d0 TH |
1081 | /** |
1082 | * find_existing_css_set - init css array and find the matching css_set | |
1083 | * @old_cset: the css_set that we're using before the cgroup transition | |
1084 | * @cgrp: the cgroup that we're moving into | |
1085 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 1086 | */ |
5abb8855 TH |
1087 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
1088 | struct cgroup *cgrp, | |
1089 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 1090 | { |
3dd06ffa | 1091 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 1092 | struct cgroup_subsys *ss; |
5abb8855 | 1093 | struct css_set *cset; |
0ac801fe | 1094 | unsigned long key; |
b326f9d0 | 1095 | int i; |
817929ec | 1096 | |
aae8aab4 BB |
1097 | /* |
1098 | * Build the set of subsystem state objects that we want to see in the | |
58315c96 | 1099 | * new css_set. While subsystems can change globally, the entries here |
aae8aab4 BB |
1100 | * won't change, so no need for locking. |
1101 | */ | |
30159ec7 | 1102 | for_each_subsys(ss, i) { |
f392e51c | 1103 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
1104 | /* |
1105 | * @ss is in this hierarchy, so we want the | |
1106 | * effective css from @cgrp. | |
1107 | */ | |
fc5a828b | 1108 | template[i] = cgroup_e_css_by_mask(cgrp, ss); |
817929ec | 1109 | } else { |
aec3dfcb TH |
1110 | /* |
1111 | * @ss is not in this hierarchy, so we don't want | |
1112 | * to change the css. | |
1113 | */ | |
5abb8855 | 1114 | template[i] = old_cset->subsys[i]; |
817929ec PM |
1115 | } |
1116 | } | |
1117 | ||
0ac801fe | 1118 | key = css_set_hash(template); |
5abb8855 TH |
1119 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
1120 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
1121 | continue; |
1122 | ||
1123 | /* This css_set matches what we need */ | |
5abb8855 | 1124 | return cset; |
472b1053 | 1125 | } |
817929ec PM |
1126 | |
1127 | /* No existing cgroup group matched */ | |
1128 | return NULL; | |
1129 | } | |
1130 | ||
69d0206c | 1131 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 1132 | { |
69d0206c | 1133 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 1134 | |
69d0206c TH |
1135 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
1136 | list_del(&link->cset_link); | |
36553434 LZ |
1137 | kfree(link); |
1138 | } | |
1139 | } | |
1140 | ||
69d0206c TH |
1141 | /** |
1142 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
1143 | * @count: the number of links to allocate | |
1144 | * @tmp_links: list_head the allocated links are put on | |
1145 | * | |
1146 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1147 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1148 | */ |
69d0206c | 1149 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1150 | { |
69d0206c | 1151 | struct cgrp_cset_link *link; |
817929ec | 1152 | int i; |
69d0206c TH |
1153 | |
1154 | INIT_LIST_HEAD(tmp_links); | |
1155 | ||
817929ec | 1156 | for (i = 0; i < count; i++) { |
f4f4be2b | 1157 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1158 | if (!link) { |
69d0206c | 1159 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1160 | return -ENOMEM; |
1161 | } | |
69d0206c | 1162 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1163 | } |
1164 | return 0; | |
1165 | } | |
1166 | ||
c12f65d4 LZ |
1167 | /** |
1168 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1169 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1170 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1171 | * @cgrp: the destination cgroup |
1172 | */ | |
69d0206c TH |
1173 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1174 | struct cgroup *cgrp) | |
c12f65d4 | 1175 | { |
69d0206c | 1176 | struct cgrp_cset_link *link; |
c12f65d4 | 1177 | |
69d0206c | 1178 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1179 | |
1180 | if (cgroup_on_dfl(cgrp)) | |
1181 | cset->dfl_cgrp = cgrp; | |
1182 | ||
69d0206c TH |
1183 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1184 | link->cset = cset; | |
7717f7ba | 1185 | link->cgrp = cgrp; |
842b597e | 1186 | |
7717f7ba | 1187 | /* |
389b9c1b | 1188 | * Always add links to the tail of the lists so that the lists are |
58315c96 | 1189 | * in chronological order. |
7717f7ba | 1190 | */ |
389b9c1b | 1191 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1192 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1193 | |
1194 | if (cgroup_parent(cgrp)) | |
a590b90d | 1195 | cgroup_get_live(cgrp); |
c12f65d4 LZ |
1196 | } |
1197 | ||
b326f9d0 TH |
1198 | /** |
1199 | * find_css_set - return a new css_set with one cgroup updated | |
1200 | * @old_cset: the baseline css_set | |
1201 | * @cgrp: the cgroup to be updated | |
1202 | * | |
1203 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1204 | * substituted into the appropriate hierarchy. | |
817929ec | 1205 | */ |
5abb8855 TH |
1206 | static struct css_set *find_css_set(struct css_set *old_cset, |
1207 | struct cgroup *cgrp) | |
817929ec | 1208 | { |
b326f9d0 | 1209 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1210 | struct css_set *cset; |
69d0206c TH |
1211 | struct list_head tmp_links; |
1212 | struct cgrp_cset_link *link; | |
2d8f243a | 1213 | struct cgroup_subsys *ss; |
0ac801fe | 1214 | unsigned long key; |
2d8f243a | 1215 | int ssid; |
472b1053 | 1216 | |
b326f9d0 TH |
1217 | lockdep_assert_held(&cgroup_mutex); |
1218 | ||
817929ec PM |
1219 | /* First see if we already have a cgroup group that matches |
1220 | * the desired set */ | |
82d6489d | 1221 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1222 | cset = find_existing_css_set(old_cset, cgrp, template); |
1223 | if (cset) | |
1224 | get_css_set(cset); | |
82d6489d | 1225 | spin_unlock_irq(&css_set_lock); |
817929ec | 1226 | |
5abb8855 TH |
1227 | if (cset) |
1228 | return cset; | |
817929ec | 1229 | |
f4f4be2b | 1230 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1231 | if (!cset) |
817929ec PM |
1232 | return NULL; |
1233 | ||
69d0206c | 1234 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1235 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1236 | kfree(cset); |
817929ec PM |
1237 | return NULL; |
1238 | } | |
1239 | ||
4b9502e6 | 1240 | refcount_set(&cset->refcount, 1); |
454000ad | 1241 | cset->dom_cset = cset; |
5abb8855 | 1242 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1243 | INIT_LIST_HEAD(&cset->mg_tasks); |
c03cd773 | 1244 | INIT_LIST_HEAD(&cset->dying_tasks); |
ed27b9f7 | 1245 | INIT_LIST_HEAD(&cset->task_iters); |
454000ad | 1246 | INIT_LIST_HEAD(&cset->threaded_csets); |
5abb8855 | 1247 | INIT_HLIST_NODE(&cset->hlist); |
5f617ebb | 1248 | INIT_LIST_HEAD(&cset->cgrp_links); |
07fd5b6c TH |
1249 | INIT_LIST_HEAD(&cset->mg_src_preload_node); |
1250 | INIT_LIST_HEAD(&cset->mg_dst_preload_node); | |
5f617ebb | 1251 | INIT_LIST_HEAD(&cset->mg_node); |
817929ec PM |
1252 | |
1253 | /* Copy the set of subsystem state objects generated in | |
1254 | * find_existing_css_set() */ | |
5abb8855 | 1255 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1256 | |
82d6489d | 1257 | spin_lock_irq(&css_set_lock); |
817929ec | 1258 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1259 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1260 | struct cgroup *c = link->cgrp; |
69d0206c | 1261 | |
7717f7ba PM |
1262 | if (c->root == cgrp->root) |
1263 | c = cgrp; | |
69d0206c | 1264 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1265 | } |
817929ec | 1266 | |
69d0206c | 1267 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1268 | |
817929ec | 1269 | css_set_count++; |
472b1053 | 1270 | |
2d8f243a | 1271 | /* Add @cset to the hash table */ |
5abb8855 TH |
1272 | key = css_set_hash(cset->subsys); |
1273 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1274 | |
53254f90 TH |
1275 | for_each_subsys(ss, ssid) { |
1276 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1277 | ||
2d8f243a | 1278 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1279 | &css->cgroup->e_csets[ssid]); |
1280 | css_get(css); | |
1281 | } | |
2d8f243a | 1282 | |
82d6489d | 1283 | spin_unlock_irq(&css_set_lock); |
817929ec | 1284 | |
454000ad TH |
1285 | /* |
1286 | * If @cset should be threaded, look up the matching dom_cset and | |
1287 | * link them up. We first fully initialize @cset then look for the | |
1288 | * dom_cset. It's simpler this way and safe as @cset is guaranteed | |
1289 | * to stay empty until we return. | |
1290 | */ | |
1291 | if (cgroup_is_threaded(cset->dfl_cgrp)) { | |
1292 | struct css_set *dcset; | |
1293 | ||
1294 | dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp); | |
1295 | if (!dcset) { | |
1296 | put_css_set(cset); | |
1297 | return NULL; | |
1298 | } | |
1299 | ||
1300 | spin_lock_irq(&css_set_lock); | |
1301 | cset->dom_cset = dcset; | |
1302 | list_add_tail(&cset->threaded_csets_node, | |
1303 | &dcset->threaded_csets); | |
1304 | spin_unlock_irq(&css_set_lock); | |
1305 | } | |
1306 | ||
5abb8855 | 1307 | return cset; |
b4f48b63 PM |
1308 | } |
1309 | ||
0a268dbd | 1310 | struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1311 | { |
f2eb478f | 1312 | struct cgroup *root_cgrp = kernfs_root_to_node(kf_root)->priv; |
2bd59d48 | 1313 | |
3dd06ffa | 1314 | return root_cgrp->root; |
2bd59d48 TH |
1315 | } |
1316 | ||
6a010a49 TH |
1317 | void cgroup_favor_dynmods(struct cgroup_root *root, bool favor) |
1318 | { | |
1319 | bool favoring = root->flags & CGRP_ROOT_FAVOR_DYNMODS; | |
1320 | ||
1321 | /* see the comment above CGRP_ROOT_FAVOR_DYNMODS definition */ | |
1322 | if (favor && !favoring) { | |
1323 | rcu_sync_enter(&cgroup_threadgroup_rwsem.rss); | |
1324 | root->flags |= CGRP_ROOT_FAVOR_DYNMODS; | |
1325 | } else if (!favor && favoring) { | |
1326 | rcu_sync_exit(&cgroup_threadgroup_rwsem.rss); | |
1327 | root->flags &= ~CGRP_ROOT_FAVOR_DYNMODS; | |
1328 | } | |
1329 | } | |
1330 | ||
3dd06ffa | 1331 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1332 | { |
1333 | int id; | |
1334 | ||
1335 | lockdep_assert_held(&cgroup_mutex); | |
1336 | ||
985ed670 | 1337 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1338 | if (id < 0) |
1339 | return id; | |
1340 | ||
1341 | root->hierarchy_id = id; | |
1342 | return 0; | |
1343 | } | |
1344 | ||
3dd06ffa | 1345 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1346 | { |
1347 | lockdep_assert_held(&cgroup_mutex); | |
1348 | ||
8c8a5502 | 1349 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1350 | } |
1351 | ||
1592c9b2 | 1352 | void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 | 1353 | { |
74321038 | 1354 | kfree(root); |
f2e85d57 TH |
1355 | } |
1356 | ||
3dd06ffa | 1357 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1358 | { |
3dd06ffa | 1359 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1360 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1361 | |
ed1777de TH |
1362 | trace_cgroup_destroy_root(root); |
1363 | ||
334c3679 | 1364 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1365 | |
776f02fa | 1366 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1367 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1368 | |
f2e85d57 | 1369 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1370 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1371 | |
7717f7ba | 1372 | /* |
f2e85d57 TH |
1373 | * Release all the links from cset_links to this hierarchy's |
1374 | * root cgroup | |
7717f7ba | 1375 | */ |
82d6489d | 1376 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1377 | |
1378 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1379 | list_del(&link->cset_link); | |
1380 | list_del(&link->cgrp_link); | |
1381 | kfree(link); | |
1382 | } | |
f0d9a5f1 | 1383 | |
82d6489d | 1384 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1385 | |
1386 | if (!list_empty(&root->root_list)) { | |
1387 | list_del(&root->root_list); | |
1388 | cgroup_root_count--; | |
1389 | } | |
1390 | ||
6a010a49 | 1391 | cgroup_favor_dynmods(root, false); |
f2e85d57 TH |
1392 | cgroup_exit_root_id(root); |
1393 | ||
1394 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1395 | |
a7df69b8 | 1396 | cgroup_rstat_exit(cgrp); |
2bd59d48 | 1397 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1398 | cgroup_free_root(root); |
1399 | } | |
1400 | ||
46307fd6 MK |
1401 | /* |
1402 | * Returned cgroup is without refcount but it's valid as long as cset pins it. | |
1403 | */ | |
e210a89f LF |
1404 | static inline struct cgroup *__cset_cgroup_from_root(struct css_set *cset, |
1405 | struct cgroup_root *root) | |
1406 | { | |
1407 | struct cgroup *res_cgroup = NULL; | |
1408 | ||
1409 | if (cset == &init_css_set) { | |
1410 | res_cgroup = &root->cgrp; | |
1411 | } else if (root == &cgrp_dfl_root) { | |
1412 | res_cgroup = cset->dfl_cgrp; | |
1413 | } else { | |
1414 | struct cgrp_cset_link *link; | |
46307fd6 | 1415 | lockdep_assert_held(&css_set_lock); |
e210a89f LF |
1416 | |
1417 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1418 | struct cgroup *c = link->cgrp; | |
1419 | ||
1420 | if (c->root == root) { | |
1421 | res_cgroup = c; | |
1422 | break; | |
1423 | } | |
1424 | } | |
1425 | } | |
1426 | ||
46307fd6 | 1427 | BUG_ON(!res_cgroup); |
e210a89f LF |
1428 | return res_cgroup; |
1429 | } | |
1430 | ||
4f41fc59 SH |
1431 | /* |
1432 | * look up cgroup associated with current task's cgroup namespace on the | |
1433 | * specified hierarchy | |
1434 | */ | |
1435 | static struct cgroup * | |
1436 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1437 | { | |
1438 | struct cgroup *res = NULL; | |
1439 | struct css_set *cset; | |
1440 | ||
1441 | lockdep_assert_held(&css_set_lock); | |
1442 | ||
1443 | rcu_read_lock(); | |
1444 | ||
1445 | cset = current->nsproxy->cgroup_ns->root_cset; | |
e210a89f | 1446 | res = __cset_cgroup_from_root(cset, root); |
4f41fc59 | 1447 | |
4f41fc59 SH |
1448 | rcu_read_unlock(); |
1449 | ||
4f41fc59 SH |
1450 | return res; |
1451 | } | |
1452 | ||
46307fd6 MK |
1453 | /* |
1454 | * Look up cgroup associated with current task's cgroup namespace on the default | |
1455 | * hierarchy. | |
1456 | * | |
1457 | * Unlike current_cgns_cgroup_from_root(), this doesn't need locks: | |
1458 | * - Internal rcu_read_lock is unnecessary because we don't dereference any rcu | |
1459 | * pointers. | |
1460 | * - css_set_lock is not needed because we just read cset->dfl_cgrp. | |
1461 | * - As a bonus returned cgrp is pinned with the current because it cannot | |
1462 | * switch cgroup_ns asynchronously. | |
1463 | */ | |
1464 | static struct cgroup *current_cgns_cgroup_dfl(void) | |
1465 | { | |
1466 | struct css_set *cset; | |
1467 | ||
1468 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1469 | return __cset_cgroup_from_root(cset, &cgrp_dfl_root); | |
1470 | } | |
1471 | ||
ceb6a081 TH |
1472 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1473 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1474 | struct cgroup_root *root) |
7717f7ba | 1475 | { |
96d365e0 | 1476 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1477 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1478 | |
46307fd6 | 1479 | return __cset_cgroup_from_root(cset, root); |
7717f7ba PM |
1480 | } |
1481 | ||
ddbcc7e8 | 1482 | /* |
ceb6a081 | 1483 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1484 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 | 1485 | */ |
0a268dbd TH |
1486 | struct cgroup *task_cgroup_from_root(struct task_struct *task, |
1487 | struct cgroup_root *root) | |
ceb6a081 TH |
1488 | { |
1489 | /* | |
e7c7b1d8 MK |
1490 | * No need to lock the task - since we hold css_set_lock the |
1491 | * task can't change groups. | |
ceb6a081 TH |
1492 | */ |
1493 | return cset_cgroup_from_root(task_css_set(task), root); | |
1494 | } | |
1495 | ||
ddbcc7e8 | 1496 | /* |
ddbcc7e8 PM |
1497 | * A task must hold cgroup_mutex to modify cgroups. |
1498 | * | |
1499 | * Any task can increment and decrement the count field without lock. | |
1500 | * So in general, code holding cgroup_mutex can't rely on the count | |
1501 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1502 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1503 | * means that no tasks are currently attached, therefore there is no |
1504 | * way a task attached to that cgroup can fork (the other way to | |
1505 | * increment the count). So code holding cgroup_mutex can safely | |
1506 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1507 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1508 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1509 | * needs that mutex. | |
1510 | * | |
ddbcc7e8 PM |
1511 | * A cgroup can only be deleted if both its 'count' of using tasks |
1512 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1513 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1514 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1515 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1516 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1517 | * |
1518 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1519 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1520 | */ |
1521 | ||
2bd59d48 | 1522 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
a424316c | 1523 | |
cf892988 JA |
1524 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1525 | char *buf) | |
ddbcc7e8 | 1526 | { |
3e1d2eed TH |
1527 | struct cgroup_subsys *ss = cft->ss; |
1528 | ||
8d7e6fb0 | 1529 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
c1bbd933 TH |
1530 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) { |
1531 | const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : ""; | |
1532 | ||
1533 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s", | |
1534 | dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, | |
cf892988 | 1535 | cft->name); |
c1bbd933 | 1536 | } else { |
cf892988 | 1537 | strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX); |
c1bbd933 | 1538 | } |
8d7e6fb0 | 1539 | return buf; |
ddbcc7e8 PM |
1540 | } |
1541 | ||
f2e85d57 TH |
1542 | /** |
1543 | * cgroup_file_mode - deduce file mode of a control file | |
1544 | * @cft: the control file in question | |
1545 | * | |
7dbdb199 | 1546 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1547 | */ |
1548 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1549 | { |
f2e85d57 | 1550 | umode_t mode = 0; |
65dff759 | 1551 | |
f2e85d57 TH |
1552 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1553 | mode |= S_IRUGO; | |
1554 | ||
7dbdb199 TH |
1555 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1556 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1557 | mode |= S_IWUGO; | |
1558 | else | |
1559 | mode |= S_IWUSR; | |
1560 | } | |
f2e85d57 TH |
1561 | |
1562 | return mode; | |
65dff759 LZ |
1563 | } |
1564 | ||
af0ba678 | 1565 | /** |
8699b776 | 1566 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1567 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1568 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1569 | * |
1570 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1571 | * enabled together through its ->depends_on mask. In such cases, more | |
1572 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1573 | * | |
0f060deb | 1574 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1575 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1576 | */ |
5ced2518 | 1577 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1578 | { |
6e5c8307 | 1579 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1580 | struct cgroup_subsys *ss; |
1581 | int ssid; | |
1582 | ||
1583 | lockdep_assert_held(&cgroup_mutex); | |
1584 | ||
f6d635ad TH |
1585 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1586 | ||
af0ba678 | 1587 | while (true) { |
6e5c8307 | 1588 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1589 | |
b4e0eeaf | 1590 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1591 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1592 | } while_each_subsys_mask(); |
af0ba678 TH |
1593 | |
1594 | /* | |
1595 | * Mask out subsystems which aren't available. This can | |
1596 | * happen only if some depended-upon subsystems were bound | |
1597 | * to non-default hierarchies. | |
1598 | */ | |
5ced2518 | 1599 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1600 | |
1601 | if (new_ss_mask == cur_ss_mask) | |
1602 | break; | |
1603 | cur_ss_mask = new_ss_mask; | |
1604 | } | |
1605 | ||
0f060deb TH |
1606 | return cur_ss_mask; |
1607 | } | |
1608 | ||
a9746d8d TH |
1609 | /** |
1610 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1611 | * @kn: the kernfs_node being serviced | |
1612 | * | |
1613 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1614 | * the method finishes if locking succeeded. Note that once this function | |
1615 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1616 | * inaccessible any time. If the caller intends to continue to access the | |
1617 | * cgroup, it should pin it before invoking this function. | |
1618 | */ | |
0a268dbd | 1619 | void cgroup_kn_unlock(struct kernfs_node *kn) |
ddbcc7e8 | 1620 | { |
a9746d8d TH |
1621 | struct cgroup *cgrp; |
1622 | ||
1623 | if (kernfs_type(kn) == KERNFS_DIR) | |
1624 | cgrp = kn->priv; | |
1625 | else | |
1626 | cgrp = kn->parent->priv; | |
1627 | ||
1628 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1629 | |
1630 | kernfs_unbreak_active_protection(kn); | |
1631 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1632 | } |
1633 | ||
a9746d8d TH |
1634 | /** |
1635 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1636 | * @kn: the kernfs_node being serviced | |
945ba199 | 1637 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1638 | * |
1639 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1640 | * @kn. It breaks the active protection, performs cgroup locking and | |
1641 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1642 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1643 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1644 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1645 | * |
1646 | * Any cgroup kernfs method implementation which requires locking the | |
1647 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1648 | * locking under kernfs active protection and allows all kernfs operations | |
1649 | * including self-removal. | |
1650 | */ | |
0a268dbd | 1651 | struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline) |
05ef1d7c | 1652 | { |
a9746d8d TH |
1653 | struct cgroup *cgrp; |
1654 | ||
1655 | if (kernfs_type(kn) == KERNFS_DIR) | |
1656 | cgrp = kn->priv; | |
1657 | else | |
1658 | cgrp = kn->parent->priv; | |
05ef1d7c | 1659 | |
2739d3cc | 1660 | /* |
01f6474c | 1661 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1662 | * active_ref. cgroup liveliness check alone provides enough |
1663 | * protection against removal. Ensure @cgrp stays accessible and | |
1664 | * break the active_ref protection. | |
2739d3cc | 1665 | */ |
aa32362f LZ |
1666 | if (!cgroup_tryget(cgrp)) |
1667 | return NULL; | |
a9746d8d TH |
1668 | kernfs_break_active_protection(kn); |
1669 | ||
945ba199 TH |
1670 | if (drain_offline) |
1671 | cgroup_lock_and_drain_offline(cgrp); | |
1672 | else | |
1673 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1674 | |
a9746d8d TH |
1675 | if (!cgroup_is_dead(cgrp)) |
1676 | return cgrp; | |
1677 | ||
1678 | cgroup_kn_unlock(kn); | |
1679 | return NULL; | |
ddbcc7e8 | 1680 | } |
05ef1d7c | 1681 | |
2739d3cc | 1682 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1683 | { |
2bd59d48 | 1684 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1685 | |
01f6474c | 1686 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1687 | |
1688 | if (cft->file_offset) { | |
1689 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1690 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1691 | ||
1692 | spin_lock_irq(&cgroup_file_kn_lock); | |
1693 | cfile->kn = NULL; | |
1694 | spin_unlock_irq(&cgroup_file_kn_lock); | |
b12e3583 TH |
1695 | |
1696 | del_timer_sync(&cfile->notify_timer); | |
34c06254 TH |
1697 | } |
1698 | ||
2bd59d48 | 1699 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1700 | } |
1701 | ||
13af07df | 1702 | /** |
4df8dc90 | 1703 | * css_clear_dir - remove subsys files in a cgroup directory |
08b2b6fd | 1704 | * @css: target css |
13af07df | 1705 | */ |
334c3679 | 1706 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1707 | { |
334c3679 | 1708 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1709 | struct cftype *cfts; |
05ef1d7c | 1710 | |
88cb04b9 TH |
1711 | if (!(css->flags & CSS_VISIBLE)) |
1712 | return; | |
1713 | ||
1714 | css->flags &= ~CSS_VISIBLE; | |
1715 | ||
5faaf05f | 1716 | if (!css->ss) { |
8a693f77 TH |
1717 | if (cgroup_on_dfl(cgrp)) { |
1718 | cgroup_addrm_files(css, cgrp, | |
1719 | cgroup_base_files, false); | |
1720 | if (cgroup_psi_enabled()) | |
1721 | cgroup_addrm_files(css, cgrp, | |
1722 | cgroup_psi_files, false); | |
1723 | } else { | |
1724 | cgroup_addrm_files(css, cgrp, | |
1725 | cgroup1_base_files, false); | |
1726 | } | |
5faaf05f TH |
1727 | } else { |
1728 | list_for_each_entry(cfts, &css->ss->cfts, node) | |
1729 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1730 | } | |
ddbcc7e8 PM |
1731 | } |
1732 | ||
ccdca218 | 1733 | /** |
4df8dc90 TH |
1734 | * css_populate_dir - create subsys files in a cgroup directory |
1735 | * @css: target css | |
ccdca218 TH |
1736 | * |
1737 | * On failure, no file is added. | |
1738 | */ | |
334c3679 | 1739 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1740 | { |
334c3679 | 1741 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1742 | struct cftype *cfts, *failed_cfts; |
1743 | int ret; | |
ccdca218 | 1744 | |
03970d3c | 1745 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1746 | return 0; |
1747 | ||
4df8dc90 | 1748 | if (!css->ss) { |
8a693f77 TH |
1749 | if (cgroup_on_dfl(cgrp)) { |
1750 | ret = cgroup_addrm_files(&cgrp->self, cgrp, | |
1751 | cgroup_base_files, true); | |
1752 | if (ret < 0) | |
1753 | return ret; | |
1754 | ||
1755 | if (cgroup_psi_enabled()) { | |
1756 | ret = cgroup_addrm_files(&cgrp->self, cgrp, | |
1757 | cgroup_psi_files, true); | |
1758 | if (ret < 0) | |
1759 | return ret; | |
1760 | } | |
1761 | } else { | |
1762 | cgroup_addrm_files(css, cgrp, | |
1763 | cgroup1_base_files, true); | |
1764 | } | |
5faaf05f TH |
1765 | } else { |
1766 | list_for_each_entry(cfts, &css->ss->cfts, node) { | |
1767 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1768 | if (ret < 0) { | |
1769 | failed_cfts = cfts; | |
1770 | goto err; | |
1771 | } | |
ccdca218 TH |
1772 | } |
1773 | } | |
88cb04b9 TH |
1774 | |
1775 | css->flags |= CSS_VISIBLE; | |
1776 | ||
ccdca218 TH |
1777 | return 0; |
1778 | err: | |
4df8dc90 TH |
1779 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1780 | if (cfts == failed_cfts) | |
1781 | break; | |
1782 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1783 | } | |
ccdca218 TH |
1784 | return ret; |
1785 | } | |
1786 | ||
0a268dbd | 1787 | int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1788 | { |
1ada4838 | 1789 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1790 | struct cgroup_subsys *ss; |
2d8f243a | 1791 | int ssid, i, ret; |
7ee28539 | 1792 | u16 dfl_disable_ss_mask = 0; |
ddbcc7e8 | 1793 | |
ace2bee8 | 1794 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1795 | |
b4e0eeaf | 1796 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1797 | /* |
1798 | * If @ss has non-root csses attached to it, can't move. | |
1799 | * If @ss is an implicit controller, it is exempt from this | |
1800 | * rule and can be stolen. | |
1801 | */ | |
1802 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1803 | !ss->implicit_on_dfl) | |
3ed80a62 | 1804 | return -EBUSY; |
1d5be6b2 | 1805 | |
5df36032 | 1806 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1807 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1808 | return -EBUSY; |
7ee28539 WL |
1809 | |
1810 | /* | |
1811 | * Collect ssid's that need to be disabled from default | |
1812 | * hierarchy. | |
1813 | */ | |
1814 | if (ss->root == &cgrp_dfl_root) | |
1815 | dfl_disable_ss_mask |= 1 << ssid; | |
1816 | ||
b4e0eeaf | 1817 | } while_each_subsys_mask(); |
ddbcc7e8 | 1818 | |
7ee28539 WL |
1819 | if (dfl_disable_ss_mask) { |
1820 | struct cgroup *scgrp = &cgrp_dfl_root.cgrp; | |
1821 | ||
1822 | /* | |
1823 | * Controllers from default hierarchy that need to be rebound | |
1824 | * are all disabled together in one go. | |
1825 | */ | |
1826 | cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask; | |
1827 | WARN_ON(cgroup_apply_control(scgrp)); | |
1828 | cgroup_finalize_control(scgrp, 0); | |
1829 | } | |
1830 | ||
b4e0eeaf | 1831 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1832 | struct cgroup_root *src_root = ss->root; |
1833 | struct cgroup *scgrp = &src_root->cgrp; | |
1834 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1835 | struct css_set *cset; |
