cgroup: update how a newly forked task gets associated with css_set
[linux-2.6-block.git] / kernel / cgroup.c
CommitLineData
ddbcc7e8 1/*
ddbcc7e8
PM
2 * Generic process-grouping system.
3 *
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
6 *
0dea1168
KS
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
10 *
ddbcc7e8
PM
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
15 *
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
18 *
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
23 *
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
27 */
28
29#include <linux/cgroup.h>
2ce9738b 30#include <linux/cred.h>
c6d57f33 31#include <linux/ctype.h>
ddbcc7e8 32#include <linux/errno.h>
2ce9738b 33#include <linux/init_task.h>
ddbcc7e8
PM
34#include <linux/kernel.h>
35#include <linux/list.h>
36#include <linux/mm.h>
37#include <linux/mutex.h>
38#include <linux/mount.h>
39#include <linux/pagemap.h>
a424316c 40#include <linux/proc_fs.h>
ddbcc7e8
PM
41#include <linux/rcupdate.h>
42#include <linux/sched.h>
ddbcc7e8 43#include <linux/slab.h>
ddbcc7e8 44#include <linux/spinlock.h>
96d365e0 45#include <linux/rwsem.h>
ddbcc7e8 46#include <linux/string.h>
bbcb81d0 47#include <linux/sort.h>
81a6a5cd 48#include <linux/kmod.h>
846c7bb0
BS
49#include <linux/delayacct.h>
50#include <linux/cgroupstats.h>
0ac801fe 51#include <linux/hashtable.h>
096b7fe0 52#include <linux/pid_namespace.h>
2c6ab6d2 53#include <linux/idr.h>
d1d9fd33 54#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
c4c27fbd 55#include <linux/kthread.h>
776f02fa 56#include <linux/delay.h>
846c7bb0 57
60063497 58#include <linux/atomic.h>
ddbcc7e8 59
b1a21367
TH
60/*
61 * pidlists linger the following amount before being destroyed. The goal
62 * is avoiding frequent destruction in the middle of consecutive read calls
63 * Expiring in the middle is a performance problem not a correctness one.
64 * 1 sec should be enough.
65 */
66#define CGROUP_PIDLIST_DESTROY_DELAY HZ
67
8d7e6fb0
TH
68#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
69 MAX_CFTYPE_NAME + 2)
70
ace2bee8
TH
71/*
72 * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file
73 * creation/removal and hierarchy changing operations including cgroup
74 * creation, removal, css association and controller rebinding. This outer
75 * lock is needed mainly to resolve the circular dependency between kernfs
76 * active ref and cgroup_mutex. cgroup_tree_mutex nests above both.
77 */
78static DEFINE_MUTEX(cgroup_tree_mutex);
79
e25e2cbb
TH
80/*
81 * cgroup_mutex is the master lock. Any modification to cgroup or its
82 * hierarchy must be performed while holding it.
e25e2cbb 83 */
2219449a
TH
84#ifdef CONFIG_PROVE_RCU
85DEFINE_MUTEX(cgroup_mutex);
8af01f56 86EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */
2219449a 87#else
81a6a5cd 88static DEFINE_MUTEX(cgroup_mutex);
2219449a
TH
89#endif
90
69e943b7
TH
91/*
92 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
93 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
94 */
95static DEFINE_SPINLOCK(release_agent_path_lock);
96
ace2bee8 97#define cgroup_assert_mutexes_or_rcu_locked() \
87fb54f1 98 rcu_lockdep_assert(rcu_read_lock_held() || \
ace2bee8 99 lockdep_is_held(&cgroup_tree_mutex) || \
87fb54f1 100 lockdep_is_held(&cgroup_mutex), \
ace2bee8 101 "cgroup_[tree_]mutex or RCU read lock required");
87fb54f1 102
e5fca243
TH
103/*
104 * cgroup destruction makes heavy use of work items and there can be a lot
105 * of concurrent destructions. Use a separate workqueue so that cgroup
106 * destruction work items don't end up filling up max_active of system_wq
107 * which may lead to deadlock.
108 */
109static struct workqueue_struct *cgroup_destroy_wq;
110
b1a21367
TH
111/*
112 * pidlist destructions need to be flushed on cgroup destruction. Use a
113 * separate workqueue as flush domain.
114 */
115static struct workqueue_struct *cgroup_pidlist_destroy_wq;
116
3ed80a62 117/* generate an array of cgroup subsystem pointers */
073219e9 118#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
3ed80a62 119static struct cgroup_subsys *cgroup_subsys[] = {
ddbcc7e8
PM
120#include <linux/cgroup_subsys.h>
121};
073219e9
TH
122#undef SUBSYS
123
124/* array of cgroup subsystem names */
125#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
126static const char *cgroup_subsys_name[] = {
127#include <linux/cgroup_subsys.h>
128};
129#undef SUBSYS
ddbcc7e8 130
ddbcc7e8 131/*
9871bf95
TH
132 * The dummy hierarchy, reserved for the subsystems that are otherwise
133 * unattached - it never has more than a single cgroup, and all tasks are
134 * part of that cgroup.
ddbcc7e8 135 */
9871bf95
TH
136static struct cgroupfs_root cgroup_dummy_root;
137
138/* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
139static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup;
ddbcc7e8
PM
140
141/* The list of hierarchy roots */
142
9871bf95
TH
143static LIST_HEAD(cgroup_roots);
144static int cgroup_root_count;
ddbcc7e8 145
3417ae1f 146/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
1a574231 147static DEFINE_IDR(cgroup_hierarchy_idr);
2c6ab6d2 148
794611a1
LZ
149/*
150 * Assign a monotonically increasing serial number to cgroups. It
151 * guarantees cgroups with bigger numbers are newer than those with smaller
152 * numbers. Also, as cgroups are always appended to the parent's
153 * ->children list, it guarantees that sibling cgroups are always sorted in
00356bd5
TH
154 * the ascending serial number order on the list. Protected by
155 * cgroup_mutex.
794611a1 156 */
00356bd5 157static u64 cgroup_serial_nr_next = 1;
794611a1 158
ddbcc7e8 159/* This flag indicates whether tasks in the fork and exit paths should
a043e3b2
LZ
160 * check for fork/exit handlers to call. This avoids us having to do
161 * extra work in the fork/exit path if none of the subsystems need to
162 * be called.
ddbcc7e8 163 */
8947f9d5 164static int need_forkexit_callback __read_mostly;
ddbcc7e8 165
628f7cd4
TH
166static struct cftype cgroup_base_files[];
167
59f5296b 168static void cgroup_put(struct cgroup *cgrp);
f2e85d57
TH
169static int rebind_subsystems(struct cgroupfs_root *root,
170 unsigned long added_mask, unsigned removed_mask);
f20104de 171static void cgroup_destroy_css_killed(struct cgroup *cgrp);
42809dd4 172static int cgroup_destroy_locked(struct cgroup *cgrp);
2bb566cb
TH
173static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
174 bool is_add);
b1a21367 175static void cgroup_pidlist_destroy_all(struct cgroup *cgrp);
42809dd4 176
95109b62
TH
177/**
178 * cgroup_css - obtain a cgroup's css for the specified subsystem
179 * @cgrp: the cgroup of interest
ca8bdcaf 180 * @ss: the subsystem of interest (%NULL returns the dummy_css)
95109b62 181 *
ca8bdcaf
TH
182 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
183 * function must be called either under cgroup_mutex or rcu_read_lock() and
184 * the caller is responsible for pinning the returned css if it wants to
185 * keep accessing it outside the said locks. This function may return
186 * %NULL if @cgrp doesn't have @subsys_id enabled.
95109b62
TH
187 */
188static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
ca8bdcaf 189 struct cgroup_subsys *ss)
95109b62 190{
ca8bdcaf 191 if (ss)
aec25020 192 return rcu_dereference_check(cgrp->subsys[ss->id],
ace2bee8
TH
193 lockdep_is_held(&cgroup_tree_mutex) ||
194 lockdep_is_held(&cgroup_mutex));
ca8bdcaf
TH
195 else
196 return &cgrp->dummy_css;
95109b62 197}
42809dd4 198
ddbcc7e8 199/* convenient tests for these bits */
54766d4a 200static inline bool cgroup_is_dead(const struct cgroup *cgrp)
ddbcc7e8 201{
54766d4a 202 return test_bit(CGRP_DEAD, &cgrp->flags);
ddbcc7e8
PM
203}
204
59f5296b
TH
205struct cgroup_subsys_state *seq_css(struct seq_file *seq)
206{
2bd59d48
TH
207 struct kernfs_open_file *of = seq->private;
208 struct cgroup *cgrp = of->kn->parent->priv;
209 struct cftype *cft = seq_cft(seq);
210
211 /*
212 * This is open and unprotected implementation of cgroup_css().
213 * seq_css() is only called from a kernfs file operation which has
214 * an active reference on the file. Because all the subsystem
215 * files are drained before a css is disassociated with a cgroup,
216 * the matching css from the cgroup's subsys table is guaranteed to
217 * be and stay valid until the enclosing operation is complete.
218 */
219 if (cft->ss)
220 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
221 else
222 return &cgrp->dummy_css;
59f5296b
TH
223}
224EXPORT_SYMBOL_GPL(seq_css);
225
78574cf9
LZ
226/**
227 * cgroup_is_descendant - test ancestry
228 * @cgrp: the cgroup to be tested
229 * @ancestor: possible ancestor of @cgrp
230 *
231 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
232 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
233 * and @ancestor are accessible.
234 */
235bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
236{
237 while (cgrp) {
238 if (cgrp == ancestor)
239 return true;
240 cgrp = cgrp->parent;
241 }
242 return false;
243}
ddbcc7e8 244
e9685a03 245static int cgroup_is_releasable(const struct cgroup *cgrp)
81a6a5cd
PM
246{
247 const int bits =
bd89aabc
PM
248 (1 << CGRP_RELEASABLE) |
249 (1 << CGRP_NOTIFY_ON_RELEASE);
250 return (cgrp->flags & bits) == bits;
81a6a5cd
PM
251}
252
e9685a03 253static int notify_on_release(const struct cgroup *cgrp)
81a6a5cd 254{
bd89aabc 255 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
81a6a5cd
PM
256}
257
1c6727af
TH
258/**
259 * for_each_css - iterate all css's of a cgroup
260 * @css: the iteration cursor
261 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
262 * @cgrp: the target cgroup to iterate css's of
263 *
264 * Should be called under cgroup_mutex.
265 */
266#define for_each_css(css, ssid, cgrp) \
267 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
268 if (!((css) = rcu_dereference_check( \
269 (cgrp)->subsys[(ssid)], \
ace2bee8 270 lockdep_is_held(&cgroup_tree_mutex) || \
1c6727af
TH
271 lockdep_is_held(&cgroup_mutex)))) { } \
272 else
273
30159ec7 274/**
3ed80a62 275 * for_each_subsys - iterate all enabled cgroup subsystems
30159ec7 276 * @ss: the iteration cursor
780cd8b3 277 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
30159ec7 278 */
780cd8b3 279#define for_each_subsys(ss, ssid) \
3ed80a62
TH
280 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
281 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
30159ec7 282
5549c497
TH
283/* iterate across the active hierarchies */
284#define for_each_active_root(root) \
285 list_for_each_entry((root), &cgroup_roots, root_list)
ddbcc7e8 286
7ae1bad9
TH
287/**
288 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
289 * @cgrp: the cgroup to be checked for liveness
290 *
47cfcd09
TH
291 * On success, returns true; the mutex should be later unlocked. On
292 * failure returns false with no lock held.
7ae1bad9 293 */
b9777cf8 294static bool cgroup_lock_live_group(struct cgroup *cgrp)
7ae1bad9
TH
295{
296 mutex_lock(&cgroup_mutex);
54766d4a 297 if (cgroup_is_dead(cgrp)) {
7ae1bad9
TH
298 mutex_unlock(&cgroup_mutex);
299 return false;
300 }
301 return true;
302}
7ae1bad9 303
81a6a5cd
PM
304/* the list of cgroups eligible for automatic release. Protected by
305 * release_list_lock */
306static LIST_HEAD(release_list);
cdcc136f 307static DEFINE_RAW_SPINLOCK(release_list_lock);
81a6a5cd
PM
308static void cgroup_release_agent(struct work_struct *work);
309static DECLARE_WORK(release_agent_work, cgroup_release_agent);
bd89aabc 310static void check_for_release(struct cgroup *cgrp);
81a6a5cd 311
69d0206c
TH
312/*
313 * A cgroup can be associated with multiple css_sets as different tasks may
314 * belong to different cgroups on different hierarchies. In the other
315 * direction, a css_set is naturally associated with multiple cgroups.
316 * This M:N relationship is represented by the following link structure
317 * which exists for each association and allows traversing the associations
318 * from both sides.
319 */
320struct cgrp_cset_link {
321 /* the cgroup and css_set this link associates */
322 struct cgroup *cgrp;
323 struct css_set *cset;
324
325 /* list of cgrp_cset_links anchored at cgrp->cset_links */
326 struct list_head cset_link;
327
328 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
329 struct list_head cgrp_link;
817929ec
PM
330};
331
332/* The default css_set - used by init and its children prior to any
333 * hierarchies being mounted. It contains a pointer to the root state
334 * for each subsystem. Also used to anchor the list of css_sets. Not
335 * reference-counted, to improve performance when child cgroups
336 * haven't been created.
337 */
338
339static struct css_set init_css_set;
69d0206c 340static struct cgrp_cset_link init_cgrp_cset_link;
817929ec 341
0942eeee 342/*
96d365e0
TH
343 * css_set_rwsem protects the list of css_set objects, and the chain of
344 * tasks off each css_set.
0942eeee 345 */
96d365e0 346static DECLARE_RWSEM(css_set_rwsem);
817929ec
PM
347static int css_set_count;
348
7717f7ba
PM
349/*
350 * hash table for cgroup groups. This improves the performance to find
351 * an existing css_set. This hash doesn't (currently) take into
352 * account cgroups in empty hierarchies.
353 */
472b1053 354#define CSS_SET_HASH_BITS 7
0ac801fe 355static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
472b1053 356
0ac801fe 357static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
472b1053 358{
0ac801fe 359 unsigned long key = 0UL;
30159ec7
TH
360 struct cgroup_subsys *ss;
361 int i;
472b1053 362
30159ec7 363 for_each_subsys(ss, i)
0ac801fe
LZ
364 key += (unsigned long)css[i];
365 key = (key >> 16) ^ key;
472b1053 366
0ac801fe 367 return key;
472b1053
LZ
368}
369
89c5509b 370static void put_css_set_locked(struct css_set *cset, bool taskexit)
b4f48b63 371{
69d0206c 372 struct cgrp_cset_link *link, *tmp_link;
5abb8855 373
89c5509b
TH
374 lockdep_assert_held(&css_set_rwsem);
375
376 if (!atomic_dec_and_test(&cset->refcount))
146aa1bd 377 return;
81a6a5cd 378
2c6ab6d2 379 /* This css_set is dead. unlink it and release cgroup refcounts */
5abb8855 380 hash_del(&cset->hlist);
2c6ab6d2
PM
381 css_set_count--;
382
69d0206c 383 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
2c6ab6d2 384 struct cgroup *cgrp = link->cgrp;
5abb8855 385
69d0206c
TH
386 list_del(&link->cset_link);
387 list_del(&link->cgrp_link);
71b5707e 388
96d365e0 389 /* @cgrp can't go away while we're holding css_set_rwsem */
6f3d828f 390 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
81a6a5cd 391 if (taskexit)
bd89aabc
PM
392 set_bit(CGRP_RELEASABLE, &cgrp->flags);
393 check_for_release(cgrp);
81a6a5cd 394 }
2c6ab6d2
PM
395
396 kfree(link);
81a6a5cd 397 }
2c6ab6d2 398
5abb8855 399 kfree_rcu(cset, rcu_head);
b4f48b63
PM
400}
401
89c5509b
TH
402static void put_css_set(struct css_set *cset, bool taskexit)
403{
404 /*
405 * Ensure that the refcount doesn't hit zero while any readers
406 * can see it. Similar to atomic_dec_and_lock(), but for an
407 * rwlock
408 */
409 if (atomic_add_unless(&cset->refcount, -1, 1))
410 return;
411
412 down_write(&css_set_rwsem);
413 put_css_set_locked(cset, taskexit);
414 up_write(&css_set_rwsem);
415}
416
817929ec
PM
417/*
418 * refcounted get/put for css_set objects
419 */
5abb8855 420static inline void get_css_set(struct css_set *cset)
817929ec 421{
5abb8855 422 atomic_inc(&cset->refcount);
817929ec
PM
423}
424
b326f9d0 425/**
7717f7ba 426 * compare_css_sets - helper function for find_existing_css_set().
5abb8855
TH
427 * @cset: candidate css_set being tested
428 * @old_cset: existing css_set for a task
7717f7ba
PM
429 * @new_cgrp: cgroup that's being entered by the task
430 * @template: desired set of css pointers in css_set (pre-calculated)
431 *
6f4b7e63 432 * Returns true if "cset" matches "old_cset" except for the hierarchy
7717f7ba
PM
433 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
434 */
5abb8855
TH
435static bool compare_css_sets(struct css_set *cset,
436 struct css_set *old_cset,
7717f7ba
PM
437 struct cgroup *new_cgrp,
438 struct cgroup_subsys_state *template[])
439{
440 struct list_head *l1, *l2;
441
5abb8855 442 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
7717f7ba
PM
443 /* Not all subsystems matched */
444 return false;
445 }
446
447 /*
448 * Compare cgroup pointers in order to distinguish between
449 * different cgroups in heirarchies with no subsystems. We
450 * could get by with just this check alone (and skip the
451 * memcmp above) but on most setups the memcmp check will
452 * avoid the need for this more expensive check on almost all
453 * candidates.
454 */
455
69d0206c
TH
456 l1 = &cset->cgrp_links;
457 l2 = &old_cset->cgrp_links;
7717f7ba 458 while (1) {
69d0206c 459 struct cgrp_cset_link *link1, *link2;
5abb8855 460 struct cgroup *cgrp1, *cgrp2;
7717f7ba
PM
461
462 l1 = l1->next;
463 l2 = l2->next;
464 /* See if we reached the end - both lists are equal length. */
69d0206c
TH
465 if (l1 == &cset->cgrp_links) {
466 BUG_ON(l2 != &old_cset->cgrp_links);
7717f7ba
PM
467 break;
468 } else {
69d0206c 469 BUG_ON(l2 == &old_cset->cgrp_links);
7717f7ba
PM
470 }
471 /* Locate the cgroups associated with these links. */
69d0206c
TH
472 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
473 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
474 cgrp1 = link1->cgrp;
475 cgrp2 = link2->cgrp;
7717f7ba 476 /* Hierarchies should be linked in the same order. */
5abb8855 477 BUG_ON(cgrp1->root != cgrp2->root);
7717f7ba
PM
478
479 /*
480 * If this hierarchy is the hierarchy of the cgroup
481 * that's changing, then we need to check that this
482 * css_set points to the new cgroup; if it's any other
483 * hierarchy, then this css_set should point to the
484 * same cgroup as the old css_set.
485 */
5abb8855
TH
486 if (cgrp1->root == new_cgrp->root) {
487 if (cgrp1 != new_cgrp)
7717f7ba
PM
488 return false;
489 } else {
5abb8855 490 if (cgrp1 != cgrp2)
7717f7ba
PM
491 return false;
492 }
493 }
494 return true;
495}
496
b326f9d0
TH
497/**
498 * find_existing_css_set - init css array and find the matching css_set
499 * @old_cset: the css_set that we're using before the cgroup transition
500 * @cgrp: the cgroup that we're moving into
501 * @template: out param for the new set of csses, should be clear on entry
817929ec 502 */
5abb8855
TH
503static struct css_set *find_existing_css_set(struct css_set *old_cset,
504 struct cgroup *cgrp,
505 struct cgroup_subsys_state *template[])
b4f48b63 506{
bd89aabc 507 struct cgroupfs_root *root = cgrp->root;
30159ec7 508 struct cgroup_subsys *ss;
5abb8855 509 struct css_set *cset;
0ac801fe 510 unsigned long key;
b326f9d0 511 int i;
817929ec 512
aae8aab4
BB
513 /*
514 * Build the set of subsystem state objects that we want to see in the
515 * new css_set. while subsystems can change globally, the entries here
516 * won't change, so no need for locking.
517 */
30159ec7 518 for_each_subsys(ss, i) {
a1a71b45 519 if (root->subsys_mask & (1UL << i)) {
817929ec
PM
520 /* Subsystem is in this hierarchy. So we want
521 * the subsystem state from the new
522 * cgroup */
ca8bdcaf 523 template[i] = cgroup_css(cgrp, ss);
817929ec
PM
524 } else {
525 /* Subsystem is not in this hierarchy, so we
526 * don't want to change the subsystem state */
5abb8855 527 template[i] = old_cset->subsys[i];
817929ec
PM
528 }
529 }
530
0ac801fe 531 key = css_set_hash(template);
5abb8855
TH
532 hash_for_each_possible(css_set_table, cset, hlist, key) {
533 if (!compare_css_sets(cset, old_cset, cgrp, template))
7717f7ba
PM
534 continue;
535
536 /* This css_set matches what we need */
5abb8855 537 return cset;
472b1053 538 }
817929ec
PM
539
540 /* No existing cgroup group matched */
541 return NULL;
542}
543
69d0206c 544static void free_cgrp_cset_links(struct list_head *links_to_free)
36553434 545{
69d0206c 546 struct cgrp_cset_link *link, *tmp_link;
36553434 547
69d0206c
TH
548 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
549 list_del(&link->cset_link);
36553434
LZ
550 kfree(link);
551 }
552}
553
69d0206c
TH
554/**
555 * allocate_cgrp_cset_links - allocate cgrp_cset_links
556 * @count: the number of links to allocate
557 * @tmp_links: list_head the allocated links are put on
558 *
559 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
560 * through ->cset_link. Returns 0 on success or -errno.
817929ec 561 */
69d0206c 562static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
817929ec 563{
69d0206c 564 struct cgrp_cset_link *link;
817929ec 565 int i;
69d0206c
TH
566
567 INIT_LIST_HEAD(tmp_links);
568
817929ec 569 for (i = 0; i < count; i++) {
f4f4be2b 570 link = kzalloc(sizeof(*link), GFP_KERNEL);
817929ec 571 if (!link) {
69d0206c 572 free_cgrp_cset_links(tmp_links);
817929ec
PM
573 return -ENOMEM;
574 }
69d0206c 575 list_add(&link->cset_link, tmp_links);
817929ec
PM
576 }
577 return 0;
578}
579
c12f65d4
LZ
580/**
581 * link_css_set - a helper function to link a css_set to a cgroup
69d0206c 582 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
5abb8855 583 * @cset: the css_set to be linked
c12f65d4
LZ
584 * @cgrp: the destination cgroup
585 */
69d0206c
TH
586static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
587 struct cgroup *cgrp)
c12f65d4 588{
69d0206c 589 struct cgrp_cset_link *link;
c12f65d4 590
69d0206c
TH
591 BUG_ON(list_empty(tmp_links));
592 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
593 link->cset = cset;
7717f7ba 594 link->cgrp = cgrp;
69d0206c 595 list_move(&link->cset_link, &cgrp->cset_links);
7717f7ba
PM
596 /*
597 * Always add links to the tail of the list so that the list
598 * is sorted by order of hierarchy creation
599 */
69d0206c 600 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
c12f65d4
LZ
601}
602
b326f9d0
TH
603/**
604 * find_css_set - return a new css_set with one cgroup updated
605 * @old_cset: the baseline css_set
606 * @cgrp: the cgroup to be updated
607 *
608 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
609 * substituted into the appropriate hierarchy.
