Branch data Line data Source code
1 : : /*
2 : : * Generic process-grouping system.
3 : : *
4 : : * Based originally on the cpuset system, extracted by Paul Menage
5 : : * Copyright (C) 2006 Google, Inc
6 : : *
7 : : * Notifications support
8 : : * Copyright (C) 2009 Nokia Corporation
9 : : * Author: Kirill A. Shutemov
10 : : *
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 : :
29 : : #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 : :
31 : : #include "cgroup-internal.h"
32 : :
33 : : #include <linux/cred.h>
34 : : #include <linux/errno.h>
35 : : #include <linux/init_task.h>
36 : : #include <linux/kernel.h>
37 : : #include <linux/magic.h>
38 : : #include <linux/mutex.h>
39 : : #include <linux/mount.h>
40 : : #include <linux/pagemap.h>
41 : : #include <linux/proc_fs.h>
42 : : #include <linux/rcupdate.h>
43 : : #include <linux/sched.h>
44 : : #include <linux/sched/task.h>
45 : : #include <linux/slab.h>
46 : : #include <linux/spinlock.h>
47 : : #include <linux/percpu-rwsem.h>
48 : : #include <linux/string.h>
49 : : #include <linux/hashtable.h>
50 : : #include <linux/idr.h>
51 : : #include <linux/kthread.h>
52 : : #include <linux/atomic.h>
53 : : #include <linux/cpuset.h>
54 : : #include <linux/proc_ns.h>
55 : : #include <linux/nsproxy.h>
56 : : #include <linux/file.h>
57 : : #include <linux/fs_parser.h>
58 : : #include <linux/sched/cputime.h>
59 : : #include <linux/psi.h>
60 : : #include <net/sock.h>
61 : :
62 : : #define CREATE_TRACE_POINTS
63 : : #include <trace/events/cgroup.h>
64 : :
65 : : #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
66 : : MAX_CFTYPE_NAME + 2)
67 : : /* let's not notify more than 100 times per second */
68 : : #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
69 : :
70 : : /*
71 : : * cgroup_mutex is the master lock. Any modification to cgroup or its
72 : : * hierarchy must be performed while holding it.
73 : : *
74 : : * css_set_lock protects task->cgroups pointer, the list of css_set
75 : : * objects, and the chain of tasks off each css_set.
76 : : *
77 : : * These locks are exported if CONFIG_PROVE_RCU so that accessors in
78 : : * cgroup.h can use them for lockdep annotations.
79 : : */
80 : : DEFINE_MUTEX(cgroup_mutex);
81 : : DEFINE_SPINLOCK(css_set_lock);
82 : :
83 : : #ifdef CONFIG_PROVE_RCU
84 : : EXPORT_SYMBOL_GPL(cgroup_mutex);
85 : : EXPORT_SYMBOL_GPL(css_set_lock);
86 : : #endif
87 : :
88 : : DEFINE_SPINLOCK(trace_cgroup_path_lock);
89 : : char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
90 : : bool cgroup_debug __read_mostly;
91 : :
92 : : /*
93 : : * Protects cgroup_idr and css_idr so that IDs can be released without
94 : : * grabbing cgroup_mutex.
95 : : */
96 : : static DEFINE_SPINLOCK(cgroup_idr_lock);
97 : :
98 : : /*
99 : : * Protects cgroup_file->kn for !self csses. It synchronizes notifications
100 : : * against file removal/re-creation across css hiding.
101 : : */
102 : : static DEFINE_SPINLOCK(cgroup_file_kn_lock);
103 : :
104 : : DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
105 : :
106 : : #define cgroup_assert_mutex_or_rcu_locked() \
107 : : RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
108 : : !lockdep_is_held(&cgroup_mutex), \
109 : : "cgroup_mutex or RCU read lock required");
110 : :
111 : : /*
112 : : * cgroup destruction makes heavy use of work items and there can be a lot
113 : : * of concurrent destructions. Use a separate workqueue so that cgroup
114 : : * destruction work items don't end up filling up max_active of system_wq
115 : : * which may lead to deadlock.
116 : : */
117 : : static struct workqueue_struct *cgroup_destroy_wq;
118 : :
119 : : /* generate an array of cgroup subsystem pointers */
120 : : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
121 : : struct cgroup_subsys *cgroup_subsys[] = {
122 : : #include <linux/cgroup_subsys.h>
123 : : };
124 : : #undef SUBSYS
125 : :
126 : : /* array of cgroup subsystem names */
127 : : #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
128 : : static const char *cgroup_subsys_name[] = {
129 : : #include <linux/cgroup_subsys.h>
130 : : };
131 : : #undef SUBSYS
132 : :
133 : : /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
134 : : #define SUBSYS(_x) \
135 : : DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
136 : : DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
137 : : EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
138 : : EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
139 : : #include <linux/cgroup_subsys.h>
140 : : #undef SUBSYS
141 : :
142 : : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
143 : : static struct static_key_true *cgroup_subsys_enabled_key[] = {
144 : : #include <linux/cgroup_subsys.h>
145 : : };
146 : : #undef SUBSYS
147 : :
148 : : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
149 : : static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
150 : : #include <linux/cgroup_subsys.h>
151 : : };
152 : : #undef SUBSYS
153 : :
154 : : static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
155 : :
156 : : /*
157 : : * The default hierarchy, reserved for the subsystems that are otherwise
158 : : * unattached - it never has more than a single cgroup, and all tasks are
159 : : * part of that cgroup.
160 : : */
161 : : struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
162 : : EXPORT_SYMBOL_GPL(cgrp_dfl_root);
163 : :
164 : : /*
165 : : * The default hierarchy always exists but is hidden until mounted for the
166 : : * first time. This is for backward compatibility.
167 : : */
168 : : static bool cgrp_dfl_visible;
169 : :
170 : : /* some controllers are not supported in the default hierarchy */
171 : : static u16 cgrp_dfl_inhibit_ss_mask;
172 : :
173 : : /* some controllers are implicitly enabled on the default hierarchy */
174 : : static u16 cgrp_dfl_implicit_ss_mask;
175 : :
176 : : /* some controllers can be threaded on the default hierarchy */
177 : : static u16 cgrp_dfl_threaded_ss_mask;
178 : :
179 : : /* The list of hierarchy roots */
180 : : LIST_HEAD(cgroup_roots);
181 : : static int cgroup_root_count;
182 : :
183 : : /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
184 : : static DEFINE_IDR(cgroup_hierarchy_idr);
185 : :
186 : : /*
187 : : * Assign a monotonically increasing serial number to csses. It guarantees
188 : : * cgroups with bigger numbers are newer than those with smaller numbers.
189 : : * Also, as csses are always appended to the parent's ->children list, it
190 : : * guarantees that sibling csses are always sorted in the ascending serial
191 : : * number order on the list. Protected by cgroup_mutex.
192 : : */
193 : : static u64 css_serial_nr_next = 1;
194 : :
195 : : /*
196 : : * These bitmasks identify subsystems with specific features to avoid
197 : : * having to do iterative checks repeatedly.
198 : : */
199 : : static u16 have_fork_callback __read_mostly;
200 : : static u16 have_exit_callback __read_mostly;
201 : : static u16 have_release_callback __read_mostly;
202 : : static u16 have_canfork_callback __read_mostly;
203 : :
204 : : /* cgroup namespace for init task */
205 : : struct cgroup_namespace init_cgroup_ns = {
206 : : .count = REFCOUNT_INIT(2),
207 : : .user_ns = &init_user_ns,
208 : : .ns.ops = &cgroupns_operations,
209 : : .ns.inum = PROC_CGROUP_INIT_INO,
210 : : .root_cset = &init_css_set,
211 : : };
212 : :
213 : : static struct file_system_type cgroup2_fs_type;
214 : : static struct cftype cgroup_base_files[];
215 : :
216 : : static int cgroup_apply_control(struct cgroup *cgrp);
217 : : static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
218 : : static void css_task_iter_skip(struct css_task_iter *it,
219 : : struct task_struct *task);
220 : : static int cgroup_destroy_locked(struct cgroup *cgrp);
221 : : static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
222 : : struct cgroup_subsys *ss);
223 : : static void css_release(struct percpu_ref *ref);
224 : : static void kill_css(struct cgroup_subsys_state *css);
225 : : static int cgroup_addrm_files(struct cgroup_subsys_state *css,
226 : : struct cgroup *cgrp, struct cftype cfts[],
227 : : bool is_add);
228 : :
229 : : /**
230 : : * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
231 : : * @ssid: subsys ID of interest
232 : : *
233 : : * cgroup_subsys_enabled() can only be used with literal subsys names which
234 : : * is fine for individual subsystems but unsuitable for cgroup core. This
235 : : * is slower static_key_enabled() based test indexed by @ssid.
236 : : */
237 : 44440 : bool cgroup_ssid_enabled(int ssid)
238 : : {
239 : : if (CGROUP_SUBSYS_COUNT == 0)
240 : : return false;
241 : :
242 : 51712 : return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
243 : : }
244 : :
245 : : /**
246 : : * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
247 : : * @cgrp: the cgroup of interest
248 : : *
249 : : * The default hierarchy is the v2 interface of cgroup and this function
250 : : * can be used to test whether a cgroup is on the default hierarchy for
251 : : * cases where a subsystem should behave differnetly depending on the
252 : : * interface version.
253 : : *
254 : : * The set of behaviors which change on the default hierarchy are still
255 : : * being determined and the mount option is prefixed with __DEVEL__.
256 : : *
257 : : * List of changed behaviors:
258 : : *
259 : : * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
260 : : * and "name" are disallowed.
261 : : *
262 : : * - When mounting an existing superblock, mount options should match.
263 : : *
264 : : * - Remount is disallowed.
265 : : *
266 : : * - rename(2) is disallowed.
267 : : *
268 : : * - "tasks" is removed. Everything should be at process granularity. Use
269 : : * "cgroup.procs" instead.
270 : : *
271 : : * - "cgroup.procs" is not sorted. pids will be unique unless they got
272 : : * recycled inbetween reads.
273 : : *
274 : : * - "release_agent" and "notify_on_release" are removed. Replacement
275 : : * notification mechanism will be implemented.
276 : : *
277 : : * - "cgroup.clone_children" is removed.
278 : : *
279 : : * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
280 : : * and its descendants contain no task; otherwise, 1. The file also
281 : : * generates kernfs notification which can be monitored through poll and
282 : : * [di]notify when the value of the file changes.
283 : : *
284 : : * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
285 : : * take masks of ancestors with non-empty cpus/mems, instead of being
286 : : * moved to an ancestor.
287 : : *
288 : : * - cpuset: a task can be moved into an empty cpuset, and again it takes
289 : : * masks of ancestors.
290 : : *
291 : : * - memcg: use_hierarchy is on by default and the cgroup file for the flag
292 : : * is not created.
293 : : *
294 : : * - blkcg: blk-throttle becomes properly hierarchical.
295 : : *
296 : : * - debug: disallowed on the default hierarchy.
297 : : */
298 : 0 : bool cgroup_on_dfl(const struct cgroup *cgrp)
299 : : {
300 : 4459618 : return cgrp->root == &cgrp_dfl_root;
301 : : }
302 : :
303 : : /* IDR wrappers which synchronize using cgroup_idr_lock */
304 : 160428 : static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
305 : : gfp_t gfp_mask)
306 : : {
307 : : int ret;
308 : :
309 : 160428 : idr_preload(gfp_mask);
310 : : spin_lock_bh(&cgroup_idr_lock);
311 : 160428 : ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
312 : : spin_unlock_bh(&cgroup_idr_lock);
313 : : idr_preload_end();
314 : 160428 : return ret;
315 : : }
316 : :
317 : 175384 : static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
318 : : {
319 : : void *ret;
320 : :
321 : : spin_lock_bh(&cgroup_idr_lock);
322 : 175384 : ret = idr_replace(idr, ptr, id);
323 : : spin_unlock_bh(&cgroup_idr_lock);
324 : 175384 : return ret;
325 : : }
326 : :
327 : 71128 : static void cgroup_idr_remove(struct idr *idr, int id)
328 : : {
329 : : spin_lock_bh(&cgroup_idr_lock);
330 : 71128 : idr_remove(idr, id);
331 : : spin_unlock_bh(&cgroup_idr_lock);
332 : 71128 : }
333 : :
334 : : static bool cgroup_has_tasks(struct cgroup *cgrp)
335 : : {
336 : 74360 : return cgrp->nr_populated_csets;
337 : : }
338 : :
339 : 0 : bool cgroup_is_threaded(struct cgroup *cgrp)
340 : : {
341 : 2304202 : return cgrp->dom_cgrp != cgrp;
342 : : }
343 : :
344 : : /* can @cgrp host both domain and threaded children? */
345 : : static bool cgroup_is_mixable(struct cgroup *cgrp)
346 : : {
347 : : /*
348 : : * Root isn't under domain level resource control exempting it from
349 : : * the no-internal-process constraint, so it can serve as a thread
350 : : * root and a parent of resource domains at the same time.
351 : : */
352 : : return !cgroup_parent(cgrp);
353 : : }
354 : :
355 : : /* can @cgrp become a thread root? should always be true for a thread root */
356 : 55368 : static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
357 : : {
358 : : /* mixables don't care */
359 [ + - ]: 55368 : if (cgroup_is_mixable(cgrp))
360 : : return true;
361 : :
362 : : /* domain roots can't be nested under threaded */
363 [ + - ]: 55368 : if (cgroup_is_threaded(cgrp))
364 : : return false;
365 : :
366 : : /* can only have either domain or threaded children */
367 [ + - ]: 55368 : if (cgrp->nr_populated_domain_children)
368 : : return false;
369 : :
370 : : /* and no domain controllers can be enabled */
371 [ + - ]: 55368 : if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
372 : : return false;
373 : :
374 : 55368 : return true;
375 : : }
376 : :
377 : : /* is @cgrp root of a threaded subtree? */
378 : 0 : bool cgroup_is_thread_root(struct cgroup *cgrp)
379 : : {
380 : : /* thread root should be a domain */
381 [ # # # # : 74360 : if (cgroup_is_threaded(cgrp))
+ - # # ]
382 : : return false;
383 : :
384 : : /* a domain w/ threaded children is a thread root */
385 [ # # # # : 74360 : if (cgrp->nr_threaded_children)
+ - # # ]
386 : : return true;
387 : :
388 : : /*
389 : : * A domain which has tasks and explicit threaded controllers
390 : : * enabled is a thread root.
391 : : */
392 [ # # # # : 75168 : if (cgroup_has_tasks(cgrp) &&
# # # # +
+ + - # #
# # ]
393 : 808 : (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
394 : : return true;
395 : :
396 : 0 : return false;
397 : : }
398 : :
399 : : /* a domain which isn't connected to the root w/o brekage can't be used */
400 : 55368 : static bool cgroup_is_valid_domain(struct cgroup *cgrp)
401 : : {
402 : : /* the cgroup itself can be a thread root */
403 [ + - ]: 55368 : if (cgroup_is_threaded(cgrp))
404 : : return false;
405 : :
406 : : /* but the ancestors can't be unless mixable */
407 [ + + ]: 185096 : while ((cgrp = cgroup_parent(cgrp))) {
408 [ + + + - ]: 204088 : if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
409 : : return false;
410 [ + - ]: 129728 : if (cgroup_is_threaded(cgrp))
411 : : return false;
412 : : }
413 : :
414 : : return true;
415 : : }
416 : :
417 : : /* subsystems visibly enabled on a cgroup */
418 : 285284 : static u16 cgroup_control(struct cgroup *cgrp)
419 : : {
420 : : struct cgroup *parent = cgroup_parent(cgrp);
421 : 285284 : u16 root_ss_mask = cgrp->root->subsys_mask;
422 : :
423 [ + + ]: 285284 : if (parent) {
424 : 227916 : u16 ss_mask = parent->subtree_control;
425 : :
426 : : /* threaded cgroups can only have threaded controllers */
427 [ - + ]: 227916 : if (cgroup_is_threaded(cgrp))
428 : 0 : ss_mask &= cgrp_dfl_threaded_ss_mask;
429 : 227916 : return ss_mask;
430 : : }
431 : :
432 [ + + ]: 57368 : if (cgroup_on_dfl(cgrp))
433 : 48480 : root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
434 : : cgrp_dfl_implicit_ss_mask);
435 : 57368 : return root_ss_mask;
436 : : }
437 : :
438 : : /* subsystems enabled on a cgroup */
439 : : static u16 cgroup_ss_mask(struct cgroup *cgrp)
440 : : {
441 : : struct cgroup *parent = cgroup_parent(cgrp);
442 : :
443 [ # # + + : 1527768 : if (parent) {
- + + + +
+ ]
444 : 1309320 : u16 ss_mask = parent->subtree_ss_mask;
445 : :
446 : : /* threaded cgroups can only have threaded controllers */
447 [ # # - + : 1309320 : if (cgroup_is_threaded(cgrp))
# # - + -
+ ]
448 : 0 : ss_mask &= cgrp_dfl_threaded_ss_mask;
449 : : return ss_mask;
450 : : }
451 : :
452 : 218448 : return cgrp->root->subsys_mask;
453 : : }
454 : :
455 : : /**
456 : : * cgroup_css - obtain a cgroup's css for the specified subsystem
457 : : * @cgrp: the cgroup of interest
458 : : * @ss: the subsystem of interest (%NULL returns @cgrp->self)
459 : : *
460 : : * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
461 : : * function must be called either under cgroup_mutex or rcu_read_lock() and
462 : : * the caller is responsible for pinning the returned css if it wants to
463 : : * keep accessing it outside the said locks. This function may return
464 : : * %NULL if @cgrp doesn't have @subsys_id enabled.
465 : : */
466 : : static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
467 : : struct cgroup_subsys *ss)
468 : : {
469 [ - + + - : 1679444 : if (ss)
+ - + - +
- + - + -
# # + - +
- + - + -
+ - + - +
+ # # # #
+ - # # ]
470 : 3305544 : return rcu_dereference_check(cgrp->subsys[ss->id],
471 : : lockdep_is_held(&cgroup_mutex));
472 : : else
473 : 268724 : return &cgrp->self;
474 : : }
475 : :
476 : : /**
477 : : * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
478 : : * @cgrp: the cgroup of interest
479 : : * @ss: the subsystem of interest
480 : : *
481 : : * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
482 : : * or is offline, %NULL is returned.
483 : : */
484 : 0 : static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
485 : : struct cgroup_subsys *ss)
486 : : {
487 : : struct cgroup_subsys_state *css;
488 : :
489 : : rcu_read_lock();
490 : : css = cgroup_css(cgrp, ss);
491 [ # # # # ]: 0 : if (css && !css_tryget_online(css))
492 : : css = NULL;
493 : : rcu_read_unlock();
494 : :
495 : 0 : return css;
496 : : }
497 : :
498 : : /**
499 : : * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
500 : : * @cgrp: the cgroup of interest
501 : : * @ss: the subsystem of interest (%NULL returns @cgrp->self)
502 : : *
503 : : * Similar to cgroup_css() but returns the effective css, which is defined
504 : : * as the matching css of the nearest ancestor including self which has @ss
505 : : * enabled. If @ss is associated with the hierarchy @cgrp is on, this
506 : : * function is guaranteed to return non-NULL css.
507 : : */
508 : 140252 : static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
509 : : struct cgroup_subsys *ss)
510 : : {
511 : : lockdep_assert_held(&cgroup_mutex);
512 : :
513 [ - + ]: 140252 : if (!ss)
514 : 0 : return &cgrp->self;
515 : :
516 : : /*
517 : : * This function is used while updating css associations and thus
518 : : * can't test the csses directly. Test ss_mask.
519 : : */
520 [ - + ]: 140252 : while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
521 : : cgrp = cgroup_parent(cgrp);
522 [ # # ]: 0 : if (!cgrp)
523 : : return NULL;
524 : : }
525 : :
526 : 140252 : return cgroup_css(cgrp, ss);
527 : : }
528 : :
529 : : /**
530 : : * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
531 : : * @cgrp: the cgroup of interest
532 : : * @ss: the subsystem of interest
533 : : *
534 : : * Find and get the effective css of @cgrp for @ss. The effective css is
535 : : * defined as the matching css of the nearest ancestor including self which
536 : : * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
537 : : * the root css is returned, so this function always returns a valid css.
538 : : *
539 : : * The returned css is not guaranteed to be online, and therefore it is the
540 : : * callers responsiblity to tryget a reference for it.
541 : : */
542 : 0 : struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
543 : : struct cgroup_subsys *ss)
544 : : {
545 : : struct cgroup_subsys_state *css;
546 : :
547 : : do {
548 : : css = cgroup_css(cgrp, ss);
549 : :
550 [ # # ]: 0 : if (css)
551 : 0 : return css;
552 : : cgrp = cgroup_parent(cgrp);
553 [ # # ]: 0 : } while (cgrp);
554 : :
555 : 0 : return init_css_set.subsys[ss->id];
556 : : }
557 : :
558 : : /**
559 : : * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
560 : : * @cgrp: the cgroup of interest
561 : : * @ss: the subsystem of interest
562 : : *
563 : : * Find and get the effective css of @cgrp for @ss. The effective css is
564 : : * defined as the matching css of the nearest ancestor including self which
565 : : * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
566 : : * the root css is returned, so this function always returns a valid css.
567 : : * The returned css must be put using css_put().
568 : : */
569 : 0 : struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
570 : : struct cgroup_subsys *ss)
571 : : {
572 : : struct cgroup_subsys_state *css;
573 : :
574 : : rcu_read_lock();
575 : :
576 : : do {
577 : : css = cgroup_css(cgrp, ss);
578 : :
579 [ # # # # ]: 0 : if (css && css_tryget_online(css))
580 : : goto out_unlock;
581 : : cgrp = cgroup_parent(cgrp);
582 [ # # ]: 0 : } while (cgrp);
583 : :
584 : 0 : css = init_css_set.subsys[ss->id];
585 : : css_get(css);
586 : : out_unlock:
587 : : rcu_read_unlock();
588 : 0 : return css;
589 : : }
590 : :
591 : 453164 : static void cgroup_get_live(struct cgroup *cgrp)
592 : : {
593 [ - + # # ]: 453164 : WARN_ON_ONCE(cgroup_is_dead(cgrp));
594 : : css_get(&cgrp->self);
595 : 453164 : }
596 : :
597 : : /**
598 : : * __cgroup_task_count - count the number of tasks in a cgroup. The caller
599 : : * is responsible for taking the css_set_lock.
600 : : * @cgrp: the cgroup in question
601 : : */
602 : 0 : int __cgroup_task_count(const struct cgroup *cgrp)
603 : : {
604 : : int count = 0;
605 : : struct cgrp_cset_link *link;
606 : :
607 : : lockdep_assert_held(&css_set_lock);
608 : :
609 [ # # # # ]: 0 : list_for_each_entry(link, &cgrp->cset_links, cset_link)
610 : 0 : count += link->cset->nr_tasks;
611 : :
612 : 0 : return count;
613 : : }
614 : :
615 : : /**
616 : : * cgroup_task_count - count the number of tasks in a cgroup.
617 : : * @cgrp: the cgroup in question
618 : : */
619 : 0 : int cgroup_task_count(const struct cgroup *cgrp)
620 : : {
621 : : int count;
622 : :
623 : : spin_lock_irq(&css_set_lock);
624 : : count = __cgroup_task_count(cgrp);
625 : : spin_unlock_irq(&css_set_lock);
626 : :
627 : 0 : return count;
628 : : }
629 : :
630 : 52128 : struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
631 : : {
632 : 383294 : struct cgroup *cgrp = of->kn->parent->priv;
633 : : struct cftype *cft = of_cft(of);
634 : :
635 : : /*
636 : : * This is open and unprotected implementation of cgroup_css().
637 : : * seq_css() is only called from a kernfs file operation which has
638 : : * an active reference on the file. Because all the subsystem
639 : : * files are drained before a css is disassociated with a cgroup,
640 : : * the matching css from the cgroup's subsys table is guaranteed to
641 : : * be and stay valid until the enclosing operation is complete.
642 : : */
643 [ - + - + : 383294 : if (cft->ss)
+ + # # #
# # # - +
# # # # #
# # # # #
+ - ]
644 : 105064 : return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
645 : : else
646 : 330762 : return &cgrp->self;
647 : : }
648 : : EXPORT_SYMBOL_GPL(of_css);
649 : :
650 : : /**
651 : : * for_each_css - iterate all css's of a cgroup
652 : : * @css: the iteration cursor
653 : : * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
654 : : * @cgrp: the target cgroup to iterate css's of
655 : : *
656 : : * Should be called under cgroup_[tree_]mutex.
657 : : */
658 : : #define for_each_css(css, ssid, cgrp) \
659 : : for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
660 : : if (!((css) = rcu_dereference_check( \
661 : : (cgrp)->subsys[(ssid)], \
662 : : lockdep_is_held(&cgroup_mutex)))) { } \
663 : : else
664 : :
665 : : /**
666 : : * for_each_e_css - iterate all effective css's of a cgroup
667 : : * @css: the iteration cursor
668 : : * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
669 : : * @cgrp: the target cgroup to iterate css's of
670 : : *
671 : : * Should be called under cgroup_[tree_]mutex.
672 : : */
673 : : #define for_each_e_css(css, ssid, cgrp) \
674 : : for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
675 : : if (!((css) = cgroup_e_css_by_mask(cgrp, \
676 : : cgroup_subsys[(ssid)]))) \
677 : : ; \
678 : : else
679 : :
680 : : /**
681 : : * do_each_subsys_mask - filter for_each_subsys with a bitmask
682 : : * @ss: the iteration cursor
683 : : * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
684 : : * @ss_mask: the bitmask
685 : : *
686 : : * The block will only run for cases where the ssid-th bit (1 << ssid) of
687 : : * @ss_mask is set.
688 : : */
689 : : #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
690 : : unsigned long __ss_mask = (ss_mask); \
691 : : if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
692 : : (ssid) = 0; \
693 : : break; \
694 : : } \
695 : : for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
696 : : (ss) = cgroup_subsys[ssid]; \
697 : : {
698 : :
699 : : #define while_each_subsys_mask() \
700 : : } \
701 : : } \
702 : : } while (false)
703 : :
704 : : /* iterate over child cgrps, lock should be held throughout iteration */
705 : : #define cgroup_for_each_live_child(child, cgrp) \
706 : : list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
707 : : if (({ lockdep_assert_held(&cgroup_mutex); \
708 : : cgroup_is_dead(child); })) \
709 : : ; \
710 : : else
711 : :
712 : : /* walk live descendants in preorder */
713 : : #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
714 : : css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
715 : : if (({ lockdep_assert_held(&cgroup_mutex); \
716 : : (dsct) = (d_css)->cgroup; \
717 : : cgroup_is_dead(dsct); })) \
718 : : ; \
719 : : else
720 : :
721 : : /* walk live descendants in postorder */
722 : : #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
723 : : css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
724 : : if (({ lockdep_assert_held(&cgroup_mutex); \
725 : : (dsct) = (d_css)->cgroup; \
726 : : cgroup_is_dead(dsct); })) \
727 : : ; \
728 : : else
729 : :
730 : : /*
731 : : * The default css_set - used by init and its children prior to any
732 : : * hierarchies being mounted. It contains a pointer to the root state
733 : : * for each subsystem. Also used to anchor the list of css_sets. Not
734 : : * reference-counted, to improve performance when child cgroups
735 : : * haven't been created.
736 : : */
737 : : struct css_set init_css_set = {
738 : : .refcount = REFCOUNT_INIT(1),
739 : : .dom_cset = &init_css_set,
740 : : .tasks = LIST_HEAD_INIT(init_css_set.tasks),
741 : : .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
742 : : .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
743 : : .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
744 : : .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
745 : : .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
746 : : .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
747 : : .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
748 : :
749 : : /*
750 : : * The following field is re-initialized when this cset gets linked
751 : : * in cgroup_init(). However, let's initialize the field
752 : : * statically too so that the default cgroup can be accessed safely
753 : : * early during boot.
754 : : */
755 : : .dfl_cgrp = &cgrp_dfl_root.cgrp,
756 : : };
757 : :
758 : : static int css_set_count = 1; /* 1 for init_css_set */
759 : :
760 : : static bool css_set_threaded(struct css_set *cset)
761 : : {
762 : 92990 : return cset->dom_cset != cset;
763 : : }
764 : :
765 : : /**
766 : : * css_set_populated - does a css_set contain any tasks?
767 : : * @cset: target css_set
768 : : *
769 : : * css_set_populated() should be the same as !!cset->nr_tasks at steady
770 : : * state. However, css_set_populated() can be called while a task is being
771 : : * added to or removed from the linked list before the nr_tasks is
772 : : * properly updated. Hence, we can't just look at ->nr_tasks here.
773 : : */
774 : : static bool css_set_populated(struct css_set *cset)
775 : : {
776 : : lockdep_assert_held(&css_set_lock);
777 : :
778 [ + + - + : 2345144 : return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
+ + - + +
+ - + + +
+ + + + +
+ ]
779 : : }
780 : :
781 : : /**
782 : : * cgroup_update_populated - update the populated count of a cgroup
783 : : * @cgrp: the target cgroup
784 : : * @populated: inc or dec populated count
785 : : *
786 : : * One of the css_sets associated with @cgrp is either getting its first
787 : : * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
788 : : * count is propagated towards root so that a given cgroup's
789 : : * nr_populated_children is zero iff none of its descendants contain any
790 : : * tasks.
791 : : *
792 : : * @cgrp's interface file "cgroup.populated" is zero if both
793 : : * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
794 : : * 1 otherwise. When the sum changes from or to zero, userland is notified
795 : : * that the content of the interface file has changed. This can be used to
796 : : * detect when @cgrp and its descendants become populated or empty.
797 : : */
798 : 1996720 : static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
799 : : {
800 : : struct cgroup *child = NULL;
801 [ + + ]: 1996720 : int adj = populated ? 1 : -1;
802 : :
803 : : lockdep_assert_held(&css_set_lock);
804 : :
805 : : do {
806 : : bool was_populated = cgroup_is_populated(cgrp);
807 : :
808 [ + + ]: 2179836 : if (!child) {
809 : 1996720 : cgrp->nr_populated_csets += adj;
810 : : } else {
811 [ - + ]: 183116 : if (cgroup_is_threaded(child))
812 : 0 : cgrp->nr_populated_threaded_children += adj;
813 : : else
814 : 183116 : cgrp->nr_populated_domain_children += adj;
815 : : }
816 : :
817 [ + + ]: 2179836 : if (was_populated == cgroup_is_populated(cgrp))
818 : : break;
819 : :
820 : 187156 : cgroup1_check_for_release(cgrp);
821 [ - + ]: 187156 : TRACE_CGROUP_PATH(notify_populated, cgrp,
822 : : cgroup_is_populated(cgrp));
823 : 187156 : cgroup_file_notify(&cgrp->events_file);
824 : :
825 : : child = cgrp;
826 : : cgrp = cgroup_parent(cgrp);
827 [ + + ]: 187156 : } while (cgrp);
828 : 1996720 : }
829 : :
830 : : /**
831 : : * css_set_update_populated - update populated state of a css_set
832 : : * @cset: target css_set
833 : : * @populated: whether @cset is populated or depopulated
834 : : *
835 : : * @cset is either getting the first task or losing the last. Update the
836 : : * populated counters of all associated cgroups accordingly.
837 : : */
838 : 199672 : static void css_set_update_populated(struct css_set *cset, bool populated)
839 : : {
840 : : struct cgrp_cset_link *link;
841 : :
842 : : lockdep_assert_held(&css_set_lock);
843 : :
844 [ + + ]: 2192756 : list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
845 : 1993084 : cgroup_update_populated(link->cgrp, populated);
846 : 199672 : }
847 : :
848 : : /*
849 : : * @task is leaving, advance task iterators which are pointing to it so
850 : : * that they can resume at the next position. Advancing an iterator might
851 : : * remove it from the list, use safe walk. See css_task_iter_skip() for
852 : : * details.
853 : : */
854 : 869056 : static void css_set_skip_task_iters(struct css_set *cset,
855 : : struct task_struct *task)
856 : : {
857 : : struct css_task_iter *it, *pos;
858 : :
859 [ + + ]: 869460 : list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
860 : : css_task_iter_skip(it, task);
861 : 869056 : }
862 : :
863 : : /**
864 : : * css_set_move_task - move a task from one css_set to another
865 : : * @task: task being moved
866 : : * @from_cset: css_set @task currently belongs to (may be NULL)
867 : : * @to_cset: new css_set @task is being moved to (may be NULL)
868 : : * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
869 : : *
870 : : * Move @task from @from_cset to @to_cset. If @task didn't belong to any
871 : : * css_set, @from_cset can be NULL. If @task is being disassociated
872 : : * instead of moved, @to_cset can be NULL.
873 : : *
874 : : * This function automatically handles populated counter updates and
875 : : * css_task_iter adjustments but the caller is responsible for managing
876 : : * @from_cset and @to_cset's reference counts.