a8a648c4 | 1836 | |
1ada4838 | 1837 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1838 | |
7ee28539 WL |
1839 | if (src_root != &cgrp_dfl_root) { |
1840 | /* disable from the source */ | |
1841 | src_root->subsys_mask &= ~(1 << ssid); | |
1842 | WARN_ON(cgroup_apply_control(scgrp)); | |
1843 | cgroup_finalize_control(scgrp, 0); | |
1844 | } | |
4df8dc90 | 1845 | |
334c3679 | 1846 | /* rebind */ |
1ada4838 TH |
1847 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1848 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1849 | ss->root = dst_root; |
1ada4838 | 1850 | css->cgroup = dcgrp; |
73e80ed8 | 1851 | |
82d6489d | 1852 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1853 | hash_for_each(css_set_table, i, cset, hlist) |
1854 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1855 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1856 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1857 | |
a7df69b8 JW |
1858 | if (ss->css_rstat_flush) { |
1859 | list_del_rcu(&css->rstat_css_node); | |
763f4fb7 | 1860 | synchronize_rcu(); |
a7df69b8 JW |
1861 | list_add_rcu(&css->rstat_css_node, |
1862 | &dcgrp->rstat_css_list); | |
1863 | } | |
1864 | ||
bd53d617 | 1865 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1866 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1867 | if (dst_root == &cgrp_dfl_root) { |
1868 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1869 | } else { | |
1ada4838 | 1870 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1871 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1872 | } |
a8a648c4 | 1873 | |
334c3679 TH |
1874 | ret = cgroup_apply_control(dcgrp); |
1875 | if (ret) | |
1876 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1877 | ss->name, ret); | |
1878 | ||
5df36032 TH |
1879 | if (ss->bind) |
1880 | ss->bind(css); | |
b4e0eeaf | 1881 | } while_each_subsys_mask(); |
ddbcc7e8 | 1882 | |
1ada4838 | 1883 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1884 | return 0; |
1885 | } | |
1886 | ||
1592c9b2 TH |
1887 | int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1888 | struct kernfs_root *kf_root) | |
4f41fc59 | 1889 | { |
09be4c82 | 1890 | int len = 0; |
4f41fc59 SH |
1891 | char *buf = NULL; |
1892 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1893 | struct cgroup *ns_cgroup; | |
1894 | ||
1895 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1896 | if (!buf) | |
1897 | return -ENOMEM; | |
1898 | ||
82d6489d | 1899 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1900 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1901 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1902 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1903 | |
1904 | if (len >= PATH_MAX) | |
1905 | len = -ERANGE; | |
1906 | else if (len > 0) { | |
1907 | seq_escape(sf, buf, " \t\n\\"); | |
1908 | len = 0; | |
1909 | } | |
1910 | kfree(buf); | |
1911 | return len; | |
1912 | } | |
1913 | ||
e34a98d5 AV |
1914 | enum cgroup2_param { |
1915 | Opt_nsdelegate, | |
c808f463 | 1916 | Opt_favordynmods, |
9852ae3f | 1917 | Opt_memory_localevents, |
8a931f80 | 1918 | Opt_memory_recursiveprot, |
e34a98d5 AV |
1919 | nr__cgroup2_params |
1920 | }; | |
5136f636 | 1921 | |
d7167b14 | 1922 | static const struct fs_parameter_spec cgroup2_fs_parameters[] = { |
9852ae3f | 1923 | fsparam_flag("nsdelegate", Opt_nsdelegate), |
6a010a49 | 1924 | fsparam_flag("favordynmods", Opt_favordynmods), |
9852ae3f | 1925 | fsparam_flag("memory_localevents", Opt_memory_localevents), |
8a931f80 | 1926 | fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot), |
e34a98d5 AV |
1927 | {} |
1928 | }; | |
5136f636 | 1929 | |
e34a98d5 AV |
1930 | static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param) |
1931 | { | |
1932 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); | |
1933 | struct fs_parse_result result; | |
1934 | int opt; | |
5136f636 | 1935 | |
d7167b14 | 1936 | opt = fs_parse(fc, cgroup2_fs_parameters, param, &result); |
e34a98d5 AV |
1937 | if (opt < 0) |
1938 | return opt; | |
5136f636 | 1939 | |
e34a98d5 AV |
1940 | switch (opt) { |
1941 | case Opt_nsdelegate: | |
1942 | ctx->flags |= CGRP_ROOT_NS_DELEGATE; | |
1943 | return 0; | |
6a010a49 TH |
1944 | case Opt_favordynmods: |
1945 | ctx->flags |= CGRP_ROOT_FAVOR_DYNMODS; | |
1946 | return 0; | |
9852ae3f CD |
1947 | case Opt_memory_localevents: |
1948 | ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
1949 | return 0; | |
8a931f80 JW |
1950 | case Opt_memory_recursiveprot: |
1951 | ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT; | |
1952 | return 0; | |
e34a98d5 AV |
1953 | } |
1954 | return -EINVAL; | |
5136f636 TH |
1955 | } |
1956 | ||
1957 | static void apply_cgroup_root_flags(unsigned int root_flags) | |
1958 | { | |
1959 | if (current->nsproxy->cgroup_ns == &init_cgroup_ns) { | |
1960 | if (root_flags & CGRP_ROOT_NS_DELEGATE) | |
1961 | cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE; | |
1962 | else | |
1963 | cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE; | |
9852ae3f | 1964 | |
6a010a49 TH |
1965 | cgroup_favor_dynmods(&cgrp_dfl_root, |
1966 | root_flags & CGRP_ROOT_FAVOR_DYNMODS); | |
1967 | ||
9852ae3f CD |
1968 | if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) |
1969 | cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
1970 | else | |
1971 | cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
8a931f80 JW |
1972 | |
1973 | if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT) | |
1974 | cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT; | |
1975 | else | |
1976 | cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT; | |
5136f636 TH |
1977 | } |
1978 | } | |
1979 | ||
1980 | static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root) | |
1981 | { | |
1982 | if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) | |
1983 | seq_puts(seq, ",nsdelegate"); | |
6a010a49 TH |
1984 | if (cgrp_dfl_root.flags & CGRP_ROOT_FAVOR_DYNMODS) |
1985 | seq_puts(seq, ",favordynmods"); | |
9852ae3f CD |
1986 | if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) |
1987 | seq_puts(seq, ",memory_localevents"); | |
8a931f80 JW |
1988 | if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT) |
1989 | seq_puts(seq, ",memory_recursiveprot"); | |
5136f636 TH |
1990 | return 0; |
1991 | } | |
1992 | ||
90129625 | 1993 | static int cgroup_reconfigure(struct fs_context *fc) |
ddbcc7e8 | 1994 | { |
90129625 | 1995 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
5136f636 | 1996 | |
f5dfb531 | 1997 | apply_cgroup_root_flags(ctx->flags); |
5136f636 | 1998 | return 0; |
ddbcc7e8 PM |
1999 | } |
2000 | ||
cc31edce PM |
2001 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
2002 | { | |
2d8f243a TH |
2003 | struct cgroup_subsys *ss; |
2004 | int ssid; | |
2005 | ||
d5c419b6 TH |
2006 | INIT_LIST_HEAD(&cgrp->self.sibling); |
2007 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 2008 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
2009 | INIT_LIST_HEAD(&cgrp->pidlists); |
2010 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 2011 | cgrp->self.cgroup = cgrp; |
184faf32 | 2012 | cgrp->self.flags |= CSS_ONLINE; |
454000ad | 2013 | cgrp->dom_cgrp = cgrp; |
1a926e0b RG |
2014 | cgrp->max_descendants = INT_MAX; |
2015 | cgrp->max_depth = INT_MAX; | |
8f53470b | 2016 | INIT_LIST_HEAD(&cgrp->rstat_css_list); |
d4ff749b | 2017 | prev_cputime_init(&cgrp->prev_cputime); |
2d8f243a TH |
2018 | |
2019 | for_each_subsys(ss, ssid) | |
2020 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
2021 | |
2022 | init_waitqueue_head(&cgrp->offline_waitq); | |
d62beb7f | 2023 | INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent); |
cc31edce | 2024 | } |
c6d57f33 | 2025 | |
cf6299b1 | 2026 | void init_cgroup_root(struct cgroup_fs_context *ctx) |
ddbcc7e8 | 2027 | { |
cf6299b1 | 2028 | struct cgroup_root *root = ctx->root; |
3dd06ffa | 2029 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 2030 | |
ddbcc7e8 | 2031 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 2032 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 2033 | cgrp->root = root; |
cc31edce | 2034 | init_cgroup_housekeeping(cgrp); |
c6d57f33 | 2035 | |
6a010a49 TH |
2036 | /* DYNMODS must be modified through cgroup_favor_dynmods() */ |
2037 | root->flags = ctx->flags & ~CGRP_ROOT_FAVOR_DYNMODS; | |
f5dfb531 AV |
2038 | if (ctx->release_agent) |
2039 | strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX); | |
2040 | if (ctx->name) | |
2041 | strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN); | |
2042 | if (ctx->cpuset_clone_children) | |
3dd06ffa | 2043 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
2044 | } |
2045 | ||
35ac1184 | 2046 | int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 2047 | { |
d427dfeb | 2048 | LIST_HEAD(tmp_links); |
3dd06ffa | 2049 | struct cgroup *root_cgrp = &root->cgrp; |
fa069904 | 2050 | struct kernfs_syscall_ops *kf_sops; |
d427dfeb | 2051 | struct css_set *cset; |
d427dfeb | 2052 | int i, ret; |
2c6ab6d2 | 2053 | |
d427dfeb | 2054 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 2055 | |
9732adc5 | 2056 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, |
35ac1184 | 2057 | 0, GFP_KERNEL); |
9d755d33 TH |
2058 | if (ret) |
2059 | goto out; | |
2060 | ||
d427dfeb | 2061 | /* |
f0d9a5f1 | 2062 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 2063 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
2064 | * cgroup_lock, and that's us. Later rebinding may disable |
2065 | * controllers on the default hierarchy and thus create new csets, | |
2066 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 2067 | */ |
04313591 | 2068 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 2069 | if (ret) |
9d755d33 | 2070 | goto cancel_ref; |
ddbcc7e8 | 2071 | |
985ed670 | 2072 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 2073 | if (ret) |
9d755d33 | 2074 | goto cancel_ref; |
ddbcc7e8 | 2075 | |
fa069904 TH |
2076 | kf_sops = root == &cgrp_dfl_root ? |
2077 | &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops; | |
2078 | ||
2079 | root->kf_root = kernfs_create_root(kf_sops, | |
aa818825 | 2080 | KERNFS_ROOT_CREATE_DEACTIVATED | |
38aca307 DX |
2081 | KERNFS_ROOT_SUPPORT_EXPORTOP | |
2082 | KERNFS_ROOT_SUPPORT_USER_XATTR, | |
2bd59d48 TH |
2083 | root_cgrp); |
2084 | if (IS_ERR(root->kf_root)) { | |
2085 | ret = PTR_ERR(root->kf_root); | |
2086 | goto exit_root_id; | |
2087 | } | |
f2eb478f | 2088 | root_cgrp->kn = kernfs_root_to_node(root->kf_root); |
d7495343 | 2089 | WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1); |
7f203bc8 | 2090 | root_cgrp->ancestors[0] = root_cgrp; |
ddbcc7e8 | 2091 | |
334c3679 | 2092 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 2093 | if (ret) |
2bd59d48 | 2094 | goto destroy_root; |
ddbcc7e8 | 2095 | |
a7df69b8 | 2096 | ret = cgroup_rstat_init(root_cgrp); |
d427dfeb | 2097 | if (ret) |
2bd59d48 | 2098 | goto destroy_root; |
ddbcc7e8 | 2099 | |
a7df69b8 JW |
2100 | ret = rebind_subsystems(root, ss_mask); |
2101 | if (ret) | |
2102 | goto exit_stats; | |
2103 | ||
324bda9e AS |
2104 | ret = cgroup_bpf_inherit(root_cgrp); |
2105 | WARN_ON_ONCE(ret); | |
2106 | ||
ed1777de TH |
2107 | trace_cgroup_setup_root(root); |
2108 | ||
d427dfeb TH |
2109 | /* |
2110 | * There must be no failure case after here, since rebinding takes | |
2111 | * care of subsystems' refcounts, which are explicitly dropped in | |
2112 | * the failure exit path. | |
2113 | */ | |
2114 | list_add(&root->root_list, &cgroup_roots); | |
2115 | cgroup_root_count++; | |
0df6a63f | 2116 | |
d427dfeb | 2117 | /* |
3dd06ffa | 2118 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2119 | * objects. |
2120 | */ | |
82d6489d | 2121 | spin_lock_irq(&css_set_lock); |
0de0942d | 2122 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2123 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2124 | if (css_set_populated(cset)) |
2125 | cgroup_update_populated(root_cgrp, true); | |
2126 | } | |
82d6489d | 2127 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 2128 | |
d5c419b6 | 2129 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2130 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2131 | |
d427dfeb | 2132 | ret = 0; |
2bd59d48 | 2133 | goto out; |
d427dfeb | 2134 | |
a7df69b8 JW |
2135 | exit_stats: |
2136 | cgroup_rstat_exit(root_cgrp); | |
2bd59d48 TH |
2137 | destroy_root: |
2138 | kernfs_destroy_root(root->kf_root); | |
2139 | root->kf_root = NULL; | |
2140 | exit_root_id: | |
d427dfeb | 2141 | cgroup_exit_root_id(root); |
9d755d33 | 2142 | cancel_ref: |
9a1049da | 2143 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2144 | out: |
d427dfeb TH |
2145 | free_cgrp_cset_links(&tmp_links); |
2146 | return ret; | |
ddbcc7e8 PM |
2147 | } |
2148 | ||
cca8f327 | 2149 | int cgroup_do_get_tree(struct fs_context *fc) |
ddbcc7e8 | 2150 | { |
71d883c3 | 2151 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
23bf1b6b | 2152 | int ret; |
ddbcc7e8 | 2153 | |
23bf1b6b | 2154 | ctx->kfc.root = ctx->root->kf_root; |
cca8f327 | 2155 | if (fc->fs_type == &cgroup2_fs_type) |
23bf1b6b | 2156 | ctx->kfc.magic = CGROUP2_SUPER_MAGIC; |
cca8f327 | 2157 | else |
23bf1b6b DH |
2158 | ctx->kfc.magic = CGROUP_SUPER_MAGIC; |
2159 | ret = kernfs_get_tree(fc); | |
ed82571b | 2160 | |
56fde9e0 | 2161 | /* |
633feee3 TH |
2162 | * In non-init cgroup namespace, instead of root cgroup's dentry, |
2163 | * we return the dentry corresponding to the cgroupns->root_cgrp. | |
56fde9e0 | 2164 | */ |
cca8f327 | 2165 | if (!ret && ctx->ns != &init_cgroup_ns) { |
633feee3 | 2166 | struct dentry *nsdentry; |
71d883c3 | 2167 | struct super_block *sb = fc->root->d_sb; |
633feee3 | 2168 | struct cgroup *cgrp; |
e37a06f1 | 2169 | |
633feee3 TH |
2170 | mutex_lock(&cgroup_mutex); |
2171 | spin_lock_irq(&css_set_lock); | |
2172 | ||
cca8f327 | 2173 | cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root); |
633feee3 TH |
2174 | |
2175 | spin_unlock_irq(&css_set_lock); | |
2176 | mutex_unlock(&cgroup_mutex); | |
2177 | ||
399504e2 | 2178 | nsdentry = kernfs_node_dentry(cgrp->kn, sb); |
71d883c3 | 2179 | dput(fc->root); |
71d883c3 | 2180 | if (IS_ERR(nsdentry)) { |
399504e2 | 2181 | deactivate_locked_super(sb); |
630faf81 AV |
2182 | ret = PTR_ERR(nsdentry); |
2183 | nsdentry = NULL; | |
71d883c3 | 2184 | } |
630faf81 | 2185 | fc->root = nsdentry; |
67e9c74b TH |
2186 | } |
2187 | ||
23bf1b6b | 2188 | if (!ctx->kfc.new_sb_created) |
71d883c3 | 2189 | cgroup_put(&ctx->root->cgrp); |
633feee3 | 2190 | |
71d883c3 | 2191 | return ret; |
633feee3 TH |
2192 | } |
2193 | ||
90129625 AV |
2194 | /* |
2195 | * Destroy a cgroup filesystem context. | |
2196 | */ | |
2197 | static void cgroup_fs_context_free(struct fs_context *fc) | |
ddbcc7e8 | 2198 | { |
90129625 AV |
2199 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
2200 | ||
f5dfb531 AV |
2201 | kfree(ctx->name); |
2202 | kfree(ctx->release_agent); | |
cca8f327 | 2203 | put_cgroup_ns(ctx->ns); |
23bf1b6b | 2204 | kernfs_free_fs_context(fc); |
90129625 AV |
2205 | kfree(ctx); |
2206 | } | |
2207 | ||
90129625 | 2208 | static int cgroup_get_tree(struct fs_context *fc) |
ddbcc7e8 | 2209 | { |
90129625 | 2210 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
5136f636 | 2211 | int ret; |
ddbcc7e8 | 2212 | |
dc79ec1b | 2213 | WRITE_ONCE(cgrp_dfl_visible, true); |
90129625 | 2214 | cgroup_get_live(&cgrp_dfl_root.cgrp); |
cf6299b1 | 2215 | ctx->root = &cgrp_dfl_root; |
ed82571b | 2216 | |
cca8f327 | 2217 | ret = cgroup_do_get_tree(fc); |
71d883c3 AV |
2218 | if (!ret) |
2219 | apply_cgroup_root_flags(ctx->flags); | |
2220 | return ret; | |
90129625 AV |
2221 | } |
2222 | ||
90129625 AV |
2223 | static const struct fs_context_operations cgroup_fs_context_ops = { |
2224 | .free = cgroup_fs_context_free, | |
e34a98d5 | 2225 | .parse_param = cgroup2_parse_param, |
90129625 AV |
2226 | .get_tree = cgroup_get_tree, |
2227 | .reconfigure = cgroup_reconfigure, | |
2228 | }; | |
2229 | ||
2230 | static const struct fs_context_operations cgroup1_fs_context_ops = { | |
2231 | .free = cgroup_fs_context_free, | |
8d2451f4 | 2232 | .parse_param = cgroup1_parse_param, |
90129625 AV |
2233 | .get_tree = cgroup1_get_tree, |
2234 | .reconfigure = cgroup1_reconfigure, | |
2235 | }; | |
2236 | ||
2237 | /* | |
23bf1b6b DH |
2238 | * Initialise the cgroup filesystem creation/reconfiguration context. Notably, |
2239 | * we select the namespace we're going to use. | |
90129625 AV |
2240 | */ |
2241 | static int cgroup_init_fs_context(struct fs_context *fc) | |
2242 | { | |
2243 | struct cgroup_fs_context *ctx; | |
2244 | ||
2245 | ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL); | |
2246 | if (!ctx) | |
2247 | return -ENOMEM; | |
ed82571b | 2248 | |
cca8f327 AV |
2249 | ctx->ns = current->nsproxy->cgroup_ns; |
2250 | get_cgroup_ns(ctx->ns); | |
23bf1b6b | 2251 | fc->fs_private = &ctx->kfc; |
90129625 AV |
2252 | if (fc->fs_type == &cgroup2_fs_type) |
2253 | fc->ops = &cgroup_fs_context_ops; | |
2254 | else | |
2255 | fc->ops = &cgroup1_fs_context_ops; | |
f7a99451 | 2256 | put_user_ns(fc->user_ns); |
23bf1b6b DH |
2257 | fc->user_ns = get_user_ns(ctx->ns->user_ns); |
2258 | fc->global = true; | |
6a010a49 TH |
2259 | |
2260 | #ifdef CONFIG_CGROUP_FAVOR_DYNMODS | |
2261 | ctx->flags |= CGRP_ROOT_FAVOR_DYNMODS; | |
2262 | #endif | |
90129625 | 2263 | return 0; |
2bd59d48 | 2264 | } |
8e30e2b8 | 2265 | |
2bd59d48 TH |
2266 | static void cgroup_kill_sb(struct super_block *sb) |
2267 | { | |
2268 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2269 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
a015edd2 | 2270 | |
970317aa | 2271 | /* |
35ac1184 | 2272 | * If @root doesn't have any children, start killing it. |
9d755d33 | 2273 | * This prevents new mounts by disabling percpu_ref_tryget_live(). |
1f779fb2 LZ |
2274 | * |
2275 | * And don't kill the default root. | |
970317aa | 2276 | */ |
35ac1184 | 2277 | if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root && |
04f8ef56 QW |
2278 | !percpu_ref_is_dying(&root->cgrp.self.refcnt)) { |
2279 | cgroup_bpf_offline(&root->cgrp); | |
9d755d33 | 2280 | percpu_ref_kill(&root->cgrp.self.refcnt); |
04f8ef56 | 2281 | } |
35ac1184 | 2282 | cgroup_put(&root->cgrp); |
2bd59d48 | 2283 | kernfs_kill_sb(sb); |
ddbcc7e8 | 2284 | } |
970317aa | 2285 | |
0a268dbd | 2286 | struct file_system_type cgroup_fs_type = { |
8d2451f4 AV |
2287 | .name = "cgroup", |
2288 | .init_fs_context = cgroup_init_fs_context, | |
d7167b14 | 2289 | .parameters = cgroup1_fs_parameters, |
8d2451f4 AV |
2290 | .kill_sb = cgroup_kill_sb, |
2291 | .fs_flags = FS_USERNS_MOUNT, | |
ddbcc7e8 | 2292 | }; |
3126121f | 2293 | |
67e9c74b | 2294 | static struct file_system_type cgroup2_fs_type = { |
e34a98d5 AV |
2295 | .name = "cgroup2", |
2296 | .init_fs_context = cgroup_init_fs_context, | |
d7167b14 | 2297 | .parameters = cgroup2_fs_parameters, |
e34a98d5 AV |
2298 | .kill_sb = cgroup_kill_sb, |
2299 | .fs_flags = FS_USERNS_MOUNT, | |
67e9c74b | 2300 | }; |
3126121f | 2301 | |
d5f68d33 AV |
2302 | #ifdef CONFIG_CPUSETS |
2303 | static const struct fs_context_operations cpuset_fs_context_ops = { | |
2304 | .get_tree = cgroup1_get_tree, | |
2305 | .free = cgroup_fs_context_free, | |
2306 | }; | |
2307 | ||
2308 | /* | |
2309 | * This is ugly, but preserves the userspace API for existing cpuset | |
2310 | * users. If someone tries to mount the "cpuset" filesystem, we | |
2311 | * silently switch it to mount "cgroup" instead | |
2312 | */ | |
2313 | static int cpuset_init_fs_context(struct fs_context *fc) | |
2314 | { | |
2315 | char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER); | |
2316 | struct cgroup_fs_context *ctx; | |
2317 | int err; | |
2318 | ||
2319 | err = cgroup_init_fs_context(fc); | |
2320 | if (err) { | |
2321 | kfree(agent); | |
2322 | return err; | |
2323 | } | |
2324 | ||
2325 | fc->ops = &cpuset_fs_context_ops; | |
2326 | ||
2327 | ctx = cgroup_fc2context(fc); | |
2328 | ctx->subsys_mask = 1 << cpuset_cgrp_id; | |
2329 | ctx->flags |= CGRP_ROOT_NOPREFIX; | |
2330 | ctx->release_agent = agent; | |
2331 | ||
2332 | get_filesystem(&cgroup_fs_type); | |
2333 | put_filesystem(fc->fs_type); | |
2334 | fc->fs_type = &cgroup_fs_type; | |
2335 | ||
2336 | return 0; | |
2337 | } | |
2338 | ||
2339 | static struct file_system_type cpuset_fs_type = { | |
2340 | .name = "cpuset", | |
2341 | .init_fs_context = cpuset_init_fs_context, | |
2342 | .fs_flags = FS_USERNS_MOUNT, | |
2343 | }; | |
2344 | #endif | |
2345 | ||
0a268dbd TH |
2346 | int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2347 | struct cgroup_namespace *ns) | |
a79a908f AK |
2348 | { |
2349 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 2350 | |
4c737b41 | 2351 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
2352 | } |
2353 | ||
4c737b41 TH |
2354 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
2355 | struct cgroup_namespace *ns) | |
a79a908f | 2356 | { |
4c737b41 | 2357 | int ret; |
a79a908f AK |
2358 | |
2359 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2360 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2361 | |
2362 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2363 | ||
82d6489d | 2364 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2365 | mutex_unlock(&cgroup_mutex); |
2366 | ||
2367 | return ret; | |
2368 | } | |
2369 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2370 | ||
857a2beb | 2371 | /** |
913ffdb5 | 2372 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2373 | * @task: target task |
857a2beb TH |
2374 | * @buf: the buffer to write the path into |
2375 | * @buflen: the length of the buffer | |
2376 | * | |
913ffdb5 TH |
2377 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2378 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2379 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2380 | * cgroup controller callbacks. | |
2381 | * | |
e61734c5 | 2382 | * Return value is the same as kernfs_path(). |
857a2beb | 2383 | */ |
4c737b41 | 2384 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2385 | { |
3dd06ffa | 2386 | struct cgroup_root *root; |
913ffdb5 | 2387 | struct cgroup *cgrp; |
e61734c5 | 2388 | int hierarchy_id = 1; |
4c737b41 | 2389 | int ret; |
857a2beb TH |
2390 | |
2391 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2392 | spin_lock_irq(&css_set_lock); |
857a2beb | 2393 | |
913ffdb5 TH |
2394 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2395 | ||
857a2beb TH |
2396 | if (root) { |
2397 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 2398 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2399 | } else { |
2400 | /* if no hierarchy exists, everyone is in "/" */ | |
8619e94d | 2401 | ret = strscpy(buf, "/", buflen); |
857a2beb TH |
2402 | } |
2403 | ||
82d6489d | 2404 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2405 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 2406 | return ret; |
857a2beb | 2407 | } |
913ffdb5 | 2408 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2409 | |
4f7e7236 TH |
2410 | /** |
2411 | * cgroup_attach_lock - Lock for ->attach() | |
2412 | * @lock_threadgroup: whether to down_write cgroup_threadgroup_rwsem | |
2413 | * | |
2414 | * cgroup migration sometimes needs to stabilize threadgroups against forks and | |
2415 | * exits by write-locking cgroup_threadgroup_rwsem. However, some ->attach() | |
2416 | * implementations (e.g. cpuset), also need to disable CPU hotplug. | |
2417 | * Unfortunately, letting ->attach() operations acquire cpus_read_lock() can | |
2418 | * lead to deadlocks. | |
2419 | * | |
2420 | * Bringing up a CPU may involve creating and destroying tasks which requires | |
2421 | * read-locking threadgroup_rwsem, so threadgroup_rwsem nests inside | |
2422 | * cpus_read_lock(). If we call an ->attach() which acquires the cpus lock while | |
2423 | * write-locking threadgroup_rwsem, the locking order is reversed and we end up | |
2424 | * waiting for an on-going CPU hotplug operation which in turn is waiting for | |
2425 | * the threadgroup_rwsem to be released to create new tasks. For more details: | |
2426 | * | |
2427 | * http://lkml.kernel.org/r/20220711174629.uehfmqegcwn2lqzu@wubuntu | |
2428 | * | |
2429 | * Resolve the situation by always acquiring cpus_read_lock() before optionally | |
2430 | * write-locking cgroup_threadgroup_rwsem. This allows ->attach() to assume that | |
2431 | * CPU hotplug is disabled on entry. | |
2432 | */ | |
075b593f | 2433 | void cgroup_attach_lock(bool lock_threadgroup) |
4f7e7236 TH |
2434 | { |
2435 | cpus_read_lock(); | |
2436 | if (lock_threadgroup) | |
2437 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
2438 | } | |
2439 | ||
2440 | /** | |
2441 | * cgroup_attach_unlock - Undo cgroup_attach_lock() | |
2442 | * @lock_threadgroup: whether to up_write cgroup_threadgroup_rwsem | |
2443 | */ | |
075b593f | 2444 | void cgroup_attach_unlock(bool lock_threadgroup) |
4f7e7236 TH |
2445 | { |
2446 | if (lock_threadgroup) | |
2447 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2448 | cpus_read_unlock(); | |
2449 | } | |
2450 | ||
adaae5dc | 2451 | /** |
e595cd70 | 2452 | * cgroup_migrate_add_task - add a migration target task to a migration context |
adaae5dc | 2453 | * @task: target task |
e595cd70 | 2454 | * @mgctx: target migration context |
adaae5dc | 2455 | * |
e595cd70 TH |
2456 | * Add @task, which is a migration target, to @mgctx->tset. This function |
2457 | * becomes noop if @task doesn't need to be migrated. @task's css_set | |
2458 | * should have been added as a migration source and @task->cg_list will be | |
2459 | * moved from the css_set's tasks list to mg_tasks one. | |
adaae5dc | 2460 | */ |
e595cd70 TH |
2461 | static void cgroup_migrate_add_task(struct task_struct *task, |
2462 | struct cgroup_mgctx *mgctx) | |
adaae5dc TH |
2463 | { |
2464 | struct css_set *cset; | |
2465 | ||
f0d9a5f1 | 2466 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2467 | |
2468 | /* @task either already exited or can't exit until the end */ | |
2469 | if (task->flags & PF_EXITING) | |
2470 | return; | |
2471 | ||
5153faac TH |
2472 | /* cgroup_threadgroup_rwsem protects racing against forks */ |
2473 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
adaae5dc TH |
2474 | |
2475 | cset = task_css_set(task); | |
2476 | if (!cset->mg_src_cgrp) | |
2477 | return; | |
2478 | ||
61046727 TH |
2479 | mgctx->tset.nr_tasks++; |
2480 | ||
adaae5dc TH |
2481 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2482 | if (list_empty(&cset->mg_node)) | |
e595cd70 TH |
2483 | list_add_tail(&cset->mg_node, |
2484 | &mgctx->tset.src_csets); | |
adaae5dc | 2485 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
d8ebf519 | 2486 | list_add_tail(&cset->mg_dst_cset->mg_node, |
e595cd70 | 2487 | &mgctx->tset.dst_csets); |
adaae5dc TH |
2488 | } |
2489 | ||
2f7ee569 TH |
2490 | /** |
2491 | * cgroup_taskset_first - reset taskset and return the first task | |
2492 | * @tset: taskset of interest | |
1f7dd3e5 | 2493 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2494 | * |
2495 | * @tset iteration is initialized and the first task is returned. | |
2496 | */ | |
1f7dd3e5 TH |
2497 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2498 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2499 | { |
b3dc094e TH |
2500 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2501 | tset->cur_task = NULL; | |
2502 | ||
1f7dd3e5 | 2503 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2504 | } |
2f7ee569 TH |
2505 | |
2506 | /** | |
2507 | * cgroup_taskset_next - iterate to the next task in taskset | |
2508 | * @tset: taskset of interest | |
1f7dd3e5 | 2509 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2510 | * |
2511 | * Return the next task in @tset. Iteration must have been initialized | |
2512 | * with cgroup_taskset_first(). | |
2513 | */ | |
1f7dd3e5 TH |
2514 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2515 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2516 | { |
b3dc094e TH |
2517 | struct css_set *cset = tset->cur_cset; |
2518 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2519 | |
d20d30eb | 2520 | while (CGROUP_HAS_SUBSYS_CONFIG && &cset->mg_node != tset->csets) { |
b3dc094e TH |
2521 | if (!task) |
2522 | task = list_first_entry(&cset->mg_tasks, | |
2523 | struct task_struct, cg_list); | |
2524 | else | |
2525 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2526 | |
b3dc094e TH |
2527 | if (&task->cg_list != &cset->mg_tasks) { |
2528 | tset->cur_cset = cset; | |
2529 | tset->cur_task = task; | |
1f7dd3e5 TH |
2530 | |
2531 | /* | |
2532 | * This function may be called both before and | |
2533 | * after cgroup_taskset_migrate(). The two cases | |
2534 | * can be distinguished by looking at whether @cset | |
2535 | * has its ->mg_dst_cset set. | |
2536 | */ | |
2537 | if (cset->mg_dst_cset) | |