817929ec 610 */
5abb8855
TH
611static struct css_set *find_css_set(struct css_set *old_cset,
612 struct cgroup *cgrp)
817929ec 613{
b326f9d0 614 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
5abb8855 615 struct css_set *cset;
69d0206c
TH
616 struct list_head tmp_links;
617 struct cgrp_cset_link *link;
0ac801fe 618 unsigned long key;
472b1053 619
b326f9d0
TH
620 lockdep_assert_held(&cgroup_mutex);
621
817929ec
PM
622 /* First see if we already have a cgroup group that matches
623 * the desired set */
96d365e0 624 down_read(&css_set_rwsem);
5abb8855
TH
625 cset = find_existing_css_set(old_cset, cgrp, template);
626 if (cset)
627 get_css_set(cset);
96d365e0 628 up_read(&css_set_rwsem);
817929ec 629
5abb8855
TH
630 if (cset)
631 return cset;
817929ec 632
f4f4be2b 633 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
5abb8855 634 if (!cset)
817929ec
PM
635 return NULL;
636
69d0206c 637 /* Allocate all the cgrp_cset_link objects that we'll need */
9871bf95 638 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
5abb8855 639 kfree(cset);
817929ec
PM
640 return NULL;
641 }
642
5abb8855 643 atomic_set(&cset->refcount, 1);
69d0206c 644 INIT_LIST_HEAD(&cset->cgrp_links);
5abb8855 645 INIT_LIST_HEAD(&cset->tasks);
c7561128 646 INIT_LIST_HEAD(&cset->mg_tasks);
1958d2d5 647 INIT_LIST_HEAD(&cset->mg_preload_node);
b3dc094e 648 INIT_LIST_HEAD(&cset->mg_node);
5abb8855 649 INIT_HLIST_NODE(&cset->hlist);
817929ec
PM
650
651 /* Copy the set of subsystem state objects generated in
652 * find_existing_css_set() */
5abb8855 653 memcpy(cset->subsys, template, sizeof(cset->subsys));
817929ec 654
96d365e0 655 down_write(&css_set_rwsem);
817929ec 656 /* Add reference counts and links from the new css_set. */
69d0206c 657 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
7717f7ba 658 struct cgroup *c = link->cgrp;
69d0206c 659
7717f7ba
PM
660 if (c->root == cgrp->root)
661 c = cgrp;
69d0206c 662 link_css_set(&tmp_links, cset, c);
7717f7ba 663 }
817929ec 664
69d0206c 665 BUG_ON(!list_empty(&tmp_links));
817929ec 666
817929ec 667 css_set_count++;
472b1053
LZ
668
669 /* Add this cgroup group to the hash table */
5abb8855
TH
670 key = css_set_hash(cset->subsys);
671 hash_add(css_set_table, &cset->hlist, key);
472b1053 672
96d365e0 673 up_write(&css_set_rwsem);
817929ec 674
5abb8855 675 return cset;
b4f48b63
PM
676}
677
2bd59d48
TH
678static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
679{
680 struct cgroup *top_cgrp = kf_root->kn->priv;
681
682 return top_cgrp->root;
683}
684
f2e85d57
TH
685static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end)
686{
687 int id;
688
689 lockdep_assert_held(&cgroup_mutex);
690
691 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end,
692 GFP_KERNEL);
693 if (id < 0)
694 return id;
695
696 root->hierarchy_id = id;
697 return 0;
698}
699
700static void cgroup_exit_root_id(struct cgroupfs_root *root)
701{
702 lockdep_assert_held(&cgroup_mutex);
703
704 if (root->hierarchy_id) {
705 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
706 root->hierarchy_id = 0;
707 }
708}
709
710static void cgroup_free_root(struct cgroupfs_root *root)
711{
712 if (root) {
713 /* hierarhcy ID shoulid already have been released */
714 WARN_ON_ONCE(root->hierarchy_id);
715
716 idr_destroy(&root->cgroup_idr);
717 kfree(root);
718 }
719}
720
776f02fa 721static void cgroup_destroy_root(struct cgroupfs_root *root)
59f5296b 722{
f2e85d57
TH
723 struct cgroup *cgrp = &root->top_cgroup;
724 struct cgrp_cset_link *link, *tmp_link;
f2e85d57 725
2bd59d48 726 mutex_lock(&cgroup_tree_mutex);
2bd59d48 727 mutex_lock(&cgroup_mutex);
f2e85d57 728
776f02fa 729 BUG_ON(atomic_read(&root->nr_cgrps));
f2e85d57
TH
730 BUG_ON(!list_empty(&cgrp->children));
731
f2e85d57 732 /* Rebind all subsystems back to the default hierarchy */
35585573 733 WARN_ON(rebind_subsystems(root, 0, root->subsys_mask));
f2e85d57
TH
734
735 /*
736 * Release all the links from cset_links to this hierarchy's
737 * root cgroup
738 */
96d365e0 739 down_write(&css_set_rwsem);
f2e85d57
TH
740
741 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
742 list_del(&link->cset_link);
743 list_del(&link->cgrp_link);
744 kfree(link);
745 }
96d365e0 746 up_write(&css_set_rwsem);
f2e85d57
TH
747
748 if (!list_empty(&root->root_list)) {
749 list_del(&root->root_list);
750 cgroup_root_count--;
751 }
752
753 cgroup_exit_root_id(root);
754
755 mutex_unlock(&cgroup_mutex);
756 mutex_unlock(&cgroup_tree_mutex);
f2e85d57 757
2bd59d48 758 kernfs_destroy_root(root->kf_root);
f2e85d57
TH
759 cgroup_free_root(root);
760}
761
ceb6a081
TH
762/* look up cgroup associated with given css_set on the specified hierarchy */
763static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
7717f7ba
PM
764 struct cgroupfs_root *root)
765{
7717f7ba
PM
766 struct cgroup *res = NULL;
767
96d365e0
TH
768 lockdep_assert_held(&cgroup_mutex);
769 lockdep_assert_held(&css_set_rwsem);
770
5abb8855 771 if (cset == &init_css_set) {
7717f7ba
PM
772 res = &root->top_cgroup;
773 } else {
69d0206c
TH
774 struct cgrp_cset_link *link;
775
776 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
7717f7ba 777 struct cgroup *c = link->cgrp;
69d0206c 778
7717f7ba
PM
779 if (c->root == root) {
780 res = c;
781 break;
782 }
783 }
784 }
96d365e0 785
7717f7ba
PM
786 BUG_ON(!res);
787 return res;
788}
789
ceb6a081
TH
790/*
791 * Return the cgroup for "task" from the given hierarchy. Must be
792 * called with cgroup_mutex and css_set_rwsem held.
793 */
794static struct cgroup *task_cgroup_from_root(struct task_struct *task,
795 struct cgroupfs_root *root)
796{
797 /*
798 * No need to lock the task - since we hold cgroup_mutex the
799 * task can't change groups, so the only thing that can happen
800 * is that it exits and its css is set back to init_css_set.
801 */
802 return cset_cgroup_from_root(task_css_set(task), root);
803}
804
ddbcc7e8
PM
805/*
806 * There is one global cgroup mutex. We also require taking
807 * task_lock() when dereferencing a task's cgroup subsys pointers.
808 * See "The task_lock() exception", at the end of this comment.
809 *
810 * A task must hold cgroup_mutex to modify cgroups.
811 *
812 * Any task can increment and decrement the count field without lock.
813 * So in general, code holding cgroup_mutex can't rely on the count
814 * field not changing. However, if the count goes to zero, then only
956db3ca 815 * cgroup_attach_task() can increment it again. Because a count of zero
ddbcc7e8
PM
816 * means that no tasks are currently attached, therefore there is no
817 * way a task attached to that cgroup can fork (the other way to
818 * increment the count). So code holding cgroup_mutex can safely
819 * assume that if the count is zero, it will stay zero. Similarly, if
820 * a task holds cgroup_mutex on a cgroup with zero count, it
821 * knows that the cgroup won't be removed, as cgroup_rmdir()
822 * needs that mutex.
823 *
ddbcc7e8
PM
824 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
825 * (usually) take cgroup_mutex. These are the two most performance
826 * critical pieces of code here. The exception occurs on cgroup_exit(),
827 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
828 * is taken, and if the cgroup count is zero, a usermode call made
a043e3b2
LZ
829 * to the release agent with the name of the cgroup (path relative to
830 * the root of cgroup file system) as the argument.
ddbcc7e8
PM
831 *
832 * A cgroup can only be deleted if both its 'count' of using tasks
833 * is zero, and its list of 'children' cgroups is empty. Since all
834 * tasks in the system use _some_ cgroup, and since there is always at
835 * least one task in the system (init, pid == 1), therefore, top_cgroup
836 * always has either children cgroups and/or using tasks. So we don't
837 * need a special hack to ensure that top_cgroup cannot be deleted.
838 *
839 * The task_lock() exception
840 *
841 * The need for this exception arises from the action of
d0b2fdd2 842 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
a043e3b2 843 * another. It does so using cgroup_mutex, however there are
ddbcc7e8
PM
844 * several performance critical places that need to reference
845 * task->cgroup without the expense of grabbing a system global
846 * mutex. Therefore except as noted below, when dereferencing or, as
d0b2fdd2 847 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
ddbcc7e8
PM
848 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
849 * the task_struct routinely used for such matters.
850 *
851 * P.S. One more locking exception. RCU is used to guard the
956db3ca 852 * update of a tasks cgroup pointer by cgroup_attach_task()
ddbcc7e8
PM
853 */
854
628f7cd4 855static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask);
2bd59d48 856static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
828c0950 857static const struct file_operations proc_cgroupstats_operations;
a424316c 858
8d7e6fb0
TH
859static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
860 char *buf)
861{
862 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
863 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
864 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
865 cft->ss->name, cft->name);
866 else
867 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
868 return buf;
869}
870
f2e85d57
TH
871/**
872 * cgroup_file_mode - deduce file mode of a control file
873 * @cft: the control file in question
874 *
875 * returns cft->mode if ->mode is not 0
876 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
877 * returns S_IRUGO if it has only a read handler
878 * returns S_IWUSR if it has only a write hander
879 */
880static umode_t cgroup_file_mode(const struct cftype *cft)
881{
882 umode_t mode = 0;
883
884 if (cft->mode)
885 return cft->mode;
886
887 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
888 mode |= S_IRUGO;
889
890 if (cft->write_u64 || cft->write_s64 || cft->write_string ||
891 cft->trigger)
892 mode |= S_IWUSR;
893
894 return mode;
895}
896
be445626
LZ
897static void cgroup_free_fn(struct work_struct *work)
898{
ea15f8cc 899 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
be445626 900
3c9c825b 901 atomic_dec(&cgrp->root->nr_cgrps);
b1a21367 902 cgroup_pidlist_destroy_all(cgrp);
be445626 903
776f02fa
TH
904 if (cgrp->parent) {
905 /*
906 * We get a ref to the parent, and put the ref when this
907 * cgroup is being freed, so it's guaranteed that the
908 * parent won't be destroyed before its children.
909 */
910 cgroup_put(cgrp->parent);
911 kernfs_put(cgrp->kn);
912 kfree(cgrp);
913 } else {
914 /*
915 * This is top cgroup's refcnt reaching zero, which
916 * indicates that the root should be released.
917 */
918 cgroup_destroy_root(cgrp->root);
919 }
be445626
LZ
920}
921
922static void cgroup_free_rcu(struct rcu_head *head)
923{
924 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
925
ea15f8cc 926 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
e5fca243 927 queue_work(cgroup_destroy_wq, &cgrp->destroy_work);
be445626
LZ
928}
929
59f5296b
TH
930static void cgroup_get(struct cgroup *cgrp)
931{
2bd59d48
TH
932 WARN_ON_ONCE(cgroup_is_dead(cgrp));
933 WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0);
934 atomic_inc(&cgrp->refcnt);
ddbcc7e8
PM
935}
936
59f5296b
TH
937static void cgroup_put(struct cgroup *cgrp)
938{
2bd59d48
TH
939 if (!atomic_dec_and_test(&cgrp->refcnt))
940 return;
776f02fa 941 if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp)))
2bd59d48 942 return;
59f5296b 943
2bd59d48
TH
944 /*
945 * XXX: cgrp->id is only used to look up css's. As cgroup and
946 * css's lifetimes will be decoupled, it should be made
947 * per-subsystem and moved to css->id so that lookups are
948 * successful until the target css is released.
949 */
950 mutex_lock(&cgroup_mutex);
951 idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
952 mutex_unlock(&cgroup_mutex);
953 cgrp->id = -1;
ddbcc7e8 954
2bd59d48 955 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
ddbcc7e8
PM
956}
957
2739d3cc 958static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
05ef1d7c 959{
2bd59d48 960 char name[CGROUP_FILE_NAME_MAX];
05ef1d7c 961
ace2bee8 962 lockdep_assert_held(&cgroup_tree_mutex);
2bd59d48 963 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
05ef1d7c
TH
964}
965
13af07df 966/**
628f7cd4 967 * cgroup_clear_dir - remove subsys files in a cgroup directory
8f89140a 968 * @cgrp: target cgroup
13af07df
AR
969 * @subsys_mask: mask of the subsystem ids whose files should be removed
970 */
628f7cd4 971static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask)
05ef1d7c 972{
13af07df 973 struct cgroup_subsys *ss;
b420ba7d 974 int i;
05ef1d7c 975
b420ba7d 976 for_each_subsys(ss, i) {
0adb0704 977 struct cftype *cfts;
b420ba7d
TH
978
979 if (!test_bit(i, &subsys_mask))
13af07df 980 continue;
0adb0704
TH
981 list_for_each_entry(cfts, &ss->cfts, node)
982 cgroup_addrm_files(cgrp, cfts, false);
13af07df 983 }
ddbcc7e8
PM
984}
985
ddbcc7e8 986static int rebind_subsystems(struct cgroupfs_root *root,
a8a648c4 987 unsigned long added_mask, unsigned removed_mask)
ddbcc7e8 988{
bd89aabc 989 struct cgroup *cgrp = &root->top_cgroup;
30159ec7 990 struct cgroup_subsys *ss;
3126121f 991 int i, ret;
ddbcc7e8 992
ace2bee8
TH
993 lockdep_assert_held(&cgroup_tree_mutex);
994 lockdep_assert_held(&cgroup_mutex);
aae8aab4 995
ddbcc7e8 996 /* Check that any added subsystems are currently free */
3ed80a62
TH
997 for_each_subsys(ss, i)
998 if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root)
999 return -EBUSY;
ddbcc7e8 1000
3126121f
TH
1001 ret = cgroup_populate_dir(cgrp, added_mask);
1002 if (ret)
3ed80a62 1003 return ret;
3126121f
TH
1004
1005 /*
1006 * Nothing can fail from this point on. Remove files for the
1007 * removed subsystems and rebind each subsystem.
1008 */
4ac06017 1009 mutex_unlock(&cgroup_mutex);
3126121f 1010 cgroup_clear_dir(cgrp, removed_mask);
4ac06017 1011 mutex_lock(&cgroup_mutex);
ddbcc7e8 1012
30159ec7 1013 for_each_subsys(ss, i) {
ddbcc7e8 1014 unsigned long bit = 1UL << i;
30159ec7 1015
a1a71b45 1016 if (bit & added_mask) {
ddbcc7e8 1017 /* We're binding this subsystem to this hierarchy */
ca8bdcaf
TH
1018 BUG_ON(cgroup_css(cgrp, ss));
1019 BUG_ON(!cgroup_css(cgroup_dummy_top, ss));
1020 BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top);
a8a648c4 1021
73e80ed8 1022 rcu_assign_pointer(cgrp->subsys[i],
ca8bdcaf
TH
1023 cgroup_css(cgroup_dummy_top, ss));
1024 cgroup_css(cgrp, ss)->cgroup = cgrp;
a8a648c4 1025
b2aa30f7 1026 ss->root = root;
ddbcc7e8 1027 if (ss->bind)
ca8bdcaf 1028 ss->bind(cgroup_css(cgrp, ss));
a8a648c4 1029
cf5d5941 1030 /* refcount was already taken, and we're keeping it */
a8a648c4 1031 root->subsys_mask |= bit;
a1a71b45 1032 } else if (bit & removed_mask) {
ddbcc7e8 1033 /* We're removing this subsystem */
ca8bdcaf
TH
1034 BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss));
1035 BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp);
a8a648c4 1036
ddbcc7e8 1037 if (ss->bind)
ca8bdcaf 1038 ss->bind(cgroup_css(cgroup_dummy_top, ss));
73e80ed8 1039
ca8bdcaf 1040 cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top;
73e80ed8
TH
1041 RCU_INIT_POINTER(cgrp->subsys[i], NULL);
1042
9871bf95 1043 cgroup_subsys[i]->root = &cgroup_dummy_root;
a8a648c4 1044 root->subsys_mask &= ~bit;
ddbcc7e8
PM
1045 }
1046 }
ddbcc7e8 1047
2bd59d48 1048 kernfs_activate(cgrp->kn);
ddbcc7e8
PM
1049 return 0;
1050}
1051
2bd59d48
TH
1052static int cgroup_show_options(struct seq_file *seq,
1053 struct kernfs_root *kf_root)
ddbcc7e8 1054{
2bd59d48 1055 struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);
ddbcc7e8 1056 struct cgroup_subsys *ss;
b85d2040 1057 int ssid;
ddbcc7e8 1058
b85d2040
TH
1059 for_each_subsys(ss, ssid)
1060 if (root->subsys_mask & (1 << ssid))
1061 seq_printf(seq, ",%s", ss->name);
873fe09e
TH
1062 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1063 seq_puts(seq, ",sane_behavior");
93438629 1064 if (root->flags & CGRP_ROOT_NOPREFIX)
ddbcc7e8 1065 seq_puts(seq, ",noprefix");
93438629 1066 if (root->flags & CGRP_ROOT_XATTR)
03b1cde6 1067 seq_puts(seq, ",xattr");
69e943b7
TH
1068
1069 spin_lock(&release_agent_path_lock);
81a6a5cd
PM
1070 if (strlen(root->release_agent_path))
1071 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
69e943b7
TH
1072 spin_unlock(&release_agent_path_lock);
1073
2260e7fc 1074 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
97978e6d 1075 seq_puts(seq, ",clone_children");
c6d57f33
PM
1076 if (strlen(root->name))
1077 seq_printf(seq, ",name=%s", root->name);
ddbcc7e8
PM
1078 return 0;
1079}
1080
1081struct cgroup_sb_opts {
a1a71b45 1082 unsigned long subsys_mask;
ddbcc7e8 1083 unsigned long flags;
81a6a5cd 1084 char *release_agent;
2260e7fc 1085 bool cpuset_clone_children;
c6d57f33 1086 char *name;
2c6ab6d2
PM
1087 /* User explicitly requested empty subsystem */
1088 bool none;
ddbcc7e8
PM
1089};
1090
aae8aab4 1091/*
9871bf95
TH
1092 * Convert a hierarchy specifier into a bitmask of subsystems and
1093 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1094 * array. This function takes refcounts on subsystems to be used, unless it
1095 * returns error, in which case no refcounts are taken.
aae8aab4 1096 */
cf5d5941 1097static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
ddbcc7e8 1098{
32a8cf23
DL
1099 char *token, *o = data;
1100 bool all_ss = false, one_ss = false;
f9ab5b5b 1101 unsigned long mask = (unsigned long)-1;
30159ec7
TH
1102 struct cgroup_subsys *ss;
1103 int i;
f9ab5b5b 1104
aae8aab4
BB
1105 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1106
f9ab5b5b 1107#ifdef CONFIG_CPUSETS
073219e9 1108 mask = ~(1UL << cpuset_cgrp_id);
f9ab5b5b 1109#endif
ddbcc7e8 1110
c6d57f33 1111 memset(opts, 0, sizeof(*opts));
ddbcc7e8
PM
1112
1113 while ((token = strsep(&o, ",")) != NULL) {
1114 if (!*token)
1115 return -EINVAL;
32a8cf23 1116 if (!strcmp(token, "none")) {
2c6ab6d2
PM
1117 /* Explicitly have no subsystems */
1118 opts->none = true;
32a8cf23
DL
1119 continue;
1120 }
1121 if (!strcmp(token, "all")) {
1122 /* Mutually exclusive option 'all' + subsystem name */
1123 if (one_ss)
1124 return -EINVAL;
1125 all_ss = true;
1126 continue;
1127 }
873fe09e
TH
1128 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1129 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1130 continue;
1131 }
32a8cf23 1132 if (!strcmp(token, "noprefix")) {
93438629 1133 opts->flags |= CGRP_ROOT_NOPREFIX;
32a8cf23
DL
1134 continue;
1135 }
1136 if (!strcmp(token, "clone_children")) {
2260e7fc 1137 opts->cpuset_clone_children = true;
32a8cf23
DL
1138 continue;
1139 }
03b1cde6 1140 if (!strcmp(token, "xattr")) {
93438629 1141 opts->flags |= CGRP_ROOT_XATTR;
03b1cde6
AR
1142 continue;
1143 }
32a8cf23 1144 if (!strncmp(token, "release_agent=", 14)) {
81a6a5cd
PM
1145 /* Specifying two release agents is forbidden */
1146 if (opts->release_agent)
1147 return -EINVAL;
c6d57f33 1148 opts->release_agent =
e400c285 1149 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
81a6a5cd
PM
1150 if (!opts->release_agent)
1151 return -ENOMEM;
32a8cf23
DL
1152 continue;
1153 }
1154 if (!strncmp(token, "name=", 5)) {
c6d57f33
PM
1155 const char *name = token + 5;
1156 /* Can't specify an empty name */
1157 if (!strlen(name))
1158 return -EINVAL;
1159 /* Must match [\w.-]+ */
1160 for (i = 0; i < strlen(name); i++) {
1161 char c = name[i];
1162 if (isalnum(c))
1163 continue;
1164 if ((c == '.') || (c == '-') || (c == '_'))
1165 continue;
1166 return -EINVAL;
1167 }
1168 /* Specifying two names is forbidden */
1169 if (opts->name)
1170 return -EINVAL;
1171 opts->name = kstrndup(name,
e400c285 1172 MAX_CGROUP_ROOT_NAMELEN - 1,
c6d57f33
PM
1173 GFP_KERNEL);
1174 if (!opts->name)
1175 return -ENOMEM;
32a8cf23
DL
1176
1177 continue;
1178 }
1179
30159ec7 1180 for_each_subsys(ss, i) {
32a8cf23
DL
1181 if (strcmp(token, ss->name))
1182 continue;
1183 if (ss->disabled)
1184 continue;
1185
1186 /* Mutually exclusive option 'all' + subsystem name */
1187 if (all_ss)
1188 return -EINVAL;
a1a71b45 1189 set_bit(i, &opts->subsys_mask);
32a8cf23
DL
1190 one_ss = true;
1191
1192 break;
1193 }
1194 if (i == CGROUP_SUBSYS_COUNT)
1195 return -ENOENT;
1196 }
1197
1198 /*
1199 * If the 'all' option was specified select all the subsystems,
0d19ea86
LZ
1200 * otherwise if 'none', 'name=' and a subsystem name options
1201 * were not specified, let's default to 'all'
32a8cf23 1202 */
30159ec7
TH
1203 if (all_ss || (!one_ss && !opts->none && !opts->name))
1204 for_each_subsys(ss, i)
1205 if (!ss->disabled)
1206 set_bit(i, &opts->subsys_mask);
ddbcc7e8 1207
2c6ab6d2
PM
1208 /* Consistency checks */
1209
873fe09e
TH
1210 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1211 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1212
d3ba07c3
TH
1213 if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) ||
1214 opts->cpuset_clone_children || opts->release_agent ||
1215 opts->name) {
1216 pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n");
873fe09e
TH
1217 return -EINVAL;
1218 }
873fe09e
TH
1219 }
1220
f9ab5b5b
LZ
1221 /*
1222 * Option noprefix was introduced just for backward compatibility
1223 * with the old cpuset, so we allow noprefix only if mounting just
1224 * the cpuset subsystem.
