blob: 53385cd4e6f02cfae85f99a2629f735094130c78 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080017 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
21 *
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
26 */
27
28#include <linux/res_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080031#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080032#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080033#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050048#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080049#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080051#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070054#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070057#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050058#include <linux/file.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080059#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000060#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070061#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000062#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080063#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080064
Balbir Singh8697d332008-02-07 00:13:59 -080065#include <asm/uaccess.h>
66
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070067#include <trace/events/vmscan.h>
68
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070069struct cgroup_subsys mem_cgroup_subsys __read_mostly;
David Rientjes68ae5642012-12-12 13:51:57 -080070EXPORT_SYMBOL(mem_cgroup_subsys);
71
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070072#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070073static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080074
Andrew Mortonc255a452012-07-31 16:43:02 -070075#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070076/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080077int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080078
79/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070080#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080081static int really_do_swap_account __initdata = 1;
82#else
83static int really_do_swap_account __initdata = 0;
84#endif
85
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080086#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070087#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088#endif
89
90
Johannes Weineraf7c4b02012-05-29 15:07:08 -070091static const char * const mem_cgroup_stat_names[] = {
92 "cache",
93 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070094 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070095 "mapped_file",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070096 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070097 "swap",
98};
99
Johannes Weinere9f89742011-03-23 16:42:37 -0700100enum mem_cgroup_events_index {
101 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
102 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700103 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
104 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700105 MEM_CGROUP_EVENTS_NSTATS,
106};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107
108static const char * const mem_cgroup_events_names[] = {
109 "pgpgin",
110 "pgpgout",
111 "pgfault",
112 "pgmajfault",
113};
114
Sha Zhengju58cf1882013-02-22 16:32:05 -0800115static const char * const mem_cgroup_lru_names[] = {
116 "inactive_anon",
117 "active_anon",
118 "inactive_file",
119 "active_file",
120 "unevictable",
121};
122
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700123/*
124 * Per memcg event counter is incremented at every pagein/pageout. With THP,
125 * it will be incremated by the number of pages. This counter is used for
126 * for trigger some periodic events. This is straightforward and better
127 * than using jiffies etc. to handle periodic memcg event.
128 */
129enum mem_cgroup_events_target {
130 MEM_CGROUP_TARGET_THRESH,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700131 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700132 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700133 MEM_CGROUP_NTARGETS,
134};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700135#define THRESHOLDS_EVENTS_TARGET 128
136#define SOFTLIMIT_EVENTS_TARGET 1024
137#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700138
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800139struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700140 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700141 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700142 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700143 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800144};
145
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800146struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700147 /*
148 * last scanned hierarchy member. Valid only if last_dead_count
149 * matches memcg->dead_count of the hierarchy root group.
150 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700151 struct mem_cgroup *last_visited;
Hugh Dickinsd2ab70a2014-01-23 15:53:30 -0800152 int last_dead_count;
Michal Hocko5f578162013-04-29 15:07:17 -0700153
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800154 /* scan generation, increased every round-trip */
155 unsigned int generation;
156};
157
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800158/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800159 * per-zone information in memory controller.
160 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800161struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800162 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700163 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800164
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800165 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
166
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700167 struct rb_node tree_node; /* RB tree node */
168 unsigned long long usage_in_excess;/* Set to the value by which */
169 /* the soft limit is exceeded*/
170 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700171 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700172 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800173};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800174
175struct mem_cgroup_per_node {
176 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
177};
178
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700179/*
180 * Cgroups above their limits are maintained in a RB-Tree, independent of
181 * their hierarchy representation
182 */
183
184struct mem_cgroup_tree_per_zone {
185 struct rb_root rb_root;
186 spinlock_t lock;
187};
188
189struct mem_cgroup_tree_per_node {
190 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
191};
192
193struct mem_cgroup_tree {
194 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
195};
196
197static struct mem_cgroup_tree soft_limit_tree __read_mostly;
198
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800199struct mem_cgroup_threshold {
200 struct eventfd_ctx *eventfd;
201 u64 threshold;
202};
203
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700204/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800205struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700206 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700207 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800208 /* Size of entries[] */
209 unsigned int size;
210 /* Array of thresholds */
211 struct mem_cgroup_threshold entries[0];
212};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700213
214struct mem_cgroup_thresholds {
215 /* Primary thresholds array */
216 struct mem_cgroup_threshold_ary *primary;
217 /*
218 * Spare threshold array.
219 * This is needed to make mem_cgroup_unregister_event() "never fail".
220 * It must be able to store at least primary->size - 1 entries.
221 */
222 struct mem_cgroup_threshold_ary *spare;
223};
224
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700225/* for OOM */
226struct mem_cgroup_eventfd_list {
227 struct list_head list;
228 struct eventfd_ctx *eventfd;
229};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800230
Tejun Heo79bd9812013-11-22 18:20:42 -0500231/*
232 * cgroup_event represents events which userspace want to receive.
233 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500234struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500235 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500236 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500237 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500238 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500239 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500240 * eventfd to signal userspace about the event.
241 */
242 struct eventfd_ctx *eventfd;
243 /*
244 * Each of these stored in a list by the cgroup.
245 */
246 struct list_head list;
247 /*
Tejun Heofba94802013-11-22 18:20:43 -0500248 * register_event() callback will be used to add new userspace
249 * waiter for changes related to this event. Use eventfd_signal()
250 * on eventfd to send notification to userspace.
251 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500252 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500253 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500254 /*
255 * unregister_event() callback will be called when userspace closes
256 * the eventfd or on cgroup removing. This callback must be set,
257 * if you want provide notification functionality.
258 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500259 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500260 struct eventfd_ctx *eventfd);
261 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500262 * All fields below needed to unregister event when
263 * userspace closes eventfd.
264 */
265 poll_table pt;
266 wait_queue_head_t *wqh;
267 wait_queue_t wait;
268 struct work_struct remove;
269};
270
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700271static void mem_cgroup_threshold(struct mem_cgroup *memcg);
272static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800273
Balbir Singhf64c3f52009-09-23 15:56:37 -0700274/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800275 * The memory controller data structure. The memory controller controls both
276 * page cache and RSS per cgroup. We would eventually like to provide
277 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
278 * to help the administrator determine what knobs to tune.
279 *
280 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800281 * we hit the water mark. May be even add a low water mark, such that
282 * no reclaim occurs from a cgroup at it's low water mark, this is
283 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800284 */
285struct mem_cgroup {
286 struct cgroup_subsys_state css;
287 /*
288 * the counter to account for memory usage
289 */
290 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700291
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700292 /* vmpressure notifications */
293 struct vmpressure vmpressure;
294
Li Zefan465939a2013-07-08 16:00:38 -0700295 /*
296 * the counter to account for mem+swap usage.
297 */
298 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700299
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800300 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800301 * the counter to account for kernel memory usage.
302 */
303 struct res_counter kmem;
304 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800305 * Should the accounting and control be hierarchical, per subtree?
306 */
307 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800308 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700309
310 bool oom_lock;
311 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700312 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700313
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700314 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700315 /* OOM-Killer disable */
316 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800317
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700318 /* set when res.limit == memsw.limit */
319 bool memsw_is_minimum;
320
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800321 /* protect arrays of thresholds */
322 struct mutex thresholds_lock;
323
324 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700325 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700326
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800327 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700328 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700329
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700330 /* For oom notifier event fd */
331 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700332
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800333 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800334 * Should we move charges of a task when a task is moved into this
335 * mem_cgroup ? And what type of charges should we move ?
336 */
Andrew Mortonf894ffa2013-09-12 15:13:35 -0700337 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800338 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700339 * set > 0 if pages under this cgroup are moving to other cgroup.
340 */
341 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700342 /* taken only while moving_account > 0 */
343 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700344 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800345 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800346 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700347 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700348 /*
349 * used when a cpu is offlined or other synchronizations
350 * See mem_cgroup_read_stat().
351 */
352 struct mem_cgroup_stat_cpu nocpu_base;
353 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000354
Michal Hocko5f578162013-04-29 15:07:17 -0700355 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700356#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700357 struct cg_proto tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000358#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800359#if defined(CONFIG_MEMCG_KMEM)
360 /* analogous to slab_common's slab_caches list. per-memcg */
361 struct list_head memcg_slab_caches;
362 /* Not a spinlock, we can take a lot of time walking the list */
363 struct mutex slab_caches_mutex;
364 /* Index in the kmem_cache->memcg_params->memcg_caches array */
365 int kmemcg_id;
366#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800367
368 int last_scanned_node;
369#if MAX_NUMNODES > 1
370 nodemask_t scan_nodes;
371 atomic_t numainfo_events;
372 atomic_t numainfo_updating;
373#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700374
Tejun Heofba94802013-11-22 18:20:43 -0500375 /* List of events which userspace want to receive */
376 struct list_head event_list;
377 spinlock_t event_list_lock;
378
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700379 struct mem_cgroup_per_node *nodeinfo[0];
380 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800381};
382
Glauber Costa510fc4e2012-12-18 14:21:47 -0800383/* internal only representation about the status of kmem accounting. */
384enum {
Vladimir Davydov6de64be2014-01-23 15:53:08 -0800385 KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */
Glauber Costa7de37682012-12-18 14:22:07 -0800386 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800387};
388
Glauber Costa510fc4e2012-12-18 14:21:47 -0800389#ifdef CONFIG_MEMCG_KMEM
390static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
391{
392 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
393}
Glauber Costa7de37682012-12-18 14:22:07 -0800394
395static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
396{
397 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
398}
399
400static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
401{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700402 /*
403 * Our caller must use css_get() first, because memcg_uncharge_kmem()
404 * will call css_put() if it sees the memcg is dead.
405 */
406 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800407 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
408 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
409}
410
411static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
412{
413 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
414 &memcg->kmem_account_flags);
415}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800416#endif
417
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800418/* Stuffs for move charges at task migration. */
419/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800420 * Types of charges to be moved. "move_charge_at_immitgrate" and
421 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800422 */
423enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800424 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700425 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800426 NR_MOVE_TYPE,
427};
428
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800429/* "mc" and its members are protected by cgroup_mutex */
430static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800431 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800432 struct mem_cgroup *from;
433 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800434 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800435 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800436 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800437 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800438 struct task_struct *moving_task; /* a task moving charges */
439 wait_queue_head_t waitq; /* a waitq for other context */
440} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700441 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800442 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
443};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800444
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700445static bool move_anon(void)
446{
Glauber Costaee5e8472013-02-22 16:34:50 -0800447 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700448}
449
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700450static bool move_file(void)
451{
Glauber Costaee5e8472013-02-22 16:34:50 -0800452 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700453}
454
Balbir Singh4e416952009-09-23 15:56:39 -0700455/*
456 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
457 * limit reclaim to prevent infinite loops, if they ever occur.
458 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700459#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700460#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700461
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800462enum charge_type {
463 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700464 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800465 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700466 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700467 NR_CHARGE_TYPE,
468};
469
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800470/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800471enum res_type {
472 _MEM,
473 _MEMSWAP,
474 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800475 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800476};
477
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700478#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
479#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800480#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700481/* Used for OOM nofiier */
482#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800483
Balbir Singh75822b42009-09-23 15:56:38 -0700484/*
485 * Reclaim flags for mem_cgroup_hierarchical_reclaim
486 */
487#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
488#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
489#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
490#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
491
Glauber Costa09998212013-02-22 16:34:55 -0800492/*
493 * The memcg_create_mutex will be held whenever a new cgroup is created.
494 * As a consequence, any change that needs to protect against new child cgroups
495 * appearing has to hold it as well.
496 */
497static DEFINE_MUTEX(memcg_create_mutex);
498
Wanpeng Lib2145142012-07-31 16:46:01 -0700499struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
500{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400501 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700502}
503
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700504/* Some nice accessors for the vmpressure. */
505struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
506{
507 if (!memcg)
508 memcg = root_mem_cgroup;
509 return &memcg->vmpressure;
510}
511
512struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
513{
514 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
515}
516
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700517static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
518{
519 return (memcg == root_mem_cgroup);
520}
521
Li Zefan4219b2d2013-09-23 16:56:29 +0800522/*
523 * We restrict the id in the range of [1, 65535], so it can fit into
524 * an unsigned short.
525 */
526#define MEM_CGROUP_ID_MAX USHRT_MAX
527
Li Zefan34c00c32013-09-23 16:56:01 +0800528static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
529{
530 /*
531 * The ID of the root cgroup is 0, but memcg treat 0 as an
532 * invalid ID, so we return (cgroup_id + 1).
533 */
534 return memcg->css.cgroup->id + 1;
535}
536
537static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
538{
539 struct cgroup_subsys_state *css;
540
541 css = css_from_id(id - 1, &mem_cgroup_subsys);
542 return mem_cgroup_from_css(css);
543}
544
Glauber Costae1aab162011-12-11 21:47:03 +0000545/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700546#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000547
Glauber Costae1aab162011-12-11 21:47:03 +0000548void sock_update_memcg(struct sock *sk)
549{
Glauber Costa376be5f2012-01-20 04:57:14 +0000550 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000551 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700552 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000553
554 BUG_ON(!sk->sk_prot->proto_cgroup);
555
Glauber Costaf3f511e2012-01-05 20:16:39 +0000556 /* Socket cloning can throw us here with sk_cgrp already
557 * filled. It won't however, necessarily happen from
558 * process context. So the test for root memcg given
559 * the current task's memcg won't help us in this case.
560 *
561 * Respecting the original socket's memcg is a better
562 * decision in this case.
563 */
564 if (sk->sk_cgrp) {
565 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700566 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000567 return;
568 }
569
Glauber Costae1aab162011-12-11 21:47:03 +0000570 rcu_read_lock();
571 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700572 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700573 if (!mem_cgroup_is_root(memcg) &&
574 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700575 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000576 }
577 rcu_read_unlock();
578 }
579}
580EXPORT_SYMBOL(sock_update_memcg);
581
582void sock_release_memcg(struct sock *sk)
583{
Glauber Costa376be5f2012-01-20 04:57:14 +0000584 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000585 struct mem_cgroup *memcg;
586 WARN_ON(!sk->sk_cgrp->memcg);
587 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700588 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000589 }
590}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000591
592struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
593{
594 if (!memcg || mem_cgroup_is_root(memcg))
595 return NULL;
596
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700597 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000598}
599EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000600
Glauber Costa3f134612012-05-29 15:07:11 -0700601static void disarm_sock_keys(struct mem_cgroup *memcg)
602{
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700603 if (!memcg_proto_activated(&memcg->tcp_mem))
Glauber Costa3f134612012-05-29 15:07:11 -0700604 return;
605 static_key_slow_dec(&memcg_socket_limit_enabled);
606}
607#else
608static void disarm_sock_keys(struct mem_cgroup *memcg)
609{
610}
611#endif
612
Glauber Costaa8964b92012-12-18 14:22:09 -0800613#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800614/*
615 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800616 * The main reason for not using cgroup id for this:
617 * this works better in sparse environments, where we have a lot of memcgs,
618 * but only a few kmem-limited. Or also, if we have, for instance, 200
619 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
620 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800621 *
622 * The current size of the caches array is stored in
623 * memcg_limited_groups_array_size. It will double each time we have to
624 * increase it.
625 */
626static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800627int memcg_limited_groups_array_size;
628
Glauber Costa55007d82012-12-18 14:22:38 -0800629/*
630 * MIN_SIZE is different than 1, because we would like to avoid going through
631 * the alloc/free process all the time. In a small machine, 4 kmem-limited
632 * cgroups is a reasonable guess. In the future, it could be a parameter or
633 * tunable, but that is strictly not necessary.
634 *
Li Zefanb8627832013-09-23 16:56:47 +0800635 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800636 * this constant directly from cgroup, but it is understandable that this is
637 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800638 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800639 * increase ours as well if it increases.
640 */
641#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800642#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800643
Glauber Costad7f25f82012-12-18 14:22:40 -0800644/*
645 * A lot of the calls to the cache allocation functions are expected to be
646 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
647 * conditional to this static branch, we'll have to allow modules that does
648 * kmem_cache_alloc and the such to see this symbol as well
649 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800650struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800651EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800652
653static void disarm_kmem_keys(struct mem_cgroup *memcg)
654{
Glauber Costa55007d82012-12-18 14:22:38 -0800655 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800656 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800657 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
658 }
Glauber Costabea207c2012-12-18 14:22:11 -0800659 /*
660 * This check can't live in kmem destruction function,
661 * since the charges will outlive the cgroup
662 */
663 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800664}
665#else
666static void disarm_kmem_keys(struct mem_cgroup *memcg)
667{
668}
669#endif /* CONFIG_MEMCG_KMEM */
670
671static void disarm_static_keys(struct mem_cgroup *memcg)
672{
673 disarm_sock_keys(memcg);
674 disarm_kmem_keys(memcg);
675}
676
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700677static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800678
Balbir Singhf64c3f52009-09-23 15:56:37 -0700679static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700681{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800682 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700683 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700684}
685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100687{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100689}
690
Balbir Singhf64c3f52009-09-23 15:56:37 -0700691static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700692page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700693{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700694 int nid = page_to_nid(page);
695 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700696
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700697 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700698}
699
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700static struct mem_cgroup_tree_per_zone *
701soft_limit_tree_node_zone(int nid, int zid)
702{
703 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
704}
705
706static struct mem_cgroup_tree_per_zone *
707soft_limit_tree_from_page(struct page *page)
708{
709 int nid = page_to_nid(page);
710 int zid = page_zonenum(page);
711
712 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
713}
714
715static void
716__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
717 struct mem_cgroup_per_zone *mz,
718 struct mem_cgroup_tree_per_zone *mctz,
719 unsigned long long new_usage_in_excess)
720{
721 struct rb_node **p = &mctz->rb_root.rb_node;
722 struct rb_node *parent = NULL;
723 struct mem_cgroup_per_zone *mz_node;
724
725 if (mz->on_tree)
726 return;
727
728 mz->usage_in_excess = new_usage_in_excess;
729 if (!mz->usage_in_excess)
730 return;
731 while (*p) {
732 parent = *p;
733 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
734 tree_node);
735 if (mz->usage_in_excess < mz_node->usage_in_excess)
736 p = &(*p)->rb_left;
737 /*
738 * We can't avoid mem cgroups that are over their soft
739 * limit by the same amount
740 */
741 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
742 p = &(*p)->rb_right;
743 }
744 rb_link_node(&mz->tree_node, parent, p);
745 rb_insert_color(&mz->tree_node, &mctz->rb_root);
746 mz->on_tree = true;
747}
748
749static void
750__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
751 struct mem_cgroup_per_zone *mz,
752 struct mem_cgroup_tree_per_zone *mctz)
753{
754 if (!mz->on_tree)
755 return;
756 rb_erase(&mz->tree_node, &mctz->rb_root);
757 mz->on_tree = false;
758}
759
760static void
761mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
762 struct mem_cgroup_per_zone *mz,
763 struct mem_cgroup_tree_per_zone *mctz)
764{
765 spin_lock(&mctz->lock);
766 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
767 spin_unlock(&mctz->lock);
768}
769
770
771static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
772{
773 unsigned long long excess;
774 struct mem_cgroup_per_zone *mz;
775 struct mem_cgroup_tree_per_zone *mctz;
776 int nid = page_to_nid(page);
777 int zid = page_zonenum(page);
778 mctz = soft_limit_tree_from_page(page);
779
780 /*
781 * Necessary to update all ancestors when hierarchy is used.
782 * because their event counter is not touched.
783 */
784 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
785 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
786 excess = res_counter_soft_limit_excess(&memcg->res);
787 /*
788 * We have to update the tree if mz is on RB-tree or
789 * mem is over its softlimit.
790 */
791 if (excess || mz->on_tree) {
792 spin_lock(&mctz->lock);
793 /* if on-tree, remove it */
794 if (mz->on_tree)
795 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
796 /*
797 * Insert again. mz->usage_in_excess will be updated.
798 * If excess is 0, no tree ops.
799 */
800 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
801 spin_unlock(&mctz->lock);
802 }
803 }
804}
805
806static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
807{
808 int node, zone;
809 struct mem_cgroup_per_zone *mz;
810 struct mem_cgroup_tree_per_zone *mctz;
811
812 for_each_node(node) {
813 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
814 mz = mem_cgroup_zoneinfo(memcg, node, zone);
815 mctz = soft_limit_tree_node_zone(node, zone);
816 mem_cgroup_remove_exceeded(memcg, mz, mctz);
817 }
818 }
819}
820
821static struct mem_cgroup_per_zone *
822__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
823{
824 struct rb_node *rightmost = NULL;
825 struct mem_cgroup_per_zone *mz;
826
827retry:
828 mz = NULL;
829 rightmost = rb_last(&mctz->rb_root);
830 if (!rightmost)
831 goto done; /* Nothing to reclaim from */
832
833 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
834 /*
835 * Remove the node now but someone else can add it back,
836 * we will to add it back at the end of reclaim to its correct
837 * position in the tree.
838 */
839 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
840 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
841 !css_tryget(&mz->memcg->css))
842 goto retry;
843done:
844 return mz;
845}
846
847static struct mem_cgroup_per_zone *
848mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
849{
850 struct mem_cgroup_per_zone *mz;
851
852 spin_lock(&mctz->lock);
853 mz = __mem_cgroup_largest_soft_limit_node(mctz);
854 spin_unlock(&mctz->lock);
855 return mz;
856}
857
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700858/*
859 * Implementation Note: reading percpu statistics for memcg.