2538 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2539 | else | |
2540 | *dst_cssp = cset->subsys[tset->ssid]; | |
2541 | ||
b3dc094e TH |
2542 | return task; |
2543 | } | |
2f7ee569 | 2544 | |
b3dc094e TH |
2545 | cset = list_next_entry(cset, mg_node); |
2546 | task = NULL; | |
2547 | } | |
2f7ee569 | 2548 | |
b3dc094e | 2549 | return NULL; |
2f7ee569 | 2550 | } |
2f7ee569 | 2551 | |
adaae5dc | 2552 | /** |
2ca11b0e | 2553 | * cgroup_migrate_execute - migrate a taskset |
e595cd70 | 2554 | * @mgctx: migration context |
adaae5dc | 2555 | * |
e595cd70 | 2556 | * Migrate tasks in @mgctx as setup by migration preparation functions. |
37ff9f8f | 2557 | * This function fails iff one of the ->can_attach callbacks fails and |
e595cd70 TH |
2558 | * guarantees that either all or none of the tasks in @mgctx are migrated. |
2559 | * @mgctx is consumed regardless of success. | |
adaae5dc | 2560 | */ |
bfc2cf6f | 2561 | static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) |
adaae5dc | 2562 | { |
e595cd70 | 2563 | struct cgroup_taskset *tset = &mgctx->tset; |
37ff9f8f | 2564 | struct cgroup_subsys *ss; |
adaae5dc TH |
2565 | struct task_struct *task, *tmp_task; |
2566 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2567 | int ssid, failed_ssid, ret; |
adaae5dc | 2568 | |
adaae5dc | 2569 | /* check that we can legitimately attach to the cgroup */ |
61046727 TH |
2570 | if (tset->nr_tasks) { |
2571 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2572 | if (ss->can_attach) { | |
2573 | tset->ssid = ssid; | |
2574 | ret = ss->can_attach(tset); | |
2575 | if (ret) { | |
2576 | failed_ssid = ssid; | |
2577 | goto out_cancel_attach; | |
2578 | } | |
adaae5dc | 2579 | } |
61046727 TH |
2580 | } while_each_subsys_mask(); |
2581 | } | |
adaae5dc TH |
2582 | |
2583 | /* | |
2584 | * Now that we're guaranteed success, proceed to move all tasks to | |
2585 | * the new cgroup. There are no failure cases after here, so this | |
2586 | * is the commit point. | |
2587 | */ | |
82d6489d | 2588 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2589 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2590 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2591 | struct css_set *from_cset = task_css_set(task); | |
2592 | struct css_set *to_cset = cset->mg_dst_cset; | |
2593 | ||
2594 | get_css_set(to_cset); | |
73a7242a | 2595 | to_cset->nr_tasks++; |
f6d7d049 | 2596 | css_set_move_task(task, from_cset, to_cset, true); |
73a7242a | 2597 | from_cset->nr_tasks--; |
76f969e8 RG |
2598 | /* |
2599 | * If the source or destination cgroup is frozen, | |
2600 | * the task might require to change its state. | |
2601 | */ | |
2602 | cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp, | |
2603 | to_cset->dfl_cgrp); | |
2604 | put_css_set_locked(from_cset); | |
2605 | ||
f6d7d049 | 2606 | } |
adaae5dc | 2607 | } |
82d6489d | 2608 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2609 | |
2610 | /* | |
2611 | * Migration is committed, all target tasks are now on dst_csets. | |
2612 | * Nothing is sensitive to fork() after this point. Notify | |
2613 | * controllers that migration is complete. | |
2614 | */ | |
2615 | tset->csets = &tset->dst_csets; | |
2616 | ||
61046727 TH |
2617 | if (tset->nr_tasks) { |
2618 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2619 | if (ss->attach) { | |
2620 | tset->ssid = ssid; | |
2621 | ss->attach(tset); | |
2622 | } | |
2623 | } while_each_subsys_mask(); | |
2624 | } | |
adaae5dc TH |
2625 | |
2626 | ret = 0; | |
2627 | goto out_release_tset; | |
2628 | ||
2629 | out_cancel_attach: | |
61046727 TH |
2630 | if (tset->nr_tasks) { |
2631 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2632 | if (ssid == failed_ssid) | |
2633 | break; | |
2634 | if (ss->cancel_attach) { | |
2635 | tset->ssid = ssid; | |
2636 | ss->cancel_attach(tset); | |
2637 | } | |
2638 | } while_each_subsys_mask(); | |
2639 | } | |
adaae5dc | 2640 | out_release_tset: |
82d6489d | 2641 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2642 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2643 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2644 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2645 | list_del_init(&cset->mg_node); | |
2646 | } | |
82d6489d | 2647 | spin_unlock_irq(&css_set_lock); |
c4fa6c43 WL |
2648 | |
2649 | /* | |
2650 | * Re-initialize the cgroup_taskset structure in case it is reused | |
2651 | * again in another cgroup_migrate_add_task()/cgroup_migrate_execute() | |
2652 | * iteration. | |
2653 | */ | |
2654 | tset->nr_tasks = 0; | |
2655 | tset->csets = &tset->src_csets; | |
adaae5dc TH |
2656 | return ret; |
2657 | } | |
2658 | ||
6c694c88 | 2659 | /** |
8cfd8147 | 2660 | * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination |
6c694c88 TH |
2661 | * @dst_cgrp: destination cgroup to test |
2662 | * | |
8cfd8147 TH |
2663 | * On the default hierarchy, except for the mixable, (possible) thread root |
2664 | * and threaded cgroups, subtree_control must be zero for migration | |
2665 | * destination cgroups with tasks so that child cgroups don't compete | |
2666 | * against tasks. | |
6c694c88 | 2667 | */ |
8cfd8147 | 2668 | int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp) |
6c694c88 | 2669 | { |
8cfd8147 TH |
2670 | /* v1 doesn't have any restriction */ |
2671 | if (!cgroup_on_dfl(dst_cgrp)) | |
2672 | return 0; | |
2673 | ||
2674 | /* verify @dst_cgrp can host resources */ | |
2675 | if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp)) | |
2676 | return -EOPNOTSUPP; | |
2677 | ||
8cfd8147 TH |
2678 | /* |
2679 | * If @dst_cgrp is already or can become a thread root or is | |
2680 | * threaded, it doesn't matter. | |
2681 | */ | |
2682 | if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp)) | |
2683 | return 0; | |
2684 | ||
2685 | /* apply no-internal-process constraint */ | |
2686 | if (dst_cgrp->subtree_control) | |
2687 | return -EBUSY; | |
2688 | ||
2689 | return 0; | |
6c694c88 TH |
2690 | } |
2691 | ||
a043e3b2 | 2692 | /** |
1958d2d5 | 2693 | * cgroup_migrate_finish - cleanup after attach |
e595cd70 | 2694 | * @mgctx: migration context |
74a1166d | 2695 | * |
1958d2d5 TH |
2696 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2697 | * those functions for details. | |
74a1166d | 2698 | */ |
e595cd70 | 2699 | void cgroup_migrate_finish(struct cgroup_mgctx *mgctx) |
74a1166d | 2700 | { |
1958d2d5 | 2701 | struct css_set *cset, *tmp_cset; |
74a1166d | 2702 | |
1958d2d5 TH |
2703 | lockdep_assert_held(&cgroup_mutex); |
2704 | ||
82d6489d | 2705 | spin_lock_irq(&css_set_lock); |
e595cd70 | 2706 | |
07fd5b6c TH |
2707 | list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_src_csets, |
2708 | mg_src_preload_node) { | |
2709 | cset->mg_src_cgrp = NULL; | |
2710 | cset->mg_dst_cgrp = NULL; | |
2711 | cset->mg_dst_cset = NULL; | |
2712 | list_del_init(&cset->mg_src_preload_node); | |
2713 | put_css_set_locked(cset); | |
2714 | } | |
e595cd70 | 2715 | |
07fd5b6c TH |
2716 | list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_dst_csets, |
2717 | mg_dst_preload_node) { | |
1958d2d5 | 2718 | cset->mg_src_cgrp = NULL; |
e4857982 | 2719 | cset->mg_dst_cgrp = NULL; |
1958d2d5 | 2720 | cset->mg_dst_cset = NULL; |
07fd5b6c | 2721 | list_del_init(&cset->mg_dst_preload_node); |
a25eb52e | 2722 | put_css_set_locked(cset); |
1958d2d5 | 2723 | } |
e595cd70 | 2724 | |
82d6489d | 2725 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2726 | } |
2727 | ||
2728 | /** | |
2729 | * cgroup_migrate_add_src - add a migration source css_set | |
2730 | * @src_cset: the source css_set to add | |
2731 | * @dst_cgrp: the destination cgroup | |
e595cd70 | 2732 | * @mgctx: migration context |
1958d2d5 TH |
2733 | * |
2734 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
e595cd70 | 2735 | * @src_cset and add it to @mgctx->src_csets, which should later be cleaned |
1958d2d5 TH |
2736 | * up by cgroup_migrate_finish(). |
2737 | * | |
1ed13287 TH |
2738 | * This function may be called without holding cgroup_threadgroup_rwsem |
2739 | * even if the target is a process. Threads may be created and destroyed | |
2740 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2741 | * into play and the preloaded css_sets are guaranteed to cover all | |
2742 | * migrations. | |
1958d2d5 | 2743 | */ |
0a268dbd TH |
2744 | void cgroup_migrate_add_src(struct css_set *src_cset, |
2745 | struct cgroup *dst_cgrp, | |
e595cd70 | 2746 | struct cgroup_mgctx *mgctx) |
1958d2d5 TH |
2747 | { |
2748 | struct cgroup *src_cgrp; | |
2749 | ||
2750 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2751 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2752 | |
2b021cbf TH |
2753 | /* |
2754 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2755 | * and doesn't contain any migratable tasks. Ignore it early so | |
2756 | * that the rest of migration path doesn't get confused by it. | |
2757 | */ | |
2758 | if (src_cset->dead) | |
2759 | return; | |
2760 | ||
07fd5b6c | 2761 | if (!list_empty(&src_cset->mg_src_preload_node)) |
1958d2d5 TH |
2762 | return; |
2763 | ||
1815775e WY |
2764 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2765 | ||
1958d2d5 | 2766 | WARN_ON(src_cset->mg_src_cgrp); |
e4857982 | 2767 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2768 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2769 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2770 | ||
2771 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2772 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 | 2773 | get_css_set(src_cset); |
07fd5b6c | 2774 | list_add_tail(&src_cset->mg_src_preload_node, &mgctx->preloaded_src_csets); |
1958d2d5 TH |
2775 | } |
2776 | ||
2777 | /** | |
2778 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
e595cd70 | 2779 | * @mgctx: migration context |
1958d2d5 | 2780 | * |
e4857982 | 2781 | * Tasks are about to be moved and all the source css_sets have been |
e595cd70 TH |
2782 | * preloaded to @mgctx->preloaded_src_csets. This function looks up and |
2783 | * pins all destination css_sets, links each to its source, and append them | |
2784 | * to @mgctx->preloaded_dst_csets. | |
1958d2d5 TH |
2785 | * |
2786 | * This function must be called after cgroup_migrate_add_src() has been | |
2787 | * called on each migration source css_set. After migration is performed | |
2788 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
e595cd70 | 2789 | * @mgctx. |
1958d2d5 | 2790 | */ |
e595cd70 | 2791 | int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx) |
1958d2d5 | 2792 | { |
f817de98 | 2793 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2794 | |
2795 | lockdep_assert_held(&cgroup_mutex); | |
2796 | ||
2797 | /* look up the dst cset for each src cset and link it to src */ | |
e595cd70 | 2798 | list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets, |
07fd5b6c | 2799 | mg_src_preload_node) { |
1958d2d5 | 2800 | struct css_set *dst_cset; |
bfc2cf6f TH |
2801 | struct cgroup_subsys *ss; |
2802 | int ssid; | |
1958d2d5 | 2803 | |
e4857982 | 2804 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 | 2805 | if (!dst_cset) |
d6e486ee | 2806 | return -ENOMEM; |
1958d2d5 TH |
2807 | |
2808 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2809 | |
2810 | /* | |
2811 | * If src cset equals dst, it's noop. Drop the src. | |
2812 | * cgroup_migrate() will skip the cset too. Note that we | |
2813 | * can't handle src == dst as some nodes are used by both. | |
2814 | */ | |
2815 | if (src_cset == dst_cset) { | |
2816 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2817 | src_cset->mg_dst_cgrp = NULL; |
07fd5b6c | 2818 | list_del_init(&src_cset->mg_src_preload_node); |
a25eb52e ZL |
2819 | put_css_set(src_cset); |
2820 | put_css_set(dst_cset); | |
f817de98 TH |
2821 | continue; |
2822 | } | |
2823 | ||
1958d2d5 TH |
2824 | src_cset->mg_dst_cset = dst_cset; |
2825 | ||
07fd5b6c TH |
2826 | if (list_empty(&dst_cset->mg_dst_preload_node)) |
2827 | list_add_tail(&dst_cset->mg_dst_preload_node, | |
e595cd70 | 2828 | &mgctx->preloaded_dst_csets); |
1958d2d5 | 2829 | else |
a25eb52e | 2830 | put_css_set(dst_cset); |
bfc2cf6f TH |
2831 | |
2832 | for_each_subsys(ss, ssid) | |
2833 | if (src_cset->subsys[ssid] != dst_cset->subsys[ssid]) | |
2834 | mgctx->ss_mask |= 1 << ssid; | |
1958d2d5 TH |
2835 | } |
2836 | ||
1958d2d5 | 2837 | return 0; |
1958d2d5 TH |
2838 | } |
2839 | ||
2840 | /** | |
2841 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2842 | * @leader: the leader of the process or the task to migrate |
2843 | * @threadgroup: whether @leader points to the whole process or a single task | |
e595cd70 | 2844 | * @mgctx: migration context |
1958d2d5 | 2845 | * |
37ff9f8f TH |
2846 | * Migrate a process or task denoted by @leader. If migrating a process, |
2847 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2848 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2849 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2850 | * function and following up with cgroup_migrate_finish(). | |
2851 | * | |
2852 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2853 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2854 | * failure, when migrating multiple targets, the success or failure can be | |
2855 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2856 | * actually starting migrating. | |
2857 | */ | |
0a268dbd | 2858 | int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
bfc2cf6f | 2859 | struct cgroup_mgctx *mgctx) |
74a1166d | 2860 | { |
adaae5dc | 2861 | struct task_struct *task; |
74a1166d | 2862 | |
fb5d2b4c MSB |
2863 | /* |
2864 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2865 | * already PF_EXITING could be freed from underneath us unless we | |
2866 | * take an rcu_read_lock. | |
2867 | */ | |
82d6489d | 2868 | spin_lock_irq(&css_set_lock); |
9db8de37 | 2869 | task = leader; |
74a1166d | 2870 | do { |
e595cd70 | 2871 | cgroup_migrate_add_task(task, mgctx); |
081aa458 LZ |
2872 | if (!threadgroup) |
2873 | break; | |
9db8de37 | 2874 | } while_each_thread(leader, task); |
82d6489d | 2875 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2876 | |
bfc2cf6f | 2877 | return cgroup_migrate_execute(mgctx); |
74a1166d BB |
2878 | } |
2879 | ||
1958d2d5 TH |
2880 | /** |
2881 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2882 | * @dst_cgrp: the cgroup to attach to | |
2883 | * @leader: the task or the leader of the threadgroup to be attached | |
2884 | * @threadgroup: attach the whole threadgroup? | |
2885 | * | |
1ed13287 | 2886 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 | 2887 | */ |
0a268dbd TH |
2888 | int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader, |
2889 | bool threadgroup) | |
1958d2d5 | 2890 | { |
e595cd70 | 2891 | DEFINE_CGROUP_MGCTX(mgctx); |
1958d2d5 | 2892 | struct task_struct *task; |
6df970e4 | 2893 | int ret = 0; |
6c694c88 | 2894 | |
1958d2d5 | 2895 | /* look up all src csets */ |
82d6489d | 2896 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2897 | rcu_read_lock(); |
2898 | task = leader; | |
2899 | do { | |
e595cd70 | 2900 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); |
1958d2d5 TH |
2901 | if (!threadgroup) |
2902 | break; | |
2903 | } while_each_thread(leader, task); | |
2904 | rcu_read_unlock(); | |
82d6489d | 2905 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2906 | |
2907 | /* prepare dst csets and commit */ | |
e595cd70 | 2908 | ret = cgroup_migrate_prepare_dst(&mgctx); |
1958d2d5 | 2909 | if (!ret) |
bfc2cf6f | 2910 | ret = cgroup_migrate(leader, threadgroup, &mgctx); |
1958d2d5 | 2911 | |
e595cd70 | 2912 | cgroup_migrate_finish(&mgctx); |
ed1777de TH |
2913 | |
2914 | if (!ret) | |
e4f8d81c | 2915 | TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup); |
ed1777de | 2916 | |
1958d2d5 | 2917 | return ret; |
74a1166d BB |
2918 | } |
2919 | ||
9a3284fa | 2920 | struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup, |
4f7e7236 | 2921 | bool *threadgroup_locked) |
bbcb81d0 | 2922 | { |
bbcb81d0 | 2923 | struct task_struct *tsk; |
acbef755 | 2924 | pid_t pid; |
bbcb81d0 | 2925 | |
acbef755 | 2926 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
715c809d | 2927 | return ERR_PTR(-EINVAL); |
74a1166d | 2928 | |
9a3284fa MK |
2929 | /* |
2930 | * If we migrate a single thread, we don't care about threadgroup | |
2931 | * stability. If the thread is `current`, it won't exit(2) under our | |
2932 | * hands or change PID through exec(2). We exclude | |
2933 | * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write | |
2934 | * callers by cgroup_mutex. | |
2935 | * Therefore, we can skip the global lock. | |
2936 | */ | |
2937 | lockdep_assert_held(&cgroup_mutex); | |
4f7e7236 TH |
2938 | *threadgroup_locked = pid || threadgroup; |
2939 | cgroup_attach_lock(*threadgroup_locked); | |
715c809d | 2940 | |
b78949eb | 2941 | rcu_read_lock(); |
bbcb81d0 | 2942 | if (pid) { |
73507f33 | 2943 | tsk = find_task_by_vpid(pid); |
74a1166d | 2944 | if (!tsk) { |
715c809d TH |
2945 | tsk = ERR_PTR(-ESRCH); |
2946 | goto out_unlock_threadgroup; | |
bbcb81d0 | 2947 | } |
dedf22e9 | 2948 | } else { |
b78949eb | 2949 | tsk = current; |
dedf22e9 | 2950 | } |
cd3d0952 TH |
2951 | |
2952 | if (threadgroup) | |
b78949eb | 2953 | tsk = tsk->group_leader; |
c4c27fbd MG |
2954 | |
2955 | /* | |
77f88796 TH |
2956 | * kthreads may acquire PF_NO_SETAFFINITY during initialization. |
2957 | * If userland migrates such a kthread to a non-root cgroup, it can | |
2958 | * become trapped in a cpuset, or RT kthread may be born in a | |
2959 | * cgroup with no rt_runtime allocated. Just say no. | |
c4c27fbd | 2960 | */ |
77f88796 | 2961 | if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) { |
715c809d TH |
2962 | tsk = ERR_PTR(-EINVAL); |
2963 | goto out_unlock_threadgroup; | |
c4c27fbd MG |
2964 | } |
2965 | ||
b78949eb | 2966 | get_task_struct(tsk); |
715c809d TH |
2967 | goto out_unlock_rcu; |
2968 | ||
2969 | out_unlock_threadgroup: | |
4f7e7236 TH |
2970 | cgroup_attach_unlock(*threadgroup_locked); |
2971 | *threadgroup_locked = false; | |
715c809d | 2972 | out_unlock_rcu: |
b78949eb | 2973 | rcu_read_unlock(); |
715c809d TH |
2974 | return tsk; |
2975 | } | |
b78949eb | 2976 | |
4f7e7236 | 2977 | void cgroup_procs_write_finish(struct task_struct *task, bool threadgroup_locked) |
715c809d TH |
2978 | { |
2979 | struct cgroup_subsys *ss; | |
2980 | int ssid; | |
081aa458 | 2981 | |
715c809d TH |
2982 | /* release reference from cgroup_procs_write_start() */ |
2983 | put_task_struct(task); | |
3014dde7 | 2984 | |
4f7e7236 TH |
2985 | cgroup_attach_unlock(threadgroup_locked); |
2986 | ||
5cf1cacb TH |
2987 | for_each_subsys(ss, ssid) |
2988 | if (ss->post_attach) | |
2989 | ss->post_attach(); | |
af351026 PM |
2990 | } |
2991 | ||
6e5c8307 | 2992 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2993 | { |
f8f22e53 TH |
2994 | struct cgroup_subsys *ss; |
2995 | bool printed = false; | |
2996 | int ssid; | |
a742c59d | 2997 | |
b4e0eeaf | 2998 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2999 | if (printed) |
3000 | seq_putc(seq, ' '); | |
85db0023 | 3001 | seq_puts(seq, ss->name); |
a966a4ed | 3002 | printed = true; |
b4e0eeaf | 3003 | } while_each_subsys_mask(); |
f8f22e53 TH |
3004 | if (printed) |
3005 | seq_putc(seq, '\n'); | |
355e0c48 PM |
3006 | } |
3007 | ||
f8f22e53 TH |
3008 | /* show controllers which are enabled from the parent */ |
3009 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3010 | { |
f8f22e53 TH |
3011 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3012 | ||
5531dc91 | 3013 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 3014 | return 0; |
ddbcc7e8 PM |
3015 | } |
3016 | ||
f8f22e53 TH |
3017 | /* show controllers which are enabled for a given cgroup's children */ |
3018 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 3019 | { |
f8f22e53 TH |
3020 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3021 | ||
667c2491 | 3022 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
3023 | return 0; |
3024 | } | |
3025 | ||
3026 | /** | |
3027 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
3028 | * @cgrp: root of the subtree to update csses for | |
3029 | * | |
54962604 TH |
3030 | * @cgrp's control masks have changed and its subtree's css associations |
3031 | * need to be updated accordingly. This function looks up all css_sets | |
3032 | * which are attached to the subtree, creates the matching updated css_sets | |
3033 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
3034 | */ |
3035 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
3036 | { | |
e595cd70 | 3037 | DEFINE_CGROUP_MGCTX(mgctx); |
54962604 TH |
3038 | struct cgroup_subsys_state *d_css; |
3039 | struct cgroup *dsct; | |
f8f22e53 | 3040 | struct css_set *src_cset; |
671c11f0 | 3041 | bool has_tasks; |
f8f22e53 TH |
3042 | int ret; |
3043 | ||
f8f22e53 TH |
3044 | lockdep_assert_held(&cgroup_mutex); |
3045 | ||
3046 | /* look up all csses currently attached to @cgrp's subtree */ | |
82d6489d | 3047 | spin_lock_irq(&css_set_lock); |
54962604 | 3048 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
3049 | struct cgrp_cset_link *link; |
3050 | ||
265792d0 WL |
3051 | /* |
3052 | * As cgroup_update_dfl_csses() is only called by | |
3053 | * cgroup_apply_control(). The csses associated with the | |
3054 | * given cgrp will not be affected by changes made to | |
3055 | * its subtree_control file. We can skip them. | |
3056 | */ | |
3057 | if (dsct == cgrp) | |
3058 | continue; | |
3059 | ||
54962604 | 3060 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
e595cd70 | 3061 | cgroup_migrate_add_src(link->cset, dsct, &mgctx); |
f8f22e53 | 3062 | } |
82d6489d | 3063 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 3064 | |
671c11f0 TH |
3065 | /* |
3066 | * We need to write-lock threadgroup_rwsem while migrating tasks. | |
3067 | * However, if there are no source csets for @cgrp, changing its | |
3068 | * controllers isn't gonna produce any task migrations and the | |
3069 | * write-locking can be skipped safely. | |
3070 | */ | |
3071 | has_tasks = !list_empty(&mgctx.preloaded_src_csets); | |
4f7e7236 | 3072 | cgroup_attach_lock(has_tasks); |
671c11f0 | 3073 | |
f8f22e53 | 3074 | /* NULL dst indicates self on default hierarchy */ |
e595cd70 | 3075 | ret = cgroup_migrate_prepare_dst(&mgctx); |
f8f22e53 TH |
3076 | if (ret) |
3077 | goto out_finish; | |
3078 | ||
82d6489d | 3079 | spin_lock_irq(&css_set_lock); |
07fd5b6c TH |
3080 | list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, |
3081 | mg_src_preload_node) { | |
10265075 | 3082 | struct task_struct *task, *ntask; |
f8f22e53 | 3083 | |
10265075 TH |
3084 | /* all tasks in src_csets need to be migrated */ |
3085 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
e595cd70 | 3086 | cgroup_migrate_add_task(task, &mgctx); |
f8f22e53 | 3087 | } |
82d6489d | 3088 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 3089 | |
bfc2cf6f | 3090 | ret = cgroup_migrate_execute(&mgctx); |
f8f22e53 | 3091 | out_finish: |
e595cd70 | 3092 | cgroup_migrate_finish(&mgctx); |
4f7e7236 | 3093 | cgroup_attach_unlock(has_tasks); |
f8f22e53 TH |
3094 | return ret; |
3095 | } | |
3096 | ||
1b9b96a1 | 3097 | /** |
945ba199 | 3098 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 3099 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
3100 | * |
3101 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
3102 | * controller while the previous css is still around. This function grabs |
3103 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 3104 | */ |
0a268dbd | 3105 | void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
945ba199 | 3106 | __acquires(&cgroup_mutex) |
1b9b96a1 TH |
3107 | { |
3108 | struct cgroup *dsct; | |
ce3f1d9d | 3109 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
3110 | struct cgroup_subsys *ss; |
3111 | int ssid; | |
3112 | ||
945ba199 TH |
3113 | restart: |
3114 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 3115 | |
ce3f1d9d | 3116 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
3117 | for_each_subsys(ss, ssid) { |
3118 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3119 | DEFINE_WAIT(wait); | |
3120 | ||
ce3f1d9d | 3121 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
3122 | continue; |
3123 | ||
a590b90d | 3124 | cgroup_get_live(dsct); |
1b9b96a1 TH |
3125 | prepare_to_wait(&dsct->offline_waitq, &wait, |
3126 | TASK_UNINTERRUPTIBLE); | |
3127 | ||
3128 | mutex_unlock(&cgroup_mutex); | |
3129 | schedule(); | |
3130 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
3131 | |
3132 | cgroup_put(dsct); | |
945ba199 | 3133 | goto restart; |
1b9b96a1 TH |
3134 | } |
3135 | } | |
1b9b96a1 TH |
3136 | } |
3137 | ||
15a27c36 | 3138 | /** |
479adb89 | 3139 | * cgroup_save_control - save control masks and dom_cgrp of a subtree |
15a27c36 TH |
3140 | * @cgrp: root of the target subtree |
3141 | * | |
479adb89 TH |
3142 | * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the |
3143 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
3144 | * itself. | |
15a27c36 TH |
3145 | */ |
3146 | static void cgroup_save_control(struct cgroup *cgrp) | |
3147 | { | |
3148 | struct cgroup *dsct; | |
3149 | struct cgroup_subsys_state *d_css; | |
3150 | ||
3151 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3152 | dsct->old_subtree_control = dsct->subtree_control; | |
3153 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
479adb89 | 3154 | dsct->old_dom_cgrp = dsct->dom_cgrp; |
15a27c36 TH |
3155 | } |
3156 | } | |
3157 | ||
3158 | /** | |
3159 | * cgroup_propagate_control - refresh control masks of a subtree | |
3160 | * @cgrp: root of the target subtree | |
3161 | * | |
3162 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3163 | * ->subtree_control and propagate controller availability through the | |
3164 | * subtree so that descendants don't have unavailable controllers enabled. | |
3165 | */ | |
3166 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3167 | { | |
3168 | struct cgroup *dsct; | |
3169 | struct cgroup_subsys_state *d_css; | |
3170 | ||
3171 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3172 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3173 | dsct->subtree_ss_mask = |
3174 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3175 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3176 | } |
3177 | } | |
3178 | ||
3179 | /** | |
479adb89 | 3180 | * cgroup_restore_control - restore control masks and dom_cgrp of a subtree |
15a27c36 TH |
3181 | * @cgrp: root of the target subtree |
3182 | * | |
479adb89 TH |
3183 | * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the |
3184 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
3185 | * itself. | |
15a27c36 TH |
3186 | */ |
3187 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3188 | { | |
3189 | struct cgroup *dsct; | |
3190 | struct cgroup_subsys_state *d_css; | |
3191 | ||
3192 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3193 | dsct->subtree_control = dsct->old_subtree_control; | |
3194 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
479adb89 | 3195 | dsct->dom_cgrp = dsct->old_dom_cgrp; |
15a27c36 TH |
3196 | } |
3197 | } | |
3198 | ||
f6d635ad TH |
3199 | static bool css_visible(struct cgroup_subsys_state *css) |
3200 | { | |
3201 | struct cgroup_subsys *ss = css->ss; | |
3202 | struct cgroup *cgrp = css->cgroup; | |
3203 | ||
3204 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3205 | return true; | |
3206 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3207 | return false; | |
3208 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3209 | } | |
3210 | ||
bdb53bd7 TH |
3211 | /** |
3212 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3213 | * @cgrp: root of the target subtree |
bdb53bd7 | 3214 | * |
ce3f1d9d | 3215 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3216 | * visible. A css is created invisible if it's being implicitly enabled |
3217 | * through dependency. An invisible css is made visible when the userland | |
3218 | * explicitly enables it. | |
3219 | * | |
3220 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3221 | * been processed already aren't cleaned up. The caller is responsible for | |
8a1115ff | 3222 | * cleaning up with cgroup_apply_control_disable(). |
bdb53bd7 TH |
3223 | */ |
3224 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3225 | { | |
3226 | struct cgroup *dsct; | |
ce3f1d9d | 3227 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3228 | struct cgroup_subsys *ss; |
3229 | int ssid, ret; | |
3230 | ||
ce3f1d9d | 3231 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3232 | for_each_subsys(ss, ssid) { |
3233 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3234 | ||
3235 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) | |
3236 | continue; | |
3237 | ||
3238 | if (!css) { | |
3239 | css = css_create(dsct, ss); | |