1225 */
93438629 1226 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
f9ab5b5b
LZ
1227 return -EINVAL;
1228
2c6ab6d2
PM
1229
1230 /* Can't specify "none" and some subsystems */
a1a71b45 1231 if (opts->subsys_mask && opts->none)
2c6ab6d2
PM
1232 return -EINVAL;
1233
1234 /*
1235 * We either have to specify by name or by subsystems. (So all
1236 * empty hierarchies must have a name).
1237 */
a1a71b45 1238 if (!opts->subsys_mask && !opts->name)
ddbcc7e8
PM
1239 return -EINVAL;
1240
1241 return 0;
1242}
1243
2bd59d48 1244static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
ddbcc7e8
PM
1245{
1246 int ret = 0;
2bd59d48 1247 struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);
ddbcc7e8 1248 struct cgroup_sb_opts opts;
a1a71b45 1249 unsigned long added_mask, removed_mask;
ddbcc7e8 1250
873fe09e
TH
1251 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1252 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1253 return -EINVAL;
1254 }
1255
ace2bee8 1256 mutex_lock(&cgroup_tree_mutex);
ddbcc7e8
PM
1257 mutex_lock(&cgroup_mutex);
1258
1259 /* See what subsystems are wanted */
1260 ret = parse_cgroupfs_options(data, &opts);
1261 if (ret)
1262 goto out_unlock;
1263
a8a648c4 1264 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
8b5a5a9d
TH
1265 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1266 task_tgid_nr(current), current->comm);
1267
a1a71b45
AR
1268 added_mask = opts.subsys_mask & ~root->subsys_mask;
1269 removed_mask = root->subsys_mask & ~opts.subsys_mask;
13af07df 1270
cf5d5941 1271 /* Don't allow flags or name to change at remount */
0ce6cba3 1272 if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) ||
cf5d5941 1273 (opts.name && strcmp(opts.name, root->name))) {
0ce6cba3
TH
1274 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1275 opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "",
1276 root->flags & CGRP_ROOT_OPTION_MASK, root->name);
c6d57f33
PM
1277 ret = -EINVAL;
1278 goto out_unlock;
1279 }
1280
f172e67c 1281 /* remounting is not allowed for populated hierarchies */
3c9c825b 1282 if (!list_empty(&root->top_cgroup.children)) {
f172e67c 1283 ret = -EBUSY;
0670e08b 1284 goto out_unlock;
cf5d5941 1285 }
ddbcc7e8 1286
a8a648c4 1287 ret = rebind_subsystems(root, added_mask, removed_mask);
3126121f 1288 if (ret)
0670e08b 1289 goto out_unlock;
ddbcc7e8 1290
69e943b7
TH
1291 if (opts.release_agent) {
1292 spin_lock(&release_agent_path_lock);
81a6a5cd 1293 strcpy(root->release_agent_path, opts.release_agent);
69e943b7
TH
1294 spin_unlock(&release_agent_path_lock);
1295 }
ddbcc7e8 1296 out_unlock:
66bdc9cf 1297 kfree(opts.release_agent);
c6d57f33 1298 kfree(opts.name);
ddbcc7e8 1299 mutex_unlock(&cgroup_mutex);
ace2bee8 1300 mutex_unlock(&cgroup_tree_mutex);
ddbcc7e8
PM
1301 return ret;
1302}
1303
afeb0f9f
TH
1304/*
1305 * To reduce the fork() overhead for systems that are not actually using
1306 * their cgroups capability, we don't maintain the lists running through
1307 * each css_set to its tasks until we see the list actually used - in other
1308 * words after the first mount.
1309 */
1310static bool use_task_css_set_links __read_mostly;
1311
1312static void cgroup_enable_task_cg_lists(void)
1313{
1314 struct task_struct *p, *g;
1315
96d365e0 1316 down_write(&css_set_rwsem);
afeb0f9f
TH
1317
1318 if (use_task_css_set_links)
1319 goto out_unlock;
1320
1321 use_task_css_set_links = true;
1322
1323 /*
1324 * We need tasklist_lock because RCU is not safe against
1325 * while_each_thread(). Besides, a forking task that has passed
1326 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1327 * is not guaranteed to have its child immediately visible in the
1328 * tasklist if we walk through it with RCU.
1329 */
1330 read_lock(&tasklist_lock);
1331 do_each_thread(g, p) {
1332 task_lock(p);
1333
1334 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1335 task_css_set(p) != &init_css_set);
1336
1337 /*
1338 * We should check if the process is exiting, otherwise
1339 * it will race with cgroup_exit() in that the list
1340 * entry won't be deleted though the process has exited.
f153ad11
TH
1341 * Do it while holding siglock so that we don't end up
1342 * racing against cgroup_exit().
afeb0f9f 1343 */
f153ad11 1344 spin_lock_irq(&p->sighand->siglock);
eaf797ab
TH
1345 if (!(p->flags & PF_EXITING)) {
1346 struct css_set *cset = task_css_set(p);
1347
1348 list_add(&p->cg_list, &cset->tasks);
1349 get_css_set(cset);
1350 }
f153ad11 1351 spin_unlock_irq(&p->sighand->siglock);
afeb0f9f
TH
1352
1353 task_unlock(p);
1354 } while_each_thread(g, p);
1355 read_unlock(&tasklist_lock);
1356out_unlock:
96d365e0 1357 up_write(&css_set_rwsem);
afeb0f9f
TH
1358}
1359
cc31edce
PM
1360static void init_cgroup_housekeeping(struct cgroup *cgrp)
1361{
2bd59d48 1362 atomic_set(&cgrp->refcnt, 1);
cc31edce
PM
1363 INIT_LIST_HEAD(&cgrp->sibling);
1364 INIT_LIST_HEAD(&cgrp->children);
69d0206c 1365 INIT_LIST_HEAD(&cgrp->cset_links);
cc31edce 1366 INIT_LIST_HEAD(&cgrp->release_list);
72a8cb30
BB
1367 INIT_LIST_HEAD(&cgrp->pidlists);
1368 mutex_init(&cgrp->pidlist_mutex);
67f4c36f 1369 cgrp->dummy_css.cgroup = cgrp;
cc31edce 1370}
c6d57f33 1371
ddbcc7e8
PM
1372static void init_cgroup_root(struct cgroupfs_root *root)
1373{
bd89aabc 1374 struct cgroup *cgrp = &root->top_cgroup;
b0ca5a84 1375
ddbcc7e8 1376 INIT_LIST_HEAD(&root->root_list);
3c9c825b 1377 atomic_set(&root->nr_cgrps, 1);
bd89aabc 1378 cgrp->root = root;
cc31edce 1379 init_cgroup_housekeeping(cgrp);
4e96ee8e 1380 idr_init(&root->cgroup_idr);
ddbcc7e8
PM
1381}
1382
c6d57f33
PM
1383static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1384{
1385 struct cgroupfs_root *root;
1386
a1a71b45 1387 if (!opts->subsys_mask && !opts->none)
2bd59d48 1388 return ERR_PTR(-EINVAL);
c6d57f33
PM
1389
1390 root = kzalloc(sizeof(*root), GFP_KERNEL);
1391 if (!root)
1392 return ERR_PTR(-ENOMEM);
1393
1394 init_cgroup_root(root);
2c6ab6d2 1395
c6d57f33
PM
1396 root->flags = opts->flags;
1397 if (opts->release_agent)
1398 strcpy(root->release_agent_path, opts->release_agent);
1399 if (opts->name)
1400 strcpy(root->name, opts->name);
2260e7fc
TH
1401 if (opts->cpuset_clone_children)
1402 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
c6d57f33
PM
1403 return root;
1404}
1405
35585573 1406static int cgroup_setup_root(struct cgroupfs_root *root, unsigned long ss_mask)
d427dfeb
TH
1407{
1408 LIST_HEAD(tmp_links);
d427dfeb 1409 struct cgroup *root_cgrp = &root->top_cgroup;
d427dfeb 1410 struct css_set *cset;
d427dfeb
TH
1411 int i, ret;
1412
1413 lockdep_assert_held(&cgroup_tree_mutex);
1414 lockdep_assert_held(&cgroup_mutex);
d427dfeb
TH
1415
1416 ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL);
1417 if (ret < 0)
2bd59d48 1418 goto out;
d427dfeb
TH
1419 root_cgrp->id = ret;
1420
d427dfeb 1421 /*
96d365e0 1422 * We're accessing css_set_count without locking css_set_rwsem here,
d427dfeb
TH
1423 * but that's OK - it can only be increased by someone holding
1424 * cgroup_lock, and that's us. The worst that can happen is that we
1425 * have some link structures left over
1426 */
1427 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1428 if (ret)
2bd59d48 1429 goto out;
d427dfeb
TH
1430
1431 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1432 ret = cgroup_init_root_id(root, 2, 0);
1433 if (ret)
2bd59d48 1434 goto out;
d427dfeb 1435
2bd59d48
TH
1436 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
1437 KERNFS_ROOT_CREATE_DEACTIVATED,
1438 root_cgrp);
1439 if (IS_ERR(root->kf_root)) {
1440 ret = PTR_ERR(root->kf_root);
1441 goto exit_root_id;
1442 }
1443 root_cgrp->kn = root->kf_root->kn;
d427dfeb
TH
1444
1445 ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true);
1446 if (ret)
2bd59d48 1447 goto destroy_root;
d427dfeb 1448
35585573 1449 ret = rebind_subsystems(root, ss_mask, 0);
d427dfeb 1450 if (ret)
2bd59d48 1451 goto destroy_root;
d427dfeb
TH
1452
1453 /*
1454 * There must be no failure case after here, since rebinding takes
1455 * care of subsystems' refcounts, which are explicitly dropped in
1456 * the failure exit path.
1457 */
1458 list_add(&root->root_list, &cgroup_roots);
1459 cgroup_root_count++;
1460
1461 /*
1462 * Link the top cgroup in this hierarchy into all the css_set
1463 * objects.
1464 */
96d365e0 1465 down_write(&css_set_rwsem);
d427dfeb
TH
1466 hash_for_each(css_set_table, i, cset, hlist)
1467 link_css_set(&tmp_links, cset, root_cgrp);
96d365e0 1468 up_write(&css_set_rwsem);
d427dfeb
TH
1469
1470 BUG_ON(!list_empty(&root_cgrp->children));
3c9c825b 1471 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
d427dfeb 1472
2bd59d48 1473 kernfs_activate(root_cgrp->kn);
d427dfeb 1474 ret = 0;
2bd59d48 1475 goto out;
d427dfeb 1476
2bd59d48
TH
1477destroy_root:
1478 kernfs_destroy_root(root->kf_root);
1479 root->kf_root = NULL;
1480exit_root_id:
d427dfeb 1481 cgroup_exit_root_id(root);
2bd59d48 1482out:
d427dfeb
TH
1483 free_cgrp_cset_links(&tmp_links);
1484 return ret;
1485}
1486
f7e83571 1487static struct dentry *cgroup_mount(struct file_system_type *fs_type,
ddbcc7e8 1488 int flags, const char *unused_dev_name,
f7e83571 1489 void *data)
ddbcc7e8 1490{
2bd59d48 1491 struct cgroupfs_root *root;
ddbcc7e8 1492 struct cgroup_sb_opts opts;
2bd59d48 1493 struct dentry *dentry;
8e30e2b8 1494 int ret;
56fde9e0
TH
1495
1496 /*
1497 * The first time anyone tries to mount a cgroup, enable the list
1498 * linking each css_set to its tasks and fix up all existing tasks.
1499 */
1500 if (!use_task_css_set_links)
1501 cgroup_enable_task_cg_lists();
776f02fa 1502retry:
8e30e2b8 1503 mutex_lock(&cgroup_tree_mutex);
aae8aab4 1504 mutex_lock(&cgroup_mutex);
8e30e2b8
TH
1505
1506 /* First find the desired set of subsystems */
ddbcc7e8 1507 ret = parse_cgroupfs_options(data, &opts);
c6d57f33 1508 if (ret)
8e30e2b8 1509 goto out_unlock;
ddbcc7e8 1510
2bd59d48
TH
1511 /* look for a matching existing root */
1512 for_each_active_root(root) {
1513 bool name_match = false;
ddbcc7e8 1514
2bd59d48
TH
1515 /*
1516 * If we asked for a name then it must match. Also, if
1517 * name matches but sybsys_mask doesn't, we should fail.
1518 * Remember whether name matched.
1519 */
1520 if (opts.name) {
1521 if (strcmp(opts.name, root->name))
1522 continue;
1523 name_match = true;
1524 }
ddbcc7e8 1525
c6d57f33 1526 /*
2bd59d48
TH
1527 * If we asked for subsystems (or explicitly for no
1528 * subsystems) then they must match.
c6d57f33 1529 */
2bd59d48
TH
1530 if ((opts.subsys_mask || opts.none) &&
1531 (opts.subsys_mask != root->subsys_mask)) {
1532 if (!name_match)
1533 continue;
1534 ret = -EBUSY;
1535 goto out_unlock;
1536 }
873fe09e 1537
c7ba8287 1538 if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) {
2a0ff3fb
JL
1539 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1540 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1541 ret = -EINVAL;
8e30e2b8 1542 goto out_unlock;
2a0ff3fb
JL
1543 } else {
1544 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1545 }
873fe09e 1546 }
2bd59d48 1547
776f02fa
TH
1548 /*
1549 * A root's lifetime is governed by its top cgroup. Zero
1550 * ref indicate that the root is being destroyed. Wait for
1551 * destruction to complete so that the subsystems are free.
1552 * We can use wait_queue for the wait but this path is
1553 * super cold. Let's just sleep for a bit and retry.
1554 */
1555 if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) {
1556 mutex_unlock(&cgroup_mutex);
1557 mutex_unlock(&cgroup_tree_mutex);
6534fd6c
LZ
1558 kfree(opts.release_agent);
1559 kfree(opts.name);
776f02fa
TH
1560 msleep(10);
1561 goto retry;
1562 }
1563
1564 ret = 0;
2bd59d48 1565 goto out_unlock;
ddbcc7e8
PM
1566 }
1567
2bd59d48
TH
1568 /* no such thing, create a new one */
1569 root = cgroup_root_from_opts(&opts);
1570 if (IS_ERR(root)) {
1571 ret = PTR_ERR(root);
1572 goto out_unlock;
1573 }
1574
35585573 1575 ret = cgroup_setup_root(root, opts.subsys_mask);
2bd59d48
TH
1576 if (ret)
1577 cgroup_free_root(root);
1578
8e30e2b8 1579out_unlock:
e25e2cbb 1580 mutex_unlock(&cgroup_mutex);
ace2bee8 1581 mutex_unlock(&cgroup_tree_mutex);
8e30e2b8 1582
c6d57f33
PM
1583 kfree(opts.release_agent);
1584 kfree(opts.name);
8e30e2b8 1585
2bd59d48 1586 if (ret)
8e30e2b8 1587 return ERR_PTR(ret);
2bd59d48
TH
1588
1589 dentry = kernfs_mount(fs_type, flags, root->kf_root);
1590 if (IS_ERR(dentry))
776f02fa 1591 cgroup_put(&root->top_cgroup);
2bd59d48
TH
1592 return dentry;
1593}
1594
1595static void cgroup_kill_sb(struct super_block *sb)
1596{
1597 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
1598 struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);
1599
776f02fa 1600 cgroup_put(&root->top_cgroup);
2bd59d48 1601 kernfs_kill_sb(sb);
ddbcc7e8
PM
1602}
1603
ddbcc7e8
PM
1604static struct file_system_type cgroup_fs_type = {
1605 .name = "cgroup",
f7e83571 1606 .mount = cgroup_mount,
ddbcc7e8
PM
1607 .kill_sb = cgroup_kill_sb,
1608};
1609
676db4af
GK
1610static struct kobject *cgroup_kobj;
1611
857a2beb 1612/**
913ffdb5 1613 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
857a2beb 1614 * @task: target task
857a2beb
TH
1615 * @buf: the buffer to write the path into
1616 * @buflen: the length of the buffer
1617 *
913ffdb5
TH
1618 * Determine @task's cgroup on the first (the one with the lowest non-zero
1619 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1620 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1621 * cgroup controller callbacks.
1622 *
e61734c5 1623 * Return value is the same as kernfs_path().
857a2beb 1624 */
e61734c5 1625char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
857a2beb
TH
1626{
1627 struct cgroupfs_root *root;
913ffdb5 1628 struct cgroup *cgrp;
e61734c5
TH
1629 int hierarchy_id = 1;
1630 char *path = NULL;
857a2beb
TH
1631
1632 mutex_lock(&cgroup_mutex);
96d365e0 1633 down_read(&css_set_rwsem);
857a2beb 1634
913ffdb5
TH
1635 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1636
857a2beb
TH
1637 if (root) {
1638 cgrp = task_cgroup_from_root(task, root);
e61734c5 1639 path = cgroup_path(cgrp, buf, buflen);
913ffdb5
TH
1640 } else {
1641 /* if no hierarchy exists, everyone is in "/" */
e61734c5
TH
1642 if (strlcpy(buf, "/", buflen) < buflen)
1643 path = buf;
857a2beb
TH
1644 }
1645
96d365e0 1646 up_read(&css_set_rwsem);
857a2beb 1647 mutex_unlock(&cgroup_mutex);
e61734c5 1648 return path;
857a2beb 1649}
913ffdb5 1650EXPORT_SYMBOL_GPL(task_cgroup_path);
857a2beb 1651
b3dc094e 1652/* used to track tasks and other necessary states during migration */
2f7ee569 1653struct cgroup_taskset {
b3dc094e
TH
1654 /* the src and dst cset list running through cset->mg_node */
1655 struct list_head src_csets;
1656 struct list_head dst_csets;
1657
1658 /*
1659 * Fields for cgroup_taskset_*() iteration.
1660 *
1661 * Before migration is committed, the target migration tasks are on
1662 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1663 * the csets on ->dst_csets. ->csets point to either ->src_csets
1664 * or ->dst_csets depending on whether migration is committed.
1665 *
1666 * ->cur_csets and ->cur_task point to the current task position
1667 * during iteration.
1668 */
1669 struct list_head *csets;
1670 struct css_set *cur_cset;
1671 struct task_struct *cur_task;
2f7ee569
TH
1672};
1673
1674/**
1675 * cgroup_taskset_first - reset taskset and return the first task
1676 * @tset: taskset of interest
1677 *
1678 * @tset iteration is initialized and the first task is returned.
1679 */
1680struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1681{
b3dc094e
TH
1682 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
1683 tset->cur_task = NULL;
1684
1685 return cgroup_taskset_next(tset);
2f7ee569 1686}
2f7ee569
TH
1687
1688/**
1689 * cgroup_taskset_next - iterate to the next task in taskset
1690 * @tset: taskset of interest
1691 *
1692 * Return the next task in @tset. Iteration must have been initialized
1693 * with cgroup_taskset_first().
1694 */
1695struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1696{
b3dc094e
TH
1697 struct css_set *cset = tset->cur_cset;
1698 struct task_struct *task = tset->cur_task;
2f7ee569 1699
b3dc094e
TH
1700 while (&cset->mg_node != tset->csets) {
1701 if (!task)
1702 task = list_first_entry(&cset->mg_tasks,
1703 struct task_struct, cg_list);
1704 else
1705 task = list_next_entry(task, cg_list);
2f7ee569 1706
b3dc094e
TH
1707 if (&task->cg_list != &cset->mg_tasks) {
1708 tset->cur_cset = cset;
1709 tset->cur_task = task;
1710 return task;
1711 }
1712
1713 cset = list_next_entry(cset, mg_node);
1714 task = NULL;
1715 }
1716
1717 return NULL;
2f7ee569 1718}
2f7ee569 1719
cb0f1fe9 1720/**
74a1166d 1721 * cgroup_task_migrate - move a task from one cgroup to another.
cb0f1fe9
TH
1722 * @old_cgrp; the cgroup @tsk is being migrated from
1723 * @tsk: the task being migrated
1724 * @new_cset: the new css_set @tsk is being attached to
74a1166d 1725 *
cb0f1fe9 1726 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
74a1166d 1727 */
5abb8855
TH
1728static void cgroup_task_migrate(struct cgroup *old_cgrp,
1729 struct task_struct *tsk,
1730 struct css_set *new_cset)
74a1166d 1731{
5abb8855 1732 struct css_set *old_cset;
74a1166d 1733
cb0f1fe9
TH
1734 lockdep_assert_held(&cgroup_mutex);
1735 lockdep_assert_held(&css_set_rwsem);
1736
74a1166d 1737 /*
026085ef
MSB
1738 * We are synchronized through threadgroup_lock() against PF_EXITING
1739 * setting such that we can't race against cgroup_exit() changing the
1740 * css_set to init_css_set and dropping the old one.
74a1166d 1741 */
c84cdf75 1742 WARN_ON_ONCE(tsk->flags & PF_EXITING);
a8ad805c 1743 old_cset = task_css_set(tsk);
74a1166d 1744
b3dc094e
TH
1745 get_css_set(new_cset);
1746
74a1166d 1747 task_lock(tsk);
5abb8855 1748 rcu_assign_pointer(tsk->cgroups, new_cset);
74a1166d
BB
1749 task_unlock(tsk);
1750
b3dc094e 1751 list_move(&tsk->cg_list, &new_cset->mg_tasks);
74a1166d
BB
1752
1753 /*
5abb8855
TH
1754 * We just gained a reference on old_cset by taking it from the
1755 * task. As trading it for new_cset is protected by cgroup_mutex,
1756 * we're safe to drop it here; it will be freed under RCU.
74a1166d 1757 */
5abb8855 1758 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
cb0f1fe9 1759 put_css_set_locked(old_cset, false);
74a1166d
BB
1760}
1761
a043e3b2 1762/**
1958d2d5
TH
1763 * cgroup_migrate_finish - cleanup after attach
1764 * @preloaded_csets: list of preloaded css_sets
74a1166d 1765 *
1958d2d5
TH
1766 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
1767 * those functions for details.
74a1166d 1768 */
1958d2d5
TH
1769static void cgroup_migrate_finish(struct list_head *preloaded_csets)
1770{
1771 struct css_set *cset, *tmp_cset;
1772
1773 lockdep_assert_held(&cgroup_mutex);
1774
1775 down_write(&css_set_rwsem);
1776 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
1777 cset->mg_src_cgrp = NULL;
1778 cset->mg_dst_cset = NULL;
1779 list_del_init(&cset->mg_preload_node);
1780 put_css_set_locked(cset, false);
1781 }
1782 up_write(&css_set_rwsem);
1783}
1784
1785/**
1786 * cgroup_migrate_add_src - add a migration source css_set
1787 * @src_cset: the source css_set to add
1788 * @dst_cgrp: the destination cgroup
1789 * @preloaded_csets: list of preloaded css_sets
1790 *
1791 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
1792 * @src_cset and add it to @preloaded_csets, which should later be cleaned
1793 * up by cgroup_migrate_finish().
1794 *
1795 * This function may be called without holding threadgroup_lock even if the
1796 * target is a process. Threads may be created and destroyed but as long
1797 * as cgroup_mutex is not dropped, no new css_set can be put into play and
1798 * the preloaded css_sets are guaranteed to cover all migrations.
