cgroup.c 123.5 KB
Newer Older
1 2 3 4 5 6
/*
 *  Generic process-grouping system.
 *
 *  Based originally on the cpuset system, extracted by Paul Menage
 *  Copyright (C) 2006 Google, Inc
 *
7 8 9 10
 *  Notifications support
 *  Copyright (C) 2009 Nokia Corporation
 *  Author: Kirill A. Shutemov
 *
11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
 *  Copyright notices from the original cpuset code:
 *  --------------------------------------------------
 *  Copyright (C) 2003 BULL SA.
 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
 *
 *  Portions derived from Patrick Mochel's sysfs code.
 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 *
 *  2003-10-10 Written by Simon Derr.
 *  2003-10-22 Updates by Stephen Hemminger.
 *  2004 May-July Rework by Paul Jackson.
 *  ---------------------------------------------------
 *
 *  This file is subject to the terms and conditions of the GNU General Public
 *  License.  See the file COPYING in the main directory of the Linux
 *  distribution for more details.
 */

#include <linux/cgroup.h>
30
#include <linux/cred.h>
31
#include <linux/ctype.h>
32
#include <linux/errno.h>
33
#include <linux/init_task.h>
34 35 36 37 38 39
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/mm.h>
#include <linux/mutex.h>
#include <linux/mount.h>
#include <linux/pagemap.h>
40
#include <linux/proc_fs.h>
41 42 43 44
#include <linux/rcupdate.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
45
#include <linux/rwsem.h>
46
#include <linux/string.h>
47
#include <linux/sort.h>
48
#include <linux/kmod.h>
B
Balbir Singh 已提交
49 50
#include <linux/delayacct.h>
#include <linux/cgroupstats.h>
51
#include <linux/hashtable.h>
L
Li Zefan 已提交
52
#include <linux/pid_namespace.h>
53
#include <linux/idr.h>
54
#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
55
#include <linux/flex_array.h> /* used in cgroup_attach_task */
56
#include <linux/kthread.h>
T
Tejun Heo 已提交
57
#include <linux/delay.h>
B
Balbir Singh 已提交
58

A
Arun Sharma 已提交
59
#include <linux/atomic.h>
60

61 62 63 64 65 66 67 68
/*
 * pidlists linger the following amount before being destroyed.  The goal
 * is avoiding frequent destruction in the middle of consecutive read calls
 * Expiring in the middle is a performance problem not a correctness one.
 * 1 sec should be enough.
 */
#define CGROUP_PIDLIST_DESTROY_DELAY	HZ

T
Tejun Heo 已提交
69 70 71
#define CGROUP_FILE_NAME_MAX		(MAX_CGROUP_TYPE_NAMELEN +	\
					 MAX_CFTYPE_NAME + 2)

T
Tejun Heo 已提交
72 73 74 75 76 77 78 79 80
/*
 * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file
 * creation/removal and hierarchy changing operations including cgroup
 * creation, removal, css association and controller rebinding.  This outer
 * lock is needed mainly to resolve the circular dependency between kernfs
 * active ref and cgroup_mutex.  cgroup_tree_mutex nests above both.
 */
static DEFINE_MUTEX(cgroup_tree_mutex);

T
Tejun Heo 已提交
81 82 83 84
/*
 * cgroup_mutex is the master lock.  Any modification to cgroup or its
 * hierarchy must be performed while holding it.
 */
T
Tejun Heo 已提交
85 86
#ifdef CONFIG_PROVE_RCU
DEFINE_MUTEX(cgroup_mutex);
87
EXPORT_SYMBOL_GPL(cgroup_mutex);	/* only for lockdep */
T
Tejun Heo 已提交
88
#else
89
static DEFINE_MUTEX(cgroup_mutex);
T
Tejun Heo 已提交
90 91
#endif

92 93 94 95 96 97
/*
 * Protects cgroup_subsys->release_agent_path.  Modifying it also requires
 * cgroup_mutex.  Reading requires either cgroup_mutex or this spinlock.
 */
static DEFINE_SPINLOCK(release_agent_path_lock);

T
Tejun Heo 已提交
98
#define cgroup_assert_mutexes_or_rcu_locked()				\
99
	rcu_lockdep_assert(rcu_read_lock_held() ||			\
T
Tejun Heo 已提交
100
			   lockdep_is_held(&cgroup_tree_mutex) ||	\
101
			   lockdep_is_held(&cgroup_mutex),		\
T
Tejun Heo 已提交
102
			   "cgroup_[tree_]mutex or RCU read lock required");
103

104 105 106 107 108 109 110 111
/*
 * cgroup destruction makes heavy use of work items and there can be a lot
 * of concurrent destructions.  Use a separate workqueue so that cgroup
 * destruction work items don't end up filling up max_active of system_wq
 * which may lead to deadlock.
 */
static struct workqueue_struct *cgroup_destroy_wq;

112 113 114 115 116 117
/*
 * pidlist destructions need to be flushed on cgroup destruction.  Use a
 * separate workqueue as flush domain.
 */
static struct workqueue_struct *cgroup_pidlist_destroy_wq;

T
Tejun Heo 已提交
118
/* generate an array of cgroup subsystem pointers */
119
#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
T
Tejun Heo 已提交
120
static struct cgroup_subsys *cgroup_subsys[] = {
121 122
#include <linux/cgroup_subsys.h>
};
123 124 125 126 127 128 129 130
#undef SUBSYS

/* array of cgroup subsystem names */
#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
static const char *cgroup_subsys_name[] = {
#include <linux/cgroup_subsys.h>
};
#undef SUBSYS
131 132

/*
133 134 135
 * The dummy hierarchy, reserved for the subsystems that are otherwise
 * unattached - it never has more than a single cgroup, and all tasks are
 * part of that cgroup.
136
 */
137 138 139 140
static struct cgroupfs_root cgroup_dummy_root;

/* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup;
141 142 143

/* The list of hierarchy roots */

144 145
static LIST_HEAD(cgroup_roots);
static int cgroup_root_count;
146

T
Tejun Heo 已提交
147
/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
148
static DEFINE_IDR(cgroup_hierarchy_idr);
149

150 151 152 153 154
/*
 * Assign a monotonically increasing serial number to cgroups.  It
 * guarantees cgroups with bigger numbers are newer than those with smaller
 * numbers.  Also, as cgroups are always appended to the parent's
 * ->children list, it guarantees that sibling cgroups are always sorted in
155 156
 * the ascending serial number order on the list.  Protected by
 * cgroup_mutex.
157
 */
158
static u64 cgroup_serial_nr_next = 1;
159

160
/* This flag indicates whether tasks in the fork and exit paths should
L
Li Zefan 已提交
161 162 163
 * check for fork/exit handlers to call. This avoids us having to do
 * extra work in the fork/exit path if none of the subsystems need to
 * be called.
164
 */
165
static int need_forkexit_callback __read_mostly;
166

167 168
static struct cftype cgroup_base_files[];

169
static void cgroup_put(struct cgroup *cgrp);
170 171
static int rebind_subsystems(struct cgroupfs_root *root,
			     unsigned long added_mask, unsigned removed_mask);
172
static void cgroup_destroy_css_killed(struct cgroup *cgrp);
173
static int cgroup_destroy_locked(struct cgroup *cgrp);
174 175
static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
			      bool is_add);
176
static void cgroup_pidlist_destroy_all(struct cgroup *cgrp);
177

T
Tejun Heo 已提交
178 179 180
/**
 * cgroup_css - obtain a cgroup's css for the specified subsystem
 * @cgrp: the cgroup of interest
181
 * @ss: the subsystem of interest (%NULL returns the dummy_css)
T
Tejun Heo 已提交
182
 *
183 184 185 186 187
 * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
 * function must be called either under cgroup_mutex or rcu_read_lock() and
 * the caller is responsible for pinning the returned css if it wants to
 * keep accessing it outside the said locks.  This function may return
 * %NULL if @cgrp doesn't have @subsys_id enabled.
T
Tejun Heo 已提交
188 189
 */
static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
190
					      struct cgroup_subsys *ss)
T
Tejun Heo 已提交
191
{
192
	if (ss)
193
		return rcu_dereference_check(cgrp->subsys[ss->id],
T
Tejun Heo 已提交
194 195
					lockdep_is_held(&cgroup_tree_mutex) ||
					lockdep_is_held(&cgroup_mutex));
196 197
	else
		return &cgrp->dummy_css;
T
Tejun Heo 已提交
198
}
199

200
/* convenient tests for these bits */
201
static inline bool cgroup_is_dead(const struct cgroup *cgrp)
202
{
203
	return test_bit(CGRP_DEAD, &cgrp->flags);
204 205
}

206 207
struct cgroup_subsys_state *seq_css(struct seq_file *seq)
{
T
Tejun Heo 已提交
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223
	struct kernfs_open_file *of = seq->private;
	struct cgroup *cgrp = of->kn->parent->priv;
	struct cftype *cft = seq_cft(seq);

	/*
	 * This is open and unprotected implementation of cgroup_css().
	 * seq_css() is only called from a kernfs file operation which has
	 * an active reference on the file.  Because all the subsystem
	 * files are drained before a css is disassociated with a cgroup,
	 * the matching css from the cgroup's subsys table is guaranteed to
	 * be and stay valid until the enclosing operation is complete.
	 */
	if (cft->ss)
		return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
	else
		return &cgrp->dummy_css;
224 225 226
}
EXPORT_SYMBOL_GPL(seq_css);

227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
/**
 * cgroup_is_descendant - test ancestry
 * @cgrp: the cgroup to be tested
 * @ancestor: possible ancestor of @cgrp
 *
 * Test whether @cgrp is a descendant of @ancestor.  It also returns %true
 * if @cgrp == @ancestor.  This function is safe to call as long as @cgrp
 * and @ancestor are accessible.
 */
bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
{
	while (cgrp) {
		if (cgrp == ancestor)
			return true;
		cgrp = cgrp->parent;
	}
	return false;
}
245

246
static int cgroup_is_releasable(const struct cgroup *cgrp)
247 248
{
	const int bits =
249 250 251
		(1 << CGRP_RELEASABLE) |
		(1 << CGRP_NOTIFY_ON_RELEASE);
	return (cgrp->flags & bits) == bits;
252 253
}

254
static int notify_on_release(const struct cgroup *cgrp)
255
{
256
	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
257 258
}

T
Tejun Heo 已提交
259 260 261 262 263 264 265 266 267 268 269 270
/**
 * for_each_css - iterate all css's of a cgroup
 * @css: the iteration cursor
 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
 * @cgrp: the target cgroup to iterate css's of
 *
 * Should be called under cgroup_mutex.
 */
#define for_each_css(css, ssid, cgrp)					\
	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)	\
		if (!((css) = rcu_dereference_check(			\
				(cgrp)->subsys[(ssid)],			\
T
Tejun Heo 已提交
271
				lockdep_is_held(&cgroup_tree_mutex) ||	\
T
Tejun Heo 已提交
272 273 274
				lockdep_is_held(&cgroup_mutex)))) { }	\
		else

275
/**
T
Tejun Heo 已提交
276
 * for_each_subsys - iterate all enabled cgroup subsystems
277
 * @ss: the iteration cursor
278
 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
279
 */
280
#define for_each_subsys(ss, ssid)					\
T
Tejun Heo 已提交
281 282
	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT &&		\
	     (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
283

284 285 286
/* iterate across the active hierarchies */
#define for_each_active_root(root)					\
	list_for_each_entry((root), &cgroup_roots, root_list)
287

288 289 290 291
/**
 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
 * @cgrp: the cgroup to be checked for liveness
 *
T
Tejun Heo 已提交
292 293
 * On success, returns true; the mutex should be later unlocked.  On
 * failure returns false with no lock held.
294
 */
295
static bool cgroup_lock_live_group(struct cgroup *cgrp)
296 297
{
	mutex_lock(&cgroup_mutex);
298
	if (cgroup_is_dead(cgrp)) {
299 300 301 302 303 304
		mutex_unlock(&cgroup_mutex);
		return false;
	}
	return true;
}

305 306 307
/* the list of cgroups eligible for automatic release. Protected by
 * release_list_lock */
static LIST_HEAD(release_list);
308
static DEFINE_RAW_SPINLOCK(release_list_lock);
309 310
static void cgroup_release_agent(struct work_struct *work);
static DECLARE_WORK(release_agent_work, cgroup_release_agent);
311
static void check_for_release(struct cgroup *cgrp);
312

313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
/*
 * A cgroup can be associated with multiple css_sets as different tasks may
 * belong to different cgroups on different hierarchies.  In the other
 * direction, a css_set is naturally associated with multiple cgroups.
 * This M:N relationship is represented by the following link structure
 * which exists for each association and allows traversing the associations
 * from both sides.
 */
struct cgrp_cset_link {
	/* the cgroup and css_set this link associates */
	struct cgroup		*cgrp;
	struct css_set		*cset;

	/* list of cgrp_cset_links anchored at cgrp->cset_links */
	struct list_head	cset_link;

	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
	struct list_head	cgrp_link;
331 332 333 334 335 336 337 338 339 340
};

/* The default css_set - used by init and its children prior to any
 * hierarchies being mounted. It contains a pointer to the root state
 * for each subsystem. Also used to anchor the list of css_sets. Not
 * reference-counted, to improve performance when child cgroups
 * haven't been created.
 */

static struct css_set init_css_set;
341
static struct cgrp_cset_link init_cgrp_cset_link;
342

343
/*
344 345
 * css_set_rwsem protects the list of css_set objects, and the chain of
 * tasks off each css_set.
346
 */
347
static DECLARE_RWSEM(css_set_rwsem);
348 349
static int css_set_count;

350 351 352 353 354
/*
 * hash table for cgroup groups. This improves the performance to find
 * an existing css_set. This hash doesn't (currently) take into
 * account cgroups in empty hierarchies.
 */
355
#define CSS_SET_HASH_BITS	7
356
static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
357

358
static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
359
{
360
	unsigned long key = 0UL;
361 362
	struct cgroup_subsys *ss;
	int i;
363

364
	for_each_subsys(ss, i)
365 366
		key += (unsigned long)css[i];
	key = (key >> 16) ^ key;
367

368
	return key;
369 370
}

371
static void put_css_set_locked(struct css_set *cset, bool taskexit)
372
{
373
	struct cgrp_cset_link *link, *tmp_link;
374

375 376 377
	lockdep_assert_held(&css_set_rwsem);

	if (!atomic_dec_and_test(&cset->refcount))
378
		return;
379

380
	/* This css_set is dead. unlink it and release cgroup refcounts */
381
	hash_del(&cset->hlist);
382 383
	css_set_count--;

384
	list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
385
		struct cgroup *cgrp = link->cgrp;
386

387 388
		list_del(&link->cset_link);
		list_del(&link->cgrp_link);
L
Li Zefan 已提交
389

390
		/* @cgrp can't go away while we're holding css_set_rwsem */
T
Tejun Heo 已提交
391
		if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
392
			if (taskexit)
393 394
				set_bit(CGRP_RELEASABLE, &cgrp->flags);
			check_for_release(cgrp);
395
		}
396 397

		kfree(link);
398
	}
399

400
	kfree_rcu(cset, rcu_head);
401 402
}

403 404 405 406 407 408 409 410 411 412 413 414 415 416 417
static void put_css_set(struct css_set *cset, bool taskexit)
{
	/*
	 * Ensure that the refcount doesn't hit zero while any readers
	 * can see it. Similar to atomic_dec_and_lock(), but for an
	 * rwlock
	 */
	if (atomic_add_unless(&cset->refcount, -1, 1))
		return;

	down_write(&css_set_rwsem);
	put_css_set_locked(cset, taskexit);
	up_write(&css_set_rwsem);
}

418 419 420
/*
 * refcounted get/put for css_set objects
 */
421
static inline void get_css_set(struct css_set *cset)
422
{
423
	atomic_inc(&cset->refcount);
424 425
}

426
/**
427
 * compare_css_sets - helper function for find_existing_css_set().
428 429
 * @cset: candidate css_set being tested
 * @old_cset: existing css_set for a task
430 431 432
 * @new_cgrp: cgroup that's being entered by the task
 * @template: desired set of css pointers in css_set (pre-calculated)
 *
L
Li Zefan 已提交
433
 * Returns true if "cset" matches "old_cset" except for the hierarchy
434 435
 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
 */
436 437
static bool compare_css_sets(struct css_set *cset,
			     struct css_set *old_cset,
438 439 440 441 442
			     struct cgroup *new_cgrp,
			     struct cgroup_subsys_state *template[])
{
	struct list_head *l1, *l2;

443
	if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
444 445 446 447 448 449 450 451 452 453 454 455 456
		/* Not all subsystems matched */
		return false;
	}

	/*
	 * Compare cgroup pointers in order to distinguish between
	 * different cgroups in heirarchies with no subsystems. We
	 * could get by with just this check alone (and skip the
	 * memcmp above) but on most setups the memcmp check will
	 * avoid the need for this more expensive check on almost all
	 * candidates.
	 */

457 458
	l1 = &cset->cgrp_links;
	l2 = &old_cset->cgrp_links;
459
	while (1) {
460
		struct cgrp_cset_link *link1, *link2;
461
		struct cgroup *cgrp1, *cgrp2;
462 463 464 465

		l1 = l1->next;
		l2 = l2->next;
		/* See if we reached the end - both lists are equal length. */
466 467
		if (l1 == &cset->cgrp_links) {
			BUG_ON(l2 != &old_cset->cgrp_links);
468 469
			break;
		} else {
470
			BUG_ON(l2 == &old_cset->cgrp_links);
471 472
		}
		/* Locate the cgroups associated with these links. */
473 474 475 476
		link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
		link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
		cgrp1 = link1->cgrp;
		cgrp2 = link2->cgrp;
477
		/* Hierarchies should be linked in the same order. */
478
		BUG_ON(cgrp1->root != cgrp2->root);
479 480 481 482 483 484 485 486

		/*
		 * If this hierarchy is the hierarchy of the cgroup
		 * that's changing, then we need to check that this
		 * css_set points to the new cgroup; if it's any other
		 * hierarchy, then this css_set should point to the
		 * same cgroup as the old css_set.
		 */
487 488
		if (cgrp1->root == new_cgrp->root) {
			if (cgrp1 != new_cgrp)
489 490
				return false;
		} else {
491
			if (cgrp1 != cgrp2)
492 493 494 495 496 497
				return false;
		}
	}
	return true;
}

498 499 500 501 502
/**
 * find_existing_css_set - init css array and find the matching css_set
 * @old_cset: the css_set that we're using before the cgroup transition
 * @cgrp: the cgroup that we're moving into
 * @template: out param for the new set of csses, should be clear on entry
503
 */
504 505 506
static struct css_set *find_existing_css_set(struct css_set *old_cset,
					struct cgroup *cgrp,
					struct cgroup_subsys_state *template[])
507
{
508
	struct cgroupfs_root *root = cgrp->root;
509
	struct cgroup_subsys *ss;
510
	struct css_set *cset;
511
	unsigned long key;
512
	int i;
513

B
Ben Blum 已提交
514 515 516 517 518
	/*
	 * Build the set of subsystem state objects that we want to see in the
	 * new css_set. while subsystems can change globally, the entries here
	 * won't change, so no need for locking.
	 */
519
	for_each_subsys(ss, i) {
520
		if (root->subsys_mask & (1UL << i)) {
521 522 523
			/* Subsystem is in this hierarchy. So we want
			 * the subsystem state from the new
			 * cgroup */
524
			template[i] = cgroup_css(cgrp, ss);
525 526 527
		} else {
			/* Subsystem is not in this hierarchy, so we
			 * don't want to change the subsystem state */
528
			template[i] = old_cset->subsys[i];
529 530 531
		}
	}

532
	key = css_set_hash(template);
533 534
	hash_for_each_possible(css_set_table, cset, hlist, key) {
		if (!compare_css_sets(cset, old_cset, cgrp, template))
535 536 537
			continue;

		/* This css_set matches what we need */
538
		return cset;
539
	}
540 541 542 543 544

	/* No existing cgroup group matched */
	return NULL;
}

545
static void free_cgrp_cset_links(struct list_head *links_to_free)
546
{
547
	struct cgrp_cset_link *link, *tmp_link;
548

549 550
	list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
		list_del(&link->cset_link);
551 552 553 554
		kfree(link);
	}
}

555 556 557 558 559 560 561
/**
 * allocate_cgrp_cset_links - allocate cgrp_cset_links
 * @count: the number of links to allocate
 * @tmp_links: list_head the allocated links are put on
 *
 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
 * through ->cset_link.  Returns 0 on success or -errno.
562
 */
563
static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
564
{
565
	struct cgrp_cset_link *link;
566
	int i;
567 568 569