860 *
861 * Both of vmstat[] and percpu_counter has threshold and do periodic
862 * synchronization to implement "quick" read. There are trade-off between
863 * reading cost and precision of value. Then, we may have a chance to implement
864 * a periodic synchronizion of counter in memcg's counter.
865 *
866 * But this _read() function is used for user interface now. The user accounts
867 * memory usage by memory cgroup and he _always_ requires exact value because
868 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
869 * have to visit all online cpus and make sum. So, for now, unnecessary
870 * synchronization is not implemented. (just implemented for cpu hotplug)
871 *
872 * If there are kernel internal actions which can make use of some not-exact
873 * value, and reading all cpu value can be performance bottleneck in some
874 * common workload, threashold and synchonization as vmstat[] should be
875 * implemented.
876 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700877static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700878 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800879{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700880 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800881 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800882
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700883 get_online_cpus();
884 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700885 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700886#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700887 spin_lock(&memcg->pcp_counter_lock);
888 val += memcg->nocpu_base.count[idx];
889 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700890#endif
891 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800892 return val;
893}
894
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700895static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700896 bool charge)
897{
898 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700899 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700900}
901
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700902static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700903 enum mem_cgroup_events_index idx)
904{
905 unsigned long val = 0;
906 int cpu;
907
David Rientjes9c567512013-10-16 13:46:43 -0700908 get_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700909 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700910 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700911#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700912 spin_lock(&memcg->pcp_counter_lock);
913 val += memcg->nocpu_base.events[idx];
914 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700915#endif
David Rientjes9c567512013-10-16 13:46:43 -0700916 put_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700917 return val;
918}
919
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700920static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700921 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700922 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800923{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800924 preempt_disable();
925
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700926 /*
927 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
928 * counted as CACHE even if it's on ANON LRU.
929 */
930 if (anon)
931 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700932 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800933 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700934 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700935 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700936
David Rientjesb070e652013-05-07 16:18:09 -0700937 if (PageTransHuge(page))
938 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
939 nr_pages);
940
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800941 /* pagein of a big page is an event. So, ignore page size */
942 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700943 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800944 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700945 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800946 nr_pages = -nr_pages; /* for event */
947 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800948
Johannes Weiner13114712012-05-29 15:07:07 -0700949 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800950
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800951 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800952}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800953
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700954unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700955mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700956{
957 struct mem_cgroup_per_zone *mz;
958
959 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
960 return mz->lru_size[lru];
961}
962
963static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700964mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700965 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700966{
967 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700968 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700969 unsigned long ret = 0;
970
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700971 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700972
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700973 for_each_lru(lru) {
974 if (BIT(lru) & lru_mask)
975 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700976 }
977 return ret;
978}
979
980static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700981mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700982 int nid, unsigned int lru_mask)
983{
Ying Han889976d2011-05-26 16:25:33 -0700984 u64 total = 0;
985 int zid;
986
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700987 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700988 total += mem_cgroup_zone_nr_lru_pages(memcg,
989 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700990
Ying Han889976d2011-05-26 16:25:33 -0700991 return total;
992}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700993
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700994static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700995 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800996{
Ying Han889976d2011-05-26 16:25:33 -0700997 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800998 u64 total = 0;
999
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001000 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001001 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001002 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -08001003}
1004
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001005static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
1006 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001007{
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001008 unsigned long val, next;
1009
Johannes Weiner13114712012-05-29 15:07:07 -07001010 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -07001011 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001012 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001013 if ((long)next - (long)val < 0) {
1014 switch (target) {
1015 case MEM_CGROUP_TARGET_THRESH:
1016 next = val + THRESHOLDS_EVENTS_TARGET;
1017 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001018 case MEM_CGROUP_TARGET_SOFTLIMIT:
1019 next = val + SOFTLIMIT_EVENTS_TARGET;
1020 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001021 case MEM_CGROUP_TARGET_NUMAINFO:
1022 next = val + NUMAINFO_EVENTS_TARGET;
1023 break;
1024 default:
1025 break;
1026 }
1027 __this_cpu_write(memcg->stat->targets[target], next);
1028 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001029 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001030 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001031}
1032
1033/*
1034 * Check events in order.
1035 *
1036 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001037static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001038{
Steven Rostedt47994012011-11-02 13:38:33 -07001039 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001040 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001041 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1042 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001043 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001044 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001045
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001046 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1047 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001048#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001049 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1050 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001051#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001052 preempt_enable();
1053
1054 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001055 if (unlikely(do_softlimit))
1056 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001057#if MAX_NUMNODES > 1
1058 if (unlikely(do_numainfo))
1059 atomic_inc(&memcg->numainfo_events);
1060#endif
1061 } else
1062 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001063}
1064
Balbir Singhcf475ad2008-04-29 01:00:16 -07001065struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001066{
Balbir Singh31a78f22008-09-28 23:09:31 +01001067 /*
1068 * mm_update_next_owner() may clear mm->owner to NULL
1069 * if it races with swapoff, page migration, etc.
1070 * So this can be called with p == NULL.
1071 */
1072 if (unlikely(!p))
1073 return NULL;
1074
Tejun Heo8af01f52013-08-08 20:11:22 -04001075 return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001076}
1077
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001078struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001079{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001080 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001081
1082 if (!mm)
1083 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001084 /*
1085 * Because we have no locks, mm->owner's may be being moved to other
1086 * cgroup. We use css_tryget() here even if this looks
1087 * pessimistic (rather than adding locks here).
1088 */
1089 rcu_read_lock();
1090 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001091 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1092 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001093 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001094 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001095 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001096 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001097}
1098
Michal Hocko16248d82013-04-29 15:07:19 -07001099/*
1100 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1101 * ref. count) or NULL if the whole root's subtree has been visited.
1102 *
1103 * helper function to be used by mem_cgroup_iter
1104 */
1105static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001106 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001107{
Tejun Heo492eb212013-08-08 20:11:25 -04001108 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001109
Tejun Heobd8815a2013-08-08 20:11:27 -04001110 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001111skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001112 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001113
1114 /*
1115 * Even if we found a group we have to make sure it is
1116 * alive. css && !memcg means that the groups should be
1117 * skipped and we should continue the tree walk.
1118 * last_visited css is safe to use because it is
1119 * protected by css_get and the tree walk is rcu safe.
Michal Hocko0eef6152014-01-23 15:53:37 -08001120 *
1121 * We do not take a reference on the root of the tree walk
1122 * because we might race with the root removal when it would
1123 * be the only node in the iterated hierarchy and mem_cgroup_iter
1124 * would end up in an endless loop because it expects that at
1125 * least one valid node will be returned. Root cannot disappear
1126 * because caller of the iterator should hold it already so
1127 * skipping css reference should be safe.
Michal Hocko16248d82013-04-29 15:07:19 -07001128 */
Tejun Heo492eb212013-08-08 20:11:25 -04001129 if (next_css) {
Michal Hocko0eef6152014-01-23 15:53:37 -08001130 if ((next_css->flags & CSS_ONLINE) &&
1131 (next_css == &root->css || css_tryget(next_css)))
Hugh Dickinsd8ad3052014-01-23 15:53:32 -08001132 return mem_cgroup_from_css(next_css);
Michal Hocko0eef6152014-01-23 15:53:37 -08001133
1134 prev_css = next_css;
1135 goto skip_node;
Michal Hocko16248d82013-04-29 15:07:19 -07001136 }
1137
1138 return NULL;
1139}
1140
Johannes Weiner519ebea2013-07-03 15:04:51 -07001141static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1142{
1143 /*
1144 * When a group in the hierarchy below root is destroyed, the
1145 * hierarchy iterator can no longer be trusted since it might
1146 * have pointed to the destroyed group. Invalidate it.
1147 */
1148 atomic_inc(&root->dead_count);
1149}
1150
1151static struct mem_cgroup *
1152mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1153 struct mem_cgroup *root,
1154 int *sequence)
1155{
1156 struct mem_cgroup *position = NULL;
1157 /*
1158 * A cgroup destruction happens in two stages: offlining and
1159 * release. They are separated by a RCU grace period.
1160 *
1161 * If the iterator is valid, we may still race with an
1162 * offlining. The RCU lock ensures the object won't be
1163 * released, tryget will fail if we lost the race.
1164 */
1165 *sequence = atomic_read(&root->dead_count);
1166 if (iter->last_dead_count == *sequence) {
1167 smp_rmb();
1168 position = iter->last_visited;
Michal Hockoecc736f2014-01-23 15:53:35 -08001169
1170 /*
1171 * We cannot take a reference to root because we might race
1172 * with root removal and returning NULL would end up in
1173 * an endless loop on the iterator user level when root
1174 * would be returned all the time.
1175 */
1176 if (position && position != root &&
1177 !css_tryget(&position->css))
Johannes Weiner519ebea2013-07-03 15:04:51 -07001178 position = NULL;
1179 }
1180 return position;
1181}
1182
1183static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1184 struct mem_cgroup *last_visited,
1185 struct mem_cgroup *new_position,
Michal Hockoecc736f2014-01-23 15:53:35 -08001186 struct mem_cgroup *root,
Johannes Weiner519ebea2013-07-03 15:04:51 -07001187 int sequence)
1188{
Michal Hockoecc736f2014-01-23 15:53:35 -08001189 /* root reference counting symmetric to mem_cgroup_iter_load */
1190 if (last_visited && last_visited != root)
Johannes Weiner519ebea2013-07-03 15:04:51 -07001191 css_put(&last_visited->css);
1192 /*
1193 * We store the sequence count from the time @last_visited was
1194 * loaded successfully instead of rereading it here so that we
1195 * don't lose destruction events in between. We could have
1196 * raced with the destruction of @new_position after all.
1197 */
1198 iter->last_visited = new_position;
1199 smp_wmb();
1200 iter->last_dead_count = sequence;
1201}
1202
Johannes Weiner56600482012-01-12 17:17:59 -08001203/**
1204 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1205 * @root: hierarchy root
1206 * @prev: previously returned memcg, NULL on first invocation
1207 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1208 *
1209 * Returns references to children of the hierarchy below @root, or
1210 * @root itself, or %NULL after a full round-trip.
1211 *
1212 * Caller must pass the return value in @prev on subsequent
1213 * invocations for reference counting, or use mem_cgroup_iter_break()
1214 * to cancel a hierarchy walk before the round-trip is complete.
1215 *
1216 * Reclaimers can specify a zone and a priority level in @reclaim to
1217 * divide up the memcgs in the hierarchy among all concurrent
1218 * reclaimers operating on the same zone and priority.
1219 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001220struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001221 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001222 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001223{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001224 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001225 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001226
Andrew Morton694fbc02013-09-24 15:27:37 -07001227 if (mem_cgroup_disabled())
1228 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001229
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001230 if (!root)
1231 root = root_mem_cgroup;
1232
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001233 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001234 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001235
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001236 if (!root->use_hierarchy && root != root_mem_cgroup) {
1237 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001238 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001239 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001240 }
1241
Michal Hocko542f85f2013-04-29 15:07:15 -07001242 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001243 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001244 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001245 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001246
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001247 if (reclaim) {
1248 int nid = zone_to_nid(reclaim->zone);
1249 int zid = zone_idx(reclaim->zone);
1250 struct mem_cgroup_per_zone *mz;
1251
1252 mz = mem_cgroup_zoneinfo(root, nid, zid);
1253 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001254 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001255 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001256 goto out_unlock;
1257 }
Michal Hocko5f578162013-04-29 15:07:17 -07001258
Johannes Weiner519ebea2013-07-03 15:04:51 -07001259 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001260 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001261
Andrew Morton694fbc02013-09-24 15:27:37 -07001262 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001263
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001264 if (reclaim) {
Michal Hockoecc736f2014-01-23 15:53:35 -08001265 mem_cgroup_iter_update(iter, last_visited, memcg, root,
1266 seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001267
Michal Hocko19f39402013-04-29 15:07:18 -07001268 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001269 iter->generation++;
1270 else if (!prev && memcg)
1271 reclaim->generation = iter->generation;
1272 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001273
Andrew Morton694fbc02013-09-24 15:27:37 -07001274 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001275 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001276 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001277out_unlock:
1278 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001279out_css_put:
1280 if (prev && prev != root)
1281 css_put(&prev->css);
1282
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001283 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001284}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001285
Johannes Weiner56600482012-01-12 17:17:59 -08001286/**
1287 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1288 * @root: hierarchy root
1289 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1290 */
1291void mem_cgroup_iter_break(struct mem_cgroup *root,
1292 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001293{
1294 if (!root)
1295 root = root_mem_cgroup;
1296 if (prev && prev != root)
1297 css_put(&prev->css);
1298}
1299
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001300/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001301 * Iteration constructs for visiting all cgroups (under a tree). If
1302 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1303 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001304 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001305#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001306 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001307 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001308 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001309
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001310#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001311 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001312 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001313 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001314
David Rientjes68ae5642012-12-12 13:51:57 -08001315void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001316{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001317 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001318
Ying Han456f9982011-05-26 16:25:38 -07001319 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001320 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1321 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001322 goto out;
1323
1324 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001325 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001326 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1327 break;
1328 case PGMAJFAULT:
1329 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001330 break;
1331 default:
1332 BUG();
1333 }
1334out:
1335 rcu_read_unlock();
1336}
David Rientjes68ae5642012-12-12 13:51:57 -08001337EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001338
Johannes Weiner925b7672012-01-12 17:18:15 -08001339/**
1340 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1341 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001342 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001343 *
1344 * Returns the lru list vector holding pages for the given @zone and
1345 * @mem. This can be the global zone lruvec, if the memory controller
1346 * is disabled.
1347 */
1348struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1349 struct mem_cgroup *memcg)
1350{
1351 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001352 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001353
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001354 if (mem_cgroup_disabled()) {
1355 lruvec = &zone->lruvec;
1356 goto out;
1357 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001358
1359 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001360 lruvec = &mz->lruvec;
1361out:
1362 /*
1363 * Since a node can be onlined after the mem_cgroup was created,
1364 * we have to be prepared to initialize lruvec->zone here;
1365 * and if offlined then reonlined, we need to reinitialize it.
1366 */
1367 if (unlikely(lruvec->zone != zone))
1368 lruvec->zone = zone;
1369 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001370}
1371
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001372/*
1373 * Following LRU functions are allowed to be used without PCG_LOCK.
1374 * Operations are called by routine of global LRU independently from memcg.
1375 * What we have to take care of here is validness of pc->mem_cgroup.
1376 *
1377 * Changes to pc->mem_cgroup happens when
1378 * 1. charge
1379 * 2. moving account
1380 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1381 * It is added to LRU before charge.
1382 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1383 * When moving account, the page is not on LRU. It's isolated.
1384 */
1385
Johannes Weiner925b7672012-01-12 17:18:15 -08001386/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001387 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001388 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001389 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001390 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001391struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001392{
1393 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001394 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001395 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001396 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001397
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001398 if (mem_cgroup_disabled()) {
1399 lruvec = &zone->lruvec;
1400 goto out;
1401 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001402
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001403 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001404 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001405
1406 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001407 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001408 * an uncharged page off lru does nothing to secure
1409 * its former mem_cgroup from sudden removal.
1410 *
1411 * Our caller holds lru_lock, and PageCgroupUsed is updated
1412 * under page_cgroup lock: between them, they make all uses
1413 * of pc->mem_cgroup safe.
1414 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001415 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001416 pc->mem_cgroup = memcg = root_mem_cgroup;
1417
Johannes Weiner925b7672012-01-12 17:18:15 -08001418 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001419 lruvec = &mz->lruvec;
1420out:
1421 /*
1422 * Since a node can be onlined after the mem_cgroup was created,
1423 * we have to be prepared to initialize lruvec->zone here;
1424 * and if offlined then reonlined, we need to reinitialize it.
1425 */
1426 if (unlikely(lruvec->zone != zone))
1427 lruvec->zone = zone;
1428 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001429}
1430
1431/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001432 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1433 * @lruvec: mem_cgroup per zone lru vector
1434 * @lru: index of lru list the page is sitting on
1435 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001436 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001437 * This function must be called when a page is added to or removed from an
1438 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001439 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001440void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1441 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001442{
1443 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001444 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001445
1446 if (mem_cgroup_disabled())
1447 return;
1448
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001449 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1450 lru_size = mz->lru_size + lru;
1451 *lru_size += nr_pages;
1452 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001453}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001454
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001455/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001456 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001457 * hierarchy subtree
1458 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001459bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1460 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001461{
Johannes Weiner91c637342012-05-29 15:06:24 -07001462 if (root_memcg == memcg)
1463 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001464 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001465 return false;
Li Zefanb47f77b2013-09-23 16:55:43 +08001466 return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001467}
1468
1469static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1470 struct mem_cgroup *memcg)
1471{
1472 bool ret;
1473
Johannes Weiner91c637342012-05-29 15:06:24 -07001474 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001475 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001476 rcu_read_unlock();
1477 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001478}
1479
David Rientjesffbdccf2013-07-03 15:01:23 -07001480bool task_in_mem_cgroup(struct task_struct *task,
1481 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001482{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001483 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001484 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001485 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001486
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001487 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001488 if (p) {
1489 curr = try_get_mem_cgroup_from_mm(p->mm);
1490 task_unlock(p);
1491 } else {
1492 /*
1493 * All threads may have already detached their mm's, but the oom
1494 * killer still needs to detect if they have already been oom
1495 * killed to prevent needlessly killing additional tasks.
1496 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001497 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001498 curr = mem_cgroup_from_task(task);
1499 if (curr)
1500 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001501 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001502 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001503 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001504 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001505 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001506 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001507 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001508 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1509 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001510 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001511 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001512 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001513 return ret;
1514}
1515
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001516int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001517{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001518 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001519 unsigned long inactive;
1520 unsigned long active;
1521 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001522
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001523 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1524 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001525
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001526 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1527 if (gb)
1528 inactive_ratio = int_sqrt(10 * gb);
1529 else
1530 inactive_ratio = 1;
1531
Johannes Weiner9b272972011-11-02 13:38:23 -07001532 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001533}
1534
Balbir Singh6d61ef42009-01-07 18:08:06 -08001535#define mem_cgroup_from_res_counter(counter, member) \
1536 container_of(counter, struct mem_cgroup, member)
1537
Johannes Weiner19942822011-02-01 15:52:43 -08001538/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001539 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001540 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001541 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001542 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001543 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001544 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001545static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001546{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001547 unsigned long long margin;
1548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001549 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001550 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001551 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001552 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001553}
1554
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001555int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001556{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001557 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001558 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001559 return vm_swappiness;
1560
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001561 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001562}
1563
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001564/*
1565 * memcg->moving_account is used for checking possibility that some thread is
1566 * calling move_account(). When a thread on CPU-A starts moving pages under
1567 * a memcg, other threads should check memcg->moving_account under
1568 * rcu_read_lock(), like this:
1569 *
1570 * CPU-A CPU-B
1571 * rcu_read_lock()
1572 * memcg->moving_account+1 if (memcg->mocing_account)
1573 * take heavy locks.
1574 * synchronize_rcu() update something.
1575 * rcu_read_unlock()
1576 * start move here.
1577 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001578
1579/* for quick checking without looking up memcg */
1580atomic_t memcg_moving __read_mostly;
1581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001582static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001583{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001584 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001585 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001586 synchronize_rcu();
1587}
1588
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001589static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001590{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001591 /*
1592 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1593 * We check NULL in callee rather than caller.
1594 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001595 if (memcg) {
1596 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001597 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001598 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001599}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001600
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001601/*
1602 * 2 routines for checking "mem" is under move_account() or not.
1603 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001604 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1605 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001606 * pc->mem_cgroup may be overwritten.
1607 *
1608 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1609 * under hierarchy of moving cgroups. This is for
1610 * waiting at hith-memory prressure caused by "move".
1611 */
1612
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001613static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001614{
1615 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001616 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001617}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001618
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001619static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001620{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001621 struct mem_cgroup *from;
1622 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001623 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001624 /*
1625 * Unlike task_move routines, we access mc.to, mc.from not under
1626 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1627 */
1628 spin_lock(&mc.lock);
1629 from = mc.from;
1630 to = mc.to;
1631 if (!from)
1632 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001633
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001634 ret = mem_cgroup_same_or_subtree(memcg, from)
1635 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001636unlock:
1637 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001638 return ret;
1639}
1640
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001641static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001642{
1643 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001644 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001645 DEFINE_WAIT(wait);
1646 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1647 /* moving charge context might have finished. */
1648 if (mc.moving_task)
1649 schedule();
1650 finish_wait(&mc.waitq, &wait);
1651 return true;
1652 }
1653 }
1654 return false;
1655}
1656
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001657/*
1658 * Take this lock when
1659 * - a code tries to modify page's memcg while it's USED.