877 : : */
878 : 914658 : static void css_set_move_task(struct task_struct *task,
879 : : struct css_set *from_cset, struct css_set *to_cset,
880 : : bool use_mg_tasks)
881 : : {
882 : : lockdep_assert_held(&css_set_lock);
883 : :
884 [ + + + + ]: 1449352 : if (to_cset && !css_set_populated(to_cset))
885 : 105088 : css_set_update_populated(to_cset, true);
886 : :
887 [ + + ]: 914658 : if (from_cset) {
888 [ - + # # ]: 978184 : WARN_ON_ONCE(list_empty(&task->cg_list));
889 : :
890 : 489092 : css_set_skip_task_iters(from_cset, task);
891 : : list_del_init(&task->cg_list);
892 [ + + ]: 489092 : if (!css_set_populated(from_cset))
893 : 94180 : css_set_update_populated(from_cset, false);
894 : : } else {
895 [ - + # # ]: 851132 : WARN_ON_ONCE(!list_empty(&task->cg_list));
896 : : }
897 : :
898 [ + + ]: 914658 : if (to_cset) {
899 : : /*
900 : : * We are synchronized through cgroup_threadgroup_rwsem
901 : : * against PF_EXITING setting such that we can't race
902 : : * against cgroup_exit() changing the css_set to
903 : : * init_css_set and dropping the old one.
904 : : */
905 [ - + # # ]: 534694 : WARN_ON_ONCE(task->flags & PF_EXITING);
906 : :
907 : : cgroup_move_task(task, to_cset);
908 [ + + ]: 534694 : list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
909 : : &to_cset->tasks);
910 : : }
911 : 914658 : }
912 : :
913 : : /*
914 : : * hash table for cgroup groups. This improves the performance to find
915 : : * an existing css_set. This hash doesn't (currently) take into
916 : : * account cgroups in empty hierarchies.
917 : : */
918 : : #define CSS_SET_HASH_BITS 7
919 : : static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
920 : :
921 : : static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
922 : : {
923 : : unsigned long key = 0UL;
924 : : struct cgroup_subsys *ss;
925 : : int i;
926 : :
927 [ + + + + : 7279270 : for_each_subsys(ss, i)
+ + + + ]
928 : 3481390 : key += (unsigned long)css[i];
929 : 316490 : key = (key >> 16) ^ key;
930 : :
931 : : return key;
932 : : }
933 : :
934 : 348152 : void put_css_set_locked(struct css_set *cset)
935 : : {
936 : : struct cgrp_cset_link *link, *tmp_link;
937 : : struct cgroup_subsys *ss;
938 : : int ssid;
939 : :
940 : : lockdep_assert_held(&css_set_lock);
941 : :
942 [ + + ]: 348152 : if (!refcount_dec_and_test(&cset->refcount))
943 : 348152 : return;
944 : :
945 [ - + # # ]: 185980 : WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
946 : :
947 : : /* This css_set is dead. unlink it and release cgroup and css refs */
948 [ + + ]: 2138770 : for_each_subsys(ss, ssid) {
949 : : list_del(&cset->e_cset_node[ssid]);
950 : 1022890 : css_put(cset->subsys[ssid]);
951 : : }
952 : : hash_del(&cset->hlist);
953 : 92990 : css_set_count--;
954 : :
955 [ + + ]: 1022890 : list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
956 : : list_del(&link->cset_link);
957 : : list_del(&link->cgrp_link);
958 [ + + ]: 1859800 : if (cgroup_parent(link->cgrp))
959 : : cgroup_put(link->cgrp);
960 : 929900 : kfree(link);
961 : : }
962 : :
963 [ - + ]: 92990 : if (css_set_threaded(cset)) {
964 : : list_del(&cset->threaded_csets_node);
965 : 0 : put_css_set_locked(cset->dom_cset);
966 : : }
967 : :
968 [ + - ]: 92990 : kfree_rcu(cset, rcu_head);
969 : : }
970 : :
971 : : /**
972 : : * compare_css_sets - helper function for find_existing_css_set().
973 : : * @cset: candidate css_set being tested
974 : : * @old_cset: existing css_set for a task
975 : : * @new_cgrp: cgroup that's being entered by the task
976 : : * @template: desired set of css pointers in css_set (pre-calculated)
977 : : *
978 : : * Returns true if "cset" matches "old_cset" except for the hierarchy
979 : : * which "new_cgrp" belongs to, for which it should match "new_cgrp".
980 : : */
981 : 276124 : static bool compare_css_sets(struct css_set *cset,
982 : : struct css_set *old_cset,
983 : : struct cgroup *new_cgrp,
984 : : struct cgroup_subsys_state *template[])
985 : : {
986 : : struct cgroup *new_dfl_cgrp;
987 : : struct list_head *l1, *l2;
988 : :
989 : : /*
990 : : * On the default hierarchy, there can be csets which are
991 : : * associated with the same set of cgroups but different csses.
992 : : * Let's first ensure that csses match.
993 : : */
994 [ + + ]: 276124 : if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
995 : : return false;
996 : :
997 : :
998 : : /* @cset's domain should match the default cgroup's */
999 [ + + ]: 266328 : if (cgroup_on_dfl(new_cgrp))
1000 : : new_dfl_cgrp = new_cgrp;
1001 : : else
1002 : 171498 : new_dfl_cgrp = old_cset->dfl_cgrp;
1003 : :
1004 [ + + ]: 266328 : if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1005 : : return false;
1006 : :
1007 : : /*
1008 : : * Compare cgroup pointers in order to distinguish between
1009 : : * different cgroups in hierarchies. As different cgroups may
1010 : : * share the same effective css, this comparison is always
1011 : : * necessary.
1012 : : */
1013 : 136986 : l1 = &cset->cgrp_links;
1014 : 136986 : l2 = &old_cset->cgrp_links;
1015 : : while (1) {
1016 : : struct cgrp_cset_link *link1, *link2;
1017 : : struct cgroup *cgrp1, *cgrp2;
1018 : :
1019 : 1226080 : l1 = l1->next;
1020 : 1226080 : l2 = l2->next;
1021 : : /* See if we reached the end - both lists are equal length. */
1022 [ + + ]: 1226080 : if (l1 == &cset->cgrp_links) {
1023 [ - + ]: 107886 : BUG_ON(l2 != &old_cset->cgrp_links);
1024 : : break;
1025 : : } else {
1026 [ - + ]: 1118194 : BUG_ON(l2 == &old_cset->cgrp_links);
1027 : : }
1028 : : /* Locate the cgroups associated with these links. */
1029 : : link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1030 : : link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1031 : 1118194 : cgrp1 = link1->cgrp;
1032 : 1118194 : cgrp2 = link2->cgrp;
1033 : : /* Hierarchies should be linked in the same order. */
1034 [ - + ]: 1118194 : BUG_ON(cgrp1->root != cgrp2->root);
1035 : :
1036 : : /*
1037 : : * If this hierarchy is the hierarchy of the cgroup
1038 : : * that's changing, then we need to check that this
1039 : : * css_set points to the new cgroup; if it's any other
1040 : : * hierarchy, then this css_set should point to the
1041 : : * same cgroup as the old css_set.
1042 : : */
1043 [ + + ]: 1118194 : if (cgrp1->root == new_cgrp->root) {
1044 [ + + ]: 136986 : if (cgrp1 != new_cgrp)
1045 : : return false;
1046 : : } else {
1047 [ + + ]: 981208 : if (cgrp1 != cgrp2)
1048 : : return false;
1049 : : }
1050 : : }
1051 : : return true;
1052 : : }
1053 : :
1054 : : /**
1055 : : * find_existing_css_set - init css array and find the matching css_set
1056 : : * @old_cset: the css_set that we're using before the cgroup transition
1057 : : * @cgrp: the cgroup that we're moving into
1058 : : * @template: out param for the new set of csses, should be clear on entry
1059 : : */
1060 : 211784 : static struct css_set *find_existing_css_set(struct css_set *old_cset,
1061 : : struct cgroup *cgrp,
1062 : : struct cgroup_subsys_state *template[])
1063 : : {
1064 : 211784 : struct cgroup_root *root = cgrp->root;
1065 : : struct cgroup_subsys *ss;
1066 : : struct css_set *cset;
1067 : : unsigned long key;
1068 : : int i;
1069 : :
1070 : : /*
1071 : : * Build the set of subsystem state objects that we want to see in the
1072 : : * new css_set. while subsystems can change globally, the entries here
1073 : : * won't change, so no need for locking.
1074 : : */
1075 [ + + ]: 2541408 : for_each_subsys(ss, i) {
1076 [ + + ]: 2329624 : if (root->subsys_mask & (1UL << i)) {
1077 : : /*
1078 : : * @ss is in this hierarchy, so we want the
1079 : : * effective css from @cgrp.
1080 : : */
1081 : 140252 : template[i] = cgroup_e_css_by_mask(cgrp, ss);
1082 : : } else {
1083 : : /*
1084 : : * @ss is not in this hierarchy, so we don't want
1085 : : * to change the css.
1086 : : */
1087 : 2189372 : template[i] = old_cset->subsys[i];
1088 : : }
1089 : : }
1090 : :
1091 : : key = css_set_hash(template);
1092 [ + + + + : 486970 : hash_for_each_possible(css_set_table, cset, hlist, key) {
+ + ]
1093 [ + + ]: 276124 : if (!compare_css_sets(cset, old_cset, cgrp, template))
1094 : 168238 : continue;
1095 : :
1096 : : /* This css_set matches what we need */
1097 : 107886 : return cset;
1098 : : }
1099 : :
1100 : : /* No existing cgroup group matched */
1101 : : return NULL;
1102 : : }
1103 : :
1104 : 4040 : static void free_cgrp_cset_links(struct list_head *links_to_free)
1105 : : {
1106 : : struct cgrp_cset_link *link, *tmp_link;
1107 : :
1108 [ + + ]: 8080 : list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1109 : : list_del(&link->cset_link);
1110 : 4040 : kfree(link);
1111 : : }
1112 : 4040 : }
1113 : :
1114 : : /**
1115 : : * allocate_cgrp_cset_links - allocate cgrp_cset_links
1116 : : * @count: the number of links to allocate
1117 : : * @tmp_links: list_head the allocated links are put on
1118 : : *
1119 : : * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1120 : : * through ->cset_link. Returns 0 on success or -errno.
1121 : : */
1122 : 107938 : static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1123 : : {
1124 : : struct cgrp_cset_link *link;
1125 : : int i;
1126 : :
1127 : : INIT_LIST_HEAD(tmp_links);
1128 : :
1129 [ + + ]: 1154998 : for (i = 0; i < count; i++) {
1130 : 1047060 : link = kzalloc(sizeof(*link), GFP_KERNEL);
1131 [ - + ]: 1047060 : if (!link) {
1132 : 0 : free_cgrp_cset_links(tmp_links);
1133 : 0 : return -ENOMEM;
1134 : : }
1135 : 1047060 : list_add(&link->cset_link, tmp_links);
1136 : : }
1137 : : return 0;
1138 : : }
1139 : :
1140 : : /**
1141 : : * link_css_set - a helper function to link a css_set to a cgroup
1142 : : * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1143 : : * @cset: the css_set to be linked
1144 : : * @cgrp: the destination cgroup
1145 : : */
1146 : 1043020 : static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1147 : : struct cgroup *cgrp)
1148 : : {
1149 : : struct cgrp_cset_link *link;
1150 : :
1151 [ - + ]: 1043020 : BUG_ON(list_empty(tmp_links));
1152 : :
1153 [ + + ]: 1043020 : if (cgroup_on_dfl(cgrp))
1154 : 104302 : cset->dfl_cgrp = cgrp;
1155 : :
1156 : 1043020 : link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1157 : 1043020 : link->cset = cset;
1158 : 1043020 : link->cgrp = cgrp;
1159 : :
1160 : : /*
1161 : : * Always add links to the tail of the lists so that the lists are
1162 : : * in choronological order.
1163 : : */
1164 : 1043020 : list_move_tail(&link->cset_link, &cgrp->cset_links);
1165 : 1043020 : list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1166 : :
1167 [ + + ]: 1043020 : if (cgroup_parent(cgrp))
1168 : 296372 : cgroup_get_live(cgrp);
1169 : 1043020 : }
1170 : :
1171 : : /**
1172 : : * find_css_set - return a new css_set with one cgroup updated
1173 : : * @old_cset: the baseline css_set
1174 : : * @cgrp: the cgroup to be updated
1175 : : *
1176 : : * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1177 : : * substituted into the appropriate hierarchy.
1178 : : */
1179 : 211784 : static struct css_set *find_css_set(struct css_set *old_cset,
1180 : : struct cgroup *cgrp)
1181 : : {
1182 : 211784 : struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1183 : : struct css_set *cset;
1184 : : struct list_head tmp_links;
1185 : : struct cgrp_cset_link *link;
1186 : : struct cgroup_subsys *ss;
1187 : : unsigned long key;
1188 : : int ssid;
1189 : :
1190 : : lockdep_assert_held(&cgroup_mutex);
1191 : :
1192 : : /* First see if we already have a cgroup group that matches
1193 : : * the desired set */
1194 : : spin_lock_irq(&css_set_lock);
1195 : 211784 : cset = find_existing_css_set(old_cset, cgrp, template);
1196 [ + + ]: 211784 : if (cset)
1197 : : get_css_set(cset);
1198 : : spin_unlock_irq(&css_set_lock);
1199 : :
1200 [ + + ]: 211784 : if (cset)
1201 : : return cset;
1202 : :
1203 : 103898 : cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1204 [ + - ]: 103898 : if (!cset)
1205 : : return NULL;
1206 : :
1207 : : /* Allocate all the cgrp_cset_link objects that we'll need */
1208 [ - + ]: 103898 : if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1209 : 0 : kfree(cset);
1210 : 0 : return NULL;
1211 : : }
1212 : :
1213 : : refcount_set(&cset->refcount, 1);
1214 : 103898 : cset->dom_cset = cset;
1215 : 103898 : INIT_LIST_HEAD(&cset->tasks);
1216 : 103898 : INIT_LIST_HEAD(&cset->mg_tasks);
1217 : 103898 : INIT_LIST_HEAD(&cset->dying_tasks);
1218 : 103898 : INIT_LIST_HEAD(&cset->task_iters);
1219 : 103898 : INIT_LIST_HEAD(&cset->threaded_csets);
1220 : : INIT_HLIST_NODE(&cset->hlist);
1221 : 103898 : INIT_LIST_HEAD(&cset->cgrp_links);
1222 : 103898 : INIT_LIST_HEAD(&cset->mg_preload_node);
1223 : 103898 : INIT_LIST_HEAD(&cset->mg_node);
1224 : :
1225 : : /* Copy the set of subsystem state objects generated in
1226 : : * find_existing_css_set() */
1227 : 103898 : memcpy(cset->subsys, template, sizeof(cset->subsys));
1228 : :
1229 : : spin_lock_irq(&css_set_lock);
1230 : : /* Add reference counts and links from the new css_set. */
1231 [ + + ]: 1142878 : list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1232 : 1038980 : struct cgroup *c = link->cgrp;
1233 : :
1234 [ + + ]: 1038980 : if (c->root == cgrp->root)
1235 : : c = cgrp;
1236 : 1038980 : link_css_set(&tmp_links, cset, c);
1237 : : }
1238 : :
1239 [ - + ]: 103898 : BUG_ON(!list_empty(&tmp_links));
1240 : :
1241 : 103898 : css_set_count++;
1242 : :
1243 : : /* Add @cset to the hash table */
1244 : : key = css_set_hash(cset->subsys);
1245 : 207796 : hash_add(css_set_table, &cset->hlist, key);
1246 : :
1247 [ + + ]: 1246776 : for_each_subsys(ss, ssid) {
1248 : 1142878 : struct cgroup_subsys_state *css = cset->subsys[ssid];
1249 : :
1250 : 2285756 : list_add_tail(&cset->e_cset_node[ssid],
1251 : 1142878 : &css->cgroup->e_csets[ssid]);
1252 : : css_get(css);
1253 : : }
1254 : :
1255 : : spin_unlock_irq(&css_set_lock);
1256 : :
1257 : : /*
1258 : : * If @cset should be threaded, look up the matching dom_cset and
1259 : : * link them up. We first fully initialize @cset then look for the
1260 : : * dom_cset. It's simpler this way and safe as @cset is guaranteed
1261 : : * to stay empty until we return.
1262 : : */
1263 [ - + ]: 207796 : if (cgroup_is_threaded(cset->dfl_cgrp)) {
1264 : : struct css_set *dcset;
1265 : :
1266 : 0 : dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1267 [ # # ]: 0 : if (!dcset) {
1268 : 0 : put_css_set(cset);
1269 : 0 : return NULL;
1270 : : }
1271 : :
1272 : : spin_lock_irq(&css_set_lock);
1273 : 0 : cset->dom_cset = dcset;
1274 : 0 : list_add_tail(&cset->threaded_csets_node,
1275 : : &dcset->threaded_csets);
1276 : : spin_unlock_irq(&css_set_lock);
1277 : : }
1278 : :
1279 : 103898 : return cset;
1280 : : }
1281 : :
1282 : 496876 : struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1283 : : {
1284 : 1036670 : struct cgroup *root_cgrp = kf_root->kn->priv;
1285 : :
1286 : 1036670 : return root_cgrp->root;
1287 : : }
1288 : :
1289 : 4040 : static int cgroup_init_root_id(struct cgroup_root *root)
1290 : : {
1291 : : int id;
1292 : :
1293 : : lockdep_assert_held(&cgroup_mutex);
1294 : :
1295 : 4040 : id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1296 [ + - ]: 4040 : if (id < 0)
1297 : : return id;
1298 : :
1299 : 4040 : root->hierarchy_id = id;
1300 : 4040 : return 0;
1301 : : }
1302 : :
1303 : : static void cgroup_exit_root_id(struct cgroup_root *root)
1304 : : {
1305 : : lockdep_assert_held(&cgroup_mutex);
1306 : :
1307 : 0 : idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1308 : : }
1309 : :
1310 : 0 : void cgroup_free_root(struct cgroup_root *root)
1311 : : {
1312 [ # # ]: 0 : if (root) {
1313 : 0 : idr_destroy(&root->cgroup_idr);
1314 : 0 : kfree(root);
1315 : : }
1316 : 0 : }
1317 : :
1318 : 0 : static void cgroup_destroy_root(struct cgroup_root *root)
1319 : : {
1320 : : struct cgroup *cgrp = &root->cgrp;
1321 : : struct cgrp_cset_link *link, *tmp_link;
1322 : :
1323 : 0 : trace_cgroup_destroy_root(root);
1324 : :
1325 : 0 : cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1326 : :
1327 [ # # ]: 0 : BUG_ON(atomic_read(&root->nr_cgrps));
1328 [ # # ]: 0 : BUG_ON(!list_empty(&cgrp->self.children));
1329 : :
1330 : : /* Rebind all subsystems back to the default hierarchy */
1331 [ # # ]: 0 : WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1332 : :
1333 : : /*
1334 : : * Release all the links from cset_links to this hierarchy's
1335 : : * root cgroup
1336 : : */
1337 : : spin_lock_irq(&css_set_lock);
1338 : :
1339 [ # # ]: 0 : list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1340 : : list_del(&link->cset_link);
1341 : : list_del(&link->cgrp_link);
1342 : 0 : kfree(link);
1343 : : }
1344 : :
1345 : : spin_unlock_irq(&css_set_lock);
1346 : :
1347 [ # # ]: 0 : if (!list_empty(&root->root_list)) {
1348 : : list_del(&root->root_list);
1349 : 0 : cgroup_root_count--;
1350 : : }
1351 : :
1352 : : cgroup_exit_root_id(root);
1353 : :
1354 : 0 : mutex_unlock(&cgroup_mutex);
1355 : :
1356 : 0 : kernfs_destroy_root(root->kf_root);
1357 : 0 : cgroup_free_root(root);
1358 : 0 : }
1359 : :
1360 : : /*
1361 : : * look up cgroup associated with current task's cgroup namespace on the
1362 : : * specified hierarchy
1363 : : */
1364 : : static struct cgroup *
1365 : 539888 : current_cgns_cgroup_from_root(struct cgroup_root *root)
1366 : : {
1367 : : struct cgroup *res = NULL;
1368 : : struct css_set *cset;
1369 : :
1370 : : lockdep_assert_held(&css_set_lock);
1371 : :
1372 : : rcu_read_lock();
1373 : :
1374 : 539888 : cset = current->nsproxy->cgroup_ns->root_cset;
1375 [ + - ]: 539888 : if (cset == &init_css_set) {
1376 : 539888 : res = &root->cgrp;
1377 : : } else {
1378 : : struct cgrp_cset_link *link;
1379 : :
1380 [ # # ]: 0 : list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1381 : 0 : struct cgroup *c = link->cgrp;
1382 : :
1383 [ # # ]: 0 : if (c->root == root) {
1384 : 0 : res = c;
1385 : 0 : break;
1386 : : }
1387 : : }
1388 : : }
1389 : : rcu_read_unlock();
1390 : :
1391 [ - + ]: 539888 : BUG_ON(!res);
1392 : 539888 : return res;
1393 : : }
1394 : :
1395 : : /* look up cgroup associated with given css_set on the specified hierarchy */
1396 : 2334316 : static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1397 : : struct cgroup_root *root)
1398 : : {
1399 : : struct cgroup *res = NULL;
1400 : :
1401 : : lockdep_assert_held(&cgroup_mutex);
1402 : : lockdep_assert_held(&css_set_lock);
1403 : :
1404 [ + + ]: 2334316 : if (cset == &init_css_set) {
1405 : 947184 : res = &root->cgrp;
1406 [ + + ]: 1387132 : } else if (root == &cgrp_dfl_root) {
1407 : 222734 : res = cset->dfl_cgrp;
1408 : : } else {
1409 : : struct cgrp_cset_link *link;
1410 : :
1411 [ + - ]: 6808614 : list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1412 : 6808614 : struct cgroup *c = link->cgrp;
1413 : :
1414 [ + + ]: 6808614 : if (c->root == root) {
1415 : 1164398 : res = c;
1416 : 1164398 : break;
1417 : : }
1418 : : }
1419 : : }
1420 : :
1421 [ - + ]: 2334316 : BUG_ON(!res);
1422 : 2334316 : return res;
1423 : : }
1424 : :
1425 : : /*
1426 : : * Return the cgroup for "task" from the given hierarchy. Must be
1427 : : * called with cgroup_mutex and css_set_lock held.
1428 : : */
1429 : 0 : struct cgroup *task_cgroup_from_root(struct task_struct *task,
1430 : : struct cgroup_root *root)
1431 : : {
1432 : : /*
1433 : : * No need to lock the task - since we hold cgroup_mutex the
1434 : : * task can't change groups, so the only thing that can happen
1435 : : * is that it exits and its css is set back to init_css_set.
1436 : : */
1437 : 1180196 : return cset_cgroup_from_root(task_css_set(task), root);
1438 : : }
1439 : :
1440 : : /*
1441 : : * A task must hold cgroup_mutex to modify cgroups.
1442 : : *
1443 : : * Any task can increment and decrement the count field without lock.
1444 : : * So in general, code holding cgroup_mutex can't rely on the count
1445 : : * field not changing. However, if the count goes to zero, then only
1446 : : * cgroup_attach_task() can increment it again. Because a count of zero
1447 : : * means that no tasks are currently attached, therefore there is no
1448 : : * way a task attached to that cgroup can fork (the other way to
1449 : : * increment the count). So code holding cgroup_mutex can safely
1450 : : * assume that if the count is zero, it will stay zero. Similarly, if
1451 : : * a task holds cgroup_mutex on a cgroup with zero count, it
1452 : : * knows that the cgroup won't be removed, as cgroup_rmdir()
1453 : : * needs that mutex.
1454 : : *
1455 : : * A cgroup can only be deleted if both its 'count' of using tasks
1456 : : * is zero, and its list of 'children' cgroups is empty. Since all
1457 : : * tasks in the system use _some_ cgroup, and since there is always at
1458 : : * least one task in the system (init, pid == 1), therefore, root cgroup
1459 : : * always has either children cgroups and/or using tasks. So we don't
1460 : : * need a special hack to ensure that root cgroup cannot be deleted.
1461 : : *
1462 : : * P.S. One more locking exception. RCU is used to guard the
1463 : : * update of a tasks cgroup pointer by cgroup_attach_task()
1464 : : */
1465 : :
1466 : : static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1467 : :
1468 : 1150104 : static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1469 : : char *buf)
1470 : : {
1471 : 1150104 : struct cgroup_subsys *ss = cft->ss;
1472 : :
1473 [ + + + - : 1376000 : if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
+ - ]
1474 : 451792 : !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1475 [ + - ]: 225896 : const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1476 : :
1477 [ + + ]: 451792 : snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1478 : : dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1479 : 225896 : cft->name);
1480 : : } else {
1481 : 924208 : strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1482 : : }
1483 : 1150104 : return buf;
1484 : : }
1485 : :
1486 : : /**
1487 : : * cgroup_file_mode - deduce file mode of a control file
1488 : : * @cft: the control file in question
1489 : : *
1490 : : * S_IRUGO for read, S_IWUSR for write.
1491 : : */
1492 : 803356 : static umode_t cgroup_file_mode(const struct cftype *cft)
1493 : : {
1494 : : umode_t mode = 0;
1495 : :
1496 [ + + + + : 803356 : if (cft->read_u64 || cft->read_s64 || cft->seq_show)
+ + ]
1497 : : mode |= S_IRUGO;
1498 : :
1499 [ + + + + : 803356 : if (cft->write_u64 || cft->write_s64 || cft->write) {
+ + ]
1500 [ - + ]: 604536 : if (cft->flags & CFTYPE_WORLD_WRITABLE)
1501 : 0 : mode |= S_IWUGO;
1502 : : else
1503 : 604536 : mode |= S_IWUSR;
1504 : : }
1505 : :
1506 : 803356 : return mode;
1507 : : }
1508 : :
1509 : : /**
1510 : : * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1511 : : * @subtree_control: the new subtree_control mask to consider
1512 : : * @this_ss_mask: available subsystems
1513 : : *
1514 : : * On the default hierarchy, a subsystem may request other subsystems to be
1515 : : * enabled together through its ->depends_on mask. In such cases, more
1516 : : * subsystems than specified in "cgroup.subtree_control" may be enabled.
1517 : : *
1518 : : * This function calculates which subsystems need to be enabled if
1519 : : * @subtree_control is to be applied while restricted to @this_ss_mask.
1520 : : */
1521 : 113148 : static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1522 : : {
1523 : : u16 cur_ss_mask = subtree_control;
1524 : : struct cgroup_subsys *ss;
1525 : : int ssid;
1526 : :
1527 : : lockdep_assert_held(&cgroup_mutex);
1528 : :
1529 : 113148 : cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1530 : :
1531 : : while (true) {
1532 : : u16 new_ss_mask = cur_ss_mask;
1533 : :
1534 [ + + ]: 439660 : do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1535 : 215788 : new_ss_mask |= ss->depends_on;
1536 : : } while_each_subsys_mask();
1537 : :
1538 : : /*
1539 : : * Mask out subsystems which aren't available. This can
1540 : : * happen only if some depended-upon subsystems were bound
1541 : : * to non-default hierarchies.
1542 : : */
1543 : 223872 : new_ss_mask &= this_ss_mask;
1544 : :
1545 [ + + ]: 223872 : if (new_ss_mask == cur_ss_mask)
1546 : : break;
1547 : : cur_ss_mask = new_ss_mask;
1548 : : }
1549 : :
1550 : 113148 : return cur_ss_mask;
1551 : : }
1552 : :
1553 : : /**
1554 : : * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1555 : : * @kn: the kernfs_node being serviced
1556 : : *
1557 : : * This helper undoes cgroup_kn_lock_live() and should be invoked before
1558 : : * the method finishes if locking succeeded. Note that once this function
1559 : : * returns the cgroup returned by cgroup_kn_lock_live() may become
1560 : : * inaccessible any time. If the caller intends to continue to access the
1561 : : * cgroup, it should pin it before invoking this function.
1562 : : */
1563 : 360500 : void cgroup_kn_unlock(struct kernfs_node *kn)
1564 : : {
1565 : : struct cgroup *cgrp;
1566 : :
1567 [ + + ]: 360500 : if (kernfs_type(kn) == KERNFS_DIR)
1568 : 152756 : cgrp = kn->priv;
1569 : : else
1570 : 207744 : cgrp = kn->parent->priv;
1571 : :
1572 : 360500 : mutex_unlock(&cgroup_mutex);
1573 : :
1574 : 360500 : kernfs_unbreak_active_protection(kn);
1575 : : cgroup_put(cgrp);
1576 : 360500 : }
1577 : :
1578 : : /**
1579 : : * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1580 : : * @kn: the kernfs_node being serviced
1581 : : * @drain_offline: perform offline draining on the cgroup
1582 : : *
1583 : : * This helper is to be used by a cgroup kernfs method currently servicing
1584 : : * @kn. It breaks the active protection, performs cgroup locking and
1585 : : * verifies that the associated cgroup is alive. Returns the cgroup if
1586 : : * alive; otherwise, %NULL. A successful return should be undone by a
1587 : : * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1588 : : * cgroup is drained of offlining csses before return.
1589 : : *
1590 : : * Any cgroup kernfs method implementation which requires locking the
1591 : : * associated cgroup should use this helper. It avoids nesting cgroup
1592 : : * locking under kernfs active protection and allows all kernfs operations
1593 : : * including self-removal.
1594 : : */
1595 : 360496 : struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1596 : : {
1597 : : struct cgroup *cgrp;
1598 : :
1599 [ + + ]: 360496 : if (kernfs_type(kn) == KERNFS_DIR)
1600 : 152756 : cgrp = kn->priv;
1601 : : else
1602 : 207740 : cgrp = kn->parent->priv;
1603 : :
1604 : : /*
1605 : : * We're gonna grab cgroup_mutex which nests outside kernfs
1606 : : * active_ref. cgroup liveliness check alone provides enough
1607 : : * protection against removal. Ensure @cgrp stays accessible and
1608 : : * break the active_ref protection.
1609 : : */
1610 [ + - ]: 360492 : if (!cgroup_tryget(cgrp))
1611 : : return NULL;
1612 : 360492 : kernfs_break_active_protection(kn);
1613 : :
1614 [ - + ]: 360498 : if (drain_offline)
1615 : 0 : cgroup_lock_and_drain_offline(cgrp);
1616 : : else
1617 : 360498 : mutex_lock(&cgroup_mutex);
1618 : :
1619 [ - + ]: 360500 : if (!cgroup_is_dead(cgrp))
1620 : : return cgrp;
1621 : :
1622 : 0 : cgroup_kn_unlock(kn);
1623 : 0 : return NULL;
1624 : : }
1625 : :
1626 : 346748 : static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1627 : : {
1628 : : char name[CGROUP_FILE_NAME_MAX];
1629 : :
1630 : : lockdep_assert_held(&cgroup_mutex);
1631 : :
1632 [ + + ]: 346748 : if (cft->file_offset) {
1633 : 36372 : struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1634 : 36372 : struct cgroup_file *cfile = (void *)css + cft->file_offset;
1635 : :
1636 : : spin_lock_irq(&cgroup_file_kn_lock);
1637 : 36372 : cfile->kn = NULL;
1638 : : spin_unlock_irq(&cgroup_file_kn_lock);
1639 : :
1640 : 36372 : del_timer_sync(&cfile->notify_timer);
1641 : : }
1642 : :
1643 : 346748 : kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1644 : 346748 : }
1645 : :
1646 : : /**
1647 : : * css_clear_dir - remove subsys files in a cgroup directory
1648 : : * @css: taget css
1649 : : */
1650 : 88018 : static void css_clear_dir(struct cgroup_subsys_state *css)
1651 : : {
1652 : 88018 : struct cgroup *cgrp = css->cgroup;
1653 : : struct cftype *cfts;
1654 : :
1655 [ + + ]: 88018 : if (!(css->flags & CSS_VISIBLE))
1656 : 88018 : return;
1657 : :
1658 : 75168 : css->flags &= ~CSS_VISIBLE;
1659 : :
1660 [ + + ]: 75168 : if (!css->ss) {
1661 [ + + ]: 47688 : if (cgroup_on_dfl(cgrp))
1662 : : cfts = cgroup_base_files;
1663 : : else
1664 : : cfts = cgroup1_base_files;
1665 : :
1666 : 47688 : cgroup_addrm_files(css, cgrp, cfts, false);
1667 : : } else {
1668 [ + + ]: 55768 : list_for_each_entry(cfts, &css->ss->cfts, node)
1669 : 28288 : cgroup_addrm_files(css, cgrp, cfts, false);
1670 : : }
1671 : : }
1672 : :
1673 : : /**
1674 : : * css_populate_dir - create subsys files in a cgroup directory
1675 : : * @css: target css
1676 : : *
1677 : : * On failure, no file is added.
