memcontrol.c 181.7 KB
Newer Older
B
Balbir Singh 已提交
1 2 3 4 5
/* memcontrol.c - Memory Controller
 *
 * Copyright IBM Corporation, 2007
 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
 *
6 7 8
 * Copyright 2007 OpenVZ SWsoft Inc
 * Author: Pavel Emelianov <xemul@openvz.org>
 *
9 10 11 12
 * Memory thresholds
 * Copyright (C) 2009 Nokia Corporation
 * Author: Kirill A. Shutemov
 *
13 14 15 16
 * Kernel Memory Controller
 * Copyright (C) 2012 Parallels Inc. and Google Inc.
 * Authors: Glauber Costa and Suleiman Souhlal
 *
B
Balbir Singh 已提交
17 18 19 20 21 22 23 24 25 26 27 28 29 30
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 */

#include <linux/res_counter.h>
#include <linux/memcontrol.h>
#include <linux/cgroup.h>
31
#include <linux/mm.h>
32
#include <linux/hugetlb.h>
K
KAMEZAWA Hiroyuki 已提交
33
#include <linux/pagemap.h>
34
#include <linux/smp.h>
35
#include <linux/page-flags.h>
36
#include <linux/backing-dev.h>
37 38
#include <linux/bit_spinlock.h>
#include <linux/rcupdate.h>
39
#include <linux/limits.h>
40
#include <linux/export.h>
41
#include <linux/mutex.h>
42
#include <linux/slab.h>
43
#include <linux/swap.h>
44
#include <linux/swapops.h>
45
#include <linux/spinlock.h>
46 47
#include <linux/eventfd.h>
#include <linux/sort.h>
48
#include <linux/fs.h>
49
#include <linux/seq_file.h>
50
#include <linux/vmalloc.h>
51
#include <linux/vmpressure.h>
52
#include <linux/mm_inline.h>
53
#include <linux/page_cgroup.h>
54
#include <linux/cpu.h>
55
#include <linux/oom.h>
K
KAMEZAWA Hiroyuki 已提交
56
#include "internal.h"
G
Glauber Costa 已提交
57
#include <net/sock.h>
M
Michal Hocko 已提交
58
#include <net/ip.h>
G
Glauber Costa 已提交
59
#include <net/tcp_memcontrol.h>
B
Balbir Singh 已提交
60

61 62
#include <asm/uaccess.h>

63 64
#include <trace/events/vmscan.h>

65
struct cgroup_subsys mem_cgroup_subsys __read_mostly;
66 67
EXPORT_SYMBOL(mem_cgroup_subsys);

68
#define MEM_CGROUP_RECLAIM_RETRIES	5
69
static struct mem_cgroup *root_mem_cgroup __read_mostly;
B
Balbir Singh 已提交
70

A
Andrew Morton 已提交
71
#ifdef CONFIG_MEMCG_SWAP
L
Li Zefan 已提交
72
/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
73
int do_swap_account __read_mostly;
74 75

/* for remember boot option*/
A
Andrew Morton 已提交
76
#ifdef CONFIG_MEMCG_SWAP_ENABLED
77 78 79 80 81
static int really_do_swap_account __initdata = 1;
#else
static int really_do_swap_account __initdata = 0;
#endif

82
#else
83
#define do_swap_account		0
84 85 86
#endif


87 88 89 90 91 92 93
/*
 * Statistics for memory cgroup.
 */
enum mem_cgroup_stat_index {
	/*
	 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
	 */
94 95 96 97 98
	MEM_CGROUP_STAT_CACHE,		/* # of pages charged as cache */
	MEM_CGROUP_STAT_RSS,		/* # of pages charged as anon rss */
	MEM_CGROUP_STAT_RSS_HUGE,	/* # of pages charged as anon huge */
	MEM_CGROUP_STAT_FILE_MAPPED,	/* # of pages charged as file rss */
	MEM_CGROUP_STAT_SWAP,		/* # of pages, swapped out */
99 100 101
	MEM_CGROUP_STAT_NSTATS,
};

102 103 104
static const char * const mem_cgroup_stat_names[] = {
	"cache",
	"rss",
105
	"rss_huge",
106 107 108 109
	"mapped_file",
	"swap",
};

110 111 112
enum mem_cgroup_events_index {
	MEM_CGROUP_EVENTS_PGPGIN,	/* # of pages paged in */
	MEM_CGROUP_EVENTS_PGPGOUT,	/* # of pages paged out */
113 114
	MEM_CGROUP_EVENTS_PGFAULT,	/* # of page-faults */
	MEM_CGROUP_EVENTS_PGMAJFAULT,	/* # of major page-faults */
115 116
	MEM_CGROUP_EVENTS_NSTATS,
};
117 118 119 120 121 122 123 124

static const char * const mem_cgroup_events_names[] = {
	"pgpgin",
	"pgpgout",
	"pgfault",
	"pgmajfault",
};

125 126 127 128 129 130 131 132
static const char * const mem_cgroup_lru_names[] = {
	"inactive_anon",
	"active_anon",
	"inactive_file",
	"active_file",
	"unevictable",
};

133 134 135 136 137 138 139 140
/*
 * Per memcg event counter is incremented at every pagein/pageout. With THP,
 * it will be incremated by the number of pages. This counter is used for
 * for trigger some periodic events. This is straightforward and better
 * than using jiffies etc. to handle periodic memcg event.
 */
enum mem_cgroup_events_target {
	MEM_CGROUP_TARGET_THRESH,
141
	MEM_CGROUP_TARGET_SOFTLIMIT,
142
	MEM_CGROUP_TARGET_NUMAINFO,
143 144
	MEM_CGROUP_NTARGETS,
};
145 146 147
#define THRESHOLDS_EVENTS_TARGET 128
#define SOFTLIMIT_EVENTS_TARGET 1024
#define NUMAINFO_EVENTS_TARGET	1024
148

149
struct mem_cgroup_stat_cpu {
150
	long count[MEM_CGROUP_STAT_NSTATS];
151
	unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
152
	unsigned long nr_page_events;
153
	unsigned long targets[MEM_CGROUP_NTARGETS];
154 155
};

156
struct mem_cgroup_reclaim_iter {
M
Michal Hocko 已提交
157 158 159 160
	/*
	 * last scanned hierarchy member. Valid only if last_dead_count
	 * matches memcg->dead_count of the hierarchy root group.
	 */
161
	struct mem_cgroup *last_visited;
M
Michal Hocko 已提交
162 163
	unsigned long last_dead_count;

164 165 166 167
	/* scan generation, increased every round-trip */
	unsigned int generation;
};

168 169 170 171
/*
 * per-zone information in memory controller.
 */
struct mem_cgroup_per_zone {
172
	struct lruvec		lruvec;
173
	unsigned long		lru_size[NR_LRU_LISTS];
K
KOSAKI Motohiro 已提交
174

175 176
	struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];

177
	struct mem_cgroup	*memcg;		/* Back pointer, we cannot */
178
						/* use container_of	   */
179 180 181 182 183 184
};

struct mem_cgroup_per_node {
	struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
};

185 186 187 188 189
struct mem_cgroup_threshold {
	struct eventfd_ctx *eventfd;
	u64 threshold;
};

K
KAMEZAWA Hiroyuki 已提交
190
/* For threshold */
191
struct mem_cgroup_threshold_ary {
192
	/* An array index points to threshold just below or equal to usage. */
193
	int current_threshold;
194 195 196 197 198
	/* Size of entries[] */
	unsigned int size;
	/* Array of thresholds */
	struct mem_cgroup_threshold entries[0];
};
199 200 201 202 203 204 205 206 207 208 209 210

struct mem_cgroup_thresholds {
	/* Primary thresholds array */
	struct mem_cgroup_threshold_ary *primary;
	/*
	 * Spare threshold array.
	 * This is needed to make mem_cgroup_unregister_event() "never fail".
	 * It must be able to store at least primary->size - 1 entries.
	 */
	struct mem_cgroup_threshold_ary *spare;
};

K
KAMEZAWA Hiroyuki 已提交
211 212 213 214 215
/* for OOM */
struct mem_cgroup_eventfd_list {
	struct list_head list;
	struct eventfd_ctx *eventfd;
};
216

217 218
static void mem_cgroup_threshold(struct mem_cgroup *memcg);
static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
219

B
Balbir Singh 已提交
220 221 222 223 224 225 226
/*
 * The memory controller data structure. The memory controller controls both
 * page cache and RSS per cgroup. We would eventually like to provide
 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
 * to help the administrator determine what knobs to tune.
 *
 * TODO: Add a water mark for the memory controller. Reclaim will begin when
227 228 229
 * we hit the water mark. May be even add a low water mark, such that
 * no reclaim occurs from a cgroup at it's low water mark, this is
 * a feature that will be implemented much later in the future.
B
Balbir Singh 已提交
230 231 232 233 234 235 236
 */
struct mem_cgroup {
	struct cgroup_subsys_state css;
	/*
	 * the counter to account for memory usage
	 */
	struct res_counter res;
237

238 239 240
	/* vmpressure notifications */
	struct vmpressure vmpressure;

241 242 243 244
	/*
	 * the counter to account for mem+swap usage.
	 */
	struct res_counter memsw;
245

246 247 248 249
	/*
	 * the counter to account for kernel memory usage.
	 */
	struct res_counter kmem;
250 251 252 253
	/*
	 * Should the accounting and control be hierarchical, per subtree?
	 */
	bool use_hierarchy;
254
	unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
255 256 257

	bool		oom_lock;
	atomic_t	under_oom;
258
	atomic_t	oom_wakeups;
259

260
	int	swappiness;
261 262
	/* OOM-Killer disable */
	int		oom_kill_disable;
K
KOSAKI Motohiro 已提交
263

264 265 266
	/* set when res.limit == memsw.limit */
	bool		memsw_is_minimum;

267 268 269 270
	/* protect arrays of thresholds */
	struct mutex thresholds_lock;

	/* thresholds for memory usage. RCU-protected */
271
	struct mem_cgroup_thresholds thresholds;
272

273
	/* thresholds for mem+swap usage. RCU-protected */
274
	struct mem_cgroup_thresholds memsw_thresholds;
275

K
KAMEZAWA Hiroyuki 已提交
276 277
	/* For oom notifier event fd */
	struct list_head oom_notify;
278

279 280 281 282
	/*
	 * Should we move charges of a task when a task is moved into this
	 * mem_cgroup ? And what type of charges should we move ?
	 */
A
Andrew Morton 已提交
283
	unsigned long move_charge_at_immigrate;
284 285 286 287
	/*
	 * set > 0 if pages under this cgroup are moving to other cgroup.
	 */
	atomic_t	moving_account;
288 289
	/* taken only while moving_account > 0 */
	spinlock_t	move_lock;
290
	/*
291
	 * percpu counter.
292
	 */
293
	struct mem_cgroup_stat_cpu __percpu *stat;
294 295 296 297 298 299
	/*
	 * used when a cpu is offlined or other synchronizations
	 * See mem_cgroup_read_stat().
	 */
	struct mem_cgroup_stat_cpu nocpu_base;
	spinlock_t pcp_counter_lock;
G
Glauber Costa 已提交
300

M
Michal Hocko 已提交
301
	atomic_t	dead_count;
M
Michal Hocko 已提交
302
#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
G
Glauber Costa 已提交
303 304
	struct tcp_memcontrol tcp_mem;
#endif
305 306 307 308 309 310 311 312
#if defined(CONFIG_MEMCG_KMEM)
	/* analogous to slab_common's slab_caches list. per-memcg */
	struct list_head memcg_slab_caches;
	/* Not a spinlock, we can take a lot of time walking the list */
	struct mutex slab_caches_mutex;
        /* Index in the kmem_cache->memcg_params->memcg_caches array */
	int kmemcg_id;
#endif
313 314 315 316 317 318 319

	int last_scanned_node;
#if MAX_NUMNODES > 1
	nodemask_t	scan_nodes;
	atomic_t	numainfo_events;
	atomic_t	numainfo_updating;
#endif
320 321 322 323 324 325 326 327
	/*
	 * Protects soft_contributed transitions.
	 * See mem_cgroup_update_soft_limit
	 */
	spinlock_t soft_lock;

	/*
	 * If true then this group has increased parents' children_in_excess
A
Andrew Morton 已提交
328
	 * when it got over the soft limit.
329 330 331 332 333 334 335
	 * When a group falls bellow the soft limit, parents' children_in_excess
	 * is decreased and soft_contributed changed to false.
	 */
	bool soft_contributed;

	/* Number of children that are in soft limit excess */
	atomic_t children_in_excess;
336

337 338
	struct mem_cgroup_per_node *nodeinfo[0];
	/* WARNING: nodeinfo must be the last member here */
B
Balbir Singh 已提交
339 340
};

341 342 343 344 345 346
static size_t memcg_size(void)
{
	return sizeof(struct mem_cgroup) +
		nr_node_ids * sizeof(struct mem_cgroup_per_node);
}

347 348 349
/* internal only representation about the status of kmem accounting. */
enum {
	KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */
350
	KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */
351
	KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
352 353
};

354 355 356
/* We account when limit is on, but only after call sites are patched */
#define KMEM_ACCOUNTED_MASK \
		((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED))
357 358 359 360 361 362

#ifdef CONFIG_MEMCG_KMEM
static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
{
	set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
}
363 364 365 366 367 368

static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
{
	return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
}

369 370 371 372 373
static void memcg_kmem_set_activated(struct mem_cgroup *memcg)
{
	set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
}

374 375 376 377 378
static void memcg_kmem_clear_activated(struct mem_cgroup *memcg)
{
	clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
}

379 380
static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
{
381 382 383 384 385
	/*
	 * Our caller must use css_get() first, because memcg_uncharge_kmem()
	 * will call css_put() if it sees the memcg is dead.
	 */
	smp_wmb();
386 387 388 389 390 391 392 393 394
	if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
		set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
}

static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
{
	return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
				  &memcg->kmem_account_flags);
}
395 396
#endif

397 398
/* Stuffs for move charges at task migration. */
/*
399 400
 * Types of charges to be moved. "move_charge_at_immitgrate" and
 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
401 402
 */
enum move_type {
403
	MOVE_CHARGE_TYPE_ANON,	/* private anonymous page and swap of it */
404
	MOVE_CHARGE_TYPE_FILE,	/* file page(including tmpfs) and swap of it */
405 406 407
	NR_MOVE_TYPE,
};

408 409
/* "mc" and its members are protected by cgroup_mutex */
static struct move_charge_struct {
410
	spinlock_t	  lock; /* for from, to */
411 412
	struct mem_cgroup *from;
	struct mem_cgroup *to;
413
	unsigned long immigrate_flags;
414
	unsigned long precharge;
415
	unsigned long moved_charge;
416
	unsigned long moved_swap;
417 418 419
	struct task_struct *moving_task;	/* a task moving charges */
	wait_queue_head_t waitq;		/* a waitq for other context */
} mc = {
420
	.lock = __SPIN_LOCK_UNLOCKED(mc.lock),
421 422
	.waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
};
423

D
Daisuke Nishimura 已提交
424 425
static bool move_anon(void)
{
426
	return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
D
Daisuke Nishimura 已提交
427 428
}

429 430
static bool move_file(void)
{
431
	return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
432 433
}

434 435 436 437
/*
 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
 * limit reclaim to prevent infinite loops, if they ever occur.
 */
438
#define	MEM_CGROUP_MAX_RECLAIM_LOOPS		100
439

440 441
enum charge_type {
	MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
442
	MEM_CGROUP_CHARGE_TYPE_ANON,
K
KAMEZAWA Hiroyuki 已提交
443
	MEM_CGROUP_CHARGE_TYPE_SWAPOUT,	/* for accounting swapcache */
K
KAMEZAWA Hiroyuki 已提交
444
	MEM_CGROUP_CHARGE_TYPE_DROP,	/* a page was unused swap cache */
445 446 447
	NR_CHARGE_TYPE,
};

448
/* for encoding cft->private value on file */
G
Glauber Costa 已提交
449 450 451 452
enum res_type {
	_MEM,
	_MEMSWAP,
	_OOM_TYPE,
453
	_KMEM,
G
Glauber Costa 已提交
454 455
};

456 457
#define MEMFILE_PRIVATE(x, val)	((x) << 16 | (val))
#define MEMFILE_TYPE(val)	((val) >> 16 & 0xffff)
458
#define MEMFILE_ATTR(val)	((val) & 0xffff)
K
KAMEZAWA Hiroyuki 已提交
459 460
/* Used for OOM nofiier */
#define OOM_CONTROL		(0)
461

462 463 464 465 466 467 468 469
/*
 * Reclaim flags for mem_cgroup_hierarchical_reclaim
 */
#define MEM_CGROUP_RECLAIM_NOSWAP_BIT	0x0
#define MEM_CGROUP_RECLAIM_NOSWAP	(1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
#define MEM_CGROUP_RECLAIM_SHRINK_BIT	0x1
#define MEM_CGROUP_RECLAIM_SHRINK	(1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)

470 471 472 473 474 475 476
/*
 * The memcg_create_mutex will be held whenever a new cgroup is created.
 * As a consequence, any change that needs to protect against new child cgroups
 * appearing has to hold it as well.
 */
static DEFINE_MUTEX(memcg_create_mutex);

477 478
struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
{
479
	return s ? container_of(s, struct mem_cgroup, css) : NULL;
480 481
}

482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
/* Some nice accessors for the vmpressure. */
struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
{
	if (!memcg)
		memcg = root_mem_cgroup;
	return &memcg->vmpressure;
}

struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
{
	return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
}

struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
{
	return &mem_cgroup_from_css(css)->vmpressure;
}

500 501 502 503 504
static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
{
	return (memcg == root_mem_cgroup);
}

G
Glauber Costa 已提交
505
/* Writing them here to avoid exposing memcg's inner layout */
M
Michal Hocko 已提交
506
#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
G
Glauber Costa 已提交
507 508 509

void sock_update_memcg(struct sock *sk)
{
510
	if (mem_cgroup_sockets_enabled) {
G
Glauber Costa 已提交
511
		struct mem_cgroup *memcg;
512
		struct cg_proto *cg_proto;
G
Glauber Costa 已提交
513 514 515

		BUG_ON(!sk->sk_prot->proto_cgroup);

516 517 518 519 520 521 522 523 524 525
		/* Socket cloning can throw us here with sk_cgrp already
		 * filled. It won't however, necessarily happen from
		 * process context. So the test for root memcg given
		 * the current task's memcg won't help us in this case.
		 *
		 * Respecting the original socket's memcg is a better
		 * decision in this case.
		 */
		if (sk->sk_cgrp) {
			BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
526
			css_get(&sk->sk_cgrp->memcg->css);
527 528 529
			return;
		}

G
Glauber Costa 已提交
530 531
		rcu_read_lock();
		memcg = mem_cgroup_from_task(current);
532
		cg_proto = sk->sk_prot->proto_cgroup(memcg);
533 534
		if (!mem_cgroup_is_root(memcg) &&
		    memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
535
			sk->sk_cgrp = cg_proto;
G
Glauber Costa 已提交
536 537 538 539 540 541 542 543
		}
		rcu_read_unlock();
	}
}
EXPORT_SYMBOL(sock_update_memcg);

void sock_release_memcg(struct sock *sk)
{
544
	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
G
Glauber Costa 已提交
545 546 547
		struct mem_cgroup *memcg;
		WARN_ON(!sk->sk_cgrp->memcg);
		memcg = sk->sk_cgrp->memcg;
548
		css_put(&sk->sk_cgrp->memcg->css);
G
Glauber Costa 已提交
549 550
	}
}
G
Glauber Costa 已提交
551 552 553 554 555 556 557 558 559

struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
{
	if (!memcg || mem_cgroup_is_root(memcg))
		return NULL;

	return &memcg->tcp_mem.cg_proto;
}
EXPORT_SYMBOL(tcp_proto_cgroup);
G
Glauber Costa 已提交
560

561 562 563 564 565 566 567 568 569 570 571 572
static void disarm_sock_keys(struct mem_cgroup *memcg)
{
	if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
		return;
	static_key_slow_dec(&memcg_socket_limit_enabled);
}
#else
static void disarm_sock_keys(struct mem_cgroup *memcg)
{
}
#endif

573
#ifdef CONFIG_MEMCG_KMEM
574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591
/*
 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
 * There are two main reasons for not using the css_id for this:
 *  1) this works better in sparse environments, where we have a lot of memcgs,
 *     but only a few kmem-limited. Or also, if we have, for instance, 200
 *     memcgs, and none but the 200th is kmem-limited, we'd have to have a
 *     200 entry array for that.
 *
 *  2) In order not to violate the cgroup API, we would like to do all memory
 *     allocation in ->create(). At that point, we haven't yet allocated the
 *     css_id. Having a separate index prevents us from messing with the cgroup
 *     core for this
 *
 * The current size of the caches array is stored in
 * memcg_limited_groups_array_size.  It will double each time we have to
 * increase it.
 */
static DEFINE_IDA(kmem_limited_groups);
592 593
int memcg_limited_groups_array_size;

594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
/*
 * MIN_SIZE is different than 1, because we would like to avoid going through
 * the alloc/free process all the time. In a small machine, 4 kmem-limited
 * cgroups is a reasonable guess. In the future, it could be a parameter or
 * tunable, but that is strictly not necessary.
 *
 * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get
 * this constant directly from cgroup, but it is understandable that this is
 * better kept as an internal representation in cgroup.c. In any case, the
 * css_id space is not getting any smaller, and we don't have to necessarily
 * increase ours as well if it increases.
 */
#define MEMCG_CACHES_MIN_SIZE 4
#define MEMCG_CACHES_MAX_SIZE 65535

609 610 611 612 613 614
/*
 * A lot of the calls to the cache allocation functions are expected to be
 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
 * conditional to this static branch, we'll have to allow modules that does
 * kmem_cache_alloc and the such to see this symbol as well
 */
615
struct static_key memcg_kmem_enabled_key;
616
EXPORT_SYMBOL(memcg_kmem_enabled_key);
617 618 619

static void disarm_kmem_keys(struct mem_cgroup *memcg)
{
620
	if (memcg_kmem_is_active(memcg)) {
621
		static_key_slow_dec(&memcg_kmem_enabled_key);
622 623
		ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
	}
624 625 626 627 628
	/*
	 * This check can't live in kmem destruction function,
	 * since the charges will outlive the cgroup
	 */
	WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
629 630 631 632 633 634 635 636 637 638 639 640 641
}
#else
static void disarm_kmem_keys(struct mem_cgroup *memcg)
{
}
#endif /* CONFIG_MEMCG_KMEM */

static void disarm_static_keys(struct mem_cgroup *memcg)
{
	disarm_sock_keys(memcg);
	disarm_kmem_keys(memcg);
}

642
static void drain_all_stock_async(struct mem_cgroup *memcg);
643

644
static struct mem_cgroup_per_zone *
645
mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
646
{
647
	VM_BUG_ON((unsigned)nid >= nr_node_ids);
648
	return &memcg->nodeinfo[nid]->zoneinfo[zid];
649 650
}

651
struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
652
{
653
	return &memcg->css;
654 655
}

656
static struct mem_cgroup_per_zone *
657
page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
658
{
659 660
	int nid = page_to_nid(page);
	int zid = page_zonenum(page);
661

662
	return mem_cgroup_zoneinfo(memcg, nid, zid);
663 664
}

665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
/*
 * Implementation Note: reading percpu statistics for memcg.
 *
 * Both of vmstat[] and percpu_counter has threshold and do periodic
 * synchronization to implement "quick" read. There are trade-off between
 * reading cost and precision of value. Then, we may have a chance to implement
 * a periodic synchronizion of counter in memcg's counter.
 *
 * But this _read() function is used for user interface now. The user accounts
 * memory usage by memory cgroup and he _always_ requires exact value because
 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
 * have to visit all online cpus and make sum. So, for now, unnecessary
 * synchronization is not implemented. (just implemented for cpu hotplug)
 *
 * If there are kernel internal actions which can make use of some not-exact
 * value, and reading all cpu value can be performance bottleneck in some
 * common workload, threashold and synchonization as vmstat[] should be
 * implemented.
 */
684
static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
685
				 enum mem_cgroup_stat_index idx)
686
{
687
	long val = 0;
688 689
	int cpu;

690 691
	get_online_cpus();
	for_each_online_cpu(cpu)
692
		val += per_cpu(memcg->stat->count[idx], cpu);
693
#ifdef CONFIG_HOTPLUG_CPU
694 695 696
	spin_lock(&memcg->pcp_counter_lock);
	val += memcg->nocpu_base.count[idx];
	spin_unlock(&memcg->pcp_counter_lock);
697 698
#endif
	put_online_cpus();
699 700 701
	return val;
}

702
static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
703 704 705
					 bool charge)
{
	int val = (charge) ? 1 : -1;
706
	this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
707 708
}

709
static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
710 711 712 713 714 715
					    enum mem_cgroup_events_index idx)
{
	unsigned long val = 0;
	int cpu;

	for_each_online_cpu(cpu)
716
		val += per_cpu(memcg->stat->events[idx], cpu);
717
#ifdef CONFIG_HOTPLUG_CPU
718 719 720
	spin_lock(&memcg->pcp_counter_lock);
	val += memcg->nocpu_base.events[idx];
	spin_unlock(&memcg->pcp_counter_lock);
721 722 723 724
#endif
	return val;
}

725
static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
726
					 struct page *page,
727
					 bool anon, int nr_pages)
728
{
729 730
	preempt_disable();

731 732 733 734 735 736
	/*
	 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
	 * counted as CACHE even if it's on ANON LRU.
	 */
	if (anon)
		__this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
737
				nr_pages);
738
	else
739
		__this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
740
				nr_pages);
741

742 743 744 745
	if (PageTransHuge(page))
		__this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
				nr_pages);

746 747
	/* pagein of a big page is an event. So, ignore page size */
	if (nr_pages > 0)
748
		__this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
749
	else {
750
		__this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
751 752
		nr_pages = -nr_pages; /* for event */
	}
753

754
	__this_cpu_add(memcg->stat->nr_page_events, nr_pages);
755

756
	preempt_enable();
757 758
}

759
unsigned long
760
mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
761 762 763 764 765 766 767 768
{
	struct mem_cgroup_per_zone *mz;

	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
	return mz->lru_size[lru];
}

static unsigned long
769
mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
770
			unsigned int lru_mask)
771 772
{
	struct mem_cgroup_per_zone *mz;
H
Hugh Dickins 已提交
773
	enum lru_list lru;
774 775
	unsigned long ret = 0;

776
	mz = mem_cgroup_zoneinfo(memcg, nid, zid);
777

H
Hugh Dickins 已提交
778 779 780
	for_each_lru(lru) {
		if (BIT(lru) & lru_mask)
			ret += mz->lru_size[lru];
781 782 783 784 785
	}
	return ret;
}

static unsigned long
786
mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
787 788
			int nid, unsigned int lru_mask)
{
789 790 791
	u64 total = 0;
	int zid;

792
	for (zid = 0; zid < MAX_NR_ZONES; zid++)
793 794
		total += mem_cgroup_zone_nr_lru_pages(memcg,
						nid, zid, lru_mask);
795

796 797
	return total;
}
798

799
static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
800
			unsigned int lru_mask)
801
{
802
	int nid;
803 804
	u64 total = 0;

805
	for_each_node_state(nid, N_MEMORY)
806
		total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
807
	return total;
808 809
}

810 811
static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
				       enum mem_cgroup_events_target target)
812 813 814
{
	unsigned long val, next;

815
	val = __this_cpu_read(memcg->stat->nr_page_events);
816
	next = __this_cpu_read(memcg->stat->targets[target]);
817
	/* from time_after() in jiffies.h */
818 819 820 821 822
	if ((long)next - (long)val < 0) {
		switch (target) {
		case MEM_CGROUP_TARGET_THRESH:
			next = val + THRESHOLDS_EVENTS_TARGET;
			break;
823 824 825
		case MEM_CGROUP_TARGET_SOFTLIMIT:
			next = val + SOFTLIMIT_EVENTS_TARGET;
			break;
826 827 828 829 830 831 832 833
		case MEM_CGROUP_TARGET_NUMAINFO:
			next = val + NUMAINFO_EVENTS_TARGET;
			break;
		default:
			break;
		}
		__this_cpu_write(memcg->stat->targets[target], next);
		return true;
834
	}
835
	return false;
836 837
}