1660 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001661 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001662 */
1663static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1664 unsigned long *flags)
1665{
1666 spin_lock_irqsave(&memcg->move_lock, *flags);
1667}
1668
1669static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1670 unsigned long *flags)
1671{
1672 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1673}
1674
Sha Zhengju58cf1882013-02-22 16:32:05 -08001675#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001676/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001677 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001678 * @memcg: The memory cgroup that went over limit
1679 * @p: Task that is going to be killed
1680 *
1681 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1682 * enabled
1683 */
1684void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1685{
Michal Hocko947b3dd2014-01-21 15:51:04 -08001686 /*
1687 * protects memcg_name and makes sure that parallel ooms do not
1688 * interleave
1689 */
1690 static DEFINE_SPINLOCK(oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001691 struct cgroup *task_cgrp;
1692 struct cgroup *mem_cgrp;
Balbir Singhe2224322009-04-02 16:57:39 -07001693 static char memcg_name[PATH_MAX];
1694 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001695 struct mem_cgroup *iter;
1696 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001697
Sha Zhengju58cf1882013-02-22 16:32:05 -08001698 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001699 return;
1700
Michal Hocko947b3dd2014-01-21 15:51:04 -08001701 spin_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001702 rcu_read_lock();
1703
1704 mem_cgrp = memcg->css.cgroup;
1705 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1706
1707 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1708 if (ret < 0) {
1709 /*
1710 * Unfortunately, we are unable to convert to a useful name
1711 * But we'll still print out the usage information
1712 */
1713 rcu_read_unlock();
1714 goto done;
1715 }
1716 rcu_read_unlock();
1717
Andrew Mortond0451972013-02-22 16:32:06 -08001718 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001719
1720 rcu_read_lock();
1721 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1722 if (ret < 0) {
1723 rcu_read_unlock();
1724 goto done;
1725 }
1726 rcu_read_unlock();
1727
1728 /*
1729 * Continues from above, so we don't need an KERN_ level
1730 */
Andrew Mortond0451972013-02-22 16:32:06 -08001731 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001732done:
1733
Andrew Mortond0451972013-02-22 16:32:06 -08001734 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001735 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1736 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1737 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001738 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001739 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1740 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1741 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001742 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001743 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1744 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1745 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001746
1747 for_each_mem_cgroup_tree(iter, memcg) {
1748 pr_info("Memory cgroup stats");
1749
1750 rcu_read_lock();
1751 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1752 if (!ret)
1753 pr_cont(" for %s", memcg_name);
1754 rcu_read_unlock();
1755 pr_cont(":");
1756
1757 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1758 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1759 continue;
1760 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1761 K(mem_cgroup_read_stat(iter, i)));
1762 }
1763
1764 for (i = 0; i < NR_LRU_LISTS; i++)
1765 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1766 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1767
1768 pr_cont("\n");
1769 }
Michal Hocko947b3dd2014-01-21 15:51:04 -08001770 spin_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001771}
1772
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001773/*
1774 * This function returns the number of memcg under hierarchy tree. Returns
1775 * 1(self count) if no children.
1776 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001777static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001778{
1779 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001780 struct mem_cgroup *iter;
1781
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001782 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001783 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001784 return num;
1785}
1786
Balbir Singh6d61ef42009-01-07 18:08:06 -08001787/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001788 * Return the memory (and swap, if configured) limit for a memcg.
1789 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001790static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001791{
1792 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001793
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001794 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001795
David Rientjesa63d83f2010-08-09 17:19:46 -07001796 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001797 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001798 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001799 if (mem_cgroup_swappiness(memcg)) {
1800 u64 memsw;
1801
1802 limit += total_swap_pages << PAGE_SHIFT;
1803 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1804
1805 /*
1806 * If memsw is finite and limits the amount of swap space
1807 * available to this memcg, return that limit.
1808 */
1809 limit = min(limit, memsw);
1810 }
1811
1812 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001813}
1814
David Rientjes19965462012-12-11 16:00:26 -08001815static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1816 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001817{
1818 struct mem_cgroup *iter;
1819 unsigned long chosen_points = 0;
1820 unsigned long totalpages;
1821 unsigned int points = 0;
1822 struct task_struct *chosen = NULL;
1823
David Rientjes876aafb2012-07-31 16:43:48 -07001824 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001825 * If current has a pending SIGKILL or is exiting, then automatically
1826 * select it. The goal is to allow it to allocate so that it may
1827 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001828 */
David Rientjes465adcf2013-04-29 15:08:45 -07001829 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001830 set_thread_flag(TIF_MEMDIE);
1831 return;
1832 }
1833
1834 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001835 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1836 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001837 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001838 struct task_struct *task;
1839
Tejun Heo72ec7022013-08-08 20:11:26 -04001840 css_task_iter_start(&iter->css, &it);
1841 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001842 switch (oom_scan_process_thread(task, totalpages, NULL,
1843 false)) {
1844 case OOM_SCAN_SELECT:
1845 if (chosen)
1846 put_task_struct(chosen);
1847 chosen = task;
1848 chosen_points = ULONG_MAX;
1849 get_task_struct(chosen);
1850 /* fall through */
1851 case OOM_SCAN_CONTINUE:
1852 continue;
1853 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001854 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001855 mem_cgroup_iter_break(memcg, iter);
1856 if (chosen)
1857 put_task_struct(chosen);
1858 return;
1859 case OOM_SCAN_OK:
1860 break;
1861 };
1862 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001863 if (!points || points < chosen_points)
1864 continue;
1865 /* Prefer thread group leaders for display purposes */
1866 if (points == chosen_points &&
1867 thread_group_leader(chosen))
1868 continue;
1869
1870 if (chosen)
1871 put_task_struct(chosen);
1872 chosen = task;
1873 chosen_points = points;
1874 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001875 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001876 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001877 }
1878
1879 if (!chosen)
1880 return;
1881 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001882 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1883 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001884}
1885
Johannes Weiner56600482012-01-12 17:17:59 -08001886static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1887 gfp_t gfp_mask,
1888 unsigned long flags)
1889{
1890 unsigned long total = 0;
1891 bool noswap = false;
1892 int loop;
1893
1894 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1895 noswap = true;
1896 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1897 noswap = true;
1898
1899 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1900 if (loop)
1901 drain_all_stock_async(memcg);
1902 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1903 /*
1904 * Allow limit shrinkers, which are triggered directly
1905 * by userspace, to catch signals and stop reclaim
1906 * after minimal progress, regardless of the margin.
1907 */
1908 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1909 break;
1910 if (mem_cgroup_margin(memcg))
1911 break;
1912 /*
1913 * If nothing was reclaimed after two attempts, there
1914 * may be no reclaimable pages in this hierarchy.
1915 */
1916 if (loop && !total)
1917 break;
1918 }
1919 return total;
1920}
1921
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001922/**
1923 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001924 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001925 * @nid: the node ID to be checked.
1926 * @noswap : specify true here if the user wants flle only information.
1927 *
1928 * This function returns whether the specified memcg contains any
1929 * reclaimable pages on a node. Returns true if there are any reclaimable
1930 * pages in the node.
1931 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001932static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001933 int nid, bool noswap)
1934{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001935 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001936 return true;
1937 if (noswap || !total_swap_pages)
1938 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001939 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001940 return true;
1941 return false;
1942
1943}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001944#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001945
1946/*
1947 * Always updating the nodemask is not very good - even if we have an empty
1948 * list or the wrong list here, we can start from some node and traverse all
1949 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1950 *
1951 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001952static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001953{
1954 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001955 /*
1956 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1957 * pagein/pageout changes since the last update.
1958 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001959 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001960 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001961 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001962 return;
1963
Ying Han889976d2011-05-26 16:25:33 -07001964 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001965 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001966
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001967 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001968
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001969 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1970 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001971 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001972
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001973 atomic_set(&memcg->numainfo_events, 0);
1974 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001975}
1976
1977/*
1978 * Selecting a node where we start reclaim from. Because what we need is just
1979 * reducing usage counter, start from anywhere is O,K. Considering
1980 * memory reclaim from current node, there are pros. and cons.
1981 *
1982 * Freeing memory from current node means freeing memory from a node which
1983 * we'll use or we've used. So, it may make LRU bad. And if several threads
1984 * hit limits, it will see a contention on a node. But freeing from remote
1985 * node means more costs for memory reclaim because of memory latency.
1986 *
1987 * Now, we use round-robin. Better algorithm is welcomed.
1988 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001989int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001990{
1991 int node;
1992
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001993 mem_cgroup_may_update_nodemask(memcg);
1994 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001995
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001996 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001997 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001998 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001999 /*
2000 * We call this when we hit limit, not when pages are added to LRU.
2001 * No LRU may hold pages because all pages are UNEVICTABLE or
2002 * memcg is too small and all pages are not on LRU. In that case,
2003 * we use curret node.
2004 */
2005 if (unlikely(node == MAX_NUMNODES))
2006 node = numa_node_id();
2007
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002008 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07002009 return node;
2010}
2011
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002012/*
2013 * Check all nodes whether it contains reclaimable pages or not.
2014 * For quick scan, we make use of scan_nodes. This will allow us to skip
2015 * unused nodes. But scan_nodes is lazily updated and may not cotain
2016 * enough new information. We need to do double check.
2017 */
2018static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
2019{
2020 int nid;
2021
2022 /*
2023 * quick check...making use of scan_node.
2024 * We can skip unused nodes.
2025 */
2026 if (!nodes_empty(memcg->scan_nodes)) {
2027 for (nid = first_node(memcg->scan_nodes);
2028 nid < MAX_NUMNODES;
2029 nid = next_node(nid, memcg->scan_nodes)) {
2030
2031 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
2032 return true;
2033 }
2034 }
2035 /*
2036 * Check rest of nodes.
2037 */
2038 for_each_node_state(nid, N_MEMORY) {
2039 if (node_isset(nid, memcg->scan_nodes))
2040 continue;
2041 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
2042 return true;
2043 }
2044 return false;
2045}
2046
Ying Han889976d2011-05-26 16:25:33 -07002047#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002048int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002049{
2050 return 0;
2051}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002052
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002053static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
2054{
2055 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
2056}
Ying Han889976d2011-05-26 16:25:33 -07002057#endif
2058
Andrew Morton0608f432013-09-24 15:27:41 -07002059static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2060 struct zone *zone,
2061 gfp_t gfp_mask,
2062 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002063{
Andrew Morton0608f432013-09-24 15:27:41 -07002064 struct mem_cgroup *victim = NULL;
2065 int total = 0;
2066 int loop = 0;
2067 unsigned long excess;
2068 unsigned long nr_scanned;
2069 struct mem_cgroup_reclaim_cookie reclaim = {
2070 .zone = zone,
2071 .priority = 0,
2072 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002073
Andrew Morton0608f432013-09-24 15:27:41 -07002074 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002075
Andrew Morton0608f432013-09-24 15:27:41 -07002076 while (1) {
2077 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
2078 if (!victim) {
2079 loop++;
2080 if (loop >= 2) {
2081 /*
2082 * If we have not been able to reclaim
2083 * anything, it might because there are
2084 * no reclaimable pages under this hierarchy
2085 */
2086 if (!total)
2087 break;
2088 /*
2089 * We want to do more targeted reclaim.
2090 * excess >> 2 is not to excessive so as to
2091 * reclaim too much, nor too less that we keep
2092 * coming back to reclaim from this cgroup
2093 */
2094 if (total >= (excess >> 2) ||
2095 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
2096 break;
2097 }
2098 continue;
2099 }
2100 if (!mem_cgroup_reclaimable(victim, false))
2101 continue;
2102 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2103 zone, &nr_scanned);
2104 *total_scanned += nr_scanned;
2105 if (!res_counter_soft_limit_excess(&root_memcg->res))
2106 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002107 }
Andrew Morton0608f432013-09-24 15:27:41 -07002108 mem_cgroup_iter_break(root_memcg, victim);
2109 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002110}
2111
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002112#ifdef CONFIG_LOCKDEP
2113static struct lockdep_map memcg_oom_lock_dep_map = {
2114 .name = "memcg_oom_lock",
2115};
2116#endif
2117
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002118static DEFINE_SPINLOCK(memcg_oom_lock);
2119
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002120/*
2121 * Check OOM-Killer is already running under our hierarchy.
2122 * If someone is running, return false.
2123 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002124static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002125{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002126 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002127
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002128 spin_lock(&memcg_oom_lock);
2129
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002130 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002131 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002132 /*
2133 * this subtree of our hierarchy is already locked
2134 * so we cannot give a lock.
2135 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002136 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002137 mem_cgroup_iter_break(memcg, iter);
2138 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002139 } else
2140 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002141 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002142
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002143 if (failed) {
2144 /*
2145 * OK, we failed to lock the whole subtree so we have
2146 * to clean up what we set up to the failing subtree
2147 */
2148 for_each_mem_cgroup_tree(iter, memcg) {
2149 if (iter == failed) {
2150 mem_cgroup_iter_break(memcg, iter);
2151 break;
2152 }
2153 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002154 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002155 } else
2156 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002157
2158 spin_unlock(&memcg_oom_lock);
2159
2160 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002161}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002162
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002163static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002164{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002165 struct mem_cgroup *iter;
2166
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002167 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002168 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002169 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002170 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002171 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002172}
2173
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002174static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002175{
2176 struct mem_cgroup *iter;
2177
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002178 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002179 atomic_inc(&iter->under_oom);
2180}
2181
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002182static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002183{
2184 struct mem_cgroup *iter;
2185
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002186 /*
2187 * When a new child is created while the hierarchy is under oom,
2188 * mem_cgroup_oom_lock() may not be called. We have to use
2189 * atomic_add_unless() here.
2190 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002191 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002192 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002193}
2194
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002195static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2196
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002197struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002198 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002199 wait_queue_t wait;
2200};
2201
2202static int memcg_oom_wake_function(wait_queue_t *wait,
2203 unsigned mode, int sync, void *arg)
2204{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002205 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2206 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002207 struct oom_wait_info *oom_wait_info;
2208
2209 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002210 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002211
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002212 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002213 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002214 * Then we can use css_is_ancestor without taking care of RCU.
2215 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002216 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2217 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002218 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002219 return autoremove_wake_function(wait, mode, sync, arg);
2220}
2221
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002222static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002223{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002224 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002225 /* for filtering, pass "memcg" as argument. */
2226 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002227}
2228
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002229static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002230{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002231 if (memcg && atomic_read(&memcg->under_oom))
2232 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002233}
2234
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002235static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002236{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002237 if (!current->memcg_oom.may_oom)
2238 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002239 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002240 * We are in the middle of the charge context here, so we
2241 * don't want to block when potentially sitting on a callstack
2242 * that holds all kinds of filesystem and mm locks.
2243 *
2244 * Also, the caller may handle a failed allocation gracefully
2245 * (like optional page cache readahead) and so an OOM killer
2246 * invocation might not even be necessary.
2247 *
2248 * That's why we don't do anything here except remember the
2249 * OOM context and then deal with it at the end of the page
2250 * fault when the stack is unwound, the locks are released,
2251 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002252 */
Johannes Weiner49426422013-10-16 13:46:59 -07002253 css_get(&memcg->css);
2254 current->memcg_oom.memcg = memcg;
2255 current->memcg_oom.gfp_mask = mask;
2256 current->memcg_oom.order = order;
2257}
2258
2259/**
2260 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2261 * @handle: actually kill/wait or just clean up the OOM state
2262 *
2263 * This has to be called at the end of a page fault if the memcg OOM
2264 * handler was enabled.
2265 *
2266 * Memcg supports userspace OOM handling where failed allocations must
2267 * sleep on a waitqueue until the userspace task resolves the
2268 * situation. Sleeping directly in the charge context with all kinds
2269 * of locks held is not a good idea, instead we remember an OOM state
2270 * in the task and mem_cgroup_oom_synchronize() has to be called at
2271 * the end of the page fault to complete the OOM handling.
2272 *
2273 * Returns %true if an ongoing memcg OOM situation was detected and
2274 * completed, %false otherwise.
2275 */
2276bool mem_cgroup_oom_synchronize(bool handle)
2277{
2278 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2279 struct oom_wait_info owait;
2280 bool locked;
2281
2282 /* OOM is global, do not handle */
2283 if (!memcg)
2284 return false;
2285
2286 if (!handle)
2287 goto cleanup;
2288
2289 owait.memcg = memcg;
2290 owait.wait.flags = 0;
2291 owait.wait.func = memcg_oom_wake_function;
2292 owait.wait.private = current;
2293 INIT_LIST_HEAD(&owait.wait.task_list);
2294
2295 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002296 mem_cgroup_mark_under_oom(memcg);
2297
2298 locked = mem_cgroup_oom_trylock(memcg);
2299
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002300 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002301 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002302
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002303 if (locked && !memcg->oom_kill_disable) {
2304 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002305 finish_wait(&memcg_oom_waitq, &owait.wait);
2306 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2307 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002308 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002309 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002310 mem_cgroup_unmark_under_oom(memcg);
2311 finish_wait(&memcg_oom_waitq, &owait.wait);
2312 }
2313
2314 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002315 mem_cgroup_oom_unlock(memcg);
2316 /*
2317 * There is no guarantee that an OOM-lock contender
2318 * sees the wakeups triggered by the OOM kill
2319 * uncharges. Wake any sleepers explicitely.
2320 */
2321 memcg_oom_recover(memcg);
2322 }
Johannes Weiner49426422013-10-16 13:46:59 -07002323cleanup:
2324 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002325 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002326 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002327}
2328
Balbir Singhd69b0422009-06-17 16:26:34 -07002329/*
2330 * Currently used to update mapped file statistics, but the routine can be
2331 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002332 *
2333 * Notes: Race condition
2334 *
2335 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2336 * it tends to be costly. But considering some conditions, we doesn't need
2337 * to do so _always_.
2338 *
2339 * Considering "charge", lock_page_cgroup() is not required because all
2340 * file-stat operations happen after a page is attached to radix-tree. There
2341 * are no race with "charge".
2342 *
2343 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2344 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2345 * if there are race with "uncharge". Statistics itself is properly handled
2346 * by flags.
2347 *
2348 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002349 * small, we check mm->moving_account and detect there are possibility of race
2350 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002351 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002352
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002353void __mem_cgroup_begin_update_page_stat(struct page *page,
2354 bool *locked, unsigned long *flags)
2355{
2356 struct mem_cgroup *memcg;
2357 struct page_cgroup *pc;
2358
2359 pc = lookup_page_cgroup(page);
2360again:
2361 memcg = pc->mem_cgroup;
2362 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2363 return;
2364 /*
2365 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002366 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002367 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002368 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002369 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002370 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002371 return;
2372
2373 move_lock_mem_cgroup(memcg, flags);
2374 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2375 move_unlock_mem_cgroup(memcg, flags);
2376 goto again;
2377 }
2378 *locked = true;
2379}
2380
2381void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2382{
2383 struct page_cgroup *pc = lookup_page_cgroup(page);
2384
2385 /*
2386 * It's guaranteed that pc->mem_cgroup never changes while
2387 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002388 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002389 */
2390 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2391}
2392
Greg Thelen2a7106f2011-01-13 15:47:37 -08002393void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002394 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002395{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002396 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002397 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002398 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002399
Johannes Weinercfa44942012-01-12 17:18:38 -08002400 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002401 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002402
Sha Zhengju658b72c2013-09-12 15:13:52 -07002403 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002404 memcg = pc->mem_cgroup;
2405 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002406 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002407
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002408 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002409}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002410
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002411/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002412 * size of first charge trial. "32" comes from vmscan.c's magic value.
2413 * TODO: maybe necessary to use big numbers in big irons.
2414 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002415#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002416struct memcg_stock_pcp {
2417 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002418 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002419 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002420 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002421#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002422};
2423static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002424static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002425
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002426/**
2427 * consume_stock: Try to consume stocked charge on this cpu.
2428 * @memcg: memcg to consume from.
2429 * @nr_pages: how many pages to charge.
2430 *
2431 * The charges will only happen if @memcg matches the current cpu's memcg
2432 * stock, and at least @nr_pages are available in that stock. Failure to
2433 * service an allocation will refill the stock.
2434 *
2435 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002436 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002437static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002438{
2439 struct memcg_stock_pcp *stock;
2440 bool ret = true;
2441
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002442 if (nr_pages > CHARGE_BATCH)
2443 return false;
2444
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002445 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002446 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2447 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002448 else /* need to call res_counter_charge */
2449 ret = false;
2450 put_cpu_var(memcg_stock);
2451 return ret;
2452}
2453
2454/*
2455 * Returns stocks cached in percpu to res_counter and reset cached information.
2456 */
2457static void drain_stock(struct memcg_stock_pcp *stock)
2458{
2459 struct mem_cgroup *old = stock->cached;
2460
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002461 if (stock->nr_pages) {
2462 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2463
2464 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002465 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002466 res_counter_uncharge(&old->memsw, bytes);
2467 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002468 }
2469 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002470}
2471
2472/*
2473 * This must be called under preempt disabled or must be called by
2474 * a thread which is pinned to local cpu.
2475 */
2476static void drain_local_stock(struct work_struct *dummy)
2477{
2478 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2479 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002480 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002481}
2482
Michal Hockoe4777492013-02-22 16:35:40 -08002483static void __init memcg_stock_init(void)
2484{
2485 int cpu;
2486
2487 for_each_possible_cpu(cpu) {
2488 struct memcg_stock_pcp *stock =
2489 &per_cpu(memcg_stock, cpu);
2490 INIT_WORK(&stock->work, drain_local_stock);
2491 }
2492}
2493
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002494/*
2495 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002496 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002497 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002498static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002499{
2500 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2501
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002502 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002503 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002504 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002505 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002506 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002507 put_cpu_var(memcg_stock);
2508}
2509
2510/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002511 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002512 * of the hierarchy under it. sync flag says whether we should block
2513 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002514 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002515static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002516{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002517 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002518
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002519 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002520 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002521 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002522 for_each_online_cpu(cpu) {
2523 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002524 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002525
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002526 memcg = stock->cached;
2527 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002528 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002529 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002530 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002531 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2532 if (cpu == curcpu)
2533 drain_local_stock(&stock->work);
2534 else
2535 schedule_work_on(cpu, &stock->work);
2536 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002537 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002538 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002539
2540 if (!sync)
2541 goto out;
2542
2543 for_each_online_cpu(cpu) {
2544 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002545 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002546 flush_work(&stock->work);
2547 }
2548out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002549 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002550}
2551
2552/*
2553 * Tries to drain stocked charges in other cpus. This function is asynchronous
2554 * and just put a work per cpu for draining localy on each cpu. Caller can
2555 * expects some charges will be back to res_counter later but cannot wait for
2556 * it.