3240 | if (IS_ERR(css)) | |
3241 | return PTR_ERR(css); | |
3242 | } | |
3243 | ||
3bc0bb36 MK |
3244 | WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt)); |
3245 | ||
f6d635ad | 3246 | if (css_visible(css)) { |
334c3679 | 3247 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3248 | if (ret) |
3249 | return ret; | |
3250 | } | |
3251 | } | |
3252 | } | |
3253 | ||
3254 | return 0; | |
3255 | } | |
3256 | ||
12b3bb6a TH |
3257 | /** |
3258 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3259 | * @cgrp: root of the target subtree |
12b3bb6a | 3260 | * |
ce3f1d9d | 3261 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3262 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3263 | * | |
3264 | * A css is hidden when the userland requests it to be disabled while other | |
3265 | * subsystems are still depending on it. The css must not actively control | |
3266 | * resources and be in the vanilla state if it's made visible again later. | |
3267 | * Controllers which may be depended upon should provide ->css_reset() for | |
3268 | * this purpose. | |
3269 | */ | |
3270 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3271 | { | |
3272 | struct cgroup *dsct; | |
ce3f1d9d | 3273 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3274 | struct cgroup_subsys *ss; |
3275 | int ssid; | |
3276 | ||
ce3f1d9d | 3277 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3278 | for_each_subsys(ss, ssid) { |
3279 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3280 | ||
3281 | if (!css) | |
3282 | continue; | |
3283 | ||
3bc0bb36 MK |
3284 | WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt)); |
3285 | ||
334c3679 TH |
3286 | if (css->parent && |
3287 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3288 | kill_css(css); |
f6d635ad | 3289 | } else if (!css_visible(css)) { |
334c3679 | 3290 | css_clear_dir(css); |
12b3bb6a TH |
3291 | if (ss->css_reset) |
3292 | ss->css_reset(css); | |
3293 | } | |
3294 | } | |
3295 | } | |
3296 | } | |
3297 | ||
f7b2814b TH |
3298 | /** |
3299 | * cgroup_apply_control - apply control mask updates to the subtree | |
3300 | * @cgrp: root of the target subtree | |
3301 | * | |
3302 | * subsystems can be enabled and disabled in a subtree using the following | |
3303 | * steps. | |
3304 | * | |
3305 | * 1. Call cgroup_save_control() to stash the current state. | |
3306 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3307 | * 3. Call cgroup_apply_control() to apply the changes. | |
3308 | * 4. Optionally perform other related operations. | |
3309 | * 5. Call cgroup_finalize_control() to finish up. | |
3310 | * | |
3311 | * This function implements step 3 and propagates the mask changes | |
3312 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3313 | * process migrations. | |
3314 | */ | |
3315 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3316 | { | |
3317 | int ret; | |
3318 | ||
3319 | cgroup_propagate_control(cgrp); | |
3320 | ||
3321 | ret = cgroup_apply_control_enable(cgrp); | |
3322 | if (ret) | |
3323 | return ret; | |
3324 | ||
3325 | /* | |
fc5a828b | 3326 | * At this point, cgroup_e_css_by_mask() results reflect the new csses |
f7b2814b TH |
3327 | * making the following cgroup_update_dfl_csses() properly update |
3328 | * css associations of all tasks in the subtree. | |
3329 | */ | |
61c41711 | 3330 | return cgroup_update_dfl_csses(cgrp); |
f7b2814b TH |
3331 | } |
3332 | ||
3333 | /** | |
3334 | * cgroup_finalize_control - finalize control mask update | |
3335 | * @cgrp: root of the target subtree | |
3336 | * @ret: the result of the update | |
3337 | * | |
3338 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3339 | */ | |
3340 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3341 | { | |
3342 | if (ret) { | |
3343 | cgroup_restore_control(cgrp); | |
3344 | cgroup_propagate_control(cgrp); | |
3345 | } | |
3346 | ||
3347 | cgroup_apply_control_disable(cgrp); | |
3348 | } | |
3349 | ||
8cfd8147 TH |
3350 | static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable) |
3351 | { | |
3352 | u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask; | |
3353 | ||
3354 | /* if nothing is getting enabled, nothing to worry about */ | |
3355 | if (!enable) | |
3356 | return 0; | |
3357 | ||
3358 | /* can @cgrp host any resources? */ | |
3359 | if (!cgroup_is_valid_domain(cgrp->dom_cgrp)) | |
3360 | return -EOPNOTSUPP; | |
3361 | ||
3362 | /* mixables don't care */ | |
3363 | if (cgroup_is_mixable(cgrp)) | |
3364 | return 0; | |
3365 | ||
3366 | if (domain_enable) { | |
3367 | /* can't enable domain controllers inside a thread subtree */ | |
3368 | if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3369 | return -EOPNOTSUPP; | |
3370 | } else { | |
3371 | /* | |
3372 | * Threaded controllers can handle internal competitions | |
3373 | * and are always allowed inside a (prospective) thread | |
3374 | * subtree. | |
3375 | */ | |
3376 | if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3377 | return 0; | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * Controllers can't be enabled for a cgroup with tasks to avoid | |
3382 | * child cgroups competing against tasks. | |
3383 | */ | |
3384 | if (cgroup_has_tasks(cgrp)) | |
3385 | return -EBUSY; | |
3386 | ||
3387 | return 0; | |
3388 | } | |
3389 | ||
f8f22e53 | 3390 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3391 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3392 | char *buf, size_t nbytes, | |
3393 | loff_t off) | |
f8f22e53 | 3394 | { |
6e5c8307 | 3395 | u16 enable = 0, disable = 0; |
a9746d8d | 3396 | struct cgroup *cgrp, *child; |
f8f22e53 | 3397 | struct cgroup_subsys *ss; |
451af504 | 3398 | char *tok; |
f8f22e53 TH |
3399 | int ssid, ret; |
3400 | ||
3401 | /* | |
d37167ab TH |
3402 | * Parse input - space separated list of subsystem names prefixed |
3403 | * with either + or -. | |
f8f22e53 | 3404 | */ |
451af504 TH |
3405 | buf = strstrip(buf); |
3406 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3407 | if (tok[0] == '\0') |
3408 | continue; | |
a7165264 | 3409 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3410 | if (!cgroup_ssid_enabled(ssid) || |
3411 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3412 | continue; |
3413 | ||
3414 | if (*tok == '+') { | |
7d331fa9 TH |
3415 | enable |= 1 << ssid; |
3416 | disable &= ~(1 << ssid); | |
f8f22e53 | 3417 | } else if (*tok == '-') { |
7d331fa9 TH |
3418 | disable |= 1 << ssid; |
3419 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3420 | } else { |
3421 | return -EINVAL; | |
3422 | } | |
3423 | break; | |
b4e0eeaf | 3424 | } while_each_subsys_mask(); |
f8f22e53 TH |
3425 | if (ssid == CGROUP_SUBSYS_COUNT) |
3426 | return -EINVAL; | |
3427 | } | |
3428 | ||
945ba199 | 3429 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3430 | if (!cgrp) |
3431 | return -ENODEV; | |
f8f22e53 TH |
3432 | |
3433 | for_each_subsys(ss, ssid) { | |
3434 | if (enable & (1 << ssid)) { | |
667c2491 | 3435 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3436 | enable &= ~(1 << ssid); |
3437 | continue; | |
3438 | } | |
3439 | ||
5531dc91 | 3440 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3441 | ret = -ENOENT; |
3442 | goto out_unlock; | |
3443 | } | |
f8f22e53 | 3444 | } else if (disable & (1 << ssid)) { |
667c2491 | 3445 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3446 | disable &= ~(1 << ssid); |
3447 | continue; | |
3448 | } | |
3449 | ||
3450 | /* a child has it enabled? */ | |
3451 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3452 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3453 | ret = -EBUSY; |
ddab2b6e | 3454 | goto out_unlock; |
f8f22e53 TH |
3455 | } |
3456 | } | |
3457 | } | |
3458 | } | |
3459 | ||
3460 | if (!enable && !disable) { | |
3461 | ret = 0; | |
ddab2b6e | 3462 | goto out_unlock; |
f8f22e53 TH |
3463 | } |
3464 | ||
8cfd8147 TH |
3465 | ret = cgroup_vet_subtree_control_enable(cgrp, enable); |
3466 | if (ret) | |
27f26753 | 3467 | goto out_unlock; |
f8f22e53 | 3468 | |
15a27c36 TH |
3469 | /* save and update control masks and prepare csses */ |
3470 | cgroup_save_control(cgrp); | |
f63070d3 | 3471 | |
15a27c36 TH |
3472 | cgrp->subtree_control |= enable; |
3473 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3474 | |
f7b2814b | 3475 | ret = cgroup_apply_control(cgrp); |
f7b2814b | 3476 | cgroup_finalize_control(cgrp, ret); |
3c745417 TH |
3477 | if (ret) |
3478 | goto out_unlock; | |
f8f22e53 TH |
3479 | |
3480 | kernfs_activate(cgrp->kn); | |
f8f22e53 | 3481 | out_unlock: |
a9746d8d | 3482 | cgroup_kn_unlock(of->kn); |
451af504 | 3483 | return ret ?: nbytes; |
f8f22e53 TH |
3484 | } |
3485 | ||
c705a00d TH |
3486 | /** |
3487 | * cgroup_enable_threaded - make @cgrp threaded | |
3488 | * @cgrp: the target cgroup | |
3489 | * | |
3490 | * Called when "threaded" is written to the cgroup.type interface file and | |
3491 | * tries to make @cgrp threaded and join the parent's resource domain. | |
3492 | * This function is never called on the root cgroup as cgroup.type doesn't | |
3493 | * exist on it. | |
3494 | */ | |
8cfd8147 TH |
3495 | static int cgroup_enable_threaded(struct cgroup *cgrp) |
3496 | { | |
3497 | struct cgroup *parent = cgroup_parent(cgrp); | |
3498 | struct cgroup *dom_cgrp = parent->dom_cgrp; | |
479adb89 TH |
3499 | struct cgroup *dsct; |
3500 | struct cgroup_subsys_state *d_css; | |
8cfd8147 TH |
3501 | int ret; |
3502 | ||
3503 | lockdep_assert_held(&cgroup_mutex); | |
3504 | ||
3505 | /* noop if already threaded */ | |
3506 | if (cgroup_is_threaded(cgrp)) | |
3507 | return 0; | |
3508 | ||
d1897c95 TH |
3509 | /* |
3510 | * If @cgroup is populated or has domain controllers enabled, it | |
3511 | * can't be switched. While the below cgroup_can_be_thread_root() | |
3512 | * test can catch the same conditions, that's only when @parent is | |
3513 | * not mixable, so let's check it explicitly. | |
3514 | */ | |
3515 | if (cgroup_is_populated(cgrp) || | |
3516 | cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
3517 | return -EOPNOTSUPP; | |
3518 | ||
8cfd8147 TH |
3519 | /* we're joining the parent's domain, ensure its validity */ |
3520 | if (!cgroup_is_valid_domain(dom_cgrp) || | |
3521 | !cgroup_can_be_thread_root(dom_cgrp)) | |
3522 | return -EOPNOTSUPP; | |
3523 | ||
8cfd8147 TH |
3524 | /* |
3525 | * The following shouldn't cause actual migrations and should | |
3526 | * always succeed. | |
3527 | */ | |
3528 | cgroup_save_control(cgrp); | |
3529 | ||
479adb89 TH |
3530 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) |
3531 | if (dsct == cgrp || cgroup_is_threaded(dsct)) | |
3532 | dsct->dom_cgrp = dom_cgrp; | |
3533 | ||
8cfd8147 TH |
3534 | ret = cgroup_apply_control(cgrp); |
3535 | if (!ret) | |
3536 | parent->nr_threaded_children++; | |
8cfd8147 TH |
3537 | |
3538 | cgroup_finalize_control(cgrp, ret); | |
3539 | return ret; | |
3540 | } | |
3541 | ||
3542 | static int cgroup_type_show(struct seq_file *seq, void *v) | |
3543 | { | |
3544 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3545 | ||
3546 | if (cgroup_is_threaded(cgrp)) | |
3547 | seq_puts(seq, "threaded\n"); | |
3548 | else if (!cgroup_is_valid_domain(cgrp)) | |
3549 | seq_puts(seq, "domain invalid\n"); | |
3550 | else if (cgroup_is_thread_root(cgrp)) | |
3551 | seq_puts(seq, "domain threaded\n"); | |
3552 | else | |
3553 | seq_puts(seq, "domain\n"); | |
3554 | ||
3555 | return 0; | |
3556 | } | |
3557 | ||
3558 | static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf, | |
3559 | size_t nbytes, loff_t off) | |
3560 | { | |
3561 | struct cgroup *cgrp; | |
3562 | int ret; | |
3563 | ||
3564 | /* only switching to threaded mode is supported */ | |
3565 | if (strcmp(strstrip(buf), "threaded")) | |
3566 | return -EINVAL; | |
3567 | ||
3bc0bb36 MK |
3568 | /* drain dying csses before we re-apply (threaded) subtree control */ |
3569 | cgrp = cgroup_kn_lock_live(of->kn, true); | |
8cfd8147 TH |
3570 | if (!cgrp) |
3571 | return -ENOENT; | |
3572 | ||
3573 | /* threaded can only be enabled */ | |
3574 | ret = cgroup_enable_threaded(cgrp); | |
3575 | ||
3576 | cgroup_kn_unlock(of->kn); | |
3577 | return ret ?: nbytes; | |
3578 | } | |
3579 | ||
1a926e0b RG |
3580 | static int cgroup_max_descendants_show(struct seq_file *seq, void *v) |
3581 | { | |
3582 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3583 | int descendants = READ_ONCE(cgrp->max_descendants); | |
3584 | ||
3585 | if (descendants == INT_MAX) | |
3586 | seq_puts(seq, "max\n"); | |
3587 | else | |
3588 | seq_printf(seq, "%d\n", descendants); | |
3589 | ||
3590 | return 0; | |
3591 | } | |
3592 | ||
3593 | static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of, | |
3594 | char *buf, size_t nbytes, loff_t off) | |
3595 | { | |
3596 | struct cgroup *cgrp; | |
3597 | int descendants; | |
3598 | ssize_t ret; | |
3599 | ||
3600 | buf = strstrip(buf); | |
3601 | if (!strcmp(buf, "max")) { | |
3602 | descendants = INT_MAX; | |
3603 | } else { | |
3604 | ret = kstrtoint(buf, 0, &descendants); | |
3605 | if (ret) | |
3606 | return ret; | |
3607 | } | |
3608 | ||
696b98f2 | 3609 | if (descendants < 0) |
1a926e0b RG |
3610 | return -ERANGE; |
3611 | ||
3612 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3613 | if (!cgrp) | |
3614 | return -ENOENT; | |
3615 | ||
3616 | cgrp->max_descendants = descendants; | |
3617 | ||
3618 | cgroup_kn_unlock(of->kn); | |
3619 | ||
3620 | return nbytes; | |
3621 | } | |
3622 | ||
3623 | static int cgroup_max_depth_show(struct seq_file *seq, void *v) | |
3624 | { | |
3625 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3626 | int depth = READ_ONCE(cgrp->max_depth); | |
3627 | ||
3628 | if (depth == INT_MAX) | |
3629 | seq_puts(seq, "max\n"); | |
3630 | else | |
3631 | seq_printf(seq, "%d\n", depth); | |
3632 | ||
3633 | return 0; | |
3634 | } | |
3635 | ||
3636 | static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of, | |
3637 | char *buf, size_t nbytes, loff_t off) | |
3638 | { | |
3639 | struct cgroup *cgrp; | |
3640 | ssize_t ret; | |
3641 | int depth; | |
3642 | ||
3643 | buf = strstrip(buf); | |
3644 | if (!strcmp(buf, "max")) { | |
3645 | depth = INT_MAX; | |
3646 | } else { | |
3647 | ret = kstrtoint(buf, 0, &depth); | |
3648 | if (ret) | |
3649 | return ret; | |
3650 | } | |
3651 | ||
696b98f2 | 3652 | if (depth < 0) |
1a926e0b RG |
3653 | return -ERANGE; |
3654 | ||
3655 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3656 | if (!cgrp) | |
3657 | return -ENOENT; | |
3658 | ||
3659 | cgrp->max_depth = depth; | |
3660 | ||
3661 | cgroup_kn_unlock(of->kn); | |
3662 | ||
3663 | return nbytes; | |
3664 | } | |
3665 | ||
4a07c222 | 3666 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3667 | { |
76f969e8 RG |
3668 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3669 | ||
3670 | seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp)); | |
3671 | seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags)); | |
3672 | ||
842b597e TH |
3673 | return 0; |
3674 | } | |
3675 | ||
3e48930c | 3676 | static int cgroup_stat_show(struct seq_file *seq, void *v) |
ec39225c RG |
3677 | { |
3678 | struct cgroup *cgroup = seq_css(seq)->cgroup; | |
3679 | ||
3680 | seq_printf(seq, "nr_descendants %d\n", | |
3681 | cgroup->nr_descendants); | |
3682 | seq_printf(seq, "nr_dying_descendants %d\n", | |
3683 | cgroup->nr_dying_descendants); | |
3684 | ||
3685 | return 0; | |
3686 | } | |
3687 | ||
d41bf8c9 TH |
3688 | static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq, |
3689 | struct cgroup *cgrp, int ssid) | |
3690 | { | |
3691 | struct cgroup_subsys *ss = cgroup_subsys[ssid]; | |
3692 | struct cgroup_subsys_state *css; | |
3693 | int ret; | |
3694 | ||
3695 | if (!ss->css_extra_stat_show) | |
3696 | return 0; | |
3697 | ||
3698 | css = cgroup_tryget_css(cgrp, ss); | |
3699 | if (!css) | |
3700 | return 0; | |
3701 | ||
3702 | ret = ss->css_extra_stat_show(seq, css); | |
3703 | css_put(css); | |
3704 | return ret; | |
3705 | } | |
3706 | ||
3707 | static int cpu_stat_show(struct seq_file *seq, void *v) | |
3708 | { | |
c3ba1329 | 3709 | struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup; |
d41bf8c9 TH |
3710 | int ret = 0; |
3711 | ||
d4ff749b | 3712 | cgroup_base_stat_cputime_show(seq); |
d41bf8c9 TH |
3713 | #ifdef CONFIG_CGROUP_SCHED |
3714 | ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id); | |
3715 | #endif | |
3716 | return ret; | |
3717 | } | |
3718 | ||
2ce7135a JW |
3719 | #ifdef CONFIG_PSI |
3720 | static int cgroup_io_pressure_show(struct seq_file *seq, void *v) | |
3721 | { | |
74321038 | 3722 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
57899a66 | 3723 | struct psi_group *psi = cgroup_psi(cgrp); |
df5ba5be DS |
3724 | |
3725 | return psi_show(seq, psi, PSI_IO); | |
2ce7135a JW |
3726 | } |
3727 | static int cgroup_memory_pressure_show(struct seq_file *seq, void *v) | |
3728 | { | |
74321038 | 3729 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
57899a66 | 3730 | struct psi_group *psi = cgroup_psi(cgrp); |
df5ba5be DS |
3731 | |
3732 | return psi_show(seq, psi, PSI_MEM); | |
2ce7135a JW |
3733 | } |
3734 | static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v) | |
3735 | { | |
74321038 | 3736 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
57899a66 | 3737 | struct psi_group *psi = cgroup_psi(cgrp); |
df5ba5be DS |
3738 | |
3739 | return psi_show(seq, psi, PSI_CPU); | |
2ce7135a | 3740 | } |
0e94682b | 3741 | |
34f26a15 CZ |
3742 | static ssize_t pressure_write(struct kernfs_open_file *of, char *buf, |
3743 | size_t nbytes, enum psi_res res) | |
0e94682b | 3744 | { |
0d2b5955 | 3745 | struct cgroup_file_ctx *ctx = of->priv; |
0e94682b SB |
3746 | struct psi_trigger *new; |
3747 | struct cgroup *cgrp; | |
385aac15 | 3748 | struct psi_group *psi; |
0e94682b SB |
3749 | |
3750 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3751 | if (!cgrp) | |
3752 | return -ENODEV; | |
3753 | ||
3754 | cgroup_get(cgrp); | |
3755 | cgroup_kn_unlock(of->kn); | |
3756 | ||
a06247c6 SB |
3757 | /* Allow only one trigger per file descriptor */ |
3758 | if (ctx->psi.trigger) { | |
3759 | cgroup_put(cgrp); | |
3760 | return -EBUSY; | |
3761 | } | |
3762 | ||
57899a66 | 3763 | psi = cgroup_psi(cgrp); |
76b079ef | 3764 | new = psi_trigger_create(psi, buf, res); |
0e94682b SB |
3765 | if (IS_ERR(new)) { |
3766 | cgroup_put(cgrp); | |
3767 | return PTR_ERR(new); | |
3768 | } | |
3769 | ||
a06247c6 | 3770 | smp_store_release(&ctx->psi.trigger, new); |
0e94682b SB |
3771 | cgroup_put(cgrp); |
3772 | ||
3773 | return nbytes; | |
3774 | } | |
3775 | ||
3776 | static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of, | |
3777 | char *buf, size_t nbytes, | |
3778 | loff_t off) | |
3779 | { | |
34f26a15 | 3780 | return pressure_write(of, buf, nbytes, PSI_IO); |
0e94682b SB |
3781 | } |
3782 | ||
3783 | static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of, | |
3784 | char *buf, size_t nbytes, | |
3785 | loff_t off) | |
3786 | { | |
34f26a15 | 3787 | return pressure_write(of, buf, nbytes, PSI_MEM); |
0e94682b SB |
3788 | } |
3789 | ||
3790 | static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of, | |
3791 | char *buf, size_t nbytes, | |
3792 | loff_t off) | |
3793 | { | |
34f26a15 | 3794 | return pressure_write(of, buf, nbytes, PSI_CPU); |
0e94682b SB |
3795 | } |
3796 | ||
52b1364b CZ |
3797 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3798 | static int cgroup_irq_pressure_show(struct seq_file *seq, void *v) | |
3799 | { | |
3800 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
57899a66 | 3801 | struct psi_group *psi = cgroup_psi(cgrp); |
52b1364b CZ |
3802 | |
3803 | return psi_show(seq, psi, PSI_IRQ); | |
3804 | } | |
3805 | ||
3806 | static ssize_t cgroup_irq_pressure_write(struct kernfs_open_file *of, | |
3807 | char *buf, size_t nbytes, | |
3808 | loff_t off) | |
3809 | { | |
34f26a15 | 3810 | return pressure_write(of, buf, nbytes, PSI_IRQ); |
52b1364b CZ |
3811 | } |
3812 | #endif | |
3813 | ||
34f26a15 CZ |
3814 | static int cgroup_pressure_show(struct seq_file *seq, void *v) |
3815 | { | |
3816 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3817 | struct psi_group *psi = cgroup_psi(cgrp); | |
3818 | ||
3819 | seq_printf(seq, "%d\n", psi->enabled); | |
3820 | ||
3821 | return 0; | |
3822 | } | |
3823 | ||
3824 | static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, | |
3825 | char *buf, size_t nbytes, | |
3826 | loff_t off) | |
3827 | { | |
3828 | ssize_t ret; | |
3829 | int enable; | |
3830 | struct cgroup *cgrp; | |
3831 | struct psi_group *psi; | |
3832 | ||
3833 | ret = kstrtoint(strstrip(buf), 0, &enable); | |
3834 | if (ret) | |
3835 | return ret; | |
3836 | ||
3837 | if (enable < 0 || enable > 1) | |
3838 | return -ERANGE; | |
3839 | ||
3840 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3841 | if (!cgrp) | |
3842 | return -ENOENT; | |
3843 | ||
3844 | psi = cgroup_psi(cgrp); | |
3845 | if (psi->enabled != enable) { | |
3846 | int i; | |
3847 | ||
3848 | /* show or hide {cpu,memory,io,irq}.pressure files */ | |
3849 | for (i = 0; i < NR_PSI_RESOURCES; i++) | |
3850 | cgroup_file_show(&cgrp->psi_files[i], enable); | |
3851 | ||
3852 | psi->enabled = enable; | |
3853 | if (enable) | |
3854 | psi_cgroup_restart(psi); | |
3855 | } | |
3856 | ||
3857 | cgroup_kn_unlock(of->kn); | |
3858 | ||
3859 | return nbytes; | |
0e94682b SB |
3860 | } |
3861 | ||
3862 | static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of, | |
3863 | poll_table *pt) | |
3864 | { | |
0d2b5955 TH |
3865 | struct cgroup_file_ctx *ctx = of->priv; |
3866 | ||
3867 | return psi_trigger_poll(&ctx->psi.trigger, of->file, pt); | |
0e94682b SB |
3868 | } |
3869 | ||
3870 | static void cgroup_pressure_release(struct kernfs_open_file *of) | |
3871 | { | |
0d2b5955 TH |
3872 | struct cgroup_file_ctx *ctx = of->priv; |
3873 | ||
a06247c6 | 3874 | psi_trigger_destroy(ctx->psi.trigger); |
0e94682b | 3875 | } |
3958e2d0 SB |
3876 | |
3877 | bool cgroup_psi_enabled(void) | |
3878 | { | |
58d8c258 CZ |
3879 | if (static_branch_likely(&psi_disabled)) |
3880 | return false; | |
3881 | ||
3958e2d0 SB |
3882 | return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0; |
3883 | } | |
3884 | ||
3885 | #else /* CONFIG_PSI */ | |
3886 | bool cgroup_psi_enabled(void) | |
3887 | { | |
3888 | return false; | |
3889 | } | |
3890 | ||
0e94682b | 3891 | #endif /* CONFIG_PSI */ |
2ce7135a | 3892 | |
76f969e8 RG |
3893 | static int cgroup_freeze_show(struct seq_file *seq, void *v) |
3894 | { | |
3895 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3896 | ||
3897 | seq_printf(seq, "%d\n", cgrp->freezer.freeze); | |
3898 | ||
3899 | return 0; | |
3900 | } | |
3901 | ||
3902 | static ssize_t cgroup_freeze_write(struct kernfs_open_file *of, | |
3903 | char *buf, size_t nbytes, loff_t off) | |
3904 | { | |
3905 | struct cgroup *cgrp; | |
3906 | ssize_t ret; | |
3907 | int freeze; | |
3908 | ||
3909 | ret = kstrtoint(strstrip(buf), 0, &freeze); | |
3910 | if (ret) | |
3911 | return ret; | |
3912 | ||
3913 | if (freeze < 0 || freeze > 1) | |
3914 | return -ERANGE; | |
3915 | ||
3916 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3917 | if (!cgrp) | |
3918 | return -ENOENT; | |
3919 | ||
3920 | cgroup_freeze(cgrp, freeze); | |
3921 | ||
3922 | cgroup_kn_unlock(of->kn); | |
3923 | ||
3924 | return nbytes; | |
3925 | } | |
3926 | ||
661ee628 CB |
3927 | static void __cgroup_kill(struct cgroup *cgrp) |
3928 | { | |
3929 | struct css_task_iter it; | |
3930 | struct task_struct *task; | |
3931 | ||
3932 | lockdep_assert_held(&cgroup_mutex); | |
3933 | ||
3934 | spin_lock_irq(&css_set_lock); | |
3935 | set_bit(CGRP_KILL, &cgrp->flags); | |
3936 | spin_unlock_irq(&css_set_lock); | |
3937 | ||
3938 | css_task_iter_start(&cgrp->self, CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED, &it); | |
3939 | while ((task = css_task_iter_next(&it))) { | |
3940 | /* Ignore kernel threads here. */ | |
3941 | if (task->flags & PF_KTHREAD) | |
3942 | continue; | |
3943 | ||
3944 | /* Skip tasks that are already dying. */ | |
3945 | if (__fatal_signal_pending(task)) | |
3946 | continue; | |
3947 | ||
3948 | send_sig(SIGKILL, task, 0); | |
3949 | } | |
3950 | css_task_iter_end(&it); | |
3951 | ||
3952 | spin_lock_irq(&css_set_lock); | |
3953 | clear_bit(CGRP_KILL, &cgrp->flags); | |
3954 | spin_unlock_irq(&css_set_lock); | |
3955 | } | |
3956 | ||
3957 | static void cgroup_kill(struct cgroup *cgrp) | |
3958 | { | |
3959 | struct cgroup_subsys_state *css; | |
3960 | struct cgroup *dsct; | |
3961 | ||
3962 | lockdep_assert_held(&cgroup_mutex); | |
3963 | ||
3964 | cgroup_for_each_live_descendant_pre(dsct, css, cgrp) | |
3965 | __cgroup_kill(dsct); | |
3966 | } | |
3967 | ||
3968 | static ssize_t cgroup_kill_write(struct kernfs_open_file *of, char *buf, | |
3969 | size_t nbytes, loff_t off) | |
3970 | { | |
3971 | ssize_t ret = 0; | |
3972 | int kill; | |
3973 | struct cgroup *cgrp; | |
3974 | ||
3975 | ret = kstrtoint(strstrip(buf), 0, &kill); | |
3976 | if (ret) | |
3977 | return ret; | |
3978 | ||
3979 | if (kill != 1) | |
3980 | return -ERANGE; | |
3981 | ||
3982 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3983 | if (!cgrp) | |
3984 | return -ENOENT; | |
3985 | ||
3986 | /* | |
3987 | * Killing is a process directed operation, i.e. the whole thread-group | |
3988 | * is taken down so act like we do for cgroup.procs and only make this | |
3989 | * writable in non-threaded cgroups. | |
3990 | */ | |
3991 | if (cgroup_is_threaded(cgrp)) | |
3992 | ret = -EOPNOTSUPP; | |
3993 | else | |
3994 | cgroup_kill(cgrp); | |
3995 | ||
3996 | cgroup_kn_unlock(of->kn); | |
3997 | ||
3998 | return ret ?: nbytes; | |
3999 | } | |
4000 | ||
e90cbebc TH |
4001 | static int cgroup_file_open(struct kernfs_open_file *of) |
4002 | { | |
5a7b5f32 | 4003 | struct cftype *cft = of_cft(of); |
0d2b5955 TH |
4004 | struct cgroup_file_ctx *ctx; |
4005 | int ret; | |
e90cbebc | 4006 | |
0d2b5955 TH |
4007 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); |
4008 | if (!ctx) | |
4009 | return -ENOMEM; | |
e5745764 TH |
4010 | |
4011 | ctx->ns = current->nsproxy->cgroup_ns; | |
4012 | get_cgroup_ns(ctx->ns); | |
0d2b5955 TH |
4013 | of->priv = ctx; |
4014 | ||
4015 | if (!cft->open) | |
4016 | return 0; | |
4017 | ||
4018 | ret = cft->open(of); | |
e5745764 TH |
4019 | if (ret) { |
4020 | put_cgroup_ns(ctx->ns); | |
0d2b5955 | 4021 | kfree(ctx); |
e5745764 | 4022 | } |
0d2b5955 | 4023 | return ret; |
e90cbebc TH |
4024 | } |
4025 | ||
4026 | static void cgroup_file_release(struct kernfs_open_file *of) | |
4027 | { | |
5a7b5f32 | 4028 | struct cftype *cft = of_cft(of); |
0d2b5955 | 4029 | struct cgroup_file_ctx *ctx = of->priv; |
e90cbebc TH |
4030 | |
4031 | if (cft->release) | |
4032 | cft->release(of); | |
e5745764 | 4033 | put_cgroup_ns(ctx->ns); |
0d2b5955 | 4034 | kfree(ctx); |
e90cbebc TH |
4035 | } |
4036 | ||
2bd59d48 TH |
4037 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
4038 | size_t nbytes, loff_t off) | |
355e0c48 | 4039 | { |
e5745764 | 4040 | struct cgroup_file_ctx *ctx = of->priv; |
2bd59d48 | 4041 | struct cgroup *cgrp = of->kn->parent->priv; |
5a7b5f32 | 4042 | struct cftype *cft = of_cft(of); |
2bd59d48 | 4043 | struct cgroup_subsys_state *css; |
a742c59d | 4044 | int ret; |
355e0c48 | 4045 | |
65026da5 JR |
4046 | if (!nbytes) |
4047 | return 0; | |
4048 | ||
5136f636 TH |
4049 | /* |
4050 | * If namespaces are delegation boundaries, disallow writes to | |
4051 | * files in an non-init namespace root from inside the namespace | |
4052 | * except for the files explicitly marked delegatable - | |
4053 | * cgroup.procs and cgroup.subtree_control. | |
4054 | */ | |
4055 | if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) && | |
4056 | !(cft->flags & CFTYPE_NS_DELEGATABLE) && | |
e5745764 | 4057 | ctx->ns != &init_cgroup_ns && ctx->ns->root_cset->dfl_cgrp == cgrp) |
5136f636 TH |
4058 | return -EPERM; |
4059 | ||
b4168640 TH |
4060 | if (cft->write) |
4061 | return cft->write(of, buf, nbytes, off); | |
4062 | ||
2bd59d48 TH |
4063 | /* |
4064 | * kernfs guarantees that a file isn't deleted with operations in | |
4065 | * flight, which means that the matching css is and stays alive and | |
4066 | * doesn't need to be pinned. The RCU locking is not necessary | |
4067 | * either. It's just for the convenience of using cgroup_css(). | |
4068 | */ | |
4069 | rcu_read_lock(); | |
4070 | css = cgroup_css(cgrp, cft->ss); | |
4071 | rcu_read_unlock(); | |
a742c59d | 4072 | |
451af504 | 4073 | if (cft->write_u64) { |
a742c59d TH |
4074 | unsigned long long v; |
4075 | ret = kstrtoull(buf, 0, &v); | |
4076 | if (!ret) | |
4077 | ret = cft->write_u64(css, cft, v); | |
4078 | } else if (cft->write_s64) { | |
4079 | long long v; | |
4080 | ret = kstrtoll(buf, 0, &v); | |
4081 | if (!ret) | |
4082 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 4083 | } else { |