838
/*
A
Andrew Morton 已提交
839
 * Called from rate-limited memcg_check_events when enough
840
 * MEM_CGROUP_TARGET_SOFTLIMIT events are accumulated and it makes sure
A
Andrew Morton 已提交
841
 * that all the parents up the hierarchy will be notified that this group
842 843
 * is in excess or that it is not in excess anymore. mmecg->soft_contributed
 * makes the transition a single action whenever the state flips from one to
A
Andrew Morton 已提交
844
 * the other.
845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867
 */
static void mem_cgroup_update_soft_limit(struct mem_cgroup *memcg)
{
	unsigned long long excess = res_counter_soft_limit_excess(&memcg->res);
	struct mem_cgroup *parent = memcg;
	int delta = 0;

	spin_lock(&memcg->soft_lock);
	if (excess) {
		if (!memcg->soft_contributed) {
			delta = 1;
			memcg->soft_contributed = true;
		}
	} else {
		if (memcg->soft_contributed) {
			delta = -1;
			memcg->soft_contributed = false;
		}
	}

	/*
	 * Necessary to update all ancestors when hierarchy is used
	 * because their event counter is not touched.
868 869 870 871
	 * We track children even outside the hierarchy for the root
	 * cgroup because tree walk starting at root should visit
	 * all cgroups and we want to prevent from pointless tree
	 * walk if no children is below the limit.
872 873 874
	 */
	while (delta && (parent = parent_mem_cgroup(parent)))
		atomic_add(delta, &parent->children_in_excess);
875 876
	if (memcg != root_mem_cgroup && !root_mem_cgroup->use_hierarchy)
		atomic_add(delta, &root_mem_cgroup->children_in_excess);
877 878 879
	spin_unlock(&memcg->soft_lock);
}

880 881 882 883
/*
 * Check events in order.
 *
 */
884
static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
885
{
886
	preempt_disable();
887
	/* threshold event is triggered in finer grain than soft limit */
888 889
	if (unlikely(mem_cgroup_event_ratelimit(memcg,
						MEM_CGROUP_TARGET_THRESH))) {
890
		bool do_softlimit;
891
		bool do_numainfo __maybe_unused;
892

893 894
		do_softlimit = mem_cgroup_event_ratelimit(memcg,
						MEM_CGROUP_TARGET_SOFTLIMIT);
895 896 897 898 899 900
#if MAX_NUMNODES > 1
		do_numainfo = mem_cgroup_event_ratelimit(memcg,
						MEM_CGROUP_TARGET_NUMAINFO);
#endif
		preempt_enable();

901
		mem_cgroup_threshold(memcg);
902 903
		if (unlikely(do_softlimit))
			mem_cgroup_update_soft_limit(memcg);
904
#if MAX_NUMNODES > 1
905
		if (unlikely(do_numainfo))
906
			atomic_inc(&memcg->numainfo_events);
907
#endif
908 909
	} else
		preempt_enable();
910 911
}

912
struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
913
{
914 915 916 917 918 919 920 921
	/*
	 * mm_update_next_owner() may clear mm->owner to NULL
	 * if it races with swapoff, page migration, etc.
	 * So this can be called with p == NULL.
	 */
	if (unlikely(!p))
		return NULL;

922
	return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id));
923 924
}

925
struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
926
{
927
	struct mem_cgroup *memcg = NULL;
928 929 930

	if (!mm)
		return NULL;
931 932 933 934 935 936 937
	/*
	 * Because we have no locks, mm->owner's may be being moved to other
	 * cgroup. We use css_tryget() here even if this looks
	 * pessimistic (rather than adding locks here).
	 */
	rcu_read_lock();
	do {
938 939
		memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
		if (unlikely(!memcg))
940
			break;
941
	} while (!css_tryget(&memcg->css));
942
	rcu_read_unlock();
943
	return memcg;
944 945
}

946 947 948 949 950 951 952 953 954
static enum mem_cgroup_filter_t
mem_cgroup_filter(struct mem_cgroup *memcg, struct mem_cgroup *root,
		mem_cgroup_iter_filter cond)
{
	if (!cond)
		return VISIT;
	return cond(memcg, root);
}

955 956 957 958 959 960 961
/*
 * Returns a next (in a pre-order walk) alive memcg (with elevated css
 * ref. count) or NULL if the whole root's subtree has been visited.
 *
 * helper function to be used by mem_cgroup_iter
 */
static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
962
		struct mem_cgroup *last_visited, mem_cgroup_iter_filter cond)
963
{
964
	struct cgroup_subsys_state *prev_css, *next_css;
965

966
	prev_css = last_visited ? &last_visited->css : NULL;
967
skip_node:
968
	next_css = css_next_descendant_pre(prev_css, &root->css);
969 970 971 972 973 974 975 976

	/*
	 * Even if we found a group we have to make sure it is
	 * alive. css && !memcg means that the groups should be
	 * skipped and we should continue the tree walk.
	 * last_visited css is safe to use because it is
	 * protected by css_get and the tree walk is rcu safe.
	 */
977 978 979
	if (next_css) {
		struct mem_cgroup *mem = mem_cgroup_from_css(next_css);

980 981
		switch (mem_cgroup_filter(mem, root, cond)) {
		case SKIP:
982
			prev_css = next_css;
983
			goto skip_node;
984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004
		case SKIP_TREE:
			if (mem == root)
				return NULL;
			/*
			 * css_rightmost_descendant is not an optimal way to
			 * skip through a subtree (especially for imbalanced
			 * trees leaning to right) but that's what we have right
			 * now. More effective solution would be traversing
			 * right-up for first non-NULL without calling
			 * css_next_descendant_pre afterwards.
			 */
			prev_css = css_rightmost_descendant(next_css);
			goto skip_node;
		case VISIT:
			if (css_tryget(&mem->css))
				return mem;
			else {
				prev_css = next_css;
				goto skip_node;
			}
			break;
1005 1006 1007 1008 1009 1010
		}
	}

	return NULL;
}

1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
{
	/*
	 * When a group in the hierarchy below root is destroyed, the
	 * hierarchy iterator can no longer be trusted since it might
	 * have pointed to the destroyed group.  Invalidate it.
	 */
	atomic_inc(&root->dead_count);
}

static struct mem_cgroup *
mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
		     struct mem_cgroup *root,
		     int *sequence)
{
	struct mem_cgroup *position = NULL;
	/*
	 * A cgroup destruction happens in two stages: offlining and
	 * release.  They are separated by a RCU grace period.
	 *
	 * If the iterator is valid, we may still race with an
	 * offlining.  The RCU lock ensures the object won't be
	 * released, tryget will fail if we lost the race.
	 */
	*sequence = atomic_read(&root->dead_count);
	if (iter->last_dead_count == *sequence) {
		smp_rmb();
		position = iter->last_visited;
		if (position && !css_tryget(&position->css))
			position = NULL;
	}
	return position;
}

static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
				   struct mem_cgroup *last_visited,
				   struct mem_cgroup *new_position,
				   int sequence)
{
	if (last_visited)
		css_put(&last_visited->css);
	/*
	 * We store the sequence count from the time @last_visited was
	 * loaded successfully instead of rereading it here so that we
	 * don't lose destruction events in between.  We could have
	 * raced with the destruction of @new_position after all.
	 */
	iter->last_visited = new_position;
	smp_wmb();
	iter->last_dead_count = sequence;
}

1063 1064 1065 1066 1067
/**
 * mem_cgroup_iter - iterate over memory cgroup hierarchy
 * @root: hierarchy root
 * @prev: previously returned memcg, NULL on first invocation
 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1068
 * @cond: filter for visited nodes, NULL for no filter
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
 *
 * Returns references to children of the hierarchy below @root, or
 * @root itself, or %NULL after a full round-trip.
 *
 * Caller must pass the return value in @prev on subsequent
 * invocations for reference counting, or use mem_cgroup_iter_break()
 * to cancel a hierarchy walk before the round-trip is complete.
 *
 * Reclaimers can specify a zone and a priority level in @reclaim to
 * divide up the memcgs in the hierarchy among all concurrent
 * reclaimers operating on the same zone and priority.
 */
1081
struct mem_cgroup *mem_cgroup_iter_cond(struct mem_cgroup *root,
1082
				   struct mem_cgroup *prev,
1083 1084
				   struct mem_cgroup_reclaim_cookie *reclaim,
				   mem_cgroup_iter_filter cond)
K
KAMEZAWA Hiroyuki 已提交
1085
{
1086
	struct mem_cgroup *memcg = NULL;
1087
	struct mem_cgroup *last_visited = NULL;
1088

1089 1090 1091 1092
	if (mem_cgroup_disabled()) {
		/* first call must return non-NULL, second return NULL */
		return (struct mem_cgroup *)(unsigned long)!prev;
	}
1093

1094 1095
	if (!root)
		root = root_mem_cgroup;
K
KAMEZAWA Hiroyuki 已提交
1096

1097
	if (prev && !reclaim)
1098
		last_visited = prev;
K
KAMEZAWA Hiroyuki 已提交
1099

1100 1101
	if (!root->use_hierarchy && root != root_mem_cgroup) {
		if (prev)
1102
			goto out_css_put;
1103 1104 1105
		if (mem_cgroup_filter(root, root, cond) == VISIT)
			return root;
		return NULL;
1106
	}
K
KAMEZAWA Hiroyuki 已提交
1107

1108
	rcu_read_lock();
1109
	while (!memcg) {
1110
		struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
1111
		int uninitialized_var(seq);
1112

1113 1114 1115 1116 1117 1118 1119
		if (reclaim) {
			int nid = zone_to_nid(reclaim->zone);
			int zid = zone_idx(reclaim->zone);
			struct mem_cgroup_per_zone *mz;

			mz = mem_cgroup_zoneinfo(root, nid, zid);
			iter = &mz->reclaim_iter[reclaim->priority];
1120
			if (prev && reclaim->generation != iter->generation) {
M
Michal Hocko 已提交
1121
				iter->last_visited = NULL;
1122 1123
				goto out_unlock;
			}
M
Michal Hocko 已提交
1124

1125
			last_visited = mem_cgroup_iter_load(iter, root, &seq);
1126
		}
K
KAMEZAWA Hiroyuki 已提交
1127

1128
		memcg = __mem_cgroup_iter_next(root, last_visited, cond);
K
KAMEZAWA Hiroyuki 已提交
1129

1130
		if (reclaim) {
1131
			mem_cgroup_iter_update(iter, last_visited, memcg, seq);
1132

M
Michal Hocko 已提交
1133
			if (!memcg)
1134 1135 1136 1137
				iter->generation++;
			else if (!prev && memcg)
				reclaim->generation = iter->generation;
		}
1138

1139 1140 1141 1142 1143
		/*
		 * We have finished the whole tree walk or no group has been
		 * visited because filter told us to skip the root node.
		 */
		if (!memcg && (prev || (cond && !last_visited)))
1144
			goto out_unlock;
1145
	}
1146 1147
out_unlock:
	rcu_read_unlock();
1148 1149 1150 1151
out_css_put:
	if (prev && prev != root)
		css_put(&prev->css);

1152
	return memcg;
K
KAMEZAWA Hiroyuki 已提交
1153
}
K
KAMEZAWA Hiroyuki 已提交
1154

1155 1156 1157 1158 1159 1160 1161
/**
 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
 * @root: hierarchy root
 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
 */
void mem_cgroup_iter_break(struct mem_cgroup *root,
			   struct mem_cgroup *prev)
1162 1163 1164 1165 1166 1167
{
	if (!root)
		root = root_mem_cgroup;
	if (prev && prev != root)
		css_put(&prev->css);
}
K
KAMEZAWA Hiroyuki 已提交
1168

1169 1170 1171 1172 1173 1174
/*
 * Iteration constructs for visiting all cgroups (under a tree).  If
 * loops are exited prematurely (break), mem_cgroup_iter_break() must
 * be used for reference counting.
 */
#define for_each_mem_cgroup_tree(iter, root)		\
1175
	for (iter = mem_cgroup_iter(root, NULL, NULL);	\
1176
	     iter != NULL;				\
1177
	     iter = mem_cgroup_iter(root, iter, NULL))
1178

1179
#define for_each_mem_cgroup(iter)			\
1180
	for (iter = mem_cgroup_iter(NULL, NULL, NULL);	\
1181
	     iter != NULL;				\
1182
	     iter = mem_cgroup_iter(NULL, iter, NULL))
K
KAMEZAWA Hiroyuki 已提交
1183

1184
void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1185
{
1186
	struct mem_cgroup *memcg;
1187 1188

	rcu_read_lock();
1189 1190
	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
	if (unlikely(!memcg))
1191 1192 1193 1194
		goto out;

	switch (idx) {
	case PGFAULT:
1195 1196 1197 1198
		this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
		break;
	case PGMAJFAULT:
		this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1199 1200 1201 1202 1203 1204 1205
		break;
	default:
		BUG();
	}
out:
	rcu_read_unlock();
}
1206
EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
1207

1208 1209 1210
/**
 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
 * @zone: zone of the wanted lruvec
1211
 * @memcg: memcg of the wanted lruvec
1212 1213 1214 1215 1216 1217 1218 1219 1220
 *
 * Returns the lru list vector holding pages for the given @zone and
 * @mem.  This can be the global zone lruvec, if the memory controller
 * is disabled.
 */
struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
				      struct mem_cgroup *memcg)
{
	struct mem_cgroup_per_zone *mz;
1221
	struct lruvec *lruvec;
1222

1223 1224 1225 1226
	if (mem_cgroup_disabled()) {
		lruvec = &zone->lruvec;
		goto out;
	}
1227 1228

	mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
	lruvec = &mz->lruvec;
out:
	/*
	 * Since a node can be onlined after the mem_cgroup was created,
	 * we have to be prepared to initialize lruvec->zone here;
	 * and if offlined then reonlined, we need to reinitialize it.
	 */
	if (unlikely(lruvec->zone != zone))
		lruvec->zone = zone;
	return lruvec;
1239 1240
}

K
KAMEZAWA Hiroyuki 已提交
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253
/*
 * Following LRU functions are allowed to be used without PCG_LOCK.
 * Operations are called by routine of global LRU independently from memcg.
 * What we have to take care of here is validness of pc->mem_cgroup.
 *
 * Changes to pc->mem_cgroup happens when
 * 1. charge
 * 2. moving account
 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
 * It is added to LRU before charge.
 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
 * When moving account, the page is not on LRU. It's isolated.
 */
1254

1255
/**
1256
 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1257
 * @page: the page
1258
 * @zone: zone of the page
1259
 */
1260
struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
K
KAMEZAWA Hiroyuki 已提交
1261 1262
{
	struct mem_cgroup_per_zone *mz;
1263 1264
	struct mem_cgroup *memcg;
	struct page_cgroup *pc;
1265
	struct lruvec *lruvec;
1266

1267 1268 1269 1270
	if (mem_cgroup_disabled()) {
		lruvec = &zone->lruvec;
		goto out;
	}
1271

K
KAMEZAWA Hiroyuki 已提交
1272
	pc = lookup_page_cgroup(page);
1273
	memcg = pc->mem_cgroup;
1274 1275

	/*
1276
	 * Surreptitiously switch any uncharged offlist page to root:
1277 1278 1279 1280 1281 1282 1283
	 * an uncharged page off lru does nothing to secure
	 * its former mem_cgroup from sudden removal.
	 *
	 * Our caller holds lru_lock, and PageCgroupUsed is updated
	 * under page_cgroup lock: between them, they make all uses
	 * of pc->mem_cgroup safe.
	 */
1284
	if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1285 1286
		pc->mem_cgroup = memcg = root_mem_cgroup;

1287
	mz = page_cgroup_zoneinfo(memcg, page);
1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
	lruvec = &mz->lruvec;
out:
	/*
	 * Since a node can be onlined after the mem_cgroup was created,
	 * we have to be prepared to initialize lruvec->zone here;
	 * and if offlined then reonlined, we need to reinitialize it.
	 */
	if (unlikely(lruvec->zone != zone))
		lruvec->zone = zone;
	return lruvec;
K
KAMEZAWA Hiroyuki 已提交
1298
}
1299

1300
/**
1301 1302 1303 1304
 * mem_cgroup_update_lru_size - account for adding or removing an lru page
 * @lruvec: mem_cgroup per zone lru vector
 * @lru: index of lru list the page is sitting on
 * @nr_pages: positive when adding or negative when removing
1305
 *
1306 1307
 * This function must be called when a page is added to or removed from an
 * lru list.
1308
 */
1309 1310
void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
				int nr_pages)
1311 1312
{
	struct mem_cgroup_per_zone *mz;
1313
	unsigned long *lru_size;
1314 1315 1316 1317

	if (mem_cgroup_disabled())
		return;

1318 1319 1320 1321
	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
	lru_size = mz->lru_size + lru;
	*lru_size += nr_pages;
	VM_BUG_ON((long)(*lru_size) < 0);
K
KAMEZAWA Hiroyuki 已提交
1322
}
1323

1324
/*
1325
 * Checks whether given mem is same or in the root_mem_cgroup's
1326 1327
 * hierarchy subtree
 */
1328 1329
bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
				  struct mem_cgroup *memcg)
1330
{
1331 1332
	if (root_memcg == memcg)
		return true;
1333
	if (!root_memcg->use_hierarchy || !memcg)
1334
		return false;
1335 1336 1337 1338 1339 1340 1341 1342
	return css_is_ancestor(&memcg->css, &root_memcg->css);
}

static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
				       struct mem_cgroup *memcg)
{
	bool ret;

1343
	rcu_read_lock();
1344
	ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
1345 1346
	rcu_read_unlock();
	return ret;
1347 1348
}

1349 1350
bool task_in_mem_cgroup(struct task_struct *task,
			const struct mem_cgroup *memcg)
1351
{
1352
	struct mem_cgroup *curr = NULL;
1353
	struct task_struct *p;
1354
	bool ret;
1355

1356
	p = find_lock_task_mm(task);
1357 1358 1359 1360 1361 1362 1363 1364 1365
	if (p) {
		curr = try_get_mem_cgroup_from_mm(p->mm);
		task_unlock(p);
	} else {
		/*
		 * All threads may have already detached their mm's, but the oom
		 * killer still needs to detect if they have already been oom
		 * killed to prevent needlessly killing additional tasks.
		 */
1366
		rcu_read_lock();
1367 1368 1369
		curr = mem_cgroup_from_task(task);
		if (curr)
			css_get(&curr->css);
1370
		rcu_read_unlock();
1371
	}
1372
	if (!curr)
1373
		return false;
1374
	/*
1375
	 * We should check use_hierarchy of "memcg" not "curr". Because checking
1376
	 * use_hierarchy of "curr" here make this function true if hierarchy is
1377 1378
	 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
	 * hierarchy(even if use_hierarchy is disabled in "memcg").
1379
	 */
1380
	ret = mem_cgroup_same_or_subtree(memcg, curr);
1381
	css_put(&curr->css);
1382 1383 1384
	return ret;
}

1385
int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1386
{
1387
	unsigned long inactive_ratio;
1388
	unsigned long inactive;
1389
	unsigned long active;
1390
	unsigned long gb;
1391

1392 1393
	inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
	active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1394

1395 1396 1397 1398 1399 1400
	gb = (inactive + active) >> (30 - PAGE_SHIFT);
	if (gb)
		inactive_ratio = int_sqrt(10 * gb);
	else
		inactive_ratio = 1;

1401
	return inactive * inactive_ratio < active;
1402 1403
}

1404 1405 1406
#define mem_cgroup_from_res_counter(counter, member)	\
	container_of(counter, struct mem_cgroup, member)

1407
/**
1408
 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
W
Wanpeng Li 已提交
1409
 * @memcg: the memory cgroup
1410
 *
1411
 * Returns the maximum amount of memory @mem can be charged with, in
1412
 * pages.
1413
 */
1414
static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1415
{
1416 1417
	unsigned long long margin;

1418
	margin = res_counter_margin(&memcg->res);
1419
	if (do_swap_account)
1420
		margin = min(margin, res_counter_margin(&memcg->memsw));
1421
	return margin >> PAGE_SHIFT;
1422 1423
}

1424
int mem_cgroup_swappiness(struct mem_cgroup *memcg)
K
KOSAKI Motohiro 已提交
1425 1426
{
	/* root ? */
T
Tejun Heo 已提交
1427
	if (!css_parent(&memcg->css))
K
KOSAKI Motohiro 已提交
1428 1429
		return vm_swappiness;

1430
	return memcg->swappiness;
K
KOSAKI Motohiro 已提交
1431 1432
}

1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446
/*
 * memcg->moving_account is used for checking possibility that some thread is
 * calling move_account(). When a thread on CPU-A starts moving pages under
 * a memcg, other threads should check memcg->moving_account under
 * rcu_read_lock(), like this:
 *
 *         CPU-A                                    CPU-B
 *                                              rcu_read_lock()
 *         memcg->moving_account+1              if (memcg->mocing_account)
 *                                                   take heavy locks.
 *         synchronize_rcu()                    update something.
 *                                              rcu_read_unlock()
 *         start move here.
 */
1447 1448 1449 1450

/* for quick checking without looking up memcg */
atomic_t memcg_moving __read_mostly;

1451
static void mem_cgroup_start_move(struct mem_cgroup *memcg)
1452
{
1453
	atomic_inc(&memcg_moving);
1454
	atomic_inc(&memcg->moving_account);
1455 1456 1457
	synchronize_rcu();
}

1458
static void mem_cgroup_end_move(struct mem_cgroup *memcg)
1459
{
1460 1461 1462 1463
	/*
	 * Now, mem_cgroup_clear_mc() may call this function with NULL.
	 * We check NULL in callee rather than caller.
	 */
1464 1465
	if (memcg) {
		atomic_dec(&memcg_moving);
1466
		atomic_dec(&memcg->moving_account);
1467
	}
1468
}
1469

1470 1471 1472
/*
 * 2 routines for checking "mem" is under move_account() or not.
 *
1473 1474
 * mem_cgroup_stolen() -  checking whether a cgroup is mc.from or not. This
 *			  is used for avoiding races in accounting.  If true,
1475 1476 1477 1478 1479 1480 1481
 *			  pc->mem_cgroup may be overwritten.
 *
 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
 *			  under hierarchy of moving cgroups. This is for
 *			  waiting at hith-memory prressure caused by "move".
 */

1482
static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
1483 1484
{
	VM_BUG_ON(!rcu_read_lock_held());
1485
	return atomic_read(&memcg->moving_account) > 0;
1486
}
1487

1488
static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1489
{
1490 1491
	struct mem_cgroup *from;
	struct mem_cgroup *to;
1492
	bool ret = false;
1493 1494 1495 1496 1497 1498 1499 1500 1501
	/*
	 * Unlike task_move routines, we access mc.to, mc.from not under
	 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
	 */
	spin_lock(&mc.lock);
	from = mc.from;
	to = mc.to;
	if (!from)
		goto unlock;
1502

1503 1504
	ret = mem_cgroup_same_or_subtree(memcg, from)
		|| mem_cgroup_same_or_subtree(memcg, to);
1505 1506
unlock:
	spin_unlock(&mc.lock);
1507 1508 1509
	return ret;
}

1510
static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1511 1512
{
	if (mc.moving_task && current != mc.moving_task) {
1513
		if (mem_cgroup_under_move(memcg)) {
1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525
			DEFINE_WAIT(wait);
			prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
			/* moving charge context might have finished. */
			if (mc.moving_task)
				schedule();
			finish_wait(&mc.waitq, &wait);
			return true;
		}
	}
	return false;
}

1526 1527 1528 1529
/*
 * Take this lock when
 * - a code tries to modify page's memcg while it's USED.
 * - a code tries to modify page state accounting in a memcg.
1530
 * see mem_cgroup_stolen(), too.
1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543
 */
static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
				  unsigned long *flags)
{
	spin_lock_irqsave(&memcg->move_lock, *flags);
}

static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
				unsigned long *flags)
{
	spin_unlock_irqrestore(&memcg->move_lock, *flags);
}

1544
#define K(x) ((x) << (PAGE_SHIFT-10))
1545
/**
1546
 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563
 * @memcg: The memory cgroup that went over limit
 * @p: Task that is going to be killed
 *
 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
 * enabled
 */
void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
{
	struct cgroup *task_cgrp;
	struct cgroup *mem_cgrp;
	/*
	 * Need a buffer in BSS, can't rely on allocations. The code relies
	 * on the assumption that OOM is serialized for memory controller.
	 * If this assumption is broken, revisit this code.
	 */
	static char memcg_name[PATH_MAX];
	int ret;
1564 1565
	struct mem_cgroup *iter;
	unsigned int i;
1566

1567
	if (!p)
1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
		return;

	rcu_read_lock();

	mem_cgrp = memcg->css.cgroup;
	task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);

	ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
	if (ret < 0) {
		/*
		 * Unfortunately, we are unable to convert to a useful name
		 * But we'll still print out the usage information
		 */
		rcu_read_unlock();
		goto done;
	}
	rcu_read_unlock();

1586
	pr_info("Task in %s killed", memcg_name);
1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598

	rcu_read_lock();
	ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
	if (ret < 0) {
		rcu_read_unlock();
		goto done;
	}
	rcu_read_unlock();

	/*
	 * Continues from above, so we don't need an KERN_ level
	 */
1599
	pr_cont(" as a result of limit of %s\n", memcg_name);
1600 1601
done:

1602
	pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
1603 1604 1605
		res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
		res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
		res_counter_read_u64(&memcg->res, RES_FAILCNT));
1606
	pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
1607 1608 1609
		res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
		res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
		res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1610
	pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
1611 1612 1613
		res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
		res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
		res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637

	for_each_mem_cgroup_tree(iter, memcg) {
		pr_info("Memory cgroup stats");

		rcu_read_lock();
		ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
		if (!ret)
			pr_cont(" for %s", memcg_name);
		rcu_read_unlock();
		pr_cont(":");

		for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
			if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
				continue;
			pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
				K(mem_cgroup_read_stat(iter, i)));
		}

		for (i = 0; i < NR_LRU_LISTS; i++)
			pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
				K(mem_cgroup_nr_lru_pages(iter, BIT(i))));

		pr_cont("\n");
	}
1638 1639
}

1640 1641 1642 1643
/*
 * This function returns the number of memcg under hierarchy tree. Returns
 * 1(self count) if no children.
 */
1644
static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1645 1646
{
	int num = 0;
K
KAMEZAWA Hiroyuki 已提交
1647 1648
	struct mem_cgroup *iter;

1649
	for_each_mem_cgroup_tree(iter, memcg)
K
KAMEZAWA Hiroyuki 已提交
1650
		num++;
1651 1652 1653
	return num;
}

D
David Rientjes 已提交
1654 1655 1656
/*
 * Return the memory (and swap, if configured) limit for a memcg.
 */
1657
static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
D
David Rientjes 已提交
1658 1659 1660
{
	u64 limit;

1661 1662
	limit = res_counter_read_u64(&memcg->res, RES_LIMIT);

D
David Rientjes 已提交
1663
	/*
1664
	 * Do not consider swap space if we cannot swap due to swappiness
D
David Rientjes 已提交
1665
	 */
1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679
	if (mem_cgroup_swappiness(memcg)) {
		u64 memsw;

		limit += total_swap_pages << PAGE_SHIFT;
		memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);

		/*
		 * If memsw is finite and limits the amount of swap space
		 * available to this memcg, return that limit.
		 */
		limit = min(limit, memsw);
	}

	return limit;
D
David Rientjes 已提交
1680 1681
}

1682 1683
static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
				     int order)
1684 1685 1686 1687 1688 1689 1690
{
	struct mem_cgroup *iter;
	unsigned long chosen_points = 0;
	unsigned long totalpages;
	unsigned int points = 0;
	struct task_struct *chosen = NULL;

1691
	/*
1692 1693 1694
	 * If current has a pending SIGKILL or is exiting, then automatically
	 * select it.  The goal is to allow it to allocate so that it may
	 * quickly exit and free its memory.
1695
	 */
1696
	if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
1697 1698 1699 1700 1701
		set_thread_flag(TIF_MEMDIE);
		return;
	}

	check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
1702 1703
	totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
	for_each_mem_cgroup_tree(iter, memcg) {
1704
		struct css_task_iter it;
1705 1706
		struct task_struct *task;

1707 1708
		css_task_iter_start(&iter->css, &it);
		while ((task = css_task_iter_next(&it))) {
1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720
			switch (oom_scan_process_thread(task, totalpages, NULL,
							false)) {
			case OOM_SCAN_SELECT:
				if (chosen)
					put_task_struct(chosen);
				chosen = task;
				chosen_points = ULONG_MAX;
				get_task_struct(chosen);
				/* fall through */
			case OOM_SCAN_CONTINUE:
				continue;
			case OOM_SCAN_ABORT:
1721
				css_task_iter_end(&it);
1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737
				mem_cgroup_iter_break(memcg, iter);
				if (chosen)
					put_task_struct(chosen);
				return;
			case OOM_SCAN_OK:
				break;
			};
			points = oom_badness(task, memcg, NULL, totalpages);
			if (points > chosen_points) {
				if (chosen)
					put_task_struct(chosen);
				chosen = task;
				chosen_points = points;
				get_task_struct(chosen);
			}
		}
1738
		css_task_iter_end(&it);
1739 1740 1741 1742 1743 1744 1745 1746 1747
	}

	if (!chosen)
		return;
	points = chosen_points * 1000 / totalpages;
	oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
			 NULL, "Memory cgroup out of memory");
}