2557 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002558static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002559{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002560 /*
2561 * If someone calls draining, avoid adding more kworker runs.
2562 */
2563 if (!mutex_trylock(&percpu_charge_mutex))
2564 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002565 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002566 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002567}
2568
2569/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002570static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002571{
2572 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002573 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002574 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002575 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002576}
2577
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002578/*
2579 * This function drains percpu counter value from DEAD cpu and
2580 * move it to local cpu. Note that this function can be preempted.
2581 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002582static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002583{
2584 int i;
2585
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002586 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002587 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002588 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002589
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002590 per_cpu(memcg->stat->count[i], cpu) = 0;
2591 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002592 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002593 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002594 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002595
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002596 per_cpu(memcg->stat->events[i], cpu) = 0;
2597 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002598 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002599 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002600}
2601
Paul Gortmaker0db06282013-06-19 14:53:51 -04002602static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002603 unsigned long action,
2604 void *hcpu)
2605{
2606 int cpu = (unsigned long)hcpu;
2607 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002608 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002609
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002610 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002611 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002612
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002613 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002614 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002615
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002616 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002617 mem_cgroup_drain_pcp_counter(iter, cpu);
2618
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002619 stock = &per_cpu(memcg_stock, cpu);
2620 drain_stock(stock);
2621 return NOTIFY_OK;
2622}
2623
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002624
2625/* See __mem_cgroup_try_charge() for details */
2626enum {
2627 CHARGE_OK, /* success */
2628 CHARGE_RETRY, /* need to retry but retry is not bad */
2629 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2630 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002631};
2632
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002633static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002634 unsigned int nr_pages, unsigned int min_pages,
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002635 bool invoke_oom)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002636{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002637 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002638 struct mem_cgroup *mem_over_limit;
2639 struct res_counter *fail_res;
2640 unsigned long flags = 0;
2641 int ret;
2642
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002643 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002644
2645 if (likely(!ret)) {
2646 if (!do_swap_account)
2647 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002648 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002649 if (likely(!ret))
2650 return CHARGE_OK;
2651
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002652 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002653 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2654 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2655 } else
2656 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002657 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002658 * Never reclaim on behalf of optional batching, retry with a
2659 * single page instead.
2660 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002661 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002662 return CHARGE_RETRY;
2663
2664 if (!(gfp_mask & __GFP_WAIT))
2665 return CHARGE_WOULDBLOCK;
2666
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002667 if (gfp_mask & __GFP_NORETRY)
2668 return CHARGE_NOMEM;
2669
Johannes Weiner56600482012-01-12 17:17:59 -08002670 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002671 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002672 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002673 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002674 * Even though the limit is exceeded at this point, reclaim
2675 * may have been able to free some pages. Retry the charge
2676 * before killing the task.
2677 *
2678 * Only for regular pages, though: huge pages are rather
2679 * unlikely to succeed so close to the limit, and we fall back
2680 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002681 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002682 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002683 return CHARGE_RETRY;
2684
2685 /*
2686 * At task move, charge accounts can be doubly counted. So, it's
2687 * better to wait until the end of task_move if something is going on.
2688 */
2689 if (mem_cgroup_wait_acct_move(mem_over_limit))
2690 return CHARGE_RETRY;
2691
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002692 if (invoke_oom)
2693 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002694
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002695 return CHARGE_NOMEM;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002696}
2697
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002698/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002699 * __mem_cgroup_try_charge() does
2700 * 1. detect memcg to be charged against from passed *mm and *ptr,
2701 * 2. update res_counter
2702 * 3. call memory reclaim if necessary.
2703 *
2704 * In some special case, if the task is fatal, fatal_signal_pending() or
2705 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2706 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2707 * as possible without any hazards. 2: all pages should have a valid
2708 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2709 * pointer, that is treated as a charge to root_mem_cgroup.
2710 *
2711 * So __mem_cgroup_try_charge() will return
2712 * 0 ... on success, filling *ptr with a valid memcg pointer.
2713 * -ENOMEM ... charge failure because of resource limits.
2714 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2715 *
2716 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2717 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002718 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002719static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002720 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002721 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002723 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002724{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002725 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002726 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002727 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002728 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002729
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002730 /*
2731 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2732 * in system level. So, allow to go ahead dying process in addition to
2733 * MEMDIE process.
2734 */
2735 if (unlikely(test_thread_flag(TIF_MEMDIE)
2736 || fatal_signal_pending(current)))
2737 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002738
Johannes Weiner49426422013-10-16 13:46:59 -07002739 if (unlikely(task_in_memcg_oom(current)))
Johannes Weiner1f14c1a2013-12-12 17:12:35 -08002740 goto nomem;
Johannes Weiner49426422013-10-16 13:46:59 -07002741
Johannes Weinera0d8b002013-12-12 17:12:20 -08002742 if (gfp_mask & __GFP_NOFAIL)
2743 oom = false;
2744
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002745 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002746 * We always charge the cgroup the mm_struct belongs to.
2747 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002748 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002749 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002750 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002751 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002752 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002753again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002754 if (*ptr) { /* css should be a valid one */
2755 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002756 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002757 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002758 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002759 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002760 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002761 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002762 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002763
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002764 rcu_read_lock();
2765 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002766 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002767 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002768 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002769 * race with swapoff. Then, we have small risk of mis-accouning.
2770 * But such kind of mis-account by race always happens because
2771 * we don't have cgroup_mutex(). It's overkill and we allo that
2772 * small race, here.
2773 * (*) swapoff at el will charge against mm-struct not against
2774 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002775 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002776 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002777 if (!memcg)
2778 memcg = root_mem_cgroup;
2779 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002780 rcu_read_unlock();
2781 goto done;
2782 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002783 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002784 /*
2785 * It seems dagerous to access memcg without css_get().
2786 * But considering how consume_stok works, it's not
2787 * necessary. If consume_stock success, some charges
2788 * from this memcg are cached on this cpu. So, we
2789 * don't need to call css_get()/css_tryget() before
2790 * calling consume_stock().
2791 */
2792 rcu_read_unlock();
2793 goto done;
2794 }
2795 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002796 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002797 rcu_read_unlock();
2798 goto again;
2799 }
2800 rcu_read_unlock();
2801 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002802
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002803 do {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002804 bool invoke_oom = oom && !nr_oom_retries;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002805
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002806 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002807 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002808 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002809 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002810 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002811
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002812 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
2813 nr_pages, invoke_oom);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002814 switch (ret) {
2815 case CHARGE_OK:
2816 break;
2817 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002818 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002819 css_put(&memcg->css);
2820 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002821 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002822 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002823 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002824 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002825 case CHARGE_NOMEM: /* OOM routine works */
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002826 if (!oom || invoke_oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002827 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002828 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002829 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002830 nr_oom_retries--;
2831 break;
Balbir Singh66e17072008-02-07 00:13:56 -08002832 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002833 } while (ret != CHARGE_OK);
2834
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002835 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002836 refill_stock(memcg, batch - nr_pages);
2837 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002838done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002839 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002840 return 0;
2841nomem:
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002842 if (!(gfp_mask & __GFP_NOFAIL)) {
2843 *ptr = NULL;
2844 return -ENOMEM;
2845 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002846bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002847 *ptr = root_mem_cgroup;
2848 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002849}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002850
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002851/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002852 * Somemtimes we have to undo a charge we got by try_charge().
2853 * This function is for that and do uncharge, put css's refcnt.
2854 * gotten by try_charge().
2855 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002856static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002857 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002858{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002859 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002860 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002861
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002862 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002863 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002864 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002865 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002866}
2867
2868/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002869 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2870 * This is useful when moving usage to parent cgroup.
2871 */
2872static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2873 unsigned int nr_pages)
2874{
2875 unsigned long bytes = nr_pages * PAGE_SIZE;
2876
2877 if (mem_cgroup_is_root(memcg))
2878 return;
2879
2880 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2881 if (do_swap_account)
2882 res_counter_uncharge_until(&memcg->memsw,
2883 memcg->memsw.parent, bytes);
2884}
2885
2886/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002887 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002888 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2889 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2890 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002891 */
2892static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2893{
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002894 /* ID 0 is unused ID */
2895 if (!id)
2896 return NULL;
Li Zefan34c00c32013-09-23 16:56:01 +08002897 return mem_cgroup_from_id(id);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002898}
2899
Wu Fengguange42d9d52009-12-16 12:19:59 +01002900struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002901{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002902 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002903 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002904 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002905 swp_entry_t ent;
2906
Sasha Levin309381fea2014-01-23 15:52:54 -08002907 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002908
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002909 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002910 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002911 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002912 memcg = pc->mem_cgroup;
2913 if (memcg && !css_tryget(&memcg->css))
2914 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002915 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002916 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002917 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002918 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002919 memcg = mem_cgroup_lookup(id);
2920 if (memcg && !css_tryget(&memcg->css))
2921 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002922 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002923 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002924 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002925 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002926}
2927
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002928static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002929 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002930 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002931 enum charge_type ctype,
2932 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002933{
Johannes Weinerce587e62012-04-24 20:22:33 +02002934 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002935 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002936 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002937 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002938 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002939
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002940 lock_page_cgroup(pc);
Sasha Levin309381fea2014-01-23 15:52:54 -08002941 VM_BUG_ON_PAGE(PageCgroupUsed(pc), page);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002942 /*
2943 * we don't need page_cgroup_lock about tail pages, becase they are not
2944 * accessed by any other context at this point.
2945 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002946
2947 /*
2948 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2949 * may already be on some other mem_cgroup's LRU. Take care of it.
2950 */
2951 if (lrucare) {
2952 zone = page_zone(page);
2953 spin_lock_irq(&zone->lru_lock);
2954 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002955 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002956 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002957 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002958 was_on_lru = true;
2959 }
2960 }
2961
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002962 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002963 /*
2964 * We access a page_cgroup asynchronously without lock_page_cgroup().
2965 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2966 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2967 * before USED bit, we need memory barrier here.
2968 * See mem_cgroup_add_lru_list(), etc.
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002969 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002970 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002971 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002972
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002973 if (lrucare) {
2974 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002975 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Sasha Levin309381fea2014-01-23 15:52:54 -08002976 VM_BUG_ON_PAGE(PageLRU(page), page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002977 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002978 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002979 }
2980 spin_unlock_irq(&zone->lru_lock);
2981 }
2982
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002983 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002984 anon = true;
2985 else
2986 anon = false;
2987
David Rientjesb070e652013-05-07 16:18:09 -07002988 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002989 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002990
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002991 /*
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002992 * "charge_statistics" updated event counter. Then, check it.
2993 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2994 * if they exceeds softlimit.
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002995 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002996 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002997}
2998
Glauber Costa7cf27982012-12-18 14:22:55 -08002999static DEFINE_MUTEX(set_limit_mutex);
3000
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003001#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08003002static DEFINE_MUTEX(activate_kmem_mutex);
3003
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003004static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
3005{
3006 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
Vladimir Davydov6de64be2014-01-23 15:53:08 -08003007 memcg_kmem_is_active(memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003008}
3009
Glauber Costa1f458cb2012-12-18 14:22:50 -08003010/*
3011 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
3012 * in the memcg_cache_params struct.
3013 */
3014static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
3015{
3016 struct kmem_cache *cachep;
3017
3018 VM_BUG_ON(p->is_root_cache);
3019 cachep = p->root_cache;
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003020 return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg));
Glauber Costa1f458cb2012-12-18 14:22:50 -08003021}
3022
Glauber Costa749c5412012-12-18 14:23:01 -08003023#ifdef CONFIG_SLABINFO
Tejun Heo2da8ca82013-12-05 12:28:04 -05003024static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v)
Glauber Costa749c5412012-12-18 14:23:01 -08003025{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003026 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Glauber Costa749c5412012-12-18 14:23:01 -08003027 struct memcg_cache_params *params;
3028
3029 if (!memcg_can_account_kmem(memcg))
3030 return -EIO;
3031
3032 print_slabinfo_header(m);
3033
3034 mutex_lock(&memcg->slab_caches_mutex);
3035 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
3036 cache_show(memcg_params_to_cache(params), m);
3037 mutex_unlock(&memcg->slab_caches_mutex);
3038
3039 return 0;
3040}
3041#endif
3042
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003043static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
3044{
3045 struct res_counter *fail_res;
3046 struct mem_cgroup *_memcg;
3047 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003048
3049 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
3050 if (ret)
3051 return ret;
3052
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003053 _memcg = memcg;
3054 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
Qiang Huangb9921ec2013-11-12 15:07:22 -08003055 &_memcg, oom_gfp_allowed(gfp));
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003056
3057 if (ret == -EINTR) {
3058 /*
3059 * __mem_cgroup_try_charge() chosed to bypass to root due to
3060 * OOM kill or fatal signal. Since our only options are to
3061 * either fail the allocation or charge it to this cgroup, do
3062 * it as a temporary condition. But we can't fail. From a
3063 * kmem/slab perspective, the cache has already been selected,
3064 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3065 * our minds.
3066 *
3067 * This condition will only trigger if the task entered
3068 * memcg_charge_kmem in a sane state, but was OOM-killed during
3069 * __mem_cgroup_try_charge() above. Tasks that were already
3070 * dying when the allocation triggers should have been already
3071 * directed to the root cgroup in memcontrol.h
3072 */
3073 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3074 if (do_swap_account)
3075 res_counter_charge_nofail(&memcg->memsw, size,
3076 &fail_res);
3077 ret = 0;
3078 } else if (ret)
3079 res_counter_uncharge(&memcg->kmem, size);
3080
3081 return ret;
3082}
3083
3084static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3085{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003086 res_counter_uncharge(&memcg->res, size);
3087 if (do_swap_account)
3088 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003089
3090 /* Not down to 0 */
3091 if (res_counter_uncharge(&memcg->kmem, size))
3092 return;
3093
Li Zefan10d5ebf2013-07-08 16:00:33 -07003094 /*
3095 * Releases a reference taken in kmem_cgroup_css_offline in case
3096 * this last uncharge is racing with the offlining code or it is
3097 * outliving the memcg existence.
3098 *
3099 * The memory barrier imposed by test&clear is paired with the
3100 * explicit one in memcg_kmem_mark_dead().
3101 */
Glauber Costa7de37682012-12-18 14:22:07 -08003102 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003103 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003104}
3105
Glauber Costa2633d7a2012-12-18 14:22:34 -08003106/*
3107 * helper for acessing a memcg's index. It will be used as an index in the
3108 * child cache array in kmem_cache, and also to derive its name. This function
3109 * will return -1 when this is not a kmem-limited memcg.
3110 */
3111int memcg_cache_id(struct mem_cgroup *memcg)
3112{
3113 return memcg ? memcg->kmemcg_id : -1;
3114}
3115
Glauber Costa55007d82012-12-18 14:22:38 -08003116static size_t memcg_caches_array_size(int num_groups)
3117{
3118 ssize_t size;
3119 if (num_groups <= 0)
3120 return 0;
3121
3122 size = 2 * num_groups;
3123 if (size < MEMCG_CACHES_MIN_SIZE)
3124 size = MEMCG_CACHES_MIN_SIZE;
3125 else if (size > MEMCG_CACHES_MAX_SIZE)
3126 size = MEMCG_CACHES_MAX_SIZE;
3127
3128 return size;
3129}
3130
3131/*
3132 * We should update the current array size iff all caches updates succeed. This
3133 * can only be done from the slab side. The slab mutex needs to be held when
3134 * calling this.
3135 */
3136void memcg_update_array_size(int num)
3137{
3138 if (num > memcg_limited_groups_array_size)
3139 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3140}
3141
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003142static void kmem_cache_destroy_work_func(struct work_struct *w);
3143
Glauber Costa55007d82012-12-18 14:22:38 -08003144int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3145{
3146 struct memcg_cache_params *cur_params = s->memcg_params;
3147
Qiang Huangf35c3a82013-11-12 15:08:22 -08003148 VM_BUG_ON(!is_root_cache(s));
Glauber Costa55007d82012-12-18 14:22:38 -08003149
3150 if (num_groups > memcg_limited_groups_array_size) {
3151 int i;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003152 struct memcg_cache_params *new_params;
Glauber Costa55007d82012-12-18 14:22:38 -08003153 ssize_t size = memcg_caches_array_size(num_groups);
3154
3155 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003156 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003157
Vladimir Davydovf8570262014-01-23 15:53:06 -08003158 new_params = kzalloc(size, GFP_KERNEL);
3159 if (!new_params)
Glauber Costa55007d82012-12-18 14:22:38 -08003160 return -ENOMEM;
Glauber Costa55007d82012-12-18 14:22:38 -08003161
Vladimir Davydovf8570262014-01-23 15:53:06 -08003162 new_params->is_root_cache = true;
Glauber Costa55007d82012-12-18 14:22:38 -08003163
3164 /*
3165 * There is the chance it will be bigger than
3166 * memcg_limited_groups_array_size, if we failed an allocation
3167 * in a cache, in which case all caches updated before it, will
3168 * have a bigger array.
3169 *
3170 * But if that is the case, the data after
3171 * memcg_limited_groups_array_size is certainly unused
3172 */
3173 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3174 if (!cur_params->memcg_caches[i])
3175 continue;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003176 new_params->memcg_caches[i] =
Glauber Costa55007d82012-12-18 14:22:38 -08003177 cur_params->memcg_caches[i];
3178 }
3179
3180 /*
3181 * Ideally, we would wait until all caches succeed, and only
3182 * then free the old one. But this is not worth the extra
3183 * pointer per-cache we'd have to have for this.
3184 *
3185 * It is not a big deal if some caches are left with a size
3186 * bigger than the others. And all updates will reset this
3187 * anyway.
3188 */
Vladimir Davydovf8570262014-01-23 15:53:06 -08003189 rcu_assign_pointer(s->memcg_params, new_params);
3190 if (cur_params)
3191 kfree_rcu(cur_params, rcu_head);
Glauber Costa55007d82012-12-18 14:22:38 -08003192 }
3193 return 0;
3194}
3195
Vladimir Davydov363a0442014-01-23 15:52:56 -08003196int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s,
3197 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003198{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003199 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003200
3201 if (!memcg_kmem_enabled())
3202 return 0;
3203
Andrey Vagin90c7a792013-09-11 14:22:18 -07003204 if (!memcg) {
3205 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003206 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003207 } else
3208 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003209
Glauber Costa2633d7a2012-12-18 14:22:34 -08003210 s->memcg_params = kzalloc(size, GFP_KERNEL);
3211 if (!s->memcg_params)
3212 return -ENOMEM;
3213
Glauber Costa943a4512012-12-18 14:23:03 -08003214 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003215 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003216 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003217 INIT_WORK(&s->memcg_params->destroy,
3218 kmem_cache_destroy_work_func);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003219 } else
3220 s->memcg_params->is_root_cache = true;
3221
Glauber Costa2633d7a2012-12-18 14:22:34 -08003222 return 0;
3223}
3224
Vladimir Davydov363a0442014-01-23 15:52:56 -08003225void memcg_free_cache_params(struct kmem_cache *s)
3226{
3227 kfree(s->memcg_params);
3228}
3229
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003230void memcg_register_cache(struct kmem_cache *s)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003231{
Glauber Costad7f25f82012-12-18 14:22:40 -08003232 struct kmem_cache *root;
3233 struct mem_cgroup *memcg;
3234 int id;
3235
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003236 if (is_root_cache(s))
Glauber Costad7f25f82012-12-18 14:22:40 -08003237 return;
3238
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003239 /*
3240 * Holding the slab_mutex assures nobody will touch the memcg_caches
3241 * array while we are modifying it.
3242 */
3243 lockdep_assert_held(&slab_mutex);
3244
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003245 root = s->memcg_params->root_cache;
3246 memcg = s->memcg_params->memcg;
3247 id = memcg_cache_id(memcg);
3248
3249 css_get(&memcg->css);
3250
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003251
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003252 /*
Vladimir Davydov959c8962014-01-23 15:52:59 -08003253 * Since readers won't lock (see cache_from_memcg_idx()), we need a
3254 * barrier here to ensure nobody will see the kmem_cache partially
3255 * initialized.
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003256 */
Vladimir Davydov959c8962014-01-23 15:52:59 -08003257 smp_wmb();
3258
Vladimir Davydov96403da2014-01-23 15:53:01 -08003259 /*
3260 * Initialize the pointer to this cache in its parent's memcg_params
3261 * before adding it to the memcg_slab_caches list, otherwise we can
3262 * fail to convert memcg_params_to_cache() while traversing the list.
3263 */
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003264 VM_BUG_ON(root->memcg_params->memcg_caches[id]);
Vladimir Davydov959c8962014-01-23 15:52:59 -08003265 root->memcg_params->memcg_caches[id] = s;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003266
3267 mutex_lock(&memcg->slab_caches_mutex);
3268 list_add(&s->memcg_params->list, &memcg->memcg_slab_caches);
3269 mutex_unlock(&memcg->slab_caches_mutex);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003270}
3271
3272void memcg_unregister_cache(struct kmem_cache *s)
3273{
3274 struct kmem_cache *root;
3275 struct mem_cgroup *memcg;
3276 int id;
3277
3278 if (is_root_cache(s))
3279 return;
Glauber Costad7f25f82012-12-18 14:22:40 -08003280
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003281 /*
3282 * Holding the slab_mutex assures nobody will touch the memcg_caches
3283 * array while we are modifying it.