	INIT_LIST_HEAD(tmp_links);

570
	for (i = 0; i < count; i++) {
571
		link = kzalloc(sizeof(*link), GFP_KERNEL);
572
		if (!link) {
573
			free_cgrp_cset_links(tmp_links);
574 575
			return -ENOMEM;
		}
576
		list_add(&link->cset_link, tmp_links);
577 578 579 580
	}
	return 0;
}

581 582
/**
 * link_css_set - a helper function to link a css_set to a cgroup
583
 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
584
 * @cset: the css_set to be linked
585 586
 * @cgrp: the destination cgroup
 */
587 588
static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
			 struct cgroup *cgrp)
589
{
590
	struct cgrp_cset_link *link;
591

592 593 594
	BUG_ON(list_empty(tmp_links));
	link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
	link->cset = cset;
595
	link->cgrp = cgrp;
596
	list_move(&link->cset_link, &cgrp->cset_links);
597 598 599 600
	/*
	 * Always add links to the tail of the list so that the list
	 * is sorted by order of hierarchy creation
	 */
601
	list_add_tail(&link->cgrp_link, &cset->cgrp_links);
602 603
}

604 605 606 607 608 609 610
/**
 * find_css_set - return a new css_set with one cgroup updated
 * @old_cset: the baseline css_set
 * @cgrp: the cgroup to be updated
 *
 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
 * substituted into the appropriate hierarchy.
611
 */
612 613
static struct css_set *find_css_set(struct css_set *old_cset,
				    struct cgroup *cgrp)
614
{
615
	struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
616
	struct css_set *cset;
617 618
	struct list_head tmp_links;
	struct cgrp_cset_link *link;
619
	unsigned long key;
620

621 622
	lockdep_assert_held(&cgroup_mutex);

623 624
	/* First see if we already have a cgroup group that matches
	 * the desired set */
625
	down_read(&css_set_rwsem);
626 627 628
	cset = find_existing_css_set(old_cset, cgrp, template);
	if (cset)
		get_css_set(cset);
629
	up_read(&css_set_rwsem);
630

631 632
	if (cset)
		return cset;
633

634
	cset = kzalloc(sizeof(*cset), GFP_KERNEL);
635
	if (!cset)
636 637
		return NULL;

638
	/* Allocate all the cgrp_cset_link objects that we'll need */
639
	if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
640
		kfree(cset);
641 642 643
		return NULL;
	}

644
	atomic_set(&cset->refcount, 1);
645
	INIT_LIST_HEAD(&cset->cgrp_links);
646 647
	INIT_LIST_HEAD(&cset->tasks);
	INIT_HLIST_NODE(&cset->hlist);
648 649 650

	/* Copy the set of subsystem state objects generated in
	 * find_existing_css_set() */
651
	memcpy(cset->subsys, template, sizeof(cset->subsys));
652

653
	down_write(&css_set_rwsem);
654
	/* Add reference counts and links from the new css_set. */
655
	list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
656
		struct cgroup *c = link->cgrp;
657

658 659
		if (c->root == cgrp->root)
			c = cgrp;
660
		link_css_set(&tmp_links, cset, c);
661
	}
662

663
	BUG_ON(!list_empty(&tmp_links));
664 665

	css_set_count++;
666 667

	/* Add this cgroup group to the hash table */
668 669
	key = css_set_hash(cset->subsys);
	hash_add(css_set_table, &cset->hlist, key);
670

671
	up_write(&css_set_rwsem);
672

673
	return cset;
674 675
}

T
Tejun Heo 已提交
676 677 678 679 680 681 682
static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
{
	struct cgroup *top_cgrp = kf_root->kn->priv;

	return top_cgrp->root;
}

683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718
static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end)
{
	int id;

	lockdep_assert_held(&cgroup_mutex);

	id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end,
			      GFP_KERNEL);
	if (id < 0)
		return id;

	root->hierarchy_id = id;
	return 0;
}

static void cgroup_exit_root_id(struct cgroupfs_root *root)
{
	lockdep_assert_held(&cgroup_mutex);

	if (root->hierarchy_id) {
		idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
		root->hierarchy_id = 0;
	}
}

static void cgroup_free_root(struct cgroupfs_root *root)
{
	if (root) {
		/* hierarhcy ID shoulid already have been released */
		WARN_ON_ONCE(root->hierarchy_id);

		idr_destroy(&root->cgroup_idr);
		kfree(root);
	}
}

T
Tejun Heo 已提交
719
static void cgroup_destroy_root(struct cgroupfs_root *root)
720
{
721 722 723
	struct cgroup *cgrp = &root->top_cgroup;
	struct cgrp_cset_link *link, *tmp_link;

T
Tejun Heo 已提交
724 725
	mutex_lock(&cgroup_tree_mutex);
	mutex_lock(&cgroup_mutex);
726

T
Tejun Heo 已提交
727
	BUG_ON(atomic_read(&root->nr_cgrps));
728 729 730
	BUG_ON(!list_empty(&cgrp->children));

	/* Rebind all subsystems back to the default hierarchy */
T
Tejun Heo 已提交
731
	WARN_ON(rebind_subsystems(root, 0, root->subsys_mask));
732 733 734 735 736

	/*
	 * Release all the links from cset_links to this hierarchy's
	 * root cgroup
	 */
737
	down_write(&css_set_rwsem);
738 739 740 741 742 743

	list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
		list_del(&link->cset_link);
		list_del(&link->cgrp_link);
		kfree(link);
	}
744
	up_write(&css_set_rwsem);
745 746 747 748 749 750 751 752 753 754 755

	if (!list_empty(&root->root_list)) {
		list_del(&root->root_list);
		cgroup_root_count--;
	}

	cgroup_exit_root_id(root);

	mutex_unlock(&cgroup_mutex);
	mutex_unlock(&cgroup_tree_mutex);

T
Tejun Heo 已提交
756
	kernfs_destroy_root(root->kf_root);
757 758 759
	cgroup_free_root(root);
}

760 761
/*
 * Return the cgroup for "task" from the given hierarchy. Must be
762
 * called with cgroup_mutex and css_set_rwsem held.
763 764 765 766
 */
static struct cgroup *task_cgroup_from_root(struct task_struct *task,
					    struct cgroupfs_root *root)
{
767
	struct css_set *cset;
768 769
	struct cgroup *res = NULL;

770 771 772
	lockdep_assert_held(&cgroup_mutex);
	lockdep_assert_held(&css_set_rwsem);

773 774 775 776 777
	/*
	 * No need to lock the task - since we hold cgroup_mutex the
	 * task can't change groups, so the only thing that can happen
	 * is that it exits and its css is set back to init_css_set.
	 */
778
	cset = task_css_set(task);
779
	if (cset == &init_css_set) {
780 781
		res = &root->top_cgroup;
	} else {
782 783 784
		struct cgrp_cset_link *link;

		list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
785
			struct cgroup *c = link->cgrp;
786

787 788 789 790 791 792
			if (c->root == root) {
				res = c;
				break;
			}
		}
	}
793

794 795 796 797
	BUG_ON(!res);
	return res;
}

798 799 800 801 802 803 804 805 806 807
/*
 * There is one global cgroup mutex. We also require taking
 * task_lock() when dereferencing a task's cgroup subsys pointers.
 * See "The task_lock() exception", at the end of this comment.
 *
 * A task must hold cgroup_mutex to modify cgroups.
 *
 * Any task can increment and decrement the count field without lock.
 * So in general, code holding cgroup_mutex can't rely on the count
 * field not changing.  However, if the count goes to zero, then only
808
 * cgroup_attach_task() can increment it again.  Because a count of zero
809 810 811 812 813 814 815 816 817 818 819 820 821
 * means that no tasks are currently attached, therefore there is no
 * way a task attached to that cgroup can fork (the other way to
 * increment the count).  So code holding cgroup_mutex can safely
 * assume that if the count is zero, it will stay zero. Similarly, if
 * a task holds cgroup_mutex on a cgroup with zero count, it
 * knows that the cgroup won't be removed, as cgroup_rmdir()
 * needs that mutex.
 *
 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
 * (usually) take cgroup_mutex.  These are the two most performance
 * critical pieces of code here.  The exception occurs on cgroup_exit(),
 * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
 * is taken, and if the cgroup count is zero, a usermode call made
L
Li Zefan 已提交
822 823
 * to the release agent with the name of the cgroup (path relative to
 * the root of cgroup file system) as the argument.
824 825 826 827 828 829 830 831 832 833 834
 *
 * A cgroup can only be deleted if both its 'count' of using tasks
 * is zero, and its list of 'children' cgroups is empty.  Since all
 * tasks in the system use _some_ cgroup, and since there is always at
 * least one task in the system (init, pid == 1), therefore, top_cgroup
 * always has either children cgroups and/or using tasks.  So we don't
 * need a special hack to ensure that top_cgroup cannot be deleted.
 *
 *	The task_lock() exception
 *
 * The need for this exception arises from the action of
835
 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
L
Li Zefan 已提交
836
 * another.  It does so using cgroup_mutex, however there are
837 838 839
 * several performance critical places that need to reference
 * task->cgroup without the expense of grabbing a system global
 * mutex.  Therefore except as noted below, when dereferencing or, as
840
 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
841 842 843 844
 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
 * the task_struct routinely used for such matters.
 *
 * P.S.  One more locking exception.  RCU is used to guard the
845
 * update of a tasks cgroup pointer by cgroup_attach_task()
846 847
 */

848
static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask);
T
Tejun Heo 已提交
849
static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
850
static const struct file_operations proc_cgroupstats_operations;
851

T
Tejun Heo 已提交
852 853 854 855 856 857 858 859 860 861 862 863
static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
			      char *buf)
{
	if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
	    !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
		snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
			 cft->ss->name, cft->name);
	else
		strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
	return buf;
}

864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889
/**
 * cgroup_file_mode - deduce file mode of a control file
 * @cft: the control file in question
 *
 * returns cft->mode if ->mode is not 0
 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
 * returns S_IRUGO if it has only a read handler
 * returns S_IWUSR if it has only a write hander
 */
static umode_t cgroup_file_mode(const struct cftype *cft)
{
	umode_t mode = 0;

	if (cft->mode)
		return cft->mode;

	if (cft->read_u64 || cft->read_s64 || cft->seq_show)
		mode |= S_IRUGO;

	if (cft->write_u64 || cft->write_s64 || cft->write_string ||
	    cft->trigger)
		mode |= S_IWUSR;

	return mode;
}

890 891
static void cgroup_free_fn(struct work_struct *work)
{
892
	struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
893

894
	atomic_dec(&cgrp->root->nr_cgrps);
895
	cgroup_pidlist_destroy_all(cgrp);
896

T
Tejun Heo 已提交
897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
	if (cgrp->parent) {
		/*
		 * We get a ref to the parent, and put the ref when this
		 * cgroup is being freed, so it's guaranteed that the
		 * parent won't be destroyed before its children.
		 */
		cgroup_put(cgrp->parent);
		kernfs_put(cgrp->kn);
		kfree(cgrp);
	} else {
		/*
		 * This is top cgroup's refcnt reaching zero, which
		 * indicates that the root should be released.
		 */
		cgroup_destroy_root(cgrp->root);
	}
913 914 915 916 917 918
}

static void cgroup_free_rcu(struct rcu_head *head)
{
	struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);

919
	INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
920
	queue_work(cgroup_destroy_wq, &cgrp->destroy_work);
921 922
}

923 924
static void cgroup_get(struct cgroup *cgrp)
{
T
Tejun Heo 已提交
925 926 927
	WARN_ON_ONCE(cgroup_is_dead(cgrp));
	WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0);
	atomic_inc(&cgrp->refcnt);
928 929
}

930 931
static void cgroup_put(struct cgroup *cgrp)
{
T
Tejun Heo 已提交
932 933
	if (!atomic_dec_and_test(&cgrp->refcnt))
		return;
T
Tejun Heo 已提交
934
	if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp)))
T
Tejun Heo 已提交
935
		return;
936

T
Tejun Heo 已提交
937 938 939 940 941 942 943 944 945 946
	/*
	 * XXX: cgrp->id is only used to look up css's.  As cgroup and
	 * css's lifetimes will be decoupled, it should be made
	 * per-subsystem and moved to css->id so that lookups are
	 * successful until the target css is released.
	 */
	mutex_lock(&cgroup_mutex);
	idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
	mutex_unlock(&cgroup_mutex);
	cgrp->id = -1;
947

T
Tejun Heo 已提交
948
	call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
949 950
}

951
static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
T
Tejun Heo 已提交
952
{
T
Tejun Heo 已提交
953
	char name[CGROUP_FILE_NAME_MAX];
T
Tejun Heo 已提交
954

T
Tejun Heo 已提交
955
	lockdep_assert_held(&cgroup_tree_mutex);
T
Tejun Heo 已提交
956
	kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
T
Tejun Heo 已提交
957 958
}

959
/**
960
 * cgroup_clear_dir - remove subsys files in a cgroup directory
961
 * @cgrp: target cgroup
962 963
 * @subsys_mask: mask of the subsystem ids whose files should be removed
 */
964
static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask)
T
Tejun Heo 已提交
965
{
966
	struct cgroup_subsys *ss;
967
	int i;
T
Tejun Heo 已提交
968

969
	for_each_subsys(ss, i) {
T
Tejun Heo 已提交
970
		struct cftype *cfts;
971 972

		if (!test_bit(i, &subsys_mask))
973
			continue;
T
Tejun Heo 已提交
974 975
		list_for_each_entry(cfts, &ss->cfts, node)
			cgroup_addrm_files(cgrp, cfts, false);
976
	}
977 978 979
}

static int rebind_subsystems(struct cgroupfs_root *root,
980
			     unsigned long added_mask, unsigned removed_mask)
981
{
982
	struct cgroup *cgrp = &root->top_cgroup;
983
	struct cgroup_subsys *ss;
984
	int i, ret;
985

T
Tejun Heo 已提交
986 987
	lockdep_assert_held(&cgroup_tree_mutex);
	lockdep_assert_held(&cgroup_mutex);
B
Ben Blum 已提交
988

989
	/* Check that any added subsystems are currently free */
T
Tejun Heo 已提交
990 991 992
	for_each_subsys(ss, i)
		if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root)
			return -EBUSY;
993

994 995
	ret = cgroup_populate_dir(cgrp, added_mask);
	if (ret)
T
Tejun Heo 已提交
996
		return ret;
997 998 999 1000 1001

	/*
	 * Nothing can fail from this point on.  Remove files for the
	 * removed subsystems and rebind each subsystem.
	 */
1002
	mutex_unlock(&cgroup_mutex);
1003
	cgroup_clear_dir(cgrp, removed_mask);
1004
	mutex_lock(&cgroup_mutex);
1005

1006
	for_each_subsys(ss, i) {
1007
		unsigned long bit = 1UL << i;
1008

1009
		if (bit & added_mask) {
1010
			/* We're binding this subsystem to this hierarchy */
1011 1012 1013
			BUG_ON(cgroup_css(cgrp, ss));
			BUG_ON(!cgroup_css(cgroup_dummy_top, ss));
			BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top);
1014

1015
			rcu_assign_pointer(cgrp->subsys[i],
1016 1017
					   cgroup_css(cgroup_dummy_top, ss));
			cgroup_css(cgrp, ss)->cgroup = cgrp;
1018

1019
			ss->root = root;
1020
			if (ss->bind)
1021
				ss->bind(cgroup_css(cgrp, ss));
1022

B
Ben Blum 已提交
1023
			/* refcount was already taken, and we're keeping it */
1024
			root->subsys_mask |= bit;
1025
		} else if (bit & removed_mask) {
1026
			/* We're removing this subsystem */
1027 1028
			BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss));
			BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp);
1029

1030
			if (ss->bind)
1031
				ss->bind(cgroup_css(cgroup_dummy_top, ss));
1032

1033
			cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top;
1034 1035
			RCU_INIT_POINTER(cgrp->subsys[i], NULL);

1036
			cgroup_subsys[i]->root = &cgroup_dummy_root;
1037
			root->subsys_mask &= ~bit;
1038 1039 1040
		}
	}

T
Tejun Heo 已提交
1041
	kernfs_activate(cgrp->kn);
1042 1043 1044
	return 0;
}

T
Tejun Heo 已提交
1045 1046
static int cgroup_show_options(struct seq_file *seq,
			       struct kernfs_root *kf_root)
1047
{
T
Tejun Heo 已提交
1048
	struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);
1049
	struct cgroup_subsys *ss;
T
Tejun Heo 已提交
1050
	int ssid;
1051

T
Tejun Heo 已提交
1052 1053 1054
	for_each_subsys(ss, ssid)
		if (root->subsys_mask & (1 << ssid))
			seq_printf(seq, ",%s", ss->name);
1055 1056
	if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
		seq_puts(seq, ",sane_behavior");
1057
	if (root->flags & CGRP_ROOT_NOPREFIX)
1058
		seq_puts(seq, ",noprefix");
1059
	if (root->flags & CGRP_ROOT_XATTR)
A
Aristeu Rozanski 已提交
1060
		seq_puts(seq, ",xattr");
1061 1062

	spin_lock(&release_agent_path_lock);
1063 1064
	if (strlen(root->release_agent_path))
		seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1065 1066
	spin_unlock(&release_agent_path_lock);

1067
	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1068
		seq_puts(seq, ",clone_children");
1069 1070
	if (strlen(root->name))
		seq_printf(seq, ",name=%s", root->name);
1071 1072 1073 1074
	return 0;
}

struct cgroup_sb_opts {
1075
	unsigned long subsys_mask;
1076
	unsigned long flags;
1077
	char *release_agent;
1078
	bool cpuset_clone_children;
1079
	char *name;
1080 1081
	/* User explicitly requested empty subsystem */
	bool none;
1082 1083
};

B
Ben Blum 已提交
1084
/*
1085 1086 1087 1088
 * Convert a hierarchy specifier into a bitmask of subsystems and
 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
 * array. This function takes refcounts on subsystems to be used, unless it
 * returns error, in which case no refcounts are taken.
B
Ben Blum 已提交
1089
 */
B
Ben Blum 已提交
1090
static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1091
{
1092 1093
	char *token, *o = data;
	bool all_ss = false, one_ss = false;
1094
	unsigned long mask = (unsigned long)-1;
1095 1096
	struct cgroup_subsys *ss;
	int i;
1097

B
Ben Blum 已提交
1098 1099
	BUG_ON(!mutex_is_locked(&cgroup_mutex));

1100
#ifdef CONFIG_CPUSETS
1101
	mask = ~(1UL << cpuset_cgrp_id);
1102
#endif
1103

1104
	memset(opts, 0, sizeof(*opts));
1105 1106 1107 1108

	while ((token = strsep(&o, ",")) != NULL) {
		if (!*token)
			return -EINVAL;
1109
		if (!strcmp(token, "none")) {
1110 1111
			/* Explicitly have no subsystems */
			opts->none = true;
1112 1113 1114 1115 1116 1117 1118 1119 1120
			continue;
		}
		if (!strcmp(token, "all")) {
			/* Mutually exclusive option 'all' + subsystem name */
			if (one_ss)
				return -EINVAL;
			all_ss = true;
			continue;
		}
1121 1122 1123 1124
		if (!strcmp(token, "__DEVEL__sane_behavior")) {
			opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
			continue;
		}
1125
		if (!strcmp(token, "noprefix")) {
1126
			opts->flags |= CGRP_ROOT_NOPREFIX;
1127 1128 1129
			continue;
		}
		if (!strcmp(token, "clone_children")) {
1130
			opts->cpuset_clone_children = true;
1131 1132
			continue;
		}
A
Aristeu Rozanski 已提交
1133
		if (!strcmp(token, "xattr")) {
1134
			opts->flags |= CGRP_ROOT_XATTR;
A
Aristeu Rozanski 已提交
1135 1136
			continue;
		}
1137
		if (!strncmp(token, "release_agent=", 14)) {
1138 1139 1140
			/* Specifying two release agents is forbidden */
			if (opts->release_agent)
				return -EINVAL;
1141
			opts->release_agent =
1142
				kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1143 1144
			if (!opts->release_agent)
				return -ENOMEM;
1145 1146 1147
			continue;
		}
		if (!strncmp(token, "name=", 5)) {
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
			const char *name = token + 5;
			/* Can't specify an empty name */
			if (!strlen(name))
				return -EINVAL;
			/* Must match [\w.-]+ */
			for (i = 0; i < strlen(name); i++) {
				char c = name[i];
				if (isalnum(c))
					continue;
				if ((c == '.') || (c == '-') || (c == '_'))
					continue;
				return -EINVAL;
			}
			/* Specifying two names is forbidden */
			if (opts->name)
				return -EINVAL;
			opts->name = kstrndup(name,
1165
					      MAX_CGROUP_ROOT_NAMELEN - 1,
1166 1167 1168
					      GFP_KERNEL);
			if (!opts->name)
				return -ENOMEM;
1169 1170 1171 1172

			continue;
		}

1173
		for_each_subsys(ss, i) {
1174 1175 1176 1177 1178 1179 1180 1181
			if (strcmp(token, ss->name))
				continue;
			if (ss->disabled)
				continue;