1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783
static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
					gfp_t gfp_mask,
					unsigned long flags)
{
	unsigned long total = 0;
	bool noswap = false;
	int loop;

	if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
		noswap = true;
	if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
		noswap = true;

	for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
		if (loop)
			drain_all_stock_async(memcg);
		total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
		/*
		 * Allow limit shrinkers, which are triggered directly
		 * by userspace, to catch signals and stop reclaim
		 * after minimal progress, regardless of the margin.
		 */
		if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
			break;
		if (mem_cgroup_margin(memcg))
			break;
		/*
		 * If nothing was reclaimed after two attempts, there
		 * may be no reclaimable pages in this hierarchy.
		 */
		if (loop && !total)
			break;
	}
	return total;
}

1784
#if MAX_NUMNODES > 1
1785 1786
/**
 * test_mem_cgroup_node_reclaimable
W
Wanpeng Li 已提交
1787
 * @memcg: the target memcg
1788 1789 1790 1791 1792 1793 1794
 * @nid: the node ID to be checked.
 * @noswap : specify true here if the user wants flle only information.
 *
 * This function returns whether the specified memcg contains any
 * reclaimable pages on a node. Returns true if there are any reclaimable
 * pages in the node.
 */
1795
static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1796 1797
		int nid, bool noswap)
{
1798
	if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1799 1800 1801
		return true;
	if (noswap || !total_swap_pages)
		return false;
1802
	if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1803 1804 1805 1806
		return true;
	return false;

}
1807 1808 1809 1810 1811 1812 1813

/*
 * Always updating the nodemask is not very good - even if we have an empty
 * list or the wrong list here, we can start from some node and traverse all
 * nodes based on the zonelist. So update the list loosely once per 10 secs.
 *
 */
1814
static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1815 1816
{
	int nid;
1817 1818 1819 1820
	/*
	 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
	 * pagein/pageout changes since the last update.
	 */
1821
	if (!atomic_read(&memcg->numainfo_events))
1822
		return;
1823
	if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1824 1825 1826
		return;

	/* make a nodemask where this memcg uses memory from */
1827
	memcg->scan_nodes = node_states[N_MEMORY];
1828

1829
	for_each_node_mask(nid, node_states[N_MEMORY]) {
1830

1831 1832
		if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
			node_clear(nid, memcg->scan_nodes);
1833
	}
1834

1835 1836
	atomic_set(&memcg->numainfo_events, 0);
	atomic_set(&memcg->numainfo_updating, 0);
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850
}

/*
 * Selecting a node where we start reclaim from. Because what we need is just
 * reducing usage counter, start from anywhere is O,K. Considering
 * memory reclaim from current node, there are pros. and cons.
 *
 * Freeing memory from current node means freeing memory from a node which
 * we'll use or we've used. So, it may make LRU bad. And if several threads
 * hit limits, it will see a contention on a node. But freeing from remote
 * node means more costs for memory reclaim because of memory latency.
 *
 * Now, we use round-robin. Better algorithm is welcomed.
 */
1851
int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1852 1853 1854
{
	int node;

1855 1856
	mem_cgroup_may_update_nodemask(memcg);
	node = memcg->last_scanned_node;
1857

1858
	node = next_node(node, memcg->scan_nodes);
1859
	if (node == MAX_NUMNODES)
1860
		node = first_node(memcg->scan_nodes);
1861 1862 1863 1864 1865 1866 1867 1868 1869
	/*
	 * We call this when we hit limit, not when pages are added to LRU.
	 * No LRU may hold pages because all pages are UNEVICTABLE or
	 * memcg is too small and all pages are not on LRU. In that case,
	 * we use curret node.
	 */
	if (unlikely(node == MAX_NUMNODES))
		node = numa_node_id();

1870
	memcg->last_scanned_node = node;
1871 1872 1873 1874
	return node;
}

#else
1875
int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1876 1877 1878
{
	return 0;
}
1879

1880 1881
#endif

1882
/*
1883 1884
 * A group is eligible for the soft limit reclaim under the given root
 * hierarchy if
A
Andrew Morton 已提交
1885 1886
 *	a) it is over its soft limit
 *	b) any parent up the hierarchy is over its soft limit
1887 1888 1889
 *
 * If the given group doesn't have any children over the limit then it
 * doesn't make any sense to iterate its subtree.
1890
 */
1891 1892
enum mem_cgroup_filter_t
mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg,
1893
		struct mem_cgroup *root)
1894
{
1895 1896 1897 1898 1899
	struct mem_cgroup *parent;

	if (!memcg)
		memcg = root_mem_cgroup;
	parent = memcg;
1900 1901

	if (res_counter_soft_limit_excess(&memcg->res))
1902
		return VISIT;
1903 1904

	/*
1905 1906
	 * If any parent up to the root in the hierarchy is over its soft limit
	 * then we have to obey and reclaim from this group as well.
1907
	 */
A
Andrew Morton 已提交
1908
	while ((parent = parent_mem_cgroup(parent))) {
1909
		if (res_counter_soft_limit_excess(&parent->res))
1910
			return VISIT;
1911 1912
		if (parent == root)
			break;
1913
	}
1914

1915 1916
	if (!atomic_read(&memcg->children_in_excess))
		return SKIP_TREE;
1917
	return SKIP;
1918 1919
}

1920 1921
static DEFINE_SPINLOCK(memcg_oom_lock);

K
KAMEZAWA Hiroyuki 已提交
1922 1923 1924 1925
/*
 * Check OOM-Killer is already running under our hierarchy.
 * If someone is running, return false.
 */
1926
static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
K
KAMEZAWA Hiroyuki 已提交
1927
{
1928
	struct mem_cgroup *iter, *failed = NULL;
1929

1930 1931
	spin_lock(&memcg_oom_lock);

1932
	for_each_mem_cgroup_tree(iter, memcg) {
1933
		if (iter->oom_lock) {
1934 1935 1936 1937 1938
			/*
			 * this subtree of our hierarchy is already locked
			 * so we cannot give a lock.
			 */
			failed = iter;
1939 1940
			mem_cgroup_iter_break(memcg, iter);
			break;
1941 1942
		} else
			iter->oom_lock = true;
K
KAMEZAWA Hiroyuki 已提交
1943
	}
K
KAMEZAWA Hiroyuki 已提交
1944

1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955
	if (failed) {
		/*
		 * OK, we failed to lock the whole subtree so we have
		 * to clean up what we set up to the failing subtree
		 */
		for_each_mem_cgroup_tree(iter, memcg) {
			if (iter == failed) {
				mem_cgroup_iter_break(memcg, iter);
				break;
			}
			iter->oom_lock = false;
1956 1957
		}
	}
1958 1959 1960 1961

	spin_unlock(&memcg_oom_lock);

	return !failed;
1962
}
1963

1964
static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1965
{
K
KAMEZAWA Hiroyuki 已提交
1966 1967
	struct mem_cgroup *iter;

1968
	spin_lock(&memcg_oom_lock);
1969
	for_each_mem_cgroup_tree(iter, memcg)
1970
		iter->oom_lock = false;
1971
	spin_unlock(&memcg_oom_lock);
1972 1973
}

1974
static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1975 1976 1977
{
	struct mem_cgroup *iter;

1978
	for_each_mem_cgroup_tree(iter, memcg)
1979 1980 1981
		atomic_inc(&iter->under_oom);
}

1982
static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1983 1984 1985
{
	struct mem_cgroup *iter;

K
KAMEZAWA Hiroyuki 已提交
1986 1987 1988 1989 1990
	/*
	 * When a new child is created while the hierarchy is under oom,
	 * mem_cgroup_oom_lock() may not be called. We have to use
	 * atomic_add_unless() here.
	 */
1991
	for_each_mem_cgroup_tree(iter, memcg)
1992
		atomic_add_unless(&iter->under_oom, -1, 0);
1993 1994
}

K
KAMEZAWA Hiroyuki 已提交
1995 1996
static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);

K
KAMEZAWA Hiroyuki 已提交
1997
struct oom_wait_info {
1998
	struct mem_cgroup *memcg;
K
KAMEZAWA Hiroyuki 已提交
1999 2000 2001 2002 2003 2004
	wait_queue_t	wait;
};

static int memcg_oom_wake_function(wait_queue_t *wait,
	unsigned mode, int sync, void *arg)
{
2005 2006
	struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
	struct mem_cgroup *oom_wait_memcg;
K
KAMEZAWA Hiroyuki 已提交
2007 2008 2009
	struct oom_wait_info *oom_wait_info;

	oom_wait_info = container_of(wait, struct oom_wait_info, wait);
2010
	oom_wait_memcg = oom_wait_info->memcg;
K
KAMEZAWA Hiroyuki 已提交
2011 2012

	/*
2013
	 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
K
KAMEZAWA Hiroyuki 已提交
2014 2015
	 * Then we can use css_is_ancestor without taking care of RCU.
	 */
2016 2017
	if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
		&& !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
K
KAMEZAWA Hiroyuki 已提交
2018 2019 2020 2021
		return 0;
	return autoremove_wake_function(wait, mode, sync, arg);
}

2022
static void memcg_wakeup_oom(struct mem_cgroup *memcg)
K
KAMEZAWA Hiroyuki 已提交
2023
{
2024
	atomic_inc(&memcg->oom_wakeups);
2025 2026
	/* for filtering, pass "memcg" as argument. */
	__wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
K
KAMEZAWA Hiroyuki 已提交
2027 2028
}

2029
static void memcg_oom_recover(struct mem_cgroup *memcg)
2030
{
2031 2032
	if (memcg && atomic_read(&memcg->under_oom))
		memcg_wakeup_oom(memcg);
2033 2034
}

K
KAMEZAWA Hiroyuki 已提交
2035
/*
2036
 * try to call OOM killer
K
KAMEZAWA Hiroyuki 已提交
2037
 */
2038
static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
2039
{
2040
	bool locked;
2041
	int wakeups;
K
KAMEZAWA Hiroyuki 已提交
2042

2043 2044 2045 2046
	if (!current->memcg_oom.may_oom)
		return;

	current->memcg_oom.in_memcg_oom = 1;
2047

K
KAMEZAWA Hiroyuki 已提交
2048
	/*
2049 2050 2051 2052 2053
	 * As with any blocking lock, a contender needs to start
	 * listening for wakeups before attempting the trylock,
	 * otherwise it can miss the wakeup from the unlock and sleep
	 * indefinitely.  This is just open-coded because our locking
	 * is so particular to memcg hierarchies.
K
KAMEZAWA Hiroyuki 已提交
2054
	 */
2055
	wakeups = atomic_read(&memcg->oom_wakeups);
2056 2057 2058 2059
	mem_cgroup_mark_under_oom(memcg);

	locked = mem_cgroup_oom_trylock(memcg);

2060
	if (locked)
2061
		mem_cgroup_oom_notify(memcg);
K
KAMEZAWA Hiroyuki 已提交
2062

2063 2064
	if (locked && !memcg->oom_kill_disable) {
		mem_cgroup_unmark_under_oom(memcg);
2065
		mem_cgroup_out_of_memory(memcg, mask, order);
2066 2067 2068 2069 2070 2071 2072
		mem_cgroup_oom_unlock(memcg);
		/*
		 * There is no guarantee that an OOM-lock contender
		 * sees the wakeups triggered by the OOM kill
		 * uncharges.  Wake any sleepers explicitely.
		 */
		memcg_oom_recover(memcg);
2073
	} else {
2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097
		/*
		 * A system call can just return -ENOMEM, but if this
		 * is a page fault and somebody else is handling the
		 * OOM already, we need to sleep on the OOM waitqueue
		 * for this memcg until the situation is resolved.
		 * Which can take some time because it might be
		 * handled by a userspace task.
		 *
		 * However, this is the charge context, which means
		 * that we may sit on a large call stack and hold
		 * various filesystem locks, the mmap_sem etc. and we
		 * don't want the OOM handler to deadlock on them
		 * while we sit here and wait.  Store the current OOM
		 * context in the task_struct, then return -ENOMEM.
		 * At the end of the page fault handler, with the
		 * stack unwound, pagefault_out_of_memory() will check
		 * back with us by calling
		 * mem_cgroup_oom_synchronize(), possibly putting the
		 * task to sleep.
		 */
		current->memcg_oom.oom_locked = locked;
		current->memcg_oom.wakeups = wakeups;
		css_get(&memcg->css);
		current->memcg_oom.wait_on_memcg = memcg;
K
KAMEZAWA Hiroyuki 已提交
2098
	}
2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144
}

/**
 * mem_cgroup_oom_synchronize - complete memcg OOM handling
 *
 * This has to be called at the end of a page fault if the the memcg
 * OOM handler was enabled and the fault is returning %VM_FAULT_OOM.
 *
 * Memcg supports userspace OOM handling, so failed allocations must
 * sleep on a waitqueue until the userspace task resolves the
 * situation.  Sleeping directly in the charge context with all kinds
 * of locks held is not a good idea, instead we remember an OOM state
 * in the task and mem_cgroup_oom_synchronize() has to be called at
 * the end of the page fault to put the task to sleep and clean up the
 * OOM state.
 *
 * Returns %true if an ongoing memcg OOM situation was detected and
 * finalized, %false otherwise.
 */
bool mem_cgroup_oom_synchronize(void)
{
	struct oom_wait_info owait;
	struct mem_cgroup *memcg;

	/* OOM is global, do not handle */
	if (!current->memcg_oom.in_memcg_oom)
		return false;

	/*
	 * We invoked the OOM killer but there is a chance that a kill
	 * did not free up any charges.  Everybody else might already
	 * be sleeping, so restart the fault and keep the rampage
	 * going until some charges are released.
	 */
	memcg = current->memcg_oom.wait_on_memcg;
	if (!memcg)
		goto out;

	if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
		goto out_memcg;

	owait.memcg = memcg;
	owait.wait.flags = 0;
	owait.wait.func = memcg_oom_wake_function;
	owait.wait.private = current;
	INIT_LIST_HEAD(&owait.wait.task_list);
K
KAMEZAWA Hiroyuki 已提交
2145

2146 2147 2148 2149 2150 2151 2152 2153
	prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
	/* Only sleep if we didn't miss any wakeups since OOM */
	if (atomic_read(&memcg->oom_wakeups) == current->memcg_oom.wakeups)
		schedule();
	finish_wait(&memcg_oom_waitq, &owait.wait);
out_memcg:
	mem_cgroup_unmark_under_oom(memcg);
	if (current->memcg_oom.oom_locked) {
2154 2155 2156 2157 2158 2159 2160 2161
		mem_cgroup_oom_unlock(memcg);
		/*
		 * There is no guarantee that an OOM-lock contender
		 * sees the wakeups triggered by the OOM kill
		 * uncharges.  Wake any sleepers explicitely.
		 */
		memcg_oom_recover(memcg);
	}
2162 2163 2164 2165
	css_put(&memcg->css);
	current->memcg_oom.wait_on_memcg = NULL;
out:
	current->memcg_oom.in_memcg_oom = 0;
K
KAMEZAWA Hiroyuki 已提交
2166
	return true;
2167 2168
}

2169 2170 2171
/*
 * Currently used to update mapped file statistics, but the routine can be
 * generalized to update other statistics as well.
2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188
 *
 * Notes: Race condition
 *
 * We usually use page_cgroup_lock() for accessing page_cgroup member but
 * it tends to be costly. But considering some conditions, we doesn't need
 * to do so _always_.
 *
 * Considering "charge", lock_page_cgroup() is not required because all
 * file-stat operations happen after a page is attached to radix-tree. There
 * are no race with "charge".
 *
 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
 * if there are race with "uncharge". Statistics itself is properly handled
 * by flags.
 *
 * Considering "move", this is an only case we see a race. To make the race
2189 2190
 * small, we check mm->moving_account and detect there are possibility of race
 * If there is, we take a lock.
2191
 */
2192

2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
void __mem_cgroup_begin_update_page_stat(struct page *page,
				bool *locked, unsigned long *flags)
{
	struct mem_cgroup *memcg;
	struct page_cgroup *pc;

	pc = lookup_page_cgroup(page);
again:
	memcg = pc->mem_cgroup;
	if (unlikely(!memcg || !PageCgroupUsed(pc)))
		return;
	/*
	 * If this memory cgroup is not under account moving, we don't
2206
	 * need to take move_lock_mem_cgroup(). Because we already hold
2207
	 * rcu_read_lock(), any calls to move_account will be delayed until
2208
	 * rcu_read_unlock() if mem_cgroup_stolen() == true.
2209
	 */
2210
	if (!mem_cgroup_stolen(memcg))
2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227
		return;

	move_lock_mem_cgroup(memcg, flags);
	if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
		move_unlock_mem_cgroup(memcg, flags);
		goto again;
	}
	*locked = true;
}

void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
{
	struct page_cgroup *pc = lookup_page_cgroup(page);

	/*
	 * It's guaranteed that pc->mem_cgroup never changes while
	 * lock is held because a routine modifies pc->mem_cgroup
2228
	 * should take move_lock_mem_cgroup().
2229 2230 2231 2232
	 */
	move_unlock_mem_cgroup(pc->mem_cgroup, flags);
}

2233 2234
void mem_cgroup_update_page_stat(struct page *page,
				 enum mem_cgroup_page_stat_item idx, int val)
2235
{
2236
	struct mem_cgroup *memcg;
2237
	struct page_cgroup *pc = lookup_page_cgroup(page);
2238
	unsigned long uninitialized_var(flags);
2239

2240
	if (mem_cgroup_disabled())
2241
		return;
2242

2243 2244
	memcg = pc->mem_cgroup;
	if (unlikely(!memcg || !PageCgroupUsed(pc)))
2245
		return;
2246 2247

	switch (idx) {
2248 2249
	case MEMCG_NR_FILE_MAPPED:
		idx = MEM_CGROUP_STAT_FILE_MAPPED;
2250 2251 2252
		break;
	default:
		BUG();
2253
	}
2254

2255
	this_cpu_add(memcg->stat->count[idx], val);
2256
}
2257

2258 2259 2260 2261
/*
 * size of first charge trial. "32" comes from vmscan.c's magic value.
 * TODO: maybe necessary to use big numbers in big irons.
 */
2262
#define CHARGE_BATCH	32U
2263 2264
struct memcg_stock_pcp {
	struct mem_cgroup *cached; /* this never be root cgroup */
2265
	unsigned int nr_pages;
2266
	struct work_struct work;
2267
	unsigned long flags;
2268
#define FLUSHING_CACHED_CHARGE	0
2269 2270
};
static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2271
static DEFINE_MUTEX(percpu_charge_mutex);
2272

2273 2274 2275 2276 2277 2278 2279 2280 2281 2282
/**
 * consume_stock: Try to consume stocked charge on this cpu.
 * @memcg: memcg to consume from.
 * @nr_pages: how many pages to charge.
 *
 * The charges will only happen if @memcg matches the current cpu's memcg
 * stock, and at least @nr_pages are available in that stock.  Failure to
 * service an allocation will refill the stock.
 *
 * returns true if successful, false otherwise.
2283
 */
2284
static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2285 2286 2287 2288
{
	struct memcg_stock_pcp *stock;
	bool ret = true;

2289 2290 2291
	if (nr_pages > CHARGE_BATCH)
		return false;

2292
	stock = &get_cpu_var(memcg_stock);
2293 2294
	if (memcg == stock->cached && stock->nr_pages >= nr_pages)
		stock->nr_pages -= nr_pages;
2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307
	else /* need to call res_counter_charge */
		ret = false;
	put_cpu_var(memcg_stock);
	return ret;
}

/*
 * Returns stocks cached in percpu to res_counter and reset cached information.
 */
static void drain_stock(struct memcg_stock_pcp *stock)
{
	struct mem_cgroup *old = stock->cached;

2308 2309 2310 2311
	if (stock->nr_pages) {
		unsigned long bytes = stock->nr_pages * PAGE_SIZE;

		res_counter_uncharge(&old->res, bytes);
2312
		if (do_swap_account)
2313 2314
			res_counter_uncharge(&old->memsw, bytes);
		stock->nr_pages = 0;
2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326
	}
	stock->cached = NULL;
}

/*
 * This must be called under preempt disabled or must be called by
 * a thread which is pinned to local cpu.
 */
static void drain_local_stock(struct work_struct *dummy)
{
	struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
	drain_stock(stock);
2327
	clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2328 2329
}

2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
static void __init memcg_stock_init(void)
{
	int cpu;

	for_each_possible_cpu(cpu) {
		struct memcg_stock_pcp *stock =
					&per_cpu(memcg_stock, cpu);
		INIT_WORK(&stock->work, drain_local_stock);
	}
}

2341 2342
/*
 * Cache charges(val) which is from res_counter, to local per_cpu area.
2343
 * This will be consumed by consume_stock() function, later.
2344
 */
2345
static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2346 2347 2348
{
	struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);

2349
	if (stock->cached != memcg) { /* reset if necessary */
2350
		drain_stock(stock);
2351
		stock->cached = memcg;
2352
	}
2353
	stock->nr_pages += nr_pages;
2354 2355 2356 2357
	put_cpu_var(memcg_stock);
}

/*
2358
 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2359 2360
 * of the hierarchy under it. sync flag says whether we should block
 * until the work is done.
2361
 */
2362
static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
2363
{
2364
	int cpu, curcpu;
2365

2366 2367
	/* Notify other cpus that system-wide "drain" is running */
	get_online_cpus();
2368
	curcpu = get_cpu();
2369 2370
	for_each_online_cpu(cpu) {
		struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2371
		struct mem_cgroup *memcg;
2372

2373 2374
		memcg = stock->cached;
		if (!memcg || !stock->nr_pages)
2375
			continue;
2376
		if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
2377
			continue;
2378 2379 2380 2381 2382 2383
		if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
			if (cpu == curcpu)
				drain_local_stock(&stock->work);
			else
				schedule_work_on(cpu, &stock->work);
		}
2384
	}
2385
	put_cpu();
2386 2387 2388 2389 2390 2391

	if (!sync)
		goto out;

	for_each_online_cpu(cpu) {
		struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2392
		if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
2393 2394 2395
			flush_work(&stock->work);
	}
out:
A
Andrew Morton 已提交
2396
	put_online_cpus();
2397 2398 2399 2400 2401 2402 2403 2404
}

/*
 * Tries to drain stocked charges in other cpus. This function is asynchronous
 * and just put a work per cpu for draining localy on each cpu. Caller can
 * expects some charges will be back to res_counter later but cannot wait for
 * it.
 */
2405
static void drain_all_stock_async(struct mem_cgroup *root_memcg)
2406
{
2407 2408 2409 2410 2411
	/*
	 * If someone calls draining, avoid adding more kworker runs.
	 */
	if (!mutex_trylock(&percpu_charge_mutex))
		return;
2412
	drain_all_stock(root_memcg, false);
2413
	mutex_unlock(&percpu_charge_mutex);
2414 2415 2416
}

/* This is a synchronous drain interface. */
2417
static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
2418 2419
{
	/* called when force_empty is called */
2420
	mutex_lock(&percpu_charge_mutex);
2421
	drain_all_stock(root_memcg, true);
2422
	mutex_unlock(&percpu_charge_mutex);
2423 2424
}

2425 2426 2427 2428
/*
 * This function drains percpu counter value from DEAD cpu and
 * move it to local cpu. Note that this function can be preempted.
 */
2429
static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2430 2431 2432
{
	int i;

2433
	spin_lock(&memcg->pcp_counter_lock);
2434
	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2435
		long x = per_cpu(memcg->stat->count[i], cpu);
2436

2437 2438
		per_cpu(memcg->stat->count[i], cpu) = 0;
		memcg->nocpu_base.count[i] += x;
2439
	}
2440
	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2441
		unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2442

2443 2444
		per_cpu(memcg->stat->events[i], cpu) = 0;
		memcg->nocpu_base.events[i] += x;
2445
	}
2446
	spin_unlock(&memcg->pcp_counter_lock);
2447 2448
}

2449
static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
2450 2451 2452 2453 2454
					unsigned long action,
					void *hcpu)
{
	int cpu = (unsigned long)hcpu;
	struct memcg_stock_pcp *stock;
2455
	struct mem_cgroup *iter;
2456

2457
	if (action == CPU_ONLINE)
2458 2459
		return NOTIFY_OK;

2460
	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2461
		return NOTIFY_OK;
2462

2463
	for_each_mem_cgroup(iter)
2464 2465
		mem_cgroup_drain_pcp_counter(iter, cpu);

2466 2467 2468 2469 2470
	stock = &per_cpu(memcg_stock, cpu);
	drain_stock(stock);
	return NOTIFY_OK;
}

2471 2472 2473 2474 2475 2476 2477 2478 2479

/* See __mem_cgroup_try_charge() for details */
enum {
	CHARGE_OK,		/* success */
	CHARGE_RETRY,		/* need to retry but retry is not bad */
	CHARGE_NOMEM,		/* we can't do more. return -ENOMEM */
	CHARGE_WOULDBLOCK,	/* GFP_WAIT wasn't set and no enough res. */
};

2480
static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2481
				unsigned int nr_pages, unsigned int min_pages,
2482
				bool invoke_oom)
2483
{
2484
	unsigned long csize = nr_pages * PAGE_SIZE;
2485 2486 2487 2488 2489
	struct mem_cgroup *mem_over_limit;
	struct res_counter *fail_res;
	unsigned long flags = 0;
	int ret;

2490
	ret = res_counter_charge(&memcg->res, csize, &fail_res);
2491 2492 2493 2494

	if (likely(!ret)) {
		if (!do_swap_account)
			return CHARGE_OK;
2495
		ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
2496 2497 2498
		if (likely(!ret))
			return CHARGE_OK;

2499
		res_counter_uncharge(&memcg->res, csize);
2500 2501 2502 2503
		mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
		flags |= MEM_CGROUP_RECLAIM_NOSWAP;
	} else
		mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2504 2505 2506 2507
	/*
	 * Never reclaim on behalf of optional batching, retry with a
	 * single page instead.
	 */
2508
	if (nr_pages > min_pages)
2509 2510 2511 2512 2513
		return CHARGE_RETRY;

	if (!(gfp_mask & __GFP_WAIT))
		return CHARGE_WOULDBLOCK;

2514 2515 2516
	if (gfp_mask & __GFP_NORETRY)
		return CHARGE_NOMEM;

2517
	ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
2518
	if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2519
		return CHARGE_RETRY;
2520
	/*
2521 2522 2523 2524 2525 2526 2527
	 * Even though the limit is exceeded at this point, reclaim
	 * may have been able to free some pages.  Retry the charge
	 * before killing the task.
	 *
	 * Only for regular pages, though: huge pages are rather
	 * unlikely to succeed so close to the limit, and we fall back
	 * to regular pages anyway in case of failure.
2528
	 */
2529
	if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
2530 2531 2532 2533 2534 2535 2536 2537 2538
		return CHARGE_RETRY;

	/*
	 * At task move, charge accounts can be doubly counted. So, it's
	 * better to wait until the end of task_move if something is going on.
	 */
	if (mem_cgroup_wait_acct_move(mem_over_limit))
		return CHARGE_RETRY;

2539 2540
	if (invoke_oom)
		mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
2541

2542
	return CHARGE_NOMEM;
2543 2544
}

2545
/*
2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564
 * __mem_cgroup_try_charge() does
 * 1. detect memcg to be charged against from passed *mm and *ptr,
 * 2. update res_counter
 * 3. call memory reclaim if necessary.
 *
 * In some special case, if the task is fatal, fatal_signal_pending() or
 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
 * as possible without any hazards. 2: all pages should have a valid
 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
 * pointer, that is treated as a charge to root_mem_cgroup.
 *
 * So __mem_cgroup_try_charge() will return
 *  0       ...  on success, filling *ptr with a valid memcg pointer.
 *  -ENOMEM ...  charge failure because of resource limits.
 *  -EINTR  ...  if thread is fatal. *ptr is filled with root_mem_cgroup.
 *
 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
 * the oom-killer can be invoked.
2565
 */
2566
static int __mem_cgroup_try_charge(struct mm_struct *mm,
A
Andrea Arcangeli 已提交
2567
				   gfp_t gfp_mask,
2568
				   unsigned int nr_pages,
2569
				   struct mem_cgroup **ptr,
2570
				   bool oom)
2571
{
2572
	unsigned int batch = max(CHARGE_BATCH, nr_pages);
2573
	int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2574
	struct mem_cgroup *memcg = NULL;
2575
	int ret;
2576

K
KAMEZAWA Hiroyuki 已提交
2577 2578 2579 2580 2581 2582 2583 2584
	/*
	 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
	 * in system level. So, allow to go ahead dying process in addition to
	 * MEMDIE process.
	 */
	if (unlikely(test_thread_flag(TIF_MEMDIE)
		     || fatal_signal_pending(current)))
		goto bypass;
2585