1799 */
1800static void cgroup_migrate_add_src(struct css_set *src_cset,
1801 struct cgroup *dst_cgrp,
1802 struct list_head *preloaded_csets)
1803{
1804 struct cgroup *src_cgrp;
1805
1806 lockdep_assert_held(&cgroup_mutex);
1807 lockdep_assert_held(&css_set_rwsem);
1808
1809 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
1810
1811 /* nothing to do if this cset already belongs to the cgroup */
1812 if (src_cgrp == dst_cgrp)
1813 return;
1814
1815 if (!list_empty(&src_cset->mg_preload_node))
1816 return;
1817
1818 WARN_ON(src_cset->mg_src_cgrp);
1819 WARN_ON(!list_empty(&src_cset->mg_tasks));
1820 WARN_ON(!list_empty(&src_cset->mg_node));
1821
1822 src_cset->mg_src_cgrp = src_cgrp;
1823 get_css_set(src_cset);
1824 list_add(&src_cset->mg_preload_node, preloaded_csets);
1825}
1826
1827/**
1828 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
1829 * @dst_cgrp: the destination cgroup
1830 * @preloaded_csets: list of preloaded source css_sets
1831 *
1832 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
1833 * have been preloaded to @preloaded_csets. This function looks up and
1834 * pins all destination css_sets, links each to its source, and put them on
1835 * @preloaded_csets.
1836 *
1837 * This function must be called after cgroup_migrate_add_src() has been
1838 * called on each migration source css_set. After migration is performed
1839 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
1840 * @preloaded_csets.
1841 */
1842static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp,
1843 struct list_head *preloaded_csets)
1844{
1845 LIST_HEAD(csets);
1846 struct css_set *src_cset;
1847
1848 lockdep_assert_held(&cgroup_mutex);
1849
1850 /* look up the dst cset for each src cset and link it to src */
1851 list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) {
1852 struct css_set *dst_cset;
1853
1854 dst_cset = find_css_set(src_cset, dst_cgrp);
1855 if (!dst_cset)
1856 goto err;
1857
1858 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
1859 src_cset->mg_dst_cset = dst_cset;
1860
1861 if (list_empty(&dst_cset->mg_preload_node))
1862 list_add(&dst_cset->mg_preload_node, &csets);
1863 else
1864 put_css_set(dst_cset, false);
1865 }
1866
1867 list_splice(&csets, preloaded_csets);
1868 return 0;
1869err:
1870 cgroup_migrate_finish(&csets);
1871 return -ENOMEM;
1872}
1873
1874/**
1875 * cgroup_migrate - migrate a process or task to a cgroup
1876 * @cgrp: the destination cgroup
1877 * @leader: the leader of the process or the task to migrate
1878 * @threadgroup: whether @leader points to the whole process or a single task
1879 *
1880 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
1881 * process, the caller must be holding threadgroup_lock of @leader. The
1882 * caller is also responsible for invoking cgroup_migrate_add_src() and
1883 * cgroup_migrate_prepare_dst() on the targets before invoking this
1884 * function and following up with cgroup_migrate_finish().
1885 *
1886 * As long as a controller's ->can_attach() doesn't fail, this function is
1887 * guaranteed to succeed. This means that, excluding ->can_attach()
1888 * failure, when migrating multiple targets, the success or failure can be
1889 * decided for all targets by invoking group_migrate_prepare_dst() before
1890 * actually starting migrating.
1891 */
1892static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader,
1893 bool threadgroup)
74a1166d 1894{
b3dc094e
TH
1895 struct cgroup_taskset tset = {
1896 .src_csets = LIST_HEAD_INIT(tset.src_csets),
1897 .dst_csets = LIST_HEAD_INIT(tset.dst_csets),
1898 .csets = &tset.src_csets,
1899 };
1c6727af 1900 struct cgroup_subsys_state *css, *failed_css = NULL;
b3dc094e
TH
1901 struct css_set *cset, *tmp_cset;
1902 struct task_struct *task, *tmp_task;
1903 int i, ret;
74a1166d 1904
fb5d2b4c
MSB
1905 /*
1906 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1907 * already PF_EXITING could be freed from underneath us unless we
1908 * take an rcu_read_lock.
1909 */
b3dc094e 1910 down_write(&css_set_rwsem);
fb5d2b4c 1911 rcu_read_lock();
9db8de37 1912 task = leader;
74a1166d 1913 do {
9db8de37
TH
1914 /* @task either already exited or can't exit until the end */
1915 if (task->flags & PF_EXITING)
ea84753c 1916 goto next;
cd3d0952 1917
eaf797ab
TH
1918 /* leave @task alone if post_fork() hasn't linked it yet */
1919 if (list_empty(&task->cg_list))
1920 goto next;
1921
b3dc094e 1922 cset = task_css_set(task);
1958d2d5 1923 if (!cset->mg_src_cgrp)
ea84753c 1924 goto next;
b3dc094e 1925
b3dc094e 1926 list_move(&task->cg_list, &cset->mg_tasks);
1958d2d5
TH
1927 list_move(&cset->mg_node, &tset.src_csets);
1928 list_move(&cset->mg_dst_cset->mg_node, &tset.dst_csets);
ea84753c 1929 next:
081aa458
LZ
1930 if (!threadgroup)
1931 break;
9db8de37 1932 } while_each_thread(leader, task);
fb5d2b4c 1933 rcu_read_unlock();
b3dc094e 1934 up_write(&css_set_rwsem);
74a1166d 1935
134d3373 1936 /* methods shouldn't be called if no task is actually migrating */
b3dc094e
TH
1937 if (list_empty(&tset.src_csets))
1938 return 0;
134d3373 1939
1958d2d5 1940 /* check that we can legitimately attach to the cgroup */
1c6727af
TH
1941 for_each_css(css, i, cgrp) {
1942 if (css->ss->can_attach) {
9db8de37
TH
1943 ret = css->ss->can_attach(css, &tset);
1944 if (ret) {
1c6727af 1945 failed_css = css;
74a1166d
BB
1946 goto out_cancel_attach;
1947 }
1948 }
74a1166d
BB
1949 }
1950
1951 /*
1958d2d5
TH
1952 * Now that we're guaranteed success, proceed to move all tasks to
1953 * the new cgroup. There are no failure cases after here, so this
1954 * is the commit point.
74a1166d 1955 */
cb0f1fe9 1956 down_write(&css_set_rwsem);
b3dc094e
TH
1957 list_for_each_entry(cset, &tset.src_csets, mg_node) {
1958 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list)
1959 cgroup_task_migrate(cset->mg_src_cgrp, task,
1960 cset->mg_dst_cset);
74a1166d 1961 }
cb0f1fe9 1962 up_write(&css_set_rwsem);
b3dc094e 1963
74a1166d 1964 /*
1958d2d5
TH
1965 * Migration is committed, all target tasks are now on dst_csets.
1966 * Nothing is sensitive to fork() after this point. Notify
1967 * controllers that migration is complete.
74a1166d 1968 */
1958d2d5
TH
1969 tset.csets = &tset.dst_csets;
1970
1c6727af
TH
1971 for_each_css(css, i, cgrp)
1972 if (css->ss->attach)
1973 css->ss->attach(css, &tset);
74a1166d 1974
9db8de37 1975 ret = 0;
b3dc094e
TH
1976 goto out_release_tset;
1977
74a1166d 1978out_cancel_attach:
b3dc094e
TH
1979 for_each_css(css, i, cgrp) {
1980 if (css == failed_css)
1981 break;
1982 if (css->ss->cancel_attach)
1983 css->ss->cancel_attach(css, &tset);
74a1166d 1984 }
b3dc094e
TH
1985out_release_tset:
1986 down_write(&css_set_rwsem);
1987 list_splice_init(&tset.dst_csets, &tset.src_csets);
1988 list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) {
1989 list_splice_init(&cset->mg_tasks, &cset->tasks);
b3dc094e 1990 list_del_init(&cset->mg_node);
b3dc094e
TH
1991 }
1992 up_write(&css_set_rwsem);
9db8de37 1993 return ret;
74a1166d
BB
1994}
1995
1958d2d5
TH
1996/**
1997 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1998 * @dst_cgrp: the cgroup to attach to
1999 * @leader: the task or the leader of the threadgroup to be attached
2000 * @threadgroup: attach the whole threadgroup?
2001 *
2002 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2003 * task_lock of @tsk or each thread in the threadgroup individually in turn.
2004 */
2005static int cgroup_attach_task(struct cgroup *dst_cgrp,
2006 struct task_struct *leader, bool threadgroup)
2007{
2008 LIST_HEAD(preloaded_csets);
2009 struct task_struct *task;
2010 int ret;
2011
2012 /* look up all src csets */
2013 down_read(&css_set_rwsem);
2014 rcu_read_lock();
2015 task = leader;
2016 do {
2017 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2018 &preloaded_csets);
2019 if (!threadgroup)
2020 break;
2021 } while_each_thread(leader, task);
2022 rcu_read_unlock();
2023 up_read(&css_set_rwsem);
2024
2025 /* prepare dst csets and commit */
2026 ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets);
2027 if (!ret)
2028 ret = cgroup_migrate(dst_cgrp, leader, threadgroup);
2029
2030 cgroup_migrate_finish(&preloaded_csets);
2031 return ret;
2032}
2033
74a1166d
BB
2034/*
2035 * Find the task_struct of the task to attach by vpid and pass it along to the
cd3d0952
TH
2036 * function to attach either it or all tasks in its threadgroup. Will lock
2037 * cgroup_mutex and threadgroup; may take task_lock of task.
bbcb81d0 2038 */
74a1166d 2039static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
bbcb81d0 2040{
bbcb81d0 2041 struct task_struct *tsk;
c69e8d9c 2042 const struct cred *cred = current_cred(), *tcred;
bbcb81d0
PM
2043 int ret;
2044
74a1166d
BB
2045 if (!cgroup_lock_live_group(cgrp))
2046 return -ENODEV;
2047
b78949eb
MSB
2048retry_find_task:
2049 rcu_read_lock();
bbcb81d0 2050 if (pid) {
73507f33 2051 tsk = find_task_by_vpid(pid);
74a1166d
BB
2052 if (!tsk) {
2053 rcu_read_unlock();
dd4b0a46 2054 ret = -ESRCH;
b78949eb 2055 goto out_unlock_cgroup;
bbcb81d0 2056 }
74a1166d
BB
2057 /*
2058 * even if we're attaching all tasks in the thread group, we
2059 * only need to check permissions on one of them.
2060 */
c69e8d9c 2061 tcred = __task_cred(tsk);
14a590c3
EB
2062 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2063 !uid_eq(cred->euid, tcred->uid) &&
2064 !uid_eq(cred->euid, tcred->suid)) {
c69e8d9c 2065 rcu_read_unlock();
b78949eb
MSB
2066 ret = -EACCES;
2067 goto out_unlock_cgroup;
bbcb81d0 2068 }
b78949eb
MSB
2069 } else
2070 tsk = current;
cd3d0952
TH
2071
2072 if (threadgroup)
b78949eb 2073 tsk = tsk->group_leader;
c4c27fbd
MG
2074
2075 /*
14a40ffc 2076 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
c4c27fbd
MG
2077 * trapped in a cpuset, or RT worker may be born in a cgroup
2078 * with no rt_runtime allocated. Just say no.
2079 */
14a40ffc 2080 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
c4c27fbd
MG
2081 ret = -EINVAL;
2082 rcu_read_unlock();
2083 goto out_unlock_cgroup;
2084 }
2085
b78949eb
MSB
2086 get_task_struct(tsk);
2087 rcu_read_unlock();
2088
2089 threadgroup_lock(tsk);
2090 if (threadgroup) {
2091 if (!thread_group_leader(tsk)) {
2092 /*
2093 * a race with de_thread from another thread's exec()
2094 * may strip us of our leadership, if this happens,
2095 * there is no choice but to throw this task away and
2096 * try again; this is
2097 * "double-double-toil-and-trouble-check locking".
2098 */
2099 threadgroup_unlock(tsk);
2100 put_task_struct(tsk);
2101 goto retry_find_task;
2102 }
081aa458
LZ
2103 }
2104
2105 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2106
cd3d0952
TH
2107 threadgroup_unlock(tsk);
2108
bbcb81d0 2109 put_task_struct(tsk);
b78949eb 2110out_unlock_cgroup:
47cfcd09 2111 mutex_unlock(&cgroup_mutex);
bbcb81d0
PM
2112 return ret;
2113}
2114
7ae1bad9
TH
2115/**
2116 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2117 * @from: attach to all cgroups of a given task
2118 * @tsk: the task to be attached
2119 */
2120int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2121{
2122 struct cgroupfs_root *root;
2123 int retval = 0;
2124
47cfcd09 2125 mutex_lock(&cgroup_mutex);
7ae1bad9 2126 for_each_active_root(root) {
96d365e0
TH
2127 struct cgroup *from_cgrp;
2128
2129 down_read(&css_set_rwsem);
2130 from_cgrp = task_cgroup_from_root(from, root);
2131 up_read(&css_set_rwsem);
7ae1bad9 2132
6f4b7e63 2133 retval = cgroup_attach_task(from_cgrp, tsk, false);
7ae1bad9
TH
2134 if (retval)
2135 break;
2136 }
47cfcd09 2137 mutex_unlock(&cgroup_mutex);
7ae1bad9
TH
2138
2139 return retval;
2140}
2141EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2142
182446d0
TH
2143static int cgroup_tasks_write(struct cgroup_subsys_state *css,
2144 struct cftype *cft, u64 pid)
74a1166d 2145{
182446d0 2146 return attach_task_by_pid(css->cgroup, pid, false);
74a1166d
BB
2147}
2148
182446d0
TH
2149static int cgroup_procs_write(struct cgroup_subsys_state *css,
2150 struct cftype *cft, u64 tgid)
af351026 2151{
182446d0 2152 return attach_task_by_pid(css->cgroup, tgid, true);
af351026
PM
2153}
2154
182446d0
TH
2155static int cgroup_release_agent_write(struct cgroup_subsys_state *css,
2156 struct cftype *cft, const char *buffer)
e788e066 2157{
5f469907
TH
2158 struct cgroupfs_root *root = css->cgroup->root;
2159
2160 BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX);
182446d0 2161 if (!cgroup_lock_live_group(css->cgroup))
e788e066 2162 return -ENODEV;
69e943b7 2163 spin_lock(&release_agent_path_lock);
5f469907
TH
2164 strlcpy(root->release_agent_path, buffer,
2165 sizeof(root->release_agent_path));
69e943b7 2166 spin_unlock(&release_agent_path_lock);
47cfcd09 2167 mutex_unlock(&cgroup_mutex);
e788e066
PM
2168 return 0;
2169}
2170
2da8ca82 2171static int cgroup_release_agent_show(struct seq_file *seq, void *v)
e788e066 2172{
2da8ca82 2173 struct cgroup *cgrp = seq_css(seq)->cgroup;
182446d0 2174
e788e066
PM
2175 if (!cgroup_lock_live_group(cgrp))
2176 return -ENODEV;
2177 seq_puts(seq, cgrp->root->release_agent_path);
2178 seq_putc(seq, '\n');
47cfcd09 2179 mutex_unlock(&cgroup_mutex);
e788e066
PM
2180 return 0;
2181}
2182
2da8ca82 2183static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
873fe09e 2184{
2da8ca82
TH
2185 struct cgroup *cgrp = seq_css(seq)->cgroup;
2186
2187 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
e788e066
PM
2188 return 0;
2189}
2190
2bd59d48
TH
2191static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
2192 size_t nbytes, loff_t off)
355e0c48 2193{
2bd59d48
TH
2194 struct cgroup *cgrp = of->kn->parent->priv;
2195 struct cftype *cft = of->kn->priv;
2196 struct cgroup_subsys_state *css;
a742c59d 2197 int ret;
355e0c48 2198
2bd59d48
TH
2199 /*
2200 * kernfs guarantees that a file isn't deleted with operations in
2201 * flight, which means that the matching css is and stays alive and
2202 * doesn't need to be pinned. The RCU locking is not necessary
2203 * either. It's just for the convenience of using cgroup_css().
2204 */
2205 rcu_read_lock();
2206 css = cgroup_css(cgrp, cft->ss);
2207 rcu_read_unlock();
a742c59d
TH
2208
2209 if (cft->write_string) {
2210 ret = cft->write_string(css, cft, strstrip(buf));
2211 } else if (cft->write_u64) {
2212 unsigned long long v;
2213 ret = kstrtoull(buf, 0, &v);
2214 if (!ret)
2215 ret = cft->write_u64(css, cft, v);
2216 } else if (cft->write_s64) {
2217 long long v;
2218 ret = kstrtoll(buf, 0, &v);
2219 if (!ret)
2220 ret = cft->write_s64(css, cft, v);
2221 } else if (cft->trigger) {
2222 ret = cft->trigger(css, (unsigned int)cft->private);
e73d2c61 2223 } else {
a742c59d 2224 ret = -EINVAL;
e73d2c61 2225 }
2bd59d48 2226
a742c59d 2227 return ret ?: nbytes;
355e0c48
PM
2228}
2229
6612f05b 2230static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
db3b1497 2231{
2bd59d48 2232 return seq_cft(seq)->seq_start(seq, ppos);
db3b1497
PM
2233}
2234
6612f05b 2235static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
ddbcc7e8 2236{
2bd59d48 2237 return seq_cft(seq)->seq_next(seq, v, ppos);
ddbcc7e8
PM
2238}
2239
6612f05b 2240static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
ddbcc7e8 2241{
2bd59d48 2242 seq_cft(seq)->seq_stop(seq, v);
ddbcc7e8
PM
2243}
2244
91796569 2245static int cgroup_seqfile_show(struct seq_file *m, void *arg)
e73d2c61 2246{
7da11279
TH
2247 struct cftype *cft = seq_cft(m);
2248 struct cgroup_subsys_state *css = seq_css(m);
e73d2c61 2249
2da8ca82
TH
2250 if (cft->seq_show)
2251 return cft->seq_show(m, arg);
e73d2c61 2252
f4c753b7 2253 if (cft->read_u64)
896f5199
TH
2254 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
2255 else if (cft->read_s64)
2256 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
2257 else
2258 return -EINVAL;
2259 return 0;
91796569
PM
2260}
2261
2bd59d48
TH
2262static struct kernfs_ops cgroup_kf_single_ops = {
2263 .atomic_write_len = PAGE_SIZE,
2264 .write = cgroup_file_write,
2265 .seq_show = cgroup_seqfile_show,
91796569
PM
2266};
2267
2bd59d48
TH
2268static struct kernfs_ops cgroup_kf_ops = {
2269 .atomic_write_len = PAGE_SIZE,
2270 .write = cgroup_file_write,
2271 .seq_start = cgroup_seqfile_start,
2272 .seq_next = cgroup_seqfile_next,
2273 .seq_stop = cgroup_seqfile_stop,
2274 .seq_show = cgroup_seqfile_show,
2275};
ddbcc7e8
PM
2276
2277/*
2278 * cgroup_rename - Only allow simple rename of directories in place.
2279 */
2bd59d48
TH
2280static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
2281 const char *new_name_str)
ddbcc7e8 2282{
2bd59d48 2283 struct cgroup *cgrp = kn->priv;
2bd59d48 2284 int ret;
65dff759 2285
2bd59d48 2286 if (kernfs_type(kn) != KERNFS_DIR)
ddbcc7e8 2287 return -ENOTDIR;
2bd59d48 2288 if (kn->parent != new_parent)
ddbcc7e8 2289 return -EIO;
65dff759 2290
6db8e85c
TH
2291 /*
2292 * This isn't a proper migration and its usefulness is very
2293 * limited. Disallow if sane_behavior.
2294 */
2295 if (cgroup_sane_behavior(cgrp))
2296 return -EPERM;
2297
2bd59d48
TH
2298 mutex_lock(&cgroup_tree_mutex);
2299 mutex_lock(&cgroup_mutex);
2300
2301 ret = kernfs_rename(kn, new_parent, new_name_str);
65dff759 2302
2bd59d48
TH
2303 mutex_unlock(&cgroup_mutex);
2304 mutex_unlock(&cgroup_tree_mutex);
2bd59d48 2305 return ret;
ddbcc7e8
PM
2306}
2307
2bb566cb 2308static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft)
ddbcc7e8 2309{
8d7e6fb0 2310 char name[CGROUP_FILE_NAME_MAX];
2bd59d48
TH
2311 struct kernfs_node *kn;
2312 struct lock_class_key *key = NULL;
05ef1d7c 2313
2bd59d48
TH
2314#ifdef CONFIG_DEBUG_LOCK_ALLOC
2315 key = &cft->lockdep_key;
2316#endif
2317 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
2318 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
2319 NULL, false, key);
430af8ad 2320 return PTR_ERR_OR_ZERO(kn);
ddbcc7e8
PM
2321}
2322
b1f28d31
TH
2323/**
2324 * cgroup_addrm_files - add or remove files to a cgroup directory
2325 * @cgrp: the target cgroup
b1f28d31
TH
2326 * @cfts: array of cftypes to be added
2327 * @is_add: whether to add or remove
2328 *
2329 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2bb566cb
TH
2330 * For removals, this function never fails. If addition fails, this
2331 * function doesn't remove files already added. The caller is responsible
2332 * for cleaning up.
b1f28d31 2333 */
2bb566cb
TH
2334static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
2335 bool is_add)
ddbcc7e8 2336{
03b1cde6 2337 struct cftype *cft;
b1f28d31
TH
2338 int ret;
2339
ace2bee8 2340 lockdep_assert_held(&cgroup_tree_mutex);
db0416b6
TH
2341
2342 for (cft = cfts; cft->name[0] != '\0'; cft++) {
f33fddc2 2343 /* does cft->flags tell us to skip this file on @cgrp? */
873fe09e
TH
2344 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2345 continue;
f33fddc2
G
2346 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2347 continue;
2348 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2349 continue;
2350
2739d3cc 2351 if (is_add) {
2bb566cb 2352 ret = cgroup_add_file(cgrp, cft);
b1f28d31 2353 if (ret) {
2739d3cc 2354 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
b1f28d31
TH
2355 cft->name, ret);
2356 return ret;
2357 }
2739d3cc
LZ
2358 } else {
2359 cgroup_rm_file(cgrp, cft);
db0416b6 2360 }
ddbcc7e8 2361 }
b1f28d31 2362 return 0;
ddbcc7e8
PM
2363}
2364
21a2d343 2365static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
8e3f6541
TH
2366{
2367 LIST_HEAD(pending);
2bb566cb 2368 struct cgroup_subsys *ss = cfts[0].ss;
492eb21b 2369 struct cgroup *root = &ss->root->top_cgroup;
492eb21b 2370 struct cgroup_subsys_state *css;
9ccece80 2371 int ret = 0;
8e3f6541 2372
21a2d343 2373 lockdep_assert_held(&cgroup_tree_mutex);
4ac06017 2374
21a2d343
TH
2375 /* don't bother if @ss isn't attached */
2376 if (ss->root == &cgroup_dummy_root)
9ccece80 2377 return 0;
e8c82d20 2378
e8c82d20 2379 /* add/rm files for all cgroups created before */
ca8bdcaf 2380 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
492eb21b
TH
2381 struct cgroup *cgrp = css->cgroup;
2382
e8c82d20
LZ
2383 if (cgroup_is_dead(cgrp))
2384 continue;
2385
21a2d343 2386 ret = cgroup_addrm_files(cgrp, cfts, is_add);
9ccece80
TH
2387 if (ret)
2388 break;
8e3f6541 2389 }
21a2d343
TH
2390
2391 if (is_add && !ret)
2392 kernfs_activate(root->kn);
9ccece80 2393 return ret;
8e3f6541
TH
2394}
2395
2da440a2
TH
2396static void cgroup_exit_cftypes(struct cftype *cfts)
2397{
2398 struct cftype *cft;
2399
2bd59d48
TH
2400 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2401 /* free copy for custom atomic_write_len, see init_cftypes() */
2402 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
2403 kfree(cft->kf_ops);
2404 cft->kf_ops = NULL;
2da440a2 2405 cft->ss = NULL;
2bd59d48 2406 }
2da440a2
TH
2407}
2408
2bd59d48 2409static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2da440a2
TH
2410{
2411 struct cftype *cft;
2412
2bd59d48
TH
2413 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2414 struct kernfs_ops *kf_ops;
2415
0adb0704
TH
2416 WARN_ON(cft->ss || cft->kf_ops);
2417
2bd59d48
TH
2418 if (cft->seq_start)
2419 kf_ops = &cgroup_kf_ops;
2420 else
2421 kf_ops = &cgroup_kf_single_ops;
2422
2423 /*
2424 * Ugh... if @cft wants a custom max_write_len, we need to
2425 * make a copy of kf_ops to set its atomic_write_len.