a742c59d | 4084 | ret = -EINVAL; |
e73d2c61 | 4085 | } |
2bd59d48 | 4086 | |
a742c59d | 4087 | return ret ?: nbytes; |
355e0c48 PM |
4088 | } |
4089 | ||
dc50537b JW |
4090 | static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt) |
4091 | { | |
5a7b5f32 | 4092 | struct cftype *cft = of_cft(of); |
dc50537b JW |
4093 | |
4094 | if (cft->poll) | |
4095 | return cft->poll(of, pt); | |
4096 | ||
4097 | return kernfs_generic_poll(of, pt); | |
4098 | } | |
4099 | ||
6612f05b | 4100 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 4101 | { |
2bd59d48 | 4102 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
4103 | } |
4104 | ||
6612f05b | 4105 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 4106 | { |
2bd59d48 | 4107 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
4108 | } |
4109 | ||
6612f05b | 4110 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 4111 | { |
e90cbebc TH |
4112 | if (seq_cft(seq)->seq_stop) |
4113 | seq_cft(seq)->seq_stop(seq, v); | |
ddbcc7e8 PM |
4114 | } |
4115 | ||
91796569 | 4116 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 4117 | { |
7da11279 TH |
4118 | struct cftype *cft = seq_cft(m); |
4119 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 4120 | |
2da8ca82 TH |
4121 | if (cft->seq_show) |
4122 | return cft->seq_show(m, arg); | |
e73d2c61 | 4123 | |
f4c753b7 | 4124 | if (cft->read_u64) |
896f5199 TH |
4125 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
4126 | else if (cft->read_s64) | |
4127 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
4128 | else | |
4129 | return -EINVAL; | |
4130 | return 0; | |
91796569 PM |
4131 | } |
4132 | ||
2bd59d48 TH |
4133 | static struct kernfs_ops cgroup_kf_single_ops = { |
4134 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
4135 | .open = cgroup_file_open, |
4136 | .release = cgroup_file_release, | |
2bd59d48 | 4137 | .write = cgroup_file_write, |
dc50537b | 4138 | .poll = cgroup_file_poll, |
2bd59d48 | 4139 | .seq_show = cgroup_seqfile_show, |
91796569 PM |
4140 | }; |
4141 | ||
2bd59d48 TH |
4142 | static struct kernfs_ops cgroup_kf_ops = { |
4143 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
4144 | .open = cgroup_file_open, |
4145 | .release = cgroup_file_release, | |
2bd59d48 | 4146 | .write = cgroup_file_write, |
dc50537b | 4147 | .poll = cgroup_file_poll, |
2bd59d48 TH |
4148 | .seq_start = cgroup_seqfile_start, |
4149 | .seq_next = cgroup_seqfile_next, | |
4150 | .seq_stop = cgroup_seqfile_stop, | |
4151 | .seq_show = cgroup_seqfile_show, | |
4152 | }; | |
ddbcc7e8 | 4153 | |
49957f8e TH |
4154 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
4155 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
4156 | { | |
4157 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
4158 | .ia_uid = current_fsuid(), | |
4159 | .ia_gid = current_fsgid(), }; | |
4160 | ||
4161 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
4162 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
4163 | return 0; | |
4164 | ||
4165 | return kernfs_setattr(kn, &iattr); | |
4166 | } | |
4167 | ||
b12e3583 TH |
4168 | static void cgroup_file_notify_timer(struct timer_list *timer) |
4169 | { | |
4170 | cgroup_file_notify(container_of(timer, struct cgroup_file, | |
4171 | notify_timer)); | |
4172 | } | |
4173 | ||
4df8dc90 TH |
4174 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
4175 | struct cftype *cft) | |
ddbcc7e8 | 4176 | { |
8d7e6fb0 | 4177 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
4178 | struct kernfs_node *kn; |
4179 | struct lock_class_key *key = NULL; | |
49957f8e | 4180 | int ret; |
05ef1d7c | 4181 | |
2bd59d48 TH |
4182 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
4183 | key = &cft->lockdep_key; | |
4184 | #endif | |
4185 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
488dee96 DT |
4186 | cgroup_file_mode(cft), |
4187 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, | |
4188 | 0, cft->kf_ops, cft, | |
dfeb0750 | 4189 | NULL, key); |
49957f8e TH |
4190 | if (IS_ERR(kn)) |
4191 | return PTR_ERR(kn); | |
4192 | ||
4193 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 4194 | if (ret) { |
49957f8e | 4195 | kernfs_remove(kn); |
f8f22e53 TH |
4196 | return ret; |
4197 | } | |
4198 | ||
6f60eade TH |
4199 | if (cft->file_offset) { |
4200 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
4201 | ||
b12e3583 TH |
4202 | timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0); |
4203 | ||
34c06254 | 4204 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 4205 | cfile->kn = kn; |
34c06254 | 4206 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
4207 | } |
4208 | ||
f8f22e53 | 4209 | return 0; |
ddbcc7e8 PM |
4210 | } |
4211 | ||
b1f28d31 TH |
4212 | /** |
4213 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
4214 | * @css: the target css |
4215 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
4216 | * @cfts: array of cftypes to be added |
4217 | * @is_add: whether to add or remove | |
4218 | * | |
4219 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 4220 | * For removals, this function never fails. |
b1f28d31 | 4221 | */ |
4df8dc90 TH |
4222 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
4223 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 4224 | bool is_add) |
ddbcc7e8 | 4225 | { |
6732ed85 | 4226 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 4227 | int ret = 0; |
b1f28d31 | 4228 | |
01f6474c | 4229 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 4230 | |
6732ed85 TH |
4231 | restart: |
4232 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 4233 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 4234 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 4235 | continue; |
05ebb6e6 | 4236 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 4237 | continue; |
d51f39b0 | 4238 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 4239 | continue; |
d51f39b0 | 4240 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 | 4241 | continue; |
5cf8114d WL |
4242 | if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug) |
4243 | continue; | |
2739d3cc | 4244 | if (is_add) { |
4df8dc90 | 4245 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 4246 | if (ret) { |
ed3d261b JP |
4247 | pr_warn("%s: failed to add %s, err=%d\n", |
4248 | __func__, cft->name, ret); | |
6732ed85 TH |
4249 | cft_end = cft; |
4250 | is_add = false; | |
4251 | goto restart; | |
b1f28d31 | 4252 | } |
2739d3cc LZ |
4253 | } else { |
4254 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 4255 | } |
ddbcc7e8 | 4256 | } |
b598dde3 | 4257 | return ret; |
ddbcc7e8 PM |
4258 | } |
4259 | ||
21a2d343 | 4260 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 | 4261 | { |
2bb566cb | 4262 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 4263 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 4264 | struct cgroup_subsys_state *css; |
9ccece80 | 4265 | int ret = 0; |
8e3f6541 | 4266 | |
01f6474c | 4267 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 4268 | |
e8c82d20 | 4269 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 4270 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
4271 | struct cgroup *cgrp = css->cgroup; |
4272 | ||
88cb04b9 | 4273 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
4274 | continue; |
4275 | ||
4df8dc90 | 4276 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
4277 | if (ret) |
4278 | break; | |
8e3f6541 | 4279 | } |
21a2d343 TH |
4280 | |
4281 | if (is_add && !ret) | |
4282 | kernfs_activate(root->kn); | |
9ccece80 | 4283 | return ret; |
8e3f6541 TH |
4284 | } |
4285 | ||
2da440a2 | 4286 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 4287 | { |
2bb566cb | 4288 | struct cftype *cft; |
8e3f6541 | 4289 | |
2bd59d48 TH |
4290 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
4291 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
4292 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
4293 | kfree(cft->kf_ops); | |
4294 | cft->kf_ops = NULL; | |
2da440a2 | 4295 | cft->ss = NULL; |
a8ddc821 TH |
4296 | |
4297 | /* revert flags set by cgroup core while adding @cfts */ | |
0083d27b TH |
4298 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL | |
4299 | __CFTYPE_ADDED); | |
2bd59d48 | 4300 | } |
2da440a2 TH |
4301 | } |
4302 | ||
2bd59d48 | 4303 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
4304 | { |
4305 | struct cftype *cft; | |
0083d27b | 4306 | int ret = 0; |
2da440a2 | 4307 | |
2bd59d48 TH |
4308 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
4309 | struct kernfs_ops *kf_ops; | |
4310 | ||
0adb0704 TH |
4311 | WARN_ON(cft->ss || cft->kf_ops); |
4312 | ||
0083d27b TH |
4313 | if (cft->flags & __CFTYPE_ADDED) { |
4314 | ret = -EBUSY; | |
4315 | break; | |
4316 | } | |
3958e2d0 | 4317 | |
2bd59d48 TH |
4318 | if (cft->seq_start) |
4319 | kf_ops = &cgroup_kf_ops; | |
4320 | else | |
4321 | kf_ops = &cgroup_kf_single_ops; | |
4322 | ||
4323 | /* | |
4324 | * Ugh... if @cft wants a custom max_write_len, we need to | |
4325 | * make a copy of kf_ops to set its atomic_write_len. | |
4326 | */ | |
4327 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
4328 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
4329 | if (!kf_ops) { | |
0083d27b TH |
4330 | ret = -ENOMEM; |
4331 | break; | |
2bd59d48 TH |
4332 | } |
4333 | kf_ops->atomic_write_len = cft->max_write_len; | |
4334 | } | |
8e3f6541 | 4335 | |
2bd59d48 | 4336 | cft->kf_ops = kf_ops; |
2bb566cb | 4337 | cft->ss = ss; |
0083d27b | 4338 | cft->flags |= __CFTYPE_ADDED; |
2bd59d48 | 4339 | } |
2bb566cb | 4340 | |
0083d27b TH |
4341 | if (ret) |
4342 | cgroup_exit_cftypes(cfts); | |
4343 | return ret; | |
2da440a2 TH |
4344 | } |
4345 | ||
21a2d343 TH |
4346 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
4347 | { | |
01f6474c | 4348 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 | 4349 | |
21a2d343 TH |
4350 | list_del(&cfts->node); |
4351 | cgroup_apply_cftypes(cfts, false); | |
4352 | cgroup_exit_cftypes(cfts); | |
4353 | return 0; | |
8e3f6541 | 4354 | } |
8e3f6541 | 4355 | |
79578621 TH |
4356 | /** |
4357 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
4358 | * @cfts: zero-length name terminated array of cftypes |
4359 | * | |
2bb566cb TH |
4360 | * Unregister @cfts. Files described by @cfts are removed from all |
4361 | * existing cgroups and all future cgroups won't have them either. This | |
4362 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
4363 | * |
4364 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 4365 | * registered. |
79578621 | 4366 | */ |
2bb566cb | 4367 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 4368 | { |
21a2d343 | 4369 | int ret; |
79578621 | 4370 | |
0083d27b TH |
4371 | if (!cfts || cfts[0].name[0] == '\0') |
4372 | return 0; | |
4373 | ||
4374 | if (!(cfts[0].flags & __CFTYPE_ADDED)) | |
4375 | return -ENOENT; | |
4376 | ||
01f6474c | 4377 | mutex_lock(&cgroup_mutex); |
21a2d343 | 4378 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 4379 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 4380 | return ret; |
80b13586 TH |
4381 | } |
4382 | ||
8e3f6541 TH |
4383 | /** |
4384 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
4385 | * @ss: target cgroup subsystem | |
4386 | * @cfts: zero-length name terminated array of cftypes | |
4387 | * | |
4388 | * Register @cfts to @ss. Files described by @cfts are created for all | |
4389 | * existing cgroups to which @ss is attached and all future cgroups will | |
4390 | * have them too. This function can be called anytime whether @ss is | |
4391 | * attached or not. | |
4392 | * | |
4393 | * Returns 0 on successful registration, -errno on failure. Note that this | |
4394 | * function currently returns 0 as long as @cfts registration is successful | |
4395 | * even if some file creation attempts on existing cgroups fail. | |
4396 | */ | |
2cf669a5 | 4397 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 4398 | { |
9ccece80 | 4399 | int ret; |
8e3f6541 | 4400 | |
fc5ed1e9 | 4401 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
4402 | return 0; |
4403 | ||
dc5736ed LZ |
4404 | if (!cfts || cfts[0].name[0] == '\0') |
4405 | return 0; | |
2bb566cb | 4406 | |
2bd59d48 TH |
4407 | ret = cgroup_init_cftypes(ss, cfts); |
4408 | if (ret) | |
4409 | return ret; | |
79578621 | 4410 | |
01f6474c | 4411 | mutex_lock(&cgroup_mutex); |
21a2d343 | 4412 | |
0adb0704 | 4413 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 4414 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 4415 | if (ret) |
21a2d343 | 4416 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 4417 | |
01f6474c | 4418 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 4419 | return ret; |
79578621 TH |
4420 | } |
4421 | ||
a8ddc821 TH |
4422 | /** |
4423 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
4424 | * @ss: target cgroup subsystem | |
4425 | * @cfts: zero-length name terminated array of cftypes | |
4426 | * | |
4427 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
4428 | * the default hierarchy. | |
4429 | */ | |
4430 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
4431 | { | |
4432 | struct cftype *cft; | |
4433 | ||
4434 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 4435 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
4436 | return cgroup_add_cftypes(ss, cfts); |
4437 | } | |
4438 | ||
4439 | /** | |
4440 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
4441 | * @ss: target cgroup subsystem | |
4442 | * @cfts: zero-length name terminated array of cftypes | |
4443 | * | |
4444 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
4445 | * the legacy hierarchies. | |
4446 | */ | |
2cf669a5 TH |
4447 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
4448 | { | |
a8ddc821 TH |
4449 | struct cftype *cft; |
4450 | ||
e4b7037c TH |
4451 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
4452 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
4453 | return cgroup_add_cftypes(ss, cfts); |
4454 | } | |
4455 | ||
34c06254 TH |
4456 | /** |
4457 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
4458 | * @cfile: target cgroup_file | |
4459 | * | |
4460 | * @cfile must have been obtained by setting cftype->file_offset. | |
4461 | */ | |
4462 | void cgroup_file_notify(struct cgroup_file *cfile) | |
4463 | { | |
4464 | unsigned long flags; | |
4465 | ||
4466 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
b12e3583 TH |
4467 | if (cfile->kn) { |
4468 | unsigned long last = cfile->notified_at; | |
4469 | unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV; | |
4470 | ||
4471 | if (time_in_range(jiffies, last, next)) { | |
4472 | timer_reduce(&cfile->notify_timer, next); | |
4473 | } else { | |
4474 | kernfs_notify(cfile->kn); | |
4475 | cfile->notified_at = jiffies; | |
4476 | } | |
4477 | } | |
34c06254 TH |
4478 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); |
4479 | } | |
4480 | ||
e2691f6b TH |
4481 | /** |
4482 | * cgroup_file_show - show or hide a hidden cgroup file | |
4483 | * @cfile: target cgroup_file obtained by setting cftype->file_offset | |
4484 | * @show: whether to show or hide | |
4485 | */ | |
4486 | void cgroup_file_show(struct cgroup_file *cfile, bool show) | |
4487 | { | |
4488 | struct kernfs_node *kn; | |
4489 | ||
4490 | spin_lock_irq(&cgroup_file_kn_lock); | |
4491 | kn = cfile->kn; | |
4492 | kernfs_get(kn); | |
4493 | spin_unlock_irq(&cgroup_file_kn_lock); | |
4494 | ||
4495 | if (kn) | |
4496 | kernfs_show(kn, show); | |
4497 | ||
4498 | kernfs_put(kn); | |
4499 | } | |
4500 | ||
53fa5261 | 4501 | /** |
492eb21b | 4502 | * css_next_child - find the next child of a given css |
c2931b70 TH |
4503 | * @pos: the current position (%NULL to initiate traversal) |
4504 | * @parent: css whose children to walk | |
53fa5261 | 4505 | * |
c2931b70 | 4506 | * This function returns the next child of @parent and should be called |
87fb54f1 | 4507 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
4508 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
4509 | * be returned regardless of their states. | |
4510 | * | |
4511 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4512 | * css which finished ->css_online() is guaranteed to be visible in the | |
4513 | * future iterations and will stay visible until the last reference is put. | |
4514 | * A css which hasn't finished ->css_online() or already finished | |
4515 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4516 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 4517 | */ |
c2931b70 TH |
4518 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
4519 | struct cgroup_subsys_state *parent) | |
53fa5261 | 4520 | { |
c2931b70 | 4521 | struct cgroup_subsys_state *next; |
53fa5261 | 4522 | |
8353da1f | 4523 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
4524 | |
4525 | /* | |
de3f0341 TH |
4526 | * @pos could already have been unlinked from the sibling list. |
4527 | * Once a cgroup is removed, its ->sibling.next is no longer | |
4528 | * updated when its next sibling changes. CSS_RELEASED is set when | |
4529 | * @pos is taken off list, at which time its next pointer is valid, | |
4530 | * and, as releases are serialized, the one pointed to by the next | |
4531 | * pointer is guaranteed to not have started release yet. This | |
4532 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
4533 | * critical section, the one pointed to by its next pointer is | |
4534 | * guaranteed to not have finished its RCU grace period even if we | |
58315c96 | 4535 | * have dropped rcu_read_lock() in-between iterations. |
3b287a50 | 4536 | * |
de3f0341 TH |
4537 | * If @pos has CSS_RELEASED set, its next pointer can't be |
4538 | * dereferenced; however, as each css is given a monotonically | |
4539 | * increasing unique serial number and always appended to the | |
4540 | * sibling list, the next one can be found by walking the parent's | |
4541 | * children until the first css with higher serial number than | |
4542 | * @pos's. While this path can be slower, it happens iff iteration | |
4543 | * races against release and the race window is very small. | |
53fa5261 | 4544 | */ |
3b287a50 | 4545 | if (!pos) { |
c2931b70 TH |
4546 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
4547 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
4548 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 4549 | } else { |
3010c5b9 MB |
4550 | list_for_each_entry_rcu(next, &parent->children, sibling, |
4551 | lockdep_is_held(&cgroup_mutex)) | |
3b287a50 TH |
4552 | if (next->serial_nr > pos->serial_nr) |
4553 | break; | |
53fa5261 TH |
4554 | } |
4555 | ||
3b281afb TH |
4556 | /* |
4557 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 4558 | * the next sibling. |
3b281afb | 4559 | */ |
c2931b70 TH |
4560 | if (&next->sibling != &parent->children) |
4561 | return next; | |
3b281afb | 4562 | return NULL; |
53fa5261 | 4563 | } |
53fa5261 | 4564 | |
574bd9f7 | 4565 | /** |
492eb21b | 4566 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 4567 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4568 | * @root: css whose descendants to walk |
574bd9f7 | 4569 | * |
492eb21b | 4570 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
4571 | * to visit for pre-order traversal of @root's descendants. @root is |
4572 | * included in the iteration and the first node to be visited. | |
75501a6d | 4573 | * |
87fb54f1 TH |
4574 | * While this function requires cgroup_mutex or RCU read locking, it |
4575 | * doesn't require the whole traversal to be contained in a single critical | |
4576 | * section. This function will return the correct next descendant as long | |
4577 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
4578 | * |
4579 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4580 | * css which finished ->css_online() is guaranteed to be visible in the | |
4581 | * future iterations and will stay visible until the last reference is put. | |
4582 | * A css which hasn't finished ->css_online() or already finished | |
4583 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4584 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4585 | */ |
492eb21b TH |
4586 | struct cgroup_subsys_state * |
4587 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
4588 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4589 | { |
492eb21b | 4590 | struct cgroup_subsys_state *next; |
574bd9f7 | 4591 | |
8353da1f | 4592 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4593 | |
bd8815a6 | 4594 | /* if first iteration, visit @root */ |
7805d000 | 4595 | if (!pos) |
bd8815a6 | 4596 | return root; |
574bd9f7 TH |
4597 | |
4598 | /* visit the first child if exists */ | |
492eb21b | 4599 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4600 | if (next) |
4601 | return next; | |
4602 | ||
4603 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4604 | while (pos != root) { |
5c9d535b | 4605 | next = css_next_child(pos, pos->parent); |
75501a6d | 4606 | if (next) |
574bd9f7 | 4607 | return next; |
5c9d535b | 4608 | pos = pos->parent; |
7805d000 | 4609 | } |
574bd9f7 TH |
4610 | |
4611 | return NULL; | |
4612 | } | |
474a2800 | 4613 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 4614 | |
12a9d2fe | 4615 | /** |
492eb21b TH |
4616 | * css_rightmost_descendant - return the rightmost descendant of a css |
4617 | * @pos: css of interest | |
12a9d2fe | 4618 | * |
492eb21b TH |
4619 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4620 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4621 | * subtree of @pos. |
75501a6d | 4622 | * |
87fb54f1 TH |
4623 | * While this function requires cgroup_mutex or RCU read locking, it |
4624 | * doesn't require the whole traversal to be contained in a single critical | |
4625 | * section. This function will return the correct rightmost descendant as | |
4626 | * long as @pos is accessible. | |
12a9d2fe | 4627 | */ |
492eb21b TH |
4628 | struct cgroup_subsys_state * |
4629 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4630 | { |
492eb21b | 4631 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4632 | |
8353da1f | 4633 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4634 | |
4635 | do { | |
4636 | last = pos; | |
4637 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4638 | pos = NULL; | |
492eb21b | 4639 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4640 | pos = tmp; |
4641 | } while (pos); | |
4642 | ||
4643 | return last; | |
4644 | } | |
12a9d2fe | 4645 | |
492eb21b TH |
4646 | static struct cgroup_subsys_state * |
4647 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4648 | { |
492eb21b | 4649 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4650 | |
4651 | do { | |
4652 | last = pos; | |
492eb21b | 4653 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4654 | } while (pos); |
4655 | ||
4656 | return last; | |
4657 | } | |
4658 | ||
4659 | /** | |
492eb21b | 4660 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4661 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4662 | * @root: css whose descendants to walk |
574bd9f7 | 4663 | * |
492eb21b | 4664 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4665 | * to visit for post-order traversal of @root's descendants. @root is |
4666 | * included in the iteration and the last node to be visited. | |
75501a6d | 4667 | * |
87fb54f1 TH |
4668 | * While this function requires cgroup_mutex or RCU read locking, it |
4669 | * doesn't require the whole traversal to be contained in a single critical | |
4670 | * section. This function will return the correct next descendant as long | |
4671 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4672 | * @cgroup. | |
c2931b70 TH |
4673 | * |
4674 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4675 | * css which finished ->css_online() is guaranteed to be visible in the | |
4676 | * future iterations and will stay visible until the last reference is put. | |
4677 | * A css which hasn't finished ->css_online() or already finished | |
4678 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4679 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4680 | */ |
492eb21b TH |
4681 | struct cgroup_subsys_state * |
4682 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4683 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4684 | { |
492eb21b | 4685 | struct cgroup_subsys_state *next; |
574bd9f7 | 4686 | |
8353da1f | 4687 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4688 | |
58b79a91 TH |
4689 | /* if first iteration, visit leftmost descendant which may be @root */ |
4690 | if (!pos) | |
4691 | return css_leftmost_descendant(root); | |
574bd9f7 | 4692 | |
bd8815a6 TH |
4693 | /* if we visited @root, we're done */ |
4694 | if (pos == root) | |
4695 | return NULL; | |
4696 | ||
574bd9f7 | 4697 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4698 | next = css_next_child(pos, pos->parent); |
75501a6d | 4699 | if (next) |
492eb21b | 4700 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4701 | |
4702 | /* no sibling left, visit parent */ | |
5c9d535b | 4703 | return pos->parent; |
574bd9f7 | 4704 | } |
574bd9f7 | 4705 | |
f3d46500 TH |
4706 | /** |
4707 | * css_has_online_children - does a css have online children | |
4708 | * @css: the target css | |
4709 | * | |
4710 | * Returns %true if @css has any online children; otherwise, %false. This | |
4711 | * function can be called from any context but the caller is responsible | |
4712 | * for synchronizing against on/offlining as necessary. | |
4713 | */ | |
4714 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4715 | { |
f3d46500 TH |
4716 | struct cgroup_subsys_state *child; |
4717 | bool ret = false; | |
cbc125ef TH |
4718 | |
4719 | rcu_read_lock(); | |
f3d46500 | 4720 | css_for_each_child(child, css) { |
99bae5f9 | 4721 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4722 | ret = true; |
4723 | break; | |
cbc125ef TH |
4724 | } |
4725 | } | |
4726 | rcu_read_unlock(); | |
f3d46500 | 4727 | return ret; |
574bd9f7 | 4728 | } |
574bd9f7 | 4729 | |
450ee0c1 TH |
4730 | static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it) |
4731 | { | |
4732 | struct list_head *l; | |
4733 | struct cgrp_cset_link *link; | |
4734 | struct css_set *cset; | |
4735 | ||
4736 | lockdep_assert_held(&css_set_lock); | |
4737 | ||
4738 | /* find the next threaded cset */ | |
4739 | if (it->tcset_pos) { | |
4740 | l = it->tcset_pos->next; | |
4741 | ||
4742 | if (l != it->tcset_head) { | |
4743 | it->tcset_pos = l; | |
4744 | return container_of(l, struct css_set, | |
4745 | threaded_csets_node); | |
4746 | } | |
4747 | ||
4748 | it->tcset_pos = NULL; | |
4749 | } | |
4750 | ||
4751 | /* find the next cset */ | |
4752 | l = it->cset_pos; | |
4753 | l = l->next; | |
4754 | if (l == it->cset_head) { | |
4755 | it->cset_pos = NULL; | |
4756 | return NULL; | |
4757 | } | |
4758 | ||
4759 | if (it->ss) { | |
4760 | cset = container_of(l, struct css_set, e_cset_node[it->ss->id]); | |
4761 | } else { | |
4762 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4763 | cset = link->cset; | |
4764 | } | |
4765 | ||
4766 | it->cset_pos = l; | |
4767 | ||
4768 | /* initialize threaded css_set walking */ | |
4769 | if (it->flags & CSS_TASK_ITER_THREADED) { | |
4770 | if (it->cur_dcset) | |
4771 | put_css_set_locked(it->cur_dcset); | |
4772 | it->cur_dcset = cset; | |
4773 | get_css_set(cset); | |
4774 | ||
4775 | it->tcset_head = &cset->threaded_csets; | |
4776 | it->tcset_pos = &cset->threaded_csets; | |
4777 | } | |
4778 | ||
4779 | return cset; | |
4780 | } | |
4781 | ||
0942eeee | 4782 | /** |
58315c96 | 4783 | * css_task_iter_advance_css_set - advance a task iterator to the next css_set |
0942eeee TH |
4784 | * @it: the iterator to advance |
4785 | * | |
4786 | * Advance @it to the next css_set to walk. | |
d515876e | 4787 | */ |
ecb9d535 | 4788 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4789 | { |
d515876e TH |
4790 | struct css_set *cset; |
4791 | ||
f0d9a5f1 | 4792 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4793 | |
f43caa2a MK |
4794 | /* Advance to the next non-empty css_set and find first non-empty tasks list*/ |
4795 | while ((cset = css_task_iter_next_css_set(it))) { | |
4796 | if (!list_empty(&cset->tasks)) { | |
4797 | it->cur_tasks_head = &cset->tasks; | |
4798 | break; | |
4799 | } else if (!list_empty(&cset->mg_tasks)) { | |
4800 | it->cur_tasks_head = &cset->mg_tasks; | |
4801 | break; | |
4802 | } else if (!list_empty(&cset->dying_tasks)) { | |
4803 | it->cur_tasks_head = &cset->dying_tasks; | |
4804 | break; | |
d515876e | 4805 | } |
9c974c77 | 4806 | } |
f43caa2a MK |
4807 | if (!cset) { |
4808 | it->task_pos = NULL; | |
4809 | return; | |
4810 | } | |
4811 | it->task_pos = it->cur_tasks_head->next; | |
ed27b9f7 TH |
4812 | |
4813 | /* | |
4814 | * We don't keep css_sets locked across iteration steps and thus | |
4815 | * need to take steps to ensure that iteration can be resumed after | |