2586
	/*
2587 2588
	 * We always charge the cgroup the mm_struct belongs to.
	 * The mm_struct's mem_cgroup changes on task migration if the
2589
	 * thread group leader migrates. It's possible that mm is not
2590
	 * set, if so charge the root memcg (happens for pagecache usage).
2591
	 */
2592
	if (!*ptr && !mm)
2593
		*ptr = root_mem_cgroup;
K
KAMEZAWA Hiroyuki 已提交
2594
again:
2595 2596 2597
	if (*ptr) { /* css should be a valid one */
		memcg = *ptr;
		if (mem_cgroup_is_root(memcg))
K
KAMEZAWA Hiroyuki 已提交
2598
			goto done;
2599
		if (consume_stock(memcg, nr_pages))
K
KAMEZAWA Hiroyuki 已提交
2600
			goto done;
2601
		css_get(&memcg->css);
2602
	} else {
K
KAMEZAWA Hiroyuki 已提交
2603
		struct task_struct *p;
2604

K
KAMEZAWA Hiroyuki 已提交
2605 2606 2607
		rcu_read_lock();
		p = rcu_dereference(mm->owner);
		/*
2608
		 * Because we don't have task_lock(), "p" can exit.
2609
		 * In that case, "memcg" can point to root or p can be NULL with
2610 2611 2612 2613 2614 2615
		 * race with swapoff. Then, we have small risk of mis-accouning.
		 * But such kind of mis-account by race always happens because
		 * we don't have cgroup_mutex(). It's overkill and we allo that
		 * small race, here.
		 * (*) swapoff at el will charge against mm-struct not against
		 * task-struct. So, mm->owner can be NULL.
K
KAMEZAWA Hiroyuki 已提交
2616
		 */
2617
		memcg = mem_cgroup_from_task(p);
2618 2619 2620
		if (!memcg)
			memcg = root_mem_cgroup;
		if (mem_cgroup_is_root(memcg)) {
K
KAMEZAWA Hiroyuki 已提交
2621 2622 2623
			rcu_read_unlock();
			goto done;
		}
2624
		if (consume_stock(memcg, nr_pages)) {
K
KAMEZAWA Hiroyuki 已提交
2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636
			/*
			 * It seems dagerous to access memcg without css_get().
			 * But considering how consume_stok works, it's not
			 * necessary. If consume_stock success, some charges
			 * from this memcg are cached on this cpu. So, we
			 * don't need to call css_get()/css_tryget() before
			 * calling consume_stock().
			 */
			rcu_read_unlock();
			goto done;
		}
		/* after here, we may be blocked. we need to get refcnt */
2637
		if (!css_tryget(&memcg->css)) {
K
KAMEZAWA Hiroyuki 已提交
2638 2639 2640 2641 2642
			rcu_read_unlock();
			goto again;
		}
		rcu_read_unlock();
	}
2643

2644
	do {
2645
		bool invoke_oom = oom && !nr_oom_retries;
2646

2647
		/* If killed, bypass charge */
K
KAMEZAWA Hiroyuki 已提交
2648
		if (fatal_signal_pending(current)) {
2649
			css_put(&memcg->css);
2650
			goto bypass;
K
KAMEZAWA Hiroyuki 已提交
2651
		}
2652

2653 2654
		ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
					   nr_pages, invoke_oom);
2655 2656 2657 2658
		switch (ret) {
		case CHARGE_OK:
			break;
		case CHARGE_RETRY: /* not in OOM situation but retry */
2659
			batch = nr_pages;
2660 2661
			css_put(&memcg->css);
			memcg = NULL;
K
KAMEZAWA Hiroyuki 已提交
2662
			goto again;
2663
		case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2664
			css_put(&memcg->css);
2665 2666
			goto nomem;
		case CHARGE_NOMEM: /* OOM routine works */
2667
			if (!oom || invoke_oom) {
2668
				css_put(&memcg->css);
K
KAMEZAWA Hiroyuki 已提交
2669
				goto nomem;
K
KAMEZAWA Hiroyuki 已提交
2670
			}
2671 2672
			nr_oom_retries--;
			break;
2673
		}
2674 2675
	} while (ret != CHARGE_OK);

2676
	if (batch > nr_pages)
2677 2678
		refill_stock(memcg, batch - nr_pages);
	css_put(&memcg->css);
2679
done:
2680
	*ptr = memcg;
2681 2682
	return 0;
nomem:
2683
	*ptr = NULL;
2684
	return -ENOMEM;
K
KAMEZAWA Hiroyuki 已提交
2685
bypass:
2686 2687
	*ptr = root_mem_cgroup;
	return -EINTR;
2688
}
2689

2690 2691 2692 2693 2694
/*
 * Somemtimes we have to undo a charge we got by try_charge().
 * This function is for that and do uncharge, put css's refcnt.
 * gotten by try_charge().
 */
2695
static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
2696
				       unsigned int nr_pages)
2697
{
2698
	if (!mem_cgroup_is_root(memcg)) {
2699 2700
		unsigned long bytes = nr_pages * PAGE_SIZE;

2701
		res_counter_uncharge(&memcg->res, bytes);
2702
		if (do_swap_account)
2703
			res_counter_uncharge(&memcg->memsw, bytes);
2704
	}
2705 2706
}

2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724
/*
 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
 * This is useful when moving usage to parent cgroup.
 */
static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
					unsigned int nr_pages)
{
	unsigned long bytes = nr_pages * PAGE_SIZE;

	if (mem_cgroup_is_root(memcg))
		return;

	res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
	if (do_swap_account)
		res_counter_uncharge_until(&memcg->memsw,
						memcg->memsw.parent, bytes);
}

2725 2726
/*
 * A helper function to get mem_cgroup from ID. must be called under
T
Tejun Heo 已提交
2727 2728 2729
 * rcu_read_lock().  The caller is responsible for calling css_tryget if
 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
 * called against removed memcg.)
2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740
 */
static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
{
	struct cgroup_subsys_state *css;

	/* ID 0 is unused ID */
	if (!id)
		return NULL;
	css = css_lookup(&mem_cgroup_subsys, id);
	if (!css)
		return NULL;
2741
	return mem_cgroup_from_css(css);
2742 2743
}

2744
struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2745
{
2746
	struct mem_cgroup *memcg = NULL;
2747
	struct page_cgroup *pc;
2748
	unsigned short id;
2749 2750
	swp_entry_t ent;

2751 2752 2753
	VM_BUG_ON(!PageLocked(page));

	pc = lookup_page_cgroup(page);
2754
	lock_page_cgroup(pc);
2755
	if (PageCgroupUsed(pc)) {
2756 2757 2758
		memcg = pc->mem_cgroup;
		if (memcg && !css_tryget(&memcg->css))
			memcg = NULL;
2759
	} else if (PageSwapCache(page)) {
2760
		ent.val = page_private(page);
2761
		id = lookup_swap_cgroup_id(ent);
2762
		rcu_read_lock();
2763 2764 2765
		memcg = mem_cgroup_lookup(id);
		if (memcg && !css_tryget(&memcg->css))
			memcg = NULL;
2766
		rcu_read_unlock();
2767
	}
2768
	unlock_page_cgroup(pc);
2769
	return memcg;
2770 2771
}

2772
static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
2773
				       struct page *page,
2774
				       unsigned int nr_pages,
2775 2776
				       enum charge_type ctype,
				       bool lrucare)
2777
{
2778
	struct page_cgroup *pc = lookup_page_cgroup(page);
2779
	struct zone *uninitialized_var(zone);
2780
	struct lruvec *lruvec;
2781
	bool was_on_lru = false;
2782
	bool anon;
2783

2784
	lock_page_cgroup(pc);
2785
	VM_BUG_ON(PageCgroupUsed(pc));
2786 2787 2788 2789
	/*
	 * we don't need page_cgroup_lock about tail pages, becase they are not
	 * accessed by any other context at this point.
	 */
2790 2791 2792 2793 2794 2795 2796 2797 2798

	/*
	 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
	 * may already be on some other mem_cgroup's LRU.  Take care of it.
	 */
	if (lrucare) {
		zone = page_zone(page);
		spin_lock_irq(&zone->lru_lock);
		if (PageLRU(page)) {
2799
			lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2800
			ClearPageLRU(page);
2801
			del_page_from_lru_list(page, lruvec, page_lru(page));
2802 2803 2804 2805
			was_on_lru = true;
		}
	}

2806
	pc->mem_cgroup = memcg;
2807 2808 2809 2810 2811 2812
	/*
	 * We access a page_cgroup asynchronously without lock_page_cgroup().
	 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
	 * is accessed after testing USED bit. To make pc->mem_cgroup visible
	 * before USED bit, we need memory barrier here.
	 * See mem_cgroup_add_lru_list(), etc.
A
Andrew Morton 已提交
2813
	 */
K
KAMEZAWA Hiroyuki 已提交
2814
	smp_wmb();
2815
	SetPageCgroupUsed(pc);
2816

2817 2818
	if (lrucare) {
		if (was_on_lru) {
2819
			lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2820 2821
			VM_BUG_ON(PageLRU(page));
			SetPageLRU(page);
2822
			add_page_to_lru_list(page, lruvec, page_lru(page));
2823 2824 2825 2826
		}
		spin_unlock_irq(&zone->lru_lock);
	}

2827
	if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
2828 2829 2830 2831
		anon = true;
	else
		anon = false;

2832
	mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
2833
	unlock_page_cgroup(pc);
2834

2835
	/*
2836
	 * "charge_statistics" updated event counter.
2837
	 */
2838
	memcg_check_events(memcg, page);
2839
}
2840

2841 2842
static DEFINE_MUTEX(set_limit_mutex);

2843 2844 2845 2846 2847 2848 2849
#ifdef CONFIG_MEMCG_KMEM
static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
{
	return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
		(memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK);
}

G
Glauber Costa 已提交
2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862
/*
 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
 * in the memcg_cache_params struct.
 */
static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
{
	struct kmem_cache *cachep;

	VM_BUG_ON(p->is_root_cache);
	cachep = p->root_cache;
	return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)];
}

2863
#ifdef CONFIG_SLABINFO
2864 2865
static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css,
				    struct cftype *cft, struct seq_file *m)
2866
{
2867
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883
	struct memcg_cache_params *params;

	if (!memcg_can_account_kmem(memcg))
		return -EIO;

	print_slabinfo_header(m);

	mutex_lock(&memcg->slab_caches_mutex);
	list_for_each_entry(params, &memcg->memcg_slab_caches, list)
		cache_show(memcg_params_to_cache(params), m);
	mutex_unlock(&memcg->slab_caches_mutex);

	return 0;
}
#endif

2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936
static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
{
	struct res_counter *fail_res;
	struct mem_cgroup *_memcg;
	int ret = 0;
	bool may_oom;

	ret = res_counter_charge(&memcg->kmem, size, &fail_res);
	if (ret)
		return ret;

	/*
	 * Conditions under which we can wait for the oom_killer. Those are
	 * the same conditions tested by the core page allocator
	 */
	may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY);

	_memcg = memcg;
	ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
				      &_memcg, may_oom);

	if (ret == -EINTR)  {
		/*
		 * __mem_cgroup_try_charge() chosed to bypass to root due to
		 * OOM kill or fatal signal.  Since our only options are to
		 * either fail the allocation or charge it to this cgroup, do
		 * it as a temporary condition. But we can't fail. From a
		 * kmem/slab perspective, the cache has already been selected,
		 * by mem_cgroup_kmem_get_cache(), so it is too late to change
		 * our minds.
		 *
		 * This condition will only trigger if the task entered
		 * memcg_charge_kmem in a sane state, but was OOM-killed during
		 * __mem_cgroup_try_charge() above. Tasks that were already
		 * dying when the allocation triggers should have been already
		 * directed to the root cgroup in memcontrol.h
		 */
		res_counter_charge_nofail(&memcg->res, size, &fail_res);
		if (do_swap_account)
			res_counter_charge_nofail(&memcg->memsw, size,
						  &fail_res);
		ret = 0;
	} else if (ret)
		res_counter_uncharge(&memcg->kmem, size);

	return ret;
}

static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
{
	res_counter_uncharge(&memcg->res, size);
	if (do_swap_account)
		res_counter_uncharge(&memcg->memsw, size);
2937 2938 2939 2940 2941

	/* Not down to 0 */
	if (res_counter_uncharge(&memcg->kmem, size))
		return;

2942 2943 2944 2945 2946 2947 2948 2949
	/*
	 * Releases a reference taken in kmem_cgroup_css_offline in case
	 * this last uncharge is racing with the offlining code or it is
	 * outliving the memcg existence.
	 *
	 * The memory barrier imposed by test&clear is paired with the
	 * explicit one in memcg_kmem_mark_dead().
	 */
2950
	if (memcg_kmem_test_and_clear_dead(memcg))
2951
		css_put(&memcg->css);
2952 2953
}

2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
{
	if (!memcg)
		return;

	mutex_lock(&memcg->slab_caches_mutex);
	list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
	mutex_unlock(&memcg->slab_caches_mutex);
}

/*
 * helper for acessing a memcg's index. It will be used as an index in the
 * child cache array in kmem_cache, and also to derive its name. This function
 * will return -1 when this is not a kmem-limited memcg.
 */
int memcg_cache_id(struct mem_cgroup *memcg)
{
	return memcg ? memcg->kmemcg_id : -1;
}

2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036
/*
 * This ends up being protected by the set_limit mutex, during normal
 * operation, because that is its main call site.
 *
 * But when we create a new cache, we can call this as well if its parent
 * is kmem-limited. That will have to hold set_limit_mutex as well.
 */
int memcg_update_cache_sizes(struct mem_cgroup *memcg)
{
	int num, ret;

	num = ida_simple_get(&kmem_limited_groups,
				0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
	if (num < 0)
		return num;
	/*
	 * After this point, kmem_accounted (that we test atomically in
	 * the beginning of this conditional), is no longer 0. This
	 * guarantees only one process will set the following boolean
	 * to true. We don't need test_and_set because we're protected
	 * by the set_limit_mutex anyway.
	 */
	memcg_kmem_set_activated(memcg);

	ret = memcg_update_all_caches(num+1);
	if (ret) {
		ida_simple_remove(&kmem_limited_groups, num);
		memcg_kmem_clear_activated(memcg);
		return ret;
	}

	memcg->kmemcg_id = num;
	INIT_LIST_HEAD(&memcg->memcg_slab_caches);
	mutex_init(&memcg->slab_caches_mutex);
	return 0;
}

static size_t memcg_caches_array_size(int num_groups)
{
	ssize_t size;
	if (num_groups <= 0)
		return 0;

	size = 2 * num_groups;
	if (size < MEMCG_CACHES_MIN_SIZE)
		size = MEMCG_CACHES_MIN_SIZE;
	else if (size > MEMCG_CACHES_MAX_SIZE)
		size = MEMCG_CACHES_MAX_SIZE;

	return size;
}

/*
 * We should update the current array size iff all caches updates succeed. This
 * can only be done from the slab side. The slab mutex needs to be held when
 * calling this.
 */
void memcg_update_array_size(int num)
{
	if (num > memcg_limited_groups_array_size)
		memcg_limited_groups_array_size = memcg_caches_array_size(num);
}

3037 3038
static void kmem_cache_destroy_work_func(struct work_struct *w);

3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049
int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
{
	struct memcg_cache_params *cur_params = s->memcg_params;

	VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache);

	if (num_groups > memcg_limited_groups_array_size) {
		int i;
		ssize_t size = memcg_caches_array_size(num_groups);

		size *= sizeof(void *);
3050
		size += offsetof(struct memcg_cache_params, memcg_caches);
3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089

		s->memcg_params = kzalloc(size, GFP_KERNEL);
		if (!s->memcg_params) {
			s->memcg_params = cur_params;
			return -ENOMEM;
		}

		s->memcg_params->is_root_cache = true;

		/*
		 * There is the chance it will be bigger than
		 * memcg_limited_groups_array_size, if we failed an allocation
		 * in a cache, in which case all caches updated before it, will
		 * have a bigger array.
		 *
		 * But if that is the case, the data after
		 * memcg_limited_groups_array_size is certainly unused
		 */
		for (i = 0; i < memcg_limited_groups_array_size; i++) {
			if (!cur_params->memcg_caches[i])
				continue;
			s->memcg_params->memcg_caches[i] =
						cur_params->memcg_caches[i];
		}

		/*
		 * Ideally, we would wait until all caches succeed, and only
		 * then free the old one. But this is not worth the extra
		 * pointer per-cache we'd have to have for this.
		 *
		 * It is not a big deal if some caches are left with a size
		 * bigger than the others. And all updates will reset this
		 * anyway.
		 */
		kfree(cur_params);
	}
	return 0;
}

G
Glauber Costa 已提交
3090 3091
int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
			 struct kmem_cache *root_cache)
3092
{
3093
	size_t size;
3094 3095 3096 3097

	if (!memcg_kmem_enabled())
		return 0;

3098 3099
	if (!memcg) {
		size = offsetof(struct memcg_cache_params, memcg_caches);
3100
		size += memcg_limited_groups_array_size * sizeof(void *);
3101 3102
	} else
		size = sizeof(struct memcg_cache_params);
3103

3104 3105 3106 3107
	s->memcg_params = kzalloc(size, GFP_KERNEL);
	if (!s->memcg_params)
		return -ENOMEM;

G
Glauber Costa 已提交
3108
	if (memcg) {
3109
		s->memcg_params->memcg = memcg;
G
Glauber Costa 已提交
3110
		s->memcg_params->root_cache = root_cache;
3111 3112
		INIT_WORK(&s->memcg_params->destroy,
				kmem_cache_destroy_work_func);
3113 3114 3115
	} else
		s->memcg_params->is_root_cache = true;

3116 3117 3118 3119 3120
	return 0;
}

void memcg_release_cache(struct kmem_cache *s)
{
3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144
	struct kmem_cache *root;
	struct mem_cgroup *memcg;
	int id;

	/*
	 * This happens, for instance, when a root cache goes away before we
	 * add any memcg.
	 */
	if (!s->memcg_params)
		return;

	if (s->memcg_params->is_root_cache)
		goto out;

	memcg = s->memcg_params->memcg;
	id  = memcg_cache_id(memcg);

	root = s->memcg_params->root_cache;
	root->memcg_params->memcg_caches[id] = NULL;

	mutex_lock(&memcg->slab_caches_mutex);
	list_del(&s->memcg_params->list);
	mutex_unlock(&memcg->slab_caches_mutex);

3145
	css_put(&memcg->css);
3146
out:
3147 3148 3149
	kfree(s->memcg_params);
}

3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180
/*
 * During the creation a new cache, we need to disable our accounting mechanism
 * altogether. This is true even if we are not creating, but rather just
 * enqueing new caches to be created.
 *
 * This is because that process will trigger allocations; some visible, like
 * explicit kmallocs to auxiliary data structures, name strings and internal
 * cache structures; some well concealed, like INIT_WORK() that can allocate
 * objects during debug.
 *
 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
 * to it. This may not be a bounded recursion: since the first cache creation
 * failed to complete (waiting on the allocation), we'll just try to create the
 * cache again, failing at the same point.
 *
 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
 * inside the following two functions.
 */
static inline void memcg_stop_kmem_account(void)
{
	VM_BUG_ON(!current->mm);
	current->memcg_kmem_skip_account++;
}

static inline void memcg_resume_kmem_account(void)
{
	VM_BUG_ON(!current->mm);
	current->memcg_kmem_skip_account--;
}

G
Glauber Costa 已提交
3181 3182 3183 3184 3185 3186 3187 3188 3189
static void kmem_cache_destroy_work_func(struct work_struct *w)
{
	struct kmem_cache *cachep;
	struct memcg_cache_params *p;

	p = container_of(w, struct memcg_cache_params, destroy);

	cachep = memcg_params_to_cache(p);

G
Glauber Costa 已提交
3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210
	/*
	 * If we get down to 0 after shrink, we could delete right away.
	 * However, memcg_release_pages() already puts us back in the workqueue
	 * in that case. If we proceed deleting, we'll get a dangling
	 * reference, and removing the object from the workqueue in that case
	 * is unnecessary complication. We are not a fast path.
	 *
	 * Note that this case is fundamentally different from racing with
	 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
	 * kmem_cache_shrink, not only we would be reinserting a dead cache
	 * into the queue, but doing so from inside the worker racing to
	 * destroy it.
	 *
	 * So if we aren't down to zero, we'll just schedule a worker and try
	 * again
	 */
	if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
		kmem_cache_shrink(cachep);
		if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
			return;
	} else
G
Glauber Costa 已提交
3211 3212 3213 3214 3215 3216 3217 3218
		kmem_cache_destroy(cachep);
}

void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
{
	if (!cachep->memcg_params->dead)
		return;

G
Glauber Costa 已提交
3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238
	/*
	 * There are many ways in which we can get here.
	 *
	 * We can get to a memory-pressure situation while the delayed work is
	 * still pending to run. The vmscan shrinkers can then release all
	 * cache memory and get us to destruction. If this is the case, we'll
	 * be executed twice, which is a bug (the second time will execute over
	 * bogus data). In this case, cancelling the work should be fine.
	 *
	 * But we can also get here from the worker itself, if
	 * kmem_cache_shrink is enough to shake all the remaining objects and
	 * get the page count to 0. In this case, we'll deadlock if we try to
	 * cancel the work (the worker runs with an internal lock held, which
	 * is the same lock we would hold for cancel_work_sync().)
	 *
	 * Since we can't possibly know who got us here, just refrain from
	 * running if there is already work pending
	 */
	if (work_pending(&cachep->memcg_params->destroy))
		return;
G
Glauber Costa 已提交
3239 3240 3241 3242 3243 3244 3245
	/*
	 * We have to defer the actual destroying to a workqueue, because
	 * we might currently be in a context that cannot sleep.
	 */
	schedule_work(&cachep->memcg_params->destroy);
}

3246 3247 3248 3249 3250 3251 3252 3253 3254
/*
 * This lock protects updaters, not readers. We want readers to be as fast as
 * they can, and they will either see NULL or a valid cache value. Our model
 * allow them to see NULL, in which case the root memcg will be selected.
 *
 * We need this lock because multiple allocations to the same cache from a non
 * will span more than one worker. Only one of them can create the cache.
 */
static DEFINE_MUTEX(memcg_cache_mutex);
3255

3256 3257 3258
/*
 * Called with memcg_cache_mutex held
 */
3259 3260 3261 3262
static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
					 struct kmem_cache *s)
{
	struct kmem_cache *new;
3263
	static char *tmp_name = NULL;
3264

3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282
	lockdep_assert_held(&memcg_cache_mutex);

	/*
	 * kmem_cache_create_memcg duplicates the given name and
	 * cgroup_name for this name requires RCU context.
	 * This static temporary buffer is used to prevent from
	 * pointless shortliving allocation.
	 */
	if (!tmp_name) {
		tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
		if (!tmp_name)
			return NULL;
	}

	rcu_read_lock();
	snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
			 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
	rcu_read_unlock();
3283

3284
	new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
G
Glauber Costa 已提交
3285
				      (s->flags & ~SLAB_PANIC), s->ctor, s);
3286

3287 3288 3289
	if (new)
		new->allocflags |= __GFP_KMEMCG;

3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304
	return new;
}

static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
						  struct kmem_cache *cachep)
{
	struct kmem_cache *new_cachep;
	int idx;

	BUG_ON(!memcg_can_account_kmem(memcg));

	idx = memcg_cache_id(memcg);

	mutex_lock(&memcg_cache_mutex);
	new_cachep = cachep->memcg_params->memcg_caches[idx];
3305 3306
	if (new_cachep) {
		css_put(&memcg->css);
3307
		goto out;
3308
	}
3309 3310 3311 3312

	new_cachep = kmem_cache_dup(memcg, cachep);
	if (new_cachep == NULL) {
		new_cachep = cachep;
3313
		css_put(&memcg->css);
3314 3315 3316
		goto out;
	}

G
Glauber Costa 已提交
3317
	atomic_set(&new_cachep->memcg_params->nr_pages , 0);
3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329

	cachep->memcg_params->memcg_caches[idx] = new_cachep;
	/*
	 * the readers won't lock, make sure everybody sees the updated value,
	 * so they won't put stuff in the queue again for no reason
	 */
	wmb();
out:
	mutex_unlock(&memcg_cache_mutex);
	return new_cachep;
}

3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368
void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
{
	struct kmem_cache *c;
	int i;

	if (!s->memcg_params)
		return;
	if (!s->memcg_params->is_root_cache)
		return;

	/*
	 * If the cache is being destroyed, we trust that there is no one else
	 * requesting objects from it. Even if there are, the sanity checks in
	 * kmem_cache_destroy should caught this ill-case.
	 *
	 * Still, we don't want anyone else freeing memcg_caches under our
	 * noses, which can happen if a new memcg comes to life. As usual,
	 * we'll take the set_limit_mutex to protect ourselves against this.
	 */
	mutex_lock(&set_limit_mutex);
	for (i = 0; i < memcg_limited_groups_array_size; i++) {
		c = s->memcg_params->memcg_caches[i];
		if (!c)
			continue;

		/*
		 * We will now manually delete the caches, so to avoid races
		 * we need to cancel all pending destruction workers and
		 * proceed with destruction ourselves.
		 *
		 * kmem_cache_destroy() will call kmem_cache_shrink internally,
		 * and that could spawn the workers again: it is likely that
		 * the cache still have active pages until this very moment.
		 * This would lead us back to mem_cgroup_destroy_cache.
		 *
		 * But that will not execute at all if the "dead" flag is not
		 * set, so flip it down to guarantee we are in control.
		 */
		c->memcg_params->dead = false;
G
Glauber Costa 已提交
3369
		cancel_work_sync(&c->memcg_params->destroy);
3370 3371 3372 3373 3374
		kmem_cache_destroy(c);
	}
	mutex_unlock(&set_limit_mutex);
}

3375 3376 3377 3378 3379 3380
struct create_work {
	struct mem_cgroup *memcg;
	struct kmem_cache *cachep;
	struct work_struct work;
};

G
Glauber Costa 已提交
3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397
static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
{
	struct kmem_cache *cachep;
	struct memcg_cache_params *params;

	if (!memcg_kmem_is_active(memcg))
		return;

	mutex_lock(&memcg->slab_caches_mutex);
	list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
		cachep = memcg_params_to_cache(params);
		cachep->memcg_params->dead = true;
		schedule_work(&cachep->memcg_params->destroy);
	}
	mutex_unlock(&memcg->slab_caches_mutex);
}

3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409
static void memcg_create_cache_work_func(struct work_struct *w)
{
	struct create_work *cw;

	cw = container_of(w, struct create_work, work);
	memcg_create_kmem_cache(cw->memcg, cw->cachep);
	kfree(cw);
}

/*
 * Enqueue the creation of a per-memcg kmem_cache.
 */
3410 3411
static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
					 struct kmem_cache *cachep)
3412 3413 3414 3415
{
	struct create_work *cw;

	cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
3416 3417
	if (cw == NULL) {
		css_put(&memcg->css);
3418 3419 3420 3421 3422 3423 3424 3425 3426 3427
		return;
	}

	cw->memcg = memcg;
	cw->cachep = cachep;

	INIT_WORK(&cw->work, memcg_create_cache_work_func);
	schedule_work(&cw->work);
}

3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445
static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
				       struct kmem_cache *cachep)
{
	/*
	 * We need to stop accounting when we kmalloc, because if the
	 * corresponding kmalloc cache is not yet created, the first allocation
	 * in __memcg_create_cache_enqueue will recurse.
	 *
	 * However, it is better to enclose the whole function. Depending on
	 * the debugging options enabled, INIT_WORK(), for instance, can
	 * trigger an allocation. This too, will make us recurse. Because at
	 * this point we can't allow ourselves back into memcg_kmem_get_cache,
	 * the safest choice is to do it like this, wrapping the whole function.
	 */
	memcg_stop_kmem_account();
	__memcg_create_cache_enqueue(memcg, cachep);
	memcg_resume_kmem_account();
}
3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467
/*
 * Return the kmem_cache we're supposed to use for a slab allocation.
 * We try to use the current memcg's version of the cache.
 *
 * If the cache does not exist yet, if we are the first user of it,
 * we either create it immediately, if possible, or create it asynchronously
 * in a workqueue.
 * In the latter case, we will let the current allocation go through with
 * the original cache.
 *
 * Can't be called in interrupt context or from kernel threads.
 * This function needs to be called with rcu_read_lock() held.
 */
struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
					  gfp_t gfp)
{
	struct mem_cgroup *memcg;
	int idx;

	VM_BUG_ON(!cachep->memcg_params);
	VM_BUG_ON(!cachep->memcg_params->is_root_cache);

3468 3469 3470
	if (!current->mm || current->memcg_kmem_skip_account)
		return cachep;

3471 3472 3473 3474
	rcu_read_lock();
	memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));

	if (!memcg_can_account_kmem(memcg))
3475
		goto out;
3476 3477 3478 3479 3480 3481 3482 3483

	idx = memcg_cache_id(memcg);