2426 */
2427 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
2428 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
2429 if (!kf_ops) {
2430 cgroup_exit_cftypes(cfts);
2431 return -ENOMEM;
2432 }
2433 kf_ops->atomic_write_len = cft->max_write_len;
2434 }
2435
2436 cft->kf_ops = kf_ops;
2da440a2 2437 cft->ss = ss;
2bd59d48
TH
2438 }
2439
2440 return 0;
2da440a2
TH
2441}
2442
21a2d343
TH
2443static int cgroup_rm_cftypes_locked(struct cftype *cfts)
2444{
2445 lockdep_assert_held(&cgroup_tree_mutex);
2446
2447 if (!cfts || !cfts[0].ss)
2448 return -ENOENT;
2449
2450 list_del(&cfts->node);
2451 cgroup_apply_cftypes(cfts, false);
2452 cgroup_exit_cftypes(cfts);
2453 return 0;
2454}
2455
80b13586
TH
2456/**
2457 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2458 * @cfts: zero-length name terminated array of cftypes
2459 *
2460 * Unregister @cfts. Files described by @cfts are removed from all
2461 * existing cgroups and all future cgroups won't have them either. This
2462 * function can be called anytime whether @cfts' subsys is attached or not.
2463 *
2464 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2465 * registered.
2466 */
2467int cgroup_rm_cftypes(struct cftype *cfts)
2468{
21a2d343 2469 int ret;
80b13586 2470
21a2d343
TH
2471 mutex_lock(&cgroup_tree_mutex);
2472 ret = cgroup_rm_cftypes_locked(cfts);
2473 mutex_unlock(&cgroup_tree_mutex);
2474 return ret;
80b13586
TH
2475}
2476
8e3f6541
TH
2477/**
2478 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2479 * @ss: target cgroup subsystem
2480 * @cfts: zero-length name terminated array of cftypes
2481 *
2482 * Register @cfts to @ss. Files described by @cfts are created for all
2483 * existing cgroups to which @ss is attached and all future cgroups will
2484 * have them too. This function can be called anytime whether @ss is
2485 * attached or not.
2486 *
2487 * Returns 0 on successful registration, -errno on failure. Note that this
2488 * function currently returns 0 as long as @cfts registration is successful
2489 * even if some file creation attempts on existing cgroups fail.
2490 */
03b1cde6 2491int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
8e3f6541 2492{
9ccece80 2493 int ret;
8e3f6541 2494
dc5736ed
LZ
2495 if (!cfts || cfts[0].name[0] == '\0')
2496 return 0;
2497
2bd59d48
TH
2498 ret = cgroup_init_cftypes(ss, cfts);
2499 if (ret)
2500 return ret;
2bb566cb 2501
21a2d343
TH
2502 mutex_lock(&cgroup_tree_mutex);
2503
0adb0704 2504 list_add_tail(&cfts->node, &ss->cfts);
21a2d343 2505 ret = cgroup_apply_cftypes(cfts, true);
9ccece80 2506 if (ret)
21a2d343
TH
2507 cgroup_rm_cftypes_locked(cfts);
2508
2509 mutex_unlock(&cgroup_tree_mutex);
9ccece80 2510 return ret;
8e3f6541 2511}
8e3f6541 2512
a043e3b2
LZ
2513/**
2514 * cgroup_task_count - count the number of tasks in a cgroup.
2515 * @cgrp: the cgroup in question
2516 *
2517 * Return the number of tasks in the cgroup.
2518 */
07bc356e 2519static int cgroup_task_count(const struct cgroup *cgrp)
bbcb81d0
PM
2520{
2521 int count = 0;
69d0206c 2522 struct cgrp_cset_link *link;
817929ec 2523
96d365e0 2524 down_read(&css_set_rwsem);
69d0206c
TH
2525 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2526 count += atomic_read(&link->cset->refcount);
96d365e0 2527 up_read(&css_set_rwsem);
bbcb81d0
PM
2528 return count;
2529}
2530
53fa5261 2531/**
492eb21b
TH
2532 * css_next_child - find the next child of a given css
2533 * @pos_css: the current position (%NULL to initiate traversal)
2534 * @parent_css: css whose children to walk
53fa5261 2535 *
492eb21b 2536 * This function returns the next child of @parent_css and should be called
87fb54f1
TH
2537 * under either cgroup_mutex or RCU read lock. The only requirement is
2538 * that @parent_css and @pos_css are accessible. The next sibling is
2539 * guaranteed to be returned regardless of their states.
53fa5261 2540 */
492eb21b
TH
2541struct cgroup_subsys_state *
2542css_next_child(struct cgroup_subsys_state *pos_css,
2543 struct cgroup_subsys_state *parent_css)
53fa5261 2544{
492eb21b
TH
2545 struct cgroup *pos = pos_css ? pos_css->cgroup : NULL;
2546 struct cgroup *cgrp = parent_css->cgroup;
53fa5261
TH
2547 struct cgroup *next;
2548
ace2bee8 2549 cgroup_assert_mutexes_or_rcu_locked();
53fa5261
TH
2550
2551 /*
2552 * @pos could already have been removed. Once a cgroup is removed,
2553 * its ->sibling.next is no longer updated when its next sibling
ea15f8cc
TH
2554 * changes. As CGRP_DEAD assertion is serialized and happens
2555 * before the cgroup is taken off the ->sibling list, if we see it
2556 * unasserted, it's guaranteed that the next sibling hasn't
2557 * finished its grace period even if it's already removed, and thus
2558 * safe to dereference from this RCU critical section. If
2559 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
2560 * to be visible as %true here.
3b287a50
TH
2561 *
2562 * If @pos is dead, its next pointer can't be dereferenced;
2563 * however, as each cgroup is given a monotonically increasing
2564 * unique serial number and always appended to the sibling list,
2565 * the next one can be found by walking the parent's children until
2566 * we see a cgroup with higher serial number than @pos's. While
2567 * this path can be slower, it's taken only when either the current
2568 * cgroup is removed or iteration and removal race.
53fa5261 2569 */
3b287a50
TH
2570 if (!pos) {
2571 next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling);
2572 } else if (likely(!cgroup_is_dead(pos))) {
53fa5261 2573 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3b287a50
TH
2574 } else {
2575 list_for_each_entry_rcu(next, &cgrp->children, sibling)
2576 if (next->serial_nr > pos->serial_nr)
2577 break;
53fa5261
TH
2578 }
2579
492eb21b
TH
2580 if (&next->sibling == &cgrp->children)
2581 return NULL;
2582
ca8bdcaf 2583 return cgroup_css(next, parent_css->ss);
53fa5261 2584}
53fa5261 2585
574bd9f7 2586/**
492eb21b 2587 * css_next_descendant_pre - find the next descendant for pre-order walk
574bd9f7 2588 * @pos: the current position (%NULL to initiate traversal)
492eb21b 2589 * @root: css whose descendants to walk
574bd9f7 2590 *
492eb21b 2591 * To be used by css_for_each_descendant_pre(). Find the next descendant
bd8815a6
TH
2592 * to visit for pre-order traversal of @root's descendants. @root is
2593 * included in the iteration and the first node to be visited.
75501a6d 2594 *
87fb54f1
TH
2595 * While this function requires cgroup_mutex or RCU read locking, it
2596 * doesn't require the whole traversal to be contained in a single critical
2597 * section. This function will return the correct next descendant as long
2598 * as both @pos and @root are accessible and @pos is a descendant of @root.
574bd9f7 2599 */
492eb21b
TH
2600struct cgroup_subsys_state *
2601css_next_descendant_pre(struct cgroup_subsys_state *pos,
2602 struct cgroup_subsys_state *root)
574bd9f7 2603{
492eb21b 2604 struct cgroup_subsys_state *next;
574bd9f7 2605
ace2bee8 2606 cgroup_assert_mutexes_or_rcu_locked();
574bd9f7 2607
bd8815a6 2608 /* if first iteration, visit @root */
7805d000 2609 if (!pos)
bd8815a6 2610 return root;
574bd9f7
TH
2611
2612 /* visit the first child if exists */
492eb21b 2613 next = css_next_child(NULL, pos);
574bd9f7
TH
2614 if (next)
2615 return next;
2616
2617 /* no child, visit my or the closest ancestor's next sibling */
492eb21b
TH
2618 while (pos != root) {
2619 next = css_next_child(pos, css_parent(pos));
75501a6d 2620 if (next)
574bd9f7 2621 return next;
492eb21b 2622 pos = css_parent(pos);
7805d000 2623 }
574bd9f7
TH
2624
2625 return NULL;
2626}
574bd9f7 2627
12a9d2fe 2628/**
492eb21b
TH
2629 * css_rightmost_descendant - return the rightmost descendant of a css
2630 * @pos: css of interest
12a9d2fe 2631 *
492eb21b
TH
2632 * Return the rightmost descendant of @pos. If there's no descendant, @pos
2633 * is returned. This can be used during pre-order traversal to skip
12a9d2fe 2634 * subtree of @pos.
75501a6d 2635 *
87fb54f1
TH
2636 * While this function requires cgroup_mutex or RCU read locking, it
2637 * doesn't require the whole traversal to be contained in a single critical
2638 * section. This function will return the correct rightmost descendant as
2639 * long as @pos is accessible.
12a9d2fe 2640 */
492eb21b
TH
2641struct cgroup_subsys_state *
2642css_rightmost_descendant(struct cgroup_subsys_state *pos)
12a9d2fe 2643{
492eb21b 2644 struct cgroup_subsys_state *last, *tmp;
12a9d2fe 2645
ace2bee8 2646 cgroup_assert_mutexes_or_rcu_locked();
12a9d2fe
TH
2647
2648 do {
2649 last = pos;
2650 /* ->prev isn't RCU safe, walk ->next till the end */
2651 pos = NULL;
492eb21b 2652 css_for_each_child(tmp, last)
12a9d2fe
TH
2653 pos = tmp;
2654 } while (pos);
2655
2656 return last;
2657}
12a9d2fe 2658
492eb21b
TH
2659static struct cgroup_subsys_state *
2660css_leftmost_descendant(struct cgroup_subsys_state *pos)
574bd9f7 2661{
492eb21b 2662 struct cgroup_subsys_state *last;
574bd9f7
TH
2663
2664 do {
2665 last = pos;
492eb21b 2666 pos = css_next_child(NULL, pos);
574bd9f7
TH
2667 } while (pos);
2668
2669 return last;
2670}
2671
2672/**
492eb21b 2673 * css_next_descendant_post - find the next descendant for post-order walk
574bd9f7 2674 * @pos: the current position (%NULL to initiate traversal)
492eb21b 2675 * @root: css whose descendants to walk
574bd9f7 2676 *
492eb21b 2677 * To be used by css_for_each_descendant_post(). Find the next descendant
bd8815a6
TH
2678 * to visit for post-order traversal of @root's descendants. @root is
2679 * included in the iteration and the last node to be visited.
75501a6d 2680 *
87fb54f1
TH
2681 * While this function requires cgroup_mutex or RCU read locking, it
2682 * doesn't require the whole traversal to be contained in a single critical
2683 * section. This function will return the correct next descendant as long
2684 * as both @pos and @cgroup are accessible and @pos is a descendant of
2685 * @cgroup.
574bd9f7 2686 */
492eb21b
TH
2687struct cgroup_subsys_state *
2688css_next_descendant_post(struct cgroup_subsys_state *pos,
2689 struct cgroup_subsys_state *root)
574bd9f7 2690{
492eb21b 2691 struct cgroup_subsys_state *next;
574bd9f7 2692
ace2bee8 2693 cgroup_assert_mutexes_or_rcu_locked();
574bd9f7 2694
58b79a91
TH
2695 /* if first iteration, visit leftmost descendant which may be @root */
2696 if (!pos)
2697 return css_leftmost_descendant(root);
574bd9f7 2698
bd8815a6
TH
2699 /* if we visited @root, we're done */
2700 if (pos == root)
2701 return NULL;
2702
574bd9f7 2703 /* if there's an unvisited sibling, visit its leftmost descendant */
492eb21b 2704 next = css_next_child(pos, css_parent(pos));
75501a6d 2705 if (next)
492eb21b 2706 return css_leftmost_descendant(next);
574bd9f7
TH
2707
2708 /* no sibling left, visit parent */
bd8815a6 2709 return css_parent(pos);
574bd9f7 2710}
574bd9f7 2711
0942eeee 2712/**
72ec7029 2713 * css_advance_task_iter - advance a task itererator to the next css_set
0942eeee
TH
2714 * @it: the iterator to advance
2715 *
2716 * Advance @it to the next css_set to walk.
d515876e 2717 */
72ec7029 2718static void css_advance_task_iter(struct css_task_iter *it)
d515876e
TH
2719{
2720 struct list_head *l = it->cset_link;
2721 struct cgrp_cset_link *link;
2722 struct css_set *cset;
2723
2724 /* Advance to the next non-empty css_set */
2725 do {
2726 l = l->next;
72ec7029 2727 if (l == &it->origin_css->cgroup->cset_links) {
d515876e
TH
2728 it->cset_link = NULL;
2729 return;
2730 }
2731 link = list_entry(l, struct cgrp_cset_link, cset_link);
2732 cset = link->cset;
c7561128
TH
2733 } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks));
2734
d515876e 2735 it->cset_link = l;
c7561128
TH
2736
2737 if (!list_empty(&cset->tasks))
2738 it->task = cset->tasks.next;
2739 else
2740 it->task = cset->mg_tasks.next;
d515876e
TH
2741}
2742
0942eeee 2743/**
72ec7029
TH
2744 * css_task_iter_start - initiate task iteration
2745 * @css: the css to walk tasks of
0942eeee
TH
2746 * @it: the task iterator to use
2747 *
72ec7029
TH
2748 * Initiate iteration through the tasks of @css. The caller can call
2749 * css_task_iter_next() to walk through the tasks until the function
2750 * returns NULL. On completion of iteration, css_task_iter_end() must be
2751 * called.
0942eeee
TH
2752 *
2753 * Note that this function acquires a lock which is released when the
2754 * iteration finishes. The caller can't sleep while iteration is in
2755 * progress.
2756 */
72ec7029
TH
2757void css_task_iter_start(struct cgroup_subsys_state *css,
2758 struct css_task_iter *it)
96d365e0 2759 __acquires(css_set_rwsem)
817929ec 2760{
56fde9e0
TH
2761 /* no one should try to iterate before mounting cgroups */
2762 WARN_ON_ONCE(!use_task_css_set_links);
31a7df01 2763
96d365e0 2764 down_read(&css_set_rwsem);
c59cd3d8 2765
72ec7029
TH
2766 it->origin_css = css;
2767 it->cset_link = &css->cgroup->cset_links;
c59cd3d8 2768
72ec7029 2769 css_advance_task_iter(it);
817929ec
PM
2770}
2771
0942eeee 2772/**
72ec7029 2773 * css_task_iter_next - return the next task for the iterator
0942eeee
TH
2774 * @it: the task iterator being iterated
2775 *
2776 * The "next" function for task iteration. @it should have been
72ec7029
TH
2777 * initialized via css_task_iter_start(). Returns NULL when the iteration
2778 * reaches the end.
0942eeee 2779 */
72ec7029 2780struct task_struct *css_task_iter_next(struct css_task_iter *it)
817929ec
PM
2781{
2782 struct task_struct *res;
2783 struct list_head *l = it->task;
c7561128
TH
2784 struct cgrp_cset_link *link = list_entry(it->cset_link,
2785 struct cgrp_cset_link, cset_link);
817929ec
PM
2786
2787 /* If the iterator cg is NULL, we have no tasks */
69d0206c 2788 if (!it->cset_link)
817929ec
PM
2789 return NULL;
2790 res = list_entry(l, struct task_struct, cg_list);
c7561128
TH
2791
2792 /*
2793 * Advance iterator to find next entry. cset->tasks is consumed
2794 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
2795 * next cset.
2796 */
817929ec 2797 l = l->next;
c7561128
TH
2798
2799 if (l == &link->cset->tasks)
2800 l = link->cset->mg_tasks.next;
2801
2802 if (l == &link->cset->mg_tasks)
72ec7029 2803 css_advance_task_iter(it);
c7561128 2804 else
817929ec 2805 it->task = l;
c7561128 2806
817929ec
PM
2807 return res;
2808}
2809
0942eeee 2810/**
72ec7029 2811 * css_task_iter_end - finish task iteration
0942eeee
TH
2812 * @it: the task iterator to finish
2813 *
72ec7029 2814 * Finish task iteration started by css_task_iter_start().
0942eeee 2815 */
72ec7029 2816void css_task_iter_end(struct css_task_iter *it)
96d365e0 2817 __releases(css_set_rwsem)
817929ec 2818{
96d365e0 2819 up_read(&css_set_rwsem);
817929ec
PM
2820}
2821
8cc99345
TH
2822/**
2823 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
2824 * @to: cgroup to which the tasks will be moved
2825 * @from: cgroup in which the tasks currently reside
eaf797ab
TH
2826 *
2827 * Locking rules between cgroup_post_fork() and the migration path
2828 * guarantee that, if a task is forking while being migrated, the new child
2829 * is guaranteed to be either visible in the source cgroup after the
2830 * parent's migration is complete or put into the target cgroup. No task
2831 * can slip out of migration through forking.
8cc99345
TH
2832 */
2833int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
2834{
e406d1cf
TH
2835 struct css_task_iter it;
2836 struct task_struct *task;
2837 int ret = 0;
2838
2839 do {
2840 css_task_iter_start(&from->dummy_css, &it);
2841 task = css_task_iter_next(&it);
2842 if (task)
2843 get_task_struct(task);
2844 css_task_iter_end(&it);
2845
2846 if (task) {
2847 mutex_lock(&cgroup_mutex);
2848 ret = cgroup_attach_task(to, task, false);
2849 mutex_unlock(&cgroup_mutex);
2850 put_task_struct(task);
2851 }
2852 } while (task && !ret);
2853
2854 return ret;
8cc99345
TH
2855}
2856
bbcb81d0 2857/*
102a775e 2858 * Stuff for reading the 'tasks'/'procs' files.
bbcb81d0
PM
2859 *
2860 * Reading this file can return large amounts of data if a cgroup has
2861 * *lots* of attached tasks. So it may need several calls to read(),
2862 * but we cannot guarantee that the information we produce is correct
2863 * unless we produce it entirely atomically.
2864 *
bbcb81d0 2865 */
bbcb81d0 2866
24528255
LZ
2867/* which pidlist file are we talking about? */
2868enum cgroup_filetype {
2869 CGROUP_FILE_PROCS,
2870 CGROUP_FILE_TASKS,
2871};
2872
2873/*
2874 * A pidlist is a list of pids that virtually represents the contents of one
2875 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2876 * a pair (one each for procs, tasks) for each pid namespace that's relevant
2877 * to the cgroup.
2878 */
2879struct cgroup_pidlist {
2880 /*
2881 * used to find which pidlist is wanted. doesn't change as long as
2882 * this particular list stays in the list.
2883 */
2884 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
2885 /* array of xids */
2886 pid_t *list;
2887 /* how many elements the above list has */
2888 int length;
24528255
LZ
2889 /* each of these stored in a list by its cgroup */
2890 struct list_head links;
2891 /* pointer to the cgroup we belong to, for list removal purposes */
2892 struct cgroup *owner;
b1a21367
TH
2893 /* for delayed destruction */
2894 struct delayed_work destroy_dwork;
24528255
LZ
2895};
2896
d1d9fd33
BB
2897/*
2898 * The following two functions "fix" the issue where there are more pids
2899 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2900 * TODO: replace with a kernel-wide solution to this problem
2901 */
2902#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2903static void *pidlist_allocate(int count)
2904{
2905 if (PIDLIST_TOO_LARGE(count))
2906 return vmalloc(count * sizeof(pid_t));
2907 else
2908 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2909}
b1a21367 2910
d1d9fd33
BB
2911static void pidlist_free(void *p)
2912{
2913 if (is_vmalloc_addr(p))
2914 vfree(p);
2915 else
2916 kfree(p);
2917}
d1d9fd33 2918
b1a21367
TH
2919/*
2920 * Used to destroy all pidlists lingering waiting for destroy timer. None
2921 * should be left afterwards.
2922 */
2923static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
2924{
2925 struct cgroup_pidlist *l, *tmp_l;
2926
2927 mutex_lock(&cgrp->pidlist_mutex);
2928 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
2929 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
2930 mutex_unlock(&cgrp->pidlist_mutex);
2931
2932 flush_workqueue(cgroup_pidlist_destroy_wq);
2933 BUG_ON(!list_empty(&cgrp->pidlists));
2934}
2935
2936static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
2937{
2938 struct delayed_work *dwork = to_delayed_work(work);
2939 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
2940 destroy_dwork);
2941 struct cgroup_pidlist *tofree = NULL;
2942
2943 mutex_lock(&l->owner->pidlist_mutex);
b1a21367
TH
2944
2945 /*
04502365
TH
2946 * Destroy iff we didn't get queued again. The state won't change
2947 * as destroy_dwork can only be queued while locked.
b1a21367 2948 */
04502365 2949 if (!delayed_work_pending(dwork)) {
b1a21367
TH
2950 list_del(&l->links);
2951 pidlist_free(l->list);
2952 put_pid_ns(l->key.ns);
2953 tofree = l;
2954 }
2955
b1a21367
TH
2956 mutex_unlock(&l->owner->pidlist_mutex);
2957 kfree(tofree);
2958}
2959
bbcb81d0 2960/*
102a775e 2961 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
6ee211ad 2962 * Returns the number of unique elements.
bbcb81d0 2963 */
6ee211ad 2964static int pidlist_uniq(pid_t *list, int length)
bbcb81d0 2965{
102a775e 2966 int src, dest = 1;
102a775e
BB
2967
2968 /*
2969 * we presume the 0th element is unique, so i starts at 1. trivial
2970 * edge cases first; no work needs to be done for either
2971 */
2972 if (length == 0 || length == 1)
2973 return length;
2974 /* src and dest walk down the list; dest counts unique elements */
2975 for (src = 1; src < length; src++) {
2976 /* find next unique element */
2977 while (list[src] == list[src-1]) {
2978 src++;
2979 if (src == length)
2980 goto after;
2981 }
2982 /* dest always points to where the next unique element goes */
2983 list[dest] = list[src];
2984 dest++;
2985 }
2986after:
102a775e
BB
2987 return dest;
2988}
2989
afb2bc14
TH
2990/*
2991 * The two pid files - task and cgroup.procs - guaranteed that the result
2992 * is sorted, which forced this whole pidlist fiasco. As pid order is
2993 * different per namespace, each namespace needs differently sorted list,
2994 * making it impossible to use, for example, single rbtree of member tasks
2995 * sorted by task pointer. As pidlists can be fairly large, allocating one
2996 * per open file is dangerous, so cgroup had to implement shared pool of
2997 * pidlists keyed by cgroup and namespace.