			/* Mutually exclusive option 'all' + subsystem name */
			if (all_ss)
				return -EINVAL;
1182
			set_bit(i, &opts->subsys_mask);
1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
			one_ss = true;

			break;
		}
		if (i == CGROUP_SUBSYS_COUNT)
			return -ENOENT;
	}

	/*
	 * If the 'all' option was specified select all the subsystems,
1193 1194
	 * otherwise if 'none', 'name=' and a subsystem name options
	 * were not specified, let's default to 'all'
1195
	 */
1196 1197 1198 1199
	if (all_ss || (!one_ss && !opts->none && !opts->name))
		for_each_subsys(ss, i)
			if (!ss->disabled)
				set_bit(i, &opts->subsys_mask);
1200

1201 1202
	/* Consistency checks */

1203 1204 1205
	if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
		pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");

1206 1207 1208 1209
		if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) ||
		    opts->cpuset_clone_children || opts->release_agent ||
		    opts->name) {
			pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n");
1210 1211 1212 1213
			return -EINVAL;
		}
	}

1214 1215 1216 1217 1218
	/*
	 * Option noprefix was introduced just for backward compatibility
	 * with the old cpuset, so we allow noprefix only if mounting just
	 * the cpuset subsystem.
	 */
1219
	if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1220 1221
		return -EINVAL;

1222 1223

	/* Can't specify "none" and some subsystems */
1224
	if (opts->subsys_mask && opts->none)
1225 1226 1227 1228 1229 1230
		return -EINVAL;

	/*
	 * We either have to specify by name or by subsystems. (So all
	 * empty hierarchies must have a name).
	 */
1231
	if (!opts->subsys_mask && !opts->name)
1232 1233 1234 1235 1236
		return -EINVAL;

	return 0;
}

T
Tejun Heo 已提交
1237
static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1238 1239
{
	int ret = 0;
T
Tejun Heo 已提交
1240
	struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);
1241
	struct cgroup_sb_opts opts;
1242
	unsigned long added_mask, removed_mask;
1243

1244 1245 1246 1247 1248
	if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
		pr_err("cgroup: sane_behavior: remount is not allowed\n");
		return -EINVAL;
	}

T
Tejun Heo 已提交
1249
	mutex_lock(&cgroup_tree_mutex);
1250 1251 1252 1253 1254 1255 1256
	mutex_lock(&cgroup_mutex);

	/* See what subsystems are wanted */
	ret = parse_cgroupfs_options(data, &opts);
	if (ret)
		goto out_unlock;

1257
	if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1258 1259 1260
		pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
			   task_tgid_nr(current), current->comm);

1261 1262
	added_mask = opts.subsys_mask & ~root->subsys_mask;
	removed_mask = root->subsys_mask & ~opts.subsys_mask;
1263

B
Ben Blum 已提交
1264
	/* Don't allow flags or name to change at remount */
1265
	if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) ||
B
Ben Blum 已提交
1266
	    (opts.name && strcmp(opts.name, root->name))) {
1267 1268 1269
		pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
		       opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "",
		       root->flags & CGRP_ROOT_OPTION_MASK, root->name);
1270 1271 1272 1273
		ret = -EINVAL;
		goto out_unlock;
	}

1274
	/* remounting is not allowed for populated hierarchies */
1275
	if (!list_empty(&root->top_cgroup.children)) {
1276
		ret = -EBUSY;
1277
		goto out_unlock;
B
Ben Blum 已提交
1278
	}
1279

1280
	ret = rebind_subsystems(root, added_mask, removed_mask);
1281
	if (ret)
1282
		goto out_unlock;
1283

1284 1285
	if (opts.release_agent) {
		spin_lock(&release_agent_path_lock);
1286
		strcpy(root->release_agent_path, opts.release_agent);
1287 1288
		spin_unlock(&release_agent_path_lock);
	}
1289
 out_unlock:
1290
	kfree(opts.release_agent);
1291
	kfree(opts.name);
1292
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
1293
	mutex_unlock(&cgroup_tree_mutex);
1294 1295 1296
	return ret;
}

1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308
/*
 * To reduce the fork() overhead for systems that are not actually using
 * their cgroups capability, we don't maintain the lists running through
 * each css_set to its tasks until we see the list actually used - in other
 * words after the first mount.
 */
static bool use_task_css_set_links __read_mostly;

static void cgroup_enable_task_cg_lists(void)
{
	struct task_struct *p, *g;

1309
	down_write(&css_set_rwsem);
1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341

	if (use_task_css_set_links)
		goto out_unlock;

	use_task_css_set_links = true;

	/*
	 * We need tasklist_lock because RCU is not safe against
	 * while_each_thread(). Besides, a forking task that has passed
	 * cgroup_post_fork() without seeing use_task_css_set_links = 1
	 * is not guaranteed to have its child immediately visible in the
	 * tasklist if we walk through it with RCU.
	 */
	read_lock(&tasklist_lock);
	do_each_thread(g, p) {
		task_lock(p);

		WARN_ON_ONCE(!list_empty(&p->cg_list) ||
			     task_css_set(p) != &init_css_set);

		/*
		 * We should check if the process is exiting, otherwise
		 * it will race with cgroup_exit() in that the list
		 * entry won't be deleted though the process has exited.
		 */
		if (!(p->flags & PF_EXITING))
			list_add(&p->cg_list, &task_css_set(p)->tasks);

		task_unlock(p);
	} while_each_thread(g, p);
	read_unlock(&tasklist_lock);
out_unlock:
1342
	up_write(&css_set_rwsem);
1343 1344
}

1345 1346
static void init_cgroup_housekeeping(struct cgroup *cgrp)
{
T
Tejun Heo 已提交
1347
	atomic_set(&cgrp->refcnt, 1);
1348 1349
	INIT_LIST_HEAD(&cgrp->sibling);
	INIT_LIST_HEAD(&cgrp->children);
1350
	INIT_LIST_HEAD(&cgrp->cset_links);
1351
	INIT_LIST_HEAD(&cgrp->release_list);
1352 1353
	INIT_LIST_HEAD(&cgrp->pidlists);
	mutex_init(&cgrp->pidlist_mutex);
T
Tejun Heo 已提交
1354
	cgrp->dummy_css.cgroup = cgrp;
1355
}
1356

1357 1358
static void init_cgroup_root(struct cgroupfs_root *root)
{
1359
	struct cgroup *cgrp = &root->top_cgroup;
1360

1361
	INIT_LIST_HEAD(&root->root_list);
1362
	atomic_set(&root->nr_cgrps, 1);
1363
	cgrp->root = root;
1364
	init_cgroup_housekeeping(cgrp);
1365
	idr_init(&root->cgroup_idr);
1366 1367
}

1368 1369 1370 1371
static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
{
	struct cgroupfs_root *root;

1372
	if (!opts->subsys_mask && !opts->none)
T
Tejun Heo 已提交
1373
		return ERR_PTR(-EINVAL);
1374 1375 1376 1377 1378 1379

	root = kzalloc(sizeof(*root), GFP_KERNEL);
	if (!root)
		return ERR_PTR(-ENOMEM);

	init_cgroup_root(root);
1380

1381 1382 1383 1384 1385
	root->flags = opts->flags;
	if (opts->release_agent)
		strcpy(root->release_agent_path, opts->release_agent);
	if (opts->name)
		strcpy(root->name, opts->name);
1386 1387
	if (opts->cpuset_clone_children)
		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1388 1389 1390
	return root;
}

T
Tejun Heo 已提交
1391
static int cgroup_setup_root(struct cgroupfs_root *root, unsigned long ss_mask)
1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
{
	LIST_HEAD(tmp_links);
	struct cgroup *root_cgrp = &root->top_cgroup;
	struct css_set *cset;
	int i, ret;

	lockdep_assert_held(&cgroup_tree_mutex);
	lockdep_assert_held(&cgroup_mutex);

	ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL);
	if (ret < 0)
T
Tejun Heo 已提交
1403
		goto out;
1404 1405 1406
	root_cgrp->id = ret;

	/*
1407
	 * We're accessing css_set_count without locking css_set_rwsem here,
1408 1409 1410 1411 1412 1413
	 * but that's OK - it can only be increased by someone holding
	 * cgroup_lock, and that's us. The worst that can happen is that we
	 * have some link structures left over
	 */
	ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
	if (ret)
T
Tejun Heo 已提交
1414
		goto out;
1415 1416 1417 1418

	/* ID 0 is reserved for dummy root, 1 for unified hierarchy */
	ret = cgroup_init_root_id(root, 2, 0);
	if (ret)
T
Tejun Heo 已提交
1419
		goto out;
1420

T
Tejun Heo 已提交
1421 1422 1423 1424 1425 1426 1427 1428
	root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
					   KERNFS_ROOT_CREATE_DEACTIVATED,
					   root_cgrp);
	if (IS_ERR(root->kf_root)) {
		ret = PTR_ERR(root->kf_root);
		goto exit_root_id;
	}
	root_cgrp->kn = root->kf_root->kn;
1429 1430 1431

	ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true);
	if (ret)
T
Tejun Heo 已提交
1432
		goto destroy_root;
1433

T
Tejun Heo 已提交
1434
	ret = rebind_subsystems(root, ss_mask, 0);
1435
	if (ret)
T
Tejun Heo 已提交
1436
		goto destroy_root;
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449

	/*
	 * There must be no failure case after here, since rebinding takes
	 * care of subsystems' refcounts, which are explicitly dropped in
	 * the failure exit path.
	 */
	list_add(&root->root_list, &cgroup_roots);
	cgroup_root_count++;

	/*
	 * Link the top cgroup in this hierarchy into all the css_set
	 * objects.
	 */
1450
	down_write(&css_set_rwsem);
1451 1452
	hash_for_each(css_set_table, i, cset, hlist)
		link_css_set(&tmp_links, cset, root_cgrp);
1453
	up_write(&css_set_rwsem);
1454 1455

	BUG_ON(!list_empty(&root_cgrp->children));
1456
	BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1457

T
Tejun Heo 已提交
1458
	kernfs_activate(root_cgrp->kn);
1459
	ret = 0;
T
Tejun Heo 已提交
1460
	goto out;
1461

T
Tejun Heo 已提交
1462 1463 1464 1465
destroy_root:
	kernfs_destroy_root(root->kf_root);
	root->kf_root = NULL;
exit_root_id:
1466
	cgroup_exit_root_id(root);
T
Tejun Heo 已提交
1467
out:
1468 1469 1470 1471
	free_cgrp_cset_links(&tmp_links);
	return ret;
}

A
Al Viro 已提交
1472
static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1473
			 int flags, const char *unused_dev_name,
A
Al Viro 已提交
1474
			 void *data)
1475
{
T
Tejun Heo 已提交
1476
	struct cgroupfs_root *root;
1477
	struct cgroup_sb_opts opts;
T
Tejun Heo 已提交
1478
	struct dentry *dentry;
1479
	int ret;
1480 1481 1482 1483 1484 1485 1486

	/*
	 * The first time anyone tries to mount a cgroup, enable the list
	 * linking each css_set to its tasks and fix up all existing tasks.
	 */
	if (!use_task_css_set_links)
		cgroup_enable_task_cg_lists();
T
Tejun Heo 已提交
1487
retry:
1488
	mutex_lock(&cgroup_tree_mutex);
B
Ben Blum 已提交
1489
	mutex_lock(&cgroup_mutex);
1490 1491

	/* First find the desired set of subsystems */
1492
	ret = parse_cgroupfs_options(data, &opts);
1493
	if (ret)
1494
		goto out_unlock;
1495

T
Tejun Heo 已提交
1496 1497 1498
	/* look for a matching existing root */
	for_each_active_root(root) {
		bool name_match = false;
1499

T
Tejun Heo 已提交
1500 1501 1502 1503 1504 1505 1506 1507 1508 1509
		/*
		 * If we asked for a name then it must match.  Also, if
		 * name matches but sybsys_mask doesn't, we should fail.
		 * Remember whether name matched.
		 */
		if (opts.name) {
			if (strcmp(opts.name, root->name))
				continue;
			name_match = true;
		}
1510

1511
		/*
T
Tejun Heo 已提交
1512 1513
		 * If we asked for subsystems (or explicitly for no
		 * subsystems) then they must match.
1514
		 */
T
Tejun Heo 已提交
1515 1516 1517 1518 1519 1520 1521
		if ((opts.subsys_mask || opts.none) &&
		    (opts.subsys_mask != root->subsys_mask)) {
			if (!name_match)
				continue;
			ret = -EBUSY;
			goto out_unlock;
		}
1522

1523
		if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) {
1524 1525 1526
			if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
				pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
				ret = -EINVAL;
1527
				goto out_unlock;
1528 1529 1530
			} else {
				pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
			}
1531
		}
T
Tejun Heo 已提交
1532

T
Tejun Heo 已提交
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542
		/*
		 * A root's lifetime is governed by its top cgroup.  Zero
		 * ref indicate that the root is being destroyed.  Wait for
		 * destruction to complete so that the subsystems are free.
		 * We can use wait_queue for the wait but this path is
		 * super cold.  Let's just sleep for a bit and retry.
		 */
		if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) {
			mutex_unlock(&cgroup_mutex);
			mutex_unlock(&cgroup_tree_mutex);
1543 1544
			kfree(opts.release_agent);
			kfree(opts.name);
T
Tejun Heo 已提交
1545 1546 1547 1548 1549
			msleep(10);
			goto retry;
		}

		ret = 0;
T
Tejun Heo 已提交
1550
		goto out_unlock;
1551 1552
	}

T
Tejun Heo 已提交
1553 1554 1555 1556 1557 1558 1559
	/* no such thing, create a new one */
	root = cgroup_root_from_opts(&opts);
	if (IS_ERR(root)) {
		ret = PTR_ERR(root);
		goto out_unlock;
	}

T
Tejun Heo 已提交
1560
	ret = cgroup_setup_root(root, opts.subsys_mask);
T
Tejun Heo 已提交
1561 1562 1563
	if (ret)
		cgroup_free_root(root);

1564
out_unlock:
T
Tejun Heo 已提交
1565
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
1566
	mutex_unlock(&cgroup_tree_mutex);
1567

1568 1569
	kfree(opts.release_agent);
	kfree(opts.name);
1570

T
Tejun Heo 已提交
1571
	if (ret)
1572
		return ERR_PTR(ret);
T
Tejun Heo 已提交
1573 1574 1575

	dentry = kernfs_mount(fs_type, flags, root->kf_root);
	if (IS_ERR(dentry))
T
Tejun Heo 已提交
1576
		cgroup_put(&root->top_cgroup);
T
Tejun Heo 已提交
1577 1578 1579 1580 1581 1582 1583 1584
	return dentry;
}

static void cgroup_kill_sb(struct super_block *sb)
{
	struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
	struct cgroupfs_root *root = cgroup_root_from_kf(kf_root);

T
Tejun Heo 已提交
1585
	cgroup_put(&root->top_cgroup);
T
Tejun Heo 已提交
1586
	kernfs_kill_sb(sb);
1587 1588 1589 1590
}

static struct file_system_type cgroup_fs_type = {
	.name = "cgroup",
A
Al Viro 已提交
1591
	.mount = cgroup_mount,
1592 1593 1594
	.kill_sb = cgroup_kill_sb,
};

1595 1596
static struct kobject *cgroup_kobj;

1597
/**
1598
 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1599 1600 1601 1602
 * @task: target task
 * @buf: the buffer to write the path into
 * @buflen: the length of the buffer
 *
1603 1604 1605 1606 1607
 * Determine @task's cgroup on the first (the one with the lowest non-zero
 * hierarchy_id) cgroup hierarchy and copy its path into @buf.  This
 * function grabs cgroup_mutex and shouldn't be used inside locks used by
 * cgroup controller callbacks.
 *
T
Tejun Heo 已提交
1608
 * Return value is the same as kernfs_path().
1609
 */
T
Tejun Heo 已提交
1610
char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
1611 1612
{
	struct cgroupfs_root *root;
1613
	struct cgroup *cgrp;
T
Tejun Heo 已提交
1614 1615
	int hierarchy_id = 1;
	char *path = NULL;
1616 1617

	mutex_lock(&cgroup_mutex);
1618
	down_read(&css_set_rwsem);
1619

1620 1621
	root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);

1622 1623
	if (root) {
		cgrp = task_cgroup_from_root(task, root);
T
Tejun Heo 已提交
1624
		path = cgroup_path(cgrp, buf, buflen);
1625 1626
	} else {
		/* if no hierarchy exists, everyone is in "/" */
T
Tejun Heo 已提交
1627 1628
		if (strlcpy(buf, "/", buflen) < buflen)
			path = buf;
1629 1630
	}

1631
	up_read(&css_set_rwsem);
1632
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
1633
	return path;
1634
}
1635
EXPORT_SYMBOL_GPL(task_cgroup_path);
1636

1637 1638 1639
/*
 * Control Group taskset
 */
1640 1641 1642
struct task_and_cgroup {
	struct task_struct	*task;
	struct cgroup		*cgrp;
L
Li Zefan 已提交
1643
	struct css_set		*cset;
1644 1645
};

1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
struct cgroup_taskset {
	struct task_and_cgroup	single;
	struct flex_array	*tc_array;
	int			tc_array_len;
	int			idx;
};

/**
 * cgroup_taskset_first - reset taskset and return the first task
 * @tset: taskset of interest
 *
 * @tset iteration is initialized and the first task is returned.
 */
struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
{
	if (tset->tc_array) {
		tset->idx = 0;
		return cgroup_taskset_next(tset);
	} else {
		return tset->single.task;
	}
}

/**
 * cgroup_taskset_next - iterate to the next task in taskset
 * @tset: taskset of interest
 *
 * Return the next task in @tset.  Iteration must have been initialized
 * with cgroup_taskset_first().
 */
struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
{
	struct task_and_cgroup *tc;

	if (!tset->tc_array || tset->idx >= tset->tc_array_len)
		return NULL;

	tc = flex_array_get(tset->tc_array, tset->idx++);
	return tc->task;
}

1687
/**
B
Ben Blum 已提交
1688
 * cgroup_task_migrate - move a task from one cgroup to another.
1689 1690 1691
 * @old_cgrp; the cgroup @tsk is being migrated from
 * @tsk: the task being migrated
 * @new_cset: the new css_set @tsk is being attached to
B
Ben Blum 已提交
1692
 *
1693
 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
B
Ben Blum 已提交
1694
 */
1695 1696 1697
static void cgroup_task_migrate(struct cgroup *old_cgrp,
				struct task_struct *tsk,
				struct css_set *new_cset)
B
Ben Blum 已提交
1698
{
1699
	struct css_set *old_cset;
B
Ben Blum 已提交
1700

1701 1702 1703
	lockdep_assert_held(&cgroup_mutex);
	lockdep_assert_held(&css_set_rwsem);

B
Ben Blum 已提交
1704
	/*
1705 1706 1707
	 * We are synchronized through threadgroup_lock() against PF_EXITING
	 * setting such that we can't race against cgroup_exit() changing the
	 * css_set to init_css_set and dropping the old one.
B
Ben Blum 已提交
1708
	 */
1709
	WARN_ON_ONCE(tsk->flags & PF_EXITING);
1710
	old_cset = task_css_set(tsk);
B
Ben Blum 已提交
1711 1712

	task_lock(tsk);
1713
	rcu_assign_pointer(tsk->cgroups, new_cset);
B
Ben Blum 已提交
1714 1715
	task_unlock(tsk);

1716
	list_move(&tsk->cg_list, &new_cset->tasks);
B
Ben Blum 已提交
1717 1718

	/*
1719 1720 1721
	 * We just gained a reference on old_cset by taking it from the
	 * task. As trading it for new_cset is protected by cgroup_mutex,
	 * we're safe to drop it here; it will be freed under RCU.
B
Ben Blum 已提交
1722
	 */
1723
	set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1724
	put_css_set_locked(old_cset, false);
B
Ben Blum 已提交
1725 1726
}