1678 : : */
1679 : 174242 : static int css_populate_dir(struct cgroup_subsys_state *css)
1680 : : {
1681 : 174242 : struct cgroup *cgrp = css->cgroup;
1682 : : struct cftype *cfts, *failed_cfts;
1683 : : int ret;
1684 : :
1685 [ + + + - ]: 174242 : if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1686 : : return 0;
1687 : :
1688 [ + + ]: 164064 : if (!css->ss) {
1689 [ + + ]: 109108 : if (cgroup_on_dfl(cgrp))
1690 : : cfts = cgroup_base_files;
1691 : : else
1692 : : cfts = cgroup1_base_files;
1693 : :
1694 : 109108 : ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1695 [ + - ]: 109108 : if (ret < 0)
1696 : : return ret;
1697 : : } else {
1698 [ + + ]: 111528 : list_for_each_entry(cfts, &css->ss->cfts, node) {
1699 : 56572 : ret = cgroup_addrm_files(css, cgrp, cfts, true);
1700 [ - + ]: 56572 : if (ret < 0) {
1701 : 0 : failed_cfts = cfts;
1702 : : goto err;
1703 : : }
1704 : : }
1705 : : }
1706 : :
1707 : 164064 : css->flags |= CSS_VISIBLE;
1708 : :
1709 : 164064 : return 0;
1710 : : err:
1711 [ # # ]: 0 : list_for_each_entry(cfts, &css->ss->cfts, node) {
1712 [ # # ]: 0 : if (cfts == failed_cfts)
1713 : : break;
1714 : 0 : cgroup_addrm_files(css, cgrp, cfts, false);
1715 : : }
1716 : 0 : return ret;
1717 : : }
1718 : :
1719 : 4040 : int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1720 : : {
1721 : 4040 : struct cgroup *dcgrp = &dst_root->cgrp;
1722 : : struct cgroup_subsys *ss;
1723 : : int ssid, i, ret;
1724 : :
1725 : : lockdep_assert_held(&cgroup_mutex);
1726 : :
1727 [ + + ]: 8080 : do_each_subsys_mask(ss, ssid, ss_mask) {
1728 : : /*
1729 : : * If @ss has non-root csses attached to it, can't move.
1730 : : * If @ss is an implicit controller, it is exempt from this
1731 : : * rule and can be stolen.
1732 : : */
1733 [ - + # # ]: 8080 : if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1734 : 0 : !ss->implicit_on_dfl)
1735 : 0 : return -EBUSY;
1736 : :
1737 : : /* can't move between two non-dummy roots either */
1738 [ - + # # ]: 4040 : if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1739 : : return -EBUSY;
1740 : : } while_each_subsys_mask();
1741 : :
1742 [ + + ]: 12120 : do_each_subsys_mask(ss, ssid, ss_mask) {
1743 : 4040 : struct cgroup_root *src_root = ss->root;
1744 : 4040 : struct cgroup *scgrp = &src_root->cgrp;
1745 : : struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1746 : : struct css_set *cset;
1747 : :
1748 [ + - - + : 8080 : WARN_ON(!css || cgroup_css(dcgrp, ss));
- + ]
1749 : :
1750 : : /* disable from the source */
1751 : 4040 : src_root->subsys_mask &= ~(1 << ssid);
1752 [ - + ]: 4040 : WARN_ON(cgroup_apply_control(scgrp));
1753 : 4040 : cgroup_finalize_control(scgrp, 0);
1754 : :
1755 : : /* rebind */
1756 : 4040 : RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1757 : 4040 : rcu_assign_pointer(dcgrp->subsys[ssid], css);
1758 : 4040 : ss->root = dst_root;
1759 : 4040 : css->cgroup = dcgrp;
1760 : :
1761 : : spin_lock_irq(&css_set_lock);
1762 [ + + - + : 8080 : hash_for_each(css_set_table, i, cset, hlist)
+ + + + ]
1763 : 4040 : list_move_tail(&cset->e_cset_node[ss->id],
1764 : : &dcgrp->e_csets[ss->id]);
1765 : : spin_unlock_irq(&css_set_lock);
1766 : :
1767 : : /* default hierarchy doesn't enable controllers by default */
1768 : 4040 : dst_root->subsys_mask |= 1 << ssid;
1769 [ - + ]: 4040 : if (dst_root == &cgrp_dfl_root) {
1770 : 0 : static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1771 : : } else {
1772 : 4040 : dcgrp->subtree_control |= 1 << ssid;
1773 : 4040 : static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1774 : : }
1775 : :
1776 : 4040 : ret = cgroup_apply_control(dcgrp);
1777 [ - + ]: 4040 : if (ret)
1778 : 0 : pr_warn("partial failure to rebind %s controller (err=%d)\n",
1779 : : ss->name, ret);
1780 : :
1781 [ + + ]: 4040 : if (ss->bind)
1782 : 808 : ss->bind(css);
1783 : : } while_each_subsys_mask();
1784 : :
1785 : 4040 : kernfs_activate(dcgrp->kn);
1786 : 4040 : return 0;
1787 : : }
1788 : :
1789 : 539794 : int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1790 : : struct kernfs_root *kf_root)
1791 : : {
1792 : : int len = 0;
1793 : : char *buf = NULL;
1794 : : struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1795 : : struct cgroup *ns_cgroup;
1796 : :
1797 : : buf = kmalloc(PATH_MAX, GFP_KERNEL);
1798 [ + + ]: 539886 : if (!buf)
1799 : : return -ENOMEM;
1800 : :
1801 : : spin_lock_irq(&css_set_lock);
1802 : 539888 : ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1803 : 539888 : len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1804 : : spin_unlock_irq(&css_set_lock);
1805 : :
1806 [ + - ]: 539884 : if (len >= PATH_MAX)
1807 : : len = -ERANGE;
1808 [ + - ]: 539886 : else if (len > 0) {
1809 : 539886 : seq_escape(sf, buf, " \t\n\\");
1810 : : len = 0;
1811 : : }
1812 : 539886 : kfree(buf);
1813 : 539886 : return len;
1814 : : }
1815 : :
1816 : : enum cgroup2_param {
1817 : : Opt_nsdelegate,
1818 : : Opt_memory_localevents,
1819 : : nr__cgroup2_params
1820 : : };
1821 : :
1822 : : static const struct fs_parameter_spec cgroup2_param_specs[] = {
1823 : : fsparam_flag("nsdelegate", Opt_nsdelegate),
1824 : : fsparam_flag("memory_localevents", Opt_memory_localevents),
1825 : : {}
1826 : : };
1827 : :
1828 : : static const struct fs_parameter_description cgroup2_fs_parameters = {
1829 : : .name = "cgroup2",
1830 : : .specs = cgroup2_param_specs,
1831 : : };
1832 : :
1833 : 808 : static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1834 : : {
1835 : : struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1836 : : struct fs_parse_result result;
1837 : : int opt;
1838 : :
1839 : 808 : opt = fs_parse(fc, &cgroup2_fs_parameters, param, &result);
1840 [ + + ]: 808 : if (opt < 0)
1841 : : return opt;
1842 : :
1843 [ + - - ]: 404 : switch (opt) {
1844 : : case Opt_nsdelegate:
1845 : 404 : ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1846 : 404 : return 0;
1847 : : case Opt_memory_localevents:
1848 : 0 : ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1849 : 0 : return 0;
1850 : : }
1851 : : return -EINVAL;
1852 : : }
1853 : :
1854 : 404 : static void apply_cgroup_root_flags(unsigned int root_flags)
1855 : : {
1856 [ + - ]: 404 : if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1857 [ + - ]: 404 : if (root_flags & CGRP_ROOT_NS_DELEGATE)
1858 : 404 : cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1859 : : else
1860 : 0 : cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1861 : :
1862 [ - + ]: 404 : if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1863 : 0 : cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1864 : : else
1865 : 404 : cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1866 : : }
1867 : 404 : }
1868 : :
1869 : 55532 : static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1870 : : {
1871 [ + + ]: 55532 : if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1872 : 55566 : seq_puts(seq, ",nsdelegate");
1873 [ - + ]: 55566 : if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1874 : 0 : seq_puts(seq, ",memory_localevents");
1875 : 55566 : return 0;
1876 : : }
1877 : :
1878 : 0 : static int cgroup_reconfigure(struct fs_context *fc)
1879 : : {
1880 : : struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1881 : :
1882 : 0 : apply_cgroup_root_flags(ctx->flags);
1883 : 0 : return 0;
1884 : : }
1885 : :
1886 : : /*
1887 : : * To reduce the fork() overhead for systems that are not actually using
1888 : : * their cgroups capability, we don't maintain the lists running through
1889 : : * each css_set to its tasks until we see the list actually used - in other
1890 : : * words after the first mount.
1891 : : */
1892 : : static bool use_task_css_set_links __read_mostly;
1893 : :
1894 : 404 : void cgroup_enable_task_cg_lists(void)
1895 : : {
1896 : : struct task_struct *p, *g;
1897 : :
1898 : : /*
1899 : : * We need tasklist_lock because RCU is not safe against
1900 : : * while_each_thread(). Besides, a forking task that has passed
1901 : : * cgroup_post_fork() without seeing use_task_css_set_links = 1
1902 : : * is not guaranteed to have its child immediately visible in the
1903 : : * tasklist if we walk through it with RCU.
1904 : : */
1905 : 404 : read_lock(&tasklist_lock);
1906 : : spin_lock_irq(&css_set_lock);
1907 : :
1908 [ + - ]: 404 : if (use_task_css_set_links)
1909 : : goto out_unlock;
1910 : :
1911 : 404 : use_task_css_set_links = true;
1912 : :
1913 [ + + ]: 26586 : do_each_thread(g, p) {
1914 [ + - - + : 77334 : WARN_ON_ONCE(!list_empty(&p->cg_list) ||
- + # # ]
1915 : : task_css_set(p) != &init_css_set);
1916 : :
1917 : : /*
1918 : : * We should check if the process is exiting, otherwise
1919 : : * it will race with cgroup_exit() in that the list
1920 : : * entry won't be deleted though the process has exited.
1921 : : * Do it while holding siglock so that we don't end up
1922 : : * racing against cgroup_exit().
1923 : : *
1924 : : * Interrupts were already disabled while acquiring
1925 : : * the css_set_lock, so we do not need to disable it
1926 : : * again when acquiring the sighand->siglock here.
1927 : : */
1928 : 25778 : spin_lock(&p->sighand->siglock);
1929 [ + - ]: 25778 : if (!(p->flags & PF_EXITING)) {
1930 : : struct css_set *cset = task_css_set(p);
1931 : :
1932 [ + + ]: 25778 : if (!css_set_populated(cset))
1933 : 404 : css_set_update_populated(cset, true);
1934 : : list_add_tail(&p->cg_list, &cset->tasks);
1935 : : get_css_set(cset);
1936 : 25778 : cset->nr_tasks++;
1937 : : }
1938 : 25778 : spin_unlock(&p->sighand->siglock);
1939 [ - + ]: 25778 : } while_each_thread(g, p);
1940 : : out_unlock:
1941 : : spin_unlock_irq(&css_set_lock);
1942 : : read_unlock(&tasklist_lock);
1943 : 404 : }
1944 : :
1945 : 109108 : static void init_cgroup_housekeeping(struct cgroup *cgrp)
1946 : : {
1947 : : struct cgroup_subsys *ss;
1948 : : int ssid;
1949 : :
1950 : 109108 : INIT_LIST_HEAD(&cgrp->self.sibling);
1951 : 109108 : INIT_LIST_HEAD(&cgrp->self.children);
1952 : 109108 : INIT_LIST_HEAD(&cgrp->cset_links);
1953 : 109108 : INIT_LIST_HEAD(&cgrp->pidlists);
1954 : 109108 : mutex_init(&cgrp->pidlist_mutex);
1955 : 109108 : cgrp->self.cgroup = cgrp;
1956 : 109108 : cgrp->self.flags |= CSS_ONLINE;
1957 : 109108 : cgrp->dom_cgrp = cgrp;
1958 : 109108 : cgrp->max_descendants = INT_MAX;
1959 : 109108 : cgrp->max_depth = INT_MAX;
1960 : 109108 : INIT_LIST_HEAD(&cgrp->rstat_css_list);
1961 : : prev_cputime_init(&cgrp->prev_cputime);
1962 : :
1963 [ + + ]: 1309296 : for_each_subsys(ss, ssid)
1964 : 1200188 : INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1965 : :
1966 : 109108 : init_waitqueue_head(&cgrp->offline_waitq);
1967 : 218216 : INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1968 : 109108 : }
1969 : :
1970 : 4040 : void init_cgroup_root(struct cgroup_fs_context *ctx)
1971 : : {
1972 : 4040 : struct cgroup_root *root = ctx->root;
1973 : 4040 : struct cgroup *cgrp = &root->cgrp;
1974 : :
1975 : 4040 : INIT_LIST_HEAD(&root->root_list);
1976 : : atomic_set(&root->nr_cgrps, 1);
1977 : 4040 : cgrp->root = root;
1978 : 4040 : init_cgroup_housekeeping(cgrp);
1979 : : idr_init(&root->cgroup_idr);
1980 : :
1981 : 4040 : root->flags = ctx->flags;
1982 [ - + ]: 4040 : if (ctx->release_agent)
1983 : 0 : strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1984 [ + + ]: 4040 : if (ctx->name)
1985 : 404 : strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1986 [ - + ]: 4040 : if (ctx->cpuset_clone_children)
1987 : 0 : set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1988 : 4040 : }
1989 : :
1990 : 4040 : int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1991 : : {
1992 : 4040 : LIST_HEAD(tmp_links);
1993 : 4040 : struct cgroup *root_cgrp = &root->cgrp;
1994 : : struct kernfs_syscall_ops *kf_sops;
1995 : : struct css_set *cset;
1996 : : int i, ret;
1997 : :
1998 : : lockdep_assert_held(&cgroup_mutex);
1999 : :
2000 : 4040 : ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
2001 [ + - ]: 4040 : if (ret < 0)
2002 : : goto out;
2003 : 4040 : root_cgrp->id = ret;
2004 : 4040 : root_cgrp->ancestor_ids[0] = ret;
2005 : :
2006 : 4040 : ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
2007 : : 0, GFP_KERNEL);
2008 [ + - ]: 4040 : if (ret)
2009 : : goto out;
2010 : :
2011 : : /*
2012 : : * We're accessing css_set_count without locking css_set_lock here,
2013 : : * but that's OK - it can only be increased by someone holding
2014 : : * cgroup_lock, and that's us. Later rebinding may disable
2015 : : * controllers on the default hierarchy and thus create new csets,
2016 : : * which can't be more than the existing ones. Allocate 2x.
2017 : : */
2018 : 4040 : ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
2019 [ + - ]: 4040 : if (ret)
2020 : : goto cancel_ref;
2021 : :
2022 : 4040 : ret = cgroup_init_root_id(root);
2023 [ + - ]: 4040 : if (ret)
2024 : : goto cancel_ref;
2025 : :
2026 : : kf_sops = root == &cgrp_dfl_root ?
2027 [ + + ]: 4040 : &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2028 : :
2029 : 4040 : root->kf_root = kernfs_create_root(kf_sops,
2030 : : KERNFS_ROOT_CREATE_DEACTIVATED |
2031 : : KERNFS_ROOT_SUPPORT_EXPORTOP,
2032 : : root_cgrp);
2033 [ - + ]: 4040 : if (IS_ERR(root->kf_root)) {
2034 : : ret = PTR_ERR(root->kf_root);
2035 : 0 : goto exit_root_id;
2036 : : }
2037 : 4040 : root_cgrp->kn = root->kf_root->kn;
2038 : :
2039 : 4040 : ret = css_populate_dir(&root_cgrp->self);
2040 [ + - ]: 4040 : if (ret)
2041 : : goto destroy_root;
2042 : :
2043 : 4040 : ret = rebind_subsystems(root, ss_mask);
2044 [ + - ]: 4040 : if (ret)
2045 : : goto destroy_root;
2046 : :
2047 : 4040 : ret = cgroup_bpf_inherit(root_cgrp);
2048 [ - + # # ]: 4040 : WARN_ON_ONCE(ret);
2049 : :
2050 : 4040 : trace_cgroup_setup_root(root);
2051 : :
2052 : : /*
2053 : : * There must be no failure case after here, since rebinding takes
2054 : : * care of subsystems' refcounts, which are explicitly dropped in
2055 : : * the failure exit path.
2056 : : */
2057 : 4040 : list_add(&root->root_list, &cgroup_roots);
2058 : 4040 : cgroup_root_count++;
2059 : :
2060 : : /*
2061 : : * Link the root cgroup in this hierarchy into all the css_set
2062 : : * objects.
2063 : : */
2064 : : spin_lock_irq(&css_set_lock);
2065 [ + + - + : 8080 : hash_for_each(css_set_table, i, cset, hlist) {
+ + + + ]
2066 : 4040 : link_css_set(&tmp_links, cset, root_cgrp);
2067 [ + + ]: 4040 : if (css_set_populated(cset))
2068 : 3636 : cgroup_update_populated(root_cgrp, true);
2069 : : }
2070 : : spin_unlock_irq(&css_set_lock);
2071 : :
2072 [ - + ]: 8080 : BUG_ON(!list_empty(&root_cgrp->self.children));
2073 [ - + ]: 4040 : BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2074 : :
2075 : 4040 : kernfs_activate(root_cgrp->kn);
2076 : : ret = 0;
2077 : 4040 : goto out;
2078 : :
2079 : : destroy_root:
2080 : 0 : kernfs_destroy_root(root->kf_root);
2081 : 0 : root->kf_root = NULL;
2082 : : exit_root_id:
2083 : : cgroup_exit_root_id(root);
2084 : : cancel_ref:
2085 : 0 : percpu_ref_exit(&root_cgrp->self.refcnt);
2086 : : out:
2087 : 4040 : free_cgrp_cset_links(&tmp_links);
2088 : 4040 : return ret;
2089 : : }
2090 : :
2091 : 4040 : int cgroup_do_get_tree(struct fs_context *fc)
2092 : : {
2093 : : struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2094 : : int ret;
2095 : :
2096 : 4040 : ctx->kfc.root = ctx->root->kf_root;
2097 [ + + ]: 4040 : if (fc->fs_type == &cgroup2_fs_type)
2098 : 404 : ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2099 : : else
2100 : 3636 : ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2101 : 4040 : ret = kernfs_get_tree(fc);
2102 : :
2103 : : /*
2104 : : * In non-init cgroup namespace, instead of root cgroup's dentry,
2105 : : * we return the dentry corresponding to the cgroupns->root_cgrp.
2106 : : */
2107 [ + - - + ]: 4040 : if (!ret && ctx->ns != &init_cgroup_ns) {
2108 : : struct dentry *nsdentry;
2109 : 0 : struct super_block *sb = fc->root->d_sb;
2110 : : struct cgroup *cgrp;
2111 : :
2112 : 0 : mutex_lock(&cgroup_mutex);
2113 : : spin_lock_irq(&css_set_lock);
2114 : :
2115 : 0 : cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2116 : :
2117 : : spin_unlock_irq(&css_set_lock);
2118 : 0 : mutex_unlock(&cgroup_mutex);
2119 : :
2120 : 0 : nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2121 : 0 : dput(fc->root);
2122 [ # # ]: 0 : if (IS_ERR(nsdentry)) {
2123 : 0 : deactivate_locked_super(sb);
2124 : : ret = PTR_ERR(nsdentry);
2125 : : nsdentry = NULL;
2126 : : }
2127 : 0 : fc->root = nsdentry;
2128 : : }
2129 : :
2130 [ - + ]: 4040 : if (!ctx->kfc.new_sb_created)
2131 : 0 : cgroup_put(&ctx->root->cgrp);
2132 : :
2133 : 4040 : return ret;
2134 : : }
2135 : :
2136 : : /*
2137 : : * Destroy a cgroup filesystem context.
2138 : : */
2139 : 4040 : static void cgroup_fs_context_free(struct fs_context *fc)
2140 : : {
2141 : : struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2142 : :
2143 : 4040 : kfree(ctx->name);
2144 : 4040 : kfree(ctx->release_agent);
2145 : 4040 : put_cgroup_ns(ctx->ns);
2146 : 4040 : kernfs_free_fs_context(fc);
2147 : 4040 : kfree(ctx);
2148 : 4040 : }
2149 : :
2150 : 404 : static int cgroup_get_tree(struct fs_context *fc)
2151 : : {
2152 : : struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2153 : : int ret;
2154 : :
2155 : 404 : cgrp_dfl_visible = true;
2156 : 404 : cgroup_get_live(&cgrp_dfl_root.cgrp);
2157 : 404 : ctx->root = &cgrp_dfl_root;
2158 : :
2159 : 404 : ret = cgroup_do_get_tree(fc);
2160 [ + - ]: 404 : if (!ret)
2161 : 404 : apply_cgroup_root_flags(ctx->flags);
2162 : 404 : return ret;
2163 : : }
2164 : :
2165 : : static const struct fs_context_operations cgroup_fs_context_ops = {
2166 : : .free = cgroup_fs_context_free,
2167 : : .parse_param = cgroup2_parse_param,
2168 : : .get_tree = cgroup_get_tree,
2169 : : .reconfigure = cgroup_reconfigure,
2170 : : };
2171 : :
2172 : : static const struct fs_context_operations cgroup1_fs_context_ops = {
2173 : : .free = cgroup_fs_context_free,
2174 : : .parse_param = cgroup1_parse_param,
2175 : : .get_tree = cgroup1_get_tree,
2176 : : .reconfigure = cgroup1_reconfigure,
2177 : : };
2178 : :
2179 : : /*
2180 : : * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2181 : : * we select the namespace we're going to use.
2182 : : */
2183 : 4040 : static int cgroup_init_fs_context(struct fs_context *fc)
2184 : : {
2185 : : struct cgroup_fs_context *ctx;
2186 : :
2187 : 4040 : ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2188 [ + - ]: 4040 : if (!ctx)
2189 : : return -ENOMEM;
2190 : :
2191 : : /*
2192 : : * The first time anyone tries to mount a cgroup, enable the list
2193 : : * linking each css_set to its tasks and fix up all existing tasks.
2194 : : */
2195 [ + + ]: 4040 : if (!use_task_css_set_links)
2196 : 404 : cgroup_enable_task_cg_lists();
2197 : :
2198 : 4040 : ctx->ns = current->nsproxy->cgroup_ns;
2199 : : get_cgroup_ns(ctx->ns);
2200 : 4040 : fc->fs_private = &ctx->kfc;
2201 [ + + ]: 4040 : if (fc->fs_type == &cgroup2_fs_type)
2202 : 404 : fc->ops = &cgroup_fs_context_ops;
2203 : : else
2204 : 3636 : fc->ops = &cgroup1_fs_context_ops;
2205 : 4040 : put_user_ns(fc->user_ns);
2206 : 4040 : fc->user_ns = get_user_ns(ctx->ns->user_ns);
2207 : 4040 : fc->global = true;
2208 : 4040 : return 0;
2209 : : }
2210 : :
2211 : 0 : static void cgroup_kill_sb(struct super_block *sb)
2212 : : {
2213 : 0 : struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2214 : : struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2215 : :
2216 : : /*
2217 : : * If @root doesn't have any children, start killing it.
2218 : : * This prevents new mounts by disabling percpu_ref_tryget_live().
2219 : : * cgroup_mount() may wait for @root's release.
2220 : : *
2221 : : * And don't kill the default root.
2222 : : */
2223 [ # # # # : 0 : if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
# # ]
2224 : : !percpu_ref_is_dying(&root->cgrp.self.refcnt))
2225 : 0 : percpu_ref_kill(&root->cgrp.self.refcnt);
2226 : : cgroup_put(&root->cgrp);
2227 : 0 : kernfs_kill_sb(sb);
2228 : 0 : }
2229 : :
2230 : : struct file_system_type cgroup_fs_type = {
2231 : : .name = "cgroup",
2232 : : .init_fs_context = cgroup_init_fs_context,
2233 : : .parameters = &cgroup1_fs_parameters,
2234 : : .kill_sb = cgroup_kill_sb,
2235 : : .fs_flags = FS_USERNS_MOUNT,
2236 : : };
2237 : :
2238 : : static struct file_system_type cgroup2_fs_type = {
2239 : : .name = "cgroup2",
2240 : : .init_fs_context = cgroup_init_fs_context,
2241 : : .parameters = &cgroup2_fs_parameters,
2242 : : .kill_sb = cgroup_kill_sb,
2243 : : .fs_flags = FS_USERNS_MOUNT,
2244 : : };
2245 : :
2246 : : #ifdef CONFIG_CPUSETS
2247 : : static const struct fs_context_operations cpuset_fs_context_ops = {
2248 : : .get_tree = cgroup1_get_tree,
2249 : : .free = cgroup_fs_context_free,
2250 : : };
2251 : :
2252 : : /*
2253 : : * This is ugly, but preserves the userspace API for existing cpuset
2254 : : * users. If someone tries to mount the "cpuset" filesystem, we
2255 : : * silently switch it to mount "cgroup" instead
2256 : : */
2257 : 0 : static int cpuset_init_fs_context(struct fs_context *fc)
2258 : : {
2259 : 0 : char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2260 : : struct cgroup_fs_context *ctx;
2261 : : int err;
2262 : :
2263 : 0 : err = cgroup_init_fs_context(fc);
2264 [ # # ]: 0 : if (err) {
2265 : 0 : kfree(agent);
2266 : 0 : return err;
2267 : : }
2268 : :
2269 : 0 : fc->ops = &cpuset_fs_context_ops;
2270 : :
2271 : : ctx = cgroup_fc2context(fc);
2272 : 0 : ctx->subsys_mask = 1 << cpuset_cgrp_id;
2273 : 0 : ctx->flags |= CGRP_ROOT_NOPREFIX;
2274 : 0 : ctx->release_agent = agent;
2275 : :
2276 : 0 : get_filesystem(&cgroup_fs_type);
2277 : 0 : put_filesystem(fc->fs_type);
2278 : 0 : fc->fs_type = &cgroup_fs_type;
2279 : :
2280 : 0 : return 0;
2281 : : }
2282 : :
2283 : : static struct file_system_type cpuset_fs_type = {
2284 : : .name = "cpuset",
2285 : : .init_fs_context = cpuset_init_fs_context,
2286 : : .fs_flags = FS_USERNS_MOUNT,
2287 : : };
2288 : : #endif
2289 : :
2290 : 937488 : int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2291 : : struct cgroup_namespace *ns)
2292 : : {
2293 : 937488 : struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2294 : :
2295 : 937488 : return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2296 : : }
2297 : :
2298 : 0 : int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2299 : : struct cgroup_namespace *ns)
2300 : : {
2301 : : int ret;
2302 : :
2303 : 0 : mutex_lock(&cgroup_mutex);
2304 : : spin_lock_irq(&css_set_lock);
2305 : :
2306 : 0 : ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2307 : :
2308 : : spin_unlock_irq(&css_set_lock);
2309 : 0 : mutex_unlock(&cgroup_mutex);
2310 : :
2311 : 0 : return ret;
2312 : : }
2313 : : EXPORT_SYMBOL_GPL(cgroup_path_ns);
2314 : :
2315 : : /**
2316 : : * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2317 : : * @task: target task
2318 : : * @buf: the buffer to write the path into
2319 : : * @buflen: the length of the buffer
2320 : : *
2321 : : * Determine @task's cgroup on the first (the one with the lowest non-zero
2322 : : * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2323 : : * function grabs cgroup_mutex and shouldn't be used inside locks used by
2324 : : * cgroup controller callbacks.
2325 : : *
2326 : : * Return value is the same as kernfs_path().
2327 : : */
2328 : 0 : int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2329 : : {
2330 : : struct cgroup_root *root;
2331 : : struct cgroup *cgrp;
2332 : 0 : int hierarchy_id = 1;
2333 : : int ret;
2334 : :
2335 : 0 : mutex_lock(&cgroup_mutex);
2336 : : spin_lock_irq(&css_set_lock);
2337 : :
2338 : 0 : root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2339 : :
2340 [ # # ]: 0 : if (root) {
2341 : : cgrp = task_cgroup_from_root(task, root);
2342 : 0 : ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2343 : : } else {
2344 : : /* if no hierarchy exists, everyone is in "/" */
2345 : 0 : ret = strlcpy(buf, "/", buflen);
2346 : : }
2347 : :
2348 : : spin_unlock_irq(&css_set_lock);
2349 : 0 : mutex_unlock(&cgroup_mutex);
2350 : 0 : return ret;
2351 : : }
2352 : : EXPORT_SYMBOL_GPL(task_cgroup_path);
2353 : :
2354 : : /**
2355 : : * cgroup_migrate_add_task - add a migration target task to a migration context
2356 : : * @task: target task
2357 : : * @mgctx: target migration context
2358 : : *
2359 : : * Add @task, which is a migration target, to @mgctx->tset. This function
2360 : : * becomes noop if @task doesn't need to be migrated. @task's css_set
2361 : : * should have been added as a migration source and @task->cg_list will be
2362 : : * moved from the css_set's tasks list to mg_tasks one.
2363 : : */
2364 : 212592 : static void cgroup_migrate_add_task(struct task_struct *task,
2365 : : struct cgroup_mgctx *mgctx)
2366 : : {
2367 : : struct css_set *cset;
2368 : :
2369 : : lockdep_assert_held(&css_set_lock);
2370 : :
2371 : : /* @task either already exited or can't exit until the end */
2372 [ + - ]: 212592 : if (task->flags & PF_EXITING)
2373 : : return;
2374 : :
2375 : : /* leave @task alone if post_fork() hasn't linked it yet */
2376 [ + - ]: 425184 : if (list_empty(&task->cg_list))
2377 : : return;
2378 : :
2379 : : cset = task_css_set(task);
2380 [ + + ]: 212592 : if (!cset->mg_src_cgrp)
2381 : : return;
2382 : :
2383 : 109128 : mgctx->tset.nr_tasks++;
2384 : :
2385 : 109128 : list_move_tail(&task->cg_list, &cset->mg_tasks);
2386 [ + + ]: 218256 : if (list_empty(&cset->mg_node))
2387 : 105896 : list_add_tail(&cset->mg_node,
2388 : : &mgctx->tset.src_csets);
2389 [ + + ]: 218256 : if (list_empty(&cset->mg_dst_cset->mg_node))
2390 : 105896 : list_add_tail(&cset->mg_dst_cset->mg_node,
2391 : : &mgctx->tset.dst_csets);
2392 : : }
2393 : :
2394 : : /**
2395 : : * cgroup_taskset_first - reset taskset and return the first task
2396 : : * @tset: taskset of interest
2397 : : * @dst_cssp: output variable for the destination css
2398 : : *
2399 : : * @tset iteration is initialized and the first task is returned.
2400 : : */
2401 : 24252 : struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2402 : : struct cgroup_subsys_state **dst_cssp)
2403 : : {
2404 : 24252 : tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2405 : 24252 : tset->cur_task = NULL;
2406 : :
2407 : 24252 : return cgroup_taskset_next(tset, dst_cssp);
2408 : : }
2409 : :
2410 : : /**
2411 : : * cgroup_taskset_next - iterate to the next task in taskset
2412 : : * @tset: taskset of interest
2413 : : * @dst_cssp: output variable for the destination css
2414 : : *
2415 : : * Return the next task in @tset. Iteration must have been initialized
2416 : : * with cgroup_taskset_first().
2417 : : */
2418 : 49312 : struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2419 : : struct cgroup_subsys_state **dst_cssp)
2420 : : {
2421 : 49312 : struct css_set *cset = tset->cur_cset;
2422 : 49312 : struct task_struct *task = tset->cur_task;
2423 : :
2424 [ + + ]: 122876 : while (&cset->mg_node != tset->csets) {
2425 [ + + ]: 49312 : if (!task)
2426 : 24252 : task = list_first_entry(&cset->mg_tasks,
2427 : : struct task_struct, cg_list);
2428 : : else
2429 : 25060 : task = list_next_entry(task, cg_list);
2430 : :
2431 [ + + ]: 49312 : if (&task->cg_list != &cset->mg_tasks) {
2432 : 25060 : tset->cur_cset = cset;
2433 : 25060 : tset->cur_task = task;
2434 : :
2435 : : /*
2436 : : * This function may be called both before and
2437 : : * after cgroup_taskset_migrate(). The two cases
2438 : : * can be distinguished by looking at whether @cset
2439 : : * has its ->mg_dst_cset set.
2440 : : */
2441 [ + - ]: 25060 : if (cset->mg_dst_cset)
2442 : 25060 : *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2443 : : else
2444 : 0 : *dst_cssp = cset->subsys[tset->ssid];
2445 : :
2446 : 25060 : return task;
2447 : : }
2448 : :
2449 : 24252 : cset = list_next_entry(cset, mg_node);
2450 : : task = NULL;
2451 : : }
2452 : :
2453 : : return NULL;
2454 : : }
2455 : :
2456 : : /**
2457 : : * cgroup_taskset_migrate - migrate a taskset
2458 : : * @mgctx: migration context
2459 : : *
2460 : : * Migrate tasks in @mgctx as setup by migration preparation functions.
2461 : : * This function fails iff one of the ->can_attach callbacks fails and
2462 : : * guarantees that either all or none of the tasks in @mgctx are migrated.
2463 : : * @mgctx is consumed regardless of success.
2464 : : */
2465 : 215824 : static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2466 : : {
2467 : 215824 : struct cgroup_taskset *tset = &mgctx->tset;
2468 : : struct cgroup_subsys *ss;
2469 : : struct task_struct *task, *tmp_task;
2470 : : struct css_set *cset, *tmp_cset;
2471 : : int ssid, failed_ssid, ret;
2472 : :
2473 : : /* check that we can legitimately attach to the cgroup */
2474 [ + + ]: 215824 : if (tset->nr_tasks) {
2475 [ + + ]: 154400 : do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2476 [ + + ]: 48504 : if (ss->can_attach) {
2477 : 24252 : tset->ssid = ssid;
2478 : 24252 : ret = ss->can_attach(tset);
2479 [ - + ]: 24252 : if (ret) {
2480 : 0 : failed_ssid = ssid;
2481 : 0 : goto out_cancel_attach;
2482 : : }
2483 : : }
2484 : : } while_each_subsys_mask();
2485 : : }
2486 : :
2487 : : /*
2488 : : * Now that we're guaranteed success, proceed to move all tasks to
2489 : : * the new cgroup. There are no failure cases after here, so this
2490 : : * is the commit point.