2998 *
2999 * All this extra complexity was caused by the original implementation
3000 * committing to an entirely unnecessary property. In the long term, we
3001 * want to do away with it. Explicitly scramble sort order if
3002 * sane_behavior so that no such expectation exists in the new interface.
3003 *
3004 * Scrambling is done by swapping every two consecutive bits, which is
3005 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3006 */
3007static pid_t pid_fry(pid_t pid)
3008{
3009 unsigned a = pid & 0x55555555;
3010 unsigned b = pid & 0xAAAAAAAA;
3011
3012 return (a << 1) | (b >> 1);
3013}
3014
3015static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
3016{
3017 if (cgroup_sane_behavior(cgrp))
3018 return pid_fry(pid);
3019 else
3020 return pid;
3021}
3022
102a775e
BB
3023static int cmppid(const void *a, const void *b)
3024{
3025 return *(pid_t *)a - *(pid_t *)b;
3026}
3027
afb2bc14
TH
3028static int fried_cmppid(const void *a, const void *b)
3029{
3030 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
3031}
3032
e6b81710
TH
3033static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3034 enum cgroup_filetype type)
3035{
3036 struct cgroup_pidlist *l;
3037 /* don't need task_nsproxy() if we're looking at ourself */
3038 struct pid_namespace *ns = task_active_pid_ns(current);
3039
3040 lockdep_assert_held(&cgrp->pidlist_mutex);
3041
3042 list_for_each_entry(l, &cgrp->pidlists, links)
3043 if (l->key.type == type && l->key.ns == ns)
3044 return l;
3045 return NULL;
3046}
3047
72a8cb30
BB
3048/*
3049 * find the appropriate pidlist for our purpose (given procs vs tasks)
3050 * returns with the lock on that pidlist already held, and takes care
3051 * of the use count, or returns NULL with no locks held if we're out of
3052 * memory.
3053 */
e6b81710
TH
3054static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
3055 enum cgroup_filetype type)
72a8cb30
BB
3056{
3057 struct cgroup_pidlist *l;
b70cc5fd 3058
e6b81710
TH
3059 lockdep_assert_held(&cgrp->pidlist_mutex);
3060
3061 l = cgroup_pidlist_find(cgrp, type);
3062 if (l)
3063 return l;
3064
72a8cb30 3065 /* entry not found; create a new one */
f4f4be2b 3066 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
e6b81710 3067 if (!l)
72a8cb30 3068 return l;
e6b81710 3069
b1a21367 3070 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
72a8cb30 3071 l->key.type = type;
e6b81710
TH
3072 /* don't need task_nsproxy() if we're looking at ourself */
3073 l->key.ns = get_pid_ns(task_active_pid_ns(current));
72a8cb30
BB
3074 l->owner = cgrp;
3075 list_add(&l->links, &cgrp->pidlists);
72a8cb30
BB
3076 return l;
3077}
3078
102a775e
BB
3079/*
3080 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3081 */
72a8cb30
BB
3082static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3083 struct cgroup_pidlist **lp)
102a775e
BB
3084{
3085 pid_t *array;
3086 int length;
3087 int pid, n = 0; /* used for populating the array */
72ec7029 3088 struct css_task_iter it;
817929ec 3089 struct task_struct *tsk;
102a775e
BB
3090 struct cgroup_pidlist *l;
3091
4bac00d1
TH
3092 lockdep_assert_held(&cgrp->pidlist_mutex);
3093
102a775e
BB
3094 /*
3095 * If cgroup gets more users after we read count, we won't have
3096 * enough space - tough. This race is indistinguishable to the
3097 * caller from the case that the additional cgroup users didn't
3098 * show up until sometime later on.
3099 */
3100 length = cgroup_task_count(cgrp);
d1d9fd33 3101 array = pidlist_allocate(length);
102a775e
BB
3102 if (!array)
3103 return -ENOMEM;
3104 /* now, populate the array */
72ec7029
TH
3105 css_task_iter_start(&cgrp->dummy_css, &it);
3106 while ((tsk = css_task_iter_next(&it))) {
102a775e 3107 if (unlikely(n == length))
817929ec 3108 break;
102a775e 3109 /* get tgid or pid for procs or tasks file respectively */
72a8cb30
BB
3110 if (type == CGROUP_FILE_PROCS)
3111 pid = task_tgid_vnr(tsk);
3112 else
3113 pid = task_pid_vnr(tsk);
102a775e
BB
3114 if (pid > 0) /* make sure to only use valid results */
3115 array[n++] = pid;
817929ec 3116 }
72ec7029 3117 css_task_iter_end(&it);
102a775e
BB
3118 length = n;
3119 /* now sort & (if procs) strip out duplicates */
afb2bc14
TH
3120 if (cgroup_sane_behavior(cgrp))
3121 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
3122 else
3123 sort(array, length, sizeof(pid_t), cmppid, NULL);
72a8cb30 3124 if (type == CGROUP_FILE_PROCS)
6ee211ad 3125 length = pidlist_uniq(array, length);
e6b81710 3126
e6b81710 3127 l = cgroup_pidlist_find_create(cgrp, type);
72a8cb30 3128 if (!l) {
e6b81710 3129 mutex_unlock(&cgrp->pidlist_mutex);
d1d9fd33 3130 pidlist_free(array);
72a8cb30 3131 return -ENOMEM;
102a775e 3132 }
e6b81710
TH
3133
3134 /* store array, freeing old if necessary */
d1d9fd33 3135 pidlist_free(l->list);
102a775e
BB
3136 l->list = array;
3137 l->length = length;
72a8cb30 3138 *lp = l;
102a775e 3139 return 0;
bbcb81d0
PM
3140}
3141
846c7bb0 3142/**
a043e3b2 3143 * cgroupstats_build - build and fill cgroupstats
846c7bb0
BS
3144 * @stats: cgroupstats to fill information into
3145 * @dentry: A dentry entry belonging to the cgroup for which stats have
3146 * been requested.
a043e3b2
LZ
3147 *
3148 * Build and fill cgroupstats so that taskstats can export it to user
3149 * space.
846c7bb0
BS
3150 */
3151int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3152{
2bd59d48 3153 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
bd89aabc 3154 struct cgroup *cgrp;
72ec7029 3155 struct css_task_iter it;
846c7bb0 3156 struct task_struct *tsk;
33d283be 3157
2bd59d48
TH
3158 /* it should be kernfs_node belonging to cgroupfs and is a directory */
3159 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
3160 kernfs_type(kn) != KERNFS_DIR)
3161 return -EINVAL;
3162
bad34660
LZ
3163 mutex_lock(&cgroup_mutex);
3164
846c7bb0 3165 /*
2bd59d48
TH
3166 * We aren't being called from kernfs and there's no guarantee on
3167 * @kn->priv's validity. For this and css_tryget_from_dir(),
3168 * @kn->priv is RCU safe. Let's do the RCU dancing.
846c7bb0 3169 */
2bd59d48
TH
3170 rcu_read_lock();
3171 cgrp = rcu_dereference(kn->priv);
bad34660 3172 if (!cgrp || cgroup_is_dead(cgrp)) {
2bd59d48 3173 rcu_read_unlock();
bad34660 3174 mutex_unlock(&cgroup_mutex);
2bd59d48
TH
3175 return -ENOENT;
3176 }
bad34660 3177 rcu_read_unlock();
846c7bb0 3178
72ec7029
TH
3179 css_task_iter_start(&cgrp->dummy_css, &it);
3180 while ((tsk = css_task_iter_next(&it))) {
846c7bb0
BS
3181 switch (tsk->state) {
3182 case TASK_RUNNING:
3183 stats->nr_running++;
3184 break;
3185 case TASK_INTERRUPTIBLE:
3186 stats->nr_sleeping++;
3187 break;
3188 case TASK_UNINTERRUPTIBLE:
3189 stats->nr_uninterruptible++;
3190 break;
3191 case TASK_STOPPED:
3192 stats->nr_stopped++;
3193 break;
3194 default:
3195 if (delayacct_is_task_waiting_on_io(tsk))
3196 stats->nr_io_wait++;
3197 break;
3198 }
3199 }
72ec7029 3200 css_task_iter_end(&it);
846c7bb0 3201
bad34660 3202 mutex_unlock(&cgroup_mutex);
2bd59d48 3203 return 0;
846c7bb0
BS
3204}
3205
8f3ff208 3206
bbcb81d0 3207/*
102a775e 3208 * seq_file methods for the tasks/procs files. The seq_file position is the
cc31edce 3209 * next pid to display; the seq_file iterator is a pointer to the pid
102a775e 3210 * in the cgroup->l->list array.
bbcb81d0 3211 */
cc31edce 3212
102a775e 3213static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
bbcb81d0 3214{
cc31edce
PM
3215 /*
3216 * Initially we receive a position value that corresponds to
3217 * one more than the last pid shown (or 0 on the first call or
3218 * after a seek to the start). Use a binary-search to find the
3219 * next pid to display, if any
3220 */
2bd59d48 3221 struct kernfs_open_file *of = s->private;
7da11279 3222 struct cgroup *cgrp = seq_css(s)->cgroup;
4bac00d1 3223 struct cgroup_pidlist *l;
7da11279 3224 enum cgroup_filetype type = seq_cft(s)->private;
cc31edce 3225 int index = 0, pid = *pos;
4bac00d1
TH
3226 int *iter, ret;
3227
3228 mutex_lock(&cgrp->pidlist_mutex);
3229
3230 /*
5d22444f 3231 * !NULL @of->priv indicates that this isn't the first start()
4bac00d1 3232 * after open. If the matching pidlist is around, we can use that.
5d22444f 3233 * Look for it. Note that @of->priv can't be used directly. It
4bac00d1
TH
3234 * could already have been destroyed.
3235 */
5d22444f
TH
3236 if (of->priv)
3237 of->priv = cgroup_pidlist_find(cgrp, type);
4bac00d1
TH
3238
3239 /*
3240 * Either this is the first start() after open or the matching
3241 * pidlist has been destroyed inbetween. Create a new one.
3242 */
5d22444f
TH
3243 if (!of->priv) {
3244 ret = pidlist_array_load(cgrp, type,
3245 (struct cgroup_pidlist **)&of->priv);
4bac00d1
TH
3246 if (ret)
3247 return ERR_PTR(ret);
3248 }
5d22444f 3249 l = of->priv;
cc31edce 3250
cc31edce 3251 if (pid) {
102a775e 3252 int end = l->length;
20777766 3253
cc31edce
PM
3254 while (index < end) {
3255 int mid = (index + end) / 2;
afb2bc14 3256 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
cc31edce
PM
3257 index = mid;
3258 break;
afb2bc14 3259 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
cc31edce
PM
3260 index = mid + 1;
3261 else
3262 end = mid;
3263 }
3264 }
3265 /* If we're off the end of the array, we're done */
102a775e 3266 if (index >= l->length)
cc31edce
PM
3267 return NULL;
3268 /* Update the abstract position to be the actual pid that we found */
102a775e 3269 iter = l->list + index;
afb2bc14 3270 *pos = cgroup_pid_fry(cgrp, *iter);
cc31edce
PM
3271 return iter;
3272}
3273
102a775e 3274static void cgroup_pidlist_stop(struct seq_file *s, void *v)
cc31edce 3275{
2bd59d48 3276 struct kernfs_open_file *of = s->private;
5d22444f 3277 struct cgroup_pidlist *l = of->priv;
62236858 3278
5d22444f
TH
3279 if (l)
3280 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
04502365 3281 CGROUP_PIDLIST_DESTROY_DELAY);
7da11279 3282 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
cc31edce
PM
3283}
3284
102a775e 3285static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
cc31edce 3286{
2bd59d48 3287 struct kernfs_open_file *of = s->private;
5d22444f 3288 struct cgroup_pidlist *l = of->priv;
102a775e
BB
3289 pid_t *p = v;
3290 pid_t *end = l->list + l->length;
cc31edce
PM
3291 /*
3292 * Advance to the next pid in the array. If this goes off the
3293 * end, we're done
3294 */
3295 p++;
3296 if (p >= end) {
3297 return NULL;
3298 } else {
7da11279 3299 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
cc31edce
PM
3300 return p;
3301 }
3302}
3303
102a775e 3304static int cgroup_pidlist_show(struct seq_file *s, void *v)
cc31edce
PM
3305{
3306 return seq_printf(s, "%d\n", *(int *)v);
3307}
bbcb81d0 3308
102a775e
BB
3309/*
3310 * seq_operations functions for iterating on pidlists through seq_file -
3311 * independent of whether it's tasks or procs
3312 */
3313static const struct seq_operations cgroup_pidlist_seq_operations = {
3314 .start = cgroup_pidlist_start,
3315 .stop = cgroup_pidlist_stop,
3316 .next = cgroup_pidlist_next,
3317 .show = cgroup_pidlist_show,
cc31edce
PM
3318};
3319
182446d0
TH
3320static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
3321 struct cftype *cft)
81a6a5cd 3322{
182446d0 3323 return notify_on_release(css->cgroup);
81a6a5cd
PM
3324}
3325
182446d0
TH
3326static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
3327 struct cftype *cft, u64 val)
6379c106 3328{
182446d0 3329 clear_bit(CGRP_RELEASABLE, &css->cgroup->flags);
6379c106 3330 if (val)
182446d0 3331 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
6379c106 3332 else
182446d0 3333 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
6379c106
PM
3334 return 0;
3335}
3336
182446d0
TH
3337static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
3338 struct cftype *cft)
97978e6d 3339{
182446d0 3340 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
97978e6d
DL
3341}
3342
182446d0
TH
3343static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
3344 struct cftype *cft, u64 val)
97978e6d
DL
3345{
3346 if (val)
182446d0 3347 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
97978e6d 3348 else
182446d0 3349 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
97978e6d
DL
3350 return 0;
3351}
3352
d5c56ced 3353static struct cftype cgroup_base_files[] = {
81a6a5cd 3354 {
d5c56ced 3355 .name = "cgroup.procs",
6612f05b
TH
3356 .seq_start = cgroup_pidlist_start,
3357 .seq_next = cgroup_pidlist_next,
3358 .seq_stop = cgroup_pidlist_stop,
3359 .seq_show = cgroup_pidlist_show,
5d22444f 3360 .private = CGROUP_FILE_PROCS,
74a1166d 3361 .write_u64 = cgroup_procs_write,
74a1166d 3362 .mode = S_IRUGO | S_IWUSR,
102a775e 3363 },
97978e6d
DL
3364 {
3365 .name = "cgroup.clone_children",
873fe09e 3366 .flags = CFTYPE_INSANE,
97978e6d
DL
3367 .read_u64 = cgroup_clone_children_read,
3368 .write_u64 = cgroup_clone_children_write,
3369 },
873fe09e
TH
3370 {
3371 .name = "cgroup.sane_behavior",
3372 .flags = CFTYPE_ONLY_ON_ROOT,
2da8ca82 3373 .seq_show = cgroup_sane_behavior_show,
873fe09e 3374 },
d5c56ced
TH
3375
3376 /*
3377 * Historical crazy stuff. These don't have "cgroup." prefix and
3378 * don't exist if sane_behavior. If you're depending on these, be
3379 * prepared to be burned.
3380 */
3381 {
3382 .name = "tasks",
3383 .flags = CFTYPE_INSANE, /* use "procs" instead */
6612f05b
TH
3384 .seq_start = cgroup_pidlist_start,
3385 .seq_next = cgroup_pidlist_next,
3386 .seq_stop = cgroup_pidlist_stop,
3387 .seq_show = cgroup_pidlist_show,
5d22444f 3388 .private = CGROUP_FILE_TASKS,
d5c56ced 3389 .write_u64 = cgroup_tasks_write,
d5c56ced
TH
3390 .mode = S_IRUGO | S_IWUSR,
3391 },
3392 {
3393 .name = "notify_on_release",
3394 .flags = CFTYPE_INSANE,
3395 .read_u64 = cgroup_read_notify_on_release,
3396 .write_u64 = cgroup_write_notify_on_release,
3397 },
6e6ff25b
TH
3398 {
3399 .name = "release_agent",
cc5943a7 3400 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
2da8ca82 3401 .seq_show = cgroup_release_agent_show,
6e6ff25b 3402 .write_string = cgroup_release_agent_write,
5f469907 3403 .max_write_len = PATH_MAX - 1,
6e6ff25b 3404 },
db0416b6 3405 { } /* terminate */
bbcb81d0
PM
3406};
3407
13af07df 3408/**
628f7cd4 3409 * cgroup_populate_dir - create subsys files in a cgroup directory
13af07df 3410 * @cgrp: target cgroup
13af07df 3411 * @subsys_mask: mask of the subsystem ids whose files should be added
bee55099
TH
3412 *
3413 * On failure, no file is added.
13af07df 3414 */
628f7cd4 3415static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask)
ddbcc7e8 3416{
ddbcc7e8 3417 struct cgroup_subsys *ss;
b420ba7d 3418 int i, ret = 0;
bbcb81d0 3419
8e3f6541 3420 /* process cftsets of each subsystem */
b420ba7d 3421 for_each_subsys(ss, i) {
0adb0704 3422 struct cftype *cfts;
b420ba7d
TH
3423
3424 if (!test_bit(i, &subsys_mask))
13af07df 3425 continue;
8e3f6541 3426
0adb0704
TH
3427 list_for_each_entry(cfts, &ss->cfts, node) {
3428 ret = cgroup_addrm_files(cgrp, cfts, true);
bee55099
TH
3429 if (ret < 0)
3430 goto err;
3431 }
ddbcc7e8 3432 }
ddbcc7e8 3433 return 0;
bee55099
TH
3434err:
3435 cgroup_clear_dir(cgrp, subsys_mask);
3436 return ret;
ddbcc7e8
PM
3437}
3438
0c21ead1
TH
3439/*
3440 * css destruction is four-stage process.
3441 *
3442 * 1. Destruction starts. Killing of the percpu_ref is initiated.
3443 * Implemented in kill_css().
3444 *
3445 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
3446 * and thus css_tryget() is guaranteed to fail, the css can be offlined
3447 * by invoking offline_css(). After offlining, the base ref is put.
3448 * Implemented in css_killed_work_fn().
3449 *
3450 * 3. When the percpu_ref reaches zero, the only possible remaining
3451 * accessors are inside RCU read sections. css_release() schedules the
3452 * RCU callback.
3453 *
3454 * 4. After the grace period, the css can be freed. Implemented in
3455 * css_free_work_fn().
3456 *
3457 * It is actually hairier because both step 2 and 4 require process context
3458 * and thus involve punting to css->destroy_work adding two additional
3459 * steps to the already complex sequence.
3460 */
35ef10da 3461static void css_free_work_fn(struct work_struct *work)
48ddbe19
TH
3462{
3463 struct cgroup_subsys_state *css =
35ef10da 3464 container_of(work, struct cgroup_subsys_state, destroy_work);
0c21ead1 3465 struct cgroup *cgrp = css->cgroup;
48ddbe19 3466
0ae78e0b
TH
3467 if (css->parent)
3468 css_put(css->parent);
3469
0c21ead1 3470 css->ss->css_free(css);
2bd59d48 3471 cgroup_put(cgrp);
48ddbe19
TH
3472}
3473
0c21ead1 3474static void css_free_rcu_fn(struct rcu_head *rcu_head)
d3daf28d
TH
3475{
3476 struct cgroup_subsys_state *css =
0c21ead1 3477 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
d3daf28d 3478
35ef10da 3479 INIT_WORK(&css->destroy_work, css_free_work_fn);
e5fca243 3480 queue_work(cgroup_destroy_wq, &css->destroy_work);
48ddbe19
TH
3481}
3482
d3daf28d
TH
3483static void css_release(struct percpu_ref *ref)
3484{
3485 struct cgroup_subsys_state *css =
3486 container_of(ref, struct cgroup_subsys_state, refcnt);
3487
aec25020 3488 rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL);
0c21ead1 3489 call_rcu(&css->rcu_head, css_free_rcu_fn);
d3daf28d
TH
3490}
3491
623f926b
TH
3492static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss,
3493 struct cgroup *cgrp)
ddbcc7e8 3494{
bd89aabc 3495 css->cgroup = cgrp;
72c97e54 3496 css->ss = ss;
ddbcc7e8 3497 css->flags = 0;
0ae78e0b
TH
3498
3499 if (cgrp->parent)
ca8bdcaf 3500 css->parent = cgroup_css(cgrp->parent, ss);
0ae78e0b 3501 else
38b53aba 3502 css->flags |= CSS_ROOT;
48ddbe19 3503
ca8bdcaf 3504 BUG_ON(cgroup_css(cgrp, ss));
ddbcc7e8
PM
3505}
3506
2a4ac633 3507/* invoke ->css_online() on a new CSS and mark it online if successful */
623f926b 3508static int online_css(struct cgroup_subsys_state *css)
a31f2d3f 3509{
623f926b 3510 struct cgroup_subsys *ss = css->ss;
b1929db4
TH
3511 int ret = 0;
3512
ace2bee8 3513 lockdep_assert_held(&cgroup_tree_mutex);
a31f2d3f
TH
3514 lockdep_assert_held(&cgroup_mutex);
3515
92fb9748 3516 if (ss->css_online)
eb95419b 3517 ret = ss->css_online(css);
ae7f164a 3518 if (!ret) {
eb95419b 3519 css->flags |= CSS_ONLINE;
f20104de 3520 css->cgroup->nr_css++;
aec25020 3521 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
ae7f164a 3522 }
b1929db4 3523 return ret;
a31f2d3f
TH
3524}
3525
2a4ac633 3526/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
623f926b 3527static void offline_css(struct cgroup_subsys_state *css)
a31f2d3f 3528{
623f926b 3529 struct cgroup_subsys *ss = css->ss;
a31f2d3f 3530
ace2bee8 3531 lockdep_assert_held(&cgroup_tree_mutex);
a31f2d3f
TH
3532 lockdep_assert_held(&cgroup_mutex);
3533
3534 if (!(css->flags & CSS_ONLINE))
3535 return;
3536
d7eeac19 3537 if (ss->css_offline)
eb95419b 3538 ss->css_offline(css);
a31f2d3f 3539
eb95419b 3540 css->flags &= ~CSS_ONLINE;
09a503ea 3541 css->cgroup->nr_css--;
aec25020 3542 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css);
a31f2d3f
TH
3543}
3544
c81c925a
TH
3545/**
3546 * create_css - create a cgroup_subsys_state
3547 * @cgrp: the cgroup new css will be associated with
3548 * @ss: the subsys of new css
3549 *
3550 * Create a new css associated with @cgrp - @ss pair. On success, the new
3551 * css is online and installed in @cgrp with all interface files created.
3552 * Returns 0 on success, -errno on failure.