3284 */
3285 lockdep_assert_held(&slab_mutex);
3286
Glauber Costad7f25f82012-12-18 14:22:40 -08003287 root = s->memcg_params->root_cache;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003288 memcg = s->memcg_params->memcg;
3289 id = memcg_cache_id(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08003290
3291 mutex_lock(&memcg->slab_caches_mutex);
3292 list_del(&s->memcg_params->list);
3293 mutex_unlock(&memcg->slab_caches_mutex);
3294
Vladimir Davydov96403da2014-01-23 15:53:01 -08003295 /*
3296 * Clear the pointer to this cache in its parent's memcg_params only
3297 * after removing it from the memcg_slab_caches list, otherwise we can
3298 * fail to convert memcg_params_to_cache() while traversing the list.
3299 */
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003300 VM_BUG_ON(!root->memcg_params->memcg_caches[id]);
Vladimir Davydov96403da2014-01-23 15:53:01 -08003301 root->memcg_params->memcg_caches[id] = NULL;
3302
Li Zefan20f05312013-07-08 16:00:31 -07003303 css_put(&memcg->css);
Glauber Costa2633d7a2012-12-18 14:22:34 -08003304}
3305
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003306/*
3307 * During the creation a new cache, we need to disable our accounting mechanism
3308 * altogether. This is true even if we are not creating, but rather just
3309 * enqueing new caches to be created.
3310 *
3311 * This is because that process will trigger allocations; some visible, like
3312 * explicit kmallocs to auxiliary data structures, name strings and internal
3313 * cache structures; some well concealed, like INIT_WORK() that can allocate
3314 * objects during debug.
3315 *
3316 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3317 * to it. This may not be a bounded recursion: since the first cache creation
3318 * failed to complete (waiting on the allocation), we'll just try to create the
3319 * cache again, failing at the same point.
3320 *
3321 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3322 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3323 * inside the following two functions.
3324 */
3325static inline void memcg_stop_kmem_account(void)
3326{
3327 VM_BUG_ON(!current->mm);
3328 current->memcg_kmem_skip_account++;
3329}
3330
3331static inline void memcg_resume_kmem_account(void)
3332{
3333 VM_BUG_ON(!current->mm);
3334 current->memcg_kmem_skip_account--;
3335}
3336
Glauber Costa1f458cb2012-12-18 14:22:50 -08003337static void kmem_cache_destroy_work_func(struct work_struct *w)
3338{
3339 struct kmem_cache *cachep;
3340 struct memcg_cache_params *p;
3341
3342 p = container_of(w, struct memcg_cache_params, destroy);
3343
3344 cachep = memcg_params_to_cache(p);
3345
Glauber Costa22933152012-12-18 14:22:59 -08003346 /*
3347 * If we get down to 0 after shrink, we could delete right away.
3348 * However, memcg_release_pages() already puts us back in the workqueue
3349 * in that case. If we proceed deleting, we'll get a dangling
3350 * reference, and removing the object from the workqueue in that case
3351 * is unnecessary complication. We are not a fast path.
3352 *
3353 * Note that this case is fundamentally different from racing with
3354 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3355 * kmem_cache_shrink, not only we would be reinserting a dead cache
3356 * into the queue, but doing so from inside the worker racing to
3357 * destroy it.
3358 *
3359 * So if we aren't down to zero, we'll just schedule a worker and try
3360 * again
3361 */
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003362 if (atomic_read(&cachep->memcg_params->nr_pages) != 0)
Glauber Costa22933152012-12-18 14:22:59 -08003363 kmem_cache_shrink(cachep);
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003364 else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003365 kmem_cache_destroy(cachep);
3366}
3367
3368void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3369{
3370 if (!cachep->memcg_params->dead)
3371 return;
3372
3373 /*
Glauber Costa22933152012-12-18 14:22:59 -08003374 * There are many ways in which we can get here.
3375 *
3376 * We can get to a memory-pressure situation while the delayed work is
3377 * still pending to run. The vmscan shrinkers can then release all
3378 * cache memory and get us to destruction. If this is the case, we'll
3379 * be executed twice, which is a bug (the second time will execute over
3380 * bogus data). In this case, cancelling the work should be fine.
3381 *
3382 * But we can also get here from the worker itself, if
3383 * kmem_cache_shrink is enough to shake all the remaining objects and
3384 * get the page count to 0. In this case, we'll deadlock if we try to
3385 * cancel the work (the worker runs with an internal lock held, which
3386 * is the same lock we would hold for cancel_work_sync().)
3387 *
3388 * Since we can't possibly know who got us here, just refrain from
3389 * running if there is already work pending
3390 */
3391 if (work_pending(&cachep->memcg_params->destroy))
3392 return;
3393 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003394 * We have to defer the actual destroying to a workqueue, because
3395 * we might currently be in a context that cannot sleep.
3396 */
3397 schedule_work(&cachep->memcg_params->destroy);
3398}
3399
Vladimir Davydov842e2872014-01-23 15:53:03 -08003400static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3401 struct kmem_cache *s)
Michal Hockod9c10dd2013-03-28 08:48:14 +01003402{
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003403 struct kmem_cache *new = NULL;
Michal Hockod9c10dd2013-03-28 08:48:14 +01003404 static char *tmp_name = NULL;
Vladimir Davydov842e2872014-01-23 15:53:03 -08003405 static DEFINE_MUTEX(mutex); /* protects tmp_name */
Michal Hockod9c10dd2013-03-28 08:48:14 +01003406
Vladimir Davydov842e2872014-01-23 15:53:03 -08003407 BUG_ON(!memcg_can_account_kmem(memcg));
Michal Hockod9c10dd2013-03-28 08:48:14 +01003408
Vladimir Davydov842e2872014-01-23 15:53:03 -08003409 mutex_lock(&mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003410 /*
3411 * kmem_cache_create_memcg duplicates the given name and
3412 * cgroup_name for this name requires RCU context.
3413 * This static temporary buffer is used to prevent from
3414 * pointless shortliving allocation.
3415 */
3416 if (!tmp_name) {
3417 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3418 if (!tmp_name)
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003419 goto out;
Michal Hockod9c10dd2013-03-28 08:48:14 +01003420 }
3421
3422 rcu_read_lock();
3423 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3424 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3425 rcu_read_unlock();
3426
3427 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3428 (s->flags & ~SLAB_PANIC), s->ctor, s);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003429 if (new)
3430 new->allocflags |= __GFP_KMEMCG;
Vladimir Davydov842e2872014-01-23 15:53:03 -08003431 else
3432 new = s;
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003433out:
Vladimir Davydov842e2872014-01-23 15:53:03 -08003434 mutex_unlock(&mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003435 return new;
3436}
3437
Glauber Costa7cf27982012-12-18 14:22:55 -08003438void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3439{
3440 struct kmem_cache *c;
3441 int i;
3442
3443 if (!s->memcg_params)
3444 return;
3445 if (!s->memcg_params->is_root_cache)
3446 return;
3447
3448 /*
3449 * If the cache is being destroyed, we trust that there is no one else
3450 * requesting objects from it. Even if there are, the sanity checks in
3451 * kmem_cache_destroy should caught this ill-case.
3452 *
3453 * Still, we don't want anyone else freeing memcg_caches under our
3454 * noses, which can happen if a new memcg comes to life. As usual,
Vladimir Davydovd6441632014-01-23 15:53:09 -08003455 * we'll take the activate_kmem_mutex to protect ourselves against
3456 * this.
Glauber Costa7cf27982012-12-18 14:22:55 -08003457 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08003458 mutex_lock(&activate_kmem_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003459 for_each_memcg_cache_index(i) {
3460 c = cache_from_memcg_idx(s, i);
Glauber Costa7cf27982012-12-18 14:22:55 -08003461 if (!c)
3462 continue;
3463
3464 /*
3465 * We will now manually delete the caches, so to avoid races
3466 * we need to cancel all pending destruction workers and
3467 * proceed with destruction ourselves.
3468 *
3469 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3470 * and that could spawn the workers again: it is likely that
3471 * the cache still have active pages until this very moment.
3472 * This would lead us back to mem_cgroup_destroy_cache.
3473 *
3474 * But that will not execute at all if the "dead" flag is not
3475 * set, so flip it down to guarantee we are in control.
3476 */
3477 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003478 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003479 kmem_cache_destroy(c);
3480 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08003481 mutex_unlock(&activate_kmem_mutex);
Glauber Costa7cf27982012-12-18 14:22:55 -08003482}
3483
Glauber Costad7f25f82012-12-18 14:22:40 -08003484struct create_work {
3485 struct mem_cgroup *memcg;
3486 struct kmem_cache *cachep;
3487 struct work_struct work;
3488};
3489
Glauber Costa1f458cb2012-12-18 14:22:50 -08003490static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3491{
3492 struct kmem_cache *cachep;
3493 struct memcg_cache_params *params;
3494
3495 if (!memcg_kmem_is_active(memcg))
3496 return;
3497
3498 mutex_lock(&memcg->slab_caches_mutex);
3499 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3500 cachep = memcg_params_to_cache(params);
3501 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003502 schedule_work(&cachep->memcg_params->destroy);
3503 }
3504 mutex_unlock(&memcg->slab_caches_mutex);
3505}
3506
Glauber Costad7f25f82012-12-18 14:22:40 -08003507static void memcg_create_cache_work_func(struct work_struct *w)
3508{
3509 struct create_work *cw;
3510
3511 cw = container_of(w, struct create_work, work);
3512 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003513 css_put(&cw->memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003514 kfree(cw);
3515}
3516
3517/*
3518 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003519 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003520static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3521 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003522{
3523 struct create_work *cw;
3524
3525 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003526 if (cw == NULL) {
3527 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003528 return;
3529 }
3530
3531 cw->memcg = memcg;
3532 cw->cachep = cachep;
3533
3534 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3535 schedule_work(&cw->work);
3536}
3537
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003538static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3539 struct kmem_cache *cachep)
3540{
3541 /*
3542 * We need to stop accounting when we kmalloc, because if the
3543 * corresponding kmalloc cache is not yet created, the first allocation
3544 * in __memcg_create_cache_enqueue will recurse.
3545 *
3546 * However, it is better to enclose the whole function. Depending on
3547 * the debugging options enabled, INIT_WORK(), for instance, can
3548 * trigger an allocation. This too, will make us recurse. Because at
3549 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3550 * the safest choice is to do it like this, wrapping the whole function.
3551 */
3552 memcg_stop_kmem_account();
3553 __memcg_create_cache_enqueue(memcg, cachep);
3554 memcg_resume_kmem_account();
3555}
Glauber Costad7f25f82012-12-18 14:22:40 -08003556/*
3557 * Return the kmem_cache we're supposed to use for a slab allocation.
3558 * We try to use the current memcg's version of the cache.
3559 *
3560 * If the cache does not exist yet, if we are the first user of it,
3561 * we either create it immediately, if possible, or create it asynchronously
3562 * in a workqueue.
3563 * In the latter case, we will let the current allocation go through with
3564 * the original cache.
3565 *
3566 * Can't be called in interrupt context or from kernel threads.
3567 * This function needs to be called with rcu_read_lock() held.
3568 */
3569struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3570 gfp_t gfp)
3571{
3572 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08003573 struct kmem_cache *memcg_cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003574
3575 VM_BUG_ON(!cachep->memcg_params);
3576 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3577
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003578 if (!current->mm || current->memcg_kmem_skip_account)
3579 return cachep;
3580
Glauber Costad7f25f82012-12-18 14:22:40 -08003581 rcu_read_lock();
3582 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003583
3584 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003585 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003586
Vladimir Davydov959c8962014-01-23 15:52:59 -08003587 memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg));
3588 if (likely(memcg_cachep)) {
3589 cachep = memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07003590 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003591 }
3592
Li Zefanca0dde92013-04-29 15:08:57 -07003593 /* The corresponding put will be done in the workqueue. */
3594 if (!css_tryget(&memcg->css))
3595 goto out;
3596 rcu_read_unlock();
3597
3598 /*
3599 * If we are in a safe context (can wait, and not in interrupt
3600 * context), we could be be predictable and return right away.
3601 * This would guarantee that the allocation being performed
3602 * already belongs in the new cache.
3603 *
3604 * However, there are some clashes that can arrive from locking.
3605 * For instance, because we acquire the slab_mutex while doing
3606 * kmem_cache_dup, this means no further allocation could happen
3607 * with the slab_mutex held.
3608 *
3609 * Also, because cache creation issue get_online_cpus(), this
3610 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3611 * that ends up reversed during cpu hotplug. (cpuset allocates
3612 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3613 * better to defer everything.
3614 */
3615 memcg_create_cache_enqueue(memcg, cachep);
3616 return cachep;
3617out:
3618 rcu_read_unlock();
3619 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003620}
3621EXPORT_SYMBOL(__memcg_kmem_get_cache);
3622
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003623/*
3624 * We need to verify if the allocation against current->mm->owner's memcg is
3625 * possible for the given order. But the page is not allocated yet, so we'll
3626 * need a further commit step to do the final arrangements.
3627 *
3628 * It is possible for the task to switch cgroups in this mean time, so at
3629 * commit time, we can't rely on task conversion any longer. We'll then use
3630 * the handle argument to return to the caller which cgroup we should commit
3631 * against. We could also return the memcg directly and avoid the pointer
3632 * passing, but a boolean return value gives better semantics considering
3633 * the compiled-out case as well.
3634 *
3635 * Returning true means the allocation is possible.
3636 */
3637bool
3638__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3639{
3640 struct mem_cgroup *memcg;
3641 int ret;
3642
3643 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003644
3645 /*
3646 * Disabling accounting is only relevant for some specific memcg
3647 * internal allocations. Therefore we would initially not have such
3648 * check here, since direct calls to the page allocator that are marked
3649 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3650 * concerned with cache allocations, and by having this test at
3651 * memcg_kmem_get_cache, we are already able to relay the allocation to
3652 * the root cache and bypass the memcg cache altogether.
3653 *
3654 * There is one exception, though: the SLUB allocator does not create
3655 * large order caches, but rather service large kmallocs directly from
3656 * the page allocator. Therefore, the following sequence when backed by
3657 * the SLUB allocator:
3658 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003659 * memcg_stop_kmem_account();
3660 * kmalloc(<large_number>)
3661 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003662 *
3663 * would effectively ignore the fact that we should skip accounting,
3664 * since it will drive us directly to this function without passing
3665 * through the cache selector memcg_kmem_get_cache. Such large
3666 * allocations are extremely rare but can happen, for instance, for the
3667 * cache arrays. We bring this test here.
3668 */
3669 if (!current->mm || current->memcg_kmem_skip_account)
3670 return true;
3671
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003672 memcg = try_get_mem_cgroup_from_mm(current->mm);
3673
3674 /*
3675 * very rare case described in mem_cgroup_from_task. Unfortunately there
3676 * isn't much we can do without complicating this too much, and it would
3677 * be gfp-dependent anyway. Just let it go
3678 */
3679 if (unlikely(!memcg))
3680 return true;
3681
3682 if (!memcg_can_account_kmem(memcg)) {
3683 css_put(&memcg->css);
3684 return true;
3685 }
3686
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003687 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3688 if (!ret)
3689 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003690
3691 css_put(&memcg->css);
3692 return (ret == 0);
3693}
3694
3695void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3696 int order)
3697{
3698 struct page_cgroup *pc;
3699
3700 VM_BUG_ON(mem_cgroup_is_root(memcg));
3701
3702 /* The page allocation failed. Revert */
3703 if (!page) {
3704 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003705 return;
3706 }
3707
3708 pc = lookup_page_cgroup(page);
3709 lock_page_cgroup(pc);
3710 pc->mem_cgroup = memcg;
3711 SetPageCgroupUsed(pc);
3712 unlock_page_cgroup(pc);
3713}
3714
3715void __memcg_kmem_uncharge_pages(struct page *page, int order)
3716{
3717 struct mem_cgroup *memcg = NULL;
3718 struct page_cgroup *pc;
3719
3720
3721 pc = lookup_page_cgroup(page);
3722 /*
3723 * Fast unlocked return. Theoretically might have changed, have to
3724 * check again after locking.
3725 */
3726 if (!PageCgroupUsed(pc))
3727 return;
3728
3729 lock_page_cgroup(pc);
3730 if (PageCgroupUsed(pc)) {
3731 memcg = pc->mem_cgroup;
3732 ClearPageCgroupUsed(pc);
3733 }
3734 unlock_page_cgroup(pc);
3735
3736 /*
3737 * We trust that only if there is a memcg associated with the page, it
3738 * is a valid allocation
3739 */
3740 if (!memcg)
3741 return;
3742
Sasha Levin309381fea2014-01-23 15:52:54 -08003743 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003744 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003745}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003746#else
3747static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3748{
3749}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003750#endif /* CONFIG_MEMCG_KMEM */
3751
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003752#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3753
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003754#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003755/*
3756 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003757 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3758 * charge/uncharge will be never happen and move_account() is done under
3759 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003760 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003761void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003762{
3763 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003764 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003765 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003766 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003767
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003768 if (mem_cgroup_disabled())
3769 return;
David Rientjesb070e652013-05-07 16:18:09 -07003770
3771 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003772 for (i = 1; i < HPAGE_PMD_NR; i++) {
3773 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003774 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003775 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003776 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3777 }
David Rientjesb070e652013-05-07 16:18:09 -07003778 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3779 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003780}
Hugh Dickins12d27102012-01-12 17:19:52 -08003781#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003782
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003783static inline
3784void mem_cgroup_move_account_page_stat(struct mem_cgroup *from,
3785 struct mem_cgroup *to,
3786 unsigned int nr_pages,
3787 enum mem_cgroup_stat_index idx)
3788{
3789 /* Update stat data for mem_cgroup */
3790 preempt_disable();
Greg Thelen5e8cfc32013-10-30 13:56:21 -07003791 __this_cpu_sub(from->stat->count[idx], nr_pages);
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003792 __this_cpu_add(to->stat->count[idx], nr_pages);
3793 preempt_enable();
3794}
3795
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003796/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003797 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003798 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003799 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003800 * @pc: page_cgroup of the page.
3801 * @from: mem_cgroup which the page is moved from.
3802 * @to: mem_cgroup which the page is moved to. @from != @to.
3803 *
3804 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003805 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003806 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003807 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003808 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3809 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003810 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003811static int mem_cgroup_move_account(struct page *page,
3812 unsigned int nr_pages,
3813 struct page_cgroup *pc,
3814 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003815 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003816{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003817 unsigned long flags;
3818 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003819 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003820
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003821 VM_BUG_ON(from == to);
Sasha Levin309381fea2014-01-23 15:52:54 -08003822 VM_BUG_ON_PAGE(PageLRU(page), page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003823 /*
3824 * The page is isolated from LRU. So, collapse function
3825 * will not handle this page. But page splitting can happen.
3826 * Do this check under compound_page_lock(). The caller should
3827 * hold it.
3828 */
3829 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003830 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003831 goto out;
3832
3833 lock_page_cgroup(pc);
3834
3835 ret = -EINVAL;
3836 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3837 goto unlock;
3838
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003839 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003840
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003841 if (!anon && page_mapped(page))
3842 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3843 MEM_CGROUP_STAT_FILE_MAPPED);
3844
3845 if (PageWriteback(page))
3846 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3847 MEM_CGROUP_STAT_WRITEBACK);
3848
David Rientjesb070e652013-05-07 16:18:09 -07003849 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003850
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003851 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003852 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003853 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003854 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003855 ret = 0;
3856unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003857 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003858 /*
3859 * check events
3860 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003861 memcg_check_events(to, page);
3862 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003863out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003864 return ret;
3865}
3866
Michal Hocko2ef37d32012-10-26 13:37:30 +02003867/**
3868 * mem_cgroup_move_parent - moves page to the parent group
3869 * @page: the page to move
3870 * @pc: page_cgroup of the page
3871 * @child: page's cgroup
3872 *
3873 * move charges to its parent or the root cgroup if the group has no
3874 * parent (aka use_hierarchy==0).
3875 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3876 * mem_cgroup_move_account fails) the failure is always temporary and
3877 * it signals a race with a page removal/uncharge or migration. In the
3878 * first case the page is on the way out and it will vanish from the LRU
3879 * on the next attempt and the call should be retried later.
3880 * Isolation from the LRU fails only if page has been isolated from
3881 * the LRU since we looked at it and that usually means either global
3882 * reclaim or migration going on. The page will either get back to the
3883 * LRU or vanish.
3884 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3885 * (!PageCgroupUsed) or moved to a different group. The page will
3886 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003887 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003888static int mem_cgroup_move_parent(struct page *page,
3889 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003890 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003891{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003892 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003893 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003894 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003895 int ret;
3896
Michal Hockod8423012012-10-26 13:37:29 +02003897 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003898
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003899 ret = -EBUSY;
3900 if (!get_page_unless_zero(page))
3901 goto out;
3902 if (isolate_lru_page(page))
3903 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003904
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003905 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003906
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003907 parent = parent_mem_cgroup(child);
3908 /*
3909 * If no parent, move charges to root cgroup.
3910 */
3911 if (!parent)
3912 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003913
Michal Hocko2ef37d32012-10-26 13:37:30 +02003914 if (nr_pages > 1) {
Sasha Levin309381fea2014-01-23 15:52:54 -08003915 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003916 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003917 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003918
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003919 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003920 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003921 if (!ret)
3922 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003923
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003924 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003925 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003926 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003927put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003928 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003929out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003930 return ret;
3931}
3932
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003933/*
3934 * Charge the memory controller for page usage.