L
Li Zefan 已提交
1727
/**
1728
 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
B
Ben Blum 已提交
1729
 * @cgrp: the cgroup to attach to
1730
 * @leader: the task or the leader of the threadgroup to be attached
1731
 * @threadgroup: attach the whole threadgroup?
B
Ben Blum 已提交
1732
 *
1733
 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1734
 * task_lock of @tsk or each thread in the threadgroup individually in turn.
B
Ben Blum 已提交
1735
 */
1736
static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *leader,
T
Tejun Heo 已提交
1737
			      bool threadgroup)
B
Ben Blum 已提交
1738
{
1739
	int ret, i, group_size;
B
Ben Blum 已提交
1740
	struct cgroupfs_root *root = cgrp->root;
T
Tejun Heo 已提交
1741
	struct cgroup_subsys_state *css, *failed_css = NULL;
B
Ben Blum 已提交
1742
	/* threadgroup list cursor and array */
1743
	struct task_struct *task;
1744
	struct task_and_cgroup *tc;
1745
	struct flex_array *group;
1746
	struct cgroup_taskset tset = { };
B
Ben Blum 已提交
1747 1748 1749 1750 1751

	/*
	 * step 0: in order to do expensive, possibly blocking operations for
	 * every thread, we cannot iterate the thread group list, since it needs
	 * rcu or tasklist locked. instead, build an array of all threads in the
1752 1753
	 * group - group_rwsem prevents new threads from appearing, and if
	 * threads exit, this will just be an over-estimate.
B
Ben Blum 已提交
1754
	 */
1755
	if (threadgroup)
1756
		group_size = get_nr_threads(leader);
1757 1758
	else
		group_size = 1;
1759
	/* flex_array supports very large thread-groups better than kmalloc. */
1760
	group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
B
Ben Blum 已提交
1761 1762
	if (!group)
		return -ENOMEM;
1763
	/* pre-allocate to guarantee space while iterating in rcu read-side. */
1764 1765
	ret = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
	if (ret)
1766
		goto out_free_group_list;
B
Ben Blum 已提交
1767 1768

	i = 0;
1769 1770 1771 1772 1773
	/*
	 * Prevent freeing of tasks while we take a snapshot. Tasks that are
	 * already PF_EXITING could be freed from underneath us unless we
	 * take an rcu_read_lock.
	 */
1774
	down_read(&css_set_rwsem);
1775
	rcu_read_lock();
1776
	task = leader;
B
Ben Blum 已提交
1777
	do {
1778 1779
		struct task_and_cgroup ent;

1780 1781
		/* @task either already exited or can't exit until the end */
		if (task->flags & PF_EXITING)
1782
			goto next;
1783

B
Ben Blum 已提交
1784 1785
		/* as per above, nr_threads may decrease, but not increase. */
		BUG_ON(i >= group_size);
1786 1787
		ent.task = task;
		ent.cgrp = task_cgroup_from_root(task, root);
1788 1789
		/* nothing to do if this task is already in the cgroup */
		if (ent.cgrp == cgrp)
1790
			goto next;
1791 1792 1793 1794
		/*
		 * saying GFP_ATOMIC has no effect here because we did prealloc
		 * earlier, but it's good form to communicate our expectations.
		 */
1795 1796
		ret = flex_array_put(group, i, &ent, GFP_ATOMIC);
		BUG_ON(ret != 0);
B
Ben Blum 已提交
1797
		i++;
1798
	next:
1799 1800
		if (!threadgroup)
			break;
1801
	} while_each_thread(leader, task);
1802
	rcu_read_unlock();
1803
	up_read(&css_set_rwsem);
B
Ben Blum 已提交
1804 1805
	/* remember the number of threads in the array for later. */
	group_size = i;
1806 1807
	tset.tc_array = group;
	tset.tc_array_len = group_size;
B
Ben Blum 已提交
1808

1809
	/* methods shouldn't be called if no task is actually migrating */
1810
	ret = 0;
1811
	if (!group_size)
1812
		goto out_free_group_list;
1813

B
Ben Blum 已提交
1814 1815 1816
	/*
	 * step 1: check that we can legitimately attach to the cgroup.
	 */
T
Tejun Heo 已提交
1817 1818
	for_each_css(css, i, cgrp) {
		if (css->ss->can_attach) {
1819 1820
			ret = css->ss->can_attach(css, &tset);
			if (ret) {
T
Tejun Heo 已提交
1821
				failed_css = css;
B
Ben Blum 已提交
1822 1823 1824 1825 1826 1827 1828 1829 1830 1831
				goto out_cancel_attach;
			}
		}
	}

	/*
	 * step 2: make sure css_sets exist for all threads to be migrated.
	 * we use find_css_set, which allocates a new one if necessary.
	 */
	for (i = 0; i < group_size; i++) {
1832 1833
		struct css_set *old_cset;

1834
		tc = flex_array_get(group, i);
1835
		old_cset = task_css_set(tc->task);
L
Li Zefan 已提交
1836 1837
		tc->cset = find_css_set(old_cset, cgrp);
		if (!tc->cset) {
1838
			ret = -ENOMEM;
1839
			goto out_put_css_set_refs;
B
Ben Blum 已提交
1840 1841 1842 1843
		}
	}

	/*
1844 1845 1846
	 * step 3: now that we're guaranteed success wrt the css_sets,
	 * proceed to move all tasks to the new cgroup.  There are no
	 * failure cases after here, so this is the commit point.
B
Ben Blum 已提交
1847
	 */
1848
	down_write(&css_set_rwsem);
B
Ben Blum 已提交
1849
	for (i = 0; i < group_size; i++) {
1850
		tc = flex_array_get(group, i);
L
Li Zefan 已提交
1851
		cgroup_task_migrate(tc->cgrp, tc->task, tc->cset);
B
Ben Blum 已提交
1852
	}
1853
	up_write(&css_set_rwsem);
B
Ben Blum 已提交
1854 1855 1856
	/* nothing is sensitive to fork() after this point. */

	/*
1857
	 * step 4: do subsystem attach callbacks.
B
Ben Blum 已提交
1858
	 */
T
Tejun Heo 已提交
1859 1860 1861
	for_each_css(css, i, cgrp)
		if (css->ss->attach)
			css->ss->attach(css, &tset);
B
Ben Blum 已提交
1862 1863 1864 1865

	/*
	 * step 5: success! and cleanup
	 */
1866
	ret = 0;
1867
out_put_css_set_refs:
1868
	if (ret) {
1869 1870
		for (i = 0; i < group_size; i++) {
			tc = flex_array_get(group, i);
L
Li Zefan 已提交
1871
			if (!tc->cset)
1872
				break;
1873
			put_css_set(tc->cset, false);
1874
		}
B
Ben Blum 已提交
1875 1876
	}
out_cancel_attach:
1877
	if (ret) {
T
Tejun Heo 已提交
1878 1879
		for_each_css(css, i, cgrp) {
			if (css == failed_css)
B
Ben Blum 已提交
1880
				break;
T
Tejun Heo 已提交
1881 1882
			if (css->ss->cancel_attach)
				css->ss->cancel_attach(css, &tset);
B
Ben Blum 已提交
1883 1884 1885
		}
	}
out_free_group_list:
1886
	flex_array_free(group);
1887
	return ret;
B
Ben Blum 已提交
1888 1889 1890 1891
}

/*
 * Find the task_struct of the task to attach by vpid and pass it along to the
1892 1893
 * function to attach either it or all tasks in its threadgroup. Will lock
 * cgroup_mutex and threadgroup; may take task_lock of task.
1894
 */
B
Ben Blum 已提交
1895
static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
1896 1897
{
	struct task_struct *tsk;
1898
	const struct cred *cred = current_cred(), *tcred;
1899 1900
	int ret;

B
Ben Blum 已提交
1901 1902 1903
	if (!cgroup_lock_live_group(cgrp))
		return -ENODEV;

1904 1905
retry_find_task:
	rcu_read_lock();
1906
	if (pid) {
1907
		tsk = find_task_by_vpid(pid);
B
Ben Blum 已提交
1908 1909
		if (!tsk) {
			rcu_read_unlock();
S
SeongJae Park 已提交
1910
			ret = -ESRCH;
1911
			goto out_unlock_cgroup;
1912
		}
B
Ben Blum 已提交
1913 1914 1915 1916
		/*
		 * even if we're attaching all tasks in the thread group, we
		 * only need to check permissions on one of them.
		 */
1917
		tcred = __task_cred(tsk);
1918 1919 1920
		if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
		    !uid_eq(cred->euid, tcred->uid) &&
		    !uid_eq(cred->euid, tcred->suid)) {
1921
			rcu_read_unlock();
1922 1923
			ret = -EACCES;
			goto out_unlock_cgroup;
1924
		}
1925 1926
	} else
		tsk = current;
1927 1928

	if (threadgroup)
1929
		tsk = tsk->group_leader;
1930 1931

	/*
1932
	 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
1933 1934 1935
	 * trapped in a cpuset, or RT worker may be born in a cgroup
	 * with no rt_runtime allocated.  Just say no.
	 */
1936
	if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
1937 1938 1939 1940 1941
		ret = -EINVAL;
		rcu_read_unlock();
		goto out_unlock_cgroup;
	}

1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958
	get_task_struct(tsk);
	rcu_read_unlock();

	threadgroup_lock(tsk);
	if (threadgroup) {
		if (!thread_group_leader(tsk)) {
			/*
			 * a race with de_thread from another thread's exec()
			 * may strip us of our leadership, if this happens,
			 * there is no choice but to throw this task away and
			 * try again; this is
			 * "double-double-toil-and-trouble-check locking".
			 */
			threadgroup_unlock(tsk);
			put_task_struct(tsk);
			goto retry_find_task;
		}
1959 1960 1961 1962
	}

	ret = cgroup_attach_task(cgrp, tsk, threadgroup);

1963 1964
	threadgroup_unlock(tsk);

1965
	put_task_struct(tsk);
1966
out_unlock_cgroup:
T
Tejun Heo 已提交
1967
	mutex_unlock(&cgroup_mutex);
1968 1969 1970
	return ret;
}

1971 1972 1973 1974 1975 1976 1977 1978 1979 1980
/**
 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
 * @from: attach to all cgroups of a given task
 * @tsk: the task to be attached
 */
int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
{
	struct cgroupfs_root *root;
	int retval = 0;

T
Tejun Heo 已提交
1981
	mutex_lock(&cgroup_mutex);
1982
	for_each_active_root(root) {
1983 1984 1985 1986 1987
		struct cgroup *from_cgrp;

		down_read(&css_set_rwsem);
		from_cgrp = task_cgroup_from_root(from, root);
		up_read(&css_set_rwsem);
1988

L
Li Zefan 已提交
1989
		retval = cgroup_attach_task(from_cgrp, tsk, false);
1990 1991 1992
		if (retval)
			break;
	}
T
Tejun Heo 已提交
1993
	mutex_unlock(&cgroup_mutex);
1994 1995 1996 1997 1998

	return retval;
}
EXPORT_SYMBOL_GPL(cgroup_attach_task_all);

1999 2000
static int cgroup_tasks_write(struct cgroup_subsys_state *css,
			      struct cftype *cft, u64 pid)
B
Ben Blum 已提交
2001
{
2002
	return attach_task_by_pid(css->cgroup, pid, false);
B
Ben Blum 已提交
2003 2004
}

2005 2006
static int cgroup_procs_write(struct cgroup_subsys_state *css,
			      struct cftype *cft, u64 tgid)
2007
{
2008
	return attach_task_by_pid(css->cgroup, tgid, true);
2009 2010
}

2011 2012
static int cgroup_release_agent_write(struct cgroup_subsys_state *css,
				      struct cftype *cft, const char *buffer)
2013
{
2014 2015 2016
	struct cgroupfs_root *root = css->cgroup->root;

	BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX);
2017
	if (!cgroup_lock_live_group(css->cgroup))
2018
		return -ENODEV;
2019
	spin_lock(&release_agent_path_lock);
2020 2021
	strlcpy(root->release_agent_path, buffer,
		sizeof(root->release_agent_path));
2022
	spin_unlock(&release_agent_path_lock);
T
Tejun Heo 已提交
2023
	mutex_unlock(&cgroup_mutex);
2024 2025 2026
	return 0;
}

2027
static int cgroup_release_agent_show(struct seq_file *seq, void *v)
2028
{
2029
	struct cgroup *cgrp = seq_css(seq)->cgroup;
2030

2031 2032 2033 2034
	if (!cgroup_lock_live_group(cgrp))
		return -ENODEV;
	seq_puts(seq, cgrp->root->release_agent_path);
	seq_putc(seq, '\n');
T
Tejun Heo 已提交
2035
	mutex_unlock(&cgroup_mutex);
2036 2037 2038
	return 0;
}

2039
static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
2040
{
2041 2042 2043
	struct cgroup *cgrp = seq_css(seq)->cgroup;

	seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2044 2045 2046
	return 0;
}

T
Tejun Heo 已提交
2047 2048
static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
				 size_t nbytes, loff_t off)
2049
{
T
Tejun Heo 已提交
2050 2051 2052
	struct cgroup *cgrp = of->kn->parent->priv;
	struct cftype *cft = of->kn->priv;
	struct cgroup_subsys_state *css;
2053
	int ret;
2054

T
Tejun Heo 已提交
2055 2056 2057 2058 2059 2060 2061 2062 2063
	/*
	 * kernfs guarantees that a file isn't deleted with operations in
	 * flight, which means that the matching css is and stays alive and
	 * doesn't need to be pinned.  The RCU locking is not necessary
	 * either.  It's just for the convenience of using cgroup_css().
	 */
	rcu_read_lock();
	css = cgroup_css(cgrp, cft->ss);
	rcu_read_unlock();
2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078

	if (cft->write_string) {
		ret = cft->write_string(css, cft, strstrip(buf));
	} else if (cft->write_u64) {
		unsigned long long v;
		ret = kstrtoull(buf, 0, &v);
		if (!ret)
			ret = cft->write_u64(css, cft, v);
	} else if (cft->write_s64) {
		long long v;
		ret = kstrtoll(buf, 0, &v);
		if (!ret)
			ret = cft->write_s64(css, cft, v);
	} else if (cft->trigger) {
		ret = cft->trigger(css, (unsigned int)cft->private);
2079
	} else {
2080
		ret = -EINVAL;
2081
	}
T
Tejun Heo 已提交
2082

2083
	return ret ?: nbytes;
2084 2085
}

2086
static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
2087
{
T
Tejun Heo 已提交
2088
	return seq_cft(seq)->seq_start(seq, ppos);
2089 2090
}

2091
static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
2092
{
T
Tejun Heo 已提交
2093
	return seq_cft(seq)->seq_next(seq, v, ppos);
2094 2095
}

2096
static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
2097
{
T
Tejun Heo 已提交
2098
	seq_cft(seq)->seq_stop(seq, v);
2099 2100
}

2101
static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2102
{
2103 2104
	struct cftype *cft = seq_cft(m);
	struct cgroup_subsys_state *css = seq_css(m);
2105

2106 2107
	if (cft->seq_show)
		return cft->seq_show(m, arg);
2108

2109
	if (cft->read_u64)
2110 2111 2112 2113 2114 2115
		seq_printf(m, "%llu\n", cft->read_u64(css, cft));
	else if (cft->read_s64)
		seq_printf(m, "%lld\n", cft->read_s64(css, cft));
	else
		return -EINVAL;
	return 0;
2116 2117
}

T
Tejun Heo 已提交
2118 2119 2120 2121
static struct kernfs_ops cgroup_kf_single_ops = {
	.atomic_write_len	= PAGE_SIZE,
	.write			= cgroup_file_write,
	.seq_show		= cgroup_seqfile_show,
2122 2123
};

T
Tejun Heo 已提交
2124 2125 2126 2127 2128 2129 2130 2131
static struct kernfs_ops cgroup_kf_ops = {
	.atomic_write_len	= PAGE_SIZE,
	.write			= cgroup_file_write,
	.seq_start		= cgroup_seqfile_start,
	.seq_next		= cgroup_seqfile_next,
	.seq_stop		= cgroup_seqfile_stop,
	.seq_show		= cgroup_seqfile_show,
};
2132 2133 2134 2135

/*
 * cgroup_rename - Only allow simple rename of directories in place.
 */
T
Tejun Heo 已提交
2136 2137
static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
			 const char *new_name_str)
2138
{
T
Tejun Heo 已提交
2139 2140
	struct cgroup *cgrp = kn->priv;
	int ret;
2141

T
Tejun Heo 已提交
2142
	if (kernfs_type(kn) != KERNFS_DIR)
2143
		return -ENOTDIR;
T
Tejun Heo 已提交
2144
	if (kn->parent != new_parent)
2145
		return -EIO;
2146

2147 2148 2149 2150 2151 2152 2153
	/*
	 * This isn't a proper migration and its usefulness is very
	 * limited.  Disallow if sane_behavior.
	 */
	if (cgroup_sane_behavior(cgrp))
		return -EPERM;

T
Tejun Heo 已提交
2154 2155 2156 2157
	mutex_lock(&cgroup_tree_mutex);
	mutex_lock(&cgroup_mutex);

	ret = kernfs_rename(kn, new_parent, new_name_str);
2158

T
Tejun Heo 已提交
2159 2160 2161
	mutex_unlock(&cgroup_mutex);
	mutex_unlock(&cgroup_tree_mutex);
	return ret;
2162 2163
}

2164
static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft)
2165
{
T
Tejun Heo 已提交
2166
	char name[CGROUP_FILE_NAME_MAX];
T
Tejun Heo 已提交
2167 2168
	struct kernfs_node *kn;
	struct lock_class_key *key = NULL;
T
Tejun Heo 已提交
2169

T
Tejun Heo 已提交
2170 2171 2172 2173 2174 2175
#ifdef CONFIG_DEBUG_LOCK_ALLOC
	key = &cft->lockdep_key;
#endif
	kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
				  cgroup_file_mode(cft), 0, cft->kf_ops, cft,
				  NULL, false, key);
F
Fengguang Wu 已提交
2176
	return PTR_ERR_OR_ZERO(kn);
2177 2178
}

2179 2180 2181 2182 2183 2184 2185
/**
 * cgroup_addrm_files - add or remove files to a cgroup directory
 * @cgrp: the target cgroup
 * @cfts: array of cftypes to be added
 * @is_add: whether to add or remove
 *
 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2186 2187 2188
 * For removals, this function never fails.  If addition fails, this
 * function doesn't remove files already added.  The caller is responsible
 * for cleaning up.
2189
 */
2190 2191
static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
			      bool is_add)
2192
{
A
Aristeu Rozanski 已提交
2193
	struct cftype *cft;
2194 2195
	int ret;

T
Tejun Heo 已提交
2196
	lockdep_assert_held(&cgroup_tree_mutex);
T
Tejun Heo 已提交
2197 2198

	for (cft = cfts; cft->name[0] != '\0'; cft++) {
2199
		/* does cft->flags tell us to skip this file on @cgrp? */
2200 2201
		if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
			continue;
2202 2203 2204 2205 2206
		if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
			continue;
		if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
			continue;

2207
		if (is_add) {
2208
			ret = cgroup_add_file(cgrp, cft);
2209
			if (ret) {
2210
				pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2211 2212 2213
					cft->name, ret);
				return ret;
			}
2214 2215
		} else {
			cgroup_rm_file(cgrp, cft);
T
Tejun Heo 已提交
2216
		}
2217
	}
2218
	return 0;
2219 2220
}

2221
static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
2222 2223
{
	LIST_HEAD(pending);
2224
	struct cgroup_subsys *ss = cfts[0].ss;
2225 2226
	struct cgroup *root = &ss->root->top_cgroup;
	struct cgroup_subsys_state *css;
2227
	int ret = 0;
2228

2229
	lockdep_assert_held(&cgroup_tree_mutex);
2230

2231 2232
	/* don't bother if @ss isn't attached */
	if (ss->root == &cgroup_dummy_root)
2233
		return 0;
2234 2235

	/* add/rm files for all cgroups created before */
2236
	css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
2237 2238
		struct cgroup *cgrp = css->cgroup;

2239 2240 2241
		if (cgroup_is_dead(cgrp))
			continue;

2242
		ret = cgroup_addrm_files(cgrp, cfts, is_add);
2243 2244
		if (ret)
			break;
2245
	}
2246 2247 2248

	if (is_add && !ret)
		kernfs_activate(root->kn);
2249
	return ret;
2250 2251
}

2252 2253 2254 2255
static void cgroup_exit_cftypes(struct cftype *cfts)
{
	struct cftype *cft;