3553 */
3554static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss)
3555{
3556 struct cgroup *parent = cgrp->parent;
3557 struct cgroup_subsys_state *css;
3558 int err;
3559
c81c925a
TH
3560 lockdep_assert_held(&cgroup_mutex);
3561
3562 css = ss->css_alloc(cgroup_css(parent, ss));
3563 if (IS_ERR(css))
3564 return PTR_ERR(css);
3565
3566 err = percpu_ref_init(&css->refcnt, css_release);
3567 if (err)
3568 goto err_free;
3569
3570 init_css(css, ss, cgrp);
3571
aec25020 3572 err = cgroup_populate_dir(cgrp, 1 << ss->id);
c81c925a
TH
3573 if (err)
3574 goto err_free;
3575
3576 err = online_css(css);
3577 if (err)
3578 goto err_free;
3579
59f5296b 3580 cgroup_get(cgrp);
c81c925a
TH
3581 css_get(css->parent);
3582
3583 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
3584 parent->parent) {
3585 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
3586 current->comm, current->pid, ss->name);
3587 if (!strcmp(ss->name, "memory"))
3588 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
3589 ss->warned_broken_hierarchy = true;
3590 }
3591
3592 return 0;
3593
3594err_free:
3595 percpu_ref_cancel_init(&css->refcnt);
3596 ss->css_free(css);
3597 return err;
3598}
3599
2bd59d48 3600/**
a043e3b2
LZ
3601 * cgroup_create - create a cgroup
3602 * @parent: cgroup that will be parent of the new cgroup
e61734c5 3603 * @name: name of the new cgroup
2bd59d48 3604 * @mode: mode to set on new cgroup
ddbcc7e8 3605 */
e61734c5 3606static long cgroup_create(struct cgroup *parent, const char *name,
2bd59d48 3607 umode_t mode)
ddbcc7e8 3608{
bd89aabc 3609 struct cgroup *cgrp;
ddbcc7e8 3610 struct cgroupfs_root *root = parent->root;
b58c8998 3611 int ssid, err;
ddbcc7e8 3612 struct cgroup_subsys *ss;
2bd59d48 3613 struct kernfs_node *kn;
ddbcc7e8 3614
0a950f65 3615 /* allocate the cgroup and its ID, 0 is reserved for the root */
bd89aabc
PM
3616 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3617 if (!cgrp)
ddbcc7e8
PM
3618 return -ENOMEM;
3619
ace2bee8
TH
3620 mutex_lock(&cgroup_tree_mutex);
3621
976c06bc
TH
3622 /*
3623 * Only live parents can have children. Note that the liveliness
3624 * check isn't strictly necessary because cgroup_mkdir() and
3625 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
3626 * anyway so that locking is contained inside cgroup proper and we
3627 * don't get nasty surprises if we ever grow another caller.
3628 */
3629 if (!cgroup_lock_live_group(parent)) {
3630 err = -ENODEV;
ace2bee8 3631 goto err_unlock_tree;
0ab02ca8
LZ
3632 }
3633
3634 /*
3635 * Temporarily set the pointer to NULL, so idr_find() won't return
3636 * a half-baked cgroup.
3637 */
3638 cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
3639 if (cgrp->id < 0) {
3640 err = -ENOMEM;
3641 goto err_unlock;
976c06bc
TH
3642 }
3643
cc31edce 3644 init_cgroup_housekeeping(cgrp);
ddbcc7e8 3645
bd89aabc 3646 cgrp->parent = parent;
0ae78e0b 3647 cgrp->dummy_css.parent = &parent->dummy_css;
bd89aabc 3648 cgrp->root = parent->root;
ddbcc7e8 3649
b6abdb0e
LZ
3650 if (notify_on_release(parent))
3651 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3652
2260e7fc
TH
3653 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
3654 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
97978e6d 3655
2bd59d48 3656 /* create the directory */
e61734c5 3657 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
2bd59d48
TH
3658 if (IS_ERR(kn)) {
3659 err = PTR_ERR(kn);
0ab02ca8 3660 goto err_free_id;
2bd59d48
TH
3661 }
3662 cgrp->kn = kn;
ddbcc7e8 3663
6f30558f
TH
3664 /*
3665 * This extra ref will be put in cgroup_free_fn() and guarantees
3666 * that @cgrp->kn is always accessible.
3667 */
3668 kernfs_get(kn);
3669
00356bd5 3670 cgrp->serial_nr = cgroup_serial_nr_next++;
53fa5261 3671
4e139afc 3672 /* allocation complete, commit to creation */
4e139afc 3673 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
3c9c825b 3674 atomic_inc(&root->nr_cgrps);
59f5296b 3675 cgroup_get(parent);
415cf07a 3676
0d80255e
TH
3677 /*
3678 * @cgrp is now fully operational. If something fails after this
3679 * point, it'll be released via the normal destruction path.
3680 */
4e96ee8e
LZ
3681 idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
3682
2bb566cb 3683 err = cgroup_addrm_files(cgrp, cgroup_base_files, true);
628f7cd4
TH
3684 if (err)
3685 goto err_destroy;
3686
9d403e99 3687 /* let's create and online css's */
b85d2040
TH
3688 for_each_subsys(ss, ssid) {
3689 if (root->subsys_mask & (1 << ssid)) {
3690 err = create_css(cgrp, ss);
3691 if (err)
3692 goto err_destroy;
3693 }
a8638030 3694 }
ddbcc7e8 3695
2bd59d48
TH
3696 kernfs_activate(kn);
3697
ddbcc7e8 3698 mutex_unlock(&cgroup_mutex);
ace2bee8 3699 mutex_unlock(&cgroup_tree_mutex);
ddbcc7e8
PM
3700
3701 return 0;
3702
0a950f65 3703err_free_id:
4e96ee8e 3704 idr_remove(&root->cgroup_idr, cgrp->id);
0ab02ca8
LZ
3705err_unlock:
3706 mutex_unlock(&cgroup_mutex);
ace2bee8
TH
3707err_unlock_tree:
3708 mutex_unlock(&cgroup_tree_mutex);
bd89aabc 3709 kfree(cgrp);
ddbcc7e8 3710 return err;
4b8b47eb
TH
3711
3712err_destroy:
3713 cgroup_destroy_locked(cgrp);
3714 mutex_unlock(&cgroup_mutex);
ace2bee8 3715 mutex_unlock(&cgroup_tree_mutex);
4b8b47eb 3716 return err;
ddbcc7e8
PM
3717}
3718
2bd59d48
TH
3719static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
3720 umode_t mode)
ddbcc7e8 3721{
2bd59d48 3722 struct cgroup *parent = parent_kn->priv;
ddbcc7e8 3723
2bd59d48 3724 return cgroup_create(parent, name, mode);
ddbcc7e8
PM
3725}
3726
223dbc38
TH
3727/*
3728 * This is called when the refcnt of a css is confirmed to be killed.
3729 * css_tryget() is now guaranteed to fail.
3730 */
3731static void css_killed_work_fn(struct work_struct *work)
d3daf28d 3732{
223dbc38
TH
3733 struct cgroup_subsys_state *css =
3734 container_of(work, struct cgroup_subsys_state, destroy_work);
3735 struct cgroup *cgrp = css->cgroup;
d3daf28d 3736
ace2bee8 3737 mutex_lock(&cgroup_tree_mutex);
f20104de
TH
3738 mutex_lock(&cgroup_mutex);
3739
09a503ea
TH
3740 /*
3741 * css_tryget() is guaranteed to fail now. Tell subsystems to
3742 * initate destruction.
3743 */
3744 offline_css(css);
3745
f20104de
TH
3746 /*
3747 * If @cgrp is marked dead, it's waiting for refs of all css's to
3748 * be disabled before proceeding to the second phase of cgroup
3749 * destruction. If we are the last one, kick it off.
3750 */
09a503ea 3751 if (!cgrp->nr_css && cgroup_is_dead(cgrp))
f20104de
TH
3752 cgroup_destroy_css_killed(cgrp);
3753
3754 mutex_unlock(&cgroup_mutex);
ace2bee8 3755 mutex_unlock(&cgroup_tree_mutex);
09a503ea
TH
3756
3757 /*
3758 * Put the css refs from kill_css(). Each css holds an extra
3759 * reference to the cgroup's dentry and cgroup removal proceeds
3760 * regardless of css refs. On the last put of each css, whenever
3761 * that may be, the extra dentry ref is put so that dentry
3762 * destruction happens only after all css's are released.
3763 */
3764 css_put(css);
d3daf28d
TH
3765}
3766
223dbc38
TH
3767/* css kill confirmation processing requires process context, bounce */
3768static void css_killed_ref_fn(struct percpu_ref *ref)
d3daf28d
TH
3769{
3770 struct cgroup_subsys_state *css =
3771 container_of(ref, struct cgroup_subsys_state, refcnt);
3772
223dbc38 3773 INIT_WORK(&css->destroy_work, css_killed_work_fn);
e5fca243 3774 queue_work(cgroup_destroy_wq, &css->destroy_work);
d3daf28d
TH
3775}
3776
edae0c33
TH
3777/**
3778 * kill_css - destroy a css
3779 * @css: css to destroy
3780 *
3c14f8b4
TH
3781 * This function initiates destruction of @css by removing cgroup interface
3782 * files and putting its base reference. ->css_offline() will be invoked
3783 * asynchronously once css_tryget() is guaranteed to fail and when the
3784 * reference count reaches zero, @css will be released.
edae0c33
TH
3785 */
3786static void kill_css(struct cgroup_subsys_state *css)
3787{
2bd59d48
TH
3788 /*
3789 * This must happen before css is disassociated with its cgroup.
3790 * See seq_css() for details.
3791 */
aec25020 3792 cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
3c14f8b4 3793
edae0c33
TH
3794 /*
3795 * Killing would put the base ref, but we need to keep it alive
3796 * until after ->css_offline().
3797 */
3798 css_get(css);
3799
3800 /*
3801 * cgroup core guarantees that, by the time ->css_offline() is
3802 * invoked, no new css reference will be given out via
3803 * css_tryget(). We can't simply call percpu_ref_kill() and
3804 * proceed to offlining css's because percpu_ref_kill() doesn't
3805 * guarantee that the ref is seen as killed on all CPUs on return.
3806 *
3807 * Use percpu_ref_kill_and_confirm() to get notifications as each
3808 * css is confirmed to be seen as killed on all CPUs.
3809 */
3810 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
d3daf28d
TH
3811}
3812
3813/**
3814 * cgroup_destroy_locked - the first stage of cgroup destruction
3815 * @cgrp: cgroup to be destroyed
3816 *
3817 * css's make use of percpu refcnts whose killing latency shouldn't be
3818 * exposed to userland and are RCU protected. Also, cgroup core needs to
3819 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
3820 * invoked. To satisfy all the requirements, destruction is implemented in
3821 * the following two steps.
3822 *
3823 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
3824 * userland visible parts and start killing the percpu refcnts of
3825 * css's. Set up so that the next stage will be kicked off once all
3826 * the percpu refcnts are confirmed to be killed.
3827 *
3828 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
3829 * rest of destruction. Once all cgroup references are gone, the
3830 * cgroup is RCU-freed.
3831 *
3832 * This function implements s1. After this step, @cgrp is gone as far as
3833 * the userland is concerned and a new cgroup with the same name may be
3834 * created. As cgroup doesn't care about the names internally, this
3835 * doesn't cause any problem.
3836 */
42809dd4
TH
3837static int cgroup_destroy_locked(struct cgroup *cgrp)
3838 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
ddbcc7e8 3839{
bb78a92f 3840 struct cgroup *child;
2bd59d48 3841 struct cgroup_subsys_state *css;
ddd69148 3842 bool empty;
1c6727af 3843 int ssid;
ddbcc7e8 3844
ace2bee8 3845 lockdep_assert_held(&cgroup_tree_mutex);
42809dd4
TH
3846 lockdep_assert_held(&cgroup_mutex);
3847
ddd69148 3848 /*
96d365e0 3849 * css_set_rwsem synchronizes access to ->cset_links and prevents
89c5509b 3850 * @cgrp from being removed while put_css_set() is in progress.
ddd69148 3851 */
96d365e0 3852 down_read(&css_set_rwsem);
bb78a92f 3853 empty = list_empty(&cgrp->cset_links);
96d365e0 3854 up_read(&css_set_rwsem);
ddd69148 3855 if (!empty)
ddbcc7e8 3856 return -EBUSY;
a043e3b2 3857
bb78a92f
HD
3858 /*
3859 * Make sure there's no live children. We can't test ->children
3860 * emptiness as dead children linger on it while being destroyed;
3861 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
3862 */
3863 empty = true;
3864 rcu_read_lock();
3865 list_for_each_entry_rcu(child, &cgrp->children, sibling) {
3866 empty = cgroup_is_dead(child);
3867 if (!empty)
3868 break;
3869 }
3870 rcu_read_unlock();
3871 if (!empty)
3872 return -EBUSY;
3873
88703267 3874 /*
edae0c33
TH
3875 * Initiate massacre of all css's. cgroup_destroy_css_killed()
3876 * will be invoked to perform the rest of destruction once the
4ac06017
TH
3877 * percpu refs of all css's are confirmed to be killed. This
3878 * involves removing the subsystem's files, drop cgroup_mutex.
88703267 3879 */
4ac06017 3880 mutex_unlock(&cgroup_mutex);
1c6727af
TH
3881 for_each_css(css, ssid, cgrp)
3882 kill_css(css);
4ac06017 3883 mutex_lock(&cgroup_mutex);
455050d2
TH
3884
3885 /*
3886 * Mark @cgrp dead. This prevents further task migration and child
3887 * creation by disabling cgroup_lock_live_group(). Note that
492eb21b 3888 * CGRP_DEAD assertion is depended upon by css_next_child() to
455050d2 3889 * resume iteration after dropping RCU read lock. See
492eb21b 3890 * css_next_child() for details.
455050d2 3891 */
54766d4a 3892 set_bit(CGRP_DEAD, &cgrp->flags);
ddbcc7e8 3893
455050d2
TH
3894 /* CGRP_DEAD is set, remove from ->release_list for the last time */
3895 raw_spin_lock(&release_list_lock);
3896 if (!list_empty(&cgrp->release_list))
3897 list_del_init(&cgrp->release_list);
3898 raw_spin_unlock(&release_list_lock);
3899
3900 /*
f20104de
TH
3901 * If @cgrp has css's attached, the second stage of cgroup
3902 * destruction is kicked off from css_killed_work_fn() after the
3903 * refs of all attached css's are killed. If @cgrp doesn't have
3904 * any css, we kick it off here.
3905 */
3906 if (!cgrp->nr_css)
3907 cgroup_destroy_css_killed(cgrp);
3908
2bd59d48
TH
3909 /* remove @cgrp directory along with the base files */
3910 mutex_unlock(&cgroup_mutex);
3911
455050d2 3912 /*
2bd59d48
TH
3913 * There are two control paths which try to determine cgroup from
3914 * dentry without going through kernfs - cgroupstats_build() and
3915 * css_tryget_from_dir(). Those are supported by RCU protecting
3916 * clearing of cgrp->kn->priv backpointer, which should happen
3917 * after all files under it have been removed.
455050d2 3918 */
6f30558f 3919 kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */
2bd59d48 3920 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL);
2bd59d48 3921
4ac06017 3922 mutex_lock(&cgroup_mutex);
455050d2 3923
ea15f8cc
TH
3924 return 0;
3925};
3926
d3daf28d 3927/**
f20104de 3928 * cgroup_destroy_css_killed - the second step of cgroup destruction
d3daf28d
TH
3929 * @work: cgroup->destroy_free_work
3930 *
3931 * This function is invoked from a work item for a cgroup which is being
09a503ea
TH
3932 * destroyed after all css's are offlined and performs the rest of
3933 * destruction. This is the second step of destruction described in the
3934 * comment above cgroup_destroy_locked().
d3daf28d 3935 */
f20104de 3936static void cgroup_destroy_css_killed(struct cgroup *cgrp)
ea15f8cc 3937{
ea15f8cc 3938 struct cgroup *parent = cgrp->parent;
ea15f8cc 3939
ace2bee8 3940 lockdep_assert_held(&cgroup_tree_mutex);
f20104de 3941 lockdep_assert_held(&cgroup_mutex);
ea15f8cc 3942
999cd8a4 3943 /* delete this cgroup from parent->children */
eb6fd504 3944 list_del_rcu(&cgrp->sibling);
ed957793 3945
59f5296b 3946 cgroup_put(cgrp);
ddbcc7e8 3947
bd89aabc 3948 set_bit(CGRP_RELEASABLE, &parent->flags);
81a6a5cd 3949 check_for_release(parent);
ddbcc7e8
PM
3950}
3951
2bd59d48 3952static int cgroup_rmdir(struct kernfs_node *kn)
42809dd4 3953{
2bd59d48
TH
3954 struct cgroup *cgrp = kn->priv;
3955 int ret = 0;
3956
3957 /*
3958 * This is self-destruction but @kn can't be removed while this
3959 * callback is in progress. Let's break active protection. Once
3960 * the protection is broken, @cgrp can be destroyed at any point.
3961 * Pin it so that it stays accessible.
3962 */
3963 cgroup_get(cgrp);
3964 kernfs_break_active_protection(kn);
42809dd4 3965
ace2bee8 3966 mutex_lock(&cgroup_tree_mutex);
42809dd4 3967 mutex_lock(&cgroup_mutex);
2bd59d48
TH
3968
3969 /*
3970 * @cgrp might already have been destroyed while we're trying to
3971 * grab the mutexes.
3972 */
3973 if (!cgroup_is_dead(cgrp))
3974 ret = cgroup_destroy_locked(cgrp);
3975
42809dd4 3976 mutex_unlock(&cgroup_mutex);
ace2bee8 3977 mutex_unlock(&cgroup_tree_mutex);
42809dd4 3978
2bd59d48
TH
3979 kernfs_unbreak_active_protection(kn);
3980 cgroup_put(cgrp);
42809dd4
TH
3981 return ret;
3982}
3983
2bd59d48
TH
3984static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
3985 .remount_fs = cgroup_remount,
3986 .show_options = cgroup_show_options,
3987 .mkdir = cgroup_mkdir,
3988 .rmdir = cgroup_rmdir,
3989 .rename = cgroup_rename,
3990};
3991
06a11920 3992static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
ddbcc7e8 3993{
ddbcc7e8 3994 struct cgroup_subsys_state *css;
cfe36bde
DC
3995
3996 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
ddbcc7e8 3997
ace2bee8 3998 mutex_lock(&cgroup_tree_mutex);
648bb56d
TH
3999 mutex_lock(&cgroup_mutex);
4000
0adb0704 4001 INIT_LIST_HEAD(&ss->cfts);
8e3f6541 4002
ddbcc7e8 4003 /* Create the top cgroup state for this subsystem */
9871bf95 4004 ss->root = &cgroup_dummy_root;
ca8bdcaf 4005 css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss));
ddbcc7e8
PM
4006 /* We don't handle early failures gracefully */
4007 BUG_ON(IS_ERR(css));
623f926b 4008 init_css(css, ss, cgroup_dummy_top);
ddbcc7e8 4009
e8d55fde 4010 /* Update the init_css_set to contain a subsys
817929ec 4011 * pointer to this state - since the subsystem is
e8d55fde
LZ
4012 * newly registered, all tasks and hence the
4013 * init_css_set is in the subsystem's top cgroup. */
aec25020 4014 init_css_set.subsys[ss->id] = css;
ddbcc7e8
PM
4015
4016 need_forkexit_callback |= ss->fork || ss->exit;
4017
e8d55fde
LZ
4018 /* At system boot, before all subsystems have been
4019 * registered, no tasks have been forked, so we don't
4020 * need to invoke fork callbacks here. */
4021 BUG_ON(!list_empty(&init_task.tasks));
4022
ae7f164a 4023 BUG_ON(online_css(css));
a8638030 4024
648bb56d 4025 mutex_unlock(&cgroup_mutex);
ace2bee8 4026 mutex_unlock(&cgroup_tree_mutex);
e6a1105b
BB
4027}
4028
ddbcc7e8 4029/**
a043e3b2
LZ
4030 * cgroup_init_early - cgroup initialization at system boot
4031 *
4032 * Initialize cgroups at system boot, and initialize any
4033 * subsystems that request early init.
ddbcc7e8
PM
4034 */
4035int __init cgroup_init_early(void)
4036{
30159ec7 4037 struct cgroup_subsys *ss;
ddbcc7e8 4038 int i;
30159ec7 4039
146aa1bd 4040 atomic_set(&init_css_set.refcount, 1);
69d0206c 4041 INIT_LIST_HEAD(&init_css_set.cgrp_links);
817929ec 4042 INIT_LIST_HEAD(&init_css_set.tasks);
b3dc094e 4043 INIT_LIST_HEAD(&init_css_set.mg_tasks);
1958d2d5 4044 INIT_LIST_HEAD(&init_css_set.mg_preload_node);
b3dc094e 4045 INIT_LIST_HEAD(&init_css_set.mg_node);
472b1053 4046 INIT_HLIST_NODE(&init_css_set.hlist);
817929ec 4047 css_set_count = 1;
9871bf95
TH
4048 init_cgroup_root(&cgroup_dummy_root);
4049 cgroup_root_count = 1;
a4ea1cc9 4050 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
817929ec 4051
69d0206c 4052 init_cgrp_cset_link.cset = &init_css_set;
9871bf95
TH
4053 init_cgrp_cset_link.cgrp = cgroup_dummy_top;
4054 list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links);
69d0206c 4055 list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
ddbcc7e8 4056
3ed80a62 4057 for_each_subsys(ss, i) {
aec25020 4058 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
073219e9
TH
4059 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4060 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
aec25020 4061 ss->id, ss->name);
073219e9
TH
4062 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
4063 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
4064
aec25020 4065 ss->id = i;
073219e9 4066 ss->name = cgroup_subsys_name[i];
ddbcc7e8
PM
4067
4068 if (ss->early_init)
4069 cgroup_init_subsys(ss);
4070 }
4071 return 0;
4072}
4073
4074/**
a043e3b2
LZ
4075 * cgroup_init - cgroup initialization
4076 *
4077 * Register cgroup filesystem and /proc file, and initialize
4078 * any subsystems that didn't request early init.
ddbcc7e8
PM
4079 */
4080int __init cgroup_init(void)
4081{
30159ec7 4082 struct cgroup_subsys *ss;
0ac801fe 4083 unsigned long key;
30159ec7 4084 int i, err;
a424316c 4085
2bd59d48 4086 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
2da440a2 4087
3ed80a62 4088 for_each_subsys(ss, i) {
ddbcc7e8
PM
4089 if (!ss->early_init)
4090 cgroup_init_subsys(ss);
de00ffa5
TH
4091
4092 /*
4093 * cftype registration needs kmalloc and can't be done
4094 * during early_init. Register base cftypes separately.
4095 */
4096 if (ss->base_cftypes)
4097 WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes));
ddbcc7e8
PM
4098 }
4099
fa3ca07e 4100 /* allocate id for the dummy hierarchy */
54e7b4eb 4101 mutex_lock(&cgroup_mutex);
54e7b4eb 4102
82fe9b0d
TH
4103 /* Add init_css_set to the hash table */
4104 key = css_set_hash(init_css_set.subsys);
4105 hash_add(css_set_table, &init_css_set.hlist, key);
4106
fc76df70 4107 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1));
676db4af 4108
4e96ee8e
LZ
4109 err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top,
4110 0, 1, GFP_KERNEL);
4111 BUG_ON(err < 0);
4112
54e7b4eb
TH
4113 mutex_unlock(&cgroup_mutex);
4114
676db4af 4115 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
2bd59d48
TH
4116 if (!cgroup_kobj)
4117 return -ENOMEM;
676db4af 4118
ddbcc7e8 4119 err = register_filesystem(&cgroup_fs_type);
676db4af
GK
4120 if (err < 0) {
4121 kobject_put(cgroup_kobj);
2bd59d48 4122 return err;
676db4af 4123 }
ddbcc7e8 4124
46ae220b 4125 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
2bd59d48 4126 return 0;
ddbcc7e8 4127}
b4f48b63 4128
e5fca243
TH
4129static int __init cgroup_wq_init(void)
4130{
4131 /*
4132 * There isn't much point in executing destruction path in
4133 * parallel. Good chunk is serialized with cgroup_mutex anyway.
1a11533f 4134 * Use 1 for @max_active.
e5fca243
TH
4135 *
4136 * We would prefer to do this in cgroup_init() above, but that
4137 * is called before init_workqueues(): so leave this until after.
4138 */
1a11533f 4139 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
e5fca243 4140 BUG_ON(!cgroup_destroy_wq);
b1a21367
TH
4141
4142 /*
4143 * Used to destroy pidlists and separate to serve as flush domain.
4144 * Cap @max_active to 1 too.