4816 | * the lock is re-acquired. Iteration is performed at two levels - | |
4817 | * css_sets and tasks in them. | |
4818 | * | |
4819 | * Once created, a css_set never leaves its cgroup lists, so a | |
4820 | * pinned css_set is guaranteed to stay put and we can resume | |
4821 | * iteration afterwards. | |
4822 | * | |
4823 | * Tasks may leave @cset across iteration steps. This is resolved | |
4824 | * by registering each iterator with the css_set currently being | |
4825 | * walked and making css_set_move_task() advance iterators whose | |
4826 | * next task is leaving. | |
4827 | */ | |
4828 | if (it->cur_cset) { | |
4829 | list_del(&it->iters_node); | |
4830 | put_css_set_locked(it->cur_cset); | |
4831 | } | |
4832 | get_css_set(cset); | |
4833 | it->cur_cset = cset; | |
4834 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4835 | } |
4836 | ||
b636fd38 TH |
4837 | static void css_task_iter_skip(struct css_task_iter *it, |
4838 | struct task_struct *task) | |
ecb9d535 | 4839 | { |
b636fd38 TH |
4840 | lockdep_assert_held(&css_set_lock); |
4841 | ||
4842 | if (it->task_pos == &task->cg_list) { | |
4843 | it->task_pos = it->task_pos->next; | |
4844 | it->flags |= CSS_TASK_ITER_SKIPPED; | |
4845 | } | |
4846 | } | |
ecb9d535 | 4847 | |
ecb9d535 TH |
4848 | static void css_task_iter_advance(struct css_task_iter *it) |
4849 | { | |
c03cd773 | 4850 | struct task_struct *task; |
ecb9d535 | 4851 | |
f0d9a5f1 | 4852 | lockdep_assert_held(&css_set_lock); |
bc2fb7ed | 4853 | repeat: |
e9d81a1b TH |
4854 | if (it->task_pos) { |
4855 | /* | |
f43caa2a MK |
4856 | * Advance iterator to find next entry. We go through cset |
4857 | * tasks, mg_tasks and dying_tasks, when consumed we move onto | |
4858 | * the next cset. | |
e9d81a1b | 4859 | */ |
b636fd38 TH |
4860 | if (it->flags & CSS_TASK_ITER_SKIPPED) |
4861 | it->flags &= ~CSS_TASK_ITER_SKIPPED; | |
4862 | else | |
4863 | it->task_pos = it->task_pos->next; | |
ecb9d535 | 4864 | |
f43caa2a MK |
4865 | if (it->task_pos == &it->cur_cset->tasks) { |
4866 | it->cur_tasks_head = &it->cur_cset->mg_tasks; | |
4867 | it->task_pos = it->cur_tasks_head->next; | |
9c974c77 | 4868 | } |
f43caa2a MK |
4869 | if (it->task_pos == &it->cur_cset->mg_tasks) { |
4870 | it->cur_tasks_head = &it->cur_cset->dying_tasks; | |
4871 | it->task_pos = it->cur_tasks_head->next; | |
9c974c77 | 4872 | } |
f43caa2a | 4873 | if (it->task_pos == &it->cur_cset->dying_tasks) |
e9d81a1b | 4874 | css_task_iter_advance_css_set(it); |
e9d81a1b TH |
4875 | } else { |
4876 | /* called from start, proceed to the first cset */ | |
ecb9d535 | 4877 | css_task_iter_advance_css_set(it); |
e9d81a1b | 4878 | } |
bc2fb7ed | 4879 | |
c03cd773 TH |
4880 | if (!it->task_pos) |
4881 | return; | |
4882 | ||
4883 | task = list_entry(it->task_pos, struct task_struct, cg_list); | |
4884 | ||
4885 | if (it->flags & CSS_TASK_ITER_PROCS) { | |
4886 | /* if PROCS, skip over tasks which aren't group leaders */ | |
4887 | if (!thread_group_leader(task)) | |
4888 | goto repeat; | |
4889 | ||
4890 | /* and dying leaders w/o live member threads */ | |
f43caa2a | 4891 | if (it->cur_tasks_head == &it->cur_cset->dying_tasks && |
9c974c77 | 4892 | !atomic_read(&task->signal->live)) |
c03cd773 TH |
4893 | goto repeat; |
4894 | } else { | |
4895 | /* skip all dying ones */ | |
f43caa2a | 4896 | if (it->cur_tasks_head == &it->cur_cset->dying_tasks) |
c03cd773 TH |
4897 | goto repeat; |
4898 | } | |
ecb9d535 TH |
4899 | } |
4900 | ||
0942eeee | 4901 | /** |
72ec7029 TH |
4902 | * css_task_iter_start - initiate task iteration |
4903 | * @css: the css to walk tasks of | |
bc2fb7ed | 4904 | * @flags: CSS_TASK_ITER_* flags |
0942eeee TH |
4905 | * @it: the task iterator to use |
4906 | * | |
72ec7029 TH |
4907 | * Initiate iteration through the tasks of @css. The caller can call |
4908 | * css_task_iter_next() to walk through the tasks until the function | |
4909 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4910 | * called. | |
0942eeee | 4911 | */ |
bc2fb7ed | 4912 | void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, |
72ec7029 | 4913 | struct css_task_iter *it) |
817929ec | 4914 | { |
ed27b9f7 TH |
4915 | memset(it, 0, sizeof(*it)); |
4916 | ||
82d6489d | 4917 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4918 | |
3ebb2b6e | 4919 | it->ss = css->ss; |
bc2fb7ed | 4920 | it->flags = flags; |
3ebb2b6e | 4921 | |
d20d30eb | 4922 | if (CGROUP_HAS_SUBSYS_CONFIG && it->ss) |
3ebb2b6e TH |
4923 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; |
4924 | else | |
4925 | it->cset_pos = &css->cgroup->cset_links; | |
4926 | ||
0f0a2b4f | 4927 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4928 | |
e9d81a1b | 4929 | css_task_iter_advance(it); |
ed27b9f7 | 4930 | |
82d6489d | 4931 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4932 | } |
4933 | ||
0942eeee | 4934 | /** |
72ec7029 | 4935 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4936 | * @it: the task iterator being iterated |
4937 | * | |
4938 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4939 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4940 | * reaches the end. | |
0942eeee | 4941 | */ |
72ec7029 | 4942 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4943 | { |
d5745675 | 4944 | if (it->cur_task) { |
ed27b9f7 | 4945 | put_task_struct(it->cur_task); |
d5745675 TH |
4946 | it->cur_task = NULL; |
4947 | } | |
ed27b9f7 | 4948 | |
82d6489d | 4949 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4950 | |
cee0c33c TH |
4951 | /* @it may be half-advanced by skips, finish advancing */ |
4952 | if (it->flags & CSS_TASK_ITER_SKIPPED) | |
4953 | css_task_iter_advance(it); | |
4954 | ||
d5745675 TH |
4955 | if (it->task_pos) { |
4956 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4957 | cg_list); | |
4958 | get_task_struct(it->cur_task); | |
4959 | css_task_iter_advance(it); | |
4960 | } | |
ed27b9f7 | 4961 | |
82d6489d | 4962 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4963 | |
4964 | return it->cur_task; | |
817929ec PM |
4965 | } |
4966 | ||
0942eeee | 4967 | /** |
72ec7029 | 4968 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4969 | * @it: the task iterator to finish |
4970 | * | |
72ec7029 | 4971 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4972 | */ |
72ec7029 | 4973 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4974 | { |
ed27b9f7 | 4975 | if (it->cur_cset) { |
82d6489d | 4976 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4977 | list_del(&it->iters_node); |
4978 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4979 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4980 | } |
4981 | ||
450ee0c1 TH |
4982 | if (it->cur_dcset) |
4983 | put_css_set(it->cur_dcset); | |
4984 | ||
ed27b9f7 TH |
4985 | if (it->cur_task) |
4986 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4987 | } |
4988 | ||
b4b90a8e | 4989 | static void cgroup_procs_release(struct kernfs_open_file *of) |
31a7df01 | 4990 | { |
0d2b5955 TH |
4991 | struct cgroup_file_ctx *ctx = of->priv; |
4992 | ||
4993 | if (ctx->procs.started) | |
4994 | css_task_iter_end(&ctx->procs.iter); | |
b4b90a8e | 4995 | } |
6c694c88 | 4996 | |
b4b90a8e TH |
4997 | static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos) |
4998 | { | |
4999 | struct kernfs_open_file *of = s->private; | |
0d2b5955 | 5000 | struct cgroup_file_ctx *ctx = of->priv; |
31a7df01 | 5001 | |
2d4ecb03 VA |
5002 | if (pos) |
5003 | (*pos)++; | |
5004 | ||
0d2b5955 | 5005 | return css_task_iter_next(&ctx->procs.iter); |
b4b90a8e | 5006 | } |
31a7df01 | 5007 | |
8cfd8147 TH |
5008 | static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos, |
5009 | unsigned int iter_flags) | |
b4b90a8e TH |
5010 | { |
5011 | struct kernfs_open_file *of = s->private; | |
5012 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
0d2b5955 TH |
5013 | struct cgroup_file_ctx *ctx = of->priv; |
5014 | struct css_task_iter *it = &ctx->procs.iter; | |
8cc99345 | 5015 | |
952aaa12 | 5016 | /* |
b4b90a8e TH |
5017 | * When a seq_file is seeked, it's always traversed sequentially |
5018 | * from position 0, so we can simply keep iterating on !0 *pos. | |
952aaa12 | 5019 | */ |
0d2b5955 | 5020 | if (!ctx->procs.started) { |
2d4ecb03 | 5021 | if (WARN_ON_ONCE((*pos))) |
b4b90a8e | 5022 | return ERR_PTR(-EINVAL); |
450ee0c1 | 5023 | css_task_iter_start(&cgrp->self, iter_flags, it); |
0d2b5955 | 5024 | ctx->procs.started = true; |
2d4ecb03 | 5025 | } else if (!(*pos)) { |
b4b90a8e | 5026 | css_task_iter_end(it); |
450ee0c1 | 5027 | css_task_iter_start(&cgrp->self, iter_flags, it); |
2d4ecb03 VA |
5028 | } else |
5029 | return it->cur_task; | |
bbcb81d0 | 5030 | |
b4b90a8e TH |
5031 | return cgroup_procs_next(s, NULL, NULL); |
5032 | } | |
24528255 | 5033 | |
8cfd8147 TH |
5034 | static void *cgroup_procs_start(struct seq_file *s, loff_t *pos) |
5035 | { | |
5036 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
5037 | ||
5038 | /* | |
5039 | * All processes of a threaded subtree belong to the domain cgroup | |
5040 | * of the subtree. Only threads can be distributed across the | |
5041 | * subtree. Reject reads on cgroup.procs in the subtree proper. | |
5042 | * They're always empty anyway. | |
5043 | */ | |
5044 | if (cgroup_is_threaded(cgrp)) | |
5045 | return ERR_PTR(-EOPNOTSUPP); | |
5046 | ||
5047 | return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS | | |
5048 | CSS_TASK_ITER_THREADED); | |
5049 | } | |
5050 | ||
b4b90a8e | 5051 | static int cgroup_procs_show(struct seq_file *s, void *v) |
bbcb81d0 | 5052 | { |
bc2fb7ed | 5053 | seq_printf(s, "%d\n", task_pid_vnr(v)); |
97978e6d DL |
5054 | return 0; |
5055 | } | |
5056 | ||
f3553220 CB |
5057 | static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb) |
5058 | { | |
5059 | int ret; | |
5060 | struct inode *inode; | |
5061 | ||
5062 | lockdep_assert_held(&cgroup_mutex); | |
5063 | ||
5064 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); | |
5065 | if (!inode) | |
5066 | return -ENOMEM; | |
5067 | ||
4609e1f1 | 5068 | ret = inode_permission(&nop_mnt_idmap, inode, MAY_WRITE); |
f3553220 CB |
5069 | iput(inode); |
5070 | return ret; | |
5071 | } | |
5072 | ||
715c809d TH |
5073 | static int cgroup_procs_write_permission(struct cgroup *src_cgrp, |
5074 | struct cgroup *dst_cgrp, | |
e5745764 TH |
5075 | struct super_block *sb, |
5076 | struct cgroup_namespace *ns) | |
715c809d | 5077 | { |
715c809d | 5078 | struct cgroup *com_cgrp = src_cgrp; |
715c809d TH |
5079 | int ret; |
5080 | ||
5081 | lockdep_assert_held(&cgroup_mutex); | |
5082 | ||
5083 | /* find the common ancestor */ | |
5084 | while (!cgroup_is_descendant(dst_cgrp, com_cgrp)) | |
5085 | com_cgrp = cgroup_parent(com_cgrp); | |
5086 | ||
5087 | /* %current should be authorized to migrate to the common ancestor */ | |
f3553220 | 5088 | ret = cgroup_may_write(com_cgrp, sb); |
715c809d TH |
5089 | if (ret) |
5090 | return ret; | |
5091 | ||
5092 | /* | |
5093 | * If namespaces are delegation boundaries, %current must be able | |
5094 | * to see both source and destination cgroups from its namespace. | |
5095 | */ | |
5096 | if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) && | |
5097 | (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) || | |
5098 | !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp))) | |
5099 | return -ENOENT; | |
5100 | ||
5101 | return 0; | |
5102 | } | |
5103 | ||
6df970e4 CB |
5104 | static int cgroup_attach_permissions(struct cgroup *src_cgrp, |
5105 | struct cgroup *dst_cgrp, | |
e5745764 TH |
5106 | struct super_block *sb, bool threadgroup, |
5107 | struct cgroup_namespace *ns) | |
6df970e4 CB |
5108 | { |
5109 | int ret = 0; | |
5110 | ||
e5745764 | 5111 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb, ns); |
6df970e4 CB |
5112 | if (ret) |
5113 | return ret; | |
5114 | ||
5115 | ret = cgroup_migrate_vet_dst(dst_cgrp); | |
5116 | if (ret) | |
5117 | return ret; | |
5118 | ||
5119 | if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp)) | |
5120 | ret = -EOPNOTSUPP; | |
5121 | ||
5122 | return ret; | |
5123 | } | |
5124 | ||
da70862e MK |
5125 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
5126 | bool threadgroup) | |
715c809d | 5127 | { |
e5745764 | 5128 | struct cgroup_file_ctx *ctx = of->priv; |
715c809d TH |
5129 | struct cgroup *src_cgrp, *dst_cgrp; |
5130 | struct task_struct *task; | |
1756d799 | 5131 | const struct cred *saved_cred; |
715c809d | 5132 | ssize_t ret; |
4f7e7236 | 5133 | bool threadgroup_locked; |
715c809d TH |
5134 | |
5135 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
5136 | if (!dst_cgrp) | |
5137 | return -ENODEV; | |
5138 | ||
4f7e7236 | 5139 | task = cgroup_procs_write_start(buf, threadgroup, &threadgroup_locked); |
715c809d TH |
5140 | ret = PTR_ERR_OR_ZERO(task); |
5141 | if (ret) | |
5142 | goto out_unlock; | |
5143 | ||
5144 | /* find the source cgroup */ | |
5145 | spin_lock_irq(&css_set_lock); | |
5146 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
5147 | spin_unlock_irq(&css_set_lock); | |
5148 | ||
1756d799 TH |
5149 | /* |
5150 | * Process and thread migrations follow same delegation rule. Check | |
5151 | * permissions using the credentials from file open to protect against | |
5152 | * inherited fd attacks. | |
5153 | */ | |
5154 | saved_cred = override_creds(of->file->f_cred); | |
6df970e4 | 5155 | ret = cgroup_attach_permissions(src_cgrp, dst_cgrp, |
e5745764 TH |
5156 | of->file->f_path.dentry->d_sb, |
5157 | threadgroup, ctx->ns); | |
1756d799 | 5158 | revert_creds(saved_cred); |
715c809d TH |
5159 | if (ret) |
5160 | goto out_finish; | |
5161 | ||
da70862e | 5162 | ret = cgroup_attach_task(dst_cgrp, task, threadgroup); |
715c809d TH |
5163 | |
5164 | out_finish: | |
4f7e7236 | 5165 | cgroup_procs_write_finish(task, threadgroup_locked); |
715c809d TH |
5166 | out_unlock: |
5167 | cgroup_kn_unlock(of->kn); | |
5168 | ||
da70862e MK |
5169 | return ret; |
5170 | } | |
5171 | ||
5172 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, | |
5173 | char *buf, size_t nbytes, loff_t off) | |
5174 | { | |
5175 | return __cgroup_procs_write(of, buf, true) ?: nbytes; | |
715c809d TH |
5176 | } |
5177 | ||
8cfd8147 TH |
5178 | static void *cgroup_threads_start(struct seq_file *s, loff_t *pos) |
5179 | { | |
5180 | return __cgroup_procs_start(s, pos, 0); | |
5181 | } | |
5182 | ||
5183 | static ssize_t cgroup_threads_write(struct kernfs_open_file *of, | |
5184 | char *buf, size_t nbytes, loff_t off) | |
5185 | { | |
da70862e | 5186 | return __cgroup_procs_write(of, buf, false) ?: nbytes; |
8cfd8147 TH |
5187 | } |
5188 | ||
a14c6874 | 5189 | /* cgroup core interface files for the default hierarchy */ |
d62beb7f | 5190 | static struct cftype cgroup_base_files[] = { |
8cfd8147 TH |
5191 | { |
5192 | .name = "cgroup.type", | |
5193 | .flags = CFTYPE_NOT_ON_ROOT, | |
5194 | .seq_show = cgroup_type_show, | |
5195 | .write = cgroup_type_write, | |
5196 | }, | |
81a6a5cd | 5197 | { |
d5c56ced | 5198 | .name = "cgroup.procs", |
5136f636 | 5199 | .flags = CFTYPE_NS_DELEGATABLE, |
6f60eade | 5200 | .file_offset = offsetof(struct cgroup, procs_file), |
b4b90a8e TH |
5201 | .release = cgroup_procs_release, |
5202 | .seq_start = cgroup_procs_start, | |
5203 | .seq_next = cgroup_procs_next, | |
5204 | .seq_show = cgroup_procs_show, | |
acbef755 | 5205 | .write = cgroup_procs_write, |
102a775e | 5206 | }, |
8cfd8147 TH |
5207 | { |
5208 | .name = "cgroup.threads", | |
4f58424d | 5209 | .flags = CFTYPE_NS_DELEGATABLE, |
8cfd8147 TH |
5210 | .release = cgroup_procs_release, |
5211 | .seq_start = cgroup_threads_start, | |
5212 | .seq_next = cgroup_procs_next, | |
5213 | .seq_show = cgroup_procs_show, | |
5214 | .write = cgroup_threads_write, | |
5215 | }, | |
f8f22e53 TH |
5216 | { |
5217 | .name = "cgroup.controllers", | |
f8f22e53 TH |
5218 | .seq_show = cgroup_controllers_show, |
5219 | }, | |
5220 | { | |
5221 | .name = "cgroup.subtree_control", | |
5136f636 | 5222 | .flags = CFTYPE_NS_DELEGATABLE, |
f8f22e53 | 5223 | .seq_show = cgroup_subtree_control_show, |
451af504 | 5224 | .write = cgroup_subtree_control_write, |
f8f22e53 | 5225 | }, |
842b597e | 5226 | { |
4a07c222 | 5227 | .name = "cgroup.events", |
a14c6874 | 5228 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 5229 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 5230 | .seq_show = cgroup_events_show, |
842b597e | 5231 | }, |
1a926e0b RG |
5232 | { |
5233 | .name = "cgroup.max.descendants", | |
5234 | .seq_show = cgroup_max_descendants_show, | |
5235 | .write = cgroup_max_descendants_write, | |
5236 | }, | |
5237 | { | |
5238 | .name = "cgroup.max.depth", | |
5239 | .seq_show = cgroup_max_depth_show, | |
5240 | .write = cgroup_max_depth_write, | |
5241 | }, | |
ec39225c RG |
5242 | { |
5243 | .name = "cgroup.stat", | |
3e48930c | 5244 | .seq_show = cgroup_stat_show, |
ec39225c | 5245 | }, |
76f969e8 RG |
5246 | { |
5247 | .name = "cgroup.freeze", | |
5248 | .flags = CFTYPE_NOT_ON_ROOT, | |
5249 | .seq_show = cgroup_freeze_show, | |
5250 | .write = cgroup_freeze_write, | |
5251 | }, | |
661ee628 CB |
5252 | { |
5253 | .name = "cgroup.kill", | |
5254 | .flags = CFTYPE_NOT_ON_ROOT, | |
5255 | .write = cgroup_kill_write, | |
5256 | }, | |
d41bf8c9 TH |
5257 | { |
5258 | .name = "cpu.stat", | |
d41bf8c9 TH |
5259 | .seq_show = cpu_stat_show, |
5260 | }, | |
8a693f77 TH |
5261 | { } /* terminate */ |
5262 | }; | |
5263 | ||
5264 | static struct cftype cgroup_psi_files[] = { | |
2ce7135a JW |
5265 | #ifdef CONFIG_PSI |
5266 | { | |
5267 | .name = "io.pressure", | |
34f26a15 | 5268 | .file_offset = offsetof(struct cgroup, psi_files[PSI_IO]), |
2ce7135a | 5269 | .seq_show = cgroup_io_pressure_show, |
0e94682b SB |
5270 | .write = cgroup_io_pressure_write, |
5271 | .poll = cgroup_pressure_poll, | |
5272 | .release = cgroup_pressure_release, | |
2ce7135a JW |
5273 | }, |
5274 | { | |
5275 | .name = "memory.pressure", | |
34f26a15 | 5276 | .file_offset = offsetof(struct cgroup, psi_files[PSI_MEM]), |
2ce7135a | 5277 | .seq_show = cgroup_memory_pressure_show, |
0e94682b SB |
5278 | .write = cgroup_memory_pressure_write, |
5279 | .poll = cgroup_pressure_poll, | |
5280 | .release = cgroup_pressure_release, | |
2ce7135a JW |
5281 | }, |
5282 | { | |
5283 | .name = "cpu.pressure", | |
34f26a15 | 5284 | .file_offset = offsetof(struct cgroup, psi_files[PSI_CPU]), |
2ce7135a | 5285 | .seq_show = cgroup_cpu_pressure_show, |
0e94682b SB |
5286 | .write = cgroup_cpu_pressure_write, |
5287 | .poll = cgroup_pressure_poll, | |
5288 | .release = cgroup_pressure_release, | |
2ce7135a | 5289 | }, |
52b1364b CZ |
5290 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
5291 | { | |
5292 | .name = "irq.pressure", | |
34f26a15 | 5293 | .file_offset = offsetof(struct cgroup, psi_files[PSI_IRQ]), |
52b1364b CZ |
5294 | .seq_show = cgroup_irq_pressure_show, |
5295 | .write = cgroup_irq_pressure_write, | |
5296 | .poll = cgroup_pressure_poll, | |
5297 | .release = cgroup_pressure_release, | |
5298 | }, | |
5299 | #endif | |
34f26a15 CZ |
5300 | { |
5301 | .name = "cgroup.pressure", | |
34f26a15 CZ |
5302 | .seq_show = cgroup_pressure_show, |
5303 | .write = cgroup_pressure_write, | |
5304 | }, | |
0e94682b | 5305 | #endif /* CONFIG_PSI */ |
a14c6874 TH |
5306 | { } /* terminate */ |
5307 | }; | |
d5c56ced | 5308 | |
0c21ead1 TH |
5309 | /* |
5310 | * css destruction is four-stage process. | |
5311 | * | |
5312 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
5313 | * Implemented in kill_css(). | |
5314 | * | |
5315 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
5316 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
5317 | * offlined by invoking offline_css(). After offlining, the base ref is | |
5318 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
5319 | * |
5320 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
5321 | * accessors are inside RCU read sections. css_release() schedules the | |
5322 | * RCU callback. | |
5323 | * | |
5324 | * 4. After the grace period, the css can be freed. Implemented in | |
5325 | * css_free_work_fn(). | |
5326 | * | |
5327 | * It is actually hairier because both step 2 and 4 require process context | |
5328 | * and thus involve punting to css->destroy_work adding two additional | |
5329 | * steps to the already complex sequence. | |
5330 | */ | |
8f36aaec | 5331 | static void css_free_rwork_fn(struct work_struct *work) |
48ddbe19 | 5332 | { |
8f36aaec TH |
5333 | struct cgroup_subsys_state *css = container_of(to_rcu_work(work), |
5334 | struct cgroup_subsys_state, destroy_rwork); | |
01e58659 | 5335 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 5336 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 5337 | |
9a1049da TH |
5338 | percpu_ref_exit(&css->refcnt); |
5339 | ||
01e58659 | 5340 | if (ss) { |
9d755d33 | 5341 | /* css free path */ |
8bb5ef79 | 5342 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
5343 | int id = css->id; |
5344 | ||
01e58659 VD |
5345 | ss->css_free(css); |
5346 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 5347 | cgroup_put(cgrp); |
8bb5ef79 TH |
5348 | |
5349 | if (parent) | |
5350 | css_put(parent); | |
9d755d33 TH |
5351 | } else { |
5352 | /* cgroup free path */ | |
5353 | atomic_dec(&cgrp->root->nr_cgrps); | |
d62beb7f | 5354 | cgroup1_pidlist_destroy_all(cgrp); |
971ff493 | 5355 | cancel_work_sync(&cgrp->release_agent_work); |
c4bcfb38 | 5356 | bpf_cgrp_storage_free(cgrp); |
9d755d33 | 5357 | |
d51f39b0 | 5358 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
5359 | /* |
5360 | * We get a ref to the parent, and put the ref when | |
5361 | * this cgroup is being freed, so it's guaranteed | |
5362 | * that the parent won't be destroyed before its | |
5363 | * children. | |
5364 | */ | |
d51f39b0 | 5365 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 | 5366 | kernfs_put(cgrp->kn); |
2ce7135a | 5367 | psi_cgroup_free(cgrp); |
a7df69b8 | 5368 | cgroup_rstat_exit(cgrp); |
9d755d33 TH |
5369 | kfree(cgrp); |
5370 | } else { | |
5371 | /* | |
5372 | * This is root cgroup's refcnt reaching zero, | |
5373 | * which indicates that the root should be | |
5374 | * released. | |
5375 | */ | |
5376 | cgroup_destroy_root(cgrp->root); | |
5377 | } | |
5378 | } | |
48ddbe19 TH |
5379 | } |
5380 | ||
25e15d83 | 5381 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
5382 | { |
5383 | struct cgroup_subsys_state *css = | |
25e15d83 | 5384 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 5385 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 5386 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 5387 | |
1fed1b2e TH |
5388 | mutex_lock(&cgroup_mutex); |
5389 | ||
de3f0341 | 5390 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
5391 | list_del_rcu(&css->sibling); |
5392 | ||
9d755d33 TH |
5393 | if (ss) { |
5394 | /* css release path */ | |
8f53470b TH |
5395 | if (!list_empty(&css->rstat_css_node)) { |
5396 | cgroup_rstat_flush(cgrp); | |
5397 | list_del_rcu(&css->rstat_css_node); | |
5398 | } | |
5399 | ||
01e58659 | 5400 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
5401 | if (ss->css_released) |
5402 | ss->css_released(css); | |
9d755d33 | 5403 | } else { |
0679dee0 RG |
5404 | struct cgroup *tcgrp; |
5405 | ||
9d755d33 | 5406 | /* cgroup release path */ |
e4f8d81c | 5407 | TRACE_CGROUP_PATH(release, cgrp); |
ed1777de | 5408 | |
a7df69b8 | 5409 | cgroup_rstat_flush(cgrp); |
041cd640 | 5410 | |
4dcabece | 5411 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
5412 | for (tcgrp = cgroup_parent(cgrp); tcgrp; |
5413 | tcgrp = cgroup_parent(tcgrp)) | |
5414 | tcgrp->nr_dying_descendants--; | |
4dcabece | 5415 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5416 | |
a4189487 LZ |
5417 | /* |
5418 | * There are two control paths which try to determine | |
5419 | * cgroup from dentry without going through kernfs - | |
5420 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
5421 | * Those are supported by RCU protecting clearing of | |
5422 | * cgrp->kn->priv backpointer. | |
5423 | */ | |
6cd0f5bb TH |
5424 | if (cgrp->kn) |
5425 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
5426 | NULL); | |
9d755d33 | 5427 | } |
d3daf28d | 5428 | |
1fed1b2e TH |
5429 | mutex_unlock(&cgroup_mutex); |
5430 | ||
8f36aaec TH |
5431 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
5432 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
d3daf28d TH |
5433 | } |
5434 | ||
d3daf28d TH |
5435 | static void css_release(struct percpu_ref *ref) |
5436 | { | |
5437 | struct cgroup_subsys_state *css = | |
5438 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5439 | ||
25e15d83 TH |
5440 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5441 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5442 | } |
5443 | ||
ddfcadab TH |
5444 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5445 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5446 | { |
0cb51d71 TH |
5447 | lockdep_assert_held(&cgroup_mutex); |
5448 | ||
a590b90d | 5449 | cgroup_get_live(cgrp); |
ddfcadab | 5450 | |
d5c419b6 | 5451 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5452 | css->cgroup = cgrp; |
72c97e54 | 5453 | css->ss = ss; |
8fa3b8d6 | 5454 | css->id = -1; |
d5c419b6 TH |
5455 | INIT_LIST_HEAD(&css->sibling); |
5456 | INIT_LIST_HEAD(&css->children); | |
8f53470b | 5457 | INIT_LIST_HEAD(&css->rstat_css_node); |
0cb51d71 | 5458 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5459 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5460 | |
d51f39b0 TH |
5461 | if (cgroup_parent(cgrp)) { |
5462 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5463 | css_get(css->parent); |
ddfcadab | 5464 | } |
48ddbe19 | 5465 | |
a7df69b8 | 5466 | if (ss->css_rstat_flush) |
8f53470b TH |
5467 | list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list); |
5468 | ||
ca8bdcaf | 5469 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5470 | } |
5471 | ||
2a4ac633 | 5472 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5473 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5474 | { |
623f926b | 5475 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5476 | int ret = 0; |
5477 | ||
a31f2d3f TH |
5478 | lockdep_assert_held(&cgroup_mutex); |
5479 | ||
92fb9748 | 5480 | if (ss->css_online) |
eb95419b | 5481 | ret = ss->css_online(css); |
ae7f164a | 5482 | if (!ret) { |
eb95419b | 5483 | css->flags |= CSS_ONLINE; |
aec25020 | 5484 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5485 | |
5486 | atomic_inc(&css->online_cnt); | |
5487 | if (css->parent) | |
5488 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5489 | } |
b1929db4 | 5490 | return ret; |
a31f2d3f TH |
5491 | } |
5492 | ||
2a4ac633 | 5493 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5494 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5495 | { |
623f926b | 5496 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5497 | |
5498 | lockdep_assert_held(&cgroup_mutex); | |
5499 | ||
5500 | if (!(css->flags & CSS_ONLINE)) | |
5501 | return; | |
5502 | ||
d7eeac19 | 5503 | if (ss->css_offline) |
eb95419b | 5504 | ss->css_offline(css); |
a31f2d3f | 5505 | |
eb95419b | 5506 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5507 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5508 | |
5509 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5510 | } |
5511 | ||
c81c925a | 5512 | /** |
6cd0f5bb | 5513 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5514 | * @cgrp: the cgroup new css will be associated with |
5515 | * @ss: the subsys of new css | |
5516 | * | |
5517 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5518 | * css is online and installed in @cgrp. This function doesn't create the |
5519 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5520 | */ |
6cd0f5bb TH |
5521 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5522 | struct cgroup_subsys *ss) | |
c81c925a | 5523 | { |
d51f39b0 | 5524 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5525 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5526 | struct cgroup_subsys_state *css; |
5527 | int err; | |
5528 | ||
c81c925a TH |
5529 | lockdep_assert_held(&cgroup_mutex); |
5530 | ||
1fed1b2e | 5531 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
5532 | if (!css) |
5533 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 5534 | if (IS_ERR(css)) |
6cd0f5bb | 5535 | return css; |
c81c925a | 5536 | |
ddfcadab | 5537 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5538 | |
2aad2a86 | 5539 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5540 | if (err) |
3eb59ec6 | 5541 | goto err_free_css; |
c81c925a | 5542 | |
cf780b7d | 5543 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 5544 | if (err < 0) |
b00c52da | 5545 | goto err_free_css; |
15a4c835 | 5546 | css->id = err; |
c81c925a | 5547 | |
15a4c835 | 5548 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5549 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5550 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5551 | |