2491 : : */
2492 : : spin_lock_irq(&css_set_lock);
2493 [ + + ]: 321720 : list_for_each_entry(cset, &tset->src_csets, mg_node) {
2494 [ + + ]: 215024 : list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2495 : : struct css_set *from_cset = task_css_set(task);
2496 : 109128 : struct css_set *to_cset = cset->mg_dst_cset;
2497 : :
2498 : : get_css_set(to_cset);
2499 : 109128 : to_cset->nr_tasks++;
2500 : 109128 : css_set_move_task(task, from_cset, to_cset, true);
2501 : 109128 : from_cset->nr_tasks--;
2502 : : /*
2503 : : * If the source or destination cgroup is frozen,
2504 : : * the task might require to change its state.
2505 : : */
2506 : 109128 : cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2507 : : to_cset->dfl_cgrp);
2508 : 109128 : put_css_set_locked(from_cset);
2509 : :
2510 : : }
2511 : : }
2512 : : spin_unlock_irq(&css_set_lock);
2513 : :
2514 : : /*
2515 : : * Migration is committed, all target tasks are now on dst_csets.
2516 : : * Nothing is sensitive to fork() after this point. Notify
2517 : : * controllers that migration is complete.
2518 : : */
2519 : 215824 : tset->csets = &tset->dst_csets;
2520 : :
2521 [ + + ]: 215824 : if (tset->nr_tasks) {
2522 [ + + ]: 154400 : do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2523 [ - + ]: 48504 : if (ss->attach) {
2524 : 0 : tset->ssid = ssid;
2525 : 0 : ss->attach(tset);
2526 : : }
2527 : : } while_each_subsys_mask();
2528 : : }
2529 : :
2530 : : ret = 0;
2531 : : goto out_release_tset;
2532 : :
2533 : : out_cancel_attach:
2534 [ # # ]: 0 : if (tset->nr_tasks) {
2535 [ # # ]: 0 : do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2536 [ # # ]: 0 : if (ssid == failed_ssid)
2537 : : break;
2538 [ # # ]: 0 : if (ss->cancel_attach) {
2539 : 0 : tset->ssid = ssid;
2540 : 0 : ss->cancel_attach(tset);
2541 : : }
2542 : : } while_each_subsys_mask();
2543 : : }
2544 : : out_release_tset:
2545 : : spin_lock_irq(&css_set_lock);
2546 : 215824 : list_splice_init(&tset->dst_csets, &tset->src_csets);
2547 [ + + ]: 427616 : list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2548 : 211792 : list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2549 : : list_del_init(&cset->mg_node);
2550 : : }
2551 : : spin_unlock_irq(&css_set_lock);
2552 : :
2553 : : /*
2554 : : * Re-initialize the cgroup_taskset structure in case it is reused
2555 : : * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2556 : : * iteration.
2557 : : */
2558 : 215824 : tset->nr_tasks = 0;
2559 : 215824 : tset->csets = &tset->src_csets;
2560 : 215824 : return ret;
2561 : : }
2562 : :
2563 : : /**
2564 : : * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2565 : : * @dst_cgrp: destination cgroup to test
2566 : : *
2567 : : * On the default hierarchy, except for the mixable, (possible) thread root
2568 : : * and threaded cgroups, subtree_control must be zero for migration
2569 : : * destination cgroups with tasks so that child cgroups don't compete
2570 : : * against tasks.
2571 : : */
2572 : 207744 : int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2573 : : {
2574 : : /* v1 doesn't have any restriction */
2575 [ + + ]: 207744 : if (!cgroup_on_dfl(dst_cgrp))
2576 : : return 0;
2577 : :
2578 : : /* verify @dst_cgrp can host resources */
2579 [ + - ]: 55368 : if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2580 : : return -EOPNOTSUPP;
2581 : :
2582 : : /* mixables don't care */
2583 [ + - ]: 55368 : if (cgroup_is_mixable(dst_cgrp))
2584 : : return 0;
2585 : :
2586 : : /*
2587 : : * If @dst_cgrp is already or can become a thread root or is
2588 : : * threaded, it doesn't matter.
2589 : : */
2590 [ - + # # ]: 55368 : if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2591 : : return 0;
2592 : :
2593 : : /* apply no-internal-process constraint */
2594 [ # # ]: 0 : if (dst_cgrp->subtree_control)
2595 : : return -EBUSY;
2596 : :
2597 : 0 : return 0;
2598 : : }
2599 : :
2600 : : /**
2601 : : * cgroup_migrate_finish - cleanup after attach
2602 : : * @mgctx: migration context
2603 : : *
2604 : : * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2605 : : * those functions for details.
2606 : : */
2607 : 215824 : void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2608 : : {
2609 : 215824 : LIST_HEAD(preloaded);
2610 : : struct css_set *cset, *tmp_cset;
2611 : :
2612 : : lockdep_assert_held(&cgroup_mutex);
2613 : :
2614 : : spin_lock_irq(&css_set_lock);
2615 : :
2616 : 215824 : list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2617 : 215824 : list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2618 : :
2619 [ + + ]: 427616 : list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2620 : 211792 : cset->mg_src_cgrp = NULL;
2621 : 211792 : cset->mg_dst_cgrp = NULL;
2622 : 211792 : cset->mg_dst_cset = NULL;
2623 : : list_del_init(&cset->mg_preload_node);
2624 : 211792 : put_css_set_locked(cset);
2625 : : }
2626 : :
2627 : : spin_unlock_irq(&css_set_lock);
2628 : 215824 : }
2629 : :
2630 : : /**
2631 : : * cgroup_migrate_add_src - add a migration source css_set
2632 : : * @src_cset: the source css_set to add
2633 : : * @dst_cgrp: the destination cgroup
2634 : : * @mgctx: migration context
2635 : : *
2636 : : * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2637 : : * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2638 : : * up by cgroup_migrate_finish().
2639 : : *
2640 : : * This function may be called without holding cgroup_threadgroup_rwsem
2641 : : * even if the target is a process. Threads may be created and destroyed
2642 : : * but as long as cgroup_mutex is not dropped, no new css_set can be put
2643 : : * into play and the preloaded css_sets are guaranteed to cover all
2644 : : * migrations.
2645 : : */
2646 : 216632 : void cgroup_migrate_add_src(struct css_set *src_cset,
2647 : : struct cgroup *dst_cgrp,
2648 : : struct cgroup_mgctx *mgctx)
2649 : : {
2650 : : struct cgroup *src_cgrp;
2651 : :
2652 : : lockdep_assert_held(&cgroup_mutex);
2653 : : lockdep_assert_held(&css_set_lock);
2654 : :
2655 : : /*
2656 : : * If ->dead, @src_set is associated with one or more dead cgroups
2657 : : * and doesn't contain any migratable tasks. Ignore it early so
2658 : : * that the rest of migration path doesn't get confused by it.
2659 : : */
2660 [ + - ]: 216632 : if (src_cset->dead)
2661 : : return;
2662 : :
2663 : 216632 : src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2664 : :
2665 [ + + ]: 433264 : if (!list_empty(&src_cset->mg_preload_node))
2666 : : return;
2667 : :
2668 [ - + ]: 211784 : WARN_ON(src_cset->mg_src_cgrp);
2669 [ - + ]: 211784 : WARN_ON(src_cset->mg_dst_cgrp);
2670 [ - + ]: 423568 : WARN_ON(!list_empty(&src_cset->mg_tasks));
2671 [ - + ]: 423568 : WARN_ON(!list_empty(&src_cset->mg_node));
2672 : :
2673 : 211784 : src_cset->mg_src_cgrp = src_cgrp;
2674 : 211784 : src_cset->mg_dst_cgrp = dst_cgrp;
2675 : : get_css_set(src_cset);
2676 : 211784 : list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2677 : : }
2678 : :
2679 : : /**
2680 : : * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2681 : : * @mgctx: migration context
2682 : : *
2683 : : * Tasks are about to be moved and all the source css_sets have been
2684 : : * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2685 : : * pins all destination css_sets, links each to its source, and append them
2686 : : * to @mgctx->preloaded_dst_csets.
2687 : : *
2688 : : * This function must be called after cgroup_migrate_add_src() has been
2689 : : * called on each migration source css_set. After migration is performed
2690 : : * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2691 : : * @mgctx.
2692 : : */
2693 : 215824 : int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2694 : : {
2695 : : struct css_set *src_cset, *tmp_cset;
2696 : :
2697 : : lockdep_assert_held(&cgroup_mutex);
2698 : :
2699 : : /* look up the dst cset for each src cset and link it to src */
2700 [ + + ]: 427608 : list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2701 : : mg_preload_node) {
2702 : : struct css_set *dst_cset;
2703 : : struct cgroup_subsys *ss;
2704 : : int ssid;
2705 : :
2706 : 211784 : dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2707 [ + - ]: 211784 : if (!dst_cset)
2708 : : return -ENOMEM;
2709 : :
2710 [ + - + - : 211784 : WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
- + # # ]
2711 : :
2712 : : /*
2713 : : * If src cset equals dst, it's noop. Drop the src.
2714 : : * cgroup_migrate() will skip the cset too. Note that we
2715 : : * can't handle src == dst as some nodes are used by both.
2716 : : */
2717 [ + + ]: 211784 : if (src_cset == dst_cset) {
2718 : 105888 : src_cset->mg_src_cgrp = NULL;
2719 : 105888 : src_cset->mg_dst_cgrp = NULL;
2720 : : list_del_init(&src_cset->mg_preload_node);
2721 : 105888 : put_css_set(src_cset);
2722 : 105888 : put_css_set(dst_cset);
2723 : 105888 : continue;
2724 : : }
2725 : :
2726 : 105896 : src_cset->mg_dst_cset = dst_cset;
2727 : :
2728 [ + - ]: 211792 : if (list_empty(&dst_cset->mg_preload_node))
2729 : 105896 : list_add_tail(&dst_cset->mg_preload_node,
2730 : : &mgctx->preloaded_dst_csets);
2731 : : else
2732 : 0 : put_css_set(dst_cset);
2733 : :
2734 [ + + ]: 1164856 : for_each_subsys(ss, ssid)
2735 [ + + ]: 1164856 : if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2736 : 48504 : mgctx->ss_mask |= 1 << ssid;
2737 : : }
2738 : :
2739 : : return 0;
2740 : : }
2741 : :
2742 : : /**
2743 : : * cgroup_migrate - migrate a process or task to a cgroup
2744 : : * @leader: the leader of the process or the task to migrate
2745 : : * @threadgroup: whether @leader points to the whole process or a single task
2746 : : * @mgctx: migration context
2747 : : *
2748 : : * Migrate a process or task denoted by @leader. If migrating a process,
2749 : : * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2750 : : * responsible for invoking cgroup_migrate_add_src() and
2751 : : * cgroup_migrate_prepare_dst() on the targets before invoking this
2752 : : * function and following up with cgroup_migrate_finish().
2753 : : *
2754 : : * As long as a controller's ->can_attach() doesn't fail, this function is
2755 : : * guaranteed to succeed. This means that, excluding ->can_attach()
2756 : : * failure, when migrating multiple targets, the success or failure can be
2757 : : * decided for all targets by invoking group_migrate_prepare_dst() before
2758 : : * actually starting migrating.
2759 : : */
2760 : 207744 : int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2761 : : struct cgroup_mgctx *mgctx)
2762 : : {
2763 : : struct task_struct *task;
2764 : :
2765 : : /*
2766 : : * Prevent freeing of tasks while we take a snapshot. Tasks that are
2767 : : * already PF_EXITING could be freed from underneath us unless we
2768 : : * take an rcu_read_lock.
2769 : : */
2770 : : spin_lock_irq(&css_set_lock);
2771 : : rcu_read_lock();
2772 : : task = leader;
2773 : : do {
2774 : 212592 : cgroup_migrate_add_task(task, mgctx);
2775 [ + - ]: 212592 : if (!threadgroup)
2776 : : break;
2777 [ + + ]: 212592 : } while_each_thread(leader, task);
2778 : : rcu_read_unlock();
2779 : : spin_unlock_irq(&css_set_lock);
2780 : :
2781 : 207744 : return cgroup_migrate_execute(mgctx);
2782 : : }
2783 : :
2784 : : /**
2785 : : * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2786 : : * @dst_cgrp: the cgroup to attach to
2787 : : * @leader: the task or the leader of the threadgroup to be attached
2788 : : * @threadgroup: attach the whole threadgroup?
2789 : : *
2790 : : * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2791 : : */
2792 : 207744 : int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2793 : : bool threadgroup)
2794 : : {
2795 : 207744 : DEFINE_CGROUP_MGCTX(mgctx);
2796 : : struct task_struct *task;
2797 : : int ret;
2798 : :
2799 : 207744 : ret = cgroup_migrate_vet_dst(dst_cgrp);
2800 [ + - ]: 207744 : if (ret)
2801 : : return ret;
2802 : :
2803 : : /* look up all src csets */
2804 : : spin_lock_irq(&css_set_lock);
2805 : : rcu_read_lock();
2806 : : task = leader;
2807 : : do {
2808 : 212592 : cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2809 [ + - ]: 212592 : if (!threadgroup)
2810 : : break;
2811 [ + + ]: 212592 : } while_each_thread(leader, task);
2812 : : rcu_read_unlock();
2813 : : spin_unlock_irq(&css_set_lock);
2814 : :
2815 : : /* prepare dst csets and commit */
2816 : 207744 : ret = cgroup_migrate_prepare_dst(&mgctx);
2817 [ + - ]: 207744 : if (!ret)
2818 : 207744 : ret = cgroup_migrate(leader, threadgroup, &mgctx);
2819 : :
2820 : 207744 : cgroup_migrate_finish(&mgctx);
2821 : :
2822 [ + - ]: 207744 : if (!ret)
2823 [ - + ]: 207744 : TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2824 : :
2825 : 207744 : return ret;
2826 : : }
2827 : :
2828 : 207744 : struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup)
2829 : : __acquires(&cgroup_threadgroup_rwsem)
2830 : : {
2831 : : struct task_struct *tsk;
2832 : : pid_t pid;
2833 : :
2834 [ + - + - ]: 207744 : if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2835 : : return ERR_PTR(-EINVAL);
2836 : :
2837 : 207744 : percpu_down_write(&cgroup_threadgroup_rwsem);
2838 : :
2839 : : rcu_read_lock();
2840 [ + - ]: 207744 : if (pid) {
2841 : 207744 : tsk = find_task_by_vpid(pid);
2842 [ + - ]: 207744 : if (!tsk) {
2843 : : tsk = ERR_PTR(-ESRCH);
2844 : : goto out_unlock_threadgroup;
2845 : : }
2846 : : } else {
2847 : 0 : tsk = current;
2848 : : }
2849 : :
2850 [ + - ]: 207744 : if (threadgroup)
2851 : 207744 : tsk = tsk->group_leader;
2852 : :
2853 : : /*
2854 : : * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2855 : : * If userland migrates such a kthread to a non-root cgroup, it can
2856 : : * become trapped in a cpuset, or RT kthread may be born in a
2857 : : * cgroup with no rt_runtime allocated. Just say no.
2858 : : */
2859 [ + - + - ]: 207744 : if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2860 : : tsk = ERR_PTR(-EINVAL);
2861 : : goto out_unlock_threadgroup;
2862 : : }
2863 : :
2864 : : get_task_struct(tsk);
2865 : : goto out_unlock_rcu;
2866 : :
2867 : : out_unlock_threadgroup:
2868 : 0 : percpu_up_write(&cgroup_threadgroup_rwsem);
2869 : : out_unlock_rcu:
2870 : : rcu_read_unlock();
2871 : 207744 : return tsk;
2872 : : }
2873 : :
2874 : 207744 : void cgroup_procs_write_finish(struct task_struct *task)
2875 : : __releases(&cgroup_threadgroup_rwsem)
2876 : : {
2877 : : struct cgroup_subsys *ss;
2878 : : int ssid;
2879 : :
2880 : : /* release reference from cgroup_procs_write_start() */
2881 : 207744 : put_task_struct(task);
2882 : :
2883 : 207744 : percpu_up_write(&cgroup_threadgroup_rwsem);
2884 [ + + ]: 2492928 : for_each_subsys(ss, ssid)
2885 [ + + ]: 2285184 : if (ss->post_attach)
2886 : 415488 : ss->post_attach();
2887 : 207744 : }
2888 : :
2889 : 0 : static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2890 : : {
2891 : : struct cgroup_subsys *ss;
2892 : : bool printed = false;
2893 : : int ssid;
2894 : :
2895 [ # # ]: 0 : do_each_subsys_mask(ss, ssid, ss_mask) {
2896 [ # # ]: 0 : if (printed)
2897 : 0 : seq_putc(seq, ' ');
2898 : 0 : seq_puts(seq, ss->name);
2899 : : printed = true;
2900 : : } while_each_subsys_mask();
2901 [ # # ]: 0 : if (printed)
2902 : 0 : seq_putc(seq, '\n');
2903 : 0 : }
2904 : :
2905 : : /* show controllers which are enabled from the parent */
2906 : 0 : static int cgroup_controllers_show(struct seq_file *seq, void *v)
2907 : : {
2908 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
2909 : :
2910 : 0 : cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2911 : 0 : return 0;
2912 : : }
2913 : :
2914 : : /* show controllers which are enabled for a given cgroup's children */
2915 : 0 : static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2916 : : {
2917 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
2918 : :
2919 : 0 : cgroup_print_ss_mask(seq, cgrp->subtree_control);
2920 : 0 : return 0;
2921 : : }
2922 : :
2923 : : /**
2924 : : * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2925 : : * @cgrp: root of the subtree to update csses for
2926 : : *
2927 : : * @cgrp's control masks have changed and its subtree's css associations
2928 : : * need to be updated accordingly. This function looks up all css_sets
2929 : : * which are attached to the subtree, creates the matching updated css_sets
2930 : : * and migrates the tasks to the new ones.
2931 : : */
2932 : 8080 : static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2933 : : {
2934 : 8080 : DEFINE_CGROUP_MGCTX(mgctx);
2935 : : struct cgroup_subsys_state *d_css;
2936 : : struct cgroup *dsct;
2937 : : struct css_set *src_cset;
2938 : : int ret;
2939 : :
2940 : : lockdep_assert_held(&cgroup_mutex);
2941 : :
2942 : 8080 : percpu_down_write(&cgroup_threadgroup_rwsem);
2943 : :
2944 : : /* look up all csses currently attached to @cgrp's subtree */
2945 : : spin_lock_irq(&css_set_lock);
2946 [ + - + + ]: 24240 : cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2947 : : struct cgrp_cset_link *link;
2948 : :
2949 [ + + ]: 12120 : list_for_each_entry(link, &dsct->cset_links, cset_link)
2950 : 4040 : cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2951 : : }
2952 : : spin_unlock_irq(&css_set_lock);
2953 : :
2954 : : /* NULL dst indicates self on default hierarchy */
2955 : 8080 : ret = cgroup_migrate_prepare_dst(&mgctx);
2956 [ + - ]: 8080 : if (ret)
2957 : : goto out_finish;
2958 : :
2959 : : spin_lock_irq(&css_set_lock);
2960 [ - + ]: 8080 : list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2961 : : struct task_struct *task, *ntask;
2962 : :
2963 : : /* all tasks in src_csets need to be migrated */
2964 [ # # ]: 0 : list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2965 : 0 : cgroup_migrate_add_task(task, &mgctx);
2966 : : }
2967 : : spin_unlock_irq(&css_set_lock);
2968 : :
2969 : 8080 : ret = cgroup_migrate_execute(&mgctx);
2970 : : out_finish:
2971 : 8080 : cgroup_migrate_finish(&mgctx);
2972 : 8080 : percpu_up_write(&cgroup_threadgroup_rwsem);
2973 : 8080 : return ret;
2974 : : }
2975 : :
2976 : : /**
2977 : : * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2978 : : * @cgrp: root of the target subtree
2979 : : *
2980 : : * Because css offlining is asynchronous, userland may try to re-enable a
2981 : : * controller while the previous css is still around. This function grabs
2982 : : * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2983 : : */
2984 : 3636 : void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2985 : : __acquires(&cgroup_mutex)
2986 : : {
2987 : : struct cgroup *dsct;
2988 : : struct cgroup_subsys_state *d_css;
2989 : : struct cgroup_subsys *ss;
2990 : : int ssid;
2991 : :
2992 : : restart:
2993 : 3636 : mutex_lock(&cgroup_mutex);
2994 : :
2995 [ + - + + ]: 10908 : cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2996 [ + + ]: 83628 : for_each_subsys(ss, ssid) {
2997 : : struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2998 : 79992 : DEFINE_WAIT(wait);
2999 : :
3000 [ + - + + ]: 65166 : if (!css || !percpu_ref_is_dying(&css->refcnt))
3001 : 39996 : continue;
3002 : :
3003 : 0 : cgroup_get_live(dsct);
3004 : 0 : prepare_to_wait(&dsct->offline_waitq, &wait,
3005 : : TASK_UNINTERRUPTIBLE);
3006 : :
3007 : 0 : mutex_unlock(&cgroup_mutex);
3008 : 0 : schedule();
3009 : 0 : finish_wait(&dsct->offline_waitq, &wait);
3010 : :
3011 : : cgroup_put(dsct);
3012 : 0 : goto restart;
3013 : : }
3014 : : }
3015 : 3636 : }
3016 : :
3017 : : /**
3018 : : * cgroup_save_control - save control masks and dom_cgrp of a subtree
3019 : : * @cgrp: root of the target subtree
3020 : : *
3021 : : * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3022 : : * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3023 : : * itself.
3024 : : */
3025 : 0 : static void cgroup_save_control(struct cgroup *cgrp)
3026 : : {
3027 : : struct cgroup *dsct;
3028 : : struct cgroup_subsys_state *d_css;
3029 : :
3030 [ # # # # ]: 0 : cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3031 : 0 : dsct->old_subtree_control = dsct->subtree_control;
3032 : 0 : dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3033 : 0 : dsct->old_dom_cgrp = dsct->dom_cgrp;
3034 : : }
3035 : 0 : }
3036 : :
3037 : : /**
3038 : : * cgroup_propagate_control - refresh control masks of a subtree
3039 : : * @cgrp: root of the target subtree
3040 : : *
3041 : : * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3042 : : * ->subtree_control and propagate controller availability through the
3043 : : * subtree so that descendants don't have unavailable controllers enabled.
3044 : : */
3045 : 113148 : static void cgroup_propagate_control(struct cgroup *cgrp)
3046 : : {
3047 : : struct cgroup *dsct;
3048 : : struct cgroup_subsys_state *d_css;
3049 : :
3050 [ + - + + ]: 339444 : cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3051 : 113148 : dsct->subtree_control &= cgroup_control(dsct);
3052 : 113148 : dsct->subtree_ss_mask =
3053 : 113148 : cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3054 : : cgroup_ss_mask(dsct));
3055 : : }
3056 : 113148 : }
3057 : :
3058 : : /**
3059 : : * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3060 : : * @cgrp: root of the target subtree
3061 : : *
3062 : : * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3063 : : * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3064 : : * itself.
3065 : : */
3066 : 0 : static void cgroup_restore_control(struct cgroup *cgrp)
3067 : : {
3068 : : struct cgroup *dsct;
3069 : : struct cgroup_subsys_state *d_css;
3070 : :
3071 [ # # # # ]: 0 : cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3072 : 0 : dsct->subtree_control = dsct->old_subtree_control;
3073 : 0 : dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3074 : 0 : dsct->dom_cgrp = dsct->old_dom_cgrp;
3075 : : }
3076 : 0 : }
3077 : :
3078 : 96164 : static bool css_visible(struct cgroup_subsys_state *css)
3079 : : {
3080 : 96164 : struct cgroup_subsys *ss = css->ss;
3081 : 96164 : struct cgroup *cgrp = css->cgroup;
3082 : :
3083 [ + + ]: 96164 : if (cgroup_control(cgrp) & (1 << ss->id))
3084 : : return true;
3085 [ + + ]: 29740 : if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3086 : : return false;
3087 [ + - + + ]: 21660 : return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3088 : : }
3089 : :
3090 : : /**
3091 : : * cgroup_apply_control_enable - enable or show csses according to control
3092 : : * @cgrp: root of the target subtree
3093 : : *
3094 : : * Walk @cgrp's subtree and create new csses or make the existing ones
3095 : : * visible. A css is created invisible if it's being implicitly enabled
3096 : : * through dependency. An invisible css is made visible when the userland
3097 : : * explicitly enables it.
3098 : : *
3099 : : * Returns 0 on success, -errno on failure. On failure, csses which have
3100 : : * been processed already aren't cleaned up. The caller is responsible for
3101 : : * cleaning up with cgroup_apply_control_disable().
3102 : : */
3103 : 113148 : static int cgroup_apply_control_enable(struct cgroup *cgrp)
3104 : : {
3105 : : struct cgroup *dsct;
3106 : : struct cgroup_subsys_state *d_css;
3107 : : struct cgroup_subsys *ss;
3108 : : int ssid, ret;
3109 : :
3110 [ + - + + ]: 339444 : cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3111 [ + + ]: 1244628 : for_each_subsys(ss, ssid) {
3112 : : struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3113 : :
3114 [ + + ]: 1244628 : if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3115 : 1174724 : continue;
3116 : :
3117 [ + + ]: 69904 : if (!css) {
3118 : 46876 : css = css_create(dsct, ss);
3119 [ - + ]: 46876 : if (IS_ERR(css))
3120 : 0 : return PTR_ERR(css);
3121 : : }
3122 : :
3123 [ - + # # ]: 69904 : WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3124 : :
3125 [ + + ]: 69904 : if (css_visible(css)) {
3126 : 61094 : ret = css_populate_dir(css);
3127 [ - + ]: 61094 : if (ret)
3128 : 0 : return ret;
3129 : : }
3130 : : }
3131 : : }
3132 : :
3133 : : return 0;
3134 : : }
3135 : :
3136 : : /**
3137 : : * cgroup_apply_control_disable - kill or hide csses according to control
3138 : : * @cgrp: root of the target subtree
3139 : : *
3140 : : * Walk @cgrp's subtree and kill and hide csses so that they match
3141 : : * cgroup_ss_mask() and cgroup_visible_mask().
3142 : : *
3143 : : * A css is hidden when the userland requests it to be disabled while other
3144 : : * subsystems are still depending on it. The css must not actively control
3145 : : * resources and be in the vanilla state if it's made visible again later.
3146 : : * Controllers which may be depended upon should provide ->css_reset() for
3147 : : * this purpose.
3148 : : */
3149 : 4040 : static void cgroup_apply_control_disable(struct cgroup *cgrp)
3150 : : {
3151 : : struct cgroup *dsct;
3152 : : struct cgroup_subsys_state *d_css;
3153 : : struct cgroup_subsys *ss;
3154 : : int ssid;
3155 : :
3156 [ + - + + ]: 12120 : cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3157 [ + + ]: 44440 : for_each_subsys(ss, ssid) {
3158 : : struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3159 : :
3160 [ + + ]: 44440 : if (!css)
3161 : 18180 : continue;
3162 : :
3163 [ - + # # ]: 26260 : WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3164 : :
3165 [ - + # # ]: 26260 : if (css->parent &&
3166 : 0 : !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3167 : 0 : kill_css(css);
3168 [ + + ]: 26260 : } else if (!css_visible(css)) {
3169 : 16890 : css_clear_dir(css);
3170 [ + + ]: 16890 : if (ss->css_reset)
3171 : 4040 : ss->css_reset(css);
3172 : : }
3173 : : }
3174 : : }
3175 : 4040 : }
3176 : :
3177 : : /**
3178 : : * cgroup_apply_control - apply control mask updates to the subtree
3179 : : * @cgrp: root of the target subtree
3180 : : *
3181 : : * subsystems can be enabled and disabled in a subtree using the following
3182 : : * steps.
3183 : : *
3184 : : * 1. Call cgroup_save_control() to stash the current state.
3185 : : * 2. Update ->subtree_control masks in the subtree as desired.
3186 : : * 3. Call cgroup_apply_control() to apply the changes.
3187 : : * 4. Optionally perform other related operations.
3188 : : * 5. Call cgroup_finalize_control() to finish up.
3189 : : *
3190 : : * This function implements step 3 and propagates the mask changes
3191 : : * throughout @cgrp's subtree, updates csses accordingly and perform
3192 : : * process migrations.
3193 : : */
3194 : 8080 : static int cgroup_apply_control(struct cgroup *cgrp)
3195 : : {
3196 : : int ret;
3197 : :
3198 : 8080 : cgroup_propagate_control(cgrp);
3199 : :
3200 : 8080 : ret = cgroup_apply_control_enable(cgrp);
3201 [ + - ]: 8080 : if (ret)
3202 : : return ret;
3203 : :
3204 : : /*
3205 : : * At this point, cgroup_e_css_by_mask() results reflect the new csses
3206 : : * making the following cgroup_update_dfl_csses() properly update
3207 : : * css associations of all tasks in the subtree.
3208 : : */
3209 : 8080 : ret = cgroup_update_dfl_csses(cgrp);
3210 [ - + ]: 8080 : if (ret)
3211 : 0 : return ret;
3212 : :
3213 : : return 0;
3214 : : }
3215 : :
3216 : : /**
3217 : : * cgroup_finalize_control - finalize control mask update
3218 : : * @cgrp: root of the target subtree
3219 : : * @ret: the result of the update
3220 : : *
3221 : : * Finalize control mask update. See cgroup_apply_control() for more info.
3222 : : */
3223 : 4040 : static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3224 : : {
3225 [ - + ]: 4040 : if (ret) {
3226 : 0 : cgroup_restore_control(cgrp);
3227 : 0 : cgroup_propagate_control(cgrp);
3228 : : }
3229 : :
3230 : 4040 : cgroup_apply_control_disable(cgrp);
3231 : 4040 : }
3232 : :
3233 : 0 : static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3234 : : {
3235 : 0 : u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3236 : :
3237 : : /* if nothing is getting enabled, nothing to worry about */
3238 [ # # ]: 0 : if (!enable)
3239 : : return 0;
3240 : :
3241 : : /* can @cgrp host any resources? */
3242 [ # # ]: 0 : if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3243 : : return -EOPNOTSUPP;
3244 : :
3245 : : /* mixables don't care */
3246 [ # # ]: 0 : if (cgroup_is_mixable(cgrp))
3247 : : return 0;
3248 : :
3249 [ # # ]: 0 : if (domain_enable) {
3250 : : /* can't enable domain controllers inside a thread subtree */
3251 [ # # # # ]: 0 : if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3252 : : return -EOPNOTSUPP;
3253 : : } else {
3254 : : /*
3255 : : * Threaded controllers can handle internal competitions
3256 : : * and are always allowed inside a (prospective) thread
3257 : : * subtree.
3258 : : */
3259 [ # # # # ]: 0 : if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3260 : : return 0;
3261 : : }
3262 : :
3263 : : /*
3264 : : * Controllers can't be enabled for a cgroup with tasks to avoid
3265 : : * child cgroups competing against tasks.
3266 : : */
3267 [ # # ]: 0 : if (cgroup_has_tasks(cgrp))
3268 : : return -EBUSY;
3269 : :
3270 : 0 : return 0;
3271 : : }
3272 : :
3273 : : /* change the enabled child controllers for a cgroup in the default hierarchy */
3274 : 0 : static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3275 : : char *buf, size_t nbytes,
3276 : : loff_t off)
3277 : : {
3278 : : u16 enable = 0, disable = 0;
3279 : : struct cgroup *cgrp, *child;
3280 : : struct cgroup_subsys *ss;
3281 : : char *tok;
3282 : : int ssid, ret;
3283 : :
3284 : : /*
3285 : : * Parse input - space separated list of subsystem names prefixed
3286 : : * with either + or -.