4145 */
4146 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
4147 0, 1);
4148 BUG_ON(!cgroup_pidlist_destroy_wq);
4149
e5fca243
TH
4150 return 0;
4151}
4152core_initcall(cgroup_wq_init);
4153
a424316c
PM
4154/*
4155 * proc_cgroup_show()
4156 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4157 * - Used for /proc/<pid>/cgroup.
4158 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4159 * doesn't really matter if tsk->cgroup changes after we read it,
956db3ca 4160 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
a424316c
PM
4161 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4162 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4163 * cgroup to top_cgroup.
4164 */
4165
4166/* TODO: Use a proper seq_file iterator */
8d8b97ba 4167int proc_cgroup_show(struct seq_file *m, void *v)
a424316c
PM
4168{
4169 struct pid *pid;
4170 struct task_struct *tsk;
e61734c5 4171 char *buf, *path;
a424316c
PM
4172 int retval;
4173 struct cgroupfs_root *root;
4174
4175 retval = -ENOMEM;
e61734c5 4176 buf = kmalloc(PATH_MAX, GFP_KERNEL);
a424316c
PM
4177 if (!buf)
4178 goto out;
4179
4180 retval = -ESRCH;
4181 pid = m->private;
4182 tsk = get_pid_task(pid, PIDTYPE_PID);
4183 if (!tsk)
4184 goto out_free;
4185
4186 retval = 0;
4187
4188 mutex_lock(&cgroup_mutex);
96d365e0 4189 down_read(&css_set_rwsem);
a424316c 4190
e5f6a860 4191 for_each_active_root(root) {
a424316c 4192 struct cgroup_subsys *ss;
bd89aabc 4193 struct cgroup *cgrp;
b85d2040 4194 int ssid, count = 0;
a424316c 4195
2c6ab6d2 4196 seq_printf(m, "%d:", root->hierarchy_id);
b85d2040
TH
4197 for_each_subsys(ss, ssid)
4198 if (root->subsys_mask & (1 << ssid))
4199 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
c6d57f33
PM
4200 if (strlen(root->name))
4201 seq_printf(m, "%sname=%s", count ? "," : "",
4202 root->name);
a424316c 4203 seq_putc(m, ':');
7717f7ba 4204 cgrp = task_cgroup_from_root(tsk, root);
e61734c5
TH
4205 path = cgroup_path(cgrp, buf, PATH_MAX);
4206 if (!path) {
4207 retval = -ENAMETOOLONG;
a424316c 4208 goto out_unlock;
e61734c5
TH
4209 }
4210 seq_puts(m, path);
a424316c
PM
4211 seq_putc(m, '\n');
4212 }
4213
4214out_unlock:
96d365e0 4215 up_read(&css_set_rwsem);
a424316c
PM
4216 mutex_unlock(&cgroup_mutex);
4217 put_task_struct(tsk);
4218out_free:
4219 kfree(buf);
4220out:
4221 return retval;
4222}
4223
a424316c
PM
4224/* Display information about each subsystem and each hierarchy */
4225static int proc_cgroupstats_show(struct seq_file *m, void *v)
4226{
30159ec7 4227 struct cgroup_subsys *ss;
a424316c 4228 int i;
a424316c 4229
8bab8dde 4230 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
aae8aab4
BB
4231 /*
4232 * ideally we don't want subsystems moving around while we do this.
4233 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4234 * subsys/hierarchy state.
4235 */
a424316c 4236 mutex_lock(&cgroup_mutex);
30159ec7
TH
4237
4238 for_each_subsys(ss, i)
2c6ab6d2
PM
4239 seq_printf(m, "%s\t%d\t%d\t%d\n",
4240 ss->name, ss->root->hierarchy_id,
3c9c825b 4241 atomic_read(&ss->root->nr_cgrps), !ss->disabled);
30159ec7 4242
a424316c
PM
4243 mutex_unlock(&cgroup_mutex);
4244 return 0;
4245}
4246
4247static int cgroupstats_open(struct inode *inode, struct file *file)
4248{
9dce07f1 4249 return single_open(file, proc_cgroupstats_show, NULL);
a424316c
PM
4250}
4251
828c0950 4252static const struct file_operations proc_cgroupstats_operations = {
a424316c
PM
4253 .open = cgroupstats_open,
4254 .read = seq_read,
4255 .llseek = seq_lseek,
4256 .release = single_release,
4257};
4258
b4f48b63 4259/**
eaf797ab 4260 * cgroup_fork - initialize cgroup related fields during copy_process()
a043e3b2 4261 * @child: pointer to task_struct of forking parent process.
b4f48b63 4262 *
eaf797ab
TH
4263 * A task is associated with the init_css_set until cgroup_post_fork()
4264 * attaches it to the parent's css_set. Empty cg_list indicates that
4265 * @child isn't holding reference to its css_set.
b4f48b63
PM
4266 */
4267void cgroup_fork(struct task_struct *child)
4268{
eaf797ab 4269 RCU_INIT_POINTER(child->cgroups, &init_css_set);
817929ec 4270 INIT_LIST_HEAD(&child->cg_list);
b4f48b63
PM
4271}
4272
817929ec 4273/**
a043e3b2
LZ
4274 * cgroup_post_fork - called on a new task after adding it to the task list
4275 * @child: the task in question
4276 *
5edee61e
TH
4277 * Adds the task to the list running through its css_set if necessary and
4278 * call the subsystem fork() callbacks. Has to be after the task is
4279 * visible on the task list in case we race with the first call to
0942eeee 4280 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5edee61e 4281 * list.
a043e3b2 4282 */
817929ec
PM
4283void cgroup_post_fork(struct task_struct *child)
4284{
30159ec7 4285 struct cgroup_subsys *ss;
5edee61e
TH
4286 int i;
4287
3ce3230a 4288 /*
eaf797ab
TH
4289 * This may race against cgroup_enable_task_cg_links(). As that
4290 * function sets use_task_css_set_links before grabbing
4291 * tasklist_lock and we just went through tasklist_lock to add
4292 * @child, it's guaranteed that either we see the set
4293 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
4294 * @child during its iteration.
4295 *
4296 * If we won the race, @child is associated with %current's
4297 * css_set. Grabbing css_set_rwsem guarantees both that the
4298 * association is stable, and, on completion of the parent's
4299 * migration, @child is visible in the source of migration or
4300 * already in the destination cgroup. This guarantee is necessary
4301 * when implementing operations which need to migrate all tasks of
4302 * a cgroup to another.
4303 *
4304 * Note that if we lose to cgroup_enable_task_cg_links(), @child
4305 * will remain in init_css_set. This is safe because all tasks are
4306 * in the init_css_set before cg_links is enabled and there's no
4307 * operation which transfers all tasks out of init_css_set.
3ce3230a 4308 */
817929ec 4309 if (use_task_css_set_links) {
eaf797ab
TH
4310 struct css_set *cset;
4311
96d365e0 4312 down_write(&css_set_rwsem);
eaf797ab
TH
4313 cset = task_css_set_check(current,
4314 lockdep_is_held(&css_set_rwsem));
d8783832 4315 task_lock(child);
eaf797ab
TH
4316 if (list_empty(&child->cg_list)) {
4317 rcu_assign_pointer(child->cgroups, cset);
4318 list_add(&child->cg_list, &cset->tasks);
4319 get_css_set(cset);
4320 }
d8783832 4321 task_unlock(child);
96d365e0 4322 up_write(&css_set_rwsem);
817929ec 4323 }
5edee61e
TH
4324
4325 /*
4326 * Call ss->fork(). This must happen after @child is linked on
4327 * css_set; otherwise, @child might change state between ->fork()
4328 * and addition to css_set.
4329 */
4330 if (need_forkexit_callback) {
3ed80a62 4331 for_each_subsys(ss, i)
5edee61e
TH
4332 if (ss->fork)
4333 ss->fork(child);
5edee61e 4334 }
817929ec 4335}
5edee61e 4336
b4f48b63
PM
4337/**
4338 * cgroup_exit - detach cgroup from exiting task
4339 * @tsk: pointer to task_struct of exiting process
a043e3b2 4340 * @run_callback: run exit callbacks?
b4f48b63
PM
4341 *
4342 * Description: Detach cgroup from @tsk and release it.
4343 *
4344 * Note that cgroups marked notify_on_release force every task in
4345 * them to take the global cgroup_mutex mutex when exiting.
4346 * This could impact scaling on very large systems. Be reluctant to
4347 * use notify_on_release cgroups where very high task exit scaling
4348 * is required on large systems.
4349 *
4350 * the_top_cgroup_hack:
4351 *
4352 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4353 *
4354 * We call cgroup_exit() while the task is still competent to
4355 * handle notify_on_release(), then leave the task attached to the
4356 * root cgroup in each hierarchy for the remainder of its exit.
4357 *
4358 * To do this properly, we would increment the reference count on
4359 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4360 * code we would add a second cgroup function call, to drop that
4361 * reference. This would just create an unnecessary hot spot on
4362 * the top_cgroup reference count, to no avail.
4363 *
4364 * Normally, holding a reference to a cgroup without bumping its
4365 * count is unsafe. The cgroup could go away, or someone could
4366 * attach us to a different cgroup, decrementing the count on
4367 * the first cgroup that we never incremented. But in this case,
4368 * top_cgroup isn't going away, and either task has PF_EXITING set,
956db3ca
CW
4369 * which wards off any cgroup_attach_task() attempts, or task is a failed
4370 * fork, never visible to cgroup_attach_task.
b4f48b63
PM
4371 */
4372void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4373{
30159ec7 4374 struct cgroup_subsys *ss;
5abb8855 4375 struct css_set *cset;
eaf797ab 4376 bool put_cset = false;
d41d5a01 4377 int i;
817929ec
PM
4378
4379 /*
96d365e0
TH
4380 * Unlink from the css_set task list if necessary. Optimistically
4381 * check cg_list before taking css_set_rwsem.
817929ec
PM
4382 */
4383 if (!list_empty(&tsk->cg_list)) {
96d365e0 4384 down_write(&css_set_rwsem);
eaf797ab 4385 if (!list_empty(&tsk->cg_list)) {
8d258797 4386 list_del_init(&tsk->cg_list);
eaf797ab
TH
4387 put_cset = true;
4388 }
96d365e0 4389 up_write(&css_set_rwsem);
817929ec
PM
4390 }
4391
b4f48b63
PM
4392 /* Reassign the task to the init_css_set. */
4393 task_lock(tsk);
a8ad805c
TH
4394 cset = task_css_set(tsk);
4395 RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
d41d5a01
PZ
4396
4397 if (run_callbacks && need_forkexit_callback) {
3ed80a62
TH
4398 /* see cgroup_post_fork() for details */
4399 for_each_subsys(ss, i) {
d41d5a01 4400 if (ss->exit) {
eb95419b
TH
4401 struct cgroup_subsys_state *old_css = cset->subsys[i];
4402 struct cgroup_subsys_state *css = task_css(tsk, i);
30159ec7 4403
eb95419b 4404 ss->exit(css, old_css, tsk);
d41d5a01
PZ
4405 }
4406 }
4407 }
b4f48b63 4408 task_unlock(tsk);
d41d5a01 4409
eaf797ab
TH
4410 if (put_cset)
4411 put_css_set(cset, true);
b4f48b63 4412}
697f4161 4413
bd89aabc 4414static void check_for_release(struct cgroup *cgrp)
81a6a5cd 4415{
f50daa70 4416 if (cgroup_is_releasable(cgrp) &&
6f3d828f 4417 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
f50daa70
LZ
4418 /*
4419 * Control Group is currently removeable. If it's not
81a6a5cd 4420 * already queued for a userspace notification, queue
f50daa70
LZ
4421 * it now
4422 */
81a6a5cd 4423 int need_schedule_work = 0;
f50daa70 4424
cdcc136f 4425 raw_spin_lock(&release_list_lock);
54766d4a 4426 if (!cgroup_is_dead(cgrp) &&
bd89aabc
PM
4427 list_empty(&cgrp->release_list)) {
4428 list_add(&cgrp->release_list, &release_list);
81a6a5cd
PM
4429 need_schedule_work = 1;
4430 }
cdcc136f 4431 raw_spin_unlock(&release_list_lock);
81a6a5cd
PM
4432 if (need_schedule_work)
4433 schedule_work(&release_agent_work);
4434 }
4435}
4436
81a6a5cd
PM
4437/*
4438 * Notify userspace when a cgroup is released, by running the
4439 * configured release agent with the name of the cgroup (path
4440 * relative to the root of cgroup file system) as the argument.
4441 *
4442 * Most likely, this user command will try to rmdir this cgroup.
4443 *
4444 * This races with the possibility that some other task will be
4445 * attached to this cgroup before it is removed, or that some other
4446 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4447 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4448 * unused, and this cgroup will be reprieved from its death sentence,
4449 * to continue to serve a useful existence. Next time it's released,
4450 * we will get notified again, if it still has 'notify_on_release' set.
4451 *
4452 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4453 * means only wait until the task is successfully execve()'d. The
4454 * separate release agent task is forked by call_usermodehelper(),
4455 * then control in this thread returns here, without waiting for the
4456 * release agent task. We don't bother to wait because the caller of
4457 * this routine has no use for the exit status of the release agent
4458 * task, so no sense holding our caller up for that.
81a6a5cd 4459 */
81a6a5cd
PM
4460static void cgroup_release_agent(struct work_struct *work)
4461{
4462 BUG_ON(work != &release_agent_work);
4463 mutex_lock(&cgroup_mutex);
cdcc136f 4464 raw_spin_lock(&release_list_lock);
81a6a5cd
PM
4465 while (!list_empty(&release_list)) {
4466 char *argv[3], *envp[3];
4467 int i;
e61734c5 4468 char *pathbuf = NULL, *agentbuf = NULL, *path;
bd89aabc 4469 struct cgroup *cgrp = list_entry(release_list.next,
81a6a5cd
PM
4470 struct cgroup,
4471 release_list);
bd89aabc 4472 list_del_init(&cgrp->release_list);
cdcc136f 4473 raw_spin_unlock(&release_list_lock);
e61734c5 4474 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
e788e066
PM
4475 if (!pathbuf)
4476 goto continue_free;
e61734c5
TH
4477 path = cgroup_path(cgrp, pathbuf, PATH_MAX);
4478 if (!path)
e788e066
PM
4479 goto continue_free;
4480 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4481 if (!agentbuf)
4482 goto continue_free;
81a6a5cd
PM
4483
4484 i = 0;
e788e066 4485 argv[i++] = agentbuf;
e61734c5 4486 argv[i++] = path;
81a6a5cd
PM
4487 argv[i] = NULL;
4488
4489 i = 0;
4490 /* minimal command environment */
4491 envp[i++] = "HOME=/";
4492 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4493 envp[i] = NULL;
4494
4495 /* Drop the lock while we invoke the usermode helper,
4496 * since the exec could involve hitting disk and hence
4497 * be a slow process */
4498 mutex_unlock(&cgroup_mutex);
4499 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
81a6a5cd 4500 mutex_lock(&cgroup_mutex);
e788e066
PM
4501 continue_free:
4502 kfree(pathbuf);
4503 kfree(agentbuf);
cdcc136f 4504 raw_spin_lock(&release_list_lock);
81a6a5cd 4505 }
cdcc136f 4506 raw_spin_unlock(&release_list_lock);
81a6a5cd
PM
4507 mutex_unlock(&cgroup_mutex);
4508}
8bab8dde
PM
4509
4510static int __init cgroup_disable(char *str)
4511{
30159ec7 4512 struct cgroup_subsys *ss;
8bab8dde 4513 char *token;
30159ec7 4514 int i;
8bab8dde
PM
4515
4516 while ((token = strsep(&str, ",")) != NULL) {
4517 if (!*token)
4518 continue;
be45c900 4519
3ed80a62 4520 for_each_subsys(ss, i) {
8bab8dde
PM
4521 if (!strcmp(token, ss->name)) {
4522 ss->disabled = 1;
4523 printk(KERN_INFO "Disabling %s control group"
4524 " subsystem\n", ss->name);
4525 break;
4526 }
4527 }
4528 }
4529 return 1;
4530}
4531__setup("cgroup_disable=", cgroup_disable);
38460b48 4532
b77d7b60 4533/**
5a17f543 4534 * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir
35cf0836
TH
4535 * @dentry: directory dentry of interest
4536 * @ss: subsystem of interest
b77d7b60 4537 *
5a17f543
TH
4538 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
4539 * to get the corresponding css and return it. If such css doesn't exist
4540 * or can't be pinned, an ERR_PTR value is returned.
e5d1367f 4541 */
5a17f543
TH
4542struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry,
4543 struct cgroup_subsys *ss)
e5d1367f 4544{
2bd59d48
TH
4545 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4546 struct cgroup_subsys_state *css = NULL;
e5d1367f 4547 struct cgroup *cgrp;
b77d7b60 4548
35cf0836 4549 /* is @dentry a cgroup dir? */
2bd59d48
TH
4550 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4551 kernfs_type(kn) != KERNFS_DIR)
e5d1367f
SE
4552 return ERR_PTR(-EBADF);
4553
5a17f543
TH
4554 rcu_read_lock();
4555
2bd59d48
TH
4556 /*
4557 * This path doesn't originate from kernfs and @kn could already
4558 * have been or be removed at any point. @kn->priv is RCU
4559 * protected for this access. See destroy_locked() for details.
4560 */
4561 cgrp = rcu_dereference(kn->priv);
4562 if (cgrp)
4563 css = cgroup_css(cgrp, ss);
5a17f543
TH
4564
4565 if (!css || !css_tryget(css))
4566 css = ERR_PTR(-ENOENT);
4567
4568 rcu_read_unlock();
4569 return css;
e5d1367f 4570}
e5d1367f 4571
1cb650b9
LZ
4572/**
4573 * css_from_id - lookup css by id
4574 * @id: the cgroup id
4575 * @ss: cgroup subsys to be looked into
4576 *
4577 * Returns the css if there's valid one with @id, otherwise returns NULL.
4578 * Should be called under rcu_read_lock().
4579 */
4580struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
4581{
4582 struct cgroup *cgrp;
4583
ace2bee8 4584 cgroup_assert_mutexes_or_rcu_locked();
1cb650b9
LZ
4585
4586 cgrp = idr_find(&ss->root->cgroup_idr, id);
4587 if (cgrp)
d1625964 4588 return cgroup_css(cgrp, ss);
1cb650b9 4589 return NULL;
e5d1367f
SE
4590}
4591
fe693435 4592#ifdef CONFIG_CGROUP_DEBUG
eb95419b
TH
4593static struct cgroup_subsys_state *
4594debug_css_alloc(struct cgroup_subsys_state *parent_css)
fe693435
PM
4595{
4596 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4597
4598 if (!css)
4599 return ERR_PTR(-ENOMEM);
4600
4601 return css;
4602}
4603
eb95419b 4604static void debug_css_free(struct cgroup_subsys_state *css)
fe693435 4605{
eb95419b 4606 kfree(css);
fe693435
PM
4607}
4608
182446d0
TH
4609static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
4610 struct cftype *cft)
fe693435 4611{
182446d0 4612 return cgroup_task_count(css->cgroup);
fe693435
PM
4613}
4614
182446d0
TH
4615static u64 current_css_set_read(struct cgroup_subsys_state *css,
4616 struct cftype *cft)
fe693435
PM
4617{
4618 return (u64)(unsigned long)current->cgroups;
4619}
4620
182446d0 4621static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
03c78cbe 4622 struct cftype *cft)
fe693435
PM
4623{
4624 u64 count;
4625
4626 rcu_read_lock();
a8ad805c 4627 count = atomic_read(&task_css_set(current)->refcount);
fe693435
PM
4628 rcu_read_unlock();
4629 return count;
4630}
4631
2da8ca82 4632static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
7717f7ba 4633{
69d0206c 4634 struct cgrp_cset_link *link;
5abb8855 4635 struct css_set *cset;
e61734c5
TH
4636 char *name_buf;
4637
4638 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
4639 if (!name_buf)
4640 return -ENOMEM;
7717f7ba 4641
96d365e0 4642 down_read(&css_set_rwsem);
7717f7ba 4643 rcu_read_lock();
5abb8855 4644 cset = rcu_dereference(current->cgroups);
69d0206c 4645 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
7717f7ba 4646 struct cgroup *c = link->cgrp;
59f5296b
TH
4647 const char *name = "?";
4648
e61734c5
TH
4649 if (c != cgroup_dummy_top) {
4650 cgroup_name(c, name_buf, NAME_MAX + 1);
4651 name = name_buf;
4652 }
7717f7ba 4653
2c6ab6d2
PM
4654 seq_printf(seq, "Root %d group %s\n",
4655 c->root->hierarchy_id, name);
7717f7ba
PM
4656 }
4657 rcu_read_unlock();
96d365e0 4658 up_read(&css_set_rwsem);
e61734c5 4659 kfree(name_buf);
7717f7ba
PM
4660 return 0;
4661}
4662
4663#define MAX_TASKS_SHOWN_PER_CSS 25
2da8ca82 4664static int cgroup_css_links_read(struct seq_file *seq, void *v)
7717f7ba 4665{
2da8ca82 4666 struct cgroup_subsys_state *css = seq_css(seq);
69d0206c 4667 struct cgrp_cset_link *link;
7717f7ba 4668
96d365e0 4669 down_read(&css_set_rwsem);
182446d0 4670 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
69d0206c 4671 struct css_set *cset = link->cset;
7717f7ba
PM
4672 struct task_struct *task;
4673 int count = 0;
c7561128 4674
5abb8855 4675 seq_printf(seq, "css_set %p\n", cset);
c7561128 4676
5abb8855 4677 list_for_each_entry(task, &cset->tasks, cg_list) {
c7561128
TH
4678 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
4679 goto overflow;
4680 seq_printf(seq, " task %d\n", task_pid_vnr(task));
4681 }
4682
4683 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
4684 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
4685 goto overflow;
4686 seq_printf(seq, " task %d\n", task_pid_vnr(task));
7717f7ba 4687 }
c7561128
TH
4688 continue;
4689 overflow:
4690 seq_puts(seq, " ...\n");
7717f7ba 4691 }
96d365e0 4692 up_read(&css_set_rwsem);
7717f7ba
PM
4693 return 0;
4694}
4695
182446d0 4696static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
fe693435 4697{
182446d0 4698 return test_bit(CGRP_RELEASABLE, &css->cgroup->flags);
fe693435
PM
4699}
4700
4701static struct cftype debug_files[] = {
fe693435
PM
4702 {
4703 .name = "taskcount",
4704 .read_u64 = debug_taskcount_read,
4705 },
4706
4707 {
4708 .name = "current_css_set",
4709 .read_u64 = current_css_set_read,
4710 },
4711
4712 {
4713 .name = "current_css_set_refcount",
4714 .read_u64 = current_css_set_refcount_read,
4715 },
4716
7717f7ba
PM
4717 {
4718 .name = "current_css_set_cg_links",
2da8ca82 4719 .seq_show = current_css_set_cg_links_read,
7717f7ba
PM
4720 },
4721
4722 {
4723 .name = "cgroup_css_links",
2da8ca82 4724 .seq_show = cgroup_css_links_read,
7717f7ba
PM
4725 },
4726
fe693435
PM
4727 {
4728 .name = "releasable",
4729 .read_u64 = releasable_read,
4730 },
fe693435 4731
4baf6e33
TH
4732 { } /* terminate */
4733};
fe693435 4734
073219e9 4735struct cgroup_subsys debug_cgrp_subsys = {
92fb9748
TH
4736 .css_alloc = debug_css_alloc,
4737 .css_free = debug_css_free,
4baf6e33 4738 .base_cftypes = debug_files,
fe693435
PM
4739};
4740#endif /* CONFIG_CGROUP_DEBUG */