T
Tejun Heo 已提交
2256 2257 2258 2259 2260
	for (cft = cfts; cft->name[0] != '\0'; cft++) {
		/* free copy for custom atomic_write_len, see init_cftypes() */
		if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
			kfree(cft->kf_ops);
		cft->kf_ops = NULL;
2261
		cft->ss = NULL;
T
Tejun Heo 已提交
2262
	}
2263 2264
}

T
Tejun Heo 已提交
2265
static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2266 2267 2268
{
	struct cftype *cft;

T
Tejun Heo 已提交
2269 2270 2271
	for (cft = cfts; cft->name[0] != '\0'; cft++) {
		struct kernfs_ops *kf_ops;

T
Tejun Heo 已提交
2272 2273
		WARN_ON(cft->ss || cft->kf_ops);

T
Tejun Heo 已提交
2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292
		if (cft->seq_start)
			kf_ops = &cgroup_kf_ops;
		else
			kf_ops = &cgroup_kf_single_ops;

		/*
		 * Ugh... if @cft wants a custom max_write_len, we need to
		 * make a copy of kf_ops to set its atomic_write_len.
		 */
		if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
			kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
			if (!kf_ops) {
				cgroup_exit_cftypes(cfts);
				return -ENOMEM;
			}
			kf_ops->atomic_write_len = cft->max_write_len;
		}

		cft->kf_ops = kf_ops;
2293
		cft->ss = ss;
T
Tejun Heo 已提交
2294 2295 2296
	}

	return 0;
2297 2298
}

2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311
static int cgroup_rm_cftypes_locked(struct cftype *cfts)
{
	lockdep_assert_held(&cgroup_tree_mutex);

	if (!cfts || !cfts[0].ss)
		return -ENOENT;

	list_del(&cfts->node);
	cgroup_apply_cftypes(cfts, false);
	cgroup_exit_cftypes(cfts);
	return 0;
}

T
Tejun Heo 已提交
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324
/**
 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
 * @cfts: zero-length name terminated array of cftypes
 *
 * Unregister @cfts.  Files described by @cfts are removed from all
 * existing cgroups and all future cgroups won't have them either.  This
 * function can be called anytime whether @cfts' subsys is attached or not.
 *
 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
 * registered.
 */
int cgroup_rm_cftypes(struct cftype *cfts)
{
2325
	int ret;
T
Tejun Heo 已提交
2326

2327 2328 2329 2330
	mutex_lock(&cgroup_tree_mutex);
	ret = cgroup_rm_cftypes_locked(cfts);
	mutex_unlock(&cgroup_tree_mutex);
	return ret;
T
Tejun Heo 已提交
2331 2332
}

2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346
/**
 * cgroup_add_cftypes - add an array of cftypes to a subsystem
 * @ss: target cgroup subsystem
 * @cfts: zero-length name terminated array of cftypes
 *
 * Register @cfts to @ss.  Files described by @cfts are created for all
 * existing cgroups to which @ss is attached and all future cgroups will
 * have them too.  This function can be called anytime whether @ss is
 * attached or not.
 *
 * Returns 0 on successful registration, -errno on failure.  Note that this
 * function currently returns 0 as long as @cfts registration is successful
 * even if some file creation attempts on existing cgroups fail.
 */
A
Aristeu Rozanski 已提交
2347
int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2348
{
2349
	int ret;
2350

T
Tejun Heo 已提交
2351 2352 2353
	ret = cgroup_init_cftypes(ss, cfts);
	if (ret)
		return ret;
2354

2355 2356
	mutex_lock(&cgroup_tree_mutex);

T
Tejun Heo 已提交
2357
	list_add_tail(&cfts->node, &ss->cfts);
2358
	ret = cgroup_apply_cftypes(cfts, true);
2359
	if (ret)
2360 2361 2362
		cgroup_rm_cftypes_locked(cfts);

	mutex_unlock(&cgroup_tree_mutex);
2363
	return ret;
2364 2365
}

L
Li Zefan 已提交
2366 2367 2368 2369 2370 2371
/**
 * cgroup_task_count - count the number of tasks in a cgroup.
 * @cgrp: the cgroup in question
 *
 * Return the number of tasks in the cgroup.
 */
2372
static int cgroup_task_count(const struct cgroup *cgrp)
2373 2374
{
	int count = 0;
2375
	struct cgrp_cset_link *link;
2376

2377
	down_read(&css_set_rwsem);
2378 2379
	list_for_each_entry(link, &cgrp->cset_links, cset_link)
		count += atomic_read(&link->cset->refcount);
2380
	up_read(&css_set_rwsem);
2381 2382 2383
	return count;
}

2384
/**
2385 2386 2387
 * css_next_child - find the next child of a given css
 * @pos_css: the current position (%NULL to initiate traversal)
 * @parent_css: css whose children to walk
2388
 *
2389
 * This function returns the next child of @parent_css and should be called
2390 2391 2392
 * under either cgroup_mutex or RCU read lock.  The only requirement is
 * that @parent_css and @pos_css are accessible.  The next sibling is
 * guaranteed to be returned regardless of their states.
2393
 */
2394 2395 2396
struct cgroup_subsys_state *
css_next_child(struct cgroup_subsys_state *pos_css,
	       struct cgroup_subsys_state *parent_css)
2397
{
2398 2399
	struct cgroup *pos = pos_css ? pos_css->cgroup : NULL;
	struct cgroup *cgrp = parent_css->cgroup;
2400 2401
	struct cgroup *next;

T
Tejun Heo 已提交
2402
	cgroup_assert_mutexes_or_rcu_locked();
2403 2404 2405 2406

	/*
	 * @pos could already have been removed.  Once a cgroup is removed,
	 * its ->sibling.next is no longer updated when its next sibling
2407 2408 2409 2410 2411 2412 2413
	 * changes.  As CGRP_DEAD assertion is serialized and happens
	 * before the cgroup is taken off the ->sibling list, if we see it
	 * unasserted, it's guaranteed that the next sibling hasn't
	 * finished its grace period even if it's already removed, and thus
	 * safe to dereference from this RCU critical section.  If
	 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
	 * to be visible as %true here.
2414 2415 2416 2417 2418 2419 2420 2421
	 *
	 * If @pos is dead, its next pointer can't be dereferenced;
	 * however, as each cgroup is given a monotonically increasing
	 * unique serial number and always appended to the sibling list,
	 * the next one can be found by walking the parent's children until
	 * we see a cgroup with higher serial number than @pos's.  While
	 * this path can be slower, it's taken only when either the current
	 * cgroup is removed or iteration and removal race.
2422
	 */
2423 2424 2425
	if (!pos) {
		next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling);
	} else if (likely(!cgroup_is_dead(pos))) {
2426
		next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
2427 2428 2429 2430
	} else {
		list_for_each_entry_rcu(next, &cgrp->children, sibling)
			if (next->serial_nr > pos->serial_nr)
				break;
2431 2432
	}

2433 2434 2435
	if (&next->sibling == &cgrp->children)
		return NULL;

2436
	return cgroup_css(next, parent_css->ss);
2437 2438
}

2439
/**
2440
 * css_next_descendant_pre - find the next descendant for pre-order walk
2441
 * @pos: the current position (%NULL to initiate traversal)
2442
 * @root: css whose descendants to walk
2443
 *
2444
 * To be used by css_for_each_descendant_pre().  Find the next descendant
2445 2446
 * to visit for pre-order traversal of @root's descendants.  @root is
 * included in the iteration and the first node to be visited.
2447
 *
2448 2449 2450 2451
 * While this function requires cgroup_mutex or RCU read locking, it
 * doesn't require the whole traversal to be contained in a single critical
 * section.  This function will return the correct next descendant as long
 * as both @pos and @root are accessible and @pos is a descendant of @root.
2452
 */
2453 2454 2455
struct cgroup_subsys_state *
css_next_descendant_pre(struct cgroup_subsys_state *pos,
			struct cgroup_subsys_state *root)
2456
{
2457
	struct cgroup_subsys_state *next;
2458

T
Tejun Heo 已提交
2459
	cgroup_assert_mutexes_or_rcu_locked();
2460

2461
	/* if first iteration, visit @root */
2462
	if (!pos)
2463
		return root;
2464 2465

	/* visit the first child if exists */
2466
	next = css_next_child(NULL, pos);
2467 2468 2469 2470
	if (next)
		return next;

	/* no child, visit my or the closest ancestor's next sibling */
2471 2472
	while (pos != root) {
		next = css_next_child(pos, css_parent(pos));
2473
		if (next)
2474
			return next;
2475
		pos = css_parent(pos);
2476
	}
2477 2478 2479 2480

	return NULL;
}

2481
/**
2482 2483
 * css_rightmost_descendant - return the rightmost descendant of a css
 * @pos: css of interest
2484
 *
2485 2486
 * Return the rightmost descendant of @pos.  If there's no descendant, @pos
 * is returned.  This can be used during pre-order traversal to skip
2487
 * subtree of @pos.
2488
 *
2489 2490 2491 2492
 * While this function requires cgroup_mutex or RCU read locking, it
 * doesn't require the whole traversal to be contained in a single critical
 * section.  This function will return the correct rightmost descendant as
 * long as @pos is accessible.
2493
 */
2494 2495
struct cgroup_subsys_state *
css_rightmost_descendant(struct cgroup_subsys_state *pos)
2496
{
2497
	struct cgroup_subsys_state *last, *tmp;
2498

T
Tejun Heo 已提交
2499
	cgroup_assert_mutexes_or_rcu_locked();
2500 2501 2502 2503 2504

	do {
		last = pos;
		/* ->prev isn't RCU safe, walk ->next till the end */
		pos = NULL;
2505
		css_for_each_child(tmp, last)
2506 2507 2508 2509 2510 2511
			pos = tmp;
	} while (pos);

	return last;
}

2512 2513
static struct cgroup_subsys_state *
css_leftmost_descendant(struct cgroup_subsys_state *pos)
2514
{
2515
	struct cgroup_subsys_state *last;
2516 2517 2518

	do {
		last = pos;
2519
		pos = css_next_child(NULL, pos);
2520 2521 2522 2523 2524 2525
	} while (pos);

	return last;
}

/**
2526
 * css_next_descendant_post - find the next descendant for post-order walk
2527
 * @pos: the current position (%NULL to initiate traversal)
2528
 * @root: css whose descendants to walk
2529
 *
2530
 * To be used by css_for_each_descendant_post().  Find the next descendant
2531 2532
 * to visit for post-order traversal of @root's descendants.  @root is
 * included in the iteration and the last node to be visited.
2533
 *
2534 2535 2536 2537 2538
 * While this function requires cgroup_mutex or RCU read locking, it
 * doesn't require the whole traversal to be contained in a single critical
 * section.  This function will return the correct next descendant as long
 * as both @pos and @cgroup are accessible and @pos is a descendant of
 * @cgroup.
2539
 */
2540 2541 2542
struct cgroup_subsys_state *
css_next_descendant_post(struct cgroup_subsys_state *pos,
			 struct cgroup_subsys_state *root)
2543
{
2544
	struct cgroup_subsys_state *next;
2545

T
Tejun Heo 已提交
2546
	cgroup_assert_mutexes_or_rcu_locked();
2547

2548 2549 2550
	/* if first iteration, visit leftmost descendant which may be @root */
	if (!pos)
		return css_leftmost_descendant(root);
2551

2552 2553 2554 2555
	/* if we visited @root, we're done */
	if (pos == root)
		return NULL;

2556
	/* if there's an unvisited sibling, visit its leftmost descendant */
2557
	next = css_next_child(pos, css_parent(pos));
2558
	if (next)
2559
		return css_leftmost_descendant(next);
2560 2561

	/* no sibling left, visit parent */
2562
	return css_parent(pos);
2563 2564
}

2565
/**
2566
 * css_advance_task_iter - advance a task itererator to the next css_set
2567 2568 2569
 * @it: the iterator to advance
 *
 * Advance @it to the next css_set to walk.
2570
 */
2571
static void css_advance_task_iter(struct css_task_iter *it)
2572 2573 2574 2575 2576 2577 2578 2579
{
	struct list_head *l = it->cset_link;
	struct cgrp_cset_link *link;
	struct css_set *cset;

	/* Advance to the next non-empty css_set */
	do {
		l = l->next;
2580
		if (l == &it->origin_css->cgroup->cset_links) {
2581 2582 2583 2584 2585 2586 2587 2588 2589 2590
			it->cset_link = NULL;
			return;
		}
		link = list_entry(l, struct cgrp_cset_link, cset_link);
		cset = link->cset;
	} while (list_empty(&cset->tasks));
	it->cset_link = l;
	it->task = cset->tasks.next;
}

2591
/**
2592 2593
 * css_task_iter_start - initiate task iteration
 * @css: the css to walk tasks of
2594 2595
 * @it: the task iterator to use
 *
2596 2597 2598 2599
 * Initiate iteration through the tasks of @css.  The caller can call
 * css_task_iter_next() to walk through the tasks until the function
 * returns NULL.  On completion of iteration, css_task_iter_end() must be
 * called.
2600 2601 2602 2603 2604
 *
 * Note that this function acquires a lock which is released when the
 * iteration finishes.  The caller can't sleep while iteration is in
 * progress.
 */
2605 2606
void css_task_iter_start(struct cgroup_subsys_state *css,
			 struct css_task_iter *it)
2607
	__acquires(css_set_rwsem)
2608
{
2609 2610
	/* no one should try to iterate before mounting cgroups */
	WARN_ON_ONCE(!use_task_css_set_links);
2611

2612
	down_read(&css_set_rwsem);
2613

2614 2615
	it->origin_css = css;
	it->cset_link = &css->cgroup->cset_links;
2616

2617
	css_advance_task_iter(it);
2618 2619
}

2620
/**
2621
 * css_task_iter_next - return the next task for the iterator
2622 2623 2624
 * @it: the task iterator being iterated
 *
 * The "next" function for task iteration.  @it should have been
2625 2626
 * initialized via css_task_iter_start().  Returns NULL when the iteration
 * reaches the end.
2627
 */
2628
struct task_struct *css_task_iter_next(struct css_task_iter *it)
2629 2630 2631
{
	struct task_struct *res;
	struct list_head *l = it->task;
2632
	struct cgrp_cset_link *link;
2633 2634

	/* If the iterator cg is NULL, we have no tasks */
2635
	if (!it->cset_link)
2636 2637 2638 2639
		return NULL;
	res = list_entry(l, struct task_struct, cg_list);
	/* Advance iterator to find next entry */
	l = l->next;
2640 2641
	link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
	if (l == &link->cset->tasks) {
2642 2643 2644 2645
		/*
		 * We reached the end of this task list - move on to the
		 * next cgrp_cset_link.
		 */
2646
		css_advance_task_iter(it);
2647 2648 2649 2650 2651 2652
	} else {
		it->task = l;
	}
	return res;
}

2653
/**
2654
 * css_task_iter_end - finish task iteration
2655 2656
 * @it: the task iterator to finish
 *
2657
 * Finish task iteration started by css_task_iter_start().
2658
 */
2659
void css_task_iter_end(struct css_task_iter *it)
2660
	__releases(css_set_rwsem)
2661
{
2662
	up_read(&css_set_rwsem);
2663 2664
}

2665 2666 2667 2668 2669 2670 2671
/**
 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
 * @to: cgroup to which the tasks will be moved
 * @from: cgroup in which the tasks currently reside
 */
int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
{
2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
	struct css_task_iter it;
	struct task_struct *task;
	int ret = 0;

	do {
		css_task_iter_start(&from->dummy_css, &it);
		task = css_task_iter_next(&it);
		if (task)
			get_task_struct(task);
		css_task_iter_end(&it);

		if (task) {
			mutex_lock(&cgroup_mutex);
			ret = cgroup_attach_task(to, task, false);
			mutex_unlock(&cgroup_mutex);
			put_task_struct(task);
		}
	} while (task && !ret);

	return ret;
2692 2693
}

2694
/*
2695
 * Stuff for reading the 'tasks'/'procs' files.
2696 2697 2698 2699 2700 2701 2702 2703
 *
 * Reading this file can return large amounts of data if a cgroup has
 * *lots* of attached tasks. So it may need several calls to read(),
 * but we cannot guarantee that the information we produce is correct
 * unless we produce it entirely atomically.
 *
 */

2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729
/* which pidlist file are we talking about? */
enum cgroup_filetype {
	CGROUP_FILE_PROCS,
	CGROUP_FILE_TASKS,
};

/*
 * A pidlist is a list of pids that virtually represents the contents of one
 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
 * a pair (one each for procs, tasks) for each pid namespace that's relevant
 * to the cgroup.
 */
struct cgroup_pidlist {
	/*
	 * used to find which pidlist is wanted. doesn't change as long as
	 * this particular list stays in the list.
	*/
	struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
	/* array of xids */
	pid_t *list;
	/* how many elements the above list has */
	int length;
	/* each of these stored in a list by its cgroup */
	struct list_head links;
	/* pointer to the cgroup we belong to, for list removal purposes */
	struct cgroup *owner;
2730 2731
	/* for delayed destruction */
	struct delayed_work destroy_dwork;
2732 2733
};

2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746
/*
 * The following two functions "fix" the issue where there are more pids
 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
 * TODO: replace with a kernel-wide solution to this problem
 */
#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
static void *pidlist_allocate(int count)
{
	if (PIDLIST_TOO_LARGE(count))
		return vmalloc(count * sizeof(pid_t));
	else
		return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
}
2747

2748 2749 2750 2751 2752 2753 2754 2755
static void pidlist_free(void *p)
{
	if (is_vmalloc_addr(p))
		vfree(p);
	else
		kfree(p);
}

2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782
/*
 * Used to destroy all pidlists lingering waiting for destroy timer.  None
 * should be left afterwards.
 */
static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
{
	struct cgroup_pidlist *l, *tmp_l;

	mutex_lock(&cgrp->pidlist_mutex);
	list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
		mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
	mutex_unlock(&cgrp->pidlist_mutex);

	flush_workqueue(cgroup_pidlist_destroy_wq);
	BUG_ON(!list_empty(&cgrp->pidlists));
}

static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
						destroy_dwork);
	struct cgroup_pidlist *tofree = NULL;

	mutex_lock(&l->owner->pidlist_mutex);

	/*
2783 2784
	 * Destroy iff we didn't get queued again.  The state won't change
	 * as destroy_dwork can only be queued while locked.
2785
	 */
2786
	if (!delayed_work_pending(dwork)) {
2787 2788 2789 2790 2791 2792 2793 2794 2795 2796
		list_del(&l->links);
		pidlist_free(l->list);
		put_pid_ns(l->key.ns);
		tofree = l;
	}

	mutex_unlock(&l->owner->pidlist_mutex);
	kfree(tofree);
}

2797
/*
2798
 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2799
 * Returns the number of unique elements.
2800
 */
2801
static int pidlist_uniq(pid_t *list, int length)
2802
{
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826
	int src, dest = 1;

	/*
	 * we presume the 0th element is unique, so i starts at 1. trivial
	 * edge cases first; no work needs to be done for either
	 */
	if (length == 0 || length == 1)
		return length;
	/* src and dest walk down the list; dest counts unique elements */
	for (src = 1; src < length; src++) {
		/* find next unique element */
		while (list[src] == list[src-1]) {
			src++;
			if (src == length)
				goto after;
		}
		/* dest always points to where the next unique element goes */
		list[dest] = list[src];
		dest++;
	}
after:
	return dest;
}

2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859
/*
 * The two pid files - task and cgroup.procs - guaranteed that the result
 * is sorted, which forced this whole pidlist fiasco.  As pid order is
 * different per namespace, each namespace needs differently sorted list,
 * making it impossible to use, for example, single rbtree of member tasks
 * sorted by task pointer.  As pidlists can be fairly large, allocating one
 * per open file is dangerous, so cgroup had to implement shared pool of
 * pidlists keyed by cgroup and namespace.
 *
 * All this extra complexity was caused by the original implementation
 * committing to an entirely unnecessary property.  In the long term, we
 * want to do away with it.  Explicitly scramble sort order if
 * sane_behavior so that no such expectation exists in the new interface.
 *
 * Scrambling is done by swapping every two consecutive bits, which is
 * non-identity one-to-one mapping which disturbs sort order sufficiently.
 */
static pid_t pid_fry(pid_t pid)
{
	unsigned a = pid & 0x55555555;
	unsigned b = pid & 0xAAAAAAAA;

	return (a << 1) | (b >> 1);
}

static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
{
	if (cgroup_sane_behavior(cgrp))
		return pid_fry(pid);
	else
		return pid;
}