3287 : : */
3288 : 0 : buf = strstrip(buf);
3289 [ # # ]: 0 : while ((tok = strsep(&buf, " "))) {
3290 [ # # ]: 0 : if (tok[0] == '\0')
3291 : 0 : continue;
3292 [ # # ]: 0 : do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3293 [ # # # # ]: 0 : if (!cgroup_ssid_enabled(ssid) ||
3294 : 0 : strcmp(tok + 1, ss->name))
3295 : 0 : continue;
3296 : :
3297 [ # # ]: 0 : if (*tok == '+') {
3298 : 0 : enable |= 1 << ssid;
3299 : 0 : disable &= ~(1 << ssid);
3300 [ # # ]: 0 : } else if (*tok == '-') {
3301 : 0 : disable |= 1 << ssid;
3302 : 0 : enable &= ~(1 << ssid);
3303 : : } else {
3304 : 0 : return -EINVAL;
3305 : : }
3306 : : break;
3307 : : } while_each_subsys_mask();
3308 [ # # ]: 0 : if (ssid == CGROUP_SUBSYS_COUNT)
3309 : : return -EINVAL;
3310 : : }
3311 : :
3312 : 0 : cgrp = cgroup_kn_lock_live(of->kn, true);
3313 [ # # ]: 0 : if (!cgrp)
3314 : : return -ENODEV;
3315 : :
3316 [ # # ]: 0 : for_each_subsys(ss, ssid) {
3317 [ # # ]: 0 : if (enable & (1 << ssid)) {
3318 [ # # ]: 0 : if (cgrp->subtree_control & (1 << ssid)) {
3319 : 0 : enable &= ~(1 << ssid);
3320 : 0 : continue;
3321 : : }
3322 : :
3323 [ # # ]: 0 : if (!(cgroup_control(cgrp) & (1 << ssid))) {
3324 : : ret = -ENOENT;
3325 : : goto out_unlock;
3326 : : }
3327 [ # # ]: 0 : } else if (disable & (1 << ssid)) {
3328 [ # # ]: 0 : if (!(cgrp->subtree_control & (1 << ssid))) {
3329 : 0 : disable &= ~(1 << ssid);
3330 : 0 : continue;
3331 : : }
3332 : :
3333 : : /* a child has it enabled? */
3334 [ # # # # ]: 0 : cgroup_for_each_live_child(child, cgrp) {
3335 [ # # ]: 0 : if (child->subtree_control & (1 << ssid)) {
3336 : : ret = -EBUSY;
3337 : : goto out_unlock;
3338 : : }
3339 : : }
3340 : : }
3341 : : }
3342 : :
3343 [ # # ]: 0 : if (!enable && !disable) {
3344 : : ret = 0;
3345 : : goto out_unlock;
3346 : : }
3347 : :
3348 : 0 : ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3349 [ # # ]: 0 : if (ret)
3350 : : goto out_unlock;
3351 : :
3352 : : /* save and update control masks and prepare csses */
3353 : 0 : cgroup_save_control(cgrp);
3354 : :
3355 : 0 : cgrp->subtree_control |= enable;
3356 : 0 : cgrp->subtree_control &= ~disable;
3357 : :
3358 : 0 : ret = cgroup_apply_control(cgrp);
3359 : 0 : cgroup_finalize_control(cgrp, ret);
3360 [ # # ]: 0 : if (ret)
3361 : : goto out_unlock;
3362 : :
3363 : 0 : kernfs_activate(cgrp->kn);
3364 : : out_unlock:
3365 : 0 : cgroup_kn_unlock(of->kn);
3366 [ # # ]: 0 : return ret ?: nbytes;
3367 : : }
3368 : :
3369 : : /**
3370 : : * cgroup_enable_threaded - make @cgrp threaded
3371 : : * @cgrp: the target cgroup
3372 : : *
3373 : : * Called when "threaded" is written to the cgroup.type interface file and
3374 : : * tries to make @cgrp threaded and join the parent's resource domain.
3375 : : * This function is never called on the root cgroup as cgroup.type doesn't
3376 : : * exist on it.
3377 : : */
3378 : 0 : static int cgroup_enable_threaded(struct cgroup *cgrp)
3379 : : {
3380 : : struct cgroup *parent = cgroup_parent(cgrp);
3381 : 0 : struct cgroup *dom_cgrp = parent->dom_cgrp;
3382 : : struct cgroup *dsct;
3383 : : struct cgroup_subsys_state *d_css;
3384 : : int ret;
3385 : :
3386 : : lockdep_assert_held(&cgroup_mutex);
3387 : :
3388 : : /* noop if already threaded */
3389 [ # # ]: 0 : if (cgroup_is_threaded(cgrp))
3390 : : return 0;
3391 : :
3392 : : /*
3393 : : * If @cgroup is populated or has domain controllers enabled, it
3394 : : * can't be switched. While the below cgroup_can_be_thread_root()
3395 : : * test can catch the same conditions, that's only when @parent is
3396 : : * not mixable, so let's check it explicitly.
3397 : : */
3398 [ # # # # ]: 0 : if (cgroup_is_populated(cgrp) ||
3399 : 0 : cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3400 : : return -EOPNOTSUPP;
3401 : :
3402 : : /* we're joining the parent's domain, ensure its validity */
3403 [ # # # # ]: 0 : if (!cgroup_is_valid_domain(dom_cgrp) ||
3404 : 0 : !cgroup_can_be_thread_root(dom_cgrp))
3405 : : return -EOPNOTSUPP;
3406 : :
3407 : : /*
3408 : : * The following shouldn't cause actual migrations and should
3409 : : * always succeed.
3410 : : */
3411 : 0 : cgroup_save_control(cgrp);
3412 : :
3413 [ # # # # ]: 0 : cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3414 [ # # # # ]: 0 : if (dsct == cgrp || cgroup_is_threaded(dsct))
3415 : 0 : dsct->dom_cgrp = dom_cgrp;
3416 : :
3417 : 0 : ret = cgroup_apply_control(cgrp);
3418 [ # # ]: 0 : if (!ret)
3419 : 0 : parent->nr_threaded_children++;
3420 : :
3421 : 0 : cgroup_finalize_control(cgrp, ret);
3422 : 0 : return ret;
3423 : : }
3424 : :
3425 : 0 : static int cgroup_type_show(struct seq_file *seq, void *v)
3426 : : {
3427 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
3428 : :
3429 [ # # ]: 0 : if (cgroup_is_threaded(cgrp))
3430 : 0 : seq_puts(seq, "threaded\n");
3431 [ # # ]: 0 : else if (!cgroup_is_valid_domain(cgrp))
3432 : 0 : seq_puts(seq, "domain invalid\n");
3433 [ # # ]: 0 : else if (cgroup_is_thread_root(cgrp))
3434 : 0 : seq_puts(seq, "domain threaded\n");
3435 : : else
3436 : 0 : seq_puts(seq, "domain\n");
3437 : :
3438 : 0 : return 0;
3439 : : }
3440 : :
3441 : 0 : static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3442 : : size_t nbytes, loff_t off)
3443 : : {
3444 : : struct cgroup *cgrp;
3445 : : int ret;
3446 : :
3447 : : /* only switching to threaded mode is supported */
3448 [ # # ]: 0 : if (strcmp(strstrip(buf), "threaded"))
3449 : : return -EINVAL;
3450 : :
3451 : : /* drain dying csses before we re-apply (threaded) subtree control */
3452 : 0 : cgrp = cgroup_kn_lock_live(of->kn, true);
3453 [ # # ]: 0 : if (!cgrp)
3454 : : return -ENOENT;
3455 : :
3456 : : /* threaded can only be enabled */
3457 : 0 : ret = cgroup_enable_threaded(cgrp);
3458 : :
3459 : 0 : cgroup_kn_unlock(of->kn);
3460 [ # # ]: 0 : return ret ?: nbytes;
3461 : : }
3462 : :
3463 : 0 : static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3464 : : {
3465 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
3466 : 0 : int descendants = READ_ONCE(cgrp->max_descendants);
3467 : :
3468 [ # # ]: 0 : if (descendants == INT_MAX)
3469 : 0 : seq_puts(seq, "max\n");
3470 : : else
3471 : 0 : seq_printf(seq, "%d\n", descendants);
3472 : :
3473 : 0 : return 0;
3474 : : }
3475 : :
3476 : 0 : static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3477 : : char *buf, size_t nbytes, loff_t off)
3478 : : {
3479 : : struct cgroup *cgrp;
3480 : : int descendants;
3481 : : ssize_t ret;
3482 : :
3483 : : buf = strstrip(buf);
3484 [ # # ]: 0 : if (!strcmp(buf, "max")) {
3485 : 0 : descendants = INT_MAX;
3486 : : } else {
3487 : 0 : ret = kstrtoint(buf, 0, &descendants);
3488 [ # # ]: 0 : if (ret)
3489 : : return ret;
3490 : : }
3491 : :
3492 [ # # ]: 0 : if (descendants < 0)
3493 : : return -ERANGE;
3494 : :
3495 : 0 : cgrp = cgroup_kn_lock_live(of->kn, false);
3496 [ # # ]: 0 : if (!cgrp)
3497 : : return -ENOENT;
3498 : :
3499 : 0 : cgrp->max_descendants = descendants;
3500 : :
3501 : 0 : cgroup_kn_unlock(of->kn);
3502 : :
3503 : 0 : return nbytes;
3504 : : }
3505 : :
3506 : 0 : static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3507 : : {
3508 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
3509 : 0 : int depth = READ_ONCE(cgrp->max_depth);
3510 : :
3511 [ # # ]: 0 : if (depth == INT_MAX)
3512 : 0 : seq_puts(seq, "max\n");
3513 : : else
3514 : 0 : seq_printf(seq, "%d\n", depth);
3515 : :
3516 : 0 : return 0;
3517 : : }
3518 : :
3519 : 0 : static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3520 : : char *buf, size_t nbytes, loff_t off)
3521 : : {
3522 : : struct cgroup *cgrp;
3523 : : ssize_t ret;
3524 : : int depth;
3525 : :
3526 : : buf = strstrip(buf);
3527 [ # # ]: 0 : if (!strcmp(buf, "max")) {
3528 : 0 : depth = INT_MAX;
3529 : : } else {
3530 : 0 : ret = kstrtoint(buf, 0, &depth);
3531 [ # # ]: 0 : if (ret)
3532 : : return ret;
3533 : : }
3534 : :
3535 [ # # ]: 0 : if (depth < 0)
3536 : : return -ERANGE;
3537 : :
3538 : 0 : cgrp = cgroup_kn_lock_live(of->kn, false);
3539 [ # # ]: 0 : if (!cgrp)
3540 : : return -ENOENT;
3541 : :
3542 : 0 : cgrp->max_depth = depth;
3543 : :
3544 : 0 : cgroup_kn_unlock(of->kn);
3545 : :
3546 : 0 : return nbytes;
3547 : : }
3548 : :
3549 : 33674 : static int cgroup_events_show(struct seq_file *seq, void *v)
3550 : : {
3551 : 33674 : struct cgroup *cgrp = seq_css(seq)->cgroup;
3552 : :
3553 : 33674 : seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3554 : 33674 : seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3555 : :
3556 : 33674 : return 0;
3557 : : }
3558 : :
3559 : 0 : static int cgroup_stat_show(struct seq_file *seq, void *v)
3560 : : {
3561 : 0 : struct cgroup *cgroup = seq_css(seq)->cgroup;
3562 : :
3563 : 0 : seq_printf(seq, "nr_descendants %d\n",
3564 : : cgroup->nr_descendants);
3565 : 0 : seq_printf(seq, "nr_dying_descendants %d\n",
3566 : : cgroup->nr_dying_descendants);
3567 : :
3568 : 0 : return 0;
3569 : : }
3570 : :
3571 : 0 : static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3572 : : struct cgroup *cgrp, int ssid)
3573 : : {
3574 : 0 : struct cgroup_subsys *ss = cgroup_subsys[ssid];
3575 : : struct cgroup_subsys_state *css;
3576 : : int ret;
3577 : :
3578 [ # # ]: 0 : if (!ss->css_extra_stat_show)
3579 : : return 0;
3580 : :
3581 : 0 : css = cgroup_tryget_css(cgrp, ss);
3582 [ # # ]: 0 : if (!css)
3583 : : return 0;
3584 : :
3585 : 0 : ret = ss->css_extra_stat_show(seq, css);
3586 : : css_put(css);
3587 : 0 : return ret;
3588 : : }
3589 : :
3590 : 0 : static int cpu_stat_show(struct seq_file *seq, void *v)
3591 : : {
3592 : 0 : struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3593 : : int ret = 0;
3594 : :
3595 : 0 : cgroup_base_stat_cputime_show(seq);
3596 : : #ifdef CONFIG_CGROUP_SCHED
3597 : 0 : ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3598 : : #endif
3599 : 0 : return ret;
3600 : : }
3601 : :
3602 : : #ifdef CONFIG_PSI
3603 : : static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3604 : : {
3605 : : struct cgroup *cgroup = seq_css(seq)->cgroup;
3606 : : struct psi_group *psi = cgroup->id == 1 ? &psi_system : &cgroup->psi;
3607 : :
3608 : : return psi_show(seq, psi, PSI_IO);
3609 : : }
3610 : : static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3611 : : {
3612 : : struct cgroup *cgroup = seq_css(seq)->cgroup;
3613 : : struct psi_group *psi = cgroup->id == 1 ? &psi_system : &cgroup->psi;
3614 : :
3615 : : return psi_show(seq, psi, PSI_MEM);
3616 : : }
3617 : : static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3618 : : {
3619 : : struct cgroup *cgroup = seq_css(seq)->cgroup;
3620 : : struct psi_group *psi = cgroup->id == 1 ? &psi_system : &cgroup->psi;
3621 : :
3622 : : return psi_show(seq, psi, PSI_CPU);
3623 : : }
3624 : :
3625 : : static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3626 : : size_t nbytes, enum psi_res res)
3627 : : {
3628 : : struct psi_trigger *new;
3629 : : struct cgroup *cgrp;
3630 : :
3631 : : cgrp = cgroup_kn_lock_live(of->kn, false);
3632 : : if (!cgrp)
3633 : : return -ENODEV;
3634 : :
3635 : : cgroup_get(cgrp);
3636 : : cgroup_kn_unlock(of->kn);
3637 : :
3638 : : new = psi_trigger_create(&cgrp->psi, buf, nbytes, res);
3639 : : if (IS_ERR(new)) {
3640 : : cgroup_put(cgrp);
3641 : : return PTR_ERR(new);
3642 : : }
3643 : :
3644 : : psi_trigger_replace(&of->priv, new);
3645 : :
3646 : : cgroup_put(cgrp);
3647 : :
3648 : : return nbytes;
3649 : : }
3650 : :
3651 : : static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3652 : : char *buf, size_t nbytes,
3653 : : loff_t off)
3654 : : {
3655 : : return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3656 : : }
3657 : :
3658 : : static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3659 : : char *buf, size_t nbytes,
3660 : : loff_t off)
3661 : : {
3662 : : return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3663 : : }
3664 : :
3665 : : static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3666 : : char *buf, size_t nbytes,
3667 : : loff_t off)
3668 : : {
3669 : : return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3670 : : }
3671 : :
3672 : : static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3673 : : poll_table *pt)
3674 : : {
3675 : : return psi_trigger_poll(&of->priv, of->file, pt);
3676 : : }
3677 : :
3678 : : static void cgroup_pressure_release(struct kernfs_open_file *of)
3679 : : {
3680 : : psi_trigger_replace(&of->priv, NULL);
3681 : : }
3682 : : #endif /* CONFIG_PSI */
3683 : :
3684 : 0 : static int cgroup_freeze_show(struct seq_file *seq, void *v)
3685 : : {
3686 : 0 : struct cgroup *cgrp = seq_css(seq)->cgroup;
3687 : :
3688 : 0 : seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3689 : :
3690 : 0 : return 0;
3691 : : }
3692 : :
3693 : 0 : static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3694 : : char *buf, size_t nbytes, loff_t off)
3695 : : {
3696 : : struct cgroup *cgrp;
3697 : : ssize_t ret;
3698 : : int freeze;
3699 : :
3700 : 0 : ret = kstrtoint(strstrip(buf), 0, &freeze);
3701 [ # # ]: 0 : if (ret)
3702 : : return ret;
3703 : :
3704 [ # # ]: 0 : if (freeze < 0 || freeze > 1)
3705 : : return -ERANGE;
3706 : :
3707 : 0 : cgrp = cgroup_kn_lock_live(of->kn, false);
3708 [ # # ]: 0 : if (!cgrp)
3709 : : return -ENOENT;
3710 : :
3711 : 0 : cgroup_freeze(cgrp, freeze);
3712 : :
3713 : 0 : cgroup_kn_unlock(of->kn);
3714 : :
3715 : 0 : return nbytes;
3716 : : }
3717 : :
3718 : 408044 : static int cgroup_file_open(struct kernfs_open_file *of)
3719 : : {
3720 : 408044 : struct cftype *cft = of->kn->priv;
3721 : :
3722 [ - + ]: 408044 : if (cft->open)
3723 : 0 : return cft->open(of);
3724 : : return 0;
3725 : : }
3726 : :
3727 : 408044 : static void cgroup_file_release(struct kernfs_open_file *of)
3728 : : {
3729 : 408044 : struct cftype *cft = of->kn->priv;
3730 : :
3731 [ + + ]: 408044 : if (cft->release)
3732 : 169462 : cft->release(of);
3733 : 408044 : }
3734 : :
3735 : 259866 : static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3736 : : size_t nbytes, loff_t off)
3737 : : {
3738 : 259866 : struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3739 : 259866 : struct cgroup *cgrp = of->kn->parent->priv;
3740 : 259866 : struct cftype *cft = of->kn->priv;
3741 : : struct cgroup_subsys_state *css;
3742 : : int ret;
3743 : :
3744 : : /*
3745 : : * If namespaces are delegation boundaries, disallow writes to
3746 : : * files in an non-init namespace root from inside the namespace
3747 : : * except for the files explicitly marked delegatable -
3748 : : * cgroup.procs and cgroup.subtree_control.
3749 : : */
3750 [ + + - + ]: 315234 : if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3751 [ # # ]: 55368 : !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3752 [ # # ]: 0 : ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3753 : : return -EPERM;
3754 : :
3755 [ + - ]: 259868 : if (cft->write)
3756 : 259868 : return cft->write(of, buf, nbytes, off);
3757 : :
3758 : : /*
3759 : : * kernfs guarantees that a file isn't deleted with operations in
3760 : : * flight, which means that the matching css is and stays alive and
3761 : : * doesn't need to be pinned. The RCU locking is not necessary
3762 : : * either. It's just for the convenience of using cgroup_css().
3763 : : */
3764 : : rcu_read_lock();
3765 : 0 : css = cgroup_css(cgrp, cft->ss);
3766 : : rcu_read_unlock();
3767 : :
3768 [ # # ]: 0 : if (cft->write_u64) {
3769 : : unsigned long long v;
3770 : 0 : ret = kstrtoull(buf, 0, &v);
3771 [ # # ]: 0 : if (!ret)
3772 : 0 : ret = cft->write_u64(css, cft, v);
3773 [ # # ]: 0 : } else if (cft->write_s64) {
3774 : : long long v;
3775 : 0 : ret = kstrtoll(buf, 0, &v);
3776 [ # # ]: 0 : if (!ret)
3777 : 0 : ret = cft->write_s64(css, cft, v);
3778 : : } else {
3779 : : ret = -EINVAL;
3780 : : }
3781 : :
3782 [ # # ]: 0 : return ret ?: nbytes;
3783 : : }
3784 : :
3785 : 0 : static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3786 : : {
3787 : 0 : struct cftype *cft = of->kn->priv;
3788 : :
3789 [ # # ]: 0 : if (cft->poll)
3790 : 0 : return cft->poll(of, pt);
3791 : :
3792 : 0 : return kernfs_generic_poll(of, pt);
3793 : : }
3794 : :
3795 : 117440 : static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3796 : : {
3797 : 117440 : return seq_cft(seq)->seq_start(seq, ppos);
3798 : : }
3799 : :
3800 : 28534 : static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3801 : : {
3802 : 28534 : return seq_cft(seq)->seq_next(seq, v, ppos);
3803 : : }
3804 : :
3805 : 117440 : static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3806 : : {
3807 [ - + ]: 117440 : if (seq_cft(seq)->seq_stop)
3808 : 0 : seq_cft(seq)->seq_stop(seq, v);
3809 : 117440 : }
3810 : :
3811 : 62612 : static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3812 : : {
3813 : : struct cftype *cft = seq_cft(m);
3814 : : struct cgroup_subsys_state *css = seq_css(m);
3815 : :
3816 [ + + ]: 62612 : if (cft->seq_show)
3817 : 62208 : return cft->seq_show(m, arg);
3818 : :
3819 [ - + ]: 404 : if (cft->read_u64)
3820 : 0 : seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3821 [ + - ]: 404 : else if (cft->read_s64)
3822 : 404 : seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3823 : : else
3824 : : return -EINVAL;
3825 : : return 0;
3826 : : }
3827 : :
3828 : : static struct kernfs_ops cgroup_kf_single_ops = {
3829 : : .atomic_write_len = PAGE_SIZE,
3830 : : .open = cgroup_file_open,
3831 : : .release = cgroup_file_release,
3832 : : .write = cgroup_file_write,
3833 : : .poll = cgroup_file_poll,
3834 : : .seq_show = cgroup_seqfile_show,
3835 : : };
3836 : :
3837 : : static struct kernfs_ops cgroup_kf_ops = {
3838 : : .atomic_write_len = PAGE_SIZE,
3839 : : .open = cgroup_file_open,
3840 : : .release = cgroup_file_release,
3841 : : .write = cgroup_file_write,
3842 : : .poll = cgroup_file_poll,
3843 : : .seq_start = cgroup_seqfile_start,
3844 : : .seq_next = cgroup_seqfile_next,
3845 : : .seq_stop = cgroup_seqfile_stop,
3846 : : .seq_show = cgroup_seqfile_show,
3847 : : };
3848 : :
3849 : : /* set uid and gid of cgroup dirs and files to that of the creator */
3850 : 908424 : static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3851 : : {
3852 : 1816848 : struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3853 : 908424 : .ia_uid = current_fsuid(),
3854 : : .ia_gid = current_fsgid(), };
3855 : :
3856 [ + + - + ]: 908424 : if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3857 : : gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3858 : : return 0;
3859 : :
3860 : 61812 : return kernfs_setattr(kn, &iattr);
3861 : : }
3862 : :
3863 : 268 : static void cgroup_file_notify_timer(struct timer_list *timer)
3864 : : {
3865 : 268 : cgroup_file_notify(container_of(timer, struct cgroup_file,
3866 : : notify_timer));
3867 : 268 : }
3868 : :
3869 : 803356 : static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3870 : : struct cftype *cft)
3871 : : {
3872 : : char name[CGROUP_FILE_NAME_MAX];
3873 : : struct kernfs_node *kn;
3874 : : struct lock_class_key *key = NULL;
3875 : : int ret;
3876 : :
3877 : : #ifdef CONFIG_DEBUG_LOCK_ALLOC
3878 : : key = &cft->lockdep_key;
3879 : : #endif
3880 : 1606712 : kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3881 : : cgroup_file_mode(cft),
3882 : 803356 : GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3883 : 803356 : 0, cft->kf_ops, cft,
3884 : : NULL, key);
3885 [ - + ]: 803356 : if (IS_ERR(kn))
3886 : 0 : return PTR_ERR(kn);
3887 : :
3888 : 803356 : ret = cgroup_kn_set_ugid(kn);
3889 [ - + ]: 803356 : if (ret) {
3890 : 0 : kernfs_remove(kn);
3891 : 0 : return ret;
3892 : : }
3893 : :
3894 [ + + ]: 803356 : if (cft->file_offset) {
3895 : 83652 : struct cgroup_file *cfile = (void *)css + cft->file_offset;
3896 : :
3897 : 83652 : timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3898 : :
3899 : : spin_lock_irq(&cgroup_file_kn_lock);
3900 : 83652 : cfile->kn = kn;
3901 : : spin_unlock_irq(&cgroup_file_kn_lock);
3902 : : }
3903 : :
3904 : : return 0;
3905 : : }
3906 : :
3907 : : /**
3908 : : * cgroup_addrm_files - add or remove files to a cgroup directory
3909 : : * @css: the target css
3910 : : * @cgrp: the target cgroup (usually css->cgroup)
3911 : : * @cfts: array of cftypes to be added
3912 : : * @is_add: whether to add or remove
3913 : : *
3914 : : * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3915 : : * For removals, this function never fails.
3916 : : */
3917 : 241656 : static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3918 : : struct cgroup *cgrp, struct cftype cfts[],
3919 : : bool is_add)
3920 : : {
3921 : : struct cftype *cft, *cft_end = NULL;
3922 : : int ret = 0;
3923 : :
3924 : : lockdep_assert_held(&cgroup_mutex);
3925 : :
3926 : : restart:
3927 [ + - + + ]: 1660888 : for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3928 : : /* does cft->flags tell us to skip this file on @cgrp? */
3929 [ + + + + ]: 1431352 : if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3930 : 4040 : continue;
3931 [ + + + + ]: 1558232 : if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3932 : 29088 : continue;
3933 [ + + + + ]: 1674644 : if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3934 : 12120 : continue;
3935 [ + + + + ]: 1604732 : if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3936 : 223072 : continue;
3937 [ + + + - ]: 1150912 : if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
3938 : 808 : continue;
3939 [ + + ]: 1150104 : if (is_add) {
3940 : 803356 : ret = cgroup_add_file(css, cgrp, cft);
3941 [ - + ]: 803356 : if (ret) {
3942 : 0 : pr_warn("%s: failed to add %s, err=%d\n",
3943 : : __func__, cft->name, ret);
3944 : 0 : cft_end = cft;
3945 : : is_add = false;
3946 : 0 : goto restart;
3947 : : }
3948 : : } else {
3949 : 346748 : cgroup_rm_file(cgrp, cft);
3950 : : }
3951 : : }
3952 : 241656 : return ret;
3953 : : }
3954 : :
3955 : 4848 : static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3956 : : {
3957 : 4848 : struct cgroup_subsys *ss = cfts[0].ss;
3958 : 4848 : struct cgroup *root = &ss->root->cgrp;
3959 : : struct cgroup_subsys_state *css;
3960 : : int ret = 0;
3961 : :
3962 : : lockdep_assert_held(&cgroup_mutex);
3963 : :
3964 : : /* add/rm files for all cgroups created before */
3965 [ + + ]: 9696 : css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3966 : 4848 : struct cgroup *cgrp = css->cgroup;
3967 : :
3968 [ + - ]: 4848 : if (!(css->flags & CSS_VISIBLE))
3969 : 4848 : continue;
3970 : :
3971 : 0 : ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3972 [ # # ]: 0 : if (ret)
3973 : : break;
3974 : : }
3975 : :
3976 [ + - ]: 4848 : if (is_add && !ret)
3977 : 4848 : kernfs_activate(root->kn);
3978 : 4848 : return ret;
3979 : : }
3980 : :
3981 : 0 : static void cgroup_exit_cftypes(struct cftype *cfts)
3982 : : {
3983 : : struct cftype *cft;
3984 : :
3985 [ # # ]: 0 : for (cft = cfts; cft->name[0] != '\0'; cft++) {
3986 : : /* free copy for custom atomic_write_len, see init_cftypes() */
3987 [ # # ]: 0 : if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3988 : 0 : kfree(cft->kf_ops);
3989 : 0 : cft->kf_ops = NULL;
3990 : 0 : cft->ss = NULL;
3991 : :
3992 : : /* revert flags set by cgroup core while adding @cfts */
3993 : 0 : cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3994 : : }
3995 : 0 : }
3996 : :
3997 : 5656 : static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3998 : : {
3999 : : struct cftype *cft;
4000 : :
4001 [ + + ]: 32320 : for (cft = cfts; cft->name[0] != '\0'; cft++) {
4002 : : struct kernfs_ops *kf_ops;
4003 : :
4004 [ + - + - : 26664 : WARN_ON(cft->ss || cft->kf_ops);
- + ]
4005 : :
4006 [ + + ]: 26664 : if (cft->seq_start)
4007 : : kf_ops = &cgroup_kf_ops;
4008 : : else
4009 : : kf_ops = &cgroup_kf_single_ops;
4010 : :
4011 : : /*
4012 : : * Ugh... if @cft wants a custom max_write_len, we need to
4013 : : * make a copy of kf_ops to set its atomic_write_len.
4014 : : */
4015 [ + + ]: 26664 : if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4016 : 2020 : kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4017 [ - + ]: 2020 : if (!kf_ops) {
4018 : 0 : cgroup_exit_cftypes(cfts);
4019 : 0 : return -ENOMEM;
4020 : : }
4021 : 2020 : kf_ops->atomic_write_len = cft->max_write_len;
4022 : : }
4023 : :
4024 : 26664 : cft->kf_ops = kf_ops;
4025 : 26664 : cft->ss = ss;
4026 : : }
4027 : :
4028 : : return 0;
4029 : : }
4030 : :
4031 : 0 : static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4032 : : {
4033 : : lockdep_assert_held(&cgroup_mutex);
4034 : :
4035 [ # # # # ]: 0 : if (!cfts || !cfts[0].ss)
4036 : : return -ENOENT;
4037 : :
4038 : : list_del(&cfts->node);
4039 : 0 : cgroup_apply_cftypes(cfts, false);
4040 : 0 : cgroup_exit_cftypes(cfts);
4041 : 0 : return 0;
4042 : : }
4043 : :
4044 : : /**
4045 : : * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4046 : : * @cfts: zero-length name terminated array of cftypes
4047 : : *
4048 : : * Unregister @cfts. Files described by @cfts are removed from all
4049 : : * existing cgroups and all future cgroups won't have them either. This
4050 : : * function can be called anytime whether @cfts' subsys is attached or not.
4051 : : *
4052 : : * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4053 : : * registered.
4054 : : */
4055 : 0 : int cgroup_rm_cftypes(struct cftype *cfts)
4056 : : {
4057 : : int ret;
4058 : :
4059 : 0 : mutex_lock(&cgroup_mutex);
4060 : 0 : ret = cgroup_rm_cftypes_locked(cfts);
4061 : 0 : mutex_unlock(&cgroup_mutex);
4062 : 0 : return ret;
4063 : : }
4064 : :
4065 : : /**
4066 : : * cgroup_add_cftypes - add an array of cftypes to a subsystem
4067 : : * @ss: target cgroup subsystem
4068 : : * @cfts: zero-length name terminated array of cftypes
4069 : : *
4070 : : * Register @cfts to @ss. Files described by @cfts are created for all
4071 : : * existing cgroups to which @ss is attached and all future cgroups will
4072 : : * have them too. This function can be called anytime whether @ss is
4073 : : * attached or not.
4074 : : *
4075 : : * Returns 0 on successful registration, -errno on failure. Note that this
4076 : : * function currently returns 0 as long as @cfts registration is successful
4077 : : * even if some file creation attempts on existing cgroups fail.
4078 : : */
4079 : 7272 : static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4080 : : {
4081 : : int ret;
4082 : :
4083 [ + - ]: 14544 : if (!cgroup_ssid_enabled(ss->id))
4084 : : return 0;
4085 : :
4086 [ + + + - ]: 7272 : if (!cfts || cfts[0].name[0] == '\0')
4087 : : return 0;
4088 : :
4089 : 4848 : ret = cgroup_init_cftypes(ss, cfts);
4090 [ + - ]: 4848 : if (ret)
4091 : : return ret;
4092 : :
4093 : 4848 : mutex_lock(&cgroup_mutex);
4094 : :
4095 : 4848 : list_add_tail(&cfts->node, &ss->cfts);
4096 : 4848 : ret = cgroup_apply_cftypes(cfts, true);
4097 [ - + ]: 4848 : if (ret)
4098 : 0 : cgroup_rm_cftypes_locked(cfts);
4099 : :
4100 : 4848 : mutex_unlock(&cgroup_mutex);
4101 : 4848 : return ret;
4102 : : }
4103 : :
4104 : : /**
4105 : : * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4106 : : * @ss: target cgroup subsystem
4107 : : * @cfts: zero-length name terminated array of cftypes
4108 : : *
4109 : : * Similar to cgroup_add_cftypes() but the added files are only used for
4110 : : * the default hierarchy.
4111 : : */
4112 : 3232 : int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4113 : : {
4114 : : struct cftype *cft;
4115 : :
4116 [ + + + + ]: 7272 : for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4117 : 4040 : cft->flags |= __CFTYPE_ONLY_ON_DFL;
4118 : 3232 : return cgroup_add_cftypes(ss, cfts);
4119 : : }
4120 : :
4121 : : /**
4122 : : * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4123 : : * @ss: target cgroup subsystem
4124 : : * @cfts: zero-length name terminated array of cftypes
4125 : : *
4126 : : * Similar to cgroup_add_cftypes() but the added files are only used for
4127 : : * the legacy hierarchies.
4128 : : */
4129 : 3232 : int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4130 : : {
4131 : : struct cftype *cft;
4132 : :
4133 [ + - + + ]: 17776 : for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4134 : 14544 : cft->flags |= __CFTYPE_NOT_ON_DFL;
4135 : 3232 : return cgroup_add_cftypes(ss, cfts);
4136 : : }
4137 : :
4138 : : /**
4139 : : * cgroup_file_notify - generate a file modified event for a cgroup_file
4140 : : * @cfile: target cgroup_file
4141 : : *
4142 : : * @cfile must have been obtained by setting cftype->file_offset.
4143 : : */
4144 : 187424 : void cgroup_file_notify(struct cgroup_file *cfile)
4145 : : {
4146 : : unsigned long flags;
4147 : :
4148 : 187424 : spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4149 [ + + ]: 187424 : if (cfile->kn) {
4150 : 49584 : unsigned long last = cfile->notified_at;
4151 : 49584 : unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4152 : :
4153 [ + + + + ]: 49584 : if (time_in_range(jiffies, last, next)) {
4154 : 268 : timer_reduce(&cfile->notify_timer, next);
4155 : : } else {
4156 : 49316 : kernfs_notify(cfile->kn);
4157 : 49316 : cfile->notified_at = jiffies;
4158 : : }
4159 : : }
4160 : : spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4161 : 187424 : }
4162 : :
4163 : : /**
4164 : : * css_next_child - find the next child of a given css
4165 : : * @pos: the current position (%NULL to initiate traversal)
4166 : : * @parent: css whose children to walk
4167 : : *
4168 : : * This function returns the next child of @parent and should be called
4169 : : * under either cgroup_mutex or RCU read lock. The only requirement is
4170 : : * that @parent and @pos are accessible. The next sibling is guaranteed to
4171 : : * be returned regardless of their states.