	/*
	 * barrier to mare sure we're always seeing the up to date value.  The
	 * code updating memcg_caches will issue a write barrier to match this.
	 */
	read_barrier_depends();
3484 3485 3486
	if (likely(cachep->memcg_params->memcg_caches[idx])) {
		cachep = cachep->memcg_params->memcg_caches[idx];
		goto out;
3487 3488
	}

3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515
	/* The corresponding put will be done in the workqueue. */
	if (!css_tryget(&memcg->css))
		goto out;
	rcu_read_unlock();

	/*
	 * If we are in a safe context (can wait, and not in interrupt
	 * context), we could be be predictable and return right away.
	 * This would guarantee that the allocation being performed
	 * already belongs in the new cache.
	 *
	 * However, there are some clashes that can arrive from locking.
	 * For instance, because we acquire the slab_mutex while doing
	 * kmem_cache_dup, this means no further allocation could happen
	 * with the slab_mutex held.
	 *
	 * Also, because cache creation issue get_online_cpus(), this
	 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
	 * that ends up reversed during cpu hotplug. (cpuset allocates
	 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
	 * better to defer everything.
	 */
	memcg_create_cache_enqueue(memcg, cachep);
	return cachep;
out:
	rcu_read_unlock();
	return cachep;
3516 3517 3518
}
EXPORT_SYMBOL(__memcg_kmem_get_cache);

3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539
/*
 * We need to verify if the allocation against current->mm->owner's memcg is
 * possible for the given order. But the page is not allocated yet, so we'll
 * need a further commit step to do the final arrangements.
 *
 * It is possible for the task to switch cgroups in this mean time, so at
 * commit time, we can't rely on task conversion any longer.  We'll then use
 * the handle argument to return to the caller which cgroup we should commit
 * against. We could also return the memcg directly and avoid the pointer
 * passing, but a boolean return value gives better semantics considering
 * the compiled-out case as well.
 *
 * Returning true means the allocation is possible.
 */
bool
__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
{
	struct mem_cgroup *memcg;
	int ret;

	*_memcg = NULL;
3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554

	/*
	 * Disabling accounting is only relevant for some specific memcg
	 * internal allocations. Therefore we would initially not have such
	 * check here, since direct calls to the page allocator that are marked
	 * with GFP_KMEMCG only happen outside memcg core. We are mostly
	 * concerned with cache allocations, and by having this test at
	 * memcg_kmem_get_cache, we are already able to relay the allocation to
	 * the root cache and bypass the memcg cache altogether.
	 *
	 * There is one exception, though: the SLUB allocator does not create
	 * large order caches, but rather service large kmallocs directly from
	 * the page allocator. Therefore, the following sequence when backed by
	 * the SLUB allocator:
	 *
A
Andrew Morton 已提交
3555 3556 3557
	 *	memcg_stop_kmem_account();
	 *	kmalloc(<large_number>)
	 *	memcg_resume_kmem_account();
3558 3559 3560 3561 3562 3563 3564 3565 3566 3567
	 *
	 * would effectively ignore the fact that we should skip accounting,
	 * since it will drive us directly to this function without passing
	 * through the cache selector memcg_kmem_get_cache. Such large
	 * allocations are extremely rare but can happen, for instance, for the
	 * cache arrays. We bring this test here.
	 */
	if (!current->mm || current->memcg_kmem_skip_account)
		return true;

3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
	memcg = try_get_mem_cgroup_from_mm(current->mm);

	/*
	 * very rare case described in mem_cgroup_from_task. Unfortunately there
	 * isn't much we can do without complicating this too much, and it would
	 * be gfp-dependent anyway. Just let it go
	 */
	if (unlikely(!memcg))
		return true;

	if (!memcg_can_account_kmem(memcg)) {
		css_put(&memcg->css);
		return true;
	}

	ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
	if (!ret)
		*_memcg = memcg;

	css_put(&memcg->css);
	return (ret == 0);
}

void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
			      int order)
{
	struct page_cgroup *pc;

	VM_BUG_ON(mem_cgroup_is_root(memcg));

	/* The page allocation failed. Revert */
	if (!page) {
		memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
		return;
	}

	pc = lookup_page_cgroup(page);
	lock_page_cgroup(pc);
	pc->mem_cgroup = memcg;
	SetPageCgroupUsed(pc);
	unlock_page_cgroup(pc);
}

void __memcg_kmem_uncharge_pages(struct page *page, int order)
{
	struct mem_cgroup *memcg = NULL;
	struct page_cgroup *pc;


	pc = lookup_page_cgroup(page);
	/*
	 * Fast unlocked return. Theoretically might have changed, have to
	 * check again after locking.
	 */
	if (!PageCgroupUsed(pc))
		return;

	lock_page_cgroup(pc);
	if (PageCgroupUsed(pc)) {
		memcg = pc->mem_cgroup;
		ClearPageCgroupUsed(pc);
	}
	unlock_page_cgroup(pc);

	/*
	 * We trust that only if there is a memcg associated with the page, it
	 * is a valid allocation
	 */
	if (!memcg)
		return;

	VM_BUG_ON(mem_cgroup_is_root(memcg));
	memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
}
G
Glauber Costa 已提交
3642 3643 3644 3645
#else
static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
{
}
3646 3647
#endif /* CONFIG_MEMCG_KMEM */

3648 3649
#ifdef CONFIG_TRANSPARENT_HUGEPAGE

3650
#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
3651 3652
/*
 * Because tail pages are not marked as "used", set it. We're under
3653 3654 3655
 * zone->lru_lock, 'splitting on pmd' and compound_lock.
 * charge/uncharge will be never happen and move_account() is done under
 * compound_lock(), so we don't have to take care of races.
3656
 */
3657
void mem_cgroup_split_huge_fixup(struct page *head)
3658 3659
{
	struct page_cgroup *head_pc = lookup_page_cgroup(head);
3660
	struct page_cgroup *pc;
3661
	struct mem_cgroup *memcg;
3662
	int i;
3663

3664 3665
	if (mem_cgroup_disabled())
		return;
3666 3667

	memcg = head_pc->mem_cgroup;
3668 3669
	for (i = 1; i < HPAGE_PMD_NR; i++) {
		pc = head_pc + i;
3670
		pc->mem_cgroup = memcg;
3671 3672 3673
		smp_wmb();/* see __commit_charge() */
		pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
	}
3674 3675
	__this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
		       HPAGE_PMD_NR);
3676
}
3677
#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3678

3679
/**
3680
 * mem_cgroup_move_account - move account of the page
3681
 * @page: the page
3682
 * @nr_pages: number of regular pages (>1 for huge pages)
3683 3684 3685 3686 3687
 * @pc:	page_cgroup of the page.
 * @from: mem_cgroup which the page is moved from.
 * @to:	mem_cgroup which the page is moved to. @from != @to.
 *
 * The caller must confirm following.
K
KAMEZAWA Hiroyuki 已提交
3688
 * - page is not on LRU (isolate_page() is useful.)
3689
 * - compound_lock is held when nr_pages > 1
3690
 *
3691 3692
 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
 * from old cgroup.
3693
 */
3694 3695 3696 3697
static int mem_cgroup_move_account(struct page *page,
				   unsigned int nr_pages,
				   struct page_cgroup *pc,
				   struct mem_cgroup *from,
3698
				   struct mem_cgroup *to)
3699
{
3700 3701
	unsigned long flags;
	int ret;
3702
	bool anon = PageAnon(page);
3703

3704
	VM_BUG_ON(from == to);
3705
	VM_BUG_ON(PageLRU(page));
3706 3707 3708 3709 3710 3711 3712
	/*
	 * The page is isolated from LRU. So, collapse function
	 * will not handle this page. But page splitting can happen.
	 * Do this check under compound_page_lock(). The caller should
	 * hold it.
	 */
	ret = -EBUSY;
3713
	if (nr_pages > 1 && !PageTransHuge(page))
3714 3715 3716 3717 3718 3719 3720 3721
		goto out;

	lock_page_cgroup(pc);

	ret = -EINVAL;
	if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
		goto unlock;

3722
	move_lock_mem_cgroup(from, &flags);
3723

3724
	if (!anon && page_mapped(page)) {
3725 3726 3727 3728 3729
		/* Update mapped_file data for mem_cgroup */
		preempt_disable();
		__this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
		__this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
		preempt_enable();
3730
	}
3731
	mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
3732

3733
	/* caller should have done css_get */
K
KAMEZAWA Hiroyuki 已提交
3734
	pc->mem_cgroup = to;
3735
	mem_cgroup_charge_statistics(to, page, anon, nr_pages);
3736
	move_unlock_mem_cgroup(from, &flags);
3737 3738
	ret = 0;
unlock:
3739
	unlock_page_cgroup(pc);
3740 3741 3742
	/*
	 * check events
	 */
3743 3744
	memcg_check_events(to, page);
	memcg_check_events(from, page);
3745
out:
3746 3747 3748
	return ret;
}

3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768
/**
 * mem_cgroup_move_parent - moves page to the parent group
 * @page: the page to move
 * @pc: page_cgroup of the page
 * @child: page's cgroup
 *
 * move charges to its parent or the root cgroup if the group has no
 * parent (aka use_hierarchy==0).
 * Although this might fail (get_page_unless_zero, isolate_lru_page or
 * mem_cgroup_move_account fails) the failure is always temporary and
 * it signals a race with a page removal/uncharge or migration. In the
 * first case the page is on the way out and it will vanish from the LRU
 * on the next attempt and the call should be retried later.
 * Isolation from the LRU fails only if page has been isolated from
 * the LRU since we looked at it and that usually means either global
 * reclaim or migration going on. The page will either get back to the
 * LRU or vanish.
 * Finaly mem_cgroup_move_account fails only if the page got uncharged
 * (!PageCgroupUsed) or moved to a different group. The page will
 * disappear in the next attempt.
3769
 */
3770 3771
static int mem_cgroup_move_parent(struct page *page,
				  struct page_cgroup *pc,
3772
				  struct mem_cgroup *child)
3773 3774
{
	struct mem_cgroup *parent;
3775
	unsigned int nr_pages;
3776
	unsigned long uninitialized_var(flags);
3777 3778
	int ret;

3779
	VM_BUG_ON(mem_cgroup_is_root(child));
3780

3781 3782 3783 3784 3785
	ret = -EBUSY;
	if (!get_page_unless_zero(page))
		goto out;
	if (isolate_lru_page(page))
		goto put;
3786

3787
	nr_pages = hpage_nr_pages(page);
K
KAMEZAWA Hiroyuki 已提交
3788

3789 3790 3791 3792 3793 3794
	parent = parent_mem_cgroup(child);
	/*
	 * If no parent, move charges to root cgroup.
	 */
	if (!parent)
		parent = root_mem_cgroup;
3795

3796 3797
	if (nr_pages > 1) {
		VM_BUG_ON(!PageTransHuge(page));
3798
		flags = compound_lock_irqsave(page);
3799
	}
3800

3801
	ret = mem_cgroup_move_account(page, nr_pages,
3802
				pc, child, parent);
3803 3804
	if (!ret)
		__mem_cgroup_cancel_local_charge(child, nr_pages);
3805

3806
	if (nr_pages > 1)
3807
		compound_unlock_irqrestore(page, flags);
K
KAMEZAWA Hiroyuki 已提交
3808
	putback_lru_page(page);
3809
put:
3810
	put_page(page);
3811
out:
3812 3813 3814
	return ret;
}

3815 3816 3817 3818 3819 3820 3821
/*
 * Charge the memory controller for page usage.
 * Return
 * 0 if the charge was successful
 * < 0 if the cgroup is over its limit
 */
static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
3822
				gfp_t gfp_mask, enum charge_type ctype)
3823
{
3824
	struct mem_cgroup *memcg = NULL;
3825
	unsigned int nr_pages = 1;
3826
	bool oom = true;
3827
	int ret;
A
Andrea Arcangeli 已提交
3828

A
Andrea Arcangeli 已提交
3829
	if (PageTransHuge(page)) {
3830
		nr_pages <<= compound_order(page);
A
Andrea Arcangeli 已提交
3831
		VM_BUG_ON(!PageTransHuge(page));
3832 3833 3834 3835 3836
		/*
		 * Never OOM-kill a process for a huge page.  The
		 * fault handler will fall back to regular pages.
		 */
		oom = false;
A
Andrea Arcangeli 已提交
3837
	}
3838

3839
	ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
3840
	if (ret == -ENOMEM)
3841
		return ret;
3842
	__mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
3843 3844 3845
	return 0;
}

3846 3847
int mem_cgroup_newpage_charge(struct page *page,
			      struct mm_struct *mm, gfp_t gfp_mask)
3848
{
3849
	if (mem_cgroup_disabled())
3850
		return 0;
3851 3852 3853
	VM_BUG_ON(page_mapped(page));
	VM_BUG_ON(page->mapping && !PageAnon(page));
	VM_BUG_ON(!mm);
3854
	return mem_cgroup_charge_common(page, mm, gfp_mask,
3855
					MEM_CGROUP_CHARGE_TYPE_ANON);
3856 3857
}

3858 3859 3860
/*
 * While swap-in, try_charge -> commit or cancel, the page is locked.
 * And when try_charge() successfully returns, one refcnt to memcg without
3861
 * struct page_cgroup is acquired. This refcnt will be consumed by
3862 3863
 * "commit()" or removed by "cancel()"
 */
3864 3865 3866 3867
static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
					  struct page *page,
					  gfp_t mask,
					  struct mem_cgroup **memcgp)
3868
{
3869
	struct mem_cgroup *memcg;
3870
	struct page_cgroup *pc;
3871
	int ret;
3872

3873 3874 3875 3876 3877 3878 3879 3880 3881 3882
	pc = lookup_page_cgroup(page);
	/*
	 * Every swap fault against a single page tries to charge the
	 * page, bail as early as possible.  shmem_unuse() encounters
	 * already charged pages, too.  The USED bit is protected by
	 * the page lock, which serializes swap cache removal, which
	 * in turn serializes uncharging.
	 */
	if (PageCgroupUsed(pc))
		return 0;
3883 3884
	if (!do_swap_account)
		goto charge_cur_mm;
3885 3886
	memcg = try_get_mem_cgroup_from_page(page);
	if (!memcg)
3887
		goto charge_cur_mm;
3888 3889
	*memcgp = memcg;
	ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
3890
	css_put(&memcg->css);
3891 3892
	if (ret == -EINTR)
		ret = 0;
3893
	return ret;
3894
charge_cur_mm:
3895 3896 3897 3898
	ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
	if (ret == -EINTR)
		ret = 0;
	return ret;
3899 3900
}

3901 3902 3903 3904 3905 3906
int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
				 gfp_t gfp_mask, struct mem_cgroup **memcgp)
{
	*memcgp = NULL;
	if (mem_cgroup_disabled())
		return 0;
3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920
	/*
	 * A racing thread's fault, or swapoff, may have already
	 * updated the pte, and even removed page from swap cache: in
	 * those cases unuse_pte()'s pte_same() test will fail; but
	 * there's also a KSM case which does need to charge the page.
	 */
	if (!PageSwapCache(page)) {
		int ret;

		ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
		if (ret == -EINTR)
			ret = 0;
		return ret;
	}
3921 3922 3923
	return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
}

3924 3925 3926 3927 3928 3929 3930 3931 3932
void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
{
	if (mem_cgroup_disabled())
		return;
	if (!memcg)
		return;
	__mem_cgroup_cancel_charge(memcg, 1);
}

D
Daisuke Nishimura 已提交
3933
static void
3934
__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
D
Daisuke Nishimura 已提交
3935
					enum charge_type ctype)
3936
{
3937
	if (mem_cgroup_disabled())
3938
		return;
3939
	if (!memcg)
3940
		return;
3941

3942
	__mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
3943 3944 3945
	/*
	 * Now swap is on-memory. This means this page may be
	 * counted both as mem and swap....double count.
3946 3947 3948
	 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
	 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
	 * may call delete_from_swap_cache() before reach here.
3949
	 */
3950
	if (do_swap_account && PageSwapCache(page)) {
3951
		swp_entry_t ent = {.val = page_private(page)};
3952
		mem_cgroup_uncharge_swap(ent);
3953
	}
3954 3955
}

3956 3957
void mem_cgroup_commit_charge_swapin(struct page *page,
				     struct mem_cgroup *memcg)
D
Daisuke Nishimura 已提交
3958
{
3959
	__mem_cgroup_commit_charge_swapin(page, memcg,
3960
					  MEM_CGROUP_CHARGE_TYPE_ANON);
D
Daisuke Nishimura 已提交
3961 3962
}

3963 3964
int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
				gfp_t gfp_mask)
3965
{
3966 3967 3968 3969
	struct mem_cgroup *memcg = NULL;
	enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
	int ret;

3970
	if (mem_cgroup_disabled())
3971 3972 3973 3974 3975 3976 3977
		return 0;
	if (PageCompound(page))
		return 0;

	if (!PageSwapCache(page))
		ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
	else { /* page is swapcache/shmem */
3978 3979
		ret = __mem_cgroup_try_charge_swapin(mm, page,
						     gfp_mask, &memcg);
3980 3981 3982 3983
		if (!ret)
			__mem_cgroup_commit_charge_swapin(page, memcg, type);
	}
	return ret;
3984 3985
}

3986
static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
3987 3988
				   unsigned int nr_pages,
				   const enum charge_type ctype)
3989 3990 3991
{
	struct memcg_batch_info *batch = NULL;
	bool uncharge_memsw = true;
3992

3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003
	/* If swapout, usage of swap doesn't decrease */
	if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
		uncharge_memsw = false;

	batch = &current->memcg_batch;
	/*
	 * In usual, we do css_get() when we remember memcg pointer.
	 * But in this case, we keep res->usage until end of a series of
	 * uncharges. Then, it's ok to ignore memcg's refcnt.
	 */
	if (!batch->memcg)
4004
		batch->memcg = memcg;
4005 4006
	/*
	 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
L
Lucas De Marchi 已提交
4007
	 * In those cases, all pages freed continuously can be expected to be in
4008 4009 4010 4011 4012 4013 4014 4015
	 * the same cgroup and we have chance to coalesce uncharges.
	 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
	 * because we want to do uncharge as soon as possible.
	 */

	if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
		goto direct_uncharge;

4016
	if (nr_pages > 1)
A
Andrea Arcangeli 已提交
4017 4018
		goto direct_uncharge;

4019 4020 4021 4022 4023
	/*
	 * In typical case, batch->memcg == mem. This means we can
	 * merge a series of uncharges to an uncharge of res_counter.
	 * If not, we uncharge res_counter ony by one.
	 */
4024
	if (batch->memcg != memcg)
4025 4026
		goto direct_uncharge;
	/* remember freed charge and uncharge it later */
4027
	batch->nr_pages++;
4028
	if (uncharge_memsw)
4029
		batch->memsw_nr_pages++;
4030 4031
	return;
direct_uncharge:
4032
	res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
4033
	if (uncharge_memsw)
4034 4035 4036
		res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
	if (unlikely(batch->memcg != memcg))
		memcg_oom_recover(memcg);
4037
}
4038

4039
/*
4040
 * uncharge if !page_mapped(page)
4041
 */
4042
static struct mem_cgroup *
4043 4044
__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
			     bool end_migration)
4045
{
4046
	struct mem_cgroup *memcg = NULL;
4047 4048
	unsigned int nr_pages = 1;
	struct page_cgroup *pc;
4049
	bool anon;
4050

4051
	if (mem_cgroup_disabled())
4052
		return NULL;
4053

A
Andrea Arcangeli 已提交
4054
	if (PageTransHuge(page)) {
4055
		nr_pages <<= compound_order(page);
A
Andrea Arcangeli 已提交
4056 4057
		VM_BUG_ON(!PageTransHuge(page));
	}
4058
	/*
4059
	 * Check if our page_cgroup is valid
4060
	 */
4061
	pc = lookup_page_cgroup(page);
4062
	if (unlikely(!PageCgroupUsed(pc)))
4063
		return NULL;
4064

4065
	lock_page_cgroup(pc);
K
KAMEZAWA Hiroyuki 已提交
4066

4067
	memcg = pc->mem_cgroup;
4068

K
KAMEZAWA Hiroyuki 已提交
4069 4070 4071
	if (!PageCgroupUsed(pc))
		goto unlock_out;

4072 4073
	anon = PageAnon(page);

K
KAMEZAWA Hiroyuki 已提交
4074
	switch (ctype) {
4075
	case MEM_CGROUP_CHARGE_TYPE_ANON:
4076 4077 4078 4079 4080
		/*
		 * Generally PageAnon tells if it's the anon statistics to be
		 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
		 * used before page reached the stage of being marked PageAnon.
		 */
4081 4082
		anon = true;
		/* fallthrough */
K
KAMEZAWA Hiroyuki 已提交
4083
	case MEM_CGROUP_CHARGE_TYPE_DROP:
4084
		/* See mem_cgroup_prepare_migration() */
4085 4086 4087 4088 4089 4090 4091 4092 4093 4094
		if (page_mapped(page))
			goto unlock_out;
		/*
		 * Pages under migration may not be uncharged.  But
		 * end_migration() /must/ be the one uncharging the
		 * unused post-migration page and so it has to call
		 * here with the migration bit still set.  See the
		 * res_counter handling below.
		 */
		if (!end_migration && PageCgroupMigration(pc))
K
KAMEZAWA Hiroyuki 已提交
4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105
			goto unlock_out;
		break;
	case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
		if (!PageAnon(page)) {	/* Shared memory */
			if (page->mapping && !page_is_file_cache(page))
				goto unlock_out;
		} else if (page_mapped(page)) /* Anon */
				goto unlock_out;
		break;
	default:
		break;
4106
	}
K
KAMEZAWA Hiroyuki 已提交
4107

4108
	mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
K
KAMEZAWA Hiroyuki 已提交
4109

4110
	ClearPageCgroupUsed(pc);
4111 4112 4113 4114 4115 4116
	/*
	 * pc->mem_cgroup is not cleared here. It will be accessed when it's
	 * freed from LRU. This is safe because uncharged page is expected not
	 * to be reused (freed soon). Exception is SwapCache, it's handled by
	 * special functions.
	 */
4117

4118
	unlock_page_cgroup(pc);
K
KAMEZAWA Hiroyuki 已提交
4119
	/*
4120
	 * even after unlock, we have memcg->res.usage here and this memcg
L
Li Zefan 已提交
4121
	 * will never be freed, so it's safe to call css_get().
K
KAMEZAWA Hiroyuki 已提交
4122
	 */
4123
	memcg_check_events(memcg, page);
K
KAMEZAWA Hiroyuki 已提交
4124
	if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
4125
		mem_cgroup_swap_statistics(memcg, true);
L
Li Zefan 已提交
4126
		css_get(&memcg->css);
K
KAMEZAWA Hiroyuki 已提交
4127
	}
4128 4129 4130 4131 4132 4133
	/*
	 * Migration does not charge the res_counter for the
	 * replacement page, so leave it alone when phasing out the
	 * page that is unused after the migration.
	 */
	if (!end_migration && !mem_cgroup_is_root(memcg))
4134
		mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
4135

4136
	return memcg;
K
KAMEZAWA Hiroyuki 已提交
4137 4138 4139

unlock_out:
	unlock_page_cgroup(pc);
4140
	return NULL;
4141 4142
}

4143 4144
void mem_cgroup_uncharge_page(struct page *page)
{
4145 4146 4147
	/* early check. */
	if (page_mapped(page))
		return;
4148
	VM_BUG_ON(page->mapping && !PageAnon(page));
4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160
	/*
	 * If the page is in swap cache, uncharge should be deferred
	 * to the swap path, which also properly accounts swap usage
	 * and handles memcg lifetime.
	 *
	 * Note that this check is not stable and reclaim may add the
	 * page to swap cache at any time after this.  However, if the
	 * page is not in swap cache by the time page->mapcount hits
	 * 0, there won't be any page table references to the swap
	 * slot, and reclaim will free it and not actually write the
	 * page to disk.
	 */
4161 4162
	if (PageSwapCache(page))
		return;
4163
	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
4164 4165 4166 4167 4168
}

void mem_cgroup_uncharge_cache_page(struct page *page)
{
	VM_BUG_ON(page_mapped(page));
4169
	VM_BUG_ON(page->mapping);
4170
	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
4171 4172
}

4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186
/*
 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
 * In that cases, pages are freed continuously and we can expect pages
 * are in the same memcg. All these calls itself limits the number of
 * pages freed at once, then uncharge_start/end() is called properly.
 * This may be called prural(2) times in a context,
 */

void mem_cgroup_uncharge_start(void)
{
	current->memcg_batch.do_batch++;
	/* We can do nest. */
	if (current->memcg_batch.do_batch == 1) {
		current->memcg_batch.memcg = NULL;
4187 4188
		current->memcg_batch.nr_pages = 0;
		current->memcg_batch.memsw_nr_pages = 0;
4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208
	}
}

void mem_cgroup_uncharge_end(void)
{
	struct memcg_batch_info *batch = &current->memcg_batch;

	if (!batch->do_batch)
		return;

	batch->do_batch--;
	if (batch->do_batch) /* If stacked, do nothing. */
		return;

	if (!batch->memcg)
		return;
	/*
	 * This "batch->memcg" is valid without any css_get/put etc...
	 * bacause we hide charges behind us.
	 */
4209 4210 4211 4212 4213 4214
	if (batch->nr_pages)
		res_counter_uncharge(&batch->memcg->res,
				     batch->nr_pages * PAGE_SIZE);
	if (batch->memsw_nr_pages)
		res_counter_uncharge(&batch->memcg->memsw,
				     batch->memsw_nr_pages * PAGE_SIZE);
4215
	memcg_oom_recover(batch->memcg);
4216 4217 4218 4219
	/* forget this pointer (for sanity check) */
	batch->memcg = NULL;
}

4220
#ifdef CONFIG_SWAP
4221
/*
4222
 * called after __delete_from_swap_cache() and drop "page" account.
4223 4224
 * memcg information is recorded to swap_cgroup of "ent"
 */
K
KAMEZAWA Hiroyuki 已提交
4225 4226
void
mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
4227 4228
{
	struct mem_cgroup *memcg;
K
KAMEZAWA Hiroyuki 已提交
4229 4230 4231 4232 4233
	int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;

	if (!swapout) /* this was a swap cache but the swap is unused ! */
		ctype = MEM_CGROUP_CHARGE_TYPE_DROP;

4234
	memcg = __mem_cgroup_uncharge_common(page, ctype, false);
4235

K
KAMEZAWA Hiroyuki 已提交
4236 4237
	/*
	 * record memcg information,  if swapout && memcg != NULL,
L
Li Zefan 已提交
4238
	 * css_get() was called in uncharge().
K
KAMEZAWA Hiroyuki 已提交
4239 4240
	 */
	if (do_swap_account && swapout && memcg)
4241
		swap_cgroup_record(ent, css_id(&memcg->css));
4242
}
4243
#endif
4244

A
Andrew Morton 已提交
4245
#ifdef CONFIG_MEMCG_SWAP
4246 4247 4248 4249 4250
/*
 * called from swap_entry_free(). remove record in swap_cgroup and
 * uncharge "memsw" account.
 */
void mem_cgroup_uncharge_swap(swp_entry_t ent)
K
KAMEZAWA Hiroyuki 已提交
4251
{
4252
	struct mem_cgroup *memcg;
4253
	unsigned short id;
4254 4255 4256 4257

	if (!do_swap_account)
		return;

4258 4259 4260
	id = swap_cgroup_record(ent, 0);
	rcu_read_lock();
	memcg = mem_cgroup_lookup(id);
4261
	if (memcg) {
4262 4263 4264 4265
		/*
		 * We uncharge this because swap is freed.
		 * This memcg can be obsolete one. We avoid calling css_tryget
		 */
4266
		if (!mem_cgroup_is_root(memcg))
4267
			res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
4268
		mem_cgroup_swap_statistics(memcg, false);
L
Li Zefan 已提交
4269
		css_put(&memcg->css);
4270
	}
4271
	rcu_read_unlock();
K
KAMEZAWA Hiroyuki 已提交
4272
}
4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288