2860 2861 2862 2863 2864
static int cmppid(const void *a, const void *b)
{
	return *(pid_t *)a - *(pid_t *)b;
}

2865 2866 2867 2868 2869
static int fried_cmppid(const void *a, const void *b)
{
	return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
}

T
Tejun Heo 已提交
2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884
static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
						  enum cgroup_filetype type)
{
	struct cgroup_pidlist *l;
	/* don't need task_nsproxy() if we're looking at ourself */
	struct pid_namespace *ns = task_active_pid_ns(current);

	lockdep_assert_held(&cgrp->pidlist_mutex);

	list_for_each_entry(l, &cgrp->pidlists, links)
		if (l->key.type == type && l->key.ns == ns)
			return l;
	return NULL;
}

2885 2886 2887 2888 2889 2890
/*
 * find the appropriate pidlist for our purpose (given procs vs tasks)
 * returns with the lock on that pidlist already held, and takes care
 * of the use count, or returns NULL with no locks held if we're out of
 * memory.
 */
T
Tejun Heo 已提交
2891 2892
static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
						enum cgroup_filetype type)
2893 2894
{
	struct cgroup_pidlist *l;
2895

T
Tejun Heo 已提交
2896 2897 2898 2899 2900 2901
	lockdep_assert_held(&cgrp->pidlist_mutex);

	l = cgroup_pidlist_find(cgrp, type);
	if (l)
		return l;

2902
	/* entry not found; create a new one */
2903
	l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
T
Tejun Heo 已提交
2904
	if (!l)
2905
		return l;
T
Tejun Heo 已提交
2906

2907
	INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
2908
	l->key.type = type;
T
Tejun Heo 已提交
2909 2910
	/* don't need task_nsproxy() if we're looking at ourself */
	l->key.ns = get_pid_ns(task_active_pid_ns(current));
2911 2912 2913 2914 2915
	l->owner = cgrp;
	list_add(&l->links, &cgrp->pidlists);
	return l;
}

2916 2917 2918
/*
 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
 */
2919 2920
static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
			      struct cgroup_pidlist **lp)
2921 2922 2923 2924
{
	pid_t *array;
	int length;
	int pid, n = 0; /* used for populating the array */
2925
	struct css_task_iter it;
2926
	struct task_struct *tsk;
2927 2928
	struct cgroup_pidlist *l;

2929 2930
	lockdep_assert_held(&cgrp->pidlist_mutex);

2931 2932 2933 2934 2935 2936 2937
	/*
	 * If cgroup gets more users after we read count, we won't have
	 * enough space - tough.  This race is indistinguishable to the
	 * caller from the case that the additional cgroup users didn't
	 * show up until sometime later on.
	 */
	length = cgroup_task_count(cgrp);
2938
	array = pidlist_allocate(length);
2939 2940 2941
	if (!array)
		return -ENOMEM;
	/* now, populate the array */
2942 2943
	css_task_iter_start(&cgrp->dummy_css, &it);
	while ((tsk = css_task_iter_next(&it))) {
2944
		if (unlikely(n == length))
2945
			break;
2946
		/* get tgid or pid for procs or tasks file respectively */
2947 2948 2949 2950
		if (type == CGROUP_FILE_PROCS)
			pid = task_tgid_vnr(tsk);
		else
			pid = task_pid_vnr(tsk);
2951 2952
		if (pid > 0) /* make sure to only use valid results */
			array[n++] = pid;
2953
	}
2954
	css_task_iter_end(&it);
2955 2956
	length = n;
	/* now sort & (if procs) strip out duplicates */
2957 2958 2959 2960
	if (cgroup_sane_behavior(cgrp))
		sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
	else
		sort(array, length, sizeof(pid_t), cmppid, NULL);
2961
	if (type == CGROUP_FILE_PROCS)
2962
		length = pidlist_uniq(array, length);
T
Tejun Heo 已提交
2963 2964

	l = cgroup_pidlist_find_create(cgrp, type);
2965
	if (!l) {
T
Tejun Heo 已提交
2966
		mutex_unlock(&cgrp->pidlist_mutex);
2967
		pidlist_free(array);
2968
		return -ENOMEM;
2969
	}
T
Tejun Heo 已提交
2970 2971

	/* store array, freeing old if necessary */
2972
	pidlist_free(l->list);
2973 2974
	l->list = array;
	l->length = length;
2975
	*lp = l;
2976
	return 0;
2977 2978
}

B
Balbir Singh 已提交
2979
/**
L
Li Zefan 已提交
2980
 * cgroupstats_build - build and fill cgroupstats
B
Balbir Singh 已提交
2981 2982 2983
 * @stats: cgroupstats to fill information into
 * @dentry: A dentry entry belonging to the cgroup for which stats have
 * been requested.
L
Li Zefan 已提交
2984 2985 2986
 *
 * Build and fill cgroupstats so that taskstats can export it to user
 * space.
B
Balbir Singh 已提交
2987 2988 2989
 */
int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
{
T
Tejun Heo 已提交
2990
	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
2991
	struct cgroup *cgrp;
2992
	struct css_task_iter it;
B
Balbir Singh 已提交
2993
	struct task_struct *tsk;
2994

T
Tejun Heo 已提交
2995 2996 2997 2998 2999
	/* it should be kernfs_node belonging to cgroupfs and is a directory */
	if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
	    kernfs_type(kn) != KERNFS_DIR)
		return -EINVAL;

3000 3001
	mutex_lock(&cgroup_mutex);

B
Balbir Singh 已提交
3002
	/*
T
Tejun Heo 已提交
3003 3004 3005
	 * We aren't being called from kernfs and there's no guarantee on
	 * @kn->priv's validity.  For this and css_tryget_from_dir(),
	 * @kn->priv is RCU safe.  Let's do the RCU dancing.
B
Balbir Singh 已提交
3006
	 */
T
Tejun Heo 已提交
3007 3008
	rcu_read_lock();
	cgrp = rcu_dereference(kn->priv);
3009
	if (!cgrp || cgroup_is_dead(cgrp)) {
T
Tejun Heo 已提交
3010
		rcu_read_unlock();
3011
		mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3012 3013
		return -ENOENT;
	}
3014
	rcu_read_unlock();
B
Balbir Singh 已提交
3015

3016 3017
	css_task_iter_start(&cgrp->dummy_css, &it);
	while ((tsk = css_task_iter_next(&it))) {
B
Balbir Singh 已提交
3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036
		switch (tsk->state) {
		case TASK_RUNNING:
			stats->nr_running++;
			break;
		case TASK_INTERRUPTIBLE:
			stats->nr_sleeping++;
			break;
		case TASK_UNINTERRUPTIBLE:
			stats->nr_uninterruptible++;
			break;
		case TASK_STOPPED:
			stats->nr_stopped++;
			break;
		default:
			if (delayacct_is_task_waiting_on_io(tsk))
				stats->nr_io_wait++;
			break;
		}
	}
3037
	css_task_iter_end(&it);
B
Balbir Singh 已提交
3038

3039
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3040
	return 0;
B
Balbir Singh 已提交
3041 3042
}

3043

3044
/*
3045
 * seq_file methods for the tasks/procs files. The seq_file position is the
3046
 * next pid to display; the seq_file iterator is a pointer to the pid
3047
 * in the cgroup->l->list array.
3048
 */
3049

3050
static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3051
{
3052 3053 3054 3055 3056 3057
	/*
	 * Initially we receive a position value that corresponds to
	 * one more than the last pid shown (or 0 on the first call or
	 * after a seek to the start). Use a binary-search to find the
	 * next pid to display, if any
	 */
T
Tejun Heo 已提交
3058
	struct kernfs_open_file *of = s->private;
3059
	struct cgroup *cgrp = seq_css(s)->cgroup;
3060
	struct cgroup_pidlist *l;
3061
	enum cgroup_filetype type = seq_cft(s)->private;
3062
	int index = 0, pid = *pos;
3063 3064 3065 3066 3067
	int *iter, ret;

	mutex_lock(&cgrp->pidlist_mutex);

	/*
3068
	 * !NULL @of->priv indicates that this isn't the first start()
3069
	 * after open.  If the matching pidlist is around, we can use that.
3070
	 * Look for it.  Note that @of->priv can't be used directly.  It
3071 3072
	 * could already have been destroyed.
	 */
3073 3074
	if (of->priv)
		of->priv = cgroup_pidlist_find(cgrp, type);
3075 3076 3077 3078 3079

	/*
	 * Either this is the first start() after open or the matching
	 * pidlist has been destroyed inbetween.  Create a new one.
	 */
3080 3081 3082
	if (!of->priv) {
		ret = pidlist_array_load(cgrp, type,
					 (struct cgroup_pidlist **)&of->priv);
3083 3084 3085
		if (ret)
			return ERR_PTR(ret);
	}
3086
	l = of->priv;
3087 3088

	if (pid) {
3089
		int end = l->length;
S
Stephen Rothwell 已提交
3090

3091 3092
		while (index < end) {
			int mid = (index + end) / 2;
3093
			if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
3094 3095
				index = mid;
				break;
3096
			} else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
3097 3098 3099 3100 3101 3102
				index = mid + 1;
			else
				end = mid;
		}
	}
	/* If we're off the end of the array, we're done */
3103
	if (index >= l->length)
3104 3105
		return NULL;
	/* Update the abstract position to be the actual pid that we found */
3106
	iter = l->list + index;
3107
	*pos = cgroup_pid_fry(cgrp, *iter);
3108 3109 3110
	return iter;
}

3111
static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3112
{
T
Tejun Heo 已提交
3113
	struct kernfs_open_file *of = s->private;
3114
	struct cgroup_pidlist *l = of->priv;
3115

3116 3117
	if (l)
		mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
3118
				 CGROUP_PIDLIST_DESTROY_DELAY);
3119
	mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
3120 3121
}

3122
static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3123
{
T
Tejun Heo 已提交
3124
	struct kernfs_open_file *of = s->private;
3125
	struct cgroup_pidlist *l = of->priv;
3126 3127
	pid_t *p = v;
	pid_t *end = l->list + l->length;
3128 3129 3130 3131 3132 3133 3134 3135
	/*
	 * Advance to the next pid in the array. If this goes off the
	 * end, we're done
	 */
	p++;
	if (p >= end) {
		return NULL;
	} else {
3136
		*pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
3137 3138 3139 3140
		return p;
	}
}

3141
static int cgroup_pidlist_show(struct seq_file *s, void *v)
3142 3143 3144
{
	return seq_printf(s, "%d\n", *(int *)v);
}
3145

3146 3147 3148 3149 3150 3151 3152 3153 3154
/*
 * seq_operations functions for iterating on pidlists through seq_file -
 * independent of whether it's tasks or procs
 */
static const struct seq_operations cgroup_pidlist_seq_operations = {
	.start = cgroup_pidlist_start,
	.stop = cgroup_pidlist_stop,
	.next = cgroup_pidlist_next,
	.show = cgroup_pidlist_show,
3155 3156
};

3157 3158
static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
					 struct cftype *cft)
3159
{
3160
	return notify_on_release(css->cgroup);
3161 3162
}

3163 3164
static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
					  struct cftype *cft, u64 val)
3165
{
3166
	clear_bit(CGRP_RELEASABLE, &css->cgroup->flags);
3167
	if (val)
3168
		set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3169
	else
3170
		clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3171 3172 3173
	return 0;
}

3174 3175
static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
				      struct cftype *cft)
3176
{
3177
	return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3178 3179
}

3180 3181
static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
				       struct cftype *cft, u64 val)
3182 3183
{
	if (val)
3184
		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3185
	else
3186
		clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3187 3188 3189
	return 0;
}

3190
static struct cftype cgroup_base_files[] = {
3191
	{
3192
		.name = "cgroup.procs",
3193 3194 3195 3196
		.seq_start = cgroup_pidlist_start,
		.seq_next = cgroup_pidlist_next,
		.seq_stop = cgroup_pidlist_stop,
		.seq_show = cgroup_pidlist_show,
3197
		.private = CGROUP_FILE_PROCS,
B
Ben Blum 已提交
3198 3199
		.write_u64 = cgroup_procs_write,
		.mode = S_IRUGO | S_IWUSR,
3200
	},
3201 3202
	{
		.name = "cgroup.clone_children",
3203
		.flags = CFTYPE_INSANE,
3204 3205 3206
		.read_u64 = cgroup_clone_children_read,
		.write_u64 = cgroup_clone_children_write,
	},
3207 3208 3209
	{
		.name = "cgroup.sane_behavior",
		.flags = CFTYPE_ONLY_ON_ROOT,
3210
		.seq_show = cgroup_sane_behavior_show,
3211
	},
3212 3213 3214 3215 3216 3217 3218 3219 3220

	/*
	 * Historical crazy stuff.  These don't have "cgroup."  prefix and
	 * don't exist if sane_behavior.  If you're depending on these, be
	 * prepared to be burned.
	 */
	{
		.name = "tasks",
		.flags = CFTYPE_INSANE,		/* use "procs" instead */
3221 3222 3223 3224
		.seq_start = cgroup_pidlist_start,
		.seq_next = cgroup_pidlist_next,
		.seq_stop = cgroup_pidlist_stop,
		.seq_show = cgroup_pidlist_show,
3225
		.private = CGROUP_FILE_TASKS,
3226 3227 3228 3229 3230 3231 3232 3233 3234
		.write_u64 = cgroup_tasks_write,
		.mode = S_IRUGO | S_IWUSR,
	},
	{
		.name = "notify_on_release",
		.flags = CFTYPE_INSANE,
		.read_u64 = cgroup_read_notify_on_release,
		.write_u64 = cgroup_write_notify_on_release,
	},
3235 3236
	{
		.name = "release_agent",
3237
		.flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
3238
		.seq_show = cgroup_release_agent_show,
3239
		.write_string = cgroup_release_agent_write,
3240
		.max_write_len = PATH_MAX - 1,
3241
	},
T
Tejun Heo 已提交
3242
	{ }	/* terminate */
3243 3244
};

3245
/**
3246
 * cgroup_populate_dir - create subsys files in a cgroup directory
3247 3248
 * @cgrp: target cgroup
 * @subsys_mask: mask of the subsystem ids whose files should be added
3249 3250
 *
 * On failure, no file is added.
3251
 */
3252
static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask)
3253 3254
{
	struct cgroup_subsys *ss;
3255
	int i, ret = 0;
3256

3257
	/* process cftsets of each subsystem */
3258
	for_each_subsys(ss, i) {
T
Tejun Heo 已提交
3259
		struct cftype *cfts;
3260 3261

		if (!test_bit(i, &subsys_mask))
3262
			continue;
3263

T
Tejun Heo 已提交
3264 3265
		list_for_each_entry(cfts, &ss->cfts, node) {
			ret = cgroup_addrm_files(cgrp, cfts, true);
3266 3267 3268
			if (ret < 0)
				goto err;
		}
3269 3270
	}
	return 0;
3271 3272 3273
err:
	cgroup_clear_dir(cgrp, subsys_mask);
	return ret;
3274 3275
}

3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297
/*
 * css destruction is four-stage process.
 *
 * 1. Destruction starts.  Killing of the percpu_ref is initiated.
 *    Implemented in kill_css().
 *
 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
 *    and thus css_tryget() is guaranteed to fail, the css can be offlined
 *    by invoking offline_css().  After offlining, the base ref is put.
 *    Implemented in css_killed_work_fn().
 *
 * 3. When the percpu_ref reaches zero, the only possible remaining
 *    accessors are inside RCU read sections.  css_release() schedules the
 *    RCU callback.
 *
 * 4. After the grace period, the css can be freed.  Implemented in
 *    css_free_work_fn().
 *
 * It is actually hairier because both step 2 and 4 require process context
 * and thus involve punting to css->destroy_work adding two additional
 * steps to the already complex sequence.
 */
3298
static void css_free_work_fn(struct work_struct *work)
3299 3300
{
	struct cgroup_subsys_state *css =
3301
		container_of(work, struct cgroup_subsys_state, destroy_work);
3302
	struct cgroup *cgrp = css->cgroup;
3303

3304 3305 3306
	if (css->parent)
		css_put(css->parent);

3307
	css->ss->css_free(css);
T
Tejun Heo 已提交
3308
	cgroup_put(cgrp);
3309 3310
}

3311
static void css_free_rcu_fn(struct rcu_head *rcu_head)
3312 3313
{
	struct cgroup_subsys_state *css =
3314
		container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
3315

3316
	INIT_WORK(&css->destroy_work, css_free_work_fn);
3317
	queue_work(cgroup_destroy_wq, &css->destroy_work);
3318 3319
}

3320 3321 3322 3323 3324
static void css_release(struct percpu_ref *ref)
{
	struct cgroup_subsys_state *css =
		container_of(ref, struct cgroup_subsys_state, refcnt);

3325
	rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL);
3326
	call_rcu(&css->rcu_head, css_free_rcu_fn);
3327 3328
}

3329 3330
static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss,
		     struct cgroup *cgrp)
3331
{
3332
	css->cgroup = cgrp;
3333
	css->ss = ss;
3334
	css->flags = 0;
3335 3336

	if (cgrp->parent)
3337
		css->parent = cgroup_css(cgrp->parent, ss);
3338
	else
3339
		css->flags |= CSS_ROOT;
3340

3341
	BUG_ON(cgroup_css(cgrp, ss));
3342 3343
}

3344
/* invoke ->css_online() on a new CSS and mark it online if successful */
3345
static int online_css(struct cgroup_subsys_state *css)
3346
{
3347
	struct cgroup_subsys *ss = css->ss;
T
Tejun Heo 已提交
3348 3349
	int ret = 0;

T
Tejun Heo 已提交
3350
	lockdep_assert_held(&cgroup_tree_mutex);
3351 3352
	lockdep_assert_held(&cgroup_mutex);

3353
	if (ss->css_online)
3354
		ret = ss->css_online(css);
3355
	if (!ret) {
3356
		css->flags |= CSS_ONLINE;
3357
		css->cgroup->nr_css++;
3358
		rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
3359
	}
T
Tejun Heo 已提交
3360
	return ret;
3361 3362
}

3363
/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
3364
static void offline_css(struct cgroup_subsys_state *css)
3365
{
3366
	struct cgroup_subsys *ss = css->ss;
3367

T
Tejun Heo 已提交
3368
	lockdep_assert_held(&cgroup_tree_mutex);
3369 3370 3371 3372 3373
	lockdep_assert_held(&cgroup_mutex);

	if (!(css->flags & CSS_ONLINE))
		return;

3374
	if (ss->css_offline)
3375
		ss->css_offline(css);
3376

3377
	css->flags &= ~CSS_ONLINE;
3378
	css->cgroup->nr_css--;
3379
	RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css);
3380 3381
}

3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408
/**
 * create_css - create a cgroup_subsys_state
 * @cgrp: the cgroup new css will be associated with
 * @ss: the subsys of new css
 *
 * Create a new css associated with @cgrp - @ss pair.  On success, the new
 * css is online and installed in @cgrp with all interface files created.
 * Returns 0 on success, -errno on failure.
 */
static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss)
{
	struct cgroup *parent = cgrp->parent;
	struct cgroup_subsys_state *css;
	int err;

	lockdep_assert_held(&cgroup_mutex);

	css = ss->css_alloc(cgroup_css(parent, ss));
	if (IS_ERR(css))
		return PTR_ERR(css);

	err = percpu_ref_init(&css->refcnt, css_release);
	if (err)
		goto err_free;

	init_css(css, ss, cgrp);

3409
	err = cgroup_populate_dir(cgrp, 1 << ss->id);
3410 3411 3412 3413 3414 3415 3416
	if (err)
		goto err_free;

	err = online_css(css);
	if (err)
		goto err_free;

3417
	cgroup_get(cgrp);
3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436
	css_get(css->parent);

	if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
	    parent->parent) {
		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",
			   current->comm, current->pid, ss->name);
		if (!strcmp(ss->name, "memory"))
			pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
		ss->warned_broken_hierarchy = true;
	}

	return 0;

err_free:
	percpu_ref_cancel_init(&css->refcnt);
	ss->css_free(css);
	return err;
}