4172 : : *
4173 : : * If a subsystem synchronizes ->css_online() and the start of iteration, a
4174 : : * css which finished ->css_online() is guaranteed to be visible in the
4175 : : * future iterations and will stay visible until the last reference is put.
4176 : : * A css which hasn't finished ->css_online() or already finished
4177 : : * ->css_offline() may show up during traversal. It's each subsystem's
4178 : : * responsibility to synchronize against on/offlining.
4179 : : */
4180 : 326104 : struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4181 : : struct cgroup_subsys_state *parent)
4182 : : {
4183 : : struct cgroup_subsys_state *next;
4184 : :
4185 : : cgroup_assert_mutex_or_rcu_locked();
4186 : :
4187 : : /*
4188 : : * @pos could already have been unlinked from the sibling list.
4189 : : * Once a cgroup is removed, its ->sibling.next is no longer
4190 : : * updated when its next sibling changes. CSS_RELEASED is set when
4191 : : * @pos is taken off list, at which time its next pointer is valid,
4192 : : * and, as releases are serialized, the one pointed to by the next
4193 : : * pointer is guaranteed to not have started release yet. This
4194 : : * implies that if we observe !CSS_RELEASED on @pos in this RCU
4195 : : * critical section, the one pointed to by its next pointer is
4196 : : * guaranteed to not have finished its RCU grace period even if we
4197 : : * have dropped rcu_read_lock() inbetween iterations.
4198 : : *
4199 : : * If @pos has CSS_RELEASED set, its next pointer can't be
4200 : : * dereferenced; however, as each css is given a monotonically
4201 : : * increasing unique serial number and always appended to the
4202 : : * sibling list, the next one can be found by walking the parent's
4203 : : * children until the first css with higher serial number than
4204 : : * @pos's. While this path can be slower, it happens iff iteration
4205 : : * races against release and the race window is very small.
4206 : : */
4207 [ + - ]: 326104 : if (!pos) {
4208 : 326104 : next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4209 [ # # ]: 0 : } else if (likely(!(pos->flags & CSS_RELEASED))) {
4210 : 0 : next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4211 : : } else {
4212 [ # # ]: 0 : list_for_each_entry_rcu(next, &parent->children, sibling)
4213 [ # # ]: 0 : if (next->serial_nr > pos->serial_nr)
4214 : : break;
4215 : : }
4216 : :
4217 : : /*
4218 : : * @next, if not pointing to the head, can be dereferenced and is
4219 : : * the next sibling.
4220 : : */
4221 [ - + ]: 326104 : if (&next->sibling != &parent->children)
4222 : 0 : return next;
4223 : : return NULL;
4224 : : }
4225 : :
4226 : : /**
4227 : : * css_next_descendant_pre - find the next descendant for pre-order walk
4228 : : * @pos: the current position (%NULL to initiate traversal)
4229 : : * @root: css whose descendants to walk
4230 : : *
4231 : : * To be used by css_for_each_descendant_pre(). Find the next descendant
4232 : : * to visit for pre-order traversal of @root's descendants. @root is
4233 : : * included in the iteration and the first node to be visited.
4234 : : *
4235 : : * While this function requires cgroup_mutex or RCU read locking, it
4236 : : * doesn't require the whole traversal to be contained in a single critical
4237 : : * section. This function will return the correct next descendant as long
4238 : : * as both @pos and @root are accessible and @pos is a descendant of @root.
4239 : : *
4240 : : * If a subsystem synchronizes ->css_online() and the start of iteration, a
4241 : : * css which finished ->css_online() is guaranteed to be visible in the
4242 : : * future iterations and will stay visible until the last reference is put.
4243 : : * A css which hasn't finished ->css_online() or already finished
4244 : : * ->css_offline() may show up during traversal. It's each subsystem's
4245 : : * responsibility to synchronize against on/offlining.
4246 : : */
4247 : : struct cgroup_subsys_state *
4248 : 488144 : css_next_descendant_pre(struct cgroup_subsys_state *pos,
4249 : : struct cgroup_subsys_state *root)
4250 : : {
4251 : : struct cgroup_subsys_state *next;
4252 : :
4253 : : cgroup_assert_mutex_or_rcu_locked();
4254 : :
4255 : : /* if first iteration, visit @root */
4256 [ + + ]: 488144 : if (!pos)
4257 : : return root;
4258 : :
4259 : : /* visit the first child if exists */
4260 : 244072 : next = css_next_child(NULL, pos);
4261 [ + - ]: 244072 : if (next)
4262 : : return next;
4263 : :
4264 : : /* no child, visit my or the closest ancestor's next sibling */
4265 [ - + ]: 244072 : while (pos != root) {
4266 : 0 : next = css_next_child(pos, pos->parent);
4267 [ # # ]: 0 : if (next)
4268 : 0 : return next;
4269 : 0 : pos = pos->parent;
4270 : : }
4271 : :
4272 : : return NULL;
4273 : : }
4274 : : EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4275 : :
4276 : : /**
4277 : : * css_rightmost_descendant - return the rightmost descendant of a css
4278 : : * @pos: css of interest
4279 : : *
4280 : : * Return the rightmost descendant of @pos. If there's no descendant, @pos
4281 : : * is returned. This can be used during pre-order traversal to skip
4282 : : * subtree of @pos.
4283 : : *
4284 : : * While this function requires cgroup_mutex or RCU read locking, it
4285 : : * doesn't require the whole traversal to be contained in a single critical
4286 : : * section. This function will return the correct rightmost descendant as
4287 : : * long as @pos is accessible.
4288 : : */
4289 : : struct cgroup_subsys_state *
4290 : 0 : css_rightmost_descendant(struct cgroup_subsys_state *pos)
4291 : : {
4292 : : struct cgroup_subsys_state *last, *tmp;
4293 : :
4294 : : cgroup_assert_mutex_or_rcu_locked();
4295 : :
4296 : : do {
4297 : : last = pos;
4298 : : /* ->prev isn't RCU safe, walk ->next till the end */
4299 : : pos = NULL;
4300 [ # # ]: 0 : css_for_each_child(tmp, last)
4301 : : pos = tmp;
4302 [ # # ]: 0 : } while (pos);
4303 : :
4304 : 0 : return last;
4305 : : }
4306 : :
4307 : : static struct cgroup_subsys_state *
4308 : : css_leftmost_descendant(struct cgroup_subsys_state *pos)
4309 : : {
4310 : : struct cgroup_subsys_state *last;
4311 : :
4312 : : do {
4313 : : last = pos;
4314 : 7676 : pos = css_next_child(NULL, pos);
4315 [ - + # # ]: 7676 : } while (pos);
4316 : :
4317 : 7676 : return last;
4318 : : }
4319 : :
4320 : : /**
4321 : : * css_next_descendant_post - find the next descendant for post-order walk
4322 : : * @pos: the current position (%NULL to initiate traversal)
4323 : : * @root: css whose descendants to walk
4324 : : *
4325 : : * To be used by css_for_each_descendant_post(). Find the next descendant
4326 : : * to visit for post-order traversal of @root's descendants. @root is
4327 : : * included in the iteration and the last node to be visited.
4328 : : *
4329 : : * While this function requires cgroup_mutex or RCU read locking, it
4330 : : * doesn't require the whole traversal to be contained in a single critical
4331 : : * section. This function will return the correct next descendant as long
4332 : : * as both @pos and @cgroup are accessible and @pos is a descendant of
4333 : : * @cgroup.
4334 : : *
4335 : : * If a subsystem synchronizes ->css_online() and the start of iteration, a
4336 : : * css which finished ->css_online() is guaranteed to be visible in the
4337 : : * future iterations and will stay visible until the last reference is put.
4338 : : * A css which hasn't finished ->css_online() or already finished
4339 : : * ->css_offline() may show up during traversal. It's each subsystem's
4340 : : * responsibility to synchronize against on/offlining.
4341 : : */
4342 : : struct cgroup_subsys_state *
4343 : 15352 : css_next_descendant_post(struct cgroup_subsys_state *pos,
4344 : : struct cgroup_subsys_state *root)
4345 : : {
4346 : : struct cgroup_subsys_state *next;
4347 : :
4348 : : cgroup_assert_mutex_or_rcu_locked();
4349 : :
4350 : : /* if first iteration, visit leftmost descendant which may be @root */
4351 [ + + ]: 15352 : if (!pos)
4352 : 7676 : return css_leftmost_descendant(root);
4353 : :
4354 : : /* if we visited @root, we're done */
4355 [ - + ]: 7676 : if (pos == root)
4356 : : return NULL;
4357 : :
4358 : : /* if there's an unvisited sibling, visit its leftmost descendant */
4359 : 0 : next = css_next_child(pos, pos->parent);
4360 [ # # ]: 0 : if (next)
4361 : 0 : return css_leftmost_descendant(next);
4362 : :
4363 : : /* no sibling left, visit parent */
4364 : 0 : return pos->parent;
4365 : : }
4366 : :
4367 : : /**
4368 : : * css_has_online_children - does a css have online children
4369 : : * @css: the target css
4370 : : *
4371 : : * Returns %true if @css has any online children; otherwise, %false. This
4372 : : * function can be called from any context but the caller is responsible
4373 : : * for synchronizing against on/offlining as necessary.
4374 : : */
4375 : 70316 : bool css_has_online_children(struct cgroup_subsys_state *css)
4376 : : {
4377 : : struct cgroup_subsys_state *child;
4378 : : bool ret = false;
4379 : :
4380 : : rcu_read_lock();
4381 [ - + ]: 70316 : css_for_each_child(child, css) {
4382 [ # # ]: 0 : if (child->flags & CSS_ONLINE) {
4383 : : ret = true;
4384 : : break;
4385 : : }
4386 : : }
4387 : : rcu_read_unlock();
4388 : 70316 : return ret;
4389 : : }
4390 : :
4391 : 125944 : static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4392 : : {
4393 : : struct list_head *l;
4394 : : struct cgrp_cset_link *link;
4395 : : struct css_set *cset;
4396 : :
4397 : : lockdep_assert_held(&css_set_lock);
4398 : :
4399 : : /* find the next threaded cset */
4400 [ + + ]: 125944 : if (it->tcset_pos) {
4401 : 11850 : l = it->tcset_pos->next;
4402 : :
4403 [ - + ]: 11850 : if (l != it->tcset_head) {
4404 : 0 : it->tcset_pos = l;
4405 : 0 : return container_of(l, struct css_set,
4406 : : threaded_csets_node);
4407 : : }
4408 : :
4409 : 11850 : it->tcset_pos = NULL;
4410 : : }
4411 : :
4412 : : /* find the next cset */
4413 : 125944 : l = it->cset_pos;
4414 : 125944 : l = l->next;
4415 [ + + ]: 125944 : if (l == it->cset_head) {
4416 : 114094 : it->cset_pos = NULL;
4417 : 114094 : return NULL;
4418 : : }
4419 : :
4420 [ - + ]: 11850 : if (it->ss) {
4421 : 0 : cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4422 : : } else {
4423 : : link = list_entry(l, struct cgrp_cset_link, cset_link);
4424 : 11850 : cset = link->cset;
4425 : : }
4426 : :
4427 : 11850 : it->cset_pos = l;
4428 : :
4429 : : /* initialize threaded css_set walking */
4430 [ + - ]: 11850 : if (it->flags & CSS_TASK_ITER_THREADED) {
4431 [ - + ]: 11850 : if (it->cur_dcset)
4432 : 0 : put_css_set_locked(it->cur_dcset);
4433 : 11850 : it->cur_dcset = cset;
4434 : : get_css_set(cset);
4435 : :
4436 : 11850 : it->tcset_head = &cset->threaded_csets;
4437 : 11850 : it->tcset_pos = &cset->threaded_csets;
4438 : : }
4439 : :
4440 : 11850 : return cset;
4441 : : }
4442 : :
4443 : : /**
4444 : : * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4445 : : * @it: the iterator to advance
4446 : : *
4447 : : * Advance @it to the next css_set to walk.
4448 : : */
4449 : 125944 : static void css_task_iter_advance_css_set(struct css_task_iter *it)
4450 : : {
4451 : : struct css_set *cset;
4452 : :
4453 : : lockdep_assert_held(&css_set_lock);
4454 : :
4455 : : /* Advance to the next non-empty css_set */
4456 : : do {
4457 : 125944 : cset = css_task_iter_next_css_set(it);
4458 [ + + ]: 125944 : if (!cset) {
4459 : 114094 : it->task_pos = NULL;
4460 : 240038 : return;
4461 : : }
4462 [ + + - + ]: 20354 : } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks));
4463 : :
4464 [ + + ]: 23700 : if (!list_empty(&cset->tasks)) {
4465 : 3346 : it->task_pos = cset->tasks.next;
4466 : 3346 : it->cur_tasks_head = &cset->tasks;
4467 [ - + ]: 17008 : } else if (!list_empty(&cset->mg_tasks)) {
4468 : 0 : it->task_pos = cset->mg_tasks.next;
4469 : 0 : it->cur_tasks_head = &cset->mg_tasks;
4470 : : } else {
4471 : 8504 : it->task_pos = cset->dying_tasks.next;
4472 : 8504 : it->cur_tasks_head = &cset->dying_tasks;
4473 : : }
4474 : :
4475 : 11850 : it->tasks_head = &cset->tasks;
4476 : 11850 : it->mg_tasks_head = &cset->mg_tasks;
4477 : 11850 : it->dying_tasks_head = &cset->dying_tasks;
4478 : :
4479 : : /*
4480 : : * We don't keep css_sets locked across iteration steps and thus
4481 : : * need to take steps to ensure that iteration can be resumed after
4482 : : * the lock is re-acquired. Iteration is performed at two levels -
4483 : : * css_sets and tasks in them.
4484 : : *
4485 : : * Once created, a css_set never leaves its cgroup lists, so a
4486 : : * pinned css_set is guaranteed to stay put and we can resume
4487 : : * iteration afterwards.
4488 : : *
4489 : : * Tasks may leave @cset across iteration steps. This is resolved
4490 : : * by registering each iterator with the css_set currently being
4491 : : * walked and making css_set_move_task() advance iterators whose
4492 : : * next task is leaving.
4493 : : */
4494 [ - + ]: 11850 : if (it->cur_cset) {
4495 : : list_del(&it->iters_node);
4496 : 0 : put_css_set_locked(it->cur_cset);
4497 : : }
4498 : : get_css_set(cset);
4499 : 11850 : it->cur_cset = cset;
4500 : 11850 : list_add(&it->iters_node, &cset->task_iters);
4501 : : }
4502 : :
4503 : : static void css_task_iter_skip(struct css_task_iter *it,
4504 : : struct task_struct *task)
4505 : : {
4506 : : lockdep_assert_held(&css_set_lock);
4507 : :
4508 [ - + ]: 404 : if (it->task_pos == &task->cg_list) {
4509 : 0 : it->task_pos = it->task_pos->next;
4510 : 0 : it->flags |= CSS_TASK_ITER_SKIPPED;
4511 : : }
4512 : : }
4513 : :
4514 : 142628 : static void css_task_iter_advance(struct css_task_iter *it)
4515 : : {
4516 : : struct task_struct *task;
4517 : :
4518 : : lockdep_assert_held(&css_set_lock);
4519 : : repeat:
4520 [ + + ]: 151536 : if (it->task_pos) {
4521 : : /*
4522 : : * Advance iterator to find next entry. cset->tasks is
4523 : : * consumed first and then ->mg_tasks. After ->mg_tasks,
4524 : : * we move onto the next cset.
4525 : : */
4526 [ - + ]: 37442 : if (it->flags & CSS_TASK_ITER_SKIPPED)
4527 : 0 : it->flags &= ~CSS_TASK_ITER_SKIPPED;
4528 : : else
4529 : 37442 : it->task_pos = it->task_pos->next;
4530 : :
4531 [ + + ]: 37442 : if (it->task_pos == it->tasks_head) {
4532 : 3346 : it->task_pos = it->mg_tasks_head->next;
4533 : 3346 : it->cur_tasks_head = it->mg_tasks_head;
4534 : : }
4535 [ + + ]: 37442 : if (it->task_pos == it->mg_tasks_head) {
4536 : 3346 : it->task_pos = it->dying_tasks_head->next;
4537 : 3346 : it->cur_tasks_head = it->dying_tasks_head;
4538 : : }
4539 [ + + ]: 37442 : if (it->task_pos == it->dying_tasks_head)
4540 : 11850 : css_task_iter_advance_css_set(it);
4541 : : } else {
4542 : : /* called from start, proceed to the first cset */
4543 : 114094 : css_task_iter_advance_css_set(it);
4544 : : }
4545 : :
4546 [ + + ]: 151536 : if (!it->task_pos)
4547 : 142628 : return;
4548 : :
4549 : : task = list_entry(it->task_pos, struct task_struct, cg_list);
4550 : :
4551 [ + - ]: 37442 : if (it->flags & CSS_TASK_ITER_PROCS) {
4552 : : /* if PROCS, skip over tasks which aren't group leaders */
4553 [ - + ]: 37442 : if (!thread_group_leader(task))
4554 : : goto repeat;
4555 : :
4556 : : /* and dying leaders w/o live member threads */
4557 [ + + + - ]: 46350 : if (it->cur_tasks_head == it->dying_tasks_head &&
4558 : 17816 : !atomic_read(&task->signal->live))
4559 : : goto repeat;
4560 : : } else {
4561 : : /* skip all dying ones */
4562 [ # # ]: 0 : if (it->cur_tasks_head == it->dying_tasks_head)
4563 : : goto repeat;
4564 : : }
4565 : : }
4566 : :
4567 : : /**
4568 : : * css_task_iter_start - initiate task iteration
4569 : : * @css: the css to walk tasks of
4570 : : * @flags: CSS_TASK_ITER_* flags
4571 : : * @it: the task iterator to use
4572 : : *
4573 : : * Initiate iteration through the tasks of @css. The caller can call
4574 : : * css_task_iter_next() to walk through the tasks until the function
4575 : : * returns NULL. On completion of iteration, css_task_iter_end() must be
4576 : : * called.
4577 : : */
4578 : 114094 : void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4579 : : struct css_task_iter *it)
4580 : : {
4581 : : /* no one should try to iterate before mounting cgroups */
4582 [ - + # # ]: 114094 : WARN_ON_ONCE(!use_task_css_set_links);
4583 : :
4584 : 114094 : memset(it, 0, sizeof(*it));
4585 : :
4586 : : spin_lock_irq(&css_set_lock);
4587 : :
4588 : 114094 : it->ss = css->ss;
4589 : 114094 : it->flags = flags;
4590 : :
4591 [ - + ]: 114094 : if (it->ss)
4592 : 0 : it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4593 : : else
4594 : 114094 : it->cset_pos = &css->cgroup->cset_links;
4595 : :
4596 : 114094 : it->cset_head = it->cset_pos;
4597 : :
4598 : 114094 : css_task_iter_advance(it);
4599 : :
4600 : : spin_unlock_irq(&css_set_lock);
4601 : 114094 : }
4602 : :
4603 : : /**
4604 : : * css_task_iter_next - return the next task for the iterator
4605 : : * @it: the task iterator being iterated
4606 : : *
4607 : : * The "next" function for task iteration. @it should have been
4608 : : * initialized via css_task_iter_start(). Returns NULL when the iteration
4609 : : * reaches the end.
4610 : : */
4611 : 142628 : struct task_struct *css_task_iter_next(struct css_task_iter *it)
4612 : : {
4613 [ + + ]: 142628 : if (it->cur_task) {
4614 : 28534 : put_task_struct(it->cur_task);
4615 : 28534 : it->cur_task = NULL;
4616 : : }
4617 : :
4618 : : spin_lock_irq(&css_set_lock);
4619 : :
4620 : : /* @it may be half-advanced by skips, finish advancing */
4621 [ - + ]: 142628 : if (it->flags & CSS_TASK_ITER_SKIPPED)
4622 : 0 : css_task_iter_advance(it);
4623 : :
4624 [ + + ]: 142628 : if (it->task_pos) {
4625 : 28534 : it->cur_task = list_entry(it->task_pos, struct task_struct,
4626 : : cg_list);
4627 : : get_task_struct(it->cur_task);
4628 : 28534 : css_task_iter_advance(it);
4629 : : }
4630 : :
4631 : : spin_unlock_irq(&css_set_lock);
4632 : :
4633 : 142628 : return it->cur_task;
4634 : : }
4635 : :
4636 : : /**
4637 : : * css_task_iter_end - finish task iteration
4638 : : * @it: the task iterator to finish
4639 : : *
4640 : : * Finish task iteration started by css_task_iter_start().
4641 : : */
4642 : 114094 : void css_task_iter_end(struct css_task_iter *it)
4643 : : {
4644 [ + + ]: 114094 : if (it->cur_cset) {
4645 : : spin_lock_irq(&css_set_lock);
4646 : : list_del(&it->iters_node);
4647 : 11850 : put_css_set_locked(it->cur_cset);
4648 : : spin_unlock_irq(&css_set_lock);
4649 : : }
4650 : :
4651 [ + + ]: 114094 : if (it->cur_dcset)
4652 : 11850 : put_css_set(it->cur_dcset);
4653 : :
4654 [ - + ]: 114094 : if (it->cur_task)
4655 : 0 : put_task_struct(it->cur_task);
4656 : 114094 : }
4657 : :
4658 : 169462 : static void cgroup_procs_release(struct kernfs_open_file *of)
4659 : : {
4660 [ + + ]: 169462 : if (of->priv) {
4661 : 114094 : css_task_iter_end(of->priv);
4662 : 114094 : kfree(of->priv);
4663 : : }
4664 : 169462 : }
4665 : :
4666 : 28534 : static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4667 : : {
4668 : 142628 : struct kernfs_open_file *of = s->private;
4669 : 142628 : struct css_task_iter *it = of->priv;
4670 : :
4671 [ + - ]: 28534 : if (pos)
4672 : 28534 : (*pos)++;
4673 : :
4674 : 142628 : return css_task_iter_next(it);
4675 : : }
4676 : :
4677 : 117440 : static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4678 : : unsigned int iter_flags)
4679 : : {
4680 : 117440 : struct kernfs_open_file *of = s->private;
4681 : 117440 : struct cgroup *cgrp = seq_css(s)->cgroup;
4682 : 117440 : struct css_task_iter *it = of->priv;
4683 : :
4684 : : /*
4685 : : * When a seq_file is seeked, it's always traversed sequentially
4686 : : * from position 0, so we can simply keep iterating on !0 *pos.
4687 : : */
4688 [ + + ]: 117440 : if (!it) {
4689 [ - + # # : 114094 : if (WARN_ON_ONCE((*pos)))
+ - ]
4690 : : return ERR_PTR(-EINVAL);
4691 : :
4692 : 114094 : it = kzalloc(sizeof(*it), GFP_KERNEL);
4693 [ + - ]: 114094 : if (!it)
4694 : : return ERR_PTR(-ENOMEM);
4695 : 114094 : of->priv = it;
4696 : 114094 : css_task_iter_start(&cgrp->self, iter_flags, it);
4697 [ - + ]: 3346 : } else if (!(*pos)) {
4698 : 0 : css_task_iter_end(it);
4699 : 0 : css_task_iter_start(&cgrp->self, iter_flags, it);
4700 : : } else
4701 : 3346 : return it->cur_task;
4702 : :
4703 : 114094 : return cgroup_procs_next(s, NULL, NULL);
4704 : : }
4705 : :
4706 : 117440 : static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4707 : : {
4708 : 117440 : struct cgroup *cgrp = seq_css(s)->cgroup;
4709 : :
4710 : : /*
4711 : : * All processes of a threaded subtree belong to the domain cgroup
4712 : : * of the subtree. Only threads can be distributed across the
4713 : : * subtree. Reject reads on cgroup.procs in the subtree proper.
4714 : : * They're always empty anyway.
4715 : : */
4716 [ + - ]: 117440 : if (cgroup_is_threaded(cgrp))
4717 : : return ERR_PTR(-EOPNOTSUPP);
4718 : :
4719 : 117440 : return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4720 : : CSS_TASK_ITER_THREADED);
4721 : : }
4722 : :
4723 : 28534 : static int cgroup_procs_show(struct seq_file *s, void *v)
4724 : : {
4725 : 28534 : seq_printf(s, "%d\n", task_pid_vnr(v));
4726 : 28534 : return 0;
4727 : : }
4728 : :
4729 : 55368 : static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4730 : : struct cgroup *dst_cgrp,
4731 : : struct super_block *sb)
4732 : : {
4733 : 55368 : struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4734 : : struct cgroup *com_cgrp = src_cgrp;
4735 : : struct inode *inode;
4736 : : int ret;
4737 : :
4738 : : lockdep_assert_held(&cgroup_mutex);
4739 : :
4740 : : /* find the common ancestor */
4741 [ + + ]: 139440 : while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4742 : : com_cgrp = cgroup_parent(com_cgrp);
4743 : :
4744 : : /* %current should be authorized to migrate to the common ancestor */
4745 : 55368 : inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn);
4746 [ + - ]: 55368 : if (!inode)
4747 : : return -ENOMEM;
4748 : :
4749 : 55368 : ret = inode_permission(inode, MAY_WRITE);
4750 : 55368 : iput(inode);
4751 [ + - ]: 55368 : if (ret)
4752 : : return ret;
4753 : :
4754 : : /*
4755 : : * If namespaces are delegation boundaries, %current must be able
4756 : : * to see both source and destination cgroups from its namespace.
4757 : : */
4758 [ + - + - ]: 110736 : if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4759 [ + - ]: 110736 : (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4760 : : !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4761 : : return -ENOENT;
4762 : :
4763 : 55368 : return 0;
4764 : : }
4765 : :
4766 : 55368 : static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4767 : : char *buf, size_t nbytes, loff_t off)
4768 : : {
4769 : : struct cgroup *src_cgrp, *dst_cgrp;
4770 : : struct task_struct *task;
4771 : : ssize_t ret;
4772 : :
4773 : 55368 : dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4774 [ + - ]: 55368 : if (!dst_cgrp)
4775 : : return -ENODEV;
4776 : :
4777 : 55368 : task = cgroup_procs_write_start(buf, true);
4778 : : ret = PTR_ERR_OR_ZERO(task);
4779 [ + - ]: 55368 : if (ret)
4780 : : goto out_unlock;
4781 : :
4782 : : /* find the source cgroup */
4783 : : spin_lock_irq(&css_set_lock);
4784 : : src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4785 : : spin_unlock_irq(&css_set_lock);
4786 : :
4787 : 55368 : ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4788 : 55368 : of->file->f_path.dentry->d_sb);
4789 [ + - ]: 55368 : if (ret)
4790 : : goto out_finish;
4791 : :
4792 : 55368 : ret = cgroup_attach_task(dst_cgrp, task, true);
4793 : :
4794 : : out_finish:
4795 : 55368 : cgroup_procs_write_finish(task);
4796 : : out_unlock:
4797 : 55368 : cgroup_kn_unlock(of->kn);
4798 : :
4799 [ + - ]: 55368 : return ret ?: nbytes;
4800 : : }
4801 : :
4802 : 0 : static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4803 : : {
4804 : 0 : return __cgroup_procs_start(s, pos, 0);
4805 : : }
4806 : :
4807 : 0 : static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4808 : : char *buf, size_t nbytes, loff_t off)
4809 : : {
4810 : : struct cgroup *src_cgrp, *dst_cgrp;
4811 : : struct task_struct *task;
4812 : : ssize_t ret;
4813 : :
4814 : : buf = strstrip(buf);
4815 : :
4816 : 0 : dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4817 [ # # ]: 0 : if (!dst_cgrp)
4818 : : return -ENODEV;
4819 : :
4820 : 0 : task = cgroup_procs_write_start(buf, false);
4821 : : ret = PTR_ERR_OR_ZERO(task);
4822 [ # # ]: 0 : if (ret)
4823 : : goto out_unlock;
4824 : :
4825 : : /* find the source cgroup */
4826 : : spin_lock_irq(&css_set_lock);
4827 : : src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4828 : : spin_unlock_irq(&css_set_lock);
4829 : :
4830 : : /* thread migrations follow the cgroup.procs delegation rule */
4831 : 0 : ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4832 : 0 : of->file->f_path.dentry->d_sb);
4833 [ # # ]: 0 : if (ret)
4834 : : goto out_finish;
4835 : :
4836 : : /* and must be contained in the same domain */
4837 : : ret = -EOPNOTSUPP;
4838 [ # # ]: 0 : if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp)
4839 : : goto out_finish;
4840 : :
4841 : 0 : ret = cgroup_attach_task(dst_cgrp, task, false);
4842 : :
4843 : : out_finish:
4844 : 0 : cgroup_procs_write_finish(task);
4845 : : out_unlock:
4846 : 0 : cgroup_kn_unlock(of->kn);
4847 : :
4848 [ # # ]: 0 : return ret ?: nbytes;
4849 : : }
4850 : :
4851 : : /* cgroup core interface files for the default hierarchy */
4852 : : static struct cftype cgroup_base_files[] = {
4853 : : {
4854 : : .name = "cgroup.type",
4855 : : .flags = CFTYPE_NOT_ON_ROOT,
4856 : : .seq_show = cgroup_type_show,
4857 : : .write = cgroup_type_write,
4858 : : },
4859 : : {
4860 : : .name = "cgroup.procs",
4861 : : .flags = CFTYPE_NS_DELEGATABLE,
4862 : : .file_offset = offsetof(struct cgroup, procs_file),
4863 : : .release = cgroup_procs_release,
4864 : : .seq_start = cgroup_procs_start,
4865 : : .seq_next = cgroup_procs_next,
4866 : : .seq_show = cgroup_procs_show,
4867 : : .write = cgroup_procs_write,
4868 : : },
4869 : : {
4870 : : .name = "cgroup.threads",
4871 : : .flags = CFTYPE_NS_DELEGATABLE,
4872 : : .release = cgroup_procs_release,
4873 : : .seq_start = cgroup_threads_start,
4874 : : .seq_next = cgroup_procs_next,
4875 : : .seq_show = cgroup_procs_show,
4876 : : .write = cgroup_threads_write,
4877 : : },
4878 : : {
4879 : : .name = "cgroup.controllers",
4880 : : .seq_show = cgroup_controllers_show,
4881 : : },
4882 : : {
4883 : : .name = "cgroup.subtree_control",
4884 : : .flags = CFTYPE_NS_DELEGATABLE,
4885 : : .seq_show = cgroup_subtree_control_show,
4886 : : .write = cgroup_subtree_control_write,
4887 : : },
4888 : : {
4889 : : .name = "cgroup.events",
4890 : : .flags = CFTYPE_NOT_ON_ROOT,
4891 : : .file_offset = offsetof(struct cgroup, events_file),
4892 : : .seq_show = cgroup_events_show,
4893 : : },
4894 : : {
4895 : : .name = "cgroup.max.descendants",
4896 : : .seq_show = cgroup_max_descendants_show,
4897 : : .write = cgroup_max_descendants_write,
4898 : : },
4899 : : {
4900 : : .name = "cgroup.max.depth",
4901 : : .seq_show = cgroup_max_depth_show,
4902 : : .write = cgroup_max_depth_write,
4903 : : },
4904 : : {
4905 : : .name = "cgroup.stat",
4906 : : .seq_show = cgroup_stat_show,
4907 : : },
4908 : : {
4909 : : .name = "cgroup.freeze",
4910 : : .flags = CFTYPE_NOT_ON_ROOT,
4911 : : .seq_show = cgroup_freeze_show,
4912 : : .write = cgroup_freeze_write,
4913 : : },
4914 : : {
4915 : : .name = "cpu.stat",
4916 : : .flags = CFTYPE_NOT_ON_ROOT,
4917 : : .seq_show = cpu_stat_show,
4918 : : },
4919 : : #ifdef CONFIG_PSI
4920 : : {
4921 : : .name = "io.pressure",
4922 : : .seq_show = cgroup_io_pressure_show,
4923 : : .write = cgroup_io_pressure_write,
4924 : : .poll = cgroup_pressure_poll,
4925 : : .release = cgroup_pressure_release,
4926 : : },
4927 : : {
4928 : : .name = "memory.pressure",
4929 : : .seq_show = cgroup_memory_pressure_show,
4930 : : .write = cgroup_memory_pressure_write,
4931 : : .poll = cgroup_pressure_poll,
4932 : : .release = cgroup_pressure_release,
4933 : : },
4934 : : {
4935 : : .name = "cpu.pressure",
4936 : : .seq_show = cgroup_cpu_pressure_show,
4937 : : .write = cgroup_cpu_pressure_write,
4938 : : .poll = cgroup_pressure_poll,
4939 : : .release = cgroup_pressure_release,
4940 : : },
4941 : : #endif /* CONFIG_PSI */
4942 : : { } /* terminate */
4943 : : };
4944 : :
4945 : : /*
4946 : : * css destruction is four-stage process.
4947 : : *
4948 : : * 1. Destruction starts. Killing of the percpu_ref is initiated.