5552 | err = online_css(css); | |
5553 | if (err) | |
1fed1b2e | 5554 | goto err_list_del; |
94419627 | 5555 | |
6cd0f5bb | 5556 | return css; |
c81c925a | 5557 | |
1fed1b2e TH |
5558 | err_list_del: |
5559 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 5560 | err_free_css: |
8f53470b | 5561 | list_del_rcu(&css->rstat_css_node); |
8f36aaec TH |
5562 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
5563 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
6cd0f5bb | 5564 | return ERR_PTR(err); |
c81c925a TH |
5565 | } |
5566 | ||
07cd1294 TH |
5567 | /* |
5568 | * The returned cgroup is fully initialized including its control mask, but | |
5569 | * it isn't associated with its kernfs_node and doesn't have the control | |
5570 | * mask applied. | |
5571 | */ | |
74321038 TH |
5572 | static struct cgroup *cgroup_create(struct cgroup *parent, const char *name, |
5573 | umode_t mode) | |
ddbcc7e8 | 5574 | { |
a5bca215 | 5575 | struct cgroup_root *root = parent->root; |
a5bca215 | 5576 | struct cgroup *cgrp, *tcgrp; |
74321038 | 5577 | struct kernfs_node *kn; |
a5bca215 | 5578 | int level = parent->level + 1; |
03970d3c | 5579 | int ret; |
ddbcc7e8 | 5580 | |
0a950f65 | 5581 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
7f203bc8 | 5582 | cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL); |
a5bca215 TH |
5583 | if (!cgrp) |
5584 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5585 | |
2aad2a86 | 5586 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5587 | if (ret) |
5588 | goto out_free_cgrp; | |
5589 | ||
a7df69b8 JW |
5590 | ret = cgroup_rstat_init(cgrp); |
5591 | if (ret) | |
5592 | goto out_cancel_ref; | |
041cd640 | 5593 | |
74321038 TH |
5594 | /* create the directory */ |
5595 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5596 | if (IS_ERR(kn)) { | |
5597 | ret = PTR_ERR(kn); | |
041cd640 | 5598 | goto out_stat_exit; |
976c06bc | 5599 | } |
74321038 | 5600 | cgrp->kn = kn; |
976c06bc | 5601 | |
cc31edce | 5602 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5603 | |
9d800df1 | 5604 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5605 | cgrp->root = root; |
b11cfb58 | 5606 | cgrp->level = level; |
2ce7135a JW |
5607 | |
5608 | ret = psi_cgroup_alloc(cgrp); | |
324bda9e | 5609 | if (ret) |
74321038 | 5610 | goto out_kernfs_remove; |
b11cfb58 | 5611 | |
2ce7135a JW |
5612 | ret = cgroup_bpf_inherit(cgrp); |
5613 | if (ret) | |
5614 | goto out_psi_free; | |
5615 | ||
76f969e8 RG |
5616 | /* |
5617 | * New cgroup inherits effective freeze counter, and | |
5618 | * if the parent has to be frozen, the child has too. | |
5619 | */ | |
5620 | cgrp->freezer.e_freeze = parent->freezer.e_freeze; | |
97a61369 RG |
5621 | if (cgrp->freezer.e_freeze) { |
5622 | /* | |
5623 | * Set the CGRP_FREEZE flag, so when a process will be | |
5624 | * attached to the child cgroup, it will become frozen. | |
5625 | * At this point the new cgroup is unpopulated, so we can | |
5626 | * consider it frozen immediately. | |
5627 | */ | |
5628 | set_bit(CGRP_FREEZE, &cgrp->flags); | |
76f969e8 | 5629 | set_bit(CGRP_FROZEN, &cgrp->flags); |
97a61369 | 5630 | } |
76f969e8 | 5631 | |
4dcabece | 5632 | spin_lock_irq(&css_set_lock); |
0679dee0 | 5633 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
7f203bc8 | 5634 | cgrp->ancestors[tcgrp->level] = tcgrp; |
ddbcc7e8 | 5635 | |
76f969e8 | 5636 | if (tcgrp != cgrp) { |
0679dee0 | 5637 | tcgrp->nr_descendants++; |
76f969e8 RG |
5638 | |
5639 | /* | |
5640 | * If the new cgroup is frozen, all ancestor cgroups | |
5641 | * get a new frozen descendant, but their state can't | |
5642 | * change because of this. | |
5643 | */ | |
5644 | if (cgrp->freezer.e_freeze) | |
5645 | tcgrp->freezer.nr_frozen_descendants++; | |
5646 | } | |
0679dee0 | 5647 | } |
4dcabece | 5648 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5649 | |
b6abdb0e LZ |
5650 | if (notify_on_release(parent)) |
5651 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5652 | ||
2260e7fc TH |
5653 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5654 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5655 | |
0cb51d71 | 5656 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5657 | |
4e139afc | 5658 | /* allocation complete, commit to creation */ |
d5c419b6 | 5659 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5660 | atomic_inc(&root->nr_cgrps); |
a590b90d | 5661 | cgroup_get_live(parent); |
415cf07a | 5662 | |
bd53d617 TH |
5663 | /* |
5664 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5665 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5666 | */ |
03970d3c | 5667 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5668 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5669 | |
5670 | cgroup_propagate_control(cgrp); | |
5671 | ||
a5bca215 TH |
5672 | return cgrp; |
5673 | ||
2ce7135a JW |
5674 | out_psi_free: |
5675 | psi_cgroup_free(cgrp); | |
74321038 TH |
5676 | out_kernfs_remove: |
5677 | kernfs_remove(cgrp->kn); | |
041cd640 | 5678 | out_stat_exit: |
a7df69b8 | 5679 | cgroup_rstat_exit(cgrp); |
a5bca215 TH |
5680 | out_cancel_ref: |
5681 | percpu_ref_exit(&cgrp->self.refcnt); | |
5682 | out_free_cgrp: | |
5683 | kfree(cgrp); | |
5684 | return ERR_PTR(ret); | |
a5bca215 TH |
5685 | } |
5686 | ||
1a926e0b RG |
5687 | static bool cgroup_check_hierarchy_limits(struct cgroup *parent) |
5688 | { | |
5689 | struct cgroup *cgroup; | |
5690 | int ret = false; | |
5691 | int level = 1; | |
5692 | ||
5693 | lockdep_assert_held(&cgroup_mutex); | |
5694 | ||
5695 | for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) { | |
5696 | if (cgroup->nr_descendants >= cgroup->max_descendants) | |
5697 | goto fail; | |
5698 | ||
5699 | if (level > cgroup->max_depth) | |
5700 | goto fail; | |
5701 | ||
5702 | level++; | |
5703 | } | |
5704 | ||
5705 | ret = true; | |
5706 | fail: | |
5707 | return ret; | |
5708 | } | |
5709 | ||
1592c9b2 | 5710 | int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode) |
a5bca215 TH |
5711 | { |
5712 | struct cgroup *parent, *cgrp; | |
03970d3c | 5713 | int ret; |
a5bca215 TH |
5714 | |
5715 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5716 | if (strchr(name, '\n')) | |
5717 | return -EINVAL; | |
5718 | ||
945ba199 | 5719 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5720 | if (!parent) |
5721 | return -ENODEV; | |
5722 | ||
1a926e0b RG |
5723 | if (!cgroup_check_hierarchy_limits(parent)) { |
5724 | ret = -EAGAIN; | |
5725 | goto out_unlock; | |
5726 | } | |
5727 | ||
74321038 | 5728 | cgrp = cgroup_create(parent, name, mode); |
a5bca215 TH |
5729 | if (IS_ERR(cgrp)) { |
5730 | ret = PTR_ERR(cgrp); | |
5731 | goto out_unlock; | |
5732 | } | |
5733 | ||
195e9b6c TH |
5734 | /* |
5735 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5736 | * that @cgrp->kn is always accessible. | |
5737 | */ | |
74321038 | 5738 | kernfs_get(cgrp->kn); |
195e9b6c | 5739 | |
74321038 | 5740 | ret = cgroup_kn_set_ugid(cgrp->kn); |
195e9b6c TH |
5741 | if (ret) |
5742 | goto out_destroy; | |
5743 | ||
334c3679 | 5744 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5745 | if (ret) |
5746 | goto out_destroy; | |
5747 | ||
03970d3c TH |
5748 | ret = cgroup_apply_control_enable(cgrp); |
5749 | if (ret) | |
5750 | goto out_destroy; | |
195e9b6c | 5751 | |
e4f8d81c | 5752 | TRACE_CGROUP_PATH(mkdir, cgrp); |
ed1777de | 5753 | |
195e9b6c | 5754 | /* let's create and online css's */ |
74321038 | 5755 | kernfs_activate(cgrp->kn); |
ddbcc7e8 | 5756 | |
ba0f4d76 TH |
5757 | ret = 0; |
5758 | goto out_unlock; | |
ddbcc7e8 | 5759 | |
a5bca215 TH |
5760 | out_destroy: |
5761 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5762 | out_unlock: |
a9746d8d | 5763 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5764 | return ret; |
ddbcc7e8 PM |
5765 | } |
5766 | ||
223dbc38 TH |
5767 | /* |
5768 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 | 5769 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
08b2b6fd | 5770 | * initiate destruction and put the css ref from kill_css(). |
223dbc38 TH |
5771 | */ |
5772 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5773 | { |
223dbc38 TH |
5774 | struct cgroup_subsys_state *css = |
5775 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5776 | |
f20104de | 5777 | mutex_lock(&cgroup_mutex); |
09a503ea | 5778 | |
aa226ff4 TH |
5779 | do { |
5780 | offline_css(css); | |
5781 | css_put(css); | |
5782 | /* @css can't go away while we're holding cgroup_mutex */ | |
5783 | css = css->parent; | |
5784 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5785 | ||
5786 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5787 | } |
5788 | ||
223dbc38 TH |
5789 | /* css kill confirmation processing requires process context, bounce */ |
5790 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5791 | { |
5792 | struct cgroup_subsys_state *css = | |
5793 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5794 | ||
aa226ff4 TH |
5795 | if (atomic_dec_and_test(&css->online_cnt)) { |
5796 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5797 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5798 | } | |
d3daf28d TH |
5799 | } |
5800 | ||
f392e51c TH |
5801 | /** |
5802 | * kill_css - destroy a css | |
5803 | * @css: css to destroy | |
5804 | * | |
5805 | * This function initiates destruction of @css by removing cgroup interface | |
5806 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5807 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5808 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5809 | */ |
5810 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5811 | { |
01f6474c | 5812 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5813 | |
33c35aa4 WL |
5814 | if (css->flags & CSS_DYING) |
5815 | return; | |
5816 | ||
5817 | css->flags |= CSS_DYING; | |
5818 | ||
2bd59d48 TH |
5819 | /* |
5820 | * This must happen before css is disassociated with its cgroup. | |
5821 | * See seq_css() for details. | |
5822 | */ | |
334c3679 | 5823 | css_clear_dir(css); |
3c14f8b4 | 5824 | |
edae0c33 TH |
5825 | /* |
5826 | * Killing would put the base ref, but we need to keep it alive | |
5827 | * until after ->css_offline(). | |
5828 | */ | |
5829 | css_get(css); | |
5830 | ||
5831 | /* | |
5832 | * cgroup core guarantees that, by the time ->css_offline() is | |
5833 | * invoked, no new css reference will be given out via | |
ec903c0c | 5834 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5835 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5836 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5837 | * | |
5838 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5839 | * css is confirmed to be seen as killed on all CPUs. | |
5840 | */ | |
5841 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5842 | } |
5843 | ||
5844 | /** | |
5845 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5846 | * @cgrp: cgroup to be destroyed | |
5847 | * | |
5848 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5849 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5850 | * guarantee that css_tryget_online() won't succeed by the time |
5851 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5852 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5853 | * |
5854 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5855 | * userland visible parts and start killing the percpu refcnts of | |
5856 | * css's. Set up so that the next stage will be kicked off once all | |
5857 | * the percpu refcnts are confirmed to be killed. | |
5858 | * | |
5859 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5860 | * rest of destruction. Once all cgroup references are gone, the | |
5861 | * cgroup is RCU-freed. | |
5862 | * | |
5863 | * This function implements s1. After this step, @cgrp is gone as far as | |
5864 | * the userland is concerned and a new cgroup with the same name may be | |
5865 | * created. As cgroup doesn't care about the names internally, this | |
5866 | * doesn't cause any problem. | |
5867 | */ | |
42809dd4 TH |
5868 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5869 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5870 | { |
0679dee0 | 5871 | struct cgroup *tcgrp, *parent = cgroup_parent(cgrp); |
2bd59d48 | 5872 | struct cgroup_subsys_state *css; |
2b021cbf | 5873 | struct cgrp_cset_link *link; |
1c6727af | 5874 | int ssid; |
ddbcc7e8 | 5875 | |
42809dd4 TH |
5876 | lockdep_assert_held(&cgroup_mutex); |
5877 | ||
91486f61 TH |
5878 | /* |
5879 | * Only migration can raise populated from zero and we're already | |
5880 | * holding cgroup_mutex. | |
5881 | */ | |
5882 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5883 | return -EBUSY; |
a043e3b2 | 5884 | |
bb78a92f | 5885 | /* |
d5c419b6 TH |
5886 | * Make sure there's no live children. We can't test emptiness of |
5887 | * ->self.children as dead children linger on it while being | |
5888 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5889 | */ |
f3d46500 | 5890 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5891 | return -EBUSY; |
5892 | ||
455050d2 | 5893 | /* |
2b021cbf TH |
5894 | * Mark @cgrp and the associated csets dead. The former prevents |
5895 | * further task migration and child creation by disabling | |
5896 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5897 | * the migration path. | |
455050d2 | 5898 | */ |
184faf32 | 5899 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5900 | |
82d6489d | 5901 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5902 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5903 | link->cset->dead = true; | |
82d6489d | 5904 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5905 | |
249f3468 | 5906 | /* initiate massacre of all css's */ |
1c6727af TH |
5907 | for_each_css(css, ssid, cgrp) |
5908 | kill_css(css); | |
455050d2 | 5909 | |
5faaf05f TH |
5910 | /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */ |
5911 | css_clear_dir(&cgrp->self); | |
01f6474c | 5912 | kernfs_remove(cgrp->kn); |
f20104de | 5913 | |
b154a017 | 5914 | if (cgroup_is_threaded(cgrp)) |
454000ad TH |
5915 | parent->nr_threaded_children--; |
5916 | ||
4dcabece | 5917 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
5918 | for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
5919 | tcgrp->nr_descendants--; | |
5920 | tcgrp->nr_dying_descendants++; | |
76f969e8 RG |
5921 | /* |
5922 | * If the dying cgroup is frozen, decrease frozen descendants | |
5923 | * counters of ancestor cgroups. | |
5924 | */ | |
5925 | if (test_bit(CGRP_FROZEN, &cgrp->flags)) | |
5926 | tcgrp->freezer.nr_frozen_descendants--; | |
0679dee0 | 5927 | } |
4dcabece | 5928 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5929 | |
5a621e6c | 5930 | cgroup1_check_for_release(parent); |
2bd59d48 | 5931 | |
4bfc0bb2 RG |
5932 | cgroup_bpf_offline(cgrp); |
5933 | ||
249f3468 | 5934 | /* put the base reference */ |
9d755d33 | 5935 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5936 | |
ea15f8cc TH |
5937 | return 0; |
5938 | }; | |
5939 | ||
1592c9b2 | 5940 | int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5941 | { |
a9746d8d | 5942 | struct cgroup *cgrp; |
2bd59d48 | 5943 | int ret = 0; |
42809dd4 | 5944 | |
945ba199 | 5945 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5946 | if (!cgrp) |
5947 | return 0; | |
42809dd4 | 5948 | |
a9746d8d | 5949 | ret = cgroup_destroy_locked(cgrp); |
ed1777de | 5950 | if (!ret) |
e4f8d81c | 5951 | TRACE_CGROUP_PATH(rmdir, cgrp); |
ed1777de | 5952 | |
a9746d8d | 5953 | cgroup_kn_unlock(kn); |
42809dd4 | 5954 | return ret; |
8e3f6541 TH |
5955 | } |
5956 | ||
2bd59d48 | 5957 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5136f636 | 5958 | .show_options = cgroup_show_options, |
2bd59d48 TH |
5959 | .mkdir = cgroup_mkdir, |
5960 | .rmdir = cgroup_rmdir, | |
4f41fc59 | 5961 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5962 | }; |
5963 | ||
15a4c835 | 5964 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5965 | { |
ddbcc7e8 | 5966 | struct cgroup_subsys_state *css; |
cfe36bde | 5967 | |
a5ae9899 | 5968 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5969 | |
648bb56d TH |
5970 | mutex_lock(&cgroup_mutex); |
5971 | ||
15a4c835 | 5972 | idr_init(&ss->css_idr); |
0adb0704 | 5973 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5974 | |
3dd06ffa TH |
5975 | /* Create the root cgroup state for this subsystem */ |
5976 | ss->root = &cgrp_dfl_root; | |
8291471e | 5977 | css = ss->css_alloc(NULL); |
ddbcc7e8 PM |
5978 | /* We don't handle early failures gracefully */ |
5979 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5980 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5981 | |
5982 | /* | |
5983 | * Root csses are never destroyed and we can't initialize | |
5984 | * percpu_ref during early init. Disable refcnting. | |
5985 | */ | |
5986 | css->flags |= CSS_NO_REF; | |
5987 | ||
15a4c835 | 5988 | if (early) { |
9395a450 | 5989 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5990 | css->id = 1; |
5991 | } else { | |
5992 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5993 | BUG_ON(css->id < 0); | |
5994 | } | |
ddbcc7e8 | 5995 | |
e8d55fde | 5996 | /* Update the init_css_set to contain a subsys |
817929ec | 5997 | * pointer to this state - since the subsystem is |
e8d55fde | 5998 | * newly registered, all tasks and hence the |
3dd06ffa | 5999 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 6000 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 6001 | |
cb4a3167 AS |
6002 | have_fork_callback |= (bool)ss->fork << ss->id; |
6003 | have_exit_callback |= (bool)ss->exit << ss->id; | |
51bee5ab | 6004 | have_release_callback |= (bool)ss->release << ss->id; |
7e47682e | 6005 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 6006 | |
e8d55fde LZ |
6007 | /* At system boot, before all subsystems have been |
6008 | * registered, no tasks have been forked, so we don't | |
6009 | * need to invoke fork callbacks here. */ | |
6010 | BUG_ON(!list_empty(&init_task.tasks)); | |
6011 | ||
ae7f164a | 6012 | BUG_ON(online_css(css)); |
a8638030 | 6013 | |
cf5d5941 BB |
6014 | mutex_unlock(&cgroup_mutex); |
6015 | } | |
cf5d5941 | 6016 | |
ddbcc7e8 | 6017 | /** |
a043e3b2 LZ |
6018 | * cgroup_init_early - cgroup initialization at system boot |
6019 | * | |
6020 | * Initialize cgroups at system boot, and initialize any | |
6021 | * subsystems that request early init. | |
ddbcc7e8 PM |
6022 | */ |
6023 | int __init cgroup_init_early(void) | |
6024 | { | |
f5dfb531 | 6025 | static struct cgroup_fs_context __initdata ctx; |
30159ec7 | 6026 | struct cgroup_subsys *ss; |
ddbcc7e8 | 6027 | int i; |
30159ec7 | 6028 | |
cf6299b1 AV |
6029 | ctx.root = &cgrp_dfl_root; |
6030 | init_cgroup_root(&ctx); | |
3b514d24 TH |
6031 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
6032 | ||
a4ea1cc9 | 6033 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 6034 | |
3ed80a62 | 6035 | for_each_subsys(ss, i) { |
aec25020 | 6036 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 6037 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 6038 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 6039 | ss->id, ss->name); |
073219e9 TH |
6040 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
6041 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
6042 | ||
aec25020 | 6043 | ss->id = i; |
073219e9 | 6044 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
6045 | if (!ss->legacy_name) |
6046 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
6047 | |
6048 | if (ss->early_init) | |
15a4c835 | 6049 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
6050 | } |
6051 | return 0; | |
6052 | } | |
6053 | ||
6054 | /** | |
a043e3b2 LZ |
6055 | * cgroup_init - cgroup initialization |
6056 | * | |
6057 | * Register cgroup filesystem and /proc file, and initialize | |
6058 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
6059 | */ |
6060 | int __init cgroup_init(void) | |
6061 | { | |
30159ec7 | 6062 | struct cgroup_subsys *ss; |
035f4f51 | 6063 | int ssid; |
ddbcc7e8 | 6064 | |
6e5c8307 | 6065 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
d62beb7f | 6066 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
8a693f77 | 6067 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_psi_files)); |
d62beb7f | 6068 | BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); |
ddbcc7e8 | 6069 | |
c58632b3 | 6070 | cgroup_rstat_boot(); |
041cd640 | 6071 | |
a79a908f AK |
6072 | get_user_ns(init_cgroup_ns.user_ns); |
6073 | ||
54e7b4eb | 6074 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 6075 | |
2378d8b8 TH |
6076 | /* |
6077 | * Add init_css_set to the hash table so that dfl_root can link to | |
6078 | * it during init. | |
6079 | */ | |
6080 | hash_add(css_set_table, &init_css_set.hlist, | |
6081 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 6082 | |
35ac1184 | 6083 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 6084 | |
54e7b4eb TH |
6085 | mutex_unlock(&cgroup_mutex); |
6086 | ||
172a2c06 | 6087 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
6088 | if (ss->early_init) { |
6089 | struct cgroup_subsys_state *css = | |
6090 | init_css_set.subsys[ss->id]; | |
6091 | ||
6092 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
6093 | GFP_KERNEL); | |
6094 | BUG_ON(css->id < 0); | |
6095 | } else { | |
6096 | cgroup_init_subsys(ss, false); | |
6097 | } | |
172a2c06 | 6098 | |
2d8f243a TH |
6099 | list_add_tail(&init_css_set.e_cset_node[ssid], |
6100 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
6101 | |
6102 | /* | |
c731ae1d LZ |
6103 | * Setting dfl_root subsys_mask needs to consider the |
6104 | * disabled flag and cftype registration needs kmalloc, | |
6105 | * both of which aren't available during early_init. | |
172a2c06 | 6106 | */ |
45e1ba40 | 6107 | if (!cgroup_ssid_enabled(ssid)) |
a8ddc821 TH |
6108 | continue; |
6109 | ||
d62beb7f | 6110 | if (cgroup1_ssid_disabled(ssid)) |
223ffb29 JW |
6111 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
6112 | ss->name); | |
6113 | ||
a8ddc821 TH |
6114 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
6115 | ||
8cfd8147 TH |
6116 | /* implicit controllers must be threaded too */ |
6117 | WARN_ON(ss->implicit_on_dfl && !ss->threaded); | |
6118 | ||
f6d635ad TH |
6119 | if (ss->implicit_on_dfl) |
6120 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
6121 | else if (!ss->dfl_cftypes) | |
a7165264 | 6122 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 6123 | |
8cfd8147 TH |
6124 | if (ss->threaded) |
6125 | cgrp_dfl_threaded_ss_mask |= 1 << ss->id; | |
6126 | ||
a8ddc821 TH |
6127 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
6128 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
6129 | } else { | |
6130 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
6131 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 6132 | } |
295458e6 VD |
6133 | |
6134 | if (ss->bind) | |
6135 | ss->bind(init_css_set.subsys[ssid]); | |
7af608e4 TH |
6136 | |
6137 | mutex_lock(&cgroup_mutex); | |
6138 | css_populate_dir(init_css_set.subsys[ssid]); | |
6139 | mutex_unlock(&cgroup_mutex); | |
676db4af GK |
6140 | } |
6141 | ||
2378d8b8 TH |
6142 | /* init_css_set.subsys[] has been updated, re-hash */ |
6143 | hash_del(&init_css_set.hlist); | |
6144 | hash_add(css_set_table, &init_css_set.hlist, | |
6145 | css_set_hash(init_css_set.subsys)); | |
6146 | ||
035f4f51 TH |
6147 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
6148 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 6149 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
3f3942ac | 6150 | WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show)); |
d5f68d33 AV |
6151 | #ifdef CONFIG_CPUSETS |
6152 | WARN_ON(register_filesystem(&cpuset_fs_type)); | |
6153 | #endif | |
ddbcc7e8 | 6154 | |
2bd59d48 | 6155 | return 0; |
ddbcc7e8 | 6156 | } |
b4f48b63 | 6157 | |
e5fca243 TH |
6158 | static int __init cgroup_wq_init(void) |
6159 | { | |
6160 | /* | |
6161 | * There isn't much point in executing destruction path in | |
6162 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 6163 | * Use 1 for @max_active. |
e5fca243 TH |
6164 | * |
6165 | * We would prefer to do this in cgroup_init() above, but that | |
6166 | * is called before init_workqueues(): so leave this until after. | |
6167 | */ | |
1a11533f | 6168 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 TH |
6169 | BUG_ON(!cgroup_destroy_wq); |
6170 | return 0; | |
6171 | } | |
6172 | core_initcall(cgroup_wq_init); | |
6173 | ||
67c0496e | 6174 | void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen) |
69fd5c39 SL |
6175 | { |
6176 | struct kernfs_node *kn; | |
6177 | ||
fe0f726c | 6178 | kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id); |
69fd5c39 SL |
6179 | if (!kn) |
6180 | return; | |
6181 | kernfs_path(kn, buf, buflen); | |
6182 | kernfs_put(kn); | |
6183 | } | |
6184 | ||
6b658c48 MK |
6185 | /* |
6186 | * cgroup_get_from_id : get the cgroup associated with cgroup id | |
6187 | * @id: cgroup id | |
fa7e439c | 6188 | * On success return the cgrp or ERR_PTR on failure |
4534dee9 | 6189 | * Only cgroups within current task's cgroup NS are valid. |
6b658c48 MK |
6190 | */ |
6191 | struct cgroup *cgroup_get_from_id(u64 id) | |
6192 | { | |
6193 | struct kernfs_node *kn; | |
7e1eb543 | 6194 | struct cgroup *cgrp, *root_cgrp; |
6b658c48 | 6195 | |
6b658c48 MK |
6196 | kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id); |
6197 | if (!kn) | |
7e1eb543 | 6198 | return ERR_PTR(-ENOENT); |
6b658c48 | 6199 | |
7e1eb543 TH |
6200 | if (kernfs_type(kn) != KERNFS_DIR) { |
6201 | kernfs_put(kn); | |
6202 | return ERR_PTR(-ENOENT); | |
6203 | } | |
df02452f | 6204 | |
be288169 | 6205 | rcu_read_lock(); |
6b658c48 | 6206 | |
be288169 SB |
6207 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
6208 | if (cgrp && !cgroup_tryget(cgrp)) | |
6b658c48 | 6209 | cgrp = NULL; |
be288169 SB |
6210 | |
6211 | rcu_read_unlock(); | |
6b658c48 | 6212 | kernfs_put(kn); |
4534dee9 MK |
6213 | |
6214 | if (!cgrp) | |
7e1eb543 | 6215 | return ERR_PTR(-ENOENT); |
4534dee9 | 6216 | |
46307fd6 | 6217 | root_cgrp = current_cgns_cgroup_dfl(); |
4534dee9 MK |
6218 | if (!cgroup_is_descendant(cgrp, root_cgrp)) { |
6219 | cgroup_put(cgrp); | |
7e1eb543 | 6220 | return ERR_PTR(-ENOENT); |
4534dee9 | 6221 | } |
7e1eb543 | 6222 | |
6b658c48 MK |
6223 | return cgrp; |
6224 | } | |
6225 | EXPORT_SYMBOL_GPL(cgroup_get_from_id); | |
6226 | ||
a424316c PM |
6227 | /* |
6228 | * proc_cgroup_show() | |
6229 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
6230 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 6231 | */ |
006f4ac4 ZL |
6232 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
6233 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 6234 | { |
4c737b41 | 6235 | char *buf; |
a424316c | 6236 | int retval; |
3dd06ffa | 6237 | struct cgroup_root *root; |
a424316c PM |
6238 | |
6239 | retval = -ENOMEM; | |
e61734c5 | 6240 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
6241 | if (!buf) |
6242 | goto out; | |
6243 | ||
a424316c | 6244 | mutex_lock(&cgroup_mutex); |
82d6489d | 6245 | spin_lock_irq(&css_set_lock); |
a424316c | 6246 | |
985ed670 | 6247 | for_each_root(root) { |
a424316c | 6248 | struct cgroup_subsys *ss; |
bd89aabc | 6249 | struct cgroup *cgrp; |
b85d2040 | 6250 | int ssid, count = 0; |
a424316c | 6251 | |
dc79ec1b | 6252 | if (root == &cgrp_dfl_root && !READ_ONCE(cgrp_dfl_visible)) |
985ed670 TH |
6253 | continue; |
6254 | ||
2c6ab6d2 | 6255 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
6256 | if (root != &cgrp_dfl_root) |
6257 | for_each_subsys(ss, ssid) | |
6258 | if (root->subsys_mask & (1 << ssid)) | |
6259 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 6260 | ss->legacy_name); |
c6d57f33 PM |
6261 | if (strlen(root->name)) |
6262 | seq_printf(m, "%sname=%s", count ? "," : "", | |
6263 | root->name); | |
a424316c | 6264 | seq_putc(m, ':'); |
2e91fa7f | 6265 | |
7717f7ba | 6266 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
6267 | |
6268 | /* | |
6269 | * On traditional hierarchies, all zombie tasks show up as | |
6270 | * belonging to the root cgroup. On the default hierarchy, | |
6271 | * while a zombie doesn't show up in "cgroup.procs" and | |
6272 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
6273 | * reporting the cgroup it belonged to before exiting. If | |
6274 | * the cgroup is removed before the zombie is reaped, | |
6275 | * " (deleted)" is appended to the cgroup path. | |
6276 | */ | |
6277 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 6278 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 6279 | current->nsproxy->cgroup_ns); |
e0223003 | 6280 | if (retval >= PATH_MAX) |
2e91fa7f | 6281 | retval = -ENAMETOOLONG; |
e0223003 | 6282 | if (retval < 0) |
2e91fa7f | 6283 | goto out_unlock; |
4c737b41 TH |
6284 | |
6285 | seq_puts(m, buf); | |