3935 * Return
3936 * 0 if the charge was successful
3937 * < 0 if the cgroup is over its limit
3938 */
3939static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003940 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003941{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003942 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003943 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003944 bool oom = true;
3945 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003946
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003947 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003948 nr_pages <<= compound_order(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08003949 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Johannes Weiner8493ae42011-02-01 15:52:44 -08003950 /*
3951 * Never OOM-kill a process for a huge page. The
3952 * fault handler will fall back to regular pages.
3953 */
3954 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003955 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003956
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003957 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003958 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003959 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003960 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003961 return 0;
3962}
3963
3964int mem_cgroup_newpage_charge(struct page *page,
3965 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003966{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003967 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003968 return 0;
Sasha Levin309381fea2014-01-23 15:52:54 -08003969 VM_BUG_ON_PAGE(page_mapped(page), page);
3970 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page);
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003971 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003972 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003973 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003974}
3975
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003976/*
3977 * While swap-in, try_charge -> commit or cancel, the page is locked.
3978 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003979 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003980 * "commit()" or removed by "cancel()"
3981 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003982static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3983 struct page *page,
3984 gfp_t mask,
3985 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003986{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003987 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003988 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003989 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003990
Johannes Weiner90deb782012-07-31 16:45:47 -07003991 pc = lookup_page_cgroup(page);
3992 /*
3993 * Every swap fault against a single page tries to charge the
3994 * page, bail as early as possible. shmem_unuse() encounters
3995 * already charged pages, too. The USED bit is protected by
3996 * the page lock, which serializes swap cache removal, which
3997 * in turn serializes uncharging.
3998 */
3999 if (PageCgroupUsed(pc))
4000 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004001 if (!do_swap_account)
4002 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004003 memcg = try_get_mem_cgroup_from_page(page);
4004 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08004005 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08004006 *memcgp = memcg;
4007 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004008 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004009 if (ret == -EINTR)
4010 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08004011 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004012charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004013 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
4014 if (ret == -EINTR)
4015 ret = 0;
4016 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004017}
4018
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004019int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
4020 gfp_t gfp_mask, struct mem_cgroup **memcgp)
4021{
4022 *memcgp = NULL;
4023 if (mem_cgroup_disabled())
4024 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07004025 /*
4026 * A racing thread's fault, or swapoff, may have already
4027 * updated the pte, and even removed page from swap cache: in
4028 * those cases unuse_pte()'s pte_same() test will fail; but
4029 * there's also a KSM case which does need to charge the page.
4030 */
4031 if (!PageSwapCache(page)) {
4032 int ret;
4033
4034 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
4035 if (ret == -EINTR)
4036 ret = 0;
4037 return ret;
4038 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004039 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
4040}
4041
Johannes Weiner827a03d2012-07-31 16:45:36 -07004042void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
4043{
4044 if (mem_cgroup_disabled())
4045 return;
4046 if (!memcg)
4047 return;
4048 __mem_cgroup_cancel_charge(memcg, 1);
4049}
4050
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004051static void
Johannes Weiner72835c82012-01-12 17:18:32 -08004052__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004053 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004054{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004055 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004056 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08004057 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004058 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07004059
Johannes Weinerce587e62012-04-24 20:22:33 +02004060 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004061 /*
4062 * Now swap is on-memory. This means this page may be
4063 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004064 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
4065 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
4066 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004067 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004068 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004069 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004070 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004071 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004072}
4073
Johannes Weiner72835c82012-01-12 17:18:32 -08004074void mem_cgroup_commit_charge_swapin(struct page *page,
4075 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004076{
Johannes Weiner72835c82012-01-12 17:18:32 -08004077 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004078 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004079}
4080
Johannes Weiner827a03d2012-07-31 16:45:36 -07004081int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4082 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004083{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004084 struct mem_cgroup *memcg = NULL;
4085 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4086 int ret;
4087
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004088 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004089 return 0;
4090 if (PageCompound(page))
4091 return 0;
4092
Johannes Weiner827a03d2012-07-31 16:45:36 -07004093 if (!PageSwapCache(page))
4094 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4095 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004096 ret = __mem_cgroup_try_charge_swapin(mm, page,
4097 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004098 if (!ret)
4099 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4100 }
4101 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004102}
4103
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004104static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004105 unsigned int nr_pages,
4106 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004107{
4108 struct memcg_batch_info *batch = NULL;
4109 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004110
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004111 /* If swapout, usage of swap doesn't decrease */
4112 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4113 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004114
4115 batch = &current->memcg_batch;
4116 /*
4117 * In usual, we do css_get() when we remember memcg pointer.
4118 * But in this case, we keep res->usage until end of a series of
4119 * uncharges. Then, it's ok to ignore memcg's refcnt.
4120 */
4121 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004122 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004123 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004124 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004125 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004126 * the same cgroup and we have chance to coalesce uncharges.
4127 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4128 * because we want to do uncharge as soon as possible.
4129 */
4130
4131 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4132 goto direct_uncharge;
4133
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004134 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004135 goto direct_uncharge;
4136
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004137 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004138 * In typical case, batch->memcg == mem. This means we can
4139 * merge a series of uncharges to an uncharge of res_counter.
4140 * If not, we uncharge res_counter ony by one.
4141 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004142 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004143 goto direct_uncharge;
4144 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004145 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004146 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004147 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004148 return;
4149direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004150 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004151 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004152 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4153 if (unlikely(batch->memcg != memcg))
4154 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004155}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004156
Balbir Singh8697d332008-02-07 00:13:59 -08004157/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004158 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004159 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004160static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004161__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4162 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004163{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004164 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004165 unsigned int nr_pages = 1;
4166 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004167 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004168
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004169 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004170 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004171
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004172 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004173 nr_pages <<= compound_order(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08004174 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004175 }
Balbir Singh8697d332008-02-07 00:13:59 -08004176 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004177 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004178 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004179 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004180 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004181 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004182
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004183 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004184
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004185 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004186
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004187 if (!PageCgroupUsed(pc))
4188 goto unlock_out;
4189
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004190 anon = PageAnon(page);
4191
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004192 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004193 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004194 /*
4195 * Generally PageAnon tells if it's the anon statistics to be
4196 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4197 * used before page reached the stage of being marked PageAnon.
4198 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004199 anon = true;
4200 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004201 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004202 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004203 if (page_mapped(page))
4204 goto unlock_out;
4205 /*
4206 * Pages under migration may not be uncharged. But
4207 * end_migration() /must/ be the one uncharging the
4208 * unused post-migration page and so it has to call
4209 * here with the migration bit still set. See the
4210 * res_counter handling below.
4211 */
4212 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004213 goto unlock_out;
4214 break;
4215 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4216 if (!PageAnon(page)) { /* Shared memory */
4217 if (page->mapping && !page_is_file_cache(page))
4218 goto unlock_out;
4219 } else if (page_mapped(page)) /* Anon */
4220 goto unlock_out;
4221 break;
4222 default:
4223 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004224 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004225
David Rientjesb070e652013-05-07 16:18:09 -07004226 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004227
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004228 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004229 /*
4230 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4231 * freed from LRU. This is safe because uncharged page is expected not
4232 * to be reused (freed soon). Exception is SwapCache, it's handled by
4233 * special functions.
4234 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004235
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004236 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004237 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004238 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004239 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004240 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004241 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004242 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004243 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004244 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004245 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004246 /*
4247 * Migration does not charge the res_counter for the
4248 * replacement page, so leave it alone when phasing out the
4249 * page that is unused after the migration.
4250 */
4251 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004252 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004253
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004254 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004255
4256unlock_out:
4257 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004258 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004259}
4260
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004261void mem_cgroup_uncharge_page(struct page *page)
4262{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004263 /* early check. */
4264 if (page_mapped(page))
4265 return;
Sasha Levin309381fea2014-01-23 15:52:54 -08004266 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page);
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004267 /*
4268 * If the page is in swap cache, uncharge should be deferred
4269 * to the swap path, which also properly accounts swap usage
4270 * and handles memcg lifetime.
4271 *
4272 * Note that this check is not stable and reclaim may add the
4273 * page to swap cache at any time after this. However, if the
4274 * page is not in swap cache by the time page->mapcount hits
4275 * 0, there won't be any page table references to the swap
4276 * slot, and reclaim will free it and not actually write the
4277 * page to disk.
4278 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004279 if (PageSwapCache(page))
4280 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004281 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004282}
4283
4284void mem_cgroup_uncharge_cache_page(struct page *page)
4285{
Sasha Levin309381fea2014-01-23 15:52:54 -08004286 VM_BUG_ON_PAGE(page_mapped(page), page);
4287 VM_BUG_ON_PAGE(page->mapping, page);
Johannes Weiner0030f532012-07-31 16:45:25 -07004288 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004289}
4290
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004291/*
4292 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4293 * In that cases, pages are freed continuously and we can expect pages
4294 * are in the same memcg. All these calls itself limits the number of
4295 * pages freed at once, then uncharge_start/end() is called properly.
4296 * This may be called prural(2) times in a context,
4297 */
4298
4299void mem_cgroup_uncharge_start(void)
4300{
4301 current->memcg_batch.do_batch++;
4302 /* We can do nest. */
4303 if (current->memcg_batch.do_batch == 1) {
4304 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004305 current->memcg_batch.nr_pages = 0;
4306 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004307 }
4308}
4309
4310void mem_cgroup_uncharge_end(void)
4311{
4312 struct memcg_batch_info *batch = &current->memcg_batch;
4313
4314 if (!batch->do_batch)
4315 return;
4316
4317 batch->do_batch--;
4318 if (batch->do_batch) /* If stacked, do nothing. */
4319 return;
4320
4321 if (!batch->memcg)
4322 return;
4323 /*
4324 * This "batch->memcg" is valid without any css_get/put etc...
4325 * bacause we hide charges behind us.
4326 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004327 if (batch->nr_pages)
4328 res_counter_uncharge(&batch->memcg->res,
4329 batch->nr_pages * PAGE_SIZE);
4330 if (batch->memsw_nr_pages)
4331 res_counter_uncharge(&batch->memcg->memsw,
4332 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004333 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004334 /* forget this pointer (for sanity check) */
4335 batch->memcg = NULL;
4336}
4337
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004338#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004339/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004340 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004341 * memcg information is recorded to swap_cgroup of "ent"
4342 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004343void
4344mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004345{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004346 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004347 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004348
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004349 if (!swapout) /* this was a swap cache but the swap is unused ! */
4350 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4351
Johannes Weiner0030f532012-07-31 16:45:25 -07004352 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004353
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004354 /*
4355 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004356 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004357 */
4358 if (do_swap_account && swapout && memcg)
Li Zefan34c00c32013-09-23 16:56:01 +08004359 swap_cgroup_record(ent, mem_cgroup_id(memcg));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004360}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004361#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004362
Andrew Mortonc255a452012-07-31 16:43:02 -07004363#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004364/*
4365 * called from swap_entry_free(). remove record in swap_cgroup and
4366 * uncharge "memsw" account.
4367 */
4368void mem_cgroup_uncharge_swap(swp_entry_t ent)
4369{
4370 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004371 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004372
4373 if (!do_swap_account)
4374 return;
4375
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004376 id = swap_cgroup_record(ent, 0);
4377 rcu_read_lock();
4378 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004379 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004380 /*
4381 * We uncharge this because swap is freed.
4382 * This memcg can be obsolete one. We avoid calling css_tryget
4383 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004384 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004385 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004386 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004387 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004388 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004389 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004390}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004391
4392/**
4393 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4394 * @entry: swap entry to be moved
4395 * @from: mem_cgroup which the entry is moved from
4396 * @to: mem_cgroup which the entry is moved to
4397 *
4398 * It succeeds only when the swap_cgroup's record for this entry is the same
4399 * as the mem_cgroup's id of @from.
4400 *
4401 * Returns 0 on success, -EINVAL on failure.
4402 *
4403 * The caller must have charged to @to, IOW, called res_counter_charge() about
4404 * both res and memsw, and called css_get().
4405 */
4406static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004407 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004408{
4409 unsigned short old_id, new_id;
4410
Li Zefan34c00c32013-09-23 16:56:01 +08004411 old_id = mem_cgroup_id(from);
4412 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004413
4414 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004415 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004416 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004417 /*
4418 * This function is only called from task migration context now.
4419 * It postpones res_counter and refcount handling till the end
4420 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004421 * improvement. But we cannot postpone css_get(to) because if
4422 * the process that has been moved to @to does swap-in, the
4423 * refcount of @to might be decreased to 0.
4424 *
4425 * We are in attach() phase, so the cgroup is guaranteed to be
4426 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004427 */
Li Zefan40503772013-07-08 16:00:34 -07004428 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004429 return 0;
4430 }
4431 return -EINVAL;
4432}
4433#else
4434static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004435 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004436{
4437 return -EINVAL;
4438}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004439#endif
4440
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004441/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004442 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4443 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004444 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004445void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4446 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004447{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004448 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004449 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004450 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004451 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004452
Johannes Weiner72835c82012-01-12 17:18:32 -08004453 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004454
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004455 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004456 return;
Balbir Singh40779602008-04-04 14:29:59 -07004457
Mel Gormanb32967f2012-11-19 12:35:47 +00004458 if (PageTransHuge(page))
4459 nr_pages <<= compound_order(page);
4460
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004461 pc = lookup_page_cgroup(page);
4462 lock_page_cgroup(pc);
4463 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004464 memcg = pc->mem_cgroup;
4465 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004466 /*
4467 * At migrating an anonymous page, its mapcount goes down
4468 * to 0 and uncharge() will be called. But, even if it's fully
4469 * unmapped, migration may fail and this page has to be
4470 * charged again. We set MIGRATION flag here and delay uncharge
4471 * until end_migration() is called
4472 *
4473 * Corner Case Thinking
4474 * A)
4475 * When the old page was mapped as Anon and it's unmap-and-freed
4476 * while migration was ongoing.
4477 * If unmap finds the old page, uncharge() of it will be delayed
4478 * until end_migration(). If unmap finds a new page, it's
4479 * uncharged when it make mapcount to be 1->0. If unmap code
4480 * finds swap_migration_entry, the new page will not be mapped
4481 * and end_migration() will find it(mapcount==0).
4482 *
4483 * B)
4484 * When the old page was mapped but migraion fails, the kernel
4485 * remaps it. A charge for it is kept by MIGRATION flag even
4486 * if mapcount goes down to 0. We can do remap successfully
4487 * without charging it again.
4488 *
4489 * C)
4490 * The "old" page is under lock_page() until the end of
4491 * migration, so, the old page itself will not be swapped-out.
4492 * If the new page is swapped out before end_migraton, our
4493 * hook to usual swap-out path will catch the event.
4494 */
4495 if (PageAnon(page))
4496 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004497 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004498 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004499 /*
4500 * If the page is not charged at this point,
4501 * we return here.
4502 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004503 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004504 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004505
Johannes Weiner72835c82012-01-12 17:18:32 -08004506 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004507 /*
4508 * We charge new page before it's used/mapped. So, even if unlock_page()
4509 * is called before end_migration, we can catch all events on this new
4510 * page. In the case new page is migrated but not remapped, new page's
4511 * mapcount will be finally 0 and we call uncharge in end_migration().
4512 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004513 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004514 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004515 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004516 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004517 /*
4518 * The page is committed to the memcg, but it's not actually
4519 * charged to the res_counter since we plan on replacing the
4520 * old one and only one page is going to be left afterwards.
4521 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004522 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004523}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004524
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004525/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004526void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004527 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004528{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004529 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004530 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004531 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004532
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004533 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004534 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004535
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004536 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004537 used = oldpage;
4538 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004539 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004540 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004541 unused = oldpage;
4542 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004543 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004544 __mem_cgroup_uncharge_common(unused,
4545 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4546 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4547 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004548 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004549 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004550 * We disallowed uncharge of pages under migration because mapcount
4551 * of the page goes down to zero, temporarly.
4552 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004553 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004554 pc = lookup_page_cgroup(oldpage);
4555 lock_page_cgroup(pc);
4556 ClearPageCgroupMigration(pc);
4557 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004558
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004559 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004560 * If a page is a file cache, radix-tree replacement is very atomic
4561 * and we can skip this check. When it was an Anon page, its mapcount
4562 * goes down to 0. But because we added MIGRATION flage, it's not
4563 * uncharged yet. There are several case but page->mapcount check
4564 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4565 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004566 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004567 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004568 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004569}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004570
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004571/*
4572 * At replace page cache, newpage is not under any memcg but it's on
4573 * LRU. So, this function doesn't touch res_counter but handles LRU
4574 * in correct way. Both pages are locked so we cannot race with uncharge.
4575 */
4576void mem_cgroup_replace_page_cache(struct page *oldpage,
4577 struct page *newpage)
4578{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004579 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004580 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004581 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004582
4583 if (mem_cgroup_disabled())
4584 return;
4585
4586 pc = lookup_page_cgroup(oldpage);
4587 /* fix accounting on old pages */
4588 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004589 if (PageCgroupUsed(pc)) {
4590 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004591 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004592 ClearPageCgroupUsed(pc);
4593 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004594 unlock_page_cgroup(pc);
4595
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004596 /*
4597 * When called from shmem_replace_page(), in some cases the
4598 * oldpage has already been charged, and in some cases not.
4599 */
4600 if (!memcg)
4601 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004602 /*
4603 * Even if newpage->mapping was NULL before starting replacement,
4604 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4605 * LRU while we overwrite pc->mem_cgroup.
4606 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004607 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004608}
4609
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004610#ifdef CONFIG_DEBUG_VM
4611static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4612{
4613 struct page_cgroup *pc;
4614
4615 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004616 /*
4617 * Can be NULL while feeding pages into the page allocator for
4618 * the first time, i.e. during boot or memory hotplug;
4619 * or when mem_cgroup_disabled().
4620 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004621 if (likely(pc) && PageCgroupUsed(pc))
4622 return pc;
4623 return NULL;
4624}
4625
4626bool mem_cgroup_bad_page_check(struct page *page)
4627{
4628 if (mem_cgroup_disabled())
4629 return false;
4630
4631 return lookup_page_cgroup_used(page) != NULL;
4632}
4633
4634void mem_cgroup_print_bad_page(struct page *page)
4635{
4636 struct page_cgroup *pc;
4637
4638 pc = lookup_page_cgroup_used(page);
4639 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004640 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4641 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004642 }
4643}
4644#endif
4645
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004646static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004647 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004648{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004649 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004650 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004651 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004652 int children = mem_cgroup_count_children(memcg);
4653 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004654 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004655
4656 /*
4657 * For keeping hierarchical_reclaim simple, how long we should retry
4658 * is depends on callers. We set our retry-count to be function
4659 * of # of children which we should visit in this loop.
4660 */
4661 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4662
4663 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004664
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004665 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004666 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004667 if (signal_pending(current)) {
4668 ret = -EINTR;
4669 break;
4670 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004671 /*
4672 * Rather than hide all in some function, I do this in
4673 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004674 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004675 */
4676 mutex_lock(&set_limit_mutex);
4677 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4678 if (memswlimit < val) {
4679 ret = -EINVAL;
4680 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004681 break;
4682 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004683
4684 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4685 if (memlimit < val)
4686 enlarge = 1;
4687
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004688 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004689 if (!ret) {
4690 if (memswlimit == val)
4691 memcg->memsw_is_minimum = true;
4692 else
4693 memcg->memsw_is_minimum = false;
4694 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004695 mutex_unlock(&set_limit_mutex);
4696
4697 if (!ret)
4698 break;
4699
Johannes Weiner56600482012-01-12 17:17:59 -08004700 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4701 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004702 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4703 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004704 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004705 retry_count--;
4706 else
4707 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004708 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004709 if (!ret && enlarge)
4710 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004711
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004712 return ret;
4713}
4714
Li Zefan338c8432009-06-17 16:27:15 -07004715static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4716 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004717{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004718 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004719 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004720 int children = mem_cgroup_count_children(memcg);
4721 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004722 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004723
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004724 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004725 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004726 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004727 while (retry_count) {
4728 if (signal_pending(current)) {
4729 ret = -EINTR;
4730 break;
4731 }
4732 /*
4733 * Rather than hide all in some function, I do this in
4734 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004735 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004736 */
4737 mutex_lock(&set_limit_mutex);
4738 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4739 if (memlimit > val) {
4740 ret = -EINVAL;
4741 mutex_unlock(&set_limit_mutex);
4742 break;
4743 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004744 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4745 if (memswlimit < val)
4746 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004747 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004748 if (!ret) {
4749 if (memlimit == val)
4750 memcg->memsw_is_minimum = true;
4751 else
4752 memcg->memsw_is_minimum = false;
4753 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004754 mutex_unlock(&set_limit_mutex);
4755
4756 if (!ret)
4757 break;
4758
Johannes Weiner56600482012-01-12 17:17:59 -08004759 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4760 MEM_CGROUP_RECLAIM_NOSWAP |
4761 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004762 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004763 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004764 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004765 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004766 else
4767 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004768 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004769 if (!ret && enlarge)
4770 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004771 return ret;
4772}
4773
Andrew Morton0608f432013-09-24 15:27:41 -07004774unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
4775 gfp_t gfp_mask,
4776 unsigned long *total_scanned)
4777{
4778 unsigned long nr_reclaimed = 0;
4779 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4780 unsigned long reclaimed;
4781 int loop = 0;
4782 struct mem_cgroup_tree_per_zone *mctz;
4783 unsigned long long excess;
4784 unsigned long nr_scanned;
4785
4786 if (order > 0)
4787 return 0;
4788
4789 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
4790 /*
4791 * This loop can run a while, specially if mem_cgroup's continuously
4792 * keep exceeding their soft limit and putting the system under
4793 * pressure
4794 */
4795 do {
4796 if (next_mz)
4797 mz = next_mz;
4798 else
4799 mz = mem_cgroup_largest_soft_limit_node(mctz);
4800 if (!mz)
4801 break;
4802
4803 nr_scanned = 0;
4804 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
4805 gfp_mask, &nr_scanned);
4806 nr_reclaimed += reclaimed;
4807 *total_scanned += nr_scanned;
4808 spin_lock(&mctz->lock);
4809
4810 /*
4811 * If we failed to reclaim anything from this memory cgroup
4812 * it is time to move on to the next cgroup
4813 */
4814 next_mz = NULL;
4815 if (!reclaimed) {
4816 do {
4817 /*
4818 * Loop until we find yet another one.