T
Tejun Heo 已提交
3437
/**
L
Li Zefan 已提交
3438 3439
 * cgroup_create - create a cgroup
 * @parent: cgroup that will be parent of the new cgroup
T
Tejun Heo 已提交
3440
 * @name: name of the new cgroup
T
Tejun Heo 已提交
3441
 * @mode: mode to set on new cgroup
3442
 */
T
Tejun Heo 已提交
3443
static long cgroup_create(struct cgroup *parent, const char *name,
T
Tejun Heo 已提交
3444
			  umode_t mode)
3445
{
3446
	struct cgroup *cgrp;
3447
	struct cgroupfs_root *root = parent->root;
3448
	int ssid, err;
3449
	struct cgroup_subsys *ss;
T
Tejun Heo 已提交
3450
	struct kernfs_node *kn;
3451

T
Tejun Heo 已提交
3452
	/* allocate the cgroup and its ID, 0 is reserved for the root */
3453 3454
	cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
	if (!cgrp)
3455 3456
		return -ENOMEM;

T
Tejun Heo 已提交
3457 3458
	mutex_lock(&cgroup_tree_mutex);

3459 3460 3461 3462 3463 3464 3465 3466 3467
	/*
	 * Only live parents can have children.  Note that the liveliness
	 * check isn't strictly necessary because cgroup_mkdir() and
	 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
	 * anyway so that locking is contained inside cgroup proper and we
	 * don't get nasty surprises if we ever grow another caller.
	 */
	if (!cgroup_lock_live_group(parent)) {
		err = -ENODEV;
T
Tejun Heo 已提交
3468
		goto err_unlock_tree;
3469 3470 3471 3472 3473 3474 3475 3476 3477 3478
	}

	/*
	 * Temporarily set the pointer to NULL, so idr_find() won't return
	 * a half-baked cgroup.
	 */
	cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
	if (cgrp->id < 0) {
		err = -ENOMEM;
		goto err_unlock;
3479 3480
	}

3481
	init_cgroup_housekeeping(cgrp);
3482

3483
	cgrp->parent = parent;
3484
	cgrp->dummy_css.parent = &parent->dummy_css;
3485
	cgrp->root = parent->root;
3486

3487 3488 3489
	if (notify_on_release(parent))
		set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);

3490 3491
	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3492

T
Tejun Heo 已提交
3493
	/* create the directory */
T
Tejun Heo 已提交
3494
	kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
T
Tejun Heo 已提交
3495 3496
	if (IS_ERR(kn)) {
		err = PTR_ERR(kn);
3497
		goto err_free_id;
T
Tejun Heo 已提交
3498 3499
	}
	cgrp->kn = kn;
3500

3501 3502 3503 3504 3505 3506
	/*
	 * This extra ref will be put in cgroup_free_fn() and guarantees
	 * that @cgrp->kn is always accessible.
	 */
	kernfs_get(kn);

3507
	cgrp->serial_nr = cgroup_serial_nr_next++;
3508

3509 3510
	/* allocation complete, commit to creation */
	list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
3511
	atomic_inc(&root->nr_cgrps);
3512
	cgroup_get(parent);
3513

3514 3515 3516 3517
	/*
	 * @cgrp is now fully operational.  If something fails after this
	 * point, it'll be released via the normal destruction path.
	 */
3518 3519
	idr_replace(&root->cgroup_idr, cgrp, cgrp->id);

3520
	err = cgroup_addrm_files(cgrp, cgroup_base_files, true);
3521 3522 3523
	if (err)
		goto err_destroy;

3524
	/* let's create and online css's */
T
Tejun Heo 已提交
3525 3526 3527 3528 3529 3530
	for_each_subsys(ss, ssid) {
		if (root->subsys_mask & (1 << ssid)) {
			err = create_css(cgrp, ss);
			if (err)
				goto err_destroy;
		}
3531
	}
3532

T
Tejun Heo 已提交
3533 3534
	kernfs_activate(kn);

3535
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3536
	mutex_unlock(&cgroup_tree_mutex);
3537 3538 3539

	return 0;

T
Tejun Heo 已提交
3540
err_free_id:
3541
	idr_remove(&root->cgroup_idr, cgrp->id);
3542 3543
err_unlock:
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3544 3545
err_unlock_tree:
	mutex_unlock(&cgroup_tree_mutex);
3546
	kfree(cgrp);
3547
	return err;
3548 3549 3550 3551

err_destroy:
	cgroup_destroy_locked(cgrp);
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3552
	mutex_unlock(&cgroup_tree_mutex);
3553
	return err;
3554 3555
}

T
Tejun Heo 已提交
3556 3557
static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
			umode_t mode)
3558
{
T
Tejun Heo 已提交
3559
	struct cgroup *parent = parent_kn->priv;
3560

T
Tejun Heo 已提交
3561
	return cgroup_create(parent, name, mode);
3562 3563
}

3564 3565 3566 3567 3568
/*
 * This is called when the refcnt of a css is confirmed to be killed.
 * css_tryget() is now guaranteed to fail.
 */
static void css_killed_work_fn(struct work_struct *work)
3569
{
3570 3571 3572
	struct cgroup_subsys_state *css =
		container_of(work, struct cgroup_subsys_state, destroy_work);
	struct cgroup *cgrp = css->cgroup;
3573

T
Tejun Heo 已提交
3574
	mutex_lock(&cgroup_tree_mutex);
3575 3576
	mutex_lock(&cgroup_mutex);

3577 3578 3579 3580 3581 3582
	/*
	 * css_tryget() is guaranteed to fail now.  Tell subsystems to
	 * initate destruction.
	 */
	offline_css(css);

3583 3584 3585 3586 3587
	/*
	 * If @cgrp is marked dead, it's waiting for refs of all css's to
	 * be disabled before proceeding to the second phase of cgroup
	 * destruction.  If we are the last one, kick it off.
	 */
3588
	if (!cgrp->nr_css && cgroup_is_dead(cgrp))
3589 3590 3591
		cgroup_destroy_css_killed(cgrp);

	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3592
	mutex_unlock(&cgroup_tree_mutex);
3593 3594 3595 3596 3597 3598 3599 3600 3601

	/*
	 * Put the css refs from kill_css().  Each css holds an extra
	 * reference to the cgroup's dentry and cgroup removal proceeds
	 * regardless of css refs.  On the last put of each css, whenever
	 * that may be, the extra dentry ref is put so that dentry
	 * destruction happens only after all css's are released.
	 */
	css_put(css);
3602 3603
}

3604 3605
/* css kill confirmation processing requires process context, bounce */
static void css_killed_ref_fn(struct percpu_ref *ref)
3606 3607 3608 3609
{
	struct cgroup_subsys_state *css =
		container_of(ref, struct cgroup_subsys_state, refcnt);

3610
	INIT_WORK(&css->destroy_work, css_killed_work_fn);
3611
	queue_work(cgroup_destroy_wq, &css->destroy_work);
3612 3613
}

T
Tejun Heo 已提交
3614 3615 3616 3617
/**
 * kill_css - destroy a css
 * @css: css to destroy
 *
3618 3619 3620 3621
 * This function initiates destruction of @css by removing cgroup interface
 * files and putting its base reference.  ->css_offline() will be invoked
 * asynchronously once css_tryget() is guaranteed to fail and when the
 * reference count reaches zero, @css will be released.
T
Tejun Heo 已提交
3622 3623 3624
 */
static void kill_css(struct cgroup_subsys_state *css)
{
T
Tejun Heo 已提交
3625 3626 3627 3628
	/*
	 * This must happen before css is disassociated with its cgroup.
	 * See seq_css() for details.
	 */
3629
	cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
3630

T
Tejun Heo 已提交
3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647
	/*
	 * Killing would put the base ref, but we need to keep it alive
	 * until after ->css_offline().
	 */
	css_get(css);

	/*
	 * cgroup core guarantees that, by the time ->css_offline() is
	 * invoked, no new css reference will be given out via
	 * css_tryget().  We can't simply call percpu_ref_kill() and
	 * proceed to offlining css's because percpu_ref_kill() doesn't
	 * guarantee that the ref is seen as killed on all CPUs on return.
	 *
	 * Use percpu_ref_kill_and_confirm() to get notifications as each
	 * css is confirmed to be seen as killed on all CPUs.
	 */
	percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
}

/**
 * cgroup_destroy_locked - the first stage of cgroup destruction
 * @cgrp: cgroup to be destroyed
 *
 * css's make use of percpu refcnts whose killing latency shouldn't be
 * exposed to userland and are RCU protected.  Also, cgroup core needs to
 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
 * invoked.  To satisfy all the requirements, destruction is implemented in
 * the following two steps.
 *
 * s1. Verify @cgrp can be destroyed and mark it dying.  Remove all
 *     userland visible parts and start killing the percpu refcnts of
 *     css's.  Set up so that the next stage will be kicked off once all
 *     the percpu refcnts are confirmed to be killed.
 *
 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
 *     rest of destruction.  Once all cgroup references are gone, the
 *     cgroup is RCU-freed.
 *
 * This function implements s1.  After this step, @cgrp is gone as far as
 * the userland is concerned and a new cgroup with the same name may be
 * created.  As cgroup doesn't care about the names internally, this
 * doesn't cause any problem.
 */
3674 3675
static int cgroup_destroy_locked(struct cgroup *cgrp)
	__releases(&cgroup_mutex) __acquires(&cgroup_mutex)
3676
{
3677
	struct cgroup *child;
T
Tejun Heo 已提交
3678
	struct cgroup_subsys_state *css;
3679
	bool empty;
T
Tejun Heo 已提交
3680
	int ssid;
3681

T
Tejun Heo 已提交
3682
	lockdep_assert_held(&cgroup_tree_mutex);
3683 3684
	lockdep_assert_held(&cgroup_mutex);

3685
	/*
3686
	 * css_set_rwsem synchronizes access to ->cset_links and prevents
3687
	 * @cgrp from being removed while put_css_set() is in progress.
3688
	 */
3689
	down_read(&css_set_rwsem);
3690
	empty = list_empty(&cgrp->cset_links);
3691
	up_read(&css_set_rwsem);
3692
	if (!empty)
3693
		return -EBUSY;
L
Li Zefan 已提交
3694

3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710
	/*
	 * Make sure there's no live children.  We can't test ->children
	 * emptiness as dead children linger on it while being destroyed;
	 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
	 */
	empty = true;
	rcu_read_lock();
	list_for_each_entry_rcu(child, &cgrp->children, sibling) {
		empty = cgroup_is_dead(child);
		if (!empty)
			break;
	}
	rcu_read_unlock();
	if (!empty)
		return -EBUSY;

3711
	/*
T
Tejun Heo 已提交
3712 3713
	 * Initiate massacre of all css's.  cgroup_destroy_css_killed()
	 * will be invoked to perform the rest of destruction once the
3714 3715
	 * percpu refs of all css's are confirmed to be killed.  This
	 * involves removing the subsystem's files, drop cgroup_mutex.
3716
	 */
3717
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3718 3719
	for_each_css(css, ssid, cgrp)
		kill_css(css);
3720
	mutex_lock(&cgroup_mutex);
3721 3722 3723 3724

	/*
	 * Mark @cgrp dead.  This prevents further task migration and child
	 * creation by disabling cgroup_lock_live_group().  Note that
3725
	 * CGRP_DEAD assertion is depended upon by css_next_child() to
3726
	 * resume iteration after dropping RCU read lock.  See
3727
	 * css_next_child() for details.
3728
	 */
3729
	set_bit(CGRP_DEAD, &cgrp->flags);
3730

3731 3732 3733 3734 3735 3736 3737
	/* CGRP_DEAD is set, remove from ->release_list for the last time */
	raw_spin_lock(&release_list_lock);
	if (!list_empty(&cgrp->release_list))
		list_del_init(&cgrp->release_list);
	raw_spin_unlock(&release_list_lock);

	/*
3738 3739 3740 3741 3742 3743 3744 3745
	 * If @cgrp has css's attached, the second stage of cgroup
	 * destruction is kicked off from css_killed_work_fn() after the
	 * refs of all attached css's are killed.  If @cgrp doesn't have
	 * any css, we kick it off here.
	 */
	if (!cgrp->nr_css)
		cgroup_destroy_css_killed(cgrp);

T
Tejun Heo 已提交
3746 3747 3748
	/* remove @cgrp directory along with the base files */
	mutex_unlock(&cgroup_mutex);

3749
	/*
T
Tejun Heo 已提交
3750 3751 3752 3753 3754
	 * There are two control paths which try to determine cgroup from
	 * dentry without going through kernfs - cgroupstats_build() and
	 * css_tryget_from_dir().  Those are supported by RCU protecting
	 * clearing of cgrp->kn->priv backpointer, which should happen
	 * after all files under it have been removed.
3755
	 */
3756
	kernfs_remove(cgrp->kn);	/* @cgrp has an extra ref on its kn */
T
Tejun Heo 已提交
3757 3758
	RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL);

3759
	mutex_lock(&cgroup_mutex);
3760

3761 3762 3763
	return 0;
};

3764
/**
3765
 * cgroup_destroy_css_killed - the second step of cgroup destruction
3766 3767 3768
 * @work: cgroup->destroy_free_work
 *
 * This function is invoked from a work item for a cgroup which is being
3769 3770 3771
 * destroyed after all css's are offlined and performs the rest of
 * destruction.  This is the second step of destruction described in the
 * comment above cgroup_destroy_locked().
3772
 */
3773
static void cgroup_destroy_css_killed(struct cgroup *cgrp)
3774 3775 3776
{
	struct cgroup *parent = cgrp->parent;

T
Tejun Heo 已提交
3777
	lockdep_assert_held(&cgroup_tree_mutex);
3778
	lockdep_assert_held(&cgroup_mutex);
3779

3780
	/* delete this cgroup from parent->children */
3781
	list_del_rcu(&cgrp->sibling);
3782

3783
	cgroup_put(cgrp);
3784

3785
	set_bit(CGRP_RELEASABLE, &parent->flags);
3786
	check_for_release(parent);
3787 3788
}

T
Tejun Heo 已提交
3789
static int cgroup_rmdir(struct kernfs_node *kn)
3790
{
T
Tejun Heo 已提交
3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801
	struct cgroup *cgrp = kn->priv;
	int ret = 0;

	/*
	 * This is self-destruction but @kn can't be removed while this
	 * callback is in progress.  Let's break active protection.  Once
	 * the protection is broken, @cgrp can be destroyed at any point.
	 * Pin it so that it stays accessible.
	 */
	cgroup_get(cgrp);
	kernfs_break_active_protection(kn);
3802

T
Tejun Heo 已提交
3803
	mutex_lock(&cgroup_tree_mutex);
3804
	mutex_lock(&cgroup_mutex);
T
Tejun Heo 已提交
3805 3806 3807 3808 3809 3810 3811 3812

	/*
	 * @cgrp might already have been destroyed while we're trying to
	 * grab the mutexes.
	 */
	if (!cgroup_is_dead(cgrp))
		ret = cgroup_destroy_locked(cgrp);

3813
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3814
	mutex_unlock(&cgroup_tree_mutex);
3815

T
Tejun Heo 已提交
3816 3817
	kernfs_unbreak_active_protection(kn);
	cgroup_put(cgrp);
3818 3819 3820
	return ret;
}

T
Tejun Heo 已提交
3821 3822 3823 3824 3825 3826 3827 3828
static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
	.remount_fs		= cgroup_remount,
	.show_options		= cgroup_show_options,
	.mkdir			= cgroup_mkdir,
	.rmdir			= cgroup_rmdir,
	.rename			= cgroup_rename,
};

3829
static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3830 3831
{
	struct cgroup_subsys_state *css;
D
Diego Calleja 已提交
3832 3833

	printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3834

T
Tejun Heo 已提交
3835
	mutex_lock(&cgroup_tree_mutex);
3836 3837
	mutex_lock(&cgroup_mutex);

T
Tejun Heo 已提交
3838
	INIT_LIST_HEAD(&ss->cfts);
3839

3840
	/* Create the top cgroup state for this subsystem */
3841
	ss->root = &cgroup_dummy_root;
3842
	css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss));
3843 3844
	/* We don't handle early failures gracefully */
	BUG_ON(IS_ERR(css));
3845
	init_css(css, ss, cgroup_dummy_top);
3846

L
Li Zefan 已提交
3847
	/* Update the init_css_set to contain a subsys
3848
	 * pointer to this state - since the subsystem is
L
Li Zefan 已提交
3849 3850
	 * newly registered, all tasks and hence the
	 * init_css_set is in the subsystem's top cgroup. */
3851
	init_css_set.subsys[ss->id] = css;
3852 3853 3854

	need_forkexit_callback |= ss->fork || ss->exit;

L
Li Zefan 已提交
3855 3856 3857 3858 3859
	/* At system boot, before all subsystems have been
	 * registered, no tasks have been forked, so we don't
	 * need to invoke fork callbacks here. */
	BUG_ON(!list_empty(&init_task.tasks));

3860
	BUG_ON(online_css(css));
3861

3862
	mutex_unlock(&cgroup_mutex);
T
Tejun Heo 已提交
3863
	mutex_unlock(&cgroup_tree_mutex);
3864 3865
}

3866
/**
L
Li Zefan 已提交
3867 3868 3869 3870
 * cgroup_init_early - cgroup initialization at system boot
 *
 * Initialize cgroups at system boot, and initialize any
 * subsystems that request early init.
3871 3872 3873
 */
int __init cgroup_init_early(void)
{
3874
	struct cgroup_subsys *ss;
3875
	int i;
3876

3877
	atomic_set(&init_css_set.refcount, 1);
3878
	INIT_LIST_HEAD(&init_css_set.cgrp_links);
3879
	INIT_LIST_HEAD(&init_css_set.tasks);
3880
	INIT_HLIST_NODE(&init_css_set.hlist);
3881
	css_set_count = 1;
3882 3883
	init_cgroup_root(&cgroup_dummy_root);
	cgroup_root_count = 1;
3884
	RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
3885

3886
	init_cgrp_cset_link.cset = &init_css_set;
3887 3888
	init_cgrp_cset_link.cgrp = cgroup_dummy_top;
	list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links);
3889
	list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
3890

T
Tejun Heo 已提交
3891
	for_each_subsys(ss, i) {
3892
		WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
3893 3894
		     "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
		     i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
3895
		     ss->id, ss->name);
3896 3897 3898
		WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
		     "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);

3899
		ss->id = i;
3900
		ss->name = cgroup_subsys_name[i];
3901 3902 3903 3904 3905 3906 3907 3908

		if (ss->early_init)
			cgroup_init_subsys(ss);
	}
	return 0;
}

/**
L
Li Zefan 已提交
3909 3910 3911 3912
 * cgroup_init - cgroup initialization
 *
 * Register cgroup filesystem and /proc file, and initialize
 * any subsystems that didn't request early init.
3913 3914 3915
 */
int __init cgroup_init(void)
{
3916
	struct cgroup_subsys *ss;
3917
	unsigned long key;
3918
	int i, err;
3919

T
Tejun Heo 已提交
3920
	BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
3921

T
Tejun Heo 已提交
3922
	for_each_subsys(ss, i) {
3923 3924
		if (!ss->early_init)
			cgroup_init_subsys(ss);
3925 3926 3927 3928 3929 3930 3931

		/*
		 * cftype registration needs kmalloc and can't be done
		 * during early_init.  Register base cftypes separately.
		 */
		if (ss->base_cftypes)
			WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes));
3932 3933
	}

3934
	/* allocate id for the dummy hierarchy */
T
Tejun Heo 已提交
3935 3936
	mutex_lock(&cgroup_mutex);

3937 3938 3939 3940
	/* Add init_css_set to the hash table */
	key = css_set_hash(init_css_set.subsys);
	hash_add(css_set_table, &init_css_set.hlist, key);

3941
	BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1));
3942

3943 3944 3945 3946
	err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top,
			0, 1, GFP_KERNEL);
	BUG_ON(err < 0);

T
Tejun Heo 已提交
3947 3948
	mutex_unlock(&cgroup_mutex);

3949
	cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
T
Tejun Heo 已提交
3950 3951
	if (!cgroup_kobj)
		return -ENOMEM;
3952

3953
	err = register_filesystem(&cgroup_fs_type);
3954 3955
	if (err < 0) {
		kobject_put(cgroup_kobj);
T
Tejun Heo 已提交
3956
		return err;
3957
	}
3958

L
Li Zefan 已提交
3959
	proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
T
Tejun Heo 已提交
3960
	return 0;
3961
}
3962