/**
 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
 * @entry: swap entry to be moved
 * @from:  mem_cgroup which the entry is moved from
 * @to:  mem_cgroup which the entry is moved to
 *
 * It succeeds only when the swap_cgroup's record for this entry is the same
 * as the mem_cgroup's id of @from.
 *
 * Returns 0 on success, -EINVAL on failure.
 *
 * The caller must have charged to @to, IOW, called res_counter_charge() about
 * both res and memsw, and called css_get().
 */
static int mem_cgroup_move_swap_account(swp_entry_t entry,
4289
				struct mem_cgroup *from, struct mem_cgroup *to)
4290 4291 4292 4293 4294 4295 4296 4297
{
	unsigned short old_id, new_id;

	old_id = css_id(&from->css);
	new_id = css_id(&to->css);

	if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
		mem_cgroup_swap_statistics(from, false);
4298
		mem_cgroup_swap_statistics(to, true);
4299
		/*
4300 4301 4302
		 * This function is only called from task migration context now.
		 * It postpones res_counter and refcount handling till the end
		 * of task migration(mem_cgroup_clear_mc()) for performance
L
Li Zefan 已提交
4303 4304 4305 4306 4307 4308
		 * improvement. But we cannot postpone css_get(to)  because if
		 * the process that has been moved to @to does swap-in, the
		 * refcount of @to might be decreased to 0.
		 *
		 * We are in attach() phase, so the cgroup is guaranteed to be
		 * alive, so we can just call css_get().
4309
		 */
L
Li Zefan 已提交
4310
		css_get(&to->css);
4311 4312 4313 4314 4315 4316
		return 0;
	}
	return -EINVAL;
}
#else
static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
4317
				struct mem_cgroup *from, struct mem_cgroup *to)
4318 4319 4320
{
	return -EINVAL;
}
4321
#endif
K
KAMEZAWA Hiroyuki 已提交
4322

4323
/*
4324 4325
 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
 * page belongs to.
4326
 */
4327 4328
void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
				  struct mem_cgroup **memcgp)
4329
{
4330
	struct mem_cgroup *memcg = NULL;
4331
	unsigned int nr_pages = 1;
4332
	struct page_cgroup *pc;
4333
	enum charge_type ctype;
4334

4335
	*memcgp = NULL;
4336

4337
	if (mem_cgroup_disabled())
4338
		return;
4339

4340 4341 4342
	if (PageTransHuge(page))
		nr_pages <<= compound_order(page);

4343 4344 4345
	pc = lookup_page_cgroup(page);
	lock_page_cgroup(pc);
	if (PageCgroupUsed(pc)) {
4346 4347
		memcg = pc->mem_cgroup;
		css_get(&memcg->css);
4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378
		/*
		 * At migrating an anonymous page, its mapcount goes down
		 * to 0 and uncharge() will be called. But, even if it's fully
		 * unmapped, migration may fail and this page has to be
		 * charged again. We set MIGRATION flag here and delay uncharge
		 * until end_migration() is called
		 *
		 * Corner Case Thinking
		 * A)
		 * When the old page was mapped as Anon and it's unmap-and-freed
		 * while migration was ongoing.
		 * If unmap finds the old page, uncharge() of it will be delayed
		 * until end_migration(). If unmap finds a new page, it's
		 * uncharged when it make mapcount to be 1->0. If unmap code
		 * finds swap_migration_entry, the new page will not be mapped
		 * and end_migration() will find it(mapcount==0).
		 *
		 * B)
		 * When the old page was mapped but migraion fails, the kernel
		 * remaps it. A charge for it is kept by MIGRATION flag even
		 * if mapcount goes down to 0. We can do remap successfully
		 * without charging it again.
		 *
		 * C)
		 * The "old" page is under lock_page() until the end of
		 * migration, so, the old page itself will not be swapped-out.
		 * If the new page is swapped out before end_migraton, our
		 * hook to usual swap-out path will catch the event.
		 */
		if (PageAnon(page))
			SetPageCgroupMigration(pc);
4379
	}
4380
	unlock_page_cgroup(pc);
4381 4382 4383 4384
	/*
	 * If the page is not charged at this point,
	 * we return here.
	 */
4385
	if (!memcg)
4386
		return;
4387

4388
	*memcgp = memcg;
4389 4390 4391 4392 4393 4394 4395
	/*
	 * We charge new page before it's used/mapped. So, even if unlock_page()
	 * is called before end_migration, we can catch all events on this new
	 * page. In the case new page is migrated but not remapped, new page's
	 * mapcount will be finally 0 and we call uncharge in end_migration().
	 */
	if (PageAnon(page))
4396
		ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
4397
	else
4398
		ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
4399 4400 4401 4402 4403
	/*
	 * The page is committed to the memcg, but it's not actually
	 * charged to the res_counter since we plan on replacing the
	 * old one and only one page is going to be left afterwards.
	 */
4404
	__mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
4405
}
4406

4407
/* remove redundant charge if migration failed*/
4408
void mem_cgroup_end_migration(struct mem_cgroup *memcg,
4409
	struct page *oldpage, struct page *newpage, bool migration_ok)
4410
{
4411
	struct page *used, *unused;
4412
	struct page_cgroup *pc;
4413
	bool anon;
4414

4415
	if (!memcg)
4416
		return;
4417

4418
	if (!migration_ok) {
4419 4420
		used = oldpage;
		unused = newpage;
4421
	} else {
4422
		used = newpage;
4423 4424
		unused = oldpage;
	}
4425
	anon = PageAnon(used);
4426 4427 4428 4429
	__mem_cgroup_uncharge_common(unused,
				     anon ? MEM_CGROUP_CHARGE_TYPE_ANON
				     : MEM_CGROUP_CHARGE_TYPE_CACHE,
				     true);
4430
	css_put(&memcg->css);
4431
	/*
4432 4433 4434
	 * We disallowed uncharge of pages under migration because mapcount
	 * of the page goes down to zero, temporarly.
	 * Clear the flag and check the page should be charged.
4435
	 */
4436 4437 4438 4439 4440
	pc = lookup_page_cgroup(oldpage);
	lock_page_cgroup(pc);
	ClearPageCgroupMigration(pc);
	unlock_page_cgroup(pc);

4441
	/*
4442 4443 4444 4445 4446 4447
	 * If a page is a file cache, radix-tree replacement is very atomic
	 * and we can skip this check. When it was an Anon page, its mapcount
	 * goes down to 0. But because we added MIGRATION flage, it's not
	 * uncharged yet. There are several case but page->mapcount check
	 * and USED bit check in mem_cgroup_uncharge_page() will do enough
	 * check. (see prepare_charge() also)
4448
	 */
4449
	if (anon)
4450
		mem_cgroup_uncharge_page(used);
4451
}
4452

4453 4454 4455 4456 4457 4458 4459 4460
/*
 * At replace page cache, newpage is not under any memcg but it's on
 * LRU. So, this function doesn't touch res_counter but handles LRU
 * in correct way. Both pages are locked so we cannot race with uncharge.
 */
void mem_cgroup_replace_page_cache(struct page *oldpage,
				  struct page *newpage)
{
4461
	struct mem_cgroup *memcg = NULL;
4462 4463 4464 4465 4466 4467 4468 4469 4470
	struct page_cgroup *pc;
	enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;

	if (mem_cgroup_disabled())
		return;

	pc = lookup_page_cgroup(oldpage);
	/* fix accounting on old pages */
	lock_page_cgroup(pc);
4471 4472
	if (PageCgroupUsed(pc)) {
		memcg = pc->mem_cgroup;
4473
		mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
4474 4475
		ClearPageCgroupUsed(pc);
	}
4476 4477
	unlock_page_cgroup(pc);

4478 4479 4480 4481 4482 4483
	/*
	 * When called from shmem_replace_page(), in some cases the
	 * oldpage has already been charged, and in some cases not.
	 */
	if (!memcg)
		return;
4484 4485 4486 4487 4488
	/*
	 * Even if newpage->mapping was NULL before starting replacement,
	 * the newpage may be on LRU(or pagevec for LRU) already. We lock
	 * LRU while we overwrite pc->mem_cgroup.
	 */
4489
	__mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
4490 4491
}

4492 4493 4494 4495 4496 4497
#ifdef CONFIG_DEBUG_VM
static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
{
	struct page_cgroup *pc;

	pc = lookup_page_cgroup(page);
4498 4499 4500 4501 4502
	/*
	 * Can be NULL while feeding pages into the page allocator for
	 * the first time, i.e. during boot or memory hotplug;
	 * or when mem_cgroup_disabled().
	 */
4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521
	if (likely(pc) && PageCgroupUsed(pc))
		return pc;
	return NULL;
}

bool mem_cgroup_bad_page_check(struct page *page)
{
	if (mem_cgroup_disabled())
		return false;

	return lookup_page_cgroup_used(page) != NULL;
}

void mem_cgroup_print_bad_page(struct page *page)
{
	struct page_cgroup *pc;

	pc = lookup_page_cgroup_used(page);
	if (pc) {
4522 4523
		pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
			 pc, pc->flags, pc->mem_cgroup);
4524 4525 4526 4527
	}
}
#endif

4528
static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
4529
				unsigned long long val)
4530
{
4531
	int retry_count;
4532
	u64 memswlimit, memlimit;
4533
	int ret = 0;
4534 4535
	int children = mem_cgroup_count_children(memcg);
	u64 curusage, oldusage;
4536
	int enlarge;
4537 4538 4539 4540 4541 4542 4543 4544 4545

	/*
	 * For keeping hierarchical_reclaim simple, how long we should retry
	 * is depends on callers. We set our retry-count to be function
	 * of # of children which we should visit in this loop.
	 */
	retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;

	oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4546

4547
	enlarge = 0;
4548
	while (retry_count) {
4549 4550 4551 4552
		if (signal_pending(current)) {
			ret = -EINTR;
			break;
		}
4553 4554 4555
		/*
		 * Rather than hide all in some function, I do this in
		 * open coded manner. You see what this really does.
4556
		 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
4557 4558 4559 4560 4561 4562
		 */
		mutex_lock(&set_limit_mutex);
		memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
		if (memswlimit < val) {
			ret = -EINVAL;
			mutex_unlock(&set_limit_mutex);
4563 4564
			break;
		}
4565 4566 4567 4568 4569

		memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
		if (memlimit < val)
			enlarge = 1;

4570
		ret = res_counter_set_limit(&memcg->res, val);
4571 4572 4573 4574 4575 4576
		if (!ret) {
			if (memswlimit == val)
				memcg->memsw_is_minimum = true;
			else
				memcg->memsw_is_minimum = false;
		}
4577 4578 4579 4580 4581
		mutex_unlock(&set_limit_mutex);

		if (!ret)
			break;

4582 4583
		mem_cgroup_reclaim(memcg, GFP_KERNEL,
				   MEM_CGROUP_RECLAIM_SHRINK);
4584 4585
		curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
		/* Usage is reduced ? */
A
Andrew Morton 已提交
4586
		if (curusage >= oldusage)
4587 4588 4589
			retry_count--;
		else
			oldusage = curusage;
4590
	}
4591 4592
	if (!ret && enlarge)
		memcg_oom_recover(memcg);
4593

4594 4595 4596
	return ret;
}

L
Li Zefan 已提交
4597 4598
static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
					unsigned long long val)
4599
{
4600
	int retry_count;
4601
	u64 memlimit, memswlimit, oldusage, curusage;
4602 4603
	int children = mem_cgroup_count_children(memcg);
	int ret = -EBUSY;
4604
	int enlarge = 0;
4605

4606
	/* see mem_cgroup_resize_res_limit */
A
Andrew Morton 已提交
4607
	retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
4608
	oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
4609 4610 4611 4612 4613 4614 4615 4616
	while (retry_count) {
		if (signal_pending(current)) {
			ret = -EINTR;
			break;
		}
		/*
		 * Rather than hide all in some function, I do this in
		 * open coded manner. You see what this really does.
4617
		 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
4618 4619 4620 4621 4622 4623 4624 4625
		 */
		mutex_lock(&set_limit_mutex);
		memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
		if (memlimit > val) {
			ret = -EINVAL;
			mutex_unlock(&set_limit_mutex);
			break;
		}
4626 4627 4628
		memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
		if (memswlimit < val)
			enlarge = 1;
4629
		ret = res_counter_set_limit(&memcg->memsw, val);
4630 4631 4632 4633 4634 4635
		if (!ret) {
			if (memlimit == val)
				memcg->memsw_is_minimum = true;
			else
				memcg->memsw_is_minimum = false;
		}
4636 4637 4638 4639 4640
		mutex_unlock(&set_limit_mutex);

		if (!ret)
			break;

4641 4642 4643
		mem_cgroup_reclaim(memcg, GFP_KERNEL,
				   MEM_CGROUP_RECLAIM_NOSWAP |
				   MEM_CGROUP_RECLAIM_SHRINK);
4644
		curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
4645
		/* Usage is reduced ? */
4646
		if (curusage >= oldusage)
4647
			retry_count--;
4648 4649
		else
			oldusage = curusage;
4650
	}
4651 4652
	if (!ret && enlarge)
		memcg_oom_recover(memcg);
4653 4654 4655
	return ret;
}

4656 4657 4658 4659 4660 4661 4662
/**
 * mem_cgroup_force_empty_list - clears LRU of a group
 * @memcg: group to clear
 * @node: NUMA node
 * @zid: zone id
 * @lru: lru to to clear
 *
4663
 * Traverse a specified page_cgroup list and try to drop them all.  This doesn't
4664 4665
 * reclaim the pages page themselves - pages are moved to the parent (or root)
 * group.
4666
 */
4667
static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
K
KAMEZAWA Hiroyuki 已提交
4668
				int node, int zid, enum lru_list lru)
4669
{
4670
	struct lruvec *lruvec;
4671
	unsigned long flags;
4672
	struct list_head *list;
4673 4674
	struct page *busy;
	struct zone *zone;
4675

K
KAMEZAWA Hiroyuki 已提交
4676
	zone = &NODE_DATA(node)->node_zones[zid];
4677 4678
	lruvec = mem_cgroup_zone_lruvec(zone, memcg);
	list = &lruvec->lists[lru];
4679

4680
	busy = NULL;
4681
	do {
4682
		struct page_cgroup *pc;
4683 4684
		struct page *page;

K
KAMEZAWA Hiroyuki 已提交
4685
		spin_lock_irqsave(&zone->lru_lock, flags);
4686
		if (list_empty(list)) {
K
KAMEZAWA Hiroyuki 已提交
4687
			spin_unlock_irqrestore(&zone->lru_lock, flags);
4688
			break;
4689
		}
4690 4691 4692
		page = list_entry(list->prev, struct page, lru);
		if (busy == page) {
			list_move(&page->lru, list);
4693
			busy = NULL;
K
KAMEZAWA Hiroyuki 已提交
4694
			spin_unlock_irqrestore(&zone->lru_lock, flags);
4695 4696
			continue;
		}
K
KAMEZAWA Hiroyuki 已提交
4697
		spin_unlock_irqrestore(&zone->lru_lock, flags);
4698

4699
		pc = lookup_page_cgroup(page);
4700

4701
		if (mem_cgroup_move_parent(page, pc, memcg)) {
4702
			/* found lock contention or "pc" is obsolete. */
4703
			busy = page;
4704 4705 4706
			cond_resched();
		} else
			busy = NULL;
4707
	} while (!list_empty(list));
4708 4709 4710
}

/*
4711 4712
 * make mem_cgroup's charge to be 0 if there is no task by moving
 * all the charges and pages to the parent.
4713
 * This enables deleting this mem_cgroup.
4714 4715
 *
 * Caller is responsible for holding css reference on the memcg.
4716
 */
4717
static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
4718
{
4719
	int node, zid;
4720
	u64 usage;
4721

4722
	do {
4723 4724
		/* This is for making all *used* pages to be on LRU. */
		lru_add_drain_all();
4725 4726
		drain_all_stock_sync(memcg);
		mem_cgroup_start_move(memcg);
4727
		for_each_node_state(node, N_MEMORY) {
4728
			for (zid = 0; zid < MAX_NR_ZONES; zid++) {
H
Hugh Dickins 已提交
4729 4730
				enum lru_list lru;
				for_each_lru(lru) {
4731
					mem_cgroup_force_empty_list(memcg,
H
Hugh Dickins 已提交
4732
							node, zid, lru);
4733
				}
4734
			}
4735
		}
4736 4737
		mem_cgroup_end_move(memcg);
		memcg_oom_recover(memcg);
4738
		cond_resched();
4739

4740
		/*
4741 4742 4743 4744 4745
		 * Kernel memory may not necessarily be trackable to a specific
		 * process. So they are not migrated, and therefore we can't
		 * expect their value to drop to 0 here.
		 * Having res filled up with kmem only is enough.
		 *
4746 4747 4748 4749 4750 4751
		 * This is a safety check because mem_cgroup_force_empty_list
		 * could have raced with mem_cgroup_replace_page_cache callers
		 * so the lru seemed empty but the page could have been added
		 * right after the check. RES_USAGE should be safe as we always
		 * charge before adding to the LRU.
		 */
4752 4753 4754
		usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
			res_counter_read_u64(&memcg->kmem, RES_USAGE);
	} while (usage > 0);
4755 4756
}

4757 4758 4759 4760 4761 4762 4763
/*
 * This mainly exists for tests during the setting of set of use_hierarchy.
 * Since this is the very setting we are changing, the current hierarchy value
 * is meaningless
 */
static inline bool __memcg_has_children(struct mem_cgroup *memcg)
{
4764
	struct cgroup_subsys_state *pos;
4765 4766

	/* bounce at first found */
4767
	css_for_each_child(pos, &memcg->css)
4768 4769 4770 4771 4772
		return true;
	return false;
}

/*
4773 4774
 * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed
 * to be already dead (as in mem_cgroup_force_empty, for instance).  This is
4775 4776 4777 4778 4779 4780 4781 4782 4783
 * from mem_cgroup_count_children(), in the sense that we don't really care how
 * many children we have; we only need to know if we have any.  It also counts
 * any memcg without hierarchy as infertile.
 */
static inline bool memcg_has_children(struct mem_cgroup *memcg)
{
	return memcg->use_hierarchy && __memcg_has_children(memcg);
}

4784 4785 4786 4787 4788 4789 4790 4791 4792 4793
/*
 * Reclaims as many pages from the given memcg as possible and moves
 * the rest to the parent.
 *
 * Caller is responsible for holding css reference for memcg.
 */
static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
{
	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
	struct cgroup *cgrp = memcg->css.cgroup;
4794

4795
	/* returns EBUSY if there is a task or if we come here twice. */
4796 4797 4798
	if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
		return -EBUSY;

4799 4800
	/* we call try-to-free pages for make this cgroup empty */
	lru_add_drain_all();
4801
	/* try to free all pages in this cgroup */
4802
	while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
4803
		int progress;
4804

4805 4806 4807
		if (signal_pending(current))
			return -EINTR;

4808
		progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
4809
						false);
4810
		if (!progress) {
4811
			nr_retries--;
4812
			/* maybe some writeback is necessary */
4813
			congestion_wait(BLK_RW_ASYNC, HZ/10);
4814
		}
4815 4816

	}
K
KAMEZAWA Hiroyuki 已提交
4817
	lru_add_drain();
4818 4819 4820
	mem_cgroup_reparent_charges(memcg);

	return 0;
4821 4822
}

4823 4824
static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
					unsigned int event)
4825
{
4826
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4827

4828 4829
	if (mem_cgroup_is_root(memcg))
		return -EINVAL;
4830
	return mem_cgroup_force_empty(memcg);
4831 4832
}

4833 4834
static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
				     struct cftype *cft)
4835
{
4836
	return mem_cgroup_from_css(css)->use_hierarchy;
4837 4838
}

4839 4840
static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
				      struct cftype *cft, u64 val)
4841 4842
{
	int retval = 0;
4843
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
T
Tejun Heo 已提交
4844
	struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
4845

4846
	mutex_lock(&memcg_create_mutex);
4847 4848 4849 4850

	if (memcg->use_hierarchy == val)
		goto out;

4851
	/*
4852
	 * If parent's use_hierarchy is set, we can't make any modifications
4853 4854 4855 4856 4857 4858
	 * in the child subtrees. If it is unset, then the change can
	 * occur, provided the current cgroup has no children.
	 *
	 * For the root cgroup, parent_mem is NULL, we allow value to be
	 * set if there are no children.
	 */
4859
	if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
4860
				(val == 1 || val == 0)) {
4861
		if (!__memcg_has_children(memcg))
4862
			memcg->use_hierarchy = val;
4863 4864 4865 4866
		else
			retval = -EBUSY;
	} else
		retval = -EINVAL;
4867 4868

out:
4869
	mutex_unlock(&memcg_create_mutex);
4870 4871 4872 4873

	return retval;
}

4874

4875
static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
4876
					       enum mem_cgroup_stat_index idx)
4877
{
K
KAMEZAWA Hiroyuki 已提交
4878
	struct mem_cgroup *iter;
4879
	long val = 0;
4880

4881
	/* Per-cpu values can be negative, use a signed accumulator */
4882
	for_each_mem_cgroup_tree(iter, memcg)
K
KAMEZAWA Hiroyuki 已提交
4883 4884 4885 4886 4887
		val += mem_cgroup_read_stat(iter, idx);

	if (val < 0) /* race ? */
		val = 0;
	return val;
4888 4889
}

4890
static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
4891
{
K
KAMEZAWA Hiroyuki 已提交
4892
	u64 val;
4893

4894
	if (!mem_cgroup_is_root(memcg)) {
4895
		if (!swap)
4896
			return res_counter_read_u64(&memcg->res, RES_USAGE);
4897
		else
4898
			return res_counter_read_u64(&memcg->memsw, RES_USAGE);
4899 4900
	}

4901 4902 4903 4904
	/*
	 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
	 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
	 */
4905 4906
	val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
	val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
4907

K
KAMEZAWA Hiroyuki 已提交
4908
	if (swap)
4909
		val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
4910 4911 4912 4913

	return val << PAGE_SHIFT;
}

4914 4915 4916
static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css,
			       struct cftype *cft, struct file *file,
			       char __user *buf, size_t nbytes, loff_t *ppos)
B
Balbir Singh 已提交
4917
{
4918
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4919
	char str[64];
4920
	u64 val;
G
Glauber Costa 已提交
4921 4922
	int name, len;
	enum res_type type;
4923 4924 4925

	type = MEMFILE_TYPE(cft->private);
	name = MEMFILE_ATTR(cft->private);
4926

4927 4928
	switch (type) {
	case _MEM:
4929
		if (name == RES_USAGE)
4930
			val = mem_cgroup_usage(memcg, false);
4931
		else
4932
			val = res_counter_read_u64(&memcg->res, name);
4933 4934
		break;
	case _MEMSWAP:
4935
		if (name == RES_USAGE)
4936
			val = mem_cgroup_usage(memcg, true);
4937
		else
4938
			val = res_counter_read_u64(&memcg->memsw, name);
4939
		break;
4940 4941 4942
	case _KMEM:
		val = res_counter_read_u64(&memcg->kmem, name);
		break;
4943 4944 4945
	default:
		BUG();
	}
4946 4947 4948

	len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
	return simple_read_from_buffer(buf, nbytes, ppos, str, len);
B
Balbir Singh 已提交
4949
}
4950

4951
static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val)
4952 4953 4954
{
	int ret = -EINVAL;
#ifdef CONFIG_MEMCG_KMEM
4955
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967
	/*
	 * For simplicity, we won't allow this to be disabled.  It also can't
	 * be changed if the cgroup has children already, or if tasks had
	 * already joined.
	 *
	 * If tasks join before we set the limit, a person looking at
	 * kmem.usage_in_bytes will have no way to determine when it took
	 * place, which makes the value quite meaningless.
	 *
	 * After it first became limited, changes in the value of the limit are
	 * of course permitted.
	 */
4968
	mutex_lock(&memcg_create_mutex);
4969
	mutex_lock(&set_limit_mutex);
4970
	if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) {
4971
		if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) {
4972 4973 4974 4975 4976 4977
			ret = -EBUSY;
			goto out;
		}
		ret = res_counter_set_limit(&memcg->kmem, val);
		VM_BUG_ON(ret);

4978 4979
		ret = memcg_update_cache_sizes(memcg);
		if (ret) {
4980
			res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX);
4981 4982
			goto out;
		}
4983 4984 4985 4986 4987 4988
		static_key_slow_inc(&memcg_kmem_enabled_key);
		/*
		 * setting the active bit after the inc will guarantee no one
		 * starts accounting before all call sites are patched
		 */
		memcg_kmem_set_active(memcg);
4989 4990 4991 4992
	} else
		ret = res_counter_set_limit(&memcg->kmem, val);
out:
	mutex_unlock(&set_limit_mutex);
4993
	mutex_unlock(&memcg_create_mutex);
4994 4995 4996 4997
#endif
	return ret;
}

4998
#ifdef CONFIG_MEMCG_KMEM
4999
static int memcg_propagate_kmem(struct mem_cgroup *memcg)
5000
{
5001
	int ret = 0;
5002 5003
	struct mem_cgroup *parent = parent_mem_cgroup(memcg);
	if (!parent)
5004 5005
		goto out;

5006
	memcg->kmem_account_flags = parent->kmem_account_flags;
5007 5008 5009 5010 5011 5012 5013 5014 5015 5016
	/*
	 * When that happen, we need to disable the static branch only on those
	 * memcgs that enabled it. To achieve this, we would be forced to
	 * complicate the code by keeping track of which memcgs were the ones
	 * that actually enabled limits, and which ones got it from its
	 * parents.
	 *
	 * It is a lot simpler just to do static_key_slow_inc() on every child
	 * that is accounted.
	 */
5017 5018 5019 5020
	if (!memcg_kmem_is_active(memcg))
		goto out;

	/*
5021 5022 5023
	 * __mem_cgroup_free() will issue static_key_slow_dec() because this
	 * memcg is active already. If the later initialization fails then the
	 * cgroup core triggers the cleanup so we do not have to do it here.
5024 5025 5026 5027
	 */
	static_key_slow_inc(&memcg_kmem_enabled_key);

	mutex_lock(&set_limit_mutex);
5028
	memcg_stop_kmem_account();
5029
	ret = memcg_update_cache_sizes(memcg);
5030
	memcg_resume_kmem_account();
5031 5032 5033
	mutex_unlock(&set_limit_mutex);
out:
	return ret;
5034
}
5035
#endif /* CONFIG_MEMCG_KMEM */
5036

5037 5038 5039 5040
/*
 * The user of this function is...
 * RES_LIMIT.
 */
5041
static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
5042
			    const char *buffer)
B
Balbir Singh 已提交
5043
{
5044
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
G
Glauber Costa 已提交
5045 5046
	enum res_type type;
	int name;
5047 5048 5049
	unsigned long long val;
	int ret;

5050 5051
	type = MEMFILE_TYPE(cft->private);
	name = MEMFILE_ATTR(cft->private);
5052

5053
	switch (name) {
5054
	case RES_LIMIT:
5055 5056 5057 5058
		if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
			ret = -EINVAL;
			break;
		}
5059 5060
		/* This function does all necessary parse...reuse it */
		ret = res_counter_memparse_write_strategy(buffer, &val);
5061 5062 5063
		if (ret)
			break;
		if (type == _MEM)
5064
			ret = mem_cgroup_resize_limit(memcg, val);
5065
		else if (type == _MEMSWAP)
5066
			ret = mem_cgroup_resize_memsw_limit(memcg, val);
5067
		else if (type == _KMEM)
5068
			ret = memcg_update_kmem_limit(css, val);
5069 5070
		else
			return -EINVAL;
5071
		break;
5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085
	case RES_SOFT_LIMIT:
		ret = res_counter_memparse_write_strategy(buffer, &val);
		if (ret)
			break;
		/*
		 * For memsw, soft limits are hard to implement in terms
		 * of semantics, for now, we support soft limits for
		 * control without swap
		 */
		if (type == _MEM)
			ret = res_counter_set_soft_limit(&memcg->res, val);
		else
			ret = -EINVAL;
		break;
5086 5087 5088 5089 5090
	default:
		ret = -EINVAL; /* should be BUG() ? */
		break;
	}
	return ret;
B
Balbir Singh 已提交
5091 5092
}

5093 5094 5095 5096 5097 5098 5099 5100 5101 5102
static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
		unsigned long long *mem_limit, unsigned long long *memsw_limit)
{
	unsigned long long min_limit, min_memsw_limit, tmp;

	min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
	min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
	if (!memcg->use_hierarchy)
		goto out;

T
Tejun Heo 已提交
5103 5104
	while (css_parent(&memcg->css)) {
		memcg = mem_cgroup_from_css(css_parent(&memcg->css));
5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116
		if (!memcg->use_hierarchy)
			break;
		tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
		min_limit = min(min_limit, tmp);
		tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
		min_memsw_limit = min(min_memsw_limit, tmp);
	}
out:
	*mem_limit = min_limit;
	*memsw_limit = min_memsw_limit;
}

5117
static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
5118
{
5119
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
G
Glauber Costa 已提交
5120 5121
	int name;
	enum res_type type;
5122

5123 5124
	type = MEMFILE_TYPE(event);
	name = MEMFILE_ATTR(event);
5125

5126
	switch (name) {
5127
	case RES_MAX_USAGE:
5128
		if (type == _MEM)
5129
			res_counter_reset_max(&memcg->res);
5130
		else if (type == _MEMSWAP)
5131
			res_counter_reset_max(&memcg->memsw);
5132 5133 5134 5135
		else if (type == _KMEM)
			res_counter_reset_max(&memcg->kmem);
		else
			return -EINVAL;
5136 5137
		break;
	case RES_FAILCNT:
5138
		if (type == _MEM)
5139
			res_counter_reset_failcnt(&memcg->res);
5140
		else if (type == _MEMSWAP)
5141
			res_counter_reset_failcnt(&memcg->memsw);
5142 5143 5144 5145
		else if (type == _KMEM)
			res_counter_reset_failcnt(&memcg->kmem);
		else
			return -EINVAL;
5146 5147
		break;
	}
5148