4819 *
4820 * By the time we get the soft_limit lock
4821 * again, someone might have aded the
4822 * group back on the RB tree. Iterate to
4823 * make sure we get a different mem.
4824 * mem_cgroup_largest_soft_limit_node returns
4825 * NULL if no other cgroup is present on
4826 * the tree
4827 */
4828 next_mz =
4829 __mem_cgroup_largest_soft_limit_node(mctz);
4830 if (next_mz == mz)
4831 css_put(&next_mz->memcg->css);
4832 else /* next_mz == NULL or other memcg */
4833 break;
4834 } while (1);
4835 }
4836 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4837 excess = res_counter_soft_limit_excess(&mz->memcg->res);
4838 /*
4839 * One school of thought says that we should not add
4840 * back the node to the tree if reclaim returns 0.
4841 * But our reclaim could return 0, simply because due
4842 * to priority we are exposing a smaller subset of
4843 * memory to reclaim from. Consider this as a longer
4844 * term TODO.
4845 */
4846 /* If excess == 0, no tree ops */
4847 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
4848 spin_unlock(&mctz->lock);
4849 css_put(&mz->memcg->css);
4850 loop++;
4851 /*
4852 * Could not reclaim anything and there are no more
4853 * mem cgroups to try or we seem to be looping without
4854 * reclaiming anything.
4855 */
4856 if (!nr_reclaimed &&
4857 (next_mz == NULL ||
4858 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4859 break;
4860 } while (!nr_reclaimed);
4861 if (next_mz)
4862 css_put(&next_mz->memcg->css);
4863 return nr_reclaimed;
4864}
4865
Michal Hocko2ef37d32012-10-26 13:37:30 +02004866/**
4867 * mem_cgroup_force_empty_list - clears LRU of a group
4868 * @memcg: group to clear
4869 * @node: NUMA node
4870 * @zid: zone id
4871 * @lru: lru to to clear
4872 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004873 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004874 * reclaim the pages page themselves - pages are moved to the parent (or root)
4875 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004876 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004877static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004878 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004879{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004880 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004881 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004882 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004883 struct page *busy;
4884 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004885
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004886 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004887 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4888 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004889
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004890 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004891 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004892 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004893 struct page *page;
4894
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004895 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004896 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004897 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004898 break;
4899 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004900 page = list_entry(list->prev, struct page, lru);
4901 if (busy == page) {
4902 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004903 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004904 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004905 continue;
4906 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004907 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004908
Johannes Weiner925b7672012-01-12 17:18:15 -08004909 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004910
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004911 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004912 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004913 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004914 cond_resched();
4915 } else
4916 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004917 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004918}
4919
4920/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004921 * make mem_cgroup's charge to be 0 if there is no task by moving
4922 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004923 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004924 *
4925 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004926 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004927static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004928{
Michal Hockoc26251f2012-10-26 13:37:28 +02004929 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004930 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004931
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004932 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004933 /* This is for making all *used* pages to be on LRU. */
4934 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004935 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004936 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004937 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004938 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004939 enum lru_list lru;
4940 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004941 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004942 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004943 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004944 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004945 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004946 mem_cgroup_end_move(memcg);
4947 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004948 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004949
Michal Hocko2ef37d32012-10-26 13:37:30 +02004950 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004951 * Kernel memory may not necessarily be trackable to a specific
4952 * process. So they are not migrated, and therefore we can't
4953 * expect their value to drop to 0 here.
4954 * Having res filled up with kmem only is enough.
4955 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004956 * This is a safety check because mem_cgroup_force_empty_list
4957 * could have raced with mem_cgroup_replace_page_cache callers
4958 * so the lru seemed empty but the page could have been added
4959 * right after the check. RES_USAGE should be safe as we always
4960 * charge before adding to the LRU.
4961 */
Glauber Costabea207c2012-12-18 14:22:11 -08004962 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4963 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4964 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004965}
4966
Glauber Costab5f99b52013-02-22 16:34:53 -08004967static inline bool memcg_has_children(struct mem_cgroup *memcg)
4968{
Johannes Weiner696ac172013-10-31 16:34:15 -07004969 lockdep_assert_held(&memcg_create_mutex);
4970 /*
4971 * The lock does not prevent addition or deletion to the list
4972 * of children, but it prevents a new child from being
4973 * initialized based on this parent in css_online(), so it's
4974 * enough to decide whether hierarchically inherited
4975 * attributes can still be changed or not.
4976 */
4977 return memcg->use_hierarchy &&
4978 !list_empty(&memcg->css.cgroup->children);
Glauber Costab5f99b52013-02-22 16:34:53 -08004979}
4980
4981/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004982 * Reclaims as many pages from the given memcg as possible and moves
4983 * the rest to the parent.
4984 *
4985 * Caller is responsible for holding css reference for memcg.
4986 */
4987static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4988{
4989 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4990 struct cgroup *cgrp = memcg->css.cgroup;
4991
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004992 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004993 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4994 return -EBUSY;
4995
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004996 /* we call try-to-free pages for make this cgroup empty */
4997 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004998 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004999 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005000 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005001
Michal Hockoc26251f2012-10-26 13:37:28 +02005002 if (signal_pending(current))
5003 return -EINTR;
5004
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005005 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07005006 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005007 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005008 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005009 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02005010 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005011 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005012
5013 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08005014 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02005015 mem_cgroup_reparent_charges(memcg);
5016
5017 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08005018}
5019
Tejun Heo182446d2013-08-08 20:11:24 -04005020static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
5021 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005022{
Tejun Heo182446d2013-08-08 20:11:24 -04005023 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02005024
Michal Hockod8423012012-10-26 13:37:29 +02005025 if (mem_cgroup_is_root(memcg))
5026 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07005027 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005028}
5029
Tejun Heo182446d2013-08-08 20:11:24 -04005030static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
5031 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005032{
Tejun Heo182446d2013-08-08 20:11:24 -04005033 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005034}
5035
Tejun Heo182446d2013-08-08 20:11:24 -04005036static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
5037 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005038{
5039 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04005040 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005041 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08005042
Glauber Costa09998212013-02-22 16:34:55 -08005043 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07005044
5045 if (memcg->use_hierarchy == val)
5046 goto out;
5047
Balbir Singh18f59ea2009-01-07 18:08:07 -08005048 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005049 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08005050 * in the child subtrees. If it is unset, then the change can
5051 * occur, provided the current cgroup has no children.
5052 *
5053 * For the root cgroup, parent_mem is NULL, we allow value to be
5054 * set if there are no children.
5055 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005056 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08005057 (val == 1 || val == 0)) {
Johannes Weiner696ac172013-10-31 16:34:15 -07005058 if (list_empty(&memcg->css.cgroup->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005059 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005060 else
5061 retval = -EBUSY;
5062 } else
5063 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005064
5065out:
Glauber Costa09998212013-02-22 16:34:55 -08005066 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005067
5068 return retval;
5069}
5070
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005071
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005073 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005074{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005075 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005076 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005077
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005078 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005079 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005080 val += mem_cgroup_read_stat(iter, idx);
5081
5082 if (val < 0) /* race ? */
5083 val = 0;
5084 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005085}
5086
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005087static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005088{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005089 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005090
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005091 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005092 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005093 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005094 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005095 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005096 }
5097
David Rientjesb070e652013-05-07 16:18:09 -07005098 /*
5099 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5100 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5101 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005102 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5103 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005104
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005105 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005106 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005107
5108 return val << PAGE_SHIFT;
5109}
5110
Tejun Heo791badb2013-12-05 12:28:02 -05005111static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
5112 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005113{
Tejun Heo182446d2013-08-08 20:11:24 -04005114 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005115 u64 val;
Tejun Heo791badb2013-12-05 12:28:02 -05005116 int name;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005117 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005118
5119 type = MEMFILE_TYPE(cft->private);
5120 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005121
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005122 switch (type) {
5123 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005124 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005125 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005126 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005127 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005128 break;
5129 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005130 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005131 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005132 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005133 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005134 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005135 case _KMEM:
5136 val = res_counter_read_u64(&memcg->kmem, name);
5137 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005138 default:
5139 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005140 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005141
Tejun Heo791badb2013-12-05 12:28:02 -05005142 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005143}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005144
Glauber Costa510fc4e2012-12-18 14:21:47 -08005145#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08005146/* should be called with activate_kmem_mutex held */
5147static int __memcg_activate_kmem(struct mem_cgroup *memcg,
5148 unsigned long long limit)
5149{
5150 int err = 0;
5151 int memcg_id;
5152
5153 if (memcg_kmem_is_active(memcg))
5154 return 0;
5155
5156 /*
5157 * We are going to allocate memory for data shared by all memory
5158 * cgroups so let's stop accounting here.
5159 */
5160 memcg_stop_kmem_account();
5161
Glauber Costa510fc4e2012-12-18 14:21:47 -08005162 /*
5163 * For simplicity, we won't allow this to be disabled. It also can't
5164 * be changed if the cgroup has children already, or if tasks had
5165 * already joined.
5166 *
5167 * If tasks join before we set the limit, a person looking at
5168 * kmem.usage_in_bytes will have no way to determine when it took
5169 * place, which makes the value quite meaningless.
5170 *
5171 * After it first became limited, changes in the value of the limit are
5172 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005173 */
Glauber Costa09998212013-02-22 16:34:55 -08005174 mutex_lock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005175 if (cgroup_task_count(memcg->css.cgroup) || memcg_has_children(memcg))
5176 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08005177 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005178 if (err)
5179 goto out;
5180
5181 memcg_id = ida_simple_get(&kmem_limited_groups,
5182 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
5183 if (memcg_id < 0) {
5184 err = memcg_id;
5185 goto out;
5186 }
5187
5188 /*
5189 * Make sure we have enough space for this cgroup in each root cache's
5190 * memcg_params.
5191 */
5192 err = memcg_update_all_caches(memcg_id + 1);
5193 if (err)
5194 goto out_rmid;
5195
5196 memcg->kmemcg_id = memcg_id;
5197 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
5198 mutex_init(&memcg->slab_caches_mutex);
5199
5200 /*
5201 * We couldn't have accounted to this cgroup, because it hasn't got the
5202 * active bit set yet, so this should succeed.
5203 */
5204 err = res_counter_set_limit(&memcg->kmem, limit);
5205 VM_BUG_ON(err);
5206
5207 static_key_slow_inc(&memcg_kmem_enabled_key);
5208 /*
5209 * Setting the active bit after enabling static branching will
5210 * guarantee no one starts accounting before all call sites are
5211 * patched.
5212 */
5213 memcg_kmem_set_active(memcg);
5214out:
5215 memcg_resume_kmem_account();
5216 return err;
5217
5218out_rmid:
5219 ida_simple_remove(&kmem_limited_groups, memcg_id);
5220 goto out;
5221}
5222
5223static int memcg_activate_kmem(struct mem_cgroup *memcg,
5224 unsigned long long limit)
5225{
5226 int ret;
5227
5228 mutex_lock(&activate_kmem_mutex);
5229 ret = __memcg_activate_kmem(memcg, limit);
5230 mutex_unlock(&activate_kmem_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005231 return ret;
5232}
5233
Vladimir Davydovd6441632014-01-23 15:53:09 -08005234static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5235 unsigned long long val)
5236{
5237 int ret;
5238
5239 if (!memcg_kmem_is_active(memcg))
5240 ret = memcg_activate_kmem(memcg, val);
5241 else
5242 ret = res_counter_set_limit(&memcg->kmem, val);
5243 return ret;
5244}
5245
Glauber Costa55007d82012-12-18 14:22:38 -08005246static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005247{
Glauber Costa55007d82012-12-18 14:22:38 -08005248 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005249 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005250
Glauber Costa510fc4e2012-12-18 14:21:47 -08005251 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005252 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08005253
Vladimir Davydovd6441632014-01-23 15:53:09 -08005254 mutex_lock(&activate_kmem_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08005255 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08005256 * If the parent cgroup is not kmem-active now, it cannot be activated
5257 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08005258 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08005259 if (memcg_kmem_is_active(parent))
5260 ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX);
5261 mutex_unlock(&activate_kmem_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005262 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005263}
Vladimir Davydovd6441632014-01-23 15:53:09 -08005264#else
5265static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5266 unsigned long long val)
5267{
5268 return -EINVAL;
5269}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005270#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005271
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005272/*
5273 * The user of this function is...
5274 * RES_LIMIT.
5275 */
Tejun Heo182446d2013-08-08 20:11:24 -04005276static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Paul Menage856c13a2008-07-25 01:47:04 -07005277 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005278{
Tejun Heo182446d2013-08-08 20:11:24 -04005279 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005280 enum res_type type;
5281 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005282 unsigned long long val;
5283 int ret;
5284
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005285 type = MEMFILE_TYPE(cft->private);
5286 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005287
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005288 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005289 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005290 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5291 ret = -EINVAL;
5292 break;
5293 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005294 /* This function does all necessary parse...reuse it */
5295 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005296 if (ret)
5297 break;
5298 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005299 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005300 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005301 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005302 else if (type == _KMEM)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005303 ret = memcg_update_kmem_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005304 else
5305 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005306 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005307 case RES_SOFT_LIMIT:
5308 ret = res_counter_memparse_write_strategy(buffer, &val);
5309 if (ret)
5310 break;
5311 /*
5312 * For memsw, soft limits are hard to implement in terms
5313 * of semantics, for now, we support soft limits for
5314 * control without swap
5315 */
5316 if (type == _MEM)
5317 ret = res_counter_set_soft_limit(&memcg->res, val);
5318 else
5319 ret = -EINVAL;
5320 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005321 default:
5322 ret = -EINVAL; /* should be BUG() ? */
5323 break;
5324 }
5325 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005326}
5327
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005328static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5329 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5330{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005331 unsigned long long min_limit, min_memsw_limit, tmp;
5332
5333 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5334 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005335 if (!memcg->use_hierarchy)
5336 goto out;
5337
Tejun Heo63876982013-08-08 20:11:23 -04005338 while (css_parent(&memcg->css)) {
5339 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005340 if (!memcg->use_hierarchy)
5341 break;
5342 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5343 min_limit = min(min_limit, tmp);
5344 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5345 min_memsw_limit = min(min_memsw_limit, tmp);
5346 }
5347out:
5348 *mem_limit = min_limit;
5349 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005350}
5351
Tejun Heo182446d2013-08-08 20:11:24 -04005352static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005353{
Tejun Heo182446d2013-08-08 20:11:24 -04005354 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005355 int name;
5356 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005357
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005358 type = MEMFILE_TYPE(event);
5359 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005360
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005361 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005362 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005363 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005364 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005365 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005366 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005367 else if (type == _KMEM)
5368 res_counter_reset_max(&memcg->kmem);
5369 else
5370 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005371 break;
5372 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005373 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005374 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005375 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005376 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005377 else if (type == _KMEM)
5378 res_counter_reset_failcnt(&memcg->kmem);
5379 else
5380 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005381 break;
5382 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005383
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005384 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005385}
5386
Tejun Heo182446d2013-08-08 20:11:24 -04005387static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005388 struct cftype *cft)
5389{
Tejun Heo182446d2013-08-08 20:11:24 -04005390 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005391}
5392
Daisuke Nishimura02491442010-03-10 15:22:17 -08005393#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005394static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005395 struct cftype *cft, u64 val)
5396{
Tejun Heo182446d2013-08-08 20:11:24 -04005397 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005398
5399 if (val >= (1 << NR_MOVE_TYPE))
5400 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005401
Glauber Costaee5e8472013-02-22 16:34:50 -08005402 /*
5403 * No kind of locking is needed in here, because ->can_attach() will
5404 * check this value once in the beginning of the process, and then carry
5405 * on with stale data. This means that changes to this value will only
5406 * affect task migrations starting after the change.
5407 */
5408 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005409 return 0;
5410}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005411#else
Tejun Heo182446d2013-08-08 20:11:24 -04005412static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005413 struct cftype *cft, u64 val)
5414{
5415 return -ENOSYS;
5416}
5417#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005418
Ying Han406eb0c2011-05-26 16:25:37 -07005419#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05005420static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07005421{
Greg Thelen25485de2013-11-12 15:07:40 -08005422 struct numa_stat {
5423 const char *name;
5424 unsigned int lru_mask;
5425 };
5426
5427 static const struct numa_stat stats[] = {
5428 { "total", LRU_ALL },
5429 { "file", LRU_ALL_FILE },
5430 { "anon", LRU_ALL_ANON },
5431 { "unevictable", BIT(LRU_UNEVICTABLE) },
5432 };
5433 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07005434 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08005435 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05005436 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07005437
Greg Thelen25485de2013-11-12 15:07:40 -08005438 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5439 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
5440 seq_printf(m, "%s=%lu", stat->name, nr);
5441 for_each_node_state(nid, N_MEMORY) {
5442 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5443 stat->lru_mask);
5444 seq_printf(m, " N%d=%lu", nid, nr);
5445 }
5446 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005447 }
Ying Han406eb0c2011-05-26 16:25:37 -07005448
Ying Han071aee12013-11-12 15:07:41 -08005449 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5450 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07005451
Ying Han071aee12013-11-12 15:07:41 -08005452 nr = 0;
5453 for_each_mem_cgroup_tree(iter, memcg)
5454 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
5455 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
5456 for_each_node_state(nid, N_MEMORY) {
5457 nr = 0;
5458 for_each_mem_cgroup_tree(iter, memcg)
5459 nr += mem_cgroup_node_nr_lru_pages(
5460 iter, nid, stat->lru_mask);
5461 seq_printf(m, " N%d=%lu", nid, nr);
5462 }
5463 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005464 }
Ying Han406eb0c2011-05-26 16:25:37 -07005465
Ying Han406eb0c2011-05-26 16:25:37 -07005466 return 0;
5467}
5468#endif /* CONFIG_NUMA */
5469
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005470static inline void mem_cgroup_lru_names_not_uptodate(void)
5471{
5472 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5473}
5474
Tejun Heo2da8ca82013-12-05 12:28:04 -05005475static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005476{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005477 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005478 struct mem_cgroup *mi;
5479 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005480
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005481 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005482 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005483 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005484 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5485 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005486 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005487
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005488 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5489 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5490 mem_cgroup_read_events(memcg, i));
5491
5492 for (i = 0; i < NR_LRU_LISTS; i++)
5493 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5494 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5495
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005496 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005497 {
5498 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005499 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005500 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005501 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005502 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5503 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005504 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005505
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005506 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5507 long long val = 0;
5508
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005509 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005510 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005511 for_each_mem_cgroup_tree(mi, memcg)
5512 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5513 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5514 }
5515
5516 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5517 unsigned long long val = 0;
5518
5519 for_each_mem_cgroup_tree(mi, memcg)
5520 val += mem_cgroup_read_events(mi, i);
5521 seq_printf(m, "total_%s %llu\n",
5522 mem_cgroup_events_names[i], val);
5523 }
5524
5525 for (i = 0; i < NR_LRU_LISTS; i++) {
5526 unsigned long long val = 0;
5527
5528 for_each_mem_cgroup_tree(mi, memcg)
5529 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5530 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005531 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005532
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005533#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005534 {
5535 int nid, zid;
5536 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005537 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005538 unsigned long recent_rotated[2] = {0, 0};
5539 unsigned long recent_scanned[2] = {0, 0};
5540
5541 for_each_online_node(nid)
5542 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005543 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005544 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005545
Hugh Dickins89abfab2012-05-29 15:06:53 -07005546 recent_rotated[0] += rstat->recent_rotated[0];
5547 recent_rotated[1] += rstat->recent_rotated[1];
5548 recent_scanned[0] += rstat->recent_scanned[0];
5549 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005550 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005551 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5552 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5553 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5554 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005555 }
5556#endif
5557
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005558 return 0;
5559}
5560
Tejun Heo182446d2013-08-08 20:11:24 -04005561static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5562 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005563{
Tejun Heo182446d2013-08-08 20:11:24 -04005564 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005565
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005566 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005567}
5568
Tejun Heo182446d2013-08-08 20:11:24 -04005569static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5570 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005571{
Tejun Heo182446d2013-08-08 20:11:24 -04005572 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005573 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005574
Tejun Heo63876982013-08-08 20:11:23 -04005575 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005576 return -EINVAL;
5577
Glauber Costa09998212013-02-22 16:34:55 -08005578 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005579
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005580 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005581 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005582 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005583 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005584 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005585
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005586 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005587
Glauber Costa09998212013-02-22 16:34:55 -08005588 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005589
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005590 return 0;
5591}
5592
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005593static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5594{
5595 struct mem_cgroup_threshold_ary *t;
5596 u64 usage;
5597 int i;
5598
5599 rcu_read_lock();
5600 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005601 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005602 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005603 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005604
5605 if (!t)
5606 goto unlock;
5607
5608 usage = mem_cgroup_usage(memcg, swap);
5609
5610 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005611 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005612 * If it's not true, a threshold was crossed after last
5613 * call of __mem_cgroup_threshold().