2e91fa7f | 6286 | } else { |
4c737b41 | 6287 | seq_puts(m, "/"); |
e61734c5 | 6288 | } |
2e91fa7f | 6289 | |
2e91fa7f TH |
6290 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
6291 | seq_puts(m, " (deleted)\n"); | |
6292 | else | |
6293 | seq_putc(m, '\n'); | |
a424316c PM |
6294 | } |
6295 | ||
006f4ac4 | 6296 | retval = 0; |
a424316c | 6297 | out_unlock: |
82d6489d | 6298 | spin_unlock_irq(&css_set_lock); |
a424316c | 6299 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
6300 | kfree(buf); |
6301 | out: | |
6302 | return retval; | |
6303 | } | |
6304 | ||
b4f48b63 | 6305 | /** |
eaf797ab | 6306 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 6307 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 6308 | * |
eaf797ab | 6309 | * A task is associated with the init_css_set until cgroup_post_fork() |
ef2c41cf | 6310 | * attaches it to the target css_set. |
b4f48b63 PM |
6311 | */ |
6312 | void cgroup_fork(struct task_struct *child) | |
6313 | { | |
eaf797ab | 6314 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 6315 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
6316 | } |
6317 | ||
a6d1ce59 YA |
6318 | /** |
6319 | * cgroup_v1v2_get_from_file - get a cgroup pointer from a file pointer | |
6320 | * @f: file corresponding to cgroup_dir | |
6321 | * | |
6322 | * Find the cgroup from a file pointer associated with a cgroup directory. | |
6323 | * Returns a pointer to the cgroup on success. ERR_PTR is returned if the | |
6324 | * cgroup cannot be found. | |
6325 | */ | |
6326 | static struct cgroup *cgroup_v1v2_get_from_file(struct file *f) | |
17703097 CB |
6327 | { |
6328 | struct cgroup_subsys_state *css; | |
17703097 CB |
6329 | |
6330 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
6331 | if (IS_ERR(css)) | |
6332 | return ERR_CAST(css); | |
6333 | ||
a6d1ce59 YA |
6334 | return css->cgroup; |
6335 | } | |
6336 | ||
6337 | /** | |
6338 | * cgroup_get_from_file - same as cgroup_v1v2_get_from_file, but only supports | |
6339 | * cgroup2. | |
b675d4bd | 6340 | * @f: file corresponding to cgroup2_dir |
a6d1ce59 YA |
6341 | */ |
6342 | static struct cgroup *cgroup_get_from_file(struct file *f) | |
6343 | { | |
6344 | struct cgroup *cgrp = cgroup_v1v2_get_from_file(f); | |
6345 | ||
6346 | if (IS_ERR(cgrp)) | |
6347 | return ERR_CAST(cgrp); | |
6348 | ||
03db7716 TH |
6349 | if (!cgroup_on_dfl(cgrp)) { |
6350 | cgroup_put(cgrp); | |
6351 | return ERR_PTR(-EBADF); | |
6352 | } | |
6353 | ||
17703097 CB |
6354 | return cgrp; |
6355 | } | |
6356 | ||
ef2c41cf CB |
6357 | /** |
6358 | * cgroup_css_set_fork - find or create a css_set for a child process | |
6359 | * @kargs: the arguments passed to create the child process | |
6360 | * | |
6361 | * This functions finds or creates a new css_set which the child | |
6362 | * process will be attached to in cgroup_post_fork(). By default, | |
6363 | * the child process will be given the same css_set as its parent. | |
6364 | * | |
6365 | * If CLONE_INTO_CGROUP is specified this function will try to find an | |
6366 | * existing css_set which includes the requested cgroup and if not create | |
6367 | * a new css_set that the child will be attached to later. If this function | |
6368 | * succeeds it will hold cgroup_threadgroup_rwsem on return. If | |
6369 | * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex | |
6370 | * before grabbing cgroup_threadgroup_rwsem and will hold a reference | |
6371 | * to the target cgroup. | |
6372 | */ | |
6373 | static int cgroup_css_set_fork(struct kernel_clone_args *kargs) | |
6374 | __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem) | |
6375 | { | |
6376 | int ret; | |
6377 | struct cgroup *dst_cgrp = NULL; | |
6378 | struct css_set *cset; | |
6379 | struct super_block *sb; | |
6380 | struct file *f; | |
6381 | ||
6382 | if (kargs->flags & CLONE_INTO_CGROUP) | |
6383 | mutex_lock(&cgroup_mutex); | |
6384 | ||
6385 | cgroup_threadgroup_change_begin(current); | |
6386 | ||
6387 | spin_lock_irq(&css_set_lock); | |
6388 | cset = task_css_set(current); | |
6389 | get_css_set(cset); | |
6390 | spin_unlock_irq(&css_set_lock); | |
6391 | ||
6392 | if (!(kargs->flags & CLONE_INTO_CGROUP)) { | |
6393 | kargs->cset = cset; | |
6394 | return 0; | |
6395 | } | |
6396 | ||
6397 | f = fget_raw(kargs->cgroup); | |
6398 | if (!f) { | |
6399 | ret = -EBADF; | |
6400 | goto err; | |
6401 | } | |
6402 | sb = f->f_path.dentry->d_sb; | |
6403 | ||
6404 | dst_cgrp = cgroup_get_from_file(f); | |
6405 | if (IS_ERR(dst_cgrp)) { | |
6406 | ret = PTR_ERR(dst_cgrp); | |
6407 | dst_cgrp = NULL; | |
6408 | goto err; | |
6409 | } | |
6410 | ||
6411 | if (cgroup_is_dead(dst_cgrp)) { | |
6412 | ret = -ENODEV; | |
6413 | goto err; | |
6414 | } | |
6415 | ||
6416 | /* | |
6417 | * Verify that we the target cgroup is writable for us. This is | |
6418 | * usually done by the vfs layer but since we're not going through | |
6419 | * the vfs layer here we need to do it "manually". | |
6420 | */ | |
6421 | ret = cgroup_may_write(dst_cgrp, sb); | |
6422 | if (ret) | |
6423 | goto err; | |
6424 | ||
6d3971da CB |
6425 | /* |
6426 | * Spawning a task directly into a cgroup works by passing a file | |
6427 | * descriptor to the target cgroup directory. This can even be an O_PATH | |
6428 | * file descriptor. But it can never be a cgroup.procs file descriptor. | |
6429 | * This was done on purpose so spawning into a cgroup could be | |
6430 | * conceptualized as an atomic | |
6431 | * | |
6432 | * fd = openat(dfd_cgroup, "cgroup.procs", ...); | |
6433 | * write(fd, <child-pid>, ...); | |
6434 | * | |
6435 | * sequence, i.e. it's a shorthand for the caller opening and writing | |
6436 | * cgroup.procs of the cgroup indicated by @dfd_cgroup. This allows us | |
6437 | * to always use the caller's credentials. | |
6438 | */ | |
ef2c41cf | 6439 | ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb, |
e5745764 TH |
6440 | !(kargs->flags & CLONE_THREAD), |
6441 | current->nsproxy->cgroup_ns); | |
ef2c41cf CB |
6442 | if (ret) |
6443 | goto err; | |
6444 | ||
6445 | kargs->cset = find_css_set(cset, dst_cgrp); | |
6446 | if (!kargs->cset) { | |
6447 | ret = -ENOMEM; | |
6448 | goto err; | |
6449 | } | |
6450 | ||
6451 | put_css_set(cset); | |
6452 | fput(f); | |
6453 | kargs->cgrp = dst_cgrp; | |
6454 | return ret; | |
6455 | ||
6456 | err: | |
6457 | cgroup_threadgroup_change_end(current); | |
6458 | mutex_unlock(&cgroup_mutex); | |
6459 | if (f) | |
6460 | fput(f); | |
6461 | if (dst_cgrp) | |
6462 | cgroup_put(dst_cgrp); | |
6463 | put_css_set(cset); | |
6464 | if (kargs->cset) | |
6465 | put_css_set(kargs->cset); | |
6466 | return ret; | |
6467 | } | |
6468 | ||
6469 | /** | |
6470 | * cgroup_css_set_put_fork - drop references we took during fork | |
6471 | * @kargs: the arguments passed to create the child process | |
6472 | * | |
6473 | * Drop references to the prepared css_set and target cgroup if | |
6474 | * CLONE_INTO_CGROUP was requested. | |
6475 | */ | |
6476 | static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs) | |
6477 | __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex) | |
6478 | { | |
6479 | cgroup_threadgroup_change_end(current); | |
6480 | ||
6481 | if (kargs->flags & CLONE_INTO_CGROUP) { | |
6482 | struct cgroup *cgrp = kargs->cgrp; | |
6483 | struct css_set *cset = kargs->cset; | |
6484 | ||
6485 | mutex_unlock(&cgroup_mutex); | |
6486 | ||
6487 | if (cset) { | |
6488 | put_css_set(cset); | |
6489 | kargs->cset = NULL; | |
6490 | } | |
6491 | ||
6492 | if (cgrp) { | |
6493 | cgroup_put(cgrp); | |
6494 | kargs->cgrp = NULL; | |
6495 | } | |
6496 | } | |
6497 | } | |
6498 | ||
7e47682e AS |
6499 | /** |
6500 | * cgroup_can_fork - called on a new task before the process is exposed | |
5a5cf5cb | 6501 | * @child: the child process |
ffacbd11 | 6502 | * @kargs: the arguments passed to create the child process |
7e47682e | 6503 | * |
ef2c41cf CB |
6504 | * This prepares a new css_set for the child process which the child will |
6505 | * be attached to in cgroup_post_fork(). | |
5a5cf5cb CB |
6506 | * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork() |
6507 | * callback returns an error, the fork aborts with that error code. This | |
6508 | * allows for a cgroup subsystem to conditionally allow or deny new forks. | |
7e47682e | 6509 | */ |
ef2c41cf | 6510 | int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs) |
7e47682e AS |
6511 | { |
6512 | struct cgroup_subsys *ss; | |
6513 | int i, j, ret; | |
6514 | ||
ef2c41cf CB |
6515 | ret = cgroup_css_set_fork(kargs); |
6516 | if (ret) | |
6517 | return ret; | |
5a5cf5cb | 6518 | |
b4e0eeaf | 6519 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
ef2c41cf | 6520 | ret = ss->can_fork(child, kargs->cset); |
7e47682e AS |
6521 | if (ret) |
6522 | goto out_revert; | |
b4e0eeaf | 6523 | } while_each_subsys_mask(); |
7e47682e AS |
6524 | |
6525 | return 0; | |
6526 | ||
6527 | out_revert: | |
6528 | for_each_subsys(ss, j) { | |
6529 | if (j >= i) | |
6530 | break; | |
6531 | if (ss->cancel_fork) | |
ef2c41cf | 6532 | ss->cancel_fork(child, kargs->cset); |
7e47682e AS |
6533 | } |
6534 | ||
ef2c41cf | 6535 | cgroup_css_set_put_fork(kargs); |
5a5cf5cb | 6536 | |
7e47682e AS |
6537 | return ret; |
6538 | } | |
6539 | ||
6540 | /** | |
6541 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
ef2c41cf CB |
6542 | * @child: the child process |
6543 | * @kargs: the arguments passed to create the child process | |
7e47682e AS |
6544 | * |
6545 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
08b2b6fd | 6546 | * cgroup_can_fork() succeeded and cleans up references we took to |
ef2c41cf | 6547 | * prepare a new css_set for the child process in cgroup_can_fork(). |
7e47682e | 6548 | */ |
ef2c41cf CB |
6549 | void cgroup_cancel_fork(struct task_struct *child, |
6550 | struct kernel_clone_args *kargs) | |
7e47682e AS |
6551 | { |
6552 | struct cgroup_subsys *ss; | |
6553 | int i; | |
6554 | ||
6555 | for_each_subsys(ss, i) | |
6556 | if (ss->cancel_fork) | |
ef2c41cf | 6557 | ss->cancel_fork(child, kargs->cset); |
5a5cf5cb | 6558 | |
ef2c41cf | 6559 | cgroup_css_set_put_fork(kargs); |
7e47682e AS |
6560 | } |
6561 | ||
817929ec | 6562 | /** |
5a5cf5cb CB |
6563 | * cgroup_post_fork - finalize cgroup setup for the child process |
6564 | * @child: the child process | |
ffacbd11 | 6565 | * @kargs: the arguments passed to create the child process |
a043e3b2 | 6566 | * |
5a5cf5cb CB |
6567 | * Attach the child process to its css_set calling the subsystem fork() |
6568 | * callbacks. | |
a043e3b2 | 6569 | */ |
ef2c41cf CB |
6570 | void cgroup_post_fork(struct task_struct *child, |
6571 | struct kernel_clone_args *kargs) | |
6572 | __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex) | |
817929ec | 6573 | { |
661ee628 CB |
6574 | unsigned long cgrp_flags = 0; |
6575 | bool kill = false; | |
30159ec7 | 6576 | struct cgroup_subsys *ss; |
5153faac | 6577 | struct css_set *cset; |
5edee61e TH |
6578 | int i; |
6579 | ||
ef2c41cf CB |
6580 | cset = kargs->cset; |
6581 | kargs->cset = NULL; | |
6582 | ||
5153faac TH |
6583 | spin_lock_irq(&css_set_lock); |
6584 | ||
0cd9d33a TH |
6585 | /* init tasks are special, only link regular threads */ |
6586 | if (likely(child->pid)) { | |
661ee628 CB |
6587 | if (kargs->cgrp) |
6588 | cgrp_flags = kargs->cgrp->flags; | |
6589 | else | |
6590 | cgrp_flags = cset->dfl_cgrp->flags; | |
6591 | ||
0cd9d33a | 6592 | WARN_ON_ONCE(!list_empty(&child->cg_list)); |
0cd9d33a TH |
6593 | cset->nr_tasks++; |
6594 | css_set_move_task(child, NULL, cset, false); | |
ef2c41cf CB |
6595 | } else { |
6596 | put_css_set(cset); | |
6597 | cset = NULL; | |
0cd9d33a | 6598 | } |
5153faac | 6599 | |
661ee628 CB |
6600 | if (!(child->flags & PF_KTHREAD)) { |
6601 | if (unlikely(test_bit(CGRP_FREEZE, &cgrp_flags))) { | |
6602 | /* | |
6603 | * If the cgroup has to be frozen, the new task has | |
6604 | * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to | |
6605 | * get the task into the frozen state. | |
6606 | */ | |
6607 | spin_lock(&child->sighand->siglock); | |
6608 | WARN_ON_ONCE(child->frozen); | |
6609 | child->jobctl |= JOBCTL_TRAP_FREEZE; | |
6610 | spin_unlock(&child->sighand->siglock); | |
6611 | ||
6612 | /* | |
6613 | * Calling cgroup_update_frozen() isn't required here, | |
6614 | * because it will be called anyway a bit later from | |
6615 | * do_freezer_trap(). So we avoid cgroup's transient | |
6616 | * switch from the frozen state and back. | |
6617 | */ | |
6618 | } | |
76f969e8 RG |
6619 | |
6620 | /* | |
661ee628 CB |
6621 | * If the cgroup is to be killed notice it now and take the |
6622 | * child down right after we finished preparing it for | |
6623 | * userspace. | |
76f969e8 | 6624 | */ |
661ee628 | 6625 | kill = test_bit(CGRP_KILL, &cgrp_flags); |
817929ec | 6626 | } |
5edee61e | 6627 | |
5153faac TH |
6628 | spin_unlock_irq(&css_set_lock); |
6629 | ||
5edee61e TH |
6630 | /* |
6631 | * Call ss->fork(). This must happen after @child is linked on | |
6632 | * css_set; otherwise, @child might change state between ->fork() | |
6633 | * and addition to css_set. | |
6634 | */ | |
b4e0eeaf | 6635 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 6636 | ss->fork(child); |
b4e0eeaf | 6637 | } while_each_subsys_mask(); |
5a5cf5cb | 6638 | |
ef2c41cf CB |
6639 | /* Make the new cset the root_cset of the new cgroup namespace. */ |
6640 | if (kargs->flags & CLONE_NEWCGROUP) { | |
6641 | struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset; | |
6642 | ||
6643 | get_css_set(cset); | |
6644 | child->nsproxy->cgroup_ns->root_cset = cset; | |
6645 | put_css_set(rcset); | |
6646 | } | |
6647 | ||
661ee628 CB |
6648 | /* Cgroup has to be killed so take down child immediately. */ |
6649 | if (unlikely(kill)) | |
6650 | do_send_sig_info(SIGKILL, SEND_SIG_NOINFO, child, PIDTYPE_TGID); | |
6651 | ||
ef2c41cf | 6652 | cgroup_css_set_put_fork(kargs); |
817929ec | 6653 | } |
5edee61e | 6654 | |
b4f48b63 PM |
6655 | /** |
6656 | * cgroup_exit - detach cgroup from exiting task | |
6657 | * @tsk: pointer to task_struct of exiting process | |
6658 | * | |
e7c7b1d8 | 6659 | * Description: Detach cgroup from @tsk. |
b4f48b63 | 6660 | * |
b4f48b63 | 6661 | */ |
1ec41830 | 6662 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 6663 | { |
30159ec7 | 6664 | struct cgroup_subsys *ss; |
5abb8855 | 6665 | struct css_set *cset; |
d41d5a01 | 6666 | int i; |
817929ec | 6667 | |
5153faac | 6668 | spin_lock_irq(&css_set_lock); |
0de0942d | 6669 | |
5153faac TH |
6670 | WARN_ON_ONCE(list_empty(&tsk->cg_list)); |
6671 | cset = task_css_set(tsk); | |
6672 | css_set_move_task(tsk, cset, NULL, false); | |
6673 | list_add_tail(&tsk->cg_list, &cset->dying_tasks); | |
6674 | cset->nr_tasks--; | |
76f969e8 | 6675 | |
5153faac | 6676 | WARN_ON_ONCE(cgroup_task_frozen(tsk)); |
f4f809f6 RG |
6677 | if (unlikely(!(tsk->flags & PF_KTHREAD) && |
6678 | test_bit(CGRP_FREEZE, &task_dfl_cgroup(tsk)->flags))) | |
5153faac | 6679 | cgroup_update_frozen(task_dfl_cgroup(tsk)); |
76f969e8 | 6680 | |
5153faac | 6681 | spin_unlock_irq(&css_set_lock); |
817929ec | 6682 | |
cb4a3167 | 6683 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 6684 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 6685 | ss->exit(tsk); |
b4e0eeaf | 6686 | } while_each_subsys_mask(); |
2e91fa7f | 6687 | } |
30159ec7 | 6688 | |
51bee5ab | 6689 | void cgroup_release(struct task_struct *task) |
2e91fa7f | 6690 | { |
afcf6c8b TH |
6691 | struct cgroup_subsys *ss; |
6692 | int ssid; | |
6693 | ||
51bee5ab ON |
6694 | do_each_subsys_mask(ss, ssid, have_release_callback) { |
6695 | ss->release(task); | |
b4e0eeaf | 6696 | } while_each_subsys_mask(); |
c03cd773 | 6697 | |
5153faac TH |
6698 | spin_lock_irq(&css_set_lock); |
6699 | css_set_skip_task_iters(task_css_set(task), task); | |
6700 | list_del_init(&task->cg_list); | |
6701 | spin_unlock_irq(&css_set_lock); | |
51bee5ab | 6702 | } |
d41d5a01 | 6703 | |
51bee5ab ON |
6704 | void cgroup_free(struct task_struct *task) |
6705 | { | |
6706 | struct css_set *cset = task_css_set(task); | |
2e91fa7f | 6707 | put_css_set(cset); |
b4f48b63 | 6708 | } |
697f4161 | 6709 | |
8bab8dde PM |
6710 | static int __init cgroup_disable(char *str) |
6711 | { | |
30159ec7 | 6712 | struct cgroup_subsys *ss; |
8bab8dde | 6713 | char *token; |
30159ec7 | 6714 | int i; |
8bab8dde PM |
6715 | |
6716 | while ((token = strsep(&str, ",")) != NULL) { | |
6717 | if (!*token) | |
6718 | continue; | |
be45c900 | 6719 | |
3ed80a62 | 6720 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6721 | if (strcmp(token, ss->name) && |
6722 | strcmp(token, ss->legacy_name)) | |
6723 | continue; | |
45e1ba40 SB |
6724 | |
6725 | static_branch_disable(cgroup_subsys_enabled_key[i]); | |
6726 | pr_info("Disabling %s control group subsystem\n", | |
6727 | ss->name); | |
8bab8dde | 6728 | } |
3958e2d0 SB |
6729 | |
6730 | for (i = 0; i < OPT_FEATURE_COUNT; i++) { | |
6731 | if (strcmp(token, cgroup_opt_feature_names[i])) | |
6732 | continue; | |
6733 | cgroup_feature_disable_mask |= 1 << i; | |
6734 | pr_info("Disabling %s control group feature\n", | |
6735 | cgroup_opt_feature_names[i]); | |
6736 | break; | |
6737 | } | |
8bab8dde PM |
6738 | } |
6739 | return 1; | |
6740 | } | |
6741 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6742 | |
5cf8114d WL |
6743 | void __init __weak enable_debug_cgroup(void) { } |
6744 | ||
6745 | static int __init enable_cgroup_debug(char *str) | |
6746 | { | |
6747 | cgroup_debug = true; | |
6748 | enable_debug_cgroup(); | |
6749 | return 1; | |
6750 | } | |
6751 | __setup("cgroup_debug", enable_cgroup_debug); | |
6752 | ||
b77d7b60 | 6753 | /** |
ec903c0c | 6754 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6755 | * @dentry: directory dentry of interest |
6756 | * @ss: subsystem of interest | |
b77d7b60 | 6757 | * |
5a17f543 TH |
6758 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6759 | * to get the corresponding css and return it. If such css doesn't exist | |
6760 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6761 | */ |
ec903c0c TH |
6762 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6763 | struct cgroup_subsys *ss) | |
e5d1367f | 6764 | { |
2bd59d48 | 6765 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6766 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6767 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6768 | struct cgroup *cgrp; |
e5d1367f | 6769 | |
35cf0836 | 6770 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6771 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6772 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6773 | return ERR_PTR(-EBADF); |
6774 | ||
5a17f543 TH |
6775 | rcu_read_lock(); |
6776 | ||
2bd59d48 TH |
6777 | /* |
6778 | * This path doesn't originate from kernfs and @kn could already | |
6779 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6780 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 | 6781 | */ |
e0aed7c7 | 6782 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
2bd59d48 TH |
6783 | if (cgrp) |
6784 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6785 | |
ec903c0c | 6786 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6787 | css = ERR_PTR(-ENOENT); |
6788 | ||
6789 | rcu_read_unlock(); | |
6790 | return css; | |
e5d1367f | 6791 | } |
e5d1367f | 6792 | |
1cb650b9 LZ |
6793 | /** |
6794 | * css_from_id - lookup css by id | |
6795 | * @id: the cgroup id | |
6796 | * @ss: cgroup subsys to be looked into | |
6797 | * | |
6798 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6799 | * Should be called under rcu_read_lock(). | |
6800 | */ | |
6801 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6802 | { | |
6fa4918d | 6803 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 6804 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
6805 | } |
6806 | ||
16af4396 TH |
6807 | /** |
6808 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6809 | * @path: path on the default hierarchy | |
6810 | * | |
6811 | * Find the cgroup at @path on the default hierarchy, increment its | |
6812 | * reference count and return it. Returns pointer to the found cgroup on | |
be288169 SB |
6813 | * success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already |
6814 | * been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory. | |
16af4396 TH |
6815 | */ |
6816 | struct cgroup *cgroup_get_from_path(const char *path) | |
6817 | { | |
6818 | struct kernfs_node *kn; | |
be288169 | 6819 | struct cgroup *cgrp = ERR_PTR(-ENOENT); |
74e4b956 | 6820 | struct cgroup *root_cgrp; |
16af4396 | 6821 | |
46307fd6 | 6822 | root_cgrp = current_cgns_cgroup_dfl(); |
74e4b956 | 6823 | kn = kernfs_walk_and_get(root_cgrp->kn, path); |
be288169 SB |
6824 | if (!kn) |
6825 | goto out; | |
6826 | ||
6827 | if (kernfs_type(kn) != KERNFS_DIR) { | |
6828 | cgrp = ERR_PTR(-ENOTDIR); | |
6829 | goto out_kernfs; | |
16af4396 TH |
6830 | } |
6831 | ||
be288169 SB |
6832 | rcu_read_lock(); |
6833 | ||
6834 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); | |
6835 | if (!cgrp || !cgroup_tryget(cgrp)) | |
6836 | cgrp = ERR_PTR(-ENOENT); | |
6837 | ||
6838 | rcu_read_unlock(); | |
6839 | ||
6840 | out_kernfs: | |
6841 | kernfs_put(kn); | |
6842 | out: | |
16af4396 TH |
6843 | return cgrp; |
6844 | } | |
6845 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6846 | ||
1f3fe7eb | 6847 | /** |
b675d4bd | 6848 | * cgroup_v1v2_get_from_fd - get a cgroup pointer from a fd |
a6d1ce59 | 6849 | * @fd: fd obtained by open(cgroup_dir) |
1f3fe7eb MKL |
6850 | * |
6851 | * Find the cgroup from a fd which should be obtained | |
6852 | * by opening a cgroup directory. Returns a pointer to the | |
6853 | * cgroup on success. ERR_PTR is returned if the cgroup | |
6854 | * cannot be found. | |
6855 | */ | |
a6d1ce59 | 6856 | struct cgroup *cgroup_v1v2_get_from_fd(int fd) |
1f3fe7eb | 6857 | { |
1f3fe7eb MKL |
6858 | struct cgroup *cgrp; |
6859 | struct file *f; | |
6860 | ||
6861 | f = fget_raw(fd); | |
6862 | if (!f) | |
6863 | return ERR_PTR(-EBADF); | |
6864 | ||
a6d1ce59 | 6865 | cgrp = cgroup_v1v2_get_from_file(f); |
1f3fe7eb | 6866 | fput(f); |
1f3fe7eb MKL |
6867 | return cgrp; |
6868 | } | |
a6d1ce59 YA |
6869 | |
6870 | /** | |
6871 | * cgroup_get_from_fd - same as cgroup_v1v2_get_from_fd, but only supports | |
6872 | * cgroup2. | |
b675d4bd | 6873 | * @fd: fd obtained by open(cgroup2_dir) |
a6d1ce59 YA |
6874 | */ |
6875 | struct cgroup *cgroup_get_from_fd(int fd) | |
6876 | { | |
6877 | struct cgroup *cgrp = cgroup_v1v2_get_from_fd(fd); | |
6878 | ||
6879 | if (IS_ERR(cgrp)) | |
6880 | return ERR_CAST(cgrp); | |
6881 | ||
6882 | if (!cgroup_on_dfl(cgrp)) { | |
6883 | cgroup_put(cgrp); | |
6884 | return ERR_PTR(-EBADF); | |
6885 | } | |
6886 | return cgrp; | |
6887 | } | |
1f3fe7eb MKL |
6888 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); |
6889 | ||
38cf3a68 TH |
6890 | static u64 power_of_ten(int power) |
6891 | { | |
6892 | u64 v = 1; | |
6893 | while (power--) | |
6894 | v *= 10; | |
6895 | return v; | |
6896 | } | |
6897 | ||
6898 | /** | |
6899 | * cgroup_parse_float - parse a floating number | |
6900 | * @input: input string | |
6901 | * @dec_shift: number of decimal digits to shift | |
6902 | * @v: output | |
6903 | * | |
6904 | * Parse a decimal floating point number in @input and store the result in | |
6905 | * @v with decimal point right shifted @dec_shift times. For example, if | |
6906 | * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345. | |
6907 | * Returns 0 on success, -errno otherwise. | |
6908 | * | |
6909 | * There's nothing cgroup specific about this function except that it's | |
6910 | * currently the only user. | |
6911 | */ | |
6912 | int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v) | |
6913 | { | |
6914 | s64 whole, frac = 0; | |
6915 | int fstart = 0, fend = 0, flen; | |
6916 | ||
6917 | if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend)) | |
6918 | return -EINVAL; | |
6919 | if (frac < 0) | |
6920 | return -EINVAL; | |
6921 | ||
6922 | flen = fend > fstart ? fend - fstart : 0; | |
6923 | if (flen < dec_shift) | |
6924 | frac *= power_of_ten(dec_shift - flen); | |
6925 | else | |
6926 | frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift)); | |
6927 | ||
6928 | *v = whole * power_of_ten(dec_shift) + frac; | |
6929 | return 0; | |
6930 | } | |
6931 | ||
bd1060a1 TH |
6932 | /* |
6933 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6934 | * definition in cgroup-defs.h. | |
6935 | */ | |
6936 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6937 | ||
bd1060a1 TH |
6938 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) |
6939 | { | |
78cc316e | 6940 | struct cgroup *cgroup; |
e876ecc6 | 6941 | |
bd1060a1 | 6942 | rcu_read_lock(); |
78cc316e DB |
6943 | /* Don't associate the sock with unrelated interrupted task's cgroup. */ |
6944 | if (in_interrupt()) { | |
6945 | cgroup = &cgrp_dfl_root.cgrp; | |
6946 | cgroup_get(cgroup); | |
6947 | goto out; | |
6948 | } | |
6949 | ||
bd1060a1 TH |
6950 | while (true) { |
6951 | struct css_set *cset; | |
6952 | ||
6953 | cset = task_css_set(current); | |
6954 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
78cc316e | 6955 | cgroup = cset->dfl_cgrp; |
bd1060a1 TH |
6956 | break; |
6957 | } | |
6958 | cpu_relax(); | |
6959 | } | |
78cc316e DB |
6960 | out: |
6961 | skcd->cgroup = cgroup; | |
6962 | cgroup_bpf_get(cgroup); | |
bd1060a1 TH |
6963 | rcu_read_unlock(); |
6964 | } | |
6965 | ||
ad0f75e5 CW |
6966 | void cgroup_sk_clone(struct sock_cgroup_data *skcd) |
6967 | { | |
8520e224 DB |
6968 | struct cgroup *cgrp = sock_cgroup_ptr(skcd); |
6969 | ||
6970 | /* | |
6971 | * We might be cloning a socket which is left in an empty | |
6972 | * cgroup and the cgroup might have already been rmdir'd. | |
6973 | * Don't use cgroup_get_live(). | |
6974 | */ | |
6975 | cgroup_get(cgrp); | |
6976 | cgroup_bpf_get(cgrp); | |
ad0f75e5 CW |
6977 | } |
6978 | ||
bd1060a1 TH |
6979 | void cgroup_sk_free(struct sock_cgroup_data *skcd) |
6980 | { | |
4bfc0bb2 RG |
6981 | struct cgroup *cgrp = sock_cgroup_ptr(skcd); |
6982 | ||
6983 | cgroup_bpf_put(cgrp); | |
6984 | cgroup_put(cgrp); | |
bd1060a1 TH |
6985 | } |
6986 | ||
6987 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6988 | ||
01ee6cfb RG |
6989 | #ifdef CONFIG_SYSFS |
6990 | static ssize_t show_delegatable_files(struct cftype *files, char *buf, | |
6991 | ssize_t size, const char *prefix) | |
6992 | { | |
6993 | struct cftype *cft; | |
6994 | ssize_t ret = 0; | |
6995 | ||
6996 | for (cft = files; cft && cft->name[0] != '\0'; cft++) { | |
6997 | if (!(cft->flags & CFTYPE_NS_DELEGATABLE)) | |
6998 | continue; | |
6999 | ||
7000 | if (prefix) | |
7001 | ret += snprintf(buf + ret, size - ret, "%s.", prefix); | |
7002 | ||
7003 | ret += snprintf(buf + ret, size - ret, "%s\n", cft->name); | |
7004 | ||
4d9ebbe2 | 7005 | if (WARN_ON(ret >= size)) |
01ee6cfb | 7006 | break; |
01ee6cfb RG |
7007 | } |
7008 | ||
7009 | return ret; | |
7010 | } | |
7011 | ||
7012 | static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr, | |
7013 | char *buf) | |
7014 | { | |
7015 | struct cgroup_subsys *ss; | |
7016 | int ssid; | |
7017 | ssize_t ret = 0; | |
7018 | ||
8a693f77 TH |
7019 | ret = show_delegatable_files(cgroup_base_files, buf + ret, |
7020 | PAGE_SIZE - ret, NULL); | |
7021 | if (cgroup_psi_enabled()) | |
7022 | ret += show_delegatable_files(cgroup_psi_files, buf + ret, | |
7023 | PAGE_SIZE - ret, NULL); | |
01ee6cfb RG |
7024 | |
7025 | for_each_subsys(ss, ssid) | |
7026 | ret += show_delegatable_files(ss->dfl_cftypes, buf + ret, | |
7027 | PAGE_SIZE - ret, | |
7028 | cgroup_subsys_name[ssid]); | |
7029 | ||
7030 | return ret; | |
7031 | } | |
7032 | static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate); | |
7033 | ||
5f2e6734 RG |
7034 | static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr, |
7035 | char *buf) | |
7036 | { | |
8a931f80 JW |
7037 | return snprintf(buf, PAGE_SIZE, |
7038 | "nsdelegate\n" | |
6a010a49 | 7039 | "favordynmods\n" |
8a931f80 JW |
7040 | "memory_localevents\n" |
7041 | "memory_recursiveprot\n"); | |
5f2e6734 RG |
7042 | } |
7043 | static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features); | |
7044 | ||
01ee6cfb RG |
7045 | static struct attribute *cgroup_sysfs_attrs[] = { |
7046 | &cgroup_delegate_attr.attr, | |
5f2e6734 | 7047 | &cgroup_features_attr.attr, |
01ee6cfb RG |
7048 | NULL, |
7049 | }; | |
7050 | ||
7051 | static const struct attribute_group cgroup_sysfs_attr_group = { | |
7052 | .attrs = cgroup_sysfs_attrs, | |
7053 | .name = "cgroup", | |
7054 | }; | |
7055 | ||
7056 | static int __init cgroup_sysfs_init(void) | |
7057 | { | |
7058 | return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group); | |
7059 | } | |
7060 | subsys_initcall(cgroup_sysfs_init); | |
a5e112e6 | 7061 | |
01ee6cfb | 7062 | #endif /* CONFIG_SYSFS */ |