5149
	return 0;
5150 5151
}

5152
static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
5153 5154
					struct cftype *cft)
{
5155
	return mem_cgroup_from_css(css)->move_charge_at_immigrate;
5156 5157
}

5158
#ifdef CONFIG_MMU
5159
static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
5160 5161
					struct cftype *cft, u64 val)
{
5162
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5163 5164 5165

	if (val >= (1 << NR_MOVE_TYPE))
		return -EINVAL;
5166

5167
	/*
5168 5169 5170 5171
	 * No kind of locking is needed in here, because ->can_attach() will
	 * check this value once in the beginning of the process, and then carry
	 * on with stale data. This means that changes to this value will only
	 * affect task migrations starting after the change.
5172
	 */
5173
	memcg->move_charge_at_immigrate = val;
5174 5175
	return 0;
}
5176
#else
5177
static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
5178 5179 5180 5181 5182
					struct cftype *cft, u64 val)
{
	return -ENOSYS;
}
#endif
5183

5184
#ifdef CONFIG_NUMA
5185 5186
static int memcg_numa_stat_show(struct cgroup_subsys_state *css,
				struct cftype *cft, struct seq_file *m)
5187 5188 5189 5190
{
	int nid;
	unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
	unsigned long node_nr;
5191
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5192

5193
	total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
5194
	seq_printf(m, "total=%lu", total_nr);
5195
	for_each_node_state(nid, N_MEMORY) {
5196
		node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
5197 5198 5199 5200
		seq_printf(m, " N%d=%lu", nid, node_nr);
	}
	seq_putc(m, '\n');

5201
	file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
5202
	seq_printf(m, "file=%lu", file_nr);
5203
	for_each_node_state(nid, N_MEMORY) {
5204
		node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5205
				LRU_ALL_FILE);
5206 5207 5208 5209
		seq_printf(m, " N%d=%lu", nid, node_nr);
	}
	seq_putc(m, '\n');

5210
	anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
5211
	seq_printf(m, "anon=%lu", anon_nr);
5212
	for_each_node_state(nid, N_MEMORY) {
5213
		node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5214
				LRU_ALL_ANON);
5215 5216 5217 5218
		seq_printf(m, " N%d=%lu", nid, node_nr);
	}
	seq_putc(m, '\n');

5219
	unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
5220
	seq_printf(m, "unevictable=%lu", unevictable_nr);
5221
	for_each_node_state(nid, N_MEMORY) {
5222
		node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5223
				BIT(LRU_UNEVICTABLE));
5224 5225 5226 5227 5228 5229 5230
		seq_printf(m, " N%d=%lu", nid, node_nr);
	}
	seq_putc(m, '\n');
	return 0;
}
#endif /* CONFIG_NUMA */

5231 5232 5233 5234 5235
static inline void mem_cgroup_lru_names_not_uptodate(void)
{
	BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
}

5236
static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft,
5237
				 struct seq_file *m)
5238
{
5239
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5240 5241
	struct mem_cgroup *mi;
	unsigned int i;
5242

5243
	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5244
		if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
5245
			continue;
5246 5247
		seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
			   mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
5248
	}
L
Lee Schermerhorn 已提交
5249

5250 5251 5252 5253 5254 5255 5256 5257
	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
		seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
			   mem_cgroup_read_events(memcg, i));

	for (i = 0; i < NR_LRU_LISTS; i++)
		seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
			   mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);

K
KAMEZAWA Hiroyuki 已提交
5258
	/* Hierarchical information */
5259 5260
	{
		unsigned long long limit, memsw_limit;
5261
		memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
5262
		seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
5263
		if (do_swap_account)
5264 5265
			seq_printf(m, "hierarchical_memsw_limit %llu\n",
				   memsw_limit);
5266
	}
K
KOSAKI Motohiro 已提交
5267

5268 5269 5270
	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
		long long val = 0;

5271
		if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
5272
			continue;
5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292
		for_each_mem_cgroup_tree(mi, memcg)
			val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
		seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
	}

	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
		unsigned long long val = 0;

		for_each_mem_cgroup_tree(mi, memcg)
			val += mem_cgroup_read_events(mi, i);
		seq_printf(m, "total_%s %llu\n",
			   mem_cgroup_events_names[i], val);
	}

	for (i = 0; i < NR_LRU_LISTS; i++) {
		unsigned long long val = 0;

		for_each_mem_cgroup_tree(mi, memcg)
			val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
		seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
5293
	}
K
KAMEZAWA Hiroyuki 已提交
5294

K
KOSAKI Motohiro 已提交
5295 5296 5297 5298
#ifdef CONFIG_DEBUG_VM
	{
		int nid, zid;
		struct mem_cgroup_per_zone *mz;
5299
		struct zone_reclaim_stat *rstat;
K
KOSAKI Motohiro 已提交
5300 5301 5302 5303 5304
		unsigned long recent_rotated[2] = {0, 0};
		unsigned long recent_scanned[2] = {0, 0};

		for_each_online_node(nid)
			for (zid = 0; zid < MAX_NR_ZONES; zid++) {
5305
				mz = mem_cgroup_zoneinfo(memcg, nid, zid);
5306
				rstat = &mz->lruvec.reclaim_stat;
K
KOSAKI Motohiro 已提交
5307

5308 5309 5310 5311
				recent_rotated[0] += rstat->recent_rotated[0];
				recent_rotated[1] += rstat->recent_rotated[1];
				recent_scanned[0] += rstat->recent_scanned[0];
				recent_scanned[1] += rstat->recent_scanned[1];
K
KOSAKI Motohiro 已提交
5312
			}
5313 5314 5315 5316
		seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
		seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
		seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
		seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
K
KOSAKI Motohiro 已提交
5317 5318 5319
	}
#endif

5320 5321 5322
	return 0;
}

5323 5324
static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
				      struct cftype *cft)
K
KOSAKI Motohiro 已提交
5325
{
5326
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
K
KOSAKI Motohiro 已提交
5327

5328
	return mem_cgroup_swappiness(memcg);
K
KOSAKI Motohiro 已提交
5329 5330
}

5331 5332
static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
				       struct cftype *cft, u64 val)
K
KOSAKI Motohiro 已提交
5333
{
5334
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
T
Tejun Heo 已提交
5335
	struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
K
KOSAKI Motohiro 已提交
5336

T
Tejun Heo 已提交
5337
	if (val > 100 || !parent)
K
KOSAKI Motohiro 已提交
5338 5339
		return -EINVAL;

5340
	mutex_lock(&memcg_create_mutex);
5341

K
KOSAKI Motohiro 已提交
5342
	/* If under hierarchy, only empty-root can set this value */
5343
	if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
5344
		mutex_unlock(&memcg_create_mutex);
K
KOSAKI Motohiro 已提交
5345
		return -EINVAL;
5346
	}
K
KOSAKI Motohiro 已提交
5347 5348 5349

	memcg->swappiness = val;

5350
	mutex_unlock(&memcg_create_mutex);
5351

K
KOSAKI Motohiro 已提交
5352 5353 5354
	return 0;
}

5355 5356 5357 5358 5359 5360 5361 5362
static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
{
	struct mem_cgroup_threshold_ary *t;
	u64 usage;
	int i;

	rcu_read_lock();
	if (!swap)
5363
		t = rcu_dereference(memcg->thresholds.primary);
5364
	else
5365
		t = rcu_dereference(memcg->memsw_thresholds.primary);
5366 5367 5368 5369 5370 5371 5372

	if (!t)
		goto unlock;

	usage = mem_cgroup_usage(memcg, swap);

	/*
5373
	 * current_threshold points to threshold just below or equal to usage.
5374 5375 5376
	 * If it's not true, a threshold was crossed after last
	 * call of __mem_cgroup_threshold().
	 */
5377
	i = t->current_threshold;
5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400

	/*
	 * Iterate backward over array of thresholds starting from
	 * current_threshold and check if a threshold is crossed.
	 * If none of thresholds below usage is crossed, we read
	 * only one element of the array here.
	 */
	for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
		eventfd_signal(t->entries[i].eventfd, 1);

	/* i = current_threshold + 1 */
	i++;

	/*
	 * Iterate forward over array of thresholds starting from
	 * current_threshold+1 and check if a threshold is crossed.
	 * If none of thresholds above usage is crossed, we read
	 * only one element of the array here.
	 */
	for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
		eventfd_signal(t->entries[i].eventfd, 1);

	/* Update current_threshold */
5401
	t->current_threshold = i - 1;
5402 5403 5404 5405 5406 5407
unlock:
	rcu_read_unlock();
}

static void mem_cgroup_threshold(struct mem_cgroup *memcg)
{
5408 5409 5410 5411 5412 5413 5414
	while (memcg) {
		__mem_cgroup_threshold(memcg, false);
		if (do_swap_account)
			__mem_cgroup_threshold(memcg, true);

		memcg = parent_mem_cgroup(memcg);
	}
5415 5416 5417 5418 5419 5420 5421
}

static int compare_thresholds(const void *a, const void *b)
{
	const struct mem_cgroup_threshold *_a = a;
	const struct mem_cgroup_threshold *_b = b;

5422 5423 5424 5425 5426 5427 5428
	if (_a->threshold > _b->threshold)
		return 1;

	if (_a->threshold < _b->threshold)
		return -1;

	return 0;
5429 5430
}

5431
static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
K
KAMEZAWA Hiroyuki 已提交
5432 5433 5434
{
	struct mem_cgroup_eventfd_list *ev;

5435
	list_for_each_entry(ev, &memcg->oom_notify, list)
K
KAMEZAWA Hiroyuki 已提交
5436 5437 5438 5439
		eventfd_signal(ev->eventfd, 1);
	return 0;
}

5440
static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
K
KAMEZAWA Hiroyuki 已提交
5441
{
K
KAMEZAWA Hiroyuki 已提交
5442 5443
	struct mem_cgroup *iter;

5444
	for_each_mem_cgroup_tree(iter, memcg)
K
KAMEZAWA Hiroyuki 已提交
5445
		mem_cgroup_oom_notify_cb(iter);
K
KAMEZAWA Hiroyuki 已提交
5446 5447
}

5448
static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css,
K
KAMEZAWA Hiroyuki 已提交
5449
	struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5450
{
5451
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5452 5453
	struct mem_cgroup_thresholds *thresholds;
	struct mem_cgroup_threshold_ary *new;
G
Glauber Costa 已提交
5454
	enum res_type type = MEMFILE_TYPE(cft->private);
5455
	u64 threshold, usage;
5456
	int i, size, ret;
5457 5458 5459 5460 5461 5462

	ret = res_counter_memparse_write_strategy(args, &threshold);
	if (ret)
		return ret;

	mutex_lock(&memcg->thresholds_lock);
5463

5464
	if (type == _MEM)
5465
		thresholds = &memcg->thresholds;
5466
	else if (type == _MEMSWAP)
5467
		thresholds = &memcg->memsw_thresholds;
5468 5469 5470 5471 5472 5473
	else
		BUG();

	usage = mem_cgroup_usage(memcg, type == _MEMSWAP);

	/* Check if a threshold crossed before adding a new one */
5474
	if (thresholds->primary)
5475 5476
		__mem_cgroup_threshold(memcg, type == _MEMSWAP);

5477
	size = thresholds->primary ? thresholds->primary->size + 1 : 1;
5478 5479

	/* Allocate memory for new array of thresholds */
5480
	new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
5481
			GFP_KERNEL);
5482
	if (!new) {
5483 5484 5485
		ret = -ENOMEM;
		goto unlock;
	}
5486
	new->size = size;
5487 5488

	/* Copy thresholds (if any) to new array */
5489 5490
	if (thresholds->primary) {
		memcpy(new->entries, thresholds->primary->entries, (size - 1) *
5491
				sizeof(struct mem_cgroup_threshold));
5492 5493
	}

5494
	/* Add new threshold */
5495 5496
	new->entries[size - 1].eventfd = eventfd;
	new->entries[size - 1].threshold = threshold;
5497 5498

	/* Sort thresholds. Registering of new threshold isn't time-critical */
5499
	sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
5500 5501 5502
			compare_thresholds, NULL);

	/* Find current threshold */
5503
	new->current_threshold = -1;
5504
	for (i = 0; i < size; i++) {
5505
		if (new->entries[i].threshold <= usage) {
5506
			/*
5507 5508
			 * new->current_threshold will not be used until
			 * rcu_assign_pointer(), so it's safe to increment
5509 5510
			 * it here.
			 */
5511
			++new->current_threshold;
5512 5513
		} else
			break;
5514 5515
	}

5516 5517 5518 5519 5520
	/* Free old spare buffer and save old primary buffer as spare */
	kfree(thresholds->spare);
	thresholds->spare = thresholds->primary;

	rcu_assign_pointer(thresholds->primary, new);
5521

5522
	/* To be sure that nobody uses thresholds */
5523 5524 5525 5526 5527 5528 5529 5530
	synchronize_rcu();

unlock:
	mutex_unlock(&memcg->thresholds_lock);

	return ret;
}

5531
static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css,
K
KAMEZAWA Hiroyuki 已提交
5532
	struct cftype *cft, struct eventfd_ctx *eventfd)
5533
{
5534
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5535 5536
	struct mem_cgroup_thresholds *thresholds;
	struct mem_cgroup_threshold_ary *new;
G
Glauber Costa 已提交
5537
	enum res_type type = MEMFILE_TYPE(cft->private);
5538
	u64 usage;
5539
	int i, j, size;
5540 5541 5542

	mutex_lock(&memcg->thresholds_lock);
	if (type == _MEM)
5543
		thresholds = &memcg->thresholds;
5544
	else if (type == _MEMSWAP)
5545
		thresholds = &memcg->memsw_thresholds;
5546 5547 5548
	else
		BUG();

5549 5550 5551
	if (!thresholds->primary)
		goto unlock;

5552 5553 5554 5555 5556 5557
	usage = mem_cgroup_usage(memcg, type == _MEMSWAP);

	/* Check if a threshold crossed before removing */
	__mem_cgroup_threshold(memcg, type == _MEMSWAP);

	/* Calculate new number of threshold */
5558 5559 5560
	size = 0;
	for (i = 0; i < thresholds->primary->size; i++) {
		if (thresholds->primary->entries[i].eventfd != eventfd)
5561 5562 5563
			size++;
	}

5564
	new = thresholds->spare;
5565

5566 5567
	/* Set thresholds array to NULL if we don't have thresholds */
	if (!size) {
5568 5569
		kfree(new);
		new = NULL;
5570
		goto swap_buffers;
5571 5572
	}

5573
	new->size = size;
5574 5575

	/* Copy thresholds and find current threshold */
5576 5577 5578
	new->current_threshold = -1;
	for (i = 0, j = 0; i < thresholds->primary->size; i++) {
		if (thresholds->primary->entries[i].eventfd == eventfd)
5579 5580
			continue;

5581
		new->entries[j] = thresholds->primary->entries[i];
5582
		if (new->entries[j].threshold <= usage) {
5583
			/*
5584
			 * new->current_threshold will not be used
5585 5586 5587
			 * until rcu_assign_pointer(), so it's safe to increment
			 * it here.
			 */
5588
			++new->current_threshold;
5589 5590 5591 5592
		}
		j++;
	}

5593
swap_buffers:
5594 5595
	/* Swap primary and spare array */
	thresholds->spare = thresholds->primary;
5596 5597 5598 5599 5600 5601
	/* If all events are unregistered, free the spare array */
	if (!new) {
		kfree(thresholds->spare);
		thresholds->spare = NULL;
	}

5602
	rcu_assign_pointer(thresholds->primary, new);
5603

5604
	/* To be sure that nobody uses thresholds */
5605
	synchronize_rcu();
5606
unlock:
5607 5608
	mutex_unlock(&memcg->thresholds_lock);
}
5609

5610
static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css,
K
KAMEZAWA Hiroyuki 已提交
5611 5612
	struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
{
5613
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
K
KAMEZAWA Hiroyuki 已提交
5614
	struct mem_cgroup_eventfd_list *event;
G
Glauber Costa 已提交
5615
	enum res_type type = MEMFILE_TYPE(cft->private);
K
KAMEZAWA Hiroyuki 已提交
5616 5617 5618 5619 5620 5621

	BUG_ON(type != _OOM_TYPE);
	event = kmalloc(sizeof(*event),	GFP_KERNEL);
	if (!event)
		return -ENOMEM;

5622
	spin_lock(&memcg_oom_lock);
K
KAMEZAWA Hiroyuki 已提交
5623 5624 5625 5626 5627

	event->eventfd = eventfd;
	list_add(&event->list, &memcg->oom_notify);

	/* already in OOM ? */
5628
	if (atomic_read(&memcg->under_oom))
K
KAMEZAWA Hiroyuki 已提交
5629
		eventfd_signal(eventfd, 1);
5630
	spin_unlock(&memcg_oom_lock);
K
KAMEZAWA Hiroyuki 已提交
5631 5632 5633 5634

	return 0;
}

5635
static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css,
K
KAMEZAWA Hiroyuki 已提交
5636 5637
	struct cftype *cft, struct eventfd_ctx *eventfd)
{
5638
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
K
KAMEZAWA Hiroyuki 已提交
5639
	struct mem_cgroup_eventfd_list *ev, *tmp;
G
Glauber Costa 已提交
5640
	enum res_type type = MEMFILE_TYPE(cft->private);
K
KAMEZAWA Hiroyuki 已提交
5641 5642 5643

	BUG_ON(type != _OOM_TYPE);

5644
	spin_lock(&memcg_oom_lock);
K
KAMEZAWA Hiroyuki 已提交
5645

5646
	list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
K
KAMEZAWA Hiroyuki 已提交
5647 5648 5649 5650 5651 5652
		if (ev->eventfd == eventfd) {
			list_del(&ev->list);
			kfree(ev);
		}
	}

5653
	spin_unlock(&memcg_oom_lock);
K
KAMEZAWA Hiroyuki 已提交
5654 5655
}

5656
static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css,
5657 5658
	struct cftype *cft,  struct cgroup_map_cb *cb)
{
5659
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5660

5661
	cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
5662

5663
	if (atomic_read(&memcg->under_oom))
5664 5665 5666 5667 5668 5669
		cb->fill(cb, "under_oom", 1);
	else
		cb->fill(cb, "under_oom", 0);
	return 0;
}

5670
static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
5671 5672
	struct cftype *cft, u64 val)
{
5673
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
T
Tejun Heo 已提交
5674
	struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
5675 5676

	/* cannot set to root cgroup and only 0 and 1 are allowed */
T
Tejun Heo 已提交
5677
	if (!parent || !((val == 0) || (val == 1)))
5678 5679
		return -EINVAL;

5680
	mutex_lock(&memcg_create_mutex);
5681
	/* oom-kill-disable is a flag for subhierarchy. */
5682
	if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
5683
		mutex_unlock(&memcg_create_mutex);
5684 5685
		return -EINVAL;
	}
5686
	memcg->oom_kill_disable = val;
5687
	if (!val)
5688
		memcg_oom_recover(memcg);
5689
	mutex_unlock(&memcg_create_mutex);
5690 5691 5692
	return 0;
}

A
Andrew Morton 已提交
5693
#ifdef CONFIG_MEMCG_KMEM
5694
static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
5695
{
5696 5697
	int ret;

5698
	memcg->kmemcg_id = -1;
5699 5700 5701
	ret = memcg_propagate_kmem(memcg);
	if (ret)
		return ret;
5702

5703
	return mem_cgroup_sockets_init(memcg, ss);
5704
}
5705

5706
static void memcg_destroy_kmem(struct mem_cgroup *memcg)
G
Glauber Costa 已提交
5707
{
5708
	mem_cgroup_sockets_destroy(memcg);
5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734
}

static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
{
	if (!memcg_kmem_is_active(memcg))
		return;

	/*
	 * kmem charges can outlive the cgroup. In the case of slab
	 * pages, for instance, a page contain objects from various
	 * processes. As we prevent from taking a reference for every
	 * such allocation we have to be careful when doing uncharge
	 * (see memcg_uncharge_kmem) and here during offlining.
	 *
	 * The idea is that that only the _last_ uncharge which sees
	 * the dead memcg will drop the last reference. An additional
	 * reference is taken here before the group is marked dead
	 * which is then paired with css_put during uncharge resp. here.
	 *
	 * Although this might sound strange as this path is called from
	 * css_offline() when the referencemight have dropped down to 0
	 * and shouldn't be incremented anymore (css_tryget would fail)
	 * we do not have other options because of the kmem allocations
	 * lifetime.
	 */
	css_get(&memcg->css);
5735 5736 5737 5738 5739 5740 5741

	memcg_kmem_mark_dead(memcg);

	if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
		return;

	if (memcg_kmem_test_and_clear_dead(memcg))
5742
		css_put(&memcg->css);
G
Glauber Costa 已提交
5743
}
5744
#else
5745
static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
5746 5747 5748
{
	return 0;
}
G
Glauber Costa 已提交
5749

5750 5751 5752 5753 5754
static void memcg_destroy_kmem(struct mem_cgroup *memcg)
{
}

static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
G
Glauber Costa 已提交
5755 5756
{
}
5757 5758
#endif

B
Balbir Singh 已提交
5759 5760
static struct cftype mem_cgroup_files[] = {
	{
5761
		.name = "usage_in_bytes",
5762
		.private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
5763
		.read = mem_cgroup_read,
K
KAMEZAWA Hiroyuki 已提交
5764 5765
		.register_event = mem_cgroup_usage_register_event,
		.unregister_event = mem_cgroup_usage_unregister_event,
B
Balbir Singh 已提交
5766
	},
5767 5768
	{
		.name = "max_usage_in_bytes",
5769
		.private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
5770
		.trigger = mem_cgroup_reset,
5771
		.read = mem_cgroup_read,
5772
	},
B
Balbir Singh 已提交
5773
	{
5774
		.name = "limit_in_bytes",
5775
		.private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
5776
		.write_string = mem_cgroup_write,
5777
		.read = mem_cgroup_read,
B
Balbir Singh 已提交
5778
	},
5779 5780 5781 5782
	{
		.name = "soft_limit_in_bytes",
		.private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
		.write_string = mem_cgroup_write,
5783
		.read = mem_cgroup_read,
5784
	},
B
Balbir Singh 已提交
5785 5786
	{
		.name = "failcnt",
5787
		.private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
5788
		.trigger = mem_cgroup_reset,
5789
		.read = mem_cgroup_read,
B
Balbir Singh 已提交
5790
	},
5791 5792
	{
		.name = "stat",
5793
		.read_seq_string = memcg_stat_show,
5794
	},
5795 5796 5797 5798
	{
		.name = "force_empty",
		.trigger = mem_cgroup_force_empty_write,
	},
5799 5800
	{
		.name = "use_hierarchy",
5801
		.flags = CFTYPE_INSANE,
5802 5803 5804
		.write_u64 = mem_cgroup_hierarchy_write,
		.read_u64 = mem_cgroup_hierarchy_read,
	},
K
KOSAKI Motohiro 已提交
5805 5806 5807 5808 5809
	{
		.name = "swappiness",
		.read_u64 = mem_cgroup_swappiness_read,
		.write_u64 = mem_cgroup_swappiness_write,
	},
5810 5811 5812 5813 5814
	{
		.name = "move_charge_at_immigrate",
		.read_u64 = mem_cgroup_move_charge_read,
		.write_u64 = mem_cgroup_move_charge_write,
	},
K
KAMEZAWA Hiroyuki 已提交
5815 5816
	{
		.name = "oom_control",
5817 5818
		.read_map = mem_cgroup_oom_control_read,
		.write_u64 = mem_cgroup_oom_control_write,
K
KAMEZAWA Hiroyuki 已提交
5819 5820 5821 5822
		.register_event = mem_cgroup_oom_register_event,
		.unregister_event = mem_cgroup_oom_unregister_event,
		.private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
	},
5823 5824 5825 5826 5827
	{
		.name = "pressure_level",
		.register_event = vmpressure_register_event,
		.unregister_event = vmpressure_unregister_event,
	},
5828 5829 5830
#ifdef CONFIG_NUMA
	{
		.name = "numa_stat",
5831
		.read_seq_string = memcg_numa_stat_show,
5832 5833
	},
#endif
5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857
#ifdef CONFIG_MEMCG_KMEM
	{
		.name = "kmem.limit_in_bytes",
		.private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
		.write_string = mem_cgroup_write,
		.read = mem_cgroup_read,
	},
	{
		.name = "kmem.usage_in_bytes",
		.private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
		.read = mem_cgroup_read,
	},
	{
		.name = "kmem.failcnt",
		.private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
		.trigger = mem_cgroup_reset,
		.read = mem_cgroup_read,
	},
	{
		.name = "kmem.max_usage_in_bytes",
		.private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
		.trigger = mem_cgroup_reset,
		.read = mem_cgroup_read,
	},
5858 5859 5860 5861 5862 5863
#ifdef CONFIG_SLABINFO
	{
		.name = "kmem.slabinfo",
		.read_seq_string = mem_cgroup_slabinfo_read,
	},
#endif
5864
#endif
5865
	{ },	/* terminate */
5866
};
5867

5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897
#ifdef CONFIG_MEMCG_SWAP
static struct cftype memsw_cgroup_files[] = {
	{
		.name = "memsw.usage_in_bytes",
		.private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
		.read = mem_cgroup_read,
		.register_event = mem_cgroup_usage_register_event,
		.unregister_event = mem_cgroup_usage_unregister_event,
	},
	{
		.name = "memsw.max_usage_in_bytes",
		.private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
		.trigger = mem_cgroup_reset,
		.read = mem_cgroup_read,
	},
	{
		.name = "memsw.limit_in_bytes",
		.private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
		.write_string = mem_cgroup_write,
		.read = mem_cgroup_read,
	},
	{
		.name = "memsw.failcnt",
		.private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
		.trigger = mem_cgroup_reset,
		.read = mem_cgroup_read,
	},
	{ },	/* terminate */
};
#endif
5898
static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
5899 5900
{
	struct mem_cgroup_per_node *pn;
5901
	struct mem_cgroup_per_zone *mz;
5902
	int zone, tmp = node;
5903 5904 5905 5906 5907 5908 5909 5910
	/*
	 * This routine is called against possible nodes.
	 * But it's BUG to call kmalloc() against offline node.
	 *
	 * TODO: this routine can waste much memory for nodes which will
	 *       never be onlined. It's better to use memory hotplug callback
	 *       function.
	 */
5911 5912
	if (!node_state(node, N_NORMAL_MEMORY))
		tmp = -1;
5913
	pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
5914 5915
	if (!pn)
		return 1;
5916 5917 5918

	for (zone = 0; zone < MAX_NR_ZONES; zone++) {
		mz = &pn->zoneinfo[zone];
5919
		lruvec_init(&mz->lruvec);
5920
		mz->memcg = memcg;
5921
	}
5922
	memcg->nodeinfo[node] = pn;
5923 5924 5925
	return 0;
}

5926
static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
5927
{
5928
	kfree(memcg->nodeinfo[node]);
5929 5930
}

5931 5932
static struct mem_cgroup *mem_cgroup_alloc(void)
{
5933
	struct mem_cgroup *memcg;
5934
	size_t size = memcg_size();
5935

5936
	/* Can be very big if nr_node_ids is very big */
5937
	if (size < PAGE_SIZE)
5938
		memcg = kzalloc(size, GFP_KERNEL);
5939
	else
5940
		memcg = vzalloc(size);
5941

5942
	if (!memcg)
5943 5944
		return NULL;

5945 5946
	memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
	if (!memcg->stat)
5947
		goto out_free;
5948 5949
	spin_lock_init(&memcg->pcp_counter_lock);
	return memcg;
5950 5951 5952

out_free:
	if (size < PAGE_SIZE)
5953
		kfree(memcg);
5954
	else
5955
		vfree(memcg);
5956
	return NULL;
5957 5958
}

5959
/*
5960 5961 5962 5963 5964 5965 5966 5967
 * At destroying mem_cgroup, references from swap_cgroup can remain.
 * (scanning all at force_empty is too costly...)
 *
 * Instead of clearing all references at force_empty, we remember
 * the number of reference from swap_cgroup and free mem_cgroup when
 * it goes down to 0.
 *
 * Removal of cgroup itself succeeds regardless of refs from swap.
5968
 */
5969 5970

static void __mem_cgroup_free(struct mem_cgroup *memcg)
5971
{
5972
	int node;
5973
	size_t size = memcg_size();
5974