3963 3964 3965 3966 3967
static int __init cgroup_wq_init(void)
{
	/*
	 * There isn't much point in executing destruction path in
	 * parallel.  Good chunk is serialized with cgroup_mutex anyway.
3968 3969 3970 3971 3972
	 *
	 * XXX: Must be ordered to make sure parent is offlined after
	 * children.  The ordering requirement is for memcg where a
	 * parent's offline may wait for a child's leading to deadlock.  In
	 * the long term, this should be fixed from memcg side.
3973 3974 3975 3976
	 *
	 * We would prefer to do this in cgroup_init() above, but that
	 * is called before init_workqueues(): so leave this until after.
	 */
3977
	cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0);
3978
	BUG_ON(!cgroup_destroy_wq);
3979 3980 3981 3982 3983 3984 3985 3986 3987

	/*
	 * Used to destroy pidlists and separate to serve as flush domain.
	 * Cap @max_active to 1 too.
	 */
	cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
						    0, 1);
	BUG_ON(!cgroup_pidlist_destroy_wq);

3988 3989 3990 3991
	return 0;
}
core_initcall(cgroup_wq_init);

3992 3993 3994 3995 3996 3997
/*
 * proc_cgroup_show()
 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
 *  - Used for /proc/<pid>/cgroup.
 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
 *    doesn't really matter if tsk->cgroup changes after we read it,
3998
 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3999 4000 4001 4002 4003 4004
 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
 *    cgroup to top_cgroup.
 */

/* TODO: Use a proper seq_file iterator */
4005
int proc_cgroup_show(struct seq_file *m, void *v)
4006 4007 4008
{
	struct pid *pid;
	struct task_struct *tsk;
T
Tejun Heo 已提交
4009
	char *buf, *path;
4010 4011 4012 4013
	int retval;
	struct cgroupfs_root *root;

	retval = -ENOMEM;
T
Tejun Heo 已提交
4014
	buf = kmalloc(PATH_MAX, GFP_KERNEL);
4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026
	if (!buf)
		goto out;

	retval = -ESRCH;
	pid = m->private;
	tsk = get_pid_task(pid, PIDTYPE_PID);
	if (!tsk)
		goto out_free;

	retval = 0;

	mutex_lock(&cgroup_mutex);
4027
	down_read(&css_set_rwsem);
4028

4029
	for_each_active_root(root) {
4030
		struct cgroup_subsys *ss;
4031
		struct cgroup *cgrp;
T
Tejun Heo 已提交
4032
		int ssid, count = 0;
4033

4034
		seq_printf(m, "%d:", root->hierarchy_id);
T
Tejun Heo 已提交
4035 4036 4037
		for_each_subsys(ss, ssid)
			if (root->subsys_mask & (1 << ssid))
				seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4038 4039 4040
		if (strlen(root->name))
			seq_printf(m, "%sname=%s", count ? "," : "",
				   root->name);
4041
		seq_putc(m, ':');
4042
		cgrp = task_cgroup_from_root(tsk, root);
T
Tejun Heo 已提交
4043 4044 4045
		path = cgroup_path(cgrp, buf, PATH_MAX);
		if (!path) {
			retval = -ENAMETOOLONG;
4046
			goto out_unlock;
T
Tejun Heo 已提交
4047 4048
		}
		seq_puts(m, path);
4049 4050 4051 4052
		seq_putc(m, '\n');
	}

out_unlock:
4053
	up_read(&css_set_rwsem);
4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064
	mutex_unlock(&cgroup_mutex);
	put_task_struct(tsk);
out_free:
	kfree(buf);
out:
	return retval;
}

/* Display information about each subsystem and each hierarchy */
static int proc_cgroupstats_show(struct seq_file *m, void *v)
{
4065
	struct cgroup_subsys *ss;
4066 4067
	int i;

4068
	seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
B
Ben Blum 已提交
4069 4070 4071 4072 4073
	/*
	 * ideally we don't want subsystems moving around while we do this.
	 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
	 * subsys/hierarchy state.
	 */
4074
	mutex_lock(&cgroup_mutex);
4075 4076

	for_each_subsys(ss, i)
4077 4078
		seq_printf(m, "%s\t%d\t%d\t%d\n",
			   ss->name, ss->root->hierarchy_id,
4079
			   atomic_read(&ss->root->nr_cgrps), !ss->disabled);
4080

4081 4082 4083 4084 4085 4086
	mutex_unlock(&cgroup_mutex);
	return 0;
}

static int cgroupstats_open(struct inode *inode, struct file *file)
{
A
Al Viro 已提交
4087
	return single_open(file, proc_cgroupstats_show, NULL);
4088 4089
}

4090
static const struct file_operations proc_cgroupstats_operations = {
4091 4092 4093 4094 4095 4096
	.open = cgroupstats_open,
	.read = seq_read,
	.llseek = seq_lseek,
	.release = single_release,
};

4097 4098
/**
 * cgroup_fork - attach newly forked task to its parents cgroup.
L
Li Zefan 已提交
4099
 * @child: pointer to task_struct of forking parent process.
4100 4101 4102 4103 4104
 *
 * Description: A task inherits its parent's cgroup at fork().
 *
 * A pointer to the shared css_set was automatically copied in
 * fork.c by dup_task_struct().  However, we ignore that copy, since
4105 4106 4107 4108
 * it was not made under the protection of RCU or cgroup_mutex, so
 * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
 * have already changed current->cgroups, allowing the previously
 * referenced cgroup group to be removed and freed.
4109 4110 4111 4112 4113 4114
 *
 * At the point that cgroup_fork() is called, 'current' is the parent
 * task, and the passed argument 'child' points to the child task.
 */
void cgroup_fork(struct task_struct *child)
{
4115
	task_lock(current);
4116
	get_css_set(task_css_set(current));
4117
	child->cgroups = current->cgroups;
4118
	task_unlock(current);
4119
	INIT_LIST_HEAD(&child->cg_list);
4120 4121
}

4122
/**
L
Li Zefan 已提交
4123 4124 4125
 * cgroup_post_fork - called on a new task after adding it to the task list
 * @child: the task in question
 *
4126 4127 4128
 * Adds the task to the list running through its css_set if necessary and
 * call the subsystem fork() callbacks.  Has to be after the task is
 * visible on the task list in case we race with the first call to
4129
 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
4130
 * list.
L
Li Zefan 已提交
4131
 */
4132 4133
void cgroup_post_fork(struct task_struct *child)
{
4134
	struct cgroup_subsys *ss;
4135 4136
	int i;

4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147
	/*
	 * use_task_css_set_links is set to 1 before we walk the tasklist
	 * under the tasklist_lock and we read it here after we added the child
	 * to the tasklist under the tasklist_lock as well. If the child wasn't
	 * yet in the tasklist when we walked through it from
	 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
	 * should be visible now due to the paired locking and barriers implied
	 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
	 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
	 * lock on fork.
	 */
4148
	if (use_task_css_set_links) {
4149
		down_write(&css_set_rwsem);
4150 4151
		task_lock(child);
		if (list_empty(&child->cg_list))
4152
			list_add(&child->cg_list, &task_css_set(child)->tasks);
4153
		task_unlock(child);
4154
		up_write(&css_set_rwsem);
4155
	}
4156 4157 4158 4159 4160 4161 4162

	/*
	 * Call ss->fork().  This must happen after @child is linked on
	 * css_set; otherwise, @child might change state between ->fork()
	 * and addition to css_set.
	 */
	if (need_forkexit_callback) {
T
Tejun Heo 已提交
4163
		for_each_subsys(ss, i)
4164 4165 4166
			if (ss->fork)
				ss->fork(child);
	}
4167
}
4168

4169 4170 4171
/**
 * cgroup_exit - detach cgroup from exiting task
 * @tsk: pointer to task_struct of exiting process
L
Li Zefan 已提交
4172
 * @run_callback: run exit callbacks?
4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200
 *
 * Description: Detach cgroup from @tsk and release it.
 *
 * Note that cgroups marked notify_on_release force every task in
 * them to take the global cgroup_mutex mutex when exiting.
 * This could impact scaling on very large systems.  Be reluctant to
 * use notify_on_release cgroups where very high task exit scaling
 * is required on large systems.
 *
 * the_top_cgroup_hack:
 *
 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
 *
 *    We call cgroup_exit() while the task is still competent to
 *    handle notify_on_release(), then leave the task attached to the
 *    root cgroup in each hierarchy for the remainder of its exit.
 *
 *    To do this properly, we would increment the reference count on
 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
 *    code we would add a second cgroup function call, to drop that
 *    reference.  This would just create an unnecessary hot spot on
 *    the top_cgroup reference count, to no avail.
 *
 *    Normally, holding a reference to a cgroup without bumping its
 *    count is unsafe.   The cgroup could go away, or someone could
 *    attach us to a different cgroup, decrementing the count on
 *    the first cgroup that we never incremented.  But in this case,
 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4201 4202
 *    which wards off any cgroup_attach_task() attempts, or task is a failed
 *    fork, never visible to cgroup_attach_task.
4203 4204 4205
 */
void cgroup_exit(struct task_struct *tsk, int run_callbacks)
{
4206
	struct cgroup_subsys *ss;
4207
	struct css_set *cset;
4208
	int i;
4209 4210

	/*
4211 4212
	 * Unlink from the css_set task list if necessary.  Optimistically
	 * check cg_list before taking css_set_rwsem.
4213 4214
	 */
	if (!list_empty(&tsk->cg_list)) {
4215
		down_write(&css_set_rwsem);
4216
		if (!list_empty(&tsk->cg_list))
4217
			list_del_init(&tsk->cg_list);
4218
		up_write(&css_set_rwsem);
4219 4220
	}

4221 4222
	/* Reassign the task to the init_css_set. */
	task_lock(tsk);
4223 4224
	cset = task_css_set(tsk);
	RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
4225 4226

	if (run_callbacks && need_forkexit_callback) {
T
Tejun Heo 已提交
4227 4228
		/* see cgroup_post_fork() for details */
		for_each_subsys(ss, i) {
4229
			if (ss->exit) {
4230 4231
				struct cgroup_subsys_state *old_css = cset->subsys[i];
				struct cgroup_subsys_state *css = task_css(tsk, i);
4232

4233
				ss->exit(css, old_css, tsk);
4234 4235 4236
			}
		}
	}
4237
	task_unlock(tsk);
4238

4239
	put_css_set(cset, true);
4240
}
4241

4242
static void check_for_release(struct cgroup *cgrp)
4243
{
4244
	if (cgroup_is_releasable(cgrp) &&
T
Tejun Heo 已提交
4245
	    list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
4246 4247
		/*
		 * Control Group is currently removeable. If it's not
4248
		 * already queued for a userspace notification, queue
4249 4250
		 * it now
		 */
4251
		int need_schedule_work = 0;
4252

4253
		raw_spin_lock(&release_list_lock);
4254
		if (!cgroup_is_dead(cgrp) &&
4255 4256
		    list_empty(&cgrp->release_list)) {
			list_add(&cgrp->release_list, &release_list);
4257 4258
			need_schedule_work = 1;
		}
4259
		raw_spin_unlock(&release_list_lock);
4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291
		if (need_schedule_work)
			schedule_work(&release_agent_work);
	}
}

/*
 * Notify userspace when a cgroup is released, by running the
 * configured release agent with the name of the cgroup (path
 * relative to the root of cgroup file system) as the argument.
 *
 * Most likely, this user command will try to rmdir this cgroup.
 *
 * This races with the possibility that some other task will be
 * attached to this cgroup before it is removed, or that some other
 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
 * unused, and this cgroup will be reprieved from its death sentence,
 * to continue to serve a useful existence.  Next time it's released,
 * we will get notified again, if it still has 'notify_on_release' set.
 *
 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
 * means only wait until the task is successfully execve()'d.  The
 * separate release agent task is forked by call_usermodehelper(),
 * then control in this thread returns here, without waiting for the
 * release agent task.  We don't bother to wait because the caller of
 * this routine has no use for the exit status of the release agent
 * task, so no sense holding our caller up for that.
 */
static void cgroup_release_agent(struct work_struct *work)
{
	BUG_ON(work != &release_agent_work);
	mutex_lock(&cgroup_mutex);
4292
	raw_spin_lock(&release_list_lock);
4293 4294 4295
	while (!list_empty(&release_list)) {
		char *argv[3], *envp[3];
		int i;
T
Tejun Heo 已提交
4296
		char *pathbuf = NULL, *agentbuf = NULL, *path;
4297
		struct cgroup *cgrp = list_entry(release_list.next,
4298 4299
						    struct cgroup,
						    release_list);
4300
		list_del_init(&cgrp->release_list);
4301
		raw_spin_unlock(&release_list_lock);
T
Tejun Heo 已提交
4302
		pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
4303 4304
		if (!pathbuf)
			goto continue_free;
T
Tejun Heo 已提交
4305 4306
		path = cgroup_path(cgrp, pathbuf, PATH_MAX);
		if (!path)
4307 4308 4309 4310
			goto continue_free;
		agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
		if (!agentbuf)
			goto continue_free;
4311 4312

		i = 0;
4313
		argv[i++] = agentbuf;
T
Tejun Heo 已提交
4314
		argv[i++] = path;
4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328
		argv[i] = NULL;

		i = 0;
		/* minimal command environment */
		envp[i++] = "HOME=/";
		envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
		envp[i] = NULL;

		/* Drop the lock while we invoke the usermode helper,
		 * since the exec could involve hitting disk and hence
		 * be a slow process */
		mutex_unlock(&cgroup_mutex);
		call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
		mutex_lock(&cgroup_mutex);
4329 4330 4331
 continue_free:
		kfree(pathbuf);
		kfree(agentbuf);
4332
		raw_spin_lock(&release_list_lock);
4333
	}
4334
	raw_spin_unlock(&release_list_lock);
4335 4336
	mutex_unlock(&cgroup_mutex);
}
4337 4338 4339

static int __init cgroup_disable(char *str)
{
4340
	struct cgroup_subsys *ss;
4341
	char *token;
4342
	int i;
4343 4344 4345 4346

	while ((token = strsep(&str, ",")) != NULL) {
		if (!*token)
			continue;
4347

T
Tejun Heo 已提交
4348
		for_each_subsys(ss, i) {
4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359
			if (!strcmp(token, ss->name)) {
				ss->disabled = 1;
				printk(KERN_INFO "Disabling %s control group"
					" subsystem\n", ss->name);
				break;
			}
		}
	}
	return 1;
}
__setup("cgroup_disable=", cgroup_disable);
K
KAMEZAWA Hiroyuki 已提交
4360

4361
/**
4362
 * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir
4363 4364
 * @dentry: directory dentry of interest
 * @ss: subsystem of interest
4365
 *
4366 4367 4368
 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
 * to get the corresponding css and return it.  If such css doesn't exist
 * or can't be pinned, an ERR_PTR value is returned.
S
Stephane Eranian 已提交
4369
 */
4370 4371
struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry,
						struct cgroup_subsys *ss)
S
Stephane Eranian 已提交
4372
{
T
Tejun Heo 已提交
4373 4374
	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
	struct cgroup_subsys_state *css = NULL;
S
Stephane Eranian 已提交
4375
	struct cgroup *cgrp;
4376

4377
	/* is @dentry a cgroup dir? */
T
Tejun Heo 已提交
4378 4379
	if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
	    kernfs_type(kn) != KERNFS_DIR)
S
Stephane Eranian 已提交
4380 4381
		return ERR_PTR(-EBADF);

4382 4383
	rcu_read_lock();

T
Tejun Heo 已提交
4384 4385 4386 4387 4388 4389 4390 4391
	/*
	 * This path doesn't originate from kernfs and @kn could already
	 * have been or be removed at any point.  @kn->priv is RCU
	 * protected for this access.  See destroy_locked() for details.
	 */
	cgrp = rcu_dereference(kn->priv);
	if (cgrp)
		css = cgroup_css(cgrp, ss);
4392 4393 4394 4395 4396 4397

	if (!css || !css_tryget(css))
		css = ERR_PTR(-ENOENT);

	rcu_read_unlock();
	return css;
S
Stephane Eranian 已提交
4398 4399
}

4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411
/**
 * css_from_id - lookup css by id
 * @id: the cgroup id
 * @ss: cgroup subsys to be looked into
 *
 * Returns the css if there's valid one with @id, otherwise returns NULL.
 * Should be called under rcu_read_lock().
 */
struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
{
	struct cgroup *cgrp;

T
Tejun Heo 已提交
4412
	cgroup_assert_mutexes_or_rcu_locked();
4413 4414 4415

	cgrp = idr_find(&ss->root->cgroup_idr, id);
	if (cgrp)
4416
		return cgroup_css(cgrp, ss);
4417
	return NULL;
S
Stephane Eranian 已提交
4418 4419
}

4420
#ifdef CONFIG_CGROUP_DEBUG
4421 4422
static struct cgroup_subsys_state *
debug_css_alloc(struct cgroup_subsys_state *parent_css)
4423 4424 4425 4426 4427 4428 4429 4430 4431
{
	struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);

	if (!css)
		return ERR_PTR(-ENOMEM);

	return css;
}

4432
static void debug_css_free(struct cgroup_subsys_state *css)
4433
{
4434
	kfree(css);
4435 4436
}

4437 4438
static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
				struct cftype *cft)
4439
{
4440
	return cgroup_task_count(css->cgroup);
4441 4442
}

4443 4444
static u64 current_css_set_read(struct cgroup_subsys_state *css,
				struct cftype *cft)
4445 4446 4447 4448
{
	return (u64)(unsigned long)current->cgroups;
}

4449
static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
L
Li Zefan 已提交
4450
					 struct cftype *cft)
4451 4452 4453 4454
{
	u64 count;

	rcu_read_lock();
4455
	count = atomic_read(&task_css_set(current)->refcount);
4456 4457 4458 4459
	rcu_read_unlock();
	return count;
}

4460
static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
4461
{
4462
	struct cgrp_cset_link *link;
4463
	struct css_set *cset;
T
Tejun Heo 已提交
4464 4465 4466 4467 4468
	char *name_buf;

	name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
	if (!name_buf)
		return -ENOMEM;
4469

4470
	down_read(&css_set_rwsem);
4471
	rcu_read_lock();
4472
	cset = rcu_dereference(current->cgroups);
4473
	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
4474
		struct cgroup *c = link->cgrp;
4475 4476
		const char *name = "?";

T
Tejun Heo 已提交
4477 4478 4479 4480
		if (c != cgroup_dummy_top) {
			cgroup_name(c, name_buf, NAME_MAX + 1);
			name = name_buf;
		}
4481

4482 4483
		seq_printf(seq, "Root %d group %s\n",
			   c->root->hierarchy_id, name);
4484 4485
	}
	rcu_read_unlock();
4486
	up_read(&css_set_rwsem);
T
Tejun Heo 已提交
4487
	kfree(name_buf);
4488 4489 4490 4491
	return 0;
}

#define MAX_TASKS_SHOWN_PER_CSS 25
4492
static int cgroup_css_links_read(struct seq_file *seq, void *v)
4493
{
4494
	struct cgroup_subsys_state *css = seq_css(seq);
4495
	struct cgrp_cset_link *link;
4496

4497
	down_read(&css_set_rwsem);
4498
	list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
4499
		struct css_set *cset = link->cset;
4500 4501
		struct task_struct *task;
		int count = 0;
4502 4503
		seq_printf(seq, "css_set %p\n", cset);
		list_for_each_entry(task, &cset->tasks, cg_list) {
4504 4505 4506 4507 4508 4509 4510 4511 4512
			if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
				seq_puts(seq, "  ...\n");
				break;
			} else {
				seq_printf(seq, "  task %d\n",
					   task_pid_vnr(task));
			}
		}
	}
4513
	up_read(&css_set_rwsem);
4514 4515 4516
	return 0;
}

4517
static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
4518
{
4519
	return test_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537
}

static struct cftype debug_files[] =  {
	{
		.name = "taskcount",
		.read_u64 = debug_taskcount_read,
	},

	{
		.name = "current_css_set",
		.read_u64 = current_css_set_read,
	},

	{
		.name = "current_css_set_refcount",
		.read_u64 = current_css_set_refcount_read,
	},

4538 4539
	{
		.name = "current_css_set_cg_links",
4540
		.seq_show = current_css_set_cg_links_read,
4541 4542 4543 4544
	},

	{
		.name = "cgroup_css_links",
4545
		.seq_show = cgroup_css_links_read,
4546 4547
	},

4548 4549 4550 4551 4552
	{
		.name = "releasable",
		.read_u64 = releasable_read,
	},

4553 4554
	{ }	/* terminate */
};
4555

4556
struct cgroup_subsys debug_cgrp_subsys = {
4557 4558
	.css_alloc = debug_css_alloc,
	.css_free = debug_css_free,
4559
	.base_cftypes = debug_files,
4560 4561
};
#endif /* CONFIG_CGROUP_DEBUG */