4949 : : * Implemented in kill_css().
4950 : : *
4951 : : * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4952 : : * and thus css_tryget_online() is guaranteed to fail, the css can be
4953 : : * offlined by invoking offline_css(). After offlining, the base ref is
4954 : : * put. Implemented in css_killed_work_fn().
4955 : : *
4956 : : * 3. When the percpu_ref reaches zero, the only possible remaining
4957 : : * accessors are inside RCU read sections. css_release() schedules the
4958 : : * RCU callback.
4959 : : *
4960 : : * 4. After the grace period, the css can be freed. Implemented in
4961 : : * css_free_work_fn().
4962 : : *
4963 : : * It is actually hairier because both step 2 and 4 require process context
4964 : : * and thus involve punting to css->destroy_work adding two additional
4965 : : * steps to the already complex sequence.
4966 : : */
4967 : 71128 : static void css_free_rwork_fn(struct work_struct *work)
4968 : : {
4969 : 71128 : struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
4970 : : struct cgroup_subsys_state, destroy_rwork);
4971 : 71128 : struct cgroup_subsys *ss = css->ss;
4972 : 71128 : struct cgroup *cgrp = css->cgroup;
4973 : :
4974 : 71128 : percpu_ref_exit(&css->refcnt);
4975 : :
4976 [ + + ]: 71128 : if (ss) {
4977 : : /* css free path */
4978 : 23440 : struct cgroup_subsys_state *parent = css->parent;
4979 : 23440 : int id = css->id;
4980 : :
4981 : 23440 : ss->css_free(css);
4982 : 23440 : cgroup_idr_remove(&ss->css_idr, id);
4983 : : cgroup_put(cgrp);
4984 : :
4985 [ + - ]: 23440 : if (parent)
4986 : : css_put(parent);
4987 : : } else {
4988 : : /* cgroup free path */
4989 : 47688 : atomic_dec(&cgrp->root->nr_cgrps);
4990 : 47688 : cgroup1_pidlist_destroy_all(cgrp);
4991 : 47688 : cancel_work_sync(&cgrp->release_agent_work);
4992 : :
4993 [ + - ]: 47688 : if (cgroup_parent(cgrp)) {
4994 : : /*
4995 : : * We get a ref to the parent, and put the ref when
4996 : : * this cgroup is being freed, so it's guaranteed
4997 : : * that the parent won't be destroyed before its
4998 : : * children.
4999 : : */
5000 : : cgroup_put(cgroup_parent(cgrp));
5001 : 47688 : kernfs_put(cgrp->kn);
5002 : : psi_cgroup_free(cgrp);
5003 [ + + ]: 47686 : if (cgroup_on_dfl(cgrp))
5004 : 12124 : cgroup_rstat_exit(cgrp);
5005 : 47686 : kfree(cgrp);
5006 : : } else {
5007 : : /*
5008 : : * This is root cgroup's refcnt reaching zero,
5009 : : * which indicates that the root should be
5010 : : * released.
5011 : : */
5012 : 0 : cgroup_destroy_root(cgrp->root);
5013 : : }
5014 : : }
5015 : 71128 : }
5016 : :
5017 : 71128 : static void css_release_work_fn(struct work_struct *work)
5018 : : {
5019 : : struct cgroup_subsys_state *css =
5020 : 71128 : container_of(work, struct cgroup_subsys_state, destroy_work);
5021 : 71128 : struct cgroup_subsys *ss = css->ss;
5022 : 71128 : struct cgroup *cgrp = css->cgroup;
5023 : :
5024 : 71128 : mutex_lock(&cgroup_mutex);
5025 : :
5026 : 71128 : css->flags |= CSS_RELEASED;
5027 : : list_del_rcu(&css->sibling);
5028 : :
5029 [ + + ]: 71128 : if (ss) {
5030 : : /* css release path */
5031 [ - + ]: 46880 : if (!list_empty(&css->rstat_css_node)) {
5032 : 0 : cgroup_rstat_flush(cgrp);
5033 : : list_del_rcu(&css->rstat_css_node);
5034 : : }
5035 : :
5036 : 23440 : cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5037 [ - + ]: 23440 : if (ss->css_released)
5038 : 0 : ss->css_released(css);
5039 : : } else {
5040 : : struct cgroup *tcgrp;
5041 : :
5042 : : /* cgroup release path */
5043 [ - + ]: 47688 : TRACE_CGROUP_PATH(release, cgrp);
5044 : :
5045 [ + + ]: 47688 : if (cgroup_on_dfl(cgrp))
5046 : 12124 : cgroup_rstat_flush(cgrp);
5047 : :
5048 : : spin_lock_irq(&css_set_lock);
5049 [ + + ]: 193176 : for (tcgrp = cgroup_parent(cgrp); tcgrp;
5050 : : tcgrp = cgroup_parent(tcgrp))
5051 : 97800 : tcgrp->nr_dying_descendants--;
5052 : : spin_unlock_irq(&css_set_lock);
5053 : :
5054 : 47688 : cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
5055 : 47688 : cgrp->id = -1;
5056 : :
5057 : : /*
5058 : : * There are two control paths which try to determine
5059 : : * cgroup from dentry without going through kernfs -
5060 : : * cgroupstats_build() and css_tryget_online_from_dir().
5061 : : * Those are supported by RCU protecting clearing of
5062 : : * cgrp->kn->priv backpointer.
5063 : : */
5064 [ + - ]: 47688 : if (cgrp->kn)
5065 : : RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5066 : : NULL);
5067 : : }
5068 : :
5069 : 71128 : mutex_unlock(&cgroup_mutex);
5070 : :
5071 : 142256 : INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5072 : 71128 : queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5073 : 71128 : }
5074 : :
5075 : 71128 : static void css_release(struct percpu_ref *ref)
5076 : : {
5077 : : struct cgroup_subsys_state *css =
5078 : : container_of(ref, struct cgroup_subsys_state, refcnt);
5079 : :
5080 : 142256 : INIT_WORK(&css->destroy_work, css_release_work_fn);
5081 : 71128 : queue_work(cgroup_destroy_wq, &css->destroy_work);
5082 : 71128 : }
5083 : :
5084 : 51320 : static void init_and_link_css(struct cgroup_subsys_state *css,
5085 : : struct cgroup_subsys *ss, struct cgroup *cgrp)
5086 : : {
5087 : : lockdep_assert_held(&cgroup_mutex);
5088 : :
5089 : 51320 : cgroup_get_live(cgrp);
5090 : :
5091 : 51320 : memset(css, 0, sizeof(*css));
5092 : 51320 : css->cgroup = cgrp;
5093 : 51320 : css->ss = ss;
5094 : 51320 : css->id = -1;
5095 : 51320 : INIT_LIST_HEAD(&css->sibling);
5096 : 51320 : INIT_LIST_HEAD(&css->children);
5097 : 51320 : INIT_LIST_HEAD(&css->rstat_css_node);
5098 : 51320 : css->serial_nr = css_serial_nr_next++;
5099 : : atomic_set(&css->online_cnt, 0);
5100 : :
5101 [ + + ]: 51320 : if (cgroup_parent(cgrp)) {
5102 : 46876 : css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5103 : : css_get(css->parent);
5104 : : }
5105 : :
5106 [ + + - + ]: 51320 : if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush)
5107 : 0 : list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5108 : :
5109 [ - + ]: 51320 : BUG_ON(cgroup_css(cgrp, ss));
5110 : 51320 : }
5111 : :
5112 : : /* invoke ->css_online() on a new CSS and mark it online if successful */
5113 : 51320 : static int online_css(struct cgroup_subsys_state *css)
5114 : : {
5115 : 51320 : struct cgroup_subsys *ss = css->ss;
5116 : : int ret = 0;
5117 : :
5118 : : lockdep_assert_held(&cgroup_mutex);
5119 : :
5120 [ + + ]: 51320 : if (ss->css_online)
5121 : 24648 : ret = ss->css_online(css);
5122 [ + - ]: 51320 : if (!ret) {
5123 : 51320 : css->flags |= CSS_ONLINE;
5124 : 51320 : rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5125 : :
5126 : 51320 : atomic_inc(&css->online_cnt);
5127 [ + + ]: 51320 : if (css->parent)
5128 : 46876 : atomic_inc(&css->parent->online_cnt);
5129 : : }
5130 : 51320 : return ret;
5131 : : }
5132 : :
5133 : : /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5134 : 23440 : static void offline_css(struct cgroup_subsys_state *css)
5135 : : {
5136 : 23440 : struct cgroup_subsys *ss = css->ss;
5137 : :
5138 : : lockdep_assert_held(&cgroup_mutex);
5139 : :
5140 [ + - ]: 23440 : if (!(css->flags & CSS_ONLINE))
5141 : 23440 : return;
5142 : :
5143 [ + + ]: 23440 : if (ss->css_offline)
5144 : 11316 : ss->css_offline(css);
5145 : :
5146 : 23440 : css->flags &= ~CSS_ONLINE;
5147 : 23440 : RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5148 : :
5149 : 23440 : wake_up_all(&css->cgroup->offline_waitq);
5150 : : }
5151 : :
5152 : : /**
5153 : : * css_create - create a cgroup_subsys_state
5154 : : * @cgrp: the cgroup new css will be associated with
5155 : : * @ss: the subsys of new css
5156 : : *
5157 : : * Create a new css associated with @cgrp - @ss pair. On success, the new
5158 : : * css is online and installed in @cgrp. This function doesn't create the
5159 : : * interface files. Returns 0 on success, -errno on failure.
5160 : : */
5161 : 46876 : static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5162 : : struct cgroup_subsys *ss)
5163 : : {
5164 : : struct cgroup *parent = cgroup_parent(cgrp);
5165 : : struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5166 : : struct cgroup_subsys_state *css;
5167 : : int err;
5168 : :
5169 : : lockdep_assert_held(&cgroup_mutex);
5170 : :
5171 : 46876 : css = ss->css_alloc(parent_css);
5172 [ - + ]: 46876 : if (!css)
5173 : : css = ERR_PTR(-ENOMEM);
5174 [ + - ]: 46876 : if (IS_ERR(css))
5175 : : return css;
5176 : :
5177 : 46876 : init_and_link_css(css, ss, cgrp);
5178 : :
5179 : 46876 : err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5180 [ + - ]: 46876 : if (err)
5181 : : goto err_free_css;
5182 : :
5183 : 46876 : err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5184 [ + - ]: 46876 : if (err < 0)
5185 : : goto err_free_css;
5186 : 46876 : css->id = err;
5187 : :
5188 : : /* @css is ready to be brought online now, make it visible */
5189 : 46876 : list_add_tail_rcu(&css->sibling, &parent_css->children);
5190 : 46876 : cgroup_idr_replace(&ss->css_idr, css, css->id);
5191 : :
5192 : 46876 : err = online_css(css);
5193 [ + - ]: 46876 : if (err)
5194 : : goto err_list_del;
5195 : :
5196 [ - + # # : 46876 : if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
# # ]
5197 : : cgroup_parent(parent)) {
5198 : 0 : pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5199 : : current->comm, current->pid, ss->name);
5200 [ # # ]: 0 : if (!strcmp(ss->name, "memory"))
5201 : 0 : pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5202 : 0 : ss->warned_broken_hierarchy = true;
5203 : : }
5204 : :
5205 : 46876 : return css;
5206 : :
5207 : : err_list_del:
5208 : : list_del_rcu(&css->sibling);
5209 : : err_free_css:
5210 : : list_del_rcu(&css->rstat_css_node);
5211 : 0 : INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5212 : 0 : queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5213 : 0 : return ERR_PTR(err);
5214 : : }
5215 : :
5216 : : /*
5217 : : * The returned cgroup is fully initialized including its control mask, but
5218 : : * it isn't associated with its kernfs_node and doesn't have the control
5219 : : * mask applied.
5220 : : */
5221 : 105068 : static struct cgroup *cgroup_create(struct cgroup *parent)
5222 : : {
5223 : 105068 : struct cgroup_root *root = parent->root;
5224 : : struct cgroup *cgrp, *tcgrp;
5225 : 105068 : int level = parent->level + 1;
5226 : : int ret;
5227 : :
5228 : : /* allocate the cgroup and its ID, 0 is reserved for the root */
5229 : 105068 : cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5230 : : GFP_KERNEL);
5231 [ + - ]: 105068 : if (!cgrp)
5232 : : return ERR_PTR(-ENOMEM);
5233 : :
5234 : 105068 : ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5235 [ + - ]: 105068 : if (ret)
5236 : : goto out_free_cgrp;
5237 : :
5238 [ + + ]: 105068 : if (cgroup_on_dfl(parent)) {
5239 : 29096 : ret = cgroup_rstat_init(cgrp);
5240 [ + - ]: 29096 : if (ret)
5241 : : goto out_cancel_ref;
5242 : : }
5243 : :
5244 : : /*
5245 : : * Temporarily set the pointer to NULL, so idr_find() won't return
5246 : : * a half-baked cgroup.
5247 : : */
5248 : 105068 : cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
5249 [ + - ]: 105068 : if (cgrp->id < 0) {
5250 : : ret = -ENOMEM;
5251 : : goto out_stat_exit;
5252 : : }
5253 : :
5254 : 105068 : init_cgroup_housekeeping(cgrp);
5255 : :
5256 : 105068 : cgrp->self.parent = &parent->self;
5257 : 105068 : cgrp->root = root;
5258 : 105068 : cgrp->level = level;
5259 : :
5260 : : ret = psi_cgroup_alloc(cgrp);
5261 : : if (ret)
5262 : : goto out_idr_free;
5263 : :
5264 : 105068 : ret = cgroup_bpf_inherit(cgrp);
5265 [ + - ]: 105068 : if (ret)
5266 : : goto out_psi_free;
5267 : :
5268 : : /*
5269 : : * New cgroup inherits effective freeze counter, and
5270 : : * if the parent has to be frozen, the child has too.
5271 : : */
5272 : 105068 : cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5273 [ - + ]: 105068 : if (cgrp->freezer.e_freeze) {
5274 : : /*
5275 : : * Set the CGRP_FREEZE flag, so when a process will be
5276 : : * attached to the child cgroup, it will become frozen.
5277 : : * At this point the new cgroup is unpopulated, so we can
5278 : : * consider it frozen immediately.
5279 : : */
5280 : 0 : set_bit(CGRP_FREEZE, &cgrp->flags);
5281 : 0 : set_bit(CGRP_FROZEN, &cgrp->flags);
5282 : : }
5283 : :
5284 : : spin_lock_irq(&css_set_lock);
5285 [ + + ]: 544340 : for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5286 : 334204 : cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
5287 : :
5288 [ + + ]: 334204 : if (tcgrp != cgrp) {
5289 : 229136 : tcgrp->nr_descendants++;
5290 : :
5291 : : /*
5292 : : * If the new cgroup is frozen, all ancestor cgroups
5293 : : * get a new frozen descendant, but their state can't
5294 : : * change because of this.
5295 : : */
5296 [ - + ]: 229136 : if (cgrp->freezer.e_freeze)
5297 : 0 : tcgrp->freezer.nr_frozen_descendants++;
5298 : : }
5299 : : }
5300 : : spin_unlock_irq(&css_set_lock);
5301 : :
5302 [ - + ]: 105068 : if (notify_on_release(parent))
5303 : 0 : set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5304 : :
5305 [ - + ]: 105068 : if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5306 : 0 : set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5307 : :
5308 : 105068 : cgrp->self.serial_nr = css_serial_nr_next++;
5309 : :
5310 : : /* allocation complete, commit to creation */
5311 : 210136 : list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5312 : 105068 : atomic_inc(&root->nr_cgrps);
5313 : 105068 : cgroup_get_live(parent);
5314 : :
5315 : : /*
5316 : : * @cgrp is now fully operational. If something fails after this
5317 : : * point, it'll be released via the normal destruction path.
5318 : : */
5319 : 105068 : cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
5320 : :
5321 : : /*
5322 : : * On the default hierarchy, a child doesn't automatically inherit
5323 : : * subtree_control from the parent. Each is configured manually.
5324 : : */
5325 [ + + ]: 105068 : if (!cgroup_on_dfl(cgrp))
5326 : 75972 : cgrp->subtree_control = cgroup_control(cgrp);
5327 : :
5328 : 105068 : cgroup_propagate_control(cgrp);
5329 : :
5330 : 105068 : return cgrp;
5331 : :
5332 : : out_psi_free:
5333 : : psi_cgroup_free(cgrp);
5334 : : out_idr_free:
5335 : 0 : cgroup_idr_remove(&root->cgroup_idr, cgrp->id);
5336 : : out_stat_exit:
5337 [ # # ]: 0 : if (cgroup_on_dfl(parent))
5338 : 0 : cgroup_rstat_exit(cgrp);
5339 : : out_cancel_ref:
5340 : 0 : percpu_ref_exit(&cgrp->self.refcnt);
5341 : : out_free_cgrp:
5342 : 0 : kfree(cgrp);
5343 : 0 : return ERR_PTR(ret);
5344 : : }
5345 : :
5346 : : static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5347 : : {
5348 : : struct cgroup *cgroup;
5349 : : int ret = false;
5350 : : int level = 1;
5351 : :
5352 : : lockdep_assert_held(&cgroup_mutex);
5353 : :
5354 [ + + ]: 334204 : for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5355 [ + - ]: 229136 : if (cgroup->nr_descendants >= cgroup->max_descendants)
5356 : : goto fail;
5357 : :
5358 [ + - ]: 229136 : if (level > cgroup->max_depth)
5359 : : goto fail;
5360 : :
5361 : 229136 : level++;
5362 : : }
5363 : :
5364 : : ret = true;
5365 : : fail:
5366 : : return ret;
5367 : : }
5368 : :
5369 : 105068 : int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5370 : : {
5371 : : struct cgroup *parent, *cgrp;
5372 : : struct kernfs_node *kn;
5373 : : int ret;
5374 : :
5375 : : /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5376 [ + - ]: 105068 : if (strchr(name, '\n'))
5377 : : return -EINVAL;
5378 : :
5379 : 105068 : parent = cgroup_kn_lock_live(parent_kn, false);
5380 [ + - ]: 105068 : if (!parent)
5381 : : return -ENODEV;
5382 : :
5383 [ + - ]: 105068 : if (!cgroup_check_hierarchy_limits(parent)) {
5384 : : ret = -EAGAIN;
5385 : : goto out_unlock;
5386 : : }
5387 : :
5388 : 105068 : cgrp = cgroup_create(parent);
5389 [ - + ]: 105068 : if (IS_ERR(cgrp)) {
5390 : : ret = PTR_ERR(cgrp);
5391 : 0 : goto out_unlock;
5392 : : }
5393 : :
5394 : : /* create the directory */
5395 : 105068 : kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5396 [ - + ]: 105068 : if (IS_ERR(kn)) {
5397 : : ret = PTR_ERR(kn);
5398 : 0 : goto out_destroy;
5399 : : }
5400 : 105068 : cgrp->kn = kn;
5401 : :
5402 : : /*
5403 : : * This extra ref will be put in cgroup_free_fn() and guarantees
5404 : : * that @cgrp->kn is always accessible.
5405 : : */
5406 : 105068 : kernfs_get(kn);
5407 : :
5408 : 105068 : ret = cgroup_kn_set_ugid(kn);
5409 [ + - ]: 105068 : if (ret)
5410 : : goto out_destroy;
5411 : :
5412 : 105068 : ret = css_populate_dir(&cgrp->self);
5413 [ + - ]: 105068 : if (ret)
5414 : : goto out_destroy;
5415 : :
5416 : 105068 : ret = cgroup_apply_control_enable(cgrp);
5417 [ + - ]: 105068 : if (ret)
5418 : : goto out_destroy;
5419 : :
5420 [ - + ]: 105068 : TRACE_CGROUP_PATH(mkdir, cgrp);
5421 : :
5422 : : /* let's create and online css's */
5423 : 105068 : kernfs_activate(kn);
5424 : :
5425 : : ret = 0;
5426 : 105068 : goto out_unlock;
5427 : :
5428 : : out_destroy:
5429 : 0 : cgroup_destroy_locked(cgrp);
5430 : : out_unlock:
5431 : 105068 : cgroup_kn_unlock(parent_kn);
5432 : 105068 : return ret;
5433 : : }
5434 : :
5435 : : /*
5436 : : * This is called when the refcnt of a css is confirmed to be killed.
5437 : : * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5438 : : * initate destruction and put the css ref from kill_css().
5439 : : */
5440 : 23440 : static void css_killed_work_fn(struct work_struct *work)
5441 : : {
5442 : : struct cgroup_subsys_state *css =
5443 : 23440 : container_of(work, struct cgroup_subsys_state, destroy_work);
5444 : :
5445 : 23440 : mutex_lock(&cgroup_mutex);
5446 : :
5447 : : do {
5448 : 23440 : offline_css(css);
5449 : : css_put(css);
5450 : : /* @css can't go away while we're holding cgroup_mutex */
5451 : 23440 : css = css->parent;
5452 [ + - - + ]: 46880 : } while (css && atomic_dec_and_test(&css->online_cnt));
5453 : :
5454 : 23440 : mutex_unlock(&cgroup_mutex);
5455 : 23440 : }
5456 : :
5457 : : /* css kill confirmation processing requires process context, bounce */
5458 : 23440 : static void css_killed_ref_fn(struct percpu_ref *ref)
5459 : : {
5460 : : struct cgroup_subsys_state *css =
5461 : : container_of(ref, struct cgroup_subsys_state, refcnt);
5462 : :
5463 [ + - ]: 46880 : if (atomic_dec_and_test(&css->online_cnt)) {
5464 : 46880 : INIT_WORK(&css->destroy_work, css_killed_work_fn);
5465 : 23440 : queue_work(cgroup_destroy_wq, &css->destroy_work);
5466 : : }
5467 : 23440 : }
5468 : :
5469 : : /**
5470 : : * kill_css - destroy a css
5471 : : * @css: css to destroy
5472 : : *
5473 : : * This function initiates destruction of @css by removing cgroup interface
5474 : : * files and putting its base reference. ->css_offline() will be invoked
5475 : : * asynchronously once css_tryget_online() is guaranteed to fail and when
5476 : : * the reference count reaches zero, @css will be released.
5477 : : */
5478 : 23440 : static void kill_css(struct cgroup_subsys_state *css)
5479 : : {
5480 : : lockdep_assert_held(&cgroup_mutex);
5481 : :
5482 [ + - ]: 23440 : if (css->flags & CSS_DYING)
5483 : 23440 : return;
5484 : :
5485 : 23440 : css->flags |= CSS_DYING;
5486 : :
5487 : : /*
5488 : : * This must happen before css is disassociated with its cgroup.
5489 : : * See seq_css() for details.
5490 : : */
5491 : 23440 : css_clear_dir(css);
5492 : :
5493 : : /*
5494 : : * Killing would put the base ref, but we need to keep it alive
5495 : : * until after ->css_offline().
5496 : : */
5497 : : css_get(css);
5498 : :
5499 : : /*
5500 : : * cgroup core guarantees that, by the time ->css_offline() is
5501 : : * invoked, no new css reference will be given out via
5502 : : * css_tryget_online(). We can't simply call percpu_ref_kill() and
5503 : : * proceed to offlining css's because percpu_ref_kill() doesn't
5504 : : * guarantee that the ref is seen as killed on all CPUs on return.
5505 : : *
5506 : : * Use percpu_ref_kill_and_confirm() to get notifications as each
5507 : : * css is confirmed to be seen as killed on all CPUs.
5508 : : */
5509 : 23440 : percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5510 : : }
5511 : :
5512 : : /**
5513 : : * cgroup_destroy_locked - the first stage of cgroup destruction
5514 : : * @cgrp: cgroup to be destroyed
5515 : : *
5516 : : * css's make use of percpu refcnts whose killing latency shouldn't be
5517 : : * exposed to userland and are RCU protected. Also, cgroup core needs to
5518 : : * guarantee that css_tryget_online() won't succeed by the time
5519 : : * ->css_offline() is invoked. To satisfy all the requirements,
5520 : : * destruction is implemented in the following two steps.
5521 : : *
5522 : : * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5523 : : * userland visible parts and start killing the percpu refcnts of
5524 : : * css's. Set up so that the next stage will be kicked off once all
5525 : : * the percpu refcnts are confirmed to be killed.
5526 : : *
5527 : : * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5528 : : * rest of destruction. Once all cgroup references are gone, the
5529 : : * cgroup is RCU-freed.
5530 : : *
5531 : : * This function implements s1. After this step, @cgrp is gone as far as
5532 : : * the userland is concerned and a new cgroup with the same name may be
5533 : : * created. As cgroup doesn't care about the names internally, this
5534 : : * doesn't cause any problem.
5535 : : */
5536 : 47688 : static int cgroup_destroy_locked(struct cgroup *cgrp)
5537 : : __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5538 : : {
5539 : : struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5540 : : struct cgroup_subsys_state *css;
5541 : : struct cgrp_cset_link *link;
5542 : : int ssid;
5543 : :
5544 : : lockdep_assert_held(&cgroup_mutex);
5545 : :
5546 : : /*
5547 : : * Only migration can raise populated from zero and we're already
5548 : : * holding cgroup_mutex.
5549 : : */
5550 [ + - ]: 47688 : if (cgroup_is_populated(cgrp))
5551 : : return -EBUSY;
5552 : :
5553 : : /*
5554 : : * Make sure there's no live children. We can't test emptiness of
5555 : : * ->self.children as dead children linger on it while being
5556 : : * drained; otherwise, "rmdir parent/child parent" may fail.
5557 : : */
5558 [ + - ]: 47688 : if (css_has_online_children(&cgrp->self))
5559 : : return -EBUSY;
5560 : :
5561 : : /*
5562 : : * Mark @cgrp and the associated csets dead. The former prevents
5563 : : * further task migration and child creation by disabling
5564 : : * cgroup_lock_live_group(). The latter makes the csets ignored by
5565 : : * the migration path.
5566 : : */
5567 : 47688 : cgrp->self.flags &= ~CSS_ONLINE;
5568 : :
5569 : : spin_lock_irq(&css_set_lock);
5570 [ + + ]: 95376 : list_for_each_entry(link, &cgrp->cset_links, cset_link)
5571 : 47688 : link->cset->dead = true;
5572 : : spin_unlock_irq(&css_set_lock);
5573 : :
5574 : : /* initiate massacre of all css's */
5575 [ + + + + ]: 1096824 : for_each_css(css, ssid, cgrp)
5576 : 23440 : kill_css(css);
5577 : :
5578 : : /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5579 : 47688 : css_clear_dir(&cgrp->self);
5580 : 47688 : kernfs_remove(cgrp->kn);
5581 : :
5582 [ + - - + ]: 95376 : if (parent && cgroup_is_threaded(cgrp))
5583 : 0 : parent->nr_threaded_children--;
5584 : :
5585 : : spin_lock_irq(&css_set_lock);
5586 [ + + ]: 193176 : for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5587 : 97800 : tcgrp->nr_descendants--;
5588 : 97800 : tcgrp->nr_dying_descendants++;
5589 : : /*
5590 : : * If the dying cgroup is frozen, decrease frozen descendants
5591 : : * counters of ancestor cgroups.
5592 : : */
5593 [ - + ]: 97800 : if (test_bit(CGRP_FROZEN, &cgrp->flags))
5594 : 0 : tcgrp->freezer.nr_frozen_descendants--;
5595 : : }
5596 : : spin_unlock_irq(&css_set_lock);
5597 : :
5598 : 47688 : cgroup1_check_for_release(parent);
5599 : :
5600 : 47688 : cgroup_bpf_offline(cgrp);
5601 : :
5602 : : /* put the base reference */
5603 : 47688 : percpu_ref_kill(&cgrp->self.refcnt);
5604 : :
5605 : 47688 : return 0;
5606 : : };
5607 : :
5608 : 47688 : int cgroup_rmdir(struct kernfs_node *kn)
5609 : : {
5610 : : struct cgroup *cgrp;
5611 : : int ret = 0;
5612 : :
5613 : 47688 : cgrp = cgroup_kn_lock_live(kn, false);
5614 [ + - ]: 47688 : if (!cgrp)
5615 : : return 0;
5616 : :
5617 : 47688 : ret = cgroup_destroy_locked(cgrp);
5618 [ + - ]: 47688 : if (!ret)
5619 [ - + ]: 47688 : TRACE_CGROUP_PATH(rmdir, cgrp);
5620 : :
5621 : 47688 : cgroup_kn_unlock(kn);
5622 : 47688 : return ret;
5623 : : }
5624 : :
5625 : : static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5626 : : .show_options = cgroup_show_options,
5627 : : .mkdir = cgroup_mkdir,
5628 : : .rmdir = cgroup_rmdir,
5629 : : .show_path = cgroup_show_path,
5630 : : };
5631 : :
5632 : 4444 : static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5633 : : {
5634 : : struct cgroup_subsys_state *css;
5635 : :
5636 : : pr_debug("Initializing cgroup subsys %s\n", ss->name);
5637 : :
5638 : 4444 : mutex_lock(&cgroup_mutex);
5639 : :
5640 : : idr_init(&ss->css_idr);
5641 : 4444 : INIT_LIST_HEAD(&ss->cfts);
5642 : :
5643 : : /* Create the root cgroup state for this subsystem */
5644 : 4444 : ss->root = &cgrp_dfl_root;
5645 : 8888 : css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5646 : : /* We don't handle early failures gracefully */
5647 [ - + ]: 4444 : BUG_ON(IS_ERR(css));
5648 : 4444 : init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5649 : :
5650 : : /*
5651 : : * Root csses are never destroyed and we can't initialize
5652 : : * percpu_ref during early init. Disable refcnting.
5653 : : */
5654 : 4444 : css->flags |= CSS_NO_REF;
5655 : :
5656 [ + + ]: 4444 : if (early) {
5657 : : /* allocation can't be done safely during early init */
5658 : 1212 : css->id = 1;
5659 : : } else {
5660 : 3232 : css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5661 [ - + ]: 3232 : BUG_ON(css->id < 0);
5662 : : }
5663 : :
5664 : : /* Update the init_css_set to contain a subsys
5665 : : * pointer to this state - since the subsystem is
5666 : : * newly registered, all tasks and hence the
5667 : : * init_css_set is in the subsystem's root cgroup. */
5668 : 4444 : init_css_set.subsys[ss->id] = css;
5669 : :
5670 : 4444 : have_fork_callback |= (bool)ss->fork << ss->id;
5671 : 4444 : have_exit_callback |= (bool)ss->exit << ss->id;
5672 : 4444 : have_release_callback |= (bool)ss->release << ss->id;
5673 : 4444 : have_canfork_callback |= (bool)ss->can_fork << ss->id;
5674 : :
5675 : : /* At system boot, before all subsystems have been
5676 : : * registered, no tasks have been forked, so we don't
5677 : : * need to invoke fork callbacks here. */
5678 [ - + ]: 4444 : BUG_ON(!list_empty(&init_task.tasks));
5679 : :
5680 [ - + ]: 4444 : BUG_ON(online_css(css));
5681 : :
5682 : 4444 : mutex_unlock(&cgroup_mutex);
5683 : 4444 : }
5684 : :
5685 : : /**
5686 : : * cgroup_init_early - cgroup initialization at system boot
5687 : : *
5688 : : * Initialize cgroups at system boot, and initialize any
5689 : : * subsystems that request early init.
5690 : : */
5691 : 404 : int __init cgroup_init_early(void)
5692 : : {
5693 : : static struct cgroup_fs_context __initdata ctx;
5694 : : struct cgroup_subsys *ss;
5695 : : int i;
5696 : :
5697 : 404 : ctx.root = &cgrp_dfl_root;
5698 : 404 : init_cgroup_root(&ctx);
5699 : 404 : cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5700 : :
5701 : 404 : RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5702 : :
5703 [ + + ]: 4848 : for_each_subsys(ss, i) {
5704 [ + - + - : 8888 : WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
+ - - + +
- ]
5705 : : "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5706 : : i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5707 : : ss->id, ss->name);
5708 [ - + ]: 4444 : WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5709 : : "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5710 : :
5711 : 4444 : ss->id = i;
5712 : 4444 : ss->name = cgroup_subsys_name[i];
5713 [ + + ]: 4444 : if (!ss->legacy_name)
5714 : 4040 : ss->legacy_name = cgroup_subsys_name[i];
5715 : :
5716 [ + + ]: 4444 : if (ss->early_init)
5717 : 1212 : cgroup_init_subsys(ss, true);
5718 : : }
5719 : 404 : return 0;
5720 : : }
5721 : :
5722 : : static u16 cgroup_disable_mask __initdata;
5723 : : static u16 cgroup_enable_mask __initdata;
5724 : : static int __init cgroup_disable(char *str);
5725 : :
5726 : : /**
5727 : : * cgroup_init - cgroup initialization
5728 : : *
5729 : : * Register cgroup filesystem and /proc file, and initialize
5730 : : * any subsystems that didn't request early init.
5731 : : */
5732 : 404 : int __init cgroup_init(void)
5733 : : {
5734 : : struct cgroup_subsys *ss;
5735 : : int ssid;
5736 : :
5737 : : BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5738 [ - + ]: 404 : BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5739 [ - + ]: 404 : BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5740 : :
5741 : 404 : cgroup_rstat_boot();
5742 : :
5743 : : /*
5744 : : * The latency of the synchronize_rcu() is too high for cgroups,
5745 : : * avoid it at the cost of forcing all readers into the slow path.