5975 5976 5977 5978 5979 5980 5981
	free_css_id(&mem_cgroup_subsys, &memcg->css);

	for_each_node(node)
		free_mem_cgroup_per_zone_info(memcg, node);

	free_percpu(memcg->stat);

5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992
	/*
	 * We need to make sure that (at least for now), the jump label
	 * destruction code runs outside of the cgroup lock. This is because
	 * get_online_cpus(), which is called from the static_branch update,
	 * can't be called inside the cgroup_lock. cpusets are the ones
	 * enforcing this dependency, so if they ever change, we might as well.
	 *
	 * schedule_work() will guarantee this happens. Be careful if you need
	 * to move this code around, and make sure it is outside
	 * the cgroup_lock.
	 */
5993
	disarm_static_keys(memcg);
5994 5995 5996 5997
	if (size < PAGE_SIZE)
		kfree(memcg);
	else
		vfree(memcg);
5998
}
5999

6000 6001 6002
/*
 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
 */
G
Glauber Costa 已提交
6003
struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
6004
{
6005
	if (!memcg->res.parent)
6006
		return NULL;
6007
	return mem_cgroup_from_res_counter(memcg->res.parent, res);
6008
}
G
Glauber Costa 已提交
6009
EXPORT_SYMBOL(parent_mem_cgroup);
6010

L
Li Zefan 已提交
6011
static struct cgroup_subsys_state * __ref
6012
mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
B
Balbir Singh 已提交
6013
{
6014
	struct mem_cgroup *memcg;
K
KAMEZAWA Hiroyuki 已提交
6015
	long error = -ENOMEM;
6016
	int node;
B
Balbir Singh 已提交
6017

6018 6019
	memcg = mem_cgroup_alloc();
	if (!memcg)
K
KAMEZAWA Hiroyuki 已提交
6020
		return ERR_PTR(error);
6021

B
Bob Liu 已提交
6022
	for_each_node(node)
6023
		if (alloc_mem_cgroup_per_zone_info(memcg, node))
6024
			goto free_out;
6025

6026
	/* root ? */
6027
	if (parent_css == NULL) {
6028
		root_mem_cgroup = memcg;
6029 6030 6031
		res_counter_init(&memcg->res, NULL);
		res_counter_init(&memcg->memsw, NULL);
		res_counter_init(&memcg->kmem, NULL);
6032
	}
6033

6034 6035 6036 6037 6038
	memcg->last_scanned_node = MAX_NUMNODES;
	INIT_LIST_HEAD(&memcg->oom_notify);
	memcg->move_charge_at_immigrate = 0;
	mutex_init(&memcg->thresholds_lock);
	spin_lock_init(&memcg->move_lock);
6039
	vmpressure_init(&memcg->vmpressure);
6040
	spin_lock_init(&memcg->soft_lock);
6041 6042 6043 6044 6045 6046 6047 6048 6049

	return &memcg->css;

free_out:
	__mem_cgroup_free(memcg);
	return ERR_PTR(error);
}

static int
6050
mem_cgroup_css_online(struct cgroup_subsys_state *css)
6051
{
6052 6053
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
	struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
6054 6055
	int error = 0;

T
Tejun Heo 已提交
6056
	if (!parent)
6057 6058
		return 0;

6059
	mutex_lock(&memcg_create_mutex);
6060 6061 6062 6063 6064 6065

	memcg->use_hierarchy = parent->use_hierarchy;
	memcg->oom_kill_disable = parent->oom_kill_disable;
	memcg->swappiness = mem_cgroup_swappiness(parent);

	if (parent->use_hierarchy) {
6066 6067
		res_counter_init(&memcg->res, &parent->res);
		res_counter_init(&memcg->memsw, &parent->memsw);
6068
		res_counter_init(&memcg->kmem, &parent->kmem);
6069

6070
		/*
6071 6072
		 * No need to take a reference to the parent because cgroup
		 * core guarantees its existence.
6073
		 */
6074
	} else {
6075 6076
		res_counter_init(&memcg->res, NULL);
		res_counter_init(&memcg->memsw, NULL);
6077
		res_counter_init(&memcg->kmem, NULL);
6078 6079 6080 6081 6082
		/*
		 * Deeper hierachy with use_hierarchy == false doesn't make
		 * much sense so let cgroup subsystem know about this
		 * unfortunate state in our controller.
		 */
6083
		if (parent != root_mem_cgroup)
6084
			mem_cgroup_subsys.broken_hierarchy = true;
6085
	}
6086 6087

	error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
6088
	mutex_unlock(&memcg_create_mutex);
6089
	return error;
B
Balbir Singh 已提交
6090 6091
}

M
Michal Hocko 已提交
6092 6093 6094 6095 6096 6097 6098 6099
/*
 * Announce all parents that a group from their hierarchy is gone.
 */
static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
{
	struct mem_cgroup *parent = memcg;

	while ((parent = parent_mem_cgroup(parent)))
6100
		mem_cgroup_iter_invalidate(parent);
M
Michal Hocko 已提交
6101 6102 6103 6104 6105 6106

	/*
	 * if the root memcg is not hierarchical we have to check it
	 * explicitely.
	 */
	if (!root_mem_cgroup->use_hierarchy)
6107
		mem_cgroup_iter_invalidate(root_mem_cgroup);
M
Michal Hocko 已提交
6108 6109
}

6110
static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
6111
{
6112
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6113

6114 6115
	kmem_cgroup_css_offline(memcg);

M
Michal Hocko 已提交
6116
	mem_cgroup_invalidate_reclaim_iterators(memcg);
6117
	mem_cgroup_reparent_charges(memcg);
6118 6119 6120
	if (memcg->soft_contributed) {
		while ((memcg = parent_mem_cgroup(memcg)))
			atomic_dec(&memcg->children_in_excess);
6121 6122 6123

		if (memcg != root_mem_cgroup && !root_mem_cgroup->use_hierarchy)
			atomic_dec(&root_mem_cgroup->children_in_excess);
6124
	}
G
Glauber Costa 已提交
6125
	mem_cgroup_destroy_all_caches(memcg);
6126
	vmpressure_cleanup(&memcg->vmpressure);
6127 6128
}

6129
static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
B
Balbir Singh 已提交
6130
{
6131
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6132

6133
	memcg_destroy_kmem(memcg);
6134
	__mem_cgroup_free(memcg);
B
Balbir Singh 已提交
6135 6136
}

6137
#ifdef CONFIG_MMU
6138
/* Handlers for move charge at task migration. */
6139 6140
#define PRECHARGE_COUNT_AT_ONCE	256
static int mem_cgroup_do_precharge(unsigned long count)
6141
{
6142 6143
	int ret = 0;
	int batch_count = PRECHARGE_COUNT_AT_ONCE;
6144
	struct mem_cgroup *memcg = mc.to;
6145

6146
	if (mem_cgroup_is_root(memcg)) {
6147 6148 6149 6150 6151 6152 6153 6154
		mc.precharge += count;
		/* we don't need css_get for root */
		return ret;
	}
	/* try to charge at once */
	if (count > 1) {
		struct res_counter *dummy;
		/*
6155
		 * "memcg" cannot be under rmdir() because we've already checked
6156 6157 6158 6159
		 * by cgroup_lock_live_cgroup() that it is not removed and we
		 * are still under the same cgroup_mutex. So we can postpone
		 * css_get().
		 */
6160
		if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
6161
			goto one_by_one;
6162
		if (do_swap_account && res_counter_charge(&memcg->memsw,
6163
						PAGE_SIZE * count, &dummy)) {
6164
			res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180
			goto one_by_one;
		}
		mc.precharge += count;
		return ret;
	}
one_by_one:
	/* fall back to one by one charge */
	while (count--) {
		if (signal_pending(current)) {
			ret = -EINTR;
			break;
		}
		if (!batch_count--) {
			batch_count = PRECHARGE_COUNT_AT_ONCE;
			cond_resched();
		}
6181 6182
		ret = __mem_cgroup_try_charge(NULL,
					GFP_KERNEL, 1, &memcg, false);
6183
		if (ret)
6184
			/* mem_cgroup_clear_mc() will do uncharge later */
6185
			return ret;
6186 6187
		mc.precharge++;
	}
6188 6189 6190 6191
	return ret;
}

/**
6192
 * get_mctgt_type - get target type of moving charge
6193 6194 6195
 * @vma: the vma the pte to be checked belongs
 * @addr: the address corresponding to the pte to be checked
 * @ptent: the pte to be checked
6196
 * @target: the pointer the target page or swap ent will be stored(can be NULL)
6197 6198 6199 6200 6201 6202
 *
 * Returns
 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
 *     move charge. if @target is not NULL, the page is stored in target->page
 *     with extra refcnt got(Callers should handle it).
6203 6204 6205
 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
 *     target for charge migration. if @target is not NULL, the entry is stored
 *     in target->ent.
6206 6207 6208 6209 6210
 *
 * Called with pte lock held.
 */
union mc_target {
	struct page	*page;
6211
	swp_entry_t	ent;
6212 6213 6214
};

enum mc_target_type {
6215
	MC_TARGET_NONE = 0,
6216
	MC_TARGET_PAGE,
6217
	MC_TARGET_SWAP,
6218 6219
};

D
Daisuke Nishimura 已提交
6220 6221
static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
						unsigned long addr, pte_t ptent)
6222
{
D
Daisuke Nishimura 已提交
6223
	struct page *page = vm_normal_page(vma, addr, ptent);
6224

D
Daisuke Nishimura 已提交
6225 6226 6227 6228
	if (!page || !page_mapped(page))
		return NULL;
	if (PageAnon(page)) {
		/* we don't move shared anon */
6229
		if (!move_anon())
D
Daisuke Nishimura 已提交
6230
			return NULL;
6231 6232
	} else if (!move_file())
		/* we ignore mapcount for file pages */
D
Daisuke Nishimura 已提交
6233 6234 6235 6236 6237 6238 6239
		return NULL;
	if (!get_page_unless_zero(page))
		return NULL;

	return page;
}

6240
#ifdef CONFIG_SWAP
D
Daisuke Nishimura 已提交
6241 6242 6243 6244 6245 6246 6247 6248
static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
			unsigned long addr, pte_t ptent, swp_entry_t *entry)
{
	struct page *page = NULL;
	swp_entry_t ent = pte_to_swp_entry(ptent);

	if (!move_anon() || non_swap_entry(ent))
		return NULL;
6249 6250 6251 6252
	/*
	 * Because lookup_swap_cache() updates some statistics counter,
	 * we call find_get_page() with swapper_space directly.
	 */
6253
	page = find_get_page(swap_address_space(ent), ent.val);
D
Daisuke Nishimura 已提交
6254 6255 6256 6257 6258
	if (do_swap_account)
		entry->val = ent.val;

	return page;
}
6259 6260 6261 6262 6263 6264 6265
#else
static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
			unsigned long addr, pte_t ptent, swp_entry_t *entry)
{
	return NULL;
}
#endif
D
Daisuke Nishimura 已提交
6266

6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285
static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
			unsigned long addr, pte_t ptent, swp_entry_t *entry)
{
	struct page *page = NULL;
	struct address_space *mapping;
	pgoff_t pgoff;

	if (!vma->vm_file) /* anonymous vma */
		return NULL;
	if (!move_file())
		return NULL;

	mapping = vma->vm_file->f_mapping;
	if (pte_none(ptent))
		pgoff = linear_page_index(vma, addr);
	else /* pte_file(ptent) is true */
		pgoff = pte_to_pgoff(ptent);

	/* page is moved even if it's not RSS of this task(page-faulted). */
6286 6287 6288 6289 6290 6291
	page = find_get_page(mapping, pgoff);

#ifdef CONFIG_SWAP
	/* shmem/tmpfs may report page out on swap: account for that too. */
	if (radix_tree_exceptional_entry(page)) {
		swp_entry_t swap = radix_to_swp_entry(page);
6292
		if (do_swap_account)
6293
			*entry = swap;
6294
		page = find_get_page(swap_address_space(swap), swap.val);
6295
	}
6296
#endif
6297 6298 6299
	return page;
}

6300
static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
D
Daisuke Nishimura 已提交
6301 6302 6303 6304
		unsigned long addr, pte_t ptent, union mc_target *target)
{
	struct page *page = NULL;
	struct page_cgroup *pc;
6305
	enum mc_target_type ret = MC_TARGET_NONE;
D
Daisuke Nishimura 已提交
6306 6307 6308 6309 6310 6311
	swp_entry_t ent = { .val = 0 };

	if (pte_present(ptent))
		page = mc_handle_present_pte(vma, addr, ptent);
	else if (is_swap_pte(ptent))
		page = mc_handle_swap_pte(vma, addr, ptent, &ent);
6312 6313
	else if (pte_none(ptent) || pte_file(ptent))
		page = mc_handle_file_pte(vma, addr, ptent, &ent);
D
Daisuke Nishimura 已提交
6314 6315

	if (!page && !ent.val)
6316
		return ret;
6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331
	if (page) {
		pc = lookup_page_cgroup(page);
		/*
		 * Do only loose check w/o page_cgroup lock.
		 * mem_cgroup_move_account() checks the pc is valid or not under
		 * the lock.
		 */
		if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
			ret = MC_TARGET_PAGE;
			if (target)
				target->page = page;
		}
		if (!ret || !target)
			put_page(page);
	}
D
Daisuke Nishimura 已提交
6332 6333
	/* There is a swap entry and a page doesn't exist or isn't charged */
	if (ent.val && !ret &&
6334
			css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
6335 6336 6337
		ret = MC_TARGET_SWAP;
		if (target)
			target->ent = ent;
6338 6339 6340 6341
	}
	return ret;
}

6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
/*
 * We don't consider swapping or file mapped pages because THP does not
 * support them for now.
 * Caller should make sure that pmd_trans_huge(pmd) is true.
 */
static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
		unsigned long addr, pmd_t pmd, union mc_target *target)
{
	struct page *page = NULL;
	struct page_cgroup *pc;
	enum mc_target_type ret = MC_TARGET_NONE;

	page = pmd_page(pmd);
	VM_BUG_ON(!page || !PageHead(page));
	if (!move_anon())
		return ret;
	pc = lookup_page_cgroup(page);
	if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
		ret = MC_TARGET_PAGE;
		if (target) {
			get_page(page);
			target->page = page;
		}
	}
	return ret;
}
#else
static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
		unsigned long addr, pmd_t pmd, union mc_target *target)
{
	return MC_TARGET_NONE;
}
#endif

6377 6378 6379 6380 6381 6382 6383 6384
static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
					unsigned long addr, unsigned long end,
					struct mm_walk *walk)
{
	struct vm_area_struct *vma = walk->private;
	pte_t *pte;
	spinlock_t *ptl;

6385 6386 6387 6388
	if (pmd_trans_huge_lock(pmd, vma) == 1) {
		if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
			mc.precharge += HPAGE_PMD_NR;
		spin_unlock(&vma->vm_mm->page_table_lock);
6389
		return 0;
6390
	}
6391

6392 6393
	if (pmd_trans_unstable(pmd))
		return 0;
6394 6395
	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
	for (; addr != end; pte++, addr += PAGE_SIZE)
6396
		if (get_mctgt_type(vma, addr, *pte, NULL))
6397 6398 6399 6400
			mc.precharge++;	/* increment precharge temporarily */
	pte_unmap_unlock(pte - 1, ptl);
	cond_resched();

6401 6402 6403
	return 0;
}

6404 6405 6406 6407 6408
static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
{
	unsigned long precharge;
	struct vm_area_struct *vma;

6409
	down_read(&mm->mmap_sem);
6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420
	for (vma = mm->mmap; vma; vma = vma->vm_next) {
		struct mm_walk mem_cgroup_count_precharge_walk = {
			.pmd_entry = mem_cgroup_count_precharge_pte_range,
			.mm = mm,
			.private = vma,
		};
		if (is_vm_hugetlb_page(vma))
			continue;
		walk_page_range(vma->vm_start, vma->vm_end,
					&mem_cgroup_count_precharge_walk);
	}
6421
	up_read(&mm->mmap_sem);
6422 6423 6424 6425 6426 6427 6428 6429 6430

	precharge = mc.precharge;
	mc.precharge = 0;

	return precharge;
}

static int mem_cgroup_precharge_mc(struct mm_struct *mm)
{
6431 6432 6433 6434 6435
	unsigned long precharge = mem_cgroup_count_precharge(mm);

	VM_BUG_ON(mc.moving_task);
	mc.moving_task = current;
	return mem_cgroup_do_precharge(precharge);
6436 6437
}

6438 6439
/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
static void __mem_cgroup_clear_mc(void)
6440
{
6441 6442
	struct mem_cgroup *from = mc.from;
	struct mem_cgroup *to = mc.to;
L
Li Zefan 已提交
6443
	int i;
6444

6445
	/* we must uncharge all the leftover precharges from mc.to */
6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456
	if (mc.precharge) {
		__mem_cgroup_cancel_charge(mc.to, mc.precharge);
		mc.precharge = 0;
	}
	/*
	 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
	 * we must uncharge here.
	 */
	if (mc.moved_charge) {
		__mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
		mc.moved_charge = 0;
6457
	}
6458 6459 6460 6461 6462 6463
	/* we must fixup refcnts and charges */
	if (mc.moved_swap) {
		/* uncharge swap account from the old cgroup */
		if (!mem_cgroup_is_root(mc.from))
			res_counter_uncharge(&mc.from->memsw,
						PAGE_SIZE * mc.moved_swap);
L
Li Zefan 已提交
6464 6465 6466

		for (i = 0; i < mc.moved_swap; i++)
			css_put(&mc.from->css);
6467 6468 6469 6470 6471 6472 6473 6474 6475

		if (!mem_cgroup_is_root(mc.to)) {
			/*
			 * we charged both to->res and to->memsw, so we should
			 * uncharge to->res.
			 */
			res_counter_uncharge(&mc.to->res,
						PAGE_SIZE * mc.moved_swap);
		}
L
Li Zefan 已提交
6476
		/* we've already done css_get(mc.to) */
6477 6478
		mc.moved_swap = 0;
	}
6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493
	memcg_oom_recover(from);
	memcg_oom_recover(to);
	wake_up_all(&mc.waitq);
}

static void mem_cgroup_clear_mc(void)
{
	struct mem_cgroup *from = mc.from;

	/*
	 * we must clear moving_task before waking up waiters at the end of
	 * task migration.
	 */
	mc.moving_task = NULL;
	__mem_cgroup_clear_mc();
6494
	spin_lock(&mc.lock);
6495 6496
	mc.from = NULL;
	mc.to = NULL;
6497
	spin_unlock(&mc.lock);
6498
	mem_cgroup_end_move(from);
6499 6500
}

6501
static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
6502
				 struct cgroup_taskset *tset)
6503
{
6504
	struct task_struct *p = cgroup_taskset_first(tset);
6505
	int ret = 0;
6506
	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6507
	unsigned long move_charge_at_immigrate;
6508

6509 6510 6511 6512 6513 6514 6515
	/*
	 * We are now commited to this value whatever it is. Changes in this
	 * tunable will only affect upcoming migrations, not the current one.
	 * So we need to save it, and keep it going.
	 */
	move_charge_at_immigrate  = memcg->move_charge_at_immigrate;
	if (move_charge_at_immigrate) {
6516 6517 6518
		struct mm_struct *mm;
		struct mem_cgroup *from = mem_cgroup_from_task(p);

6519
		VM_BUG_ON(from == memcg);
6520 6521 6522 6523 6524

		mm = get_task_mm(p);
		if (!mm)
			return 0;
		/* We move charges only when we move a owner of the mm */
6525 6526 6527 6528
		if (mm->owner == p) {
			VM_BUG_ON(mc.from);
			VM_BUG_ON(mc.to);
			VM_BUG_ON(mc.precharge);
6529
			VM_BUG_ON(mc.moved_charge);
6530
			VM_BUG_ON(mc.moved_swap);
6531
			mem_cgroup_start_move(from);
6532
			spin_lock(&mc.lock);
6533
			mc.from = from;
6534
			mc.to = memcg;
6535
			mc.immigrate_flags = move_charge_at_immigrate;
6536
			spin_unlock(&mc.lock);
6537
			/* We set mc.moving_task later */
6538 6539 6540 6541

			ret = mem_cgroup_precharge_mc(mm);
			if (ret)
				mem_cgroup_clear_mc();
6542 6543
		}
		mmput(mm);
6544 6545 6546 6547
	}
	return ret;
}

6548
static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
6549
				     struct cgroup_taskset *tset)
6550
{
6551
	mem_cgroup_clear_mc();
6552 6553
}

6554 6555 6556
static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
				unsigned long addr, unsigned long end,
				struct mm_walk *walk)
6557
{
6558 6559 6560 6561
	int ret = 0;
	struct vm_area_struct *vma = walk->private;
	pte_t *pte;
	spinlock_t *ptl;
6562 6563 6564 6565
	enum mc_target_type target_type;
	union mc_target target;
	struct page *page;
	struct page_cgroup *pc;
6566

6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577
	/*
	 * We don't take compound_lock() here but no race with splitting thp
	 * happens because:
	 *  - if pmd_trans_huge_lock() returns 1, the relevant thp is not
	 *    under splitting, which means there's no concurrent thp split,
	 *  - if another thread runs into split_huge_page() just after we
	 *    entered this if-block, the thread must wait for page table lock
	 *    to be unlocked in __split_huge_page_splitting(), where the main
	 *    part of thp split is not executed yet.
	 */
	if (pmd_trans_huge_lock(pmd, vma) == 1) {
6578
		if (mc.precharge < HPAGE_PMD_NR) {
6579 6580 6581 6582 6583 6584 6585 6586 6587
			spin_unlock(&vma->vm_mm->page_table_lock);
			return 0;
		}
		target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
		if (target_type == MC_TARGET_PAGE) {
			page = target.page;
			if (!isolate_lru_page(page)) {
				pc = lookup_page_cgroup(page);
				if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
6588
							pc, mc.from, mc.to)) {
6589 6590 6591 6592 6593 6594 6595 6596
					mc.precharge -= HPAGE_PMD_NR;
					mc.moved_charge += HPAGE_PMD_NR;
				}
				putback_lru_page(page);
			}
			put_page(page);
		}
		spin_unlock(&vma->vm_mm->page_table_lock);
6597
		return 0;
6598 6599
	}

6600 6601
	if (pmd_trans_unstable(pmd))
		return 0;
6602 6603 6604 6605
retry:
	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
	for (; addr != end; addr += PAGE_SIZE) {
		pte_t ptent = *(pte++);
6606
		swp_entry_t ent;
6607 6608 6609 6610

		if (!mc.precharge)
			break;

6611
		switch (get_mctgt_type(vma, addr, ptent, &target)) {
6612 6613 6614 6615 6616
		case MC_TARGET_PAGE:
			page = target.page;
			if (isolate_lru_page(page))
				goto put;
			pc = lookup_page_cgroup(page);
6617
			if (!mem_cgroup_move_account(page, 1, pc,
6618
						     mc.from, mc.to)) {
6619
				mc.precharge--;
6620 6621
				/* we uncharge from mc.from later. */
				mc.moved_charge++;
6622 6623
			}
			putback_lru_page(page);
6624
put:			/* get_mctgt_type() gets the page */
6625 6626
			put_page(page);
			break;
6627 6628
		case MC_TARGET_SWAP:
			ent = target.ent;
6629
			if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
6630
				mc.precharge--;
6631 6632 6633
				/* we fixup refcnts and charges later. */
				mc.moved_swap++;
			}
6634
			break;
6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648
		default:
			break;
		}
	}
	pte_unmap_unlock(pte - 1, ptl);
	cond_resched();

	if (addr != end) {
		/*
		 * We have consumed all precharges we got in can_attach().
		 * We try charge one by one, but don't do any additional
		 * charges to mc.to if we have failed in charge once in attach()
		 * phase.
		 */
6649
		ret = mem_cgroup_do_precharge(1);
6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661
		if (!ret)
			goto retry;
	}

	return ret;
}

static void mem_cgroup_move_charge(struct mm_struct *mm)
{
	struct vm_area_struct *vma;

	lru_add_drain_all();
6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674
retry:
	if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
		/*
		 * Someone who are holding the mmap_sem might be waiting in
		 * waitq. So we cancel all extra charges, wake up all waiters,
		 * and retry. Because we cancel precharges, we might not be able
		 * to move enough charges, but moving charge is a best-effort
		 * feature anyway, so it wouldn't be a big problem.
		 */
		__mem_cgroup_clear_mc();
		cond_resched();
		goto retry;
	}
6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692
	for (vma = mm->mmap; vma; vma = vma->vm_next) {
		int ret;
		struct mm_walk mem_cgroup_move_charge_walk = {
			.pmd_entry = mem_cgroup_move_charge_pte_range,
			.mm = mm,
			.private = vma,
		};
		if (is_vm_hugetlb_page(vma))
			continue;
		ret = walk_page_range(vma->vm_start, vma->vm_end,
						&mem_cgroup_move_charge_walk);
		if (ret)
			/*
			 * means we have consumed all precharges and failed in
			 * doing additional charge. Just abandon here.
			 */
			break;
	}
6693
	up_read(&mm->mmap_sem);
6694 6695
}

6696
static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
6697
				 struct cgroup_taskset *tset)
B
Balbir Singh 已提交
6698
{
6699
	struct task_struct *p = cgroup_taskset_first(tset);
6700
	struct mm_struct *mm = get_task_mm(p);
6701 6702

	if (mm) {
6703 6704
		if (mc.to)
			mem_cgroup_move_charge(mm);
6705 6706
		mmput(mm);
	}
6707 6708
	if (mc.to)
		mem_cgroup_clear_mc();
B
Balbir Singh 已提交
6709
}
6710
#else	/* !CONFIG_MMU */
6711
static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
6712
				 struct cgroup_taskset *tset)
6713 6714 6715
{
	return 0;
}
6716
static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
6717
				     struct cgroup_taskset *tset)
6718 6719
{
}
6720
static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
6721
				 struct cgroup_taskset *tset)
6722 6723 6724
{
}
#endif
B
Balbir Singh 已提交
6725

6726 6727 6728 6729
/*
 * Cgroup retains root cgroups across [un]mount cycles making it necessary
 * to verify sane_behavior flag on each mount attempt.
 */
6730
static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
6731 6732 6733 6734 6735 6736
{
	/*
	 * use_hierarchy is forced with sane_behavior.  cgroup core
	 * guarantees that @root doesn't have any children, so turning it
	 * on for the root memcg is enough.
	 */
6737 6738
	if (cgroup_sane_behavior(root_css->cgroup))
		mem_cgroup_from_css(root_css)->use_hierarchy = true;
6739 6740
}

B
Balbir Singh 已提交
6741 6742 6743
struct cgroup_subsys mem_cgroup_subsys = {
	.name = "memory",
	.subsys_id = mem_cgroup_subsys_id,
6744
	.css_alloc = mem_cgroup_css_alloc,
6745
	.css_online = mem_cgroup_css_online,
6746 6747
	.css_offline = mem_cgroup_css_offline,
	.css_free = mem_cgroup_css_free,
6748 6749
	.can_attach = mem_cgroup_can_attach,
	.cancel_attach = mem_cgroup_cancel_attach,
B
Balbir Singh 已提交
6750
	.attach = mem_cgroup_move_task,
6751
	.bind = mem_cgroup_bind,
6752
	.base_cftypes = mem_cgroup_files,
6753
	.early_init = 0,
K
KAMEZAWA Hiroyuki 已提交
6754
	.use_id = 1,
B
Balbir Singh 已提交
6755
};
6756

A
Andrew Morton 已提交
6757
#ifdef CONFIG_MEMCG_SWAP
6758 6759
static int __init enable_swap_account(char *s)
{
6760
	if (!strcmp(s, "1"))
6761
		really_do_swap_account = 1;
6762
	else if (!strcmp(s, "0"))
6763 6764 6765
		really_do_swap_account = 0;
	return 1;
}
6766
__setup("swapaccount=", enable_swap_account);
6767

6768 6769
static void __init memsw_file_init(void)
{
6770 6771 6772 6773 6774 6775 6776 6777 6778
	WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
}

static void __init enable_swap_cgroup(void)
{
	if (!mem_cgroup_disabled() && really_do_swap_account) {
		do_swap_account = 1;
		memsw_file_init();
	}
6779
}
6780

6781
#else
6782
static void __init enable_swap_cgroup(void)
6783 6784
{
}
6785
#endif
6786 6787

/*
6788 6789 6790 6791 6792 6793
 * subsys_initcall() for memory controller.
 *
 * Some parts like hotcpu_notifier() have to be initialized from this context
 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
 * everything that doesn't depend on a specific mem_cgroup structure should
 * be initialized from here.
6794 6795 6796 6797
 */
static int __init mem_cgroup_init(void)
{
	hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
6798
	enable_swap_cgroup();
6799
	memcg_stock_init();
6800 6801 6802
	return 0;
}
subsys_initcall(mem_cgroup_init);