5746 : : */
5747 : 404 : rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5748 : :
5749 : 404 : get_user_ns(init_cgroup_ns.user_ns);
5750 : :
5751 : 404 : mutex_lock(&cgroup_mutex);
5752 : :
5753 : : /*
5754 : : * Add init_css_set to the hash table so that dfl_root can link to
5755 : : * it during init.
5756 : : */
5757 : 404 : hash_add(css_set_table, &init_css_set.hlist,
5758 : : css_set_hash(init_css_set.subsys));
5759 : :
5760 [ - + ]: 404 : BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5761 : :
5762 : 404 : mutex_unlock(&cgroup_mutex);
5763 : :
5764 : : /* Apply an implicit disable... */
5765 : 404 : cgroup_disable("memory");
5766 : :
5767 : : /* ...knowing that an explicit enable will override it. */
5768 : 404 : cgroup_disable_mask &= ~cgroup_enable_mask;
5769 : :
5770 [ + + ]: 4848 : for_each_subsys(ss, ssid) {
5771 [ + + ]: 4444 : if (ss->early_init) {
5772 : 1212 : struct cgroup_subsys_state *css =
5773 : 1212 : init_css_set.subsys[ss->id];
5774 : :
5775 : 1212 : css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5776 : : GFP_KERNEL);
5777 [ - + ]: 1212 : BUG_ON(css->id < 0);
5778 : : } else {
5779 : 3232 : cgroup_init_subsys(ss, false);
5780 : : }
5781 : :
5782 : 4444 : list_add_tail(&init_css_set.e_cset_node[ssid],
5783 : : &cgrp_dfl_root.cgrp.e_csets[ssid]);
5784 : :
5785 : : /*
5786 : : * Setting dfl_root subsys_mask needs to consider the
5787 : : * disabled flag and cftype registration needs kmalloc,
5788 : : * both of which aren't available during early_init.
5789 : : */
5790 [ + + ]: 4444 : if (cgroup_disable_mask & (1 << ssid)) {
5791 : 404 : static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5792 : 404 : printk(KERN_INFO "Disabling %s control group subsystem\n",
5793 : : ss->name);
5794 : 404 : continue;
5795 : : }
5796 : :
5797 [ - + ]: 4040 : if (cgroup1_ssid_disabled(ssid))
5798 : 0 : printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5799 : : ss->name);
5800 : :
5801 : 4040 : cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5802 : :
5803 : : /* implicit controllers must be threaded too */
5804 [ + + + - : 4040 : WARN_ON(ss->implicit_on_dfl && !ss->threaded);
- + ]
5805 : :
5806 [ + + ]: 4040 : if (ss->implicit_on_dfl)
5807 : 404 : cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5808 [ + + ]: 3636 : else if (!ss->dfl_cftypes)
5809 : 2020 : cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5810 : :
5811 [ + + ]: 4040 : if (ss->threaded)
5812 : 1616 : cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5813 : :
5814 [ + + ]: 4040 : if (ss->dfl_cftypes == ss->legacy_cftypes) {
5815 [ - + ]: 808 : WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5816 : : } else {
5817 [ - + ]: 3232 : WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5818 [ - + ]: 3232 : WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5819 : : }
5820 : :
5821 [ + + ]: 4040 : if (ss->bind)
5822 : 808 : ss->bind(init_css_set.subsys[ssid]);
5823 : :
5824 : 4040 : mutex_lock(&cgroup_mutex);
5825 : 4040 : css_populate_dir(init_css_set.subsys[ssid]);
5826 : 4040 : mutex_unlock(&cgroup_mutex);
5827 : : }
5828 : :
5829 : : /* init_css_set.subsys[] has been updated, re-hash */
5830 : : hash_del(&init_css_set.hlist);
5831 : 404 : hash_add(css_set_table, &init_css_set.hlist,
5832 : : css_set_hash(init_css_set.subsys));
5833 : :
5834 [ - + ]: 404 : WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5835 [ - + ]: 404 : WARN_ON(register_filesystem(&cgroup_fs_type));
5836 [ - + ]: 404 : WARN_ON(register_filesystem(&cgroup2_fs_type));
5837 [ - + ]: 404 : WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5838 : : #ifdef CONFIG_CPUSETS
5839 [ - + ]: 404 : WARN_ON(register_filesystem(&cpuset_fs_type));
5840 : : #endif
5841 : :
5842 : 404 : return 0;
5843 : : }
5844 : :
5845 : 404 : static int __init cgroup_wq_init(void)
5846 : : {
5847 : : /*
5848 : : * There isn't much point in executing destruction path in
5849 : : * parallel. Good chunk is serialized with cgroup_mutex anyway.
5850 : : * Use 1 for @max_active.
5851 : : *
5852 : : * We would prefer to do this in cgroup_init() above, but that
5853 : : * is called before init_workqueues(): so leave this until after.
5854 : : */
5855 : 404 : cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5856 [ - + ]: 404 : BUG_ON(!cgroup_destroy_wq);
5857 : 404 : return 0;
5858 : : }
5859 : : core_initcall(cgroup_wq_init);
5860 : :
5861 : 0 : void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
5862 : : char *buf, size_t buflen)
5863 : : {
5864 : : struct kernfs_node *kn;
5865 : :
5866 : 0 : kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id);
5867 [ # # ]: 0 : if (!kn)
5868 : 0 : return;
5869 : : kernfs_path(kn, buf, buflen);
5870 : 0 : kernfs_put(kn);
5871 : : }
5872 : :
5873 : : /*
5874 : : * proc_cgroup_show()
5875 : : * - Print task's cgroup paths into seq_file, one line for each hierarchy
5876 : : * - Used for /proc/<pid>/cgroup.
5877 : : */
5878 : 112806 : int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5879 : : struct pid *pid, struct task_struct *tsk)
5880 : : {
5881 : : char *buf;
5882 : : int retval;
5883 : : struct cgroup_root *root;
5884 : :
5885 : : retval = -ENOMEM;
5886 : : buf = kmalloc(PATH_MAX, GFP_KERNEL);
5887 [ + - ]: 112806 : if (!buf)
5888 : : goto out;
5889 : :
5890 : 112806 : mutex_lock(&cgroup_mutex);
5891 : : spin_lock_irq(&css_set_lock);
5892 : :
5893 [ + + ]: 1237634 : for_each_root(root) {
5894 : : struct cgroup_subsys *ss;
5895 : : struct cgroup *cgrp;
5896 : : int ssid, count = 0;
5897 : :
5898 [ + + - + ]: 1124828 : if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5899 : 0 : continue;
5900 : :
5901 : 1124828 : seq_printf(m, "%d:", root->hierarchy_id);
5902 [ + + ]: 1124828 : if (root != &cgrp_dfl_root)
5903 [ + + ]: 11132242 : for_each_subsys(ss, ssid)
5904 [ + + ]: 11132242 : if (root->subsys_mask & (1 << ssid))
5905 [ + + ]: 1124020 : seq_printf(m, "%s%s", count++ ? "," : "",
5906 : : ss->legacy_name);
5907 [ + + ]: 1124828 : if (strlen(root->name))
5908 [ + - ]: 225612 : seq_printf(m, "%sname=%s", count ? "," : "",
5909 : 112806 : root->name);
5910 : 1124828 : seq_putc(m, ':');
5911 : :
5912 : : cgrp = task_cgroup_from_root(tsk, root);
5913 : :
5914 : : /*
5915 : : * On traditional hierarchies, all zombie tasks show up as
5916 : : * belonging to the root cgroup. On the default hierarchy,
5917 : : * while a zombie doesn't show up in "cgroup.procs" and
5918 : : * thus can't be migrated, its /proc/PID/cgroup keeps
5919 : : * reporting the cgroup it belonged to before exiting. If
5920 : : * the cgroup is removed before the zombie is reaped,
5921 : : * " (deleted)" is appended to the cgroup path.
5922 : : */
5923 [ + + + + ]: 1124828 : if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5924 : 937488 : retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5925 : 937488 : current->nsproxy->cgroup_ns);
5926 [ - + ]: 937488 : if (retval >= PATH_MAX)
5927 : : retval = -ENAMETOOLONG;
5928 [ + - ]: 937488 : if (retval < 0)
5929 : : goto out_unlock;
5930 : :
5931 : 937488 : seq_puts(m, buf);
5932 : : } else {
5933 : 187340 : seq_puts(m, "/");
5934 : : }
5935 : :
5936 [ + + + + ]: 1237634 : if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5937 : 94 : seq_puts(m, " (deleted)\n");
5938 : : else
5939 : 1124734 : seq_putc(m, '\n');
5940 : : }
5941 : :
5942 : : retval = 0;
5943 : : out_unlock:
5944 : : spin_unlock_irq(&css_set_lock);
5945 : 112806 : mutex_unlock(&cgroup_mutex);
5946 : 112806 : kfree(buf);
5947 : : out:
5948 : 112806 : return retval;
5949 : : }
5950 : :
5951 : : /**
5952 : : * cgroup_fork - initialize cgroup related fields during copy_process()
5953 : : * @child: pointer to task_struct of forking parent process.
5954 : : *
5955 : : * A task is associated with the init_css_set until cgroup_post_fork()
5956 : : * attaches it to the parent's css_set. Empty cg_list indicates that
5957 : : * @child isn't holding reference to its css_set.
5958 : : */
5959 : 459424 : void cgroup_fork(struct task_struct *child)
5960 : : {
5961 : 459424 : RCU_INIT_POINTER(child->cgroups, &init_css_set);
5962 : 459424 : INIT_LIST_HEAD(&child->cg_list);
5963 : 459424 : }
5964 : :
5965 : : /**
5966 : : * cgroup_can_fork - called on a new task before the process is exposed
5967 : : * @child: the task in question.
5968 : : *
5969 : : * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5970 : : * returns an error, the fork aborts with that error code. This allows for
5971 : : * a cgroup subsystem to conditionally allow or deny new forks.
5972 : : */
5973 : 459424 : int cgroup_can_fork(struct task_struct *child)
5974 : : {
5975 : : struct cgroup_subsys *ss;
5976 : : int i, j, ret;
5977 : :
5978 [ + + ]: 918848 : do_each_subsys_mask(ss, i, have_canfork_callback) {
5979 : 459418 : ret = ss->can_fork(child);
5980 [ + - ]: 459424 : if (ret)
5981 : : goto out_revert;
5982 : : } while_each_subsys_mask();
5983 : :
5984 : 459424 : return 0;
5985 : :
5986 : : out_revert:
5987 [ # # ]: 0 : for_each_subsys(ss, j) {
5988 [ # # ]: 0 : if (j >= i)
5989 : : break;
5990 [ # # ]: 0 : if (ss->cancel_fork)
5991 : 0 : ss->cancel_fork(child);
5992 : : }
5993 : :
5994 : 0 : return ret;
5995 : : }
5996 : :
5997 : : /**
5998 : : * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5999 : : * @child: the task in question
6000 : : *
6001 : : * This calls the cancel_fork() callbacks if a fork failed *after*
6002 : : * cgroup_can_fork() succeded.
6003 : : */
6004 : 0 : void cgroup_cancel_fork(struct task_struct *child)
6005 : : {
6006 : : struct cgroup_subsys *ss;
6007 : : int i;
6008 : :
6009 [ # # ]: 0 : for_each_subsys(ss, i)
6010 [ # # ]: 0 : if (ss->cancel_fork)
6011 : 0 : ss->cancel_fork(child);
6012 : 0 : }
6013 : :
6014 : : /**
6015 : : * cgroup_post_fork - called on a new task after adding it to the task list
6016 : : * @child: the task in question
6017 : : *
6018 : : * Adds the task to the list running through its css_set if necessary and
6019 : : * call the subsystem fork() callbacks. Has to be after the task is
6020 : : * visible on the task list in case we race with the first call to
6021 : : * cgroup_task_iter_start() - to guarantee that the new task ends up on its
6022 : : * list.
6023 : : */
6024 : 459424 : void cgroup_post_fork(struct task_struct *child)
6025 : : {
6026 : : struct cgroup_subsys *ss;
6027 : : int i;
6028 : :
6029 : : /*
6030 : : * This may race against cgroup_enable_task_cg_lists(). As that
6031 : : * function sets use_task_css_set_links before grabbing
6032 : : * tasklist_lock and we just went through tasklist_lock to add
6033 : : * @child, it's guaranteed that either we see the set
6034 : : * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
6035 : : * @child during its iteration.
6036 : : *
6037 : : * If we won the race, @child is associated with %current's
6038 : : * css_set. Grabbing css_set_lock guarantees both that the
6039 : : * association is stable, and, on completion of the parent's
6040 : : * migration, @child is visible in the source of migration or
6041 : : * already in the destination cgroup. This guarantee is necessary
6042 : : * when implementing operations which need to migrate all tasks of
6043 : : * a cgroup to another.
6044 : : *
6045 : : * Note that if we lose to cgroup_enable_task_cg_lists(), @child
6046 : : * will remain in init_css_set. This is safe because all tasks are
6047 : : * in the init_css_set before cg_links is enabled and there's no
6048 : : * operation which transfers all tasks out of init_css_set.
6049 : : */
6050 [ + + ]: 459424 : if (use_task_css_set_links) {
6051 : : struct css_set *cset;
6052 : :
6053 : : spin_lock_irq(&css_set_lock);
6054 : 425566 : cset = task_css_set(current);
6055 [ + - ]: 851132 : if (list_empty(&child->cg_list)) {
6056 : : get_css_set(cset);
6057 : 425566 : cset->nr_tasks++;
6058 : 425566 : css_set_move_task(child, NULL, cset, false);
6059 : : }
6060 : :
6061 : : /*
6062 : : * If the cgroup has to be frozen, the new task has too.
6063 : : * Let's set the JOBCTL_TRAP_FREEZE jobctl bit to get
6064 : : * the task into the frozen state.
6065 : : */
6066 [ - + ]: 425566 : if (unlikely(cgroup_task_freeze(child))) {
6067 : 0 : spin_lock(&child->sighand->siglock);
6068 [ # # # # ]: 0 : WARN_ON_ONCE(child->frozen);
6069 : 0 : child->jobctl |= JOBCTL_TRAP_FREEZE;
6070 : 0 : spin_unlock(&child->sighand->siglock);
6071 : :
6072 : : /*
6073 : : * Calling cgroup_update_frozen() isn't required here,
6074 : : * because it will be called anyway a bit later
6075 : : * from do_freezer_trap(). So we avoid cgroup's
6076 : : * transient switch from the frozen state and back.
6077 : : */
6078 : : }
6079 : :
6080 : : spin_unlock_irq(&css_set_lock);
6081 : : }
6082 : :
6083 : : /*
6084 : : * Call ss->fork(). This must happen after @child is linked on
6085 : : * css_set; otherwise, @child might change state between ->fork()
6086 : : * and addition to css_set.
6087 : : */
6088 [ + + ]: 1837696 : do_each_subsys_mask(ss, i, have_fork_callback) {
6089 : 1378268 : ss->fork(child);
6090 : : } while_each_subsys_mask();
6091 : 459418 : }
6092 : :
6093 : : /**
6094 : : * cgroup_exit - detach cgroup from exiting task
6095 : : * @tsk: pointer to task_struct of exiting process
6096 : : *
6097 : : * Description: Detach cgroup from @tsk and release it.
6098 : : *
6099 : : * Note that cgroups marked notify_on_release force every task in
6100 : : * them to take the global cgroup_mutex mutex when exiting.
6101 : : * This could impact scaling on very large systems. Be reluctant to
6102 : : * use notify_on_release cgroups where very high task exit scaling
6103 : : * is required on large systems.
6104 : : *
6105 : : * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
6106 : : * call cgroup_exit() while the task is still competent to handle
6107 : : * notify_on_release(), then leave the task attached to the root cgroup in
6108 : : * each hierarchy for the remainder of its exit. No need to bother with
6109 : : * init_css_set refcnting. init_css_set never goes away and we can't race
6110 : : * with migration path - PF_EXITING is visible to migration path.
6111 : : */
6112 : 386798 : void cgroup_exit(struct task_struct *tsk)
6113 : : {
6114 : : struct cgroup_subsys *ss;
6115 : : struct css_set *cset;
6116 : : int i;
6117 : :
6118 : : /*
6119 : : * Unlink from @tsk from its css_set. As migration path can't race
6120 : : * with us, we can check css_set and cg_list without synchronization.
6121 : : */
6122 : : cset = task_css_set(tsk);
6123 : :
6124 [ + + ]: 773596 : if (!list_empty(&tsk->cg_list)) {
6125 : : spin_lock_irq(&css_set_lock);
6126 : 379964 : css_set_move_task(tsk, cset, NULL, false);
6127 : 379964 : list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6128 : 379964 : cset->nr_tasks--;
6129 : :
6130 [ - + # # ]: 379964 : WARN_ON_ONCE(cgroup_task_frozen(tsk));
6131 [ - + ]: 379964 : if (unlikely(cgroup_task_freeze(tsk)))
6132 : 0 : cgroup_update_frozen(task_dfl_cgroup(tsk));
6133 : :
6134 : : spin_unlock_irq(&css_set_lock);
6135 : : } else {
6136 : : get_css_set(cset);
6137 : : }
6138 : :
6139 : : /* see cgroup_post_fork() for details */
6140 [ + + ]: 773650 : do_each_subsys_mask(ss, i, have_exit_callback) {
6141 : 386722 : ss->exit(tsk);
6142 : : } while_each_subsys_mask();
6143 : 386822 : }
6144 : :
6145 : 386832 : void cgroup_release(struct task_struct *task)
6146 : : {
6147 : : struct cgroup_subsys *ss;
6148 : : int ssid;
6149 : :
6150 [ + + ]: 773640 : do_each_subsys_mask(ss, ssid, have_release_callback) {
6151 : 386786 : ss->release(task);
6152 : : } while_each_subsys_mask();
6153 : :
6154 [ + + ]: 386814 : if (use_task_css_set_links) {
6155 : : spin_lock_irq(&css_set_lock);
6156 : 379964 : css_set_skip_task_iters(task_css_set(task), task);
6157 : 379964 : list_del_init(&task->cg_list);
6158 : : spin_unlock_irq(&css_set_lock);
6159 : : }
6160 : 386816 : }
6161 : :
6162 : 385550 : void cgroup_free(struct task_struct *task)
6163 : : {
6164 : : struct css_set *cset = task_css_set(task);
6165 : 385550 : put_css_set(cset);
6166 : 385902 : }
6167 : :
6168 : 404 : static int __init cgroup_disable(char *str)
6169 : : {
6170 : : struct cgroup_subsys *ss;
6171 : : char *token;
6172 : : int i;
6173 : :
6174 [ + + ]: 1212 : while ((token = strsep(&str, ",")) != NULL) {
6175 [ + - ]: 404 : if (!*token)
6176 : 0 : continue;
6177 : :
6178 [ + + ]: 4444 : for_each_subsys(ss, i) {
6179 [ + - + + ]: 8484 : if (strcmp(token, ss->name) &&
6180 : 4040 : strcmp(token, ss->legacy_name))
6181 : 4040 : continue;
6182 : 404 : cgroup_disable_mask |= 1 << i;
6183 : : }
6184 : : }
6185 : 404 : return 1;
6186 : : }
6187 : : __setup("cgroup_disable=", cgroup_disable);
6188 : :
6189 : 0 : static int __init cgroup_enable(char *str)
6190 : : {
6191 : : struct cgroup_subsys *ss;
6192 : : char *token;
6193 : : int i;
6194 : :
6195 [ # # ]: 0 : while ((token = strsep(&str, ",")) != NULL) {
6196 [ # # ]: 0 : if (!*token)
6197 : 0 : continue;
6198 : :
6199 [ # # ]: 0 : for_each_subsys(ss, i) {
6200 [ # # # # ]: 0 : if (strcmp(token, ss->name) &&
6201 : 0 : strcmp(token, ss->legacy_name))
6202 : 0 : continue;
6203 : :
6204 : 0 : cgroup_enable_mask |= 1 << i;
6205 : : }
6206 : : }
6207 : 0 : return 1;
6208 : : }
6209 : : __setup("cgroup_enable=", cgroup_enable);
6210 : :
6211 : 0 : void __init __weak enable_debug_cgroup(void) { }
6212 : :
6213 : 0 : static int __init enable_cgroup_debug(char *str)
6214 : : {
6215 : 0 : cgroup_debug = true;
6216 : 0 : enable_debug_cgroup();
6217 : 0 : return 1;
6218 : : }
6219 : : __setup("cgroup_debug", enable_cgroup_debug);
6220 : :
6221 : : /**
6222 : : * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6223 : : * @dentry: directory dentry of interest
6224 : : * @ss: subsystem of interest
6225 : : *
6226 : : * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6227 : : * to get the corresponding css and return it. If such css doesn't exist
6228 : : * or can't be pinned, an ERR_PTR value is returned.
6229 : : */
6230 : 2424 : struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6231 : : struct cgroup_subsys *ss)
6232 : : {
6233 : 2424 : struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6234 : 2424 : struct file_system_type *s_type = dentry->d_sb->s_type;
6235 : : struct cgroup_subsys_state *css = NULL;
6236 : : struct cgroup *cgrp;
6237 : :
6238 : : /* is @dentry a cgroup dir? */
6239 [ + - + - : 2424 : if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
+ - ]
6240 [ + - ]: 2424 : !kn || kernfs_type(kn) != KERNFS_DIR)
6241 : : return ERR_PTR(-EBADF);
6242 : :
6243 : : rcu_read_lock();
6244 : :
6245 : : /*
6246 : : * This path doesn't originate from kernfs and @kn could already
6247 : : * have been or be removed at any point. @kn->priv is RCU
6248 : : * protected for this access. See css_release_work_fn() for details.
6249 : : */
6250 : : cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6251 [ + - ]: 2424 : if (cgrp)
6252 : : css = cgroup_css(cgrp, ss);
6253 : :
6254 [ + - - + ]: 4848 : if (!css || !css_tryget_online(css))
6255 : : css = ERR_PTR(-ENOENT);
6256 : :
6257 : : rcu_read_unlock();
6258 : 2424 : return css;
6259 : : }
6260 : :
6261 : : /**
6262 : : * css_from_id - lookup css by id
6263 : : * @id: the cgroup id
6264 : : * @ss: cgroup subsys to be looked into
6265 : : *
6266 : : * Returns the css if there's valid one with @id, otherwise returns NULL.
6267 : : * Should be called under rcu_read_lock().
6268 : : */
6269 : 0 : struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6270 : : {
6271 : : WARN_ON_ONCE(!rcu_read_lock_held());
6272 : 0 : return idr_find(&ss->css_idr, id);
6273 : : }
6274 : :
6275 : : /**
6276 : : * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6277 : : * @path: path on the default hierarchy
6278 : : *
6279 : : * Find the cgroup at @path on the default hierarchy, increment its
6280 : : * reference count and return it. Returns pointer to the found cgroup on
6281 : : * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6282 : : * if @path points to a non-directory.
6283 : : */
6284 : 0 : struct cgroup *cgroup_get_from_path(const char *path)
6285 : : {
6286 : : struct kernfs_node *kn;
6287 : : struct cgroup *cgrp;
6288 : :
6289 : 0 : mutex_lock(&cgroup_mutex);
6290 : :
6291 : 0 : kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6292 [ # # ]: 0 : if (kn) {
6293 [ # # ]: 0 : if (kernfs_type(kn) == KERNFS_DIR) {
6294 : 0 : cgrp = kn->priv;
6295 : 0 : cgroup_get_live(cgrp);
6296 : : } else {
6297 : : cgrp = ERR_PTR(-ENOTDIR);
6298 : : }
6299 : 0 : kernfs_put(kn);
6300 : : } else {
6301 : : cgrp = ERR_PTR(-ENOENT);
6302 : : }
6303 : :
6304 : 0 : mutex_unlock(&cgroup_mutex);
6305 : 0 : return cgrp;
6306 : : }
6307 : : EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6308 : :
6309 : : /**
6310 : : * cgroup_get_from_fd - get a cgroup pointer from a fd
6311 : : * @fd: fd obtained by open(cgroup2_dir)
6312 : : *
6313 : : * Find the cgroup from a fd which should be obtained
6314 : : * by opening a cgroup directory. Returns a pointer to the
6315 : : * cgroup on success. ERR_PTR is returned if the cgroup
6316 : : * cannot be found.
6317 : : */
6318 : 2828 : struct cgroup *cgroup_get_from_fd(int fd)
6319 : : {
6320 : : struct cgroup_subsys_state *css;
6321 : : struct cgroup *cgrp;
6322 : : struct file *f;
6323 : :
6324 : 2828 : f = fget_raw(fd);
6325 [ + + ]: 2828 : if (!f)
6326 : : return ERR_PTR(-EBADF);
6327 : :
6328 : 2424 : css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6329 : 2424 : fput(f);
6330 [ + - ]: 2424 : if (IS_ERR(css))
6331 : : return ERR_CAST(css);
6332 : :
6333 : 2424 : cgrp = css->cgroup;
6334 [ - + ]: 2424 : if (!cgroup_on_dfl(cgrp)) {
6335 : : cgroup_put(cgrp);
6336 : : return ERR_PTR(-EBADF);
6337 : : }
6338 : :
6339 : : return cgrp;
6340 : : }
6341 : : EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6342 : :
6343 : : static u64 power_of_ten(int power)
6344 : : {
6345 : : u64 v = 1;
6346 [ # # # # : 0 : while (power--)
# # ]
6347 : 0 : v *= 10;
6348 : 0 : return v;
6349 : : }
6350 : :
6351 : : /**
6352 : : * cgroup_parse_float - parse a floating number
6353 : : * @input: input string
6354 : : * @dec_shift: number of decimal digits to shift
6355 : : * @v: output
6356 : : *
6357 : : * Parse a decimal floating point number in @input and store the result in
6358 : : * @v with decimal point right shifted @dec_shift times. For example, if
6359 : : * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6360 : : * Returns 0 on success, -errno otherwise.
6361 : : *
6362 : : * There's nothing cgroup specific about this function except that it's
6363 : : * currently the only user.
6364 : : */
6365 : 0 : int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6366 : : {
6367 : 0 : s64 whole, frac = 0;
6368 : 0 : int fstart = 0, fend = 0, flen;
6369 : :
6370 [ # # ]: 0 : if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6371 : : return -EINVAL;
6372 [ # # ]: 0 : if (frac < 0)
6373 : : return -EINVAL;
6374 : :
6375 [ # # ]: 0 : flen = fend > fstart ? fend - fstart : 0;
6376 [ # # ]: 0 : if (flen < dec_shift)
6377 : 0 : frac *= power_of_ten(dec_shift - flen);
6378 : : else
6379 [ # # # # : 0 : frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # ]
6380 : :
6381 : 0 : *v = whole * power_of_ten(dec_shift) + frac;
6382 : 0 : return 0;
6383 : : }
6384 : :
6385 : : /*
6386 : : * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6387 : : * definition in cgroup-defs.h.
6388 : : */
6389 : : #ifdef CONFIG_SOCK_CGROUP_DATA
6390 : :
6391 : : #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6392 : :
6393 : : DEFINE_SPINLOCK(cgroup_sk_update_lock);
6394 : : static bool cgroup_sk_alloc_disabled __read_mostly;
6395 : :
6396 : 0 : void cgroup_sk_alloc_disable(void)
6397 : : {
6398 [ # # ]: 0 : if (cgroup_sk_alloc_disabled)
6399 : 0 : return;
6400 : 0 : pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6401 : 0 : cgroup_sk_alloc_disabled = true;
6402 : : }
6403 : :
6404 : : #else
6405 : :
6406 : : #define cgroup_sk_alloc_disabled false
6407 : :
6408 : : #endif
6409 : :
6410 : 419798 : void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6411 : : {
6412 [ - + ]: 419798 : if (cgroup_sk_alloc_disabled) {
6413 : 0 : skcd->no_refcnt = 1;
6414 : 0 : return;
6415 : : }
6416 : :
6417 : : /* Don't associate the sock with unrelated interrupted task's cgroup. */
6418 [ + + ]: 419798 : if (in_interrupt())
6419 : : return;
6420 : :
6421 : : rcu_read_lock();
6422 : :
6423 : : while (true) {
6424 : : struct css_set *cset;
6425 : :
6426 : 419792 : cset = task_css_set(current);
6427 [ + - ]: 839582 : if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6428 : 419790 : skcd->val = (unsigned long)cset->dfl_cgrp;
6429 : 419790 : cgroup_bpf_get(cset->dfl_cgrp);
6430 : : break;
6431 : : }
6432 : 0 : cpu_relax();
6433 : 0 : }
6434 : :
6435 : : rcu_read_unlock();
6436 : : }
6437 : :
6438 : 0 : void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6439 : : {
6440 [ # # ]: 0 : if (skcd->val) {
6441 [ # # ]: 0 : if (skcd->no_refcnt)
6442 : 0 : return;
6443 : : /*
6444 : : * We might be cloning a socket which is left in an empty
6445 : : * cgroup and the cgroup might have already been rmdir'd.
6446 : : * Don't use cgroup_get_live().
6447 : : */
6448 : : cgroup_get(sock_cgroup_ptr(skcd));
6449 : : cgroup_bpf_get(sock_cgroup_ptr(skcd));
6450 : : }
6451 : : }
6452 : :
6453 : 326048 : void cgroup_sk_free(struct sock_cgroup_data *skcd)
6454 : : {
6455 : : struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6456 : :
6457 [ + + ]: 326048 : if (skcd->no_refcnt)
6458 : 326056 : return;
6459 : : cgroup_bpf_put(cgrp);
6460 : : cgroup_put(cgrp);
6461 : : }
6462 : :
6463 : : #endif /* CONFIG_SOCK_CGROUP_DATA */
6464 : :
6465 : : #ifdef CONFIG_CGROUP_BPF
6466 : 2424 : int cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog,
6467 : : enum bpf_attach_type type, u32 flags)
6468 : : {
6469 : : int ret;
6470 : :
6471 : 2424 : mutex_lock(&cgroup_mutex);
6472 : 2424 : ret = __cgroup_bpf_attach(cgrp, prog, type, flags);
6473 : 2424 : mutex_unlock(&cgroup_mutex);
6474 : 2424 : return ret;
6475 : : }
6476 : 0 : int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6477 : : enum bpf_attach_type type, u32 flags)
6478 : : {
6479 : : int ret;
6480 : :
6481 : 0 : mutex_lock(&cgroup_mutex);
6482 : 0 : ret = __cgroup_bpf_detach(cgrp, prog, type);
6483 : 0 : mutex_unlock(&cgroup_mutex);
6484 : 0 : return ret;
6485 : : }
6486 : 0 : int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6487 : : union bpf_attr __user *uattr)
6488 : : {
6489 : : int ret;
6490 : :
6491 : 0 : mutex_lock(&cgroup_mutex);
6492 : 0 : ret = __cgroup_bpf_query(cgrp, attr, uattr);
6493 : 0 : mutex_unlock(&cgroup_mutex);
6494 : 0 : return ret;
6495 : : }
6496 : : #endif /* CONFIG_CGROUP_BPF */
6497 : :
6498 : : #ifdef CONFIG_SYSFS
6499 : 0 : static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6500 : : ssize_t size, const char *prefix)
6501 : : {
6502 : : struct cftype *cft;
6503 : : ssize_t ret = 0;
6504 : :
6505 [ # # # # ]: 0 : for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6506 [ # # ]: 0 : if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6507 : 0 : continue;
6508 : :
6509 [ # # ]: 0 : if (prefix)
6510 : 0 : ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6511 : :
6512 : 0 : ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6513 : :
6514 [ # # # # ]: 0 : if (WARN_ON(ret >= size))
6515 : : break;
6516 : : }
6517 : :
6518 : 0 : return ret;
6519 : : }
6520 : :
6521 : 0 : static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6522 : : char *buf)
6523 : : {
6524 : : struct cgroup_subsys *ss;
6525 : : int ssid;
6526 : : ssize_t ret = 0;
6527 : :
6528 : 0 : ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6529 : : NULL);
6530 : :
6531 [ # # ]: 0 : for_each_subsys(ss, ssid)
6532 : 0 : ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6533 : 0 : PAGE_SIZE - ret,
6534 : : cgroup_subsys_name[ssid]);
6535 : :
6536 : 0 : return ret;
6537 : : }
6538 : : static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6539 : :
6540 : 0 : static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6541 : : char *buf)
6542 : : {
6543 : 0 : return snprintf(buf, PAGE_SIZE, "nsdelegate\nmemory_localevents\n");
6544 : : }
6545 : : static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6546 : :
6547 : : static struct attribute *cgroup_sysfs_attrs[] = {
6548 : : &cgroup_delegate_attr.attr,
6549 : : &cgroup_features_attr.attr,
6550 : : NULL,
6551 : : };
6552 : :
6553 : : static const struct attribute_group cgroup_sysfs_attr_group = {
6554 : : .attrs = cgroup_sysfs_attrs,
6555 : : .name = "cgroup",
6556 : : };
6557 : :
6558 : 404 : static int __init cgroup_sysfs_init(void)
6559 : : {
6560 : 404 : return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6561 : : }
6562 : : subsys_initcall(cgroup_sysfs_init);
6563 : :
6564 : : #endif /* CONFIG_SYSFS */
|