5614 */
Phil Carmody5407a562010-05-26 14:42:42 -07005615 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005616
5617 /*
5618 * Iterate backward over array of thresholds starting from
5619 * current_threshold and check if a threshold is crossed.
5620 * If none of thresholds below usage is crossed, we read
5621 * only one element of the array here.
5622 */
5623 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5624 eventfd_signal(t->entries[i].eventfd, 1);
5625
5626 /* i = current_threshold + 1 */
5627 i++;
5628
5629 /*
5630 * Iterate forward over array of thresholds starting from
5631 * current_threshold+1 and check if a threshold is crossed.
5632 * If none of thresholds above usage is crossed, we read
5633 * only one element of the array here.
5634 */
5635 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5636 eventfd_signal(t->entries[i].eventfd, 1);
5637
5638 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005639 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005640unlock:
5641 rcu_read_unlock();
5642}
5643
5644static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5645{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005646 while (memcg) {
5647 __mem_cgroup_threshold(memcg, false);
5648 if (do_swap_account)
5649 __mem_cgroup_threshold(memcg, true);
5650
5651 memcg = parent_mem_cgroup(memcg);
5652 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005653}
5654
5655static int compare_thresholds(const void *a, const void *b)
5656{
5657 const struct mem_cgroup_threshold *_a = a;
5658 const struct mem_cgroup_threshold *_b = b;
5659
Greg Thelen2bff24a2013-09-11 14:23:08 -07005660 if (_a->threshold > _b->threshold)
5661 return 1;
5662
5663 if (_a->threshold < _b->threshold)
5664 return -1;
5665
5666 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005667}
5668
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005669static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005670{
5671 struct mem_cgroup_eventfd_list *ev;
5672
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005673 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005674 eventfd_signal(ev->eventfd, 1);
5675 return 0;
5676}
5677
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005678static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005679{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005680 struct mem_cgroup *iter;
5681
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005682 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005683 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005684}
5685
Tejun Heo59b6f872013-11-22 18:20:43 -05005686static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005687 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005688{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005689 struct mem_cgroup_thresholds *thresholds;
5690 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005691 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005692 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005693
5694 ret = res_counter_memparse_write_strategy(args, &threshold);
5695 if (ret)
5696 return ret;
5697
5698 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005699
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005700 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005701 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005702 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005703 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005704 else
5705 BUG();
5706
5707 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5708
5709 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005710 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005711 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5712
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005713 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005714
5715 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005716 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005717 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005718 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005719 ret = -ENOMEM;
5720 goto unlock;
5721 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005722 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005723
5724 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005725 if (thresholds->primary) {
5726 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005727 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005728 }
5729
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005730 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005731 new->entries[size - 1].eventfd = eventfd;
5732 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005733
5734 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005735 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005736 compare_thresholds, NULL);
5737
5738 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005739 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005740 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005741 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005742 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005743 * new->current_threshold will not be used until
5744 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005745 * it here.
5746 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005747 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005748 } else
5749 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005750 }
5751
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005752 /* Free old spare buffer and save old primary buffer as spare */
5753 kfree(thresholds->spare);
5754 thresholds->spare = thresholds->primary;
5755
5756 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005757
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005758 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005759 synchronize_rcu();
5760
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005761unlock:
5762 mutex_unlock(&memcg->thresholds_lock);
5763
5764 return ret;
5765}
5766
Tejun Heo59b6f872013-11-22 18:20:43 -05005767static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005768 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005769{
Tejun Heo59b6f872013-11-22 18:20:43 -05005770 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005771}
5772
Tejun Heo59b6f872013-11-22 18:20:43 -05005773static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005774 struct eventfd_ctx *eventfd, const char *args)
5775{
Tejun Heo59b6f872013-11-22 18:20:43 -05005776 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005777}
5778
Tejun Heo59b6f872013-11-22 18:20:43 -05005779static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005780 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005781{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005782 struct mem_cgroup_thresholds *thresholds;
5783 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005784 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005785 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005786
5787 mutex_lock(&memcg->thresholds_lock);
5788 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005789 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005790 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005791 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005792 else
5793 BUG();
5794
Anton Vorontsov371528c2012-02-24 05:14:46 +04005795 if (!thresholds->primary)
5796 goto unlock;
5797
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005798 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5799
5800 /* Check if a threshold crossed before removing */
5801 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5802
5803 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005804 size = 0;
5805 for (i = 0; i < thresholds->primary->size; i++) {
5806 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005807 size++;
5808 }
5809
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005810 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005811
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005812 /* Set thresholds array to NULL if we don't have thresholds */
5813 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005814 kfree(new);
5815 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005816 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005817 }
5818
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005819 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005820
5821 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005822 new->current_threshold = -1;
5823 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5824 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005825 continue;
5826
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005827 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005828 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005829 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005830 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005831 * until rcu_assign_pointer(), so it's safe to increment
5832 * it here.
5833 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005834 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005835 }
5836 j++;
5837 }
5838
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005839swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005840 /* Swap primary and spare array */
5841 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005842 /* If all events are unregistered, free the spare array */
5843 if (!new) {
5844 kfree(thresholds->spare);
5845 thresholds->spare = NULL;
5846 }
5847
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005848 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005849
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005850 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005851 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005852unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005853 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005854}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005855
Tejun Heo59b6f872013-11-22 18:20:43 -05005856static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005857 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005858{
Tejun Heo59b6f872013-11-22 18:20:43 -05005859 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005860}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005861
Tejun Heo59b6f872013-11-22 18:20:43 -05005862static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005863 struct eventfd_ctx *eventfd)
5864{
Tejun Heo59b6f872013-11-22 18:20:43 -05005865 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005866}
5867
Tejun Heo59b6f872013-11-22 18:20:43 -05005868static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005869 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005870{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005871 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005872
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005873 event = kmalloc(sizeof(*event), GFP_KERNEL);
5874 if (!event)
5875 return -ENOMEM;
5876
Michal Hocko1af8efe2011-07-26 16:08:24 -07005877 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005878
5879 event->eventfd = eventfd;
5880 list_add(&event->list, &memcg->oom_notify);
5881
5882 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005883 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005884 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005885 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005886
5887 return 0;
5888}
5889
Tejun Heo59b6f872013-11-22 18:20:43 -05005890static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005891 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005892{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005893 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005894
Michal Hocko1af8efe2011-07-26 16:08:24 -07005895 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005896
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005897 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005898 if (ev->eventfd == eventfd) {
5899 list_del(&ev->list);
5900 kfree(ev);
5901 }
5902 }
5903
Michal Hocko1af8efe2011-07-26 16:08:24 -07005904 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005905}
5906
Tejun Heo2da8ca82013-12-05 12:28:04 -05005907static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005908{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005909 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005910
Tejun Heo791badb2013-12-05 12:28:02 -05005911 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
5912 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005913 return 0;
5914}
5915
Tejun Heo182446d2013-08-08 20:11:24 -04005916static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005917 struct cftype *cft, u64 val)
5918{
Tejun Heo182446d2013-08-08 20:11:24 -04005919 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005920 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005921
5922 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005923 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005924 return -EINVAL;
5925
Glauber Costa09998212013-02-22 16:34:55 -08005926 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005927 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005928 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005929 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005930 return -EINVAL;
5931 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005932 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005933 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005934 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005935 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005936 return 0;
5937}
5938
Andrew Mortonc255a452012-07-31 16:43:02 -07005939#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005940static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005941{
Glauber Costa55007d82012-12-18 14:22:38 -08005942 int ret;
5943
Glauber Costa2633d7a2012-12-18 14:22:34 -08005944 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005945 ret = memcg_propagate_kmem(memcg);
5946 if (ret)
5947 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005948
Glauber Costa1d62e432012-04-09 19:36:33 -03005949 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005950}
Glauber Costae5671df2011-12-11 21:47:01 +00005951
Li Zefan10d5ebf2013-07-08 16:00:33 -07005952static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005953{
Glauber Costa1d62e432012-04-09 19:36:33 -03005954 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005955}
5956
5957static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5958{
5959 if (!memcg_kmem_is_active(memcg))
5960 return;
5961
5962 /*
5963 * kmem charges can outlive the cgroup. In the case of slab
5964 * pages, for instance, a page contain objects from various
5965 * processes. As we prevent from taking a reference for every
5966 * such allocation we have to be careful when doing uncharge
5967 * (see memcg_uncharge_kmem) and here during offlining.
5968 *
5969 * The idea is that that only the _last_ uncharge which sees
5970 * the dead memcg will drop the last reference. An additional
5971 * reference is taken here before the group is marked dead
5972 * which is then paired with css_put during uncharge resp. here.
5973 *
5974 * Although this might sound strange as this path is called from
5975 * css_offline() when the referencemight have dropped down to 0
5976 * and shouldn't be incremented anymore (css_tryget would fail)
5977 * we do not have other options because of the kmem allocations
5978 * lifetime.
5979 */
5980 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005981
5982 memcg_kmem_mark_dead(memcg);
5983
5984 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5985 return;
5986
Glauber Costa7de37682012-12-18 14:22:07 -08005987 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005988 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005989}
Glauber Costae5671df2011-12-11 21:47:01 +00005990#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005991static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005992{
5993 return 0;
5994}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005995
Li Zefan10d5ebf2013-07-08 16:00:33 -07005996static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5997{
5998}
5999
6000static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00006001{
6002}
Glauber Costae5671df2011-12-11 21:47:01 +00006003#endif
6004
Tejun Heo79bd9812013-11-22 18:20:42 -05006005/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05006006 * DO NOT USE IN NEW FILES.
6007 *
6008 * "cgroup.event_control" implementation.
6009 *
6010 * This is way over-engineered. It tries to support fully configurable
6011 * events for each user. Such level of flexibility is completely
6012 * unnecessary especially in the light of the planned unified hierarchy.
6013 *
6014 * Please deprecate this and replace with something simpler if at all
6015 * possible.
6016 */
6017
6018/*
Tejun Heo79bd9812013-11-22 18:20:42 -05006019 * Unregister event and free resources.
6020 *
6021 * Gets called from workqueue.
6022 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05006023static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05006024{
Tejun Heo3bc942f2013-11-22 18:20:44 -05006025 struct mem_cgroup_event *event =
6026 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05006027 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05006028
6029 remove_wait_queue(event->wqh, &event->wait);
6030
Tejun Heo59b6f872013-11-22 18:20:43 -05006031 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05006032
6033 /* Notify userspace the event is going away. */
6034 eventfd_signal(event->eventfd, 1);
6035
6036 eventfd_ctx_put(event->eventfd);
6037 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05006038 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05006039}
6040
6041/*
6042 * Gets called on POLLHUP on eventfd when user closes it.
6043 *
6044 * Called with wqh->lock held and interrupts disabled.
6045 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05006046static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
6047 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05006048{
Tejun Heo3bc942f2013-11-22 18:20:44 -05006049 struct mem_cgroup_event *event =
6050 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05006051 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05006052 unsigned long flags = (unsigned long)key;
6053
6054 if (flags & POLLHUP) {
6055 /*
6056 * If the event has been detached at cgroup removal, we
6057 * can simply return knowing the other side will cleanup
6058 * for us.
6059 *
6060 * We can't race against event freeing since the other
6061 * side will require wqh->lock via remove_wait_queue(),
6062 * which we hold.
6063 */
Tejun Heofba94802013-11-22 18:20:43 -05006064 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006065 if (!list_empty(&event->list)) {
6066 list_del_init(&event->list);
6067 /*
6068 * We are in atomic context, but cgroup_event_remove()
6069 * may sleep, so we have to call it in workqueue.
6070 */
6071 schedule_work(&event->remove);
6072 }
Tejun Heofba94802013-11-22 18:20:43 -05006073 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006074 }
6075
6076 return 0;
6077}
6078
Tejun Heo3bc942f2013-11-22 18:20:44 -05006079static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05006080 wait_queue_head_t *wqh, poll_table *pt)
6081{
Tejun Heo3bc942f2013-11-22 18:20:44 -05006082 struct mem_cgroup_event *event =
6083 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05006084
6085 event->wqh = wqh;
6086 add_wait_queue(wqh, &event->wait);
6087}
6088
6089/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05006090 * DO NOT USE IN NEW FILES.
6091 *
Tejun Heo79bd9812013-11-22 18:20:42 -05006092 * Parse input and register new cgroup event handler.
6093 *
6094 * Input must be in format '<event_fd> <control_fd> <args>'.
6095 * Interpretation of args is defined by control file implementation.
6096 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05006097static int memcg_write_event_control(struct cgroup_subsys_state *css,
6098 struct cftype *cft, const char *buffer)
Tejun Heo79bd9812013-11-22 18:20:42 -05006099{
Tejun Heofba94802013-11-22 18:20:43 -05006100 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006101 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05006102 struct cgroup_subsys_state *cfile_css;
6103 unsigned int efd, cfd;
6104 struct fd efile;
6105 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05006106 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05006107 char *endp;
6108 int ret;
6109
6110 efd = simple_strtoul(buffer, &endp, 10);
6111 if (*endp != ' ')
6112 return -EINVAL;
6113 buffer = endp + 1;
6114
6115 cfd = simple_strtoul(buffer, &endp, 10);
6116 if ((*endp != ' ') && (*endp != '\0'))
6117 return -EINVAL;
6118 buffer = endp + 1;
6119
6120 event = kzalloc(sizeof(*event), GFP_KERNEL);
6121 if (!event)
6122 return -ENOMEM;
6123
Tejun Heo59b6f872013-11-22 18:20:43 -05006124 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05006125 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006126 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
6127 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
6128 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05006129
6130 efile = fdget(efd);
6131 if (!efile.file) {
6132 ret = -EBADF;
6133 goto out_kfree;
6134 }
6135
6136 event->eventfd = eventfd_ctx_fileget(efile.file);
6137 if (IS_ERR(event->eventfd)) {
6138 ret = PTR_ERR(event->eventfd);
6139 goto out_put_efile;
6140 }
6141
6142 cfile = fdget(cfd);
6143 if (!cfile.file) {
6144 ret = -EBADF;
6145 goto out_put_eventfd;
6146 }
6147
6148 /* the process need read permission on control file */
6149 /* AV: shouldn't we check that it's been opened for read instead? */
6150 ret = inode_permission(file_inode(cfile.file), MAY_READ);
6151 if (ret < 0)
6152 goto out_put_cfile;
6153
Tejun Heo79bd9812013-11-22 18:20:42 -05006154 /*
Tejun Heofba94802013-11-22 18:20:43 -05006155 * Determine the event callbacks and set them in @event. This used
6156 * to be done via struct cftype but cgroup core no longer knows
6157 * about these events. The following is crude but the whole thing
6158 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05006159 *
6160 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05006161 */
6162 name = cfile.file->f_dentry->d_name.name;
6163
6164 if (!strcmp(name, "memory.usage_in_bytes")) {
6165 event->register_event = mem_cgroup_usage_register_event;
6166 event->unregister_event = mem_cgroup_usage_unregister_event;
6167 } else if (!strcmp(name, "memory.oom_control")) {
6168 event->register_event = mem_cgroup_oom_register_event;
6169 event->unregister_event = mem_cgroup_oom_unregister_event;
6170 } else if (!strcmp(name, "memory.pressure_level")) {
6171 event->register_event = vmpressure_register_event;
6172 event->unregister_event = vmpressure_unregister_event;
6173 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05006174 event->register_event = memsw_cgroup_usage_register_event;
6175 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05006176 } else {
6177 ret = -EINVAL;
6178 goto out_put_cfile;
6179 }
6180
6181 /*
Tejun Heob5557c42013-11-22 18:20:42 -05006182 * Verify @cfile should belong to @css. Also, remaining events are
6183 * automatically removed on cgroup destruction but the removal is
6184 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05006185 */
6186 rcu_read_lock();
6187
6188 ret = -EINVAL;
Tejun Heob5557c42013-11-22 18:20:42 -05006189 cfile_css = css_from_dir(cfile.file->f_dentry->d_parent,
6190 &mem_cgroup_subsys);
6191 if (cfile_css == css && css_tryget(css))
Tejun Heo79bd9812013-11-22 18:20:42 -05006192 ret = 0;
6193
6194 rcu_read_unlock();
6195 if (ret)
6196 goto out_put_cfile;
6197
Tejun Heo59b6f872013-11-22 18:20:43 -05006198 ret = event->register_event(memcg, event->eventfd, buffer);
Tejun Heo79bd9812013-11-22 18:20:42 -05006199 if (ret)
6200 goto out_put_css;
6201
6202 efile.file->f_op->poll(efile.file, &event->pt);
6203
Tejun Heofba94802013-11-22 18:20:43 -05006204 spin_lock(&memcg->event_list_lock);
6205 list_add(&event->list, &memcg->event_list);
6206 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006207
6208 fdput(cfile);
6209 fdput(efile);
6210
6211 return 0;
6212
6213out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05006214 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05006215out_put_cfile:
6216 fdput(cfile);
6217out_put_eventfd:
6218 eventfd_ctx_put(event->eventfd);
6219out_put_efile:
6220 fdput(efile);
6221out_kfree:
6222 kfree(event);
6223
6224 return ret;
6225}
6226
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006227static struct cftype mem_cgroup_files[] = {
6228 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006229 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006230 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006231 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006232 },
6233 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006234 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006235 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006236 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006237 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006238 },
6239 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006240 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006241 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07006242 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006243 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006244 },
6245 {
Balbir Singh296c81d2009-09-23 15:56:36 -07006246 .name = "soft_limit_in_bytes",
6247 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
6248 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006249 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07006250 },
6251 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006252 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006253 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006254 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006255 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006256 },
Balbir Singh8697d332008-02-07 00:13:59 -08006257 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006258 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006259 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006260 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08006261 {
6262 .name = "force_empty",
6263 .trigger = mem_cgroup_force_empty_write,
6264 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08006265 {
6266 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07006267 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08006268 .write_u64 = mem_cgroup_hierarchy_write,
6269 .read_u64 = mem_cgroup_hierarchy_read,
6270 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006271 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05006272 .name = "cgroup.event_control", /* XXX: for compat */
6273 .write_string = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05006274 .flags = CFTYPE_NO_PREFIX,
6275 .mode = S_IWUGO,
6276 },
6277 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006278 .name = "swappiness",
6279 .read_u64 = mem_cgroup_swappiness_read,
6280 .write_u64 = mem_cgroup_swappiness_write,
6281 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006282 {
6283 .name = "move_charge_at_immigrate",
6284 .read_u64 = mem_cgroup_move_charge_read,
6285 .write_u64 = mem_cgroup_move_charge_write,
6286 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006287 {
6288 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006289 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006290 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006291 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6292 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006293 {
6294 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006295 },
Ying Han406eb0c2011-05-26 16:25:37 -07006296#ifdef CONFIG_NUMA
6297 {
6298 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006299 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006300 },
6301#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006302#ifdef CONFIG_MEMCG_KMEM
6303 {
6304 .name = "kmem.limit_in_bytes",
6305 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6306 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006307 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006308 },
6309 {
6310 .name = "kmem.usage_in_bytes",
6311 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006312 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006313 },
6314 {
6315 .name = "kmem.failcnt",
6316 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6317 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006318 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006319 },
6320 {
6321 .name = "kmem.max_usage_in_bytes",
6322 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6323 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006324 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006325 },
Glauber Costa749c5412012-12-18 14:23:01 -08006326#ifdef CONFIG_SLABINFO
6327 {
6328 .name = "kmem.slabinfo",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006329 .seq_show = mem_cgroup_slabinfo_read,
Glauber Costa749c5412012-12-18 14:23:01 -08006330 },
6331#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006332#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006333 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006334};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006335
Michal Hocko2d110852013-02-22 16:34:43 -08006336#ifdef CONFIG_MEMCG_SWAP
6337static struct cftype memsw_cgroup_files[] = {
6338 {
6339 .name = "memsw.usage_in_bytes",
6340 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006341 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006342 },
6343 {
6344 .name = "memsw.max_usage_in_bytes",
6345 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6346 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006347 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006348 },
6349 {
6350 .name = "memsw.limit_in_bytes",
6351 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6352 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006353 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006354 },
6355 {
6356 .name = "memsw.failcnt",
6357 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6358 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006359 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006360 },
6361 { }, /* terminate */
6362};
6363#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006364static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006365{
6366 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006367 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006368 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006369 /*
6370 * This routine is called against possible nodes.
6371 * But it's BUG to call kmalloc() against offline node.
6372 *
6373 * TODO: this routine can waste much memory for nodes which will
6374 * never be onlined. It's better to use memory hotplug callback
6375 * function.
6376 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006377 if (!node_state(node, N_NORMAL_MEMORY))
6378 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006379 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006380 if (!pn)
6381 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006382
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006383 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6384 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006385 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006386 mz->usage_in_excess = 0;
6387 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006388 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006389 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006390 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006391 return 0;
6392}
6393
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006394static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006395{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006396 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006397}
6398
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006399static struct mem_cgroup *mem_cgroup_alloc(void)
6400{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006401 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006402 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006403
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006404 size = sizeof(struct mem_cgroup);
6405 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006406
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006407 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006408 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006409 return NULL;
6410
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006411 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6412 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006413 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006414 spin_lock_init(&memcg->pcp_counter_lock);
6415 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006416
6417out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006418 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006419 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006420}
6421
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006422/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006423 * At destroying mem_cgroup, references from swap_cgroup can remain.
6424 * (scanning all at force_empty is too costly...)
6425 *
6426 * Instead of clearing all references at force_empty, we remember
6427 * the number of reference from swap_cgroup and free mem_cgroup when
6428 * it goes down to 0.
6429 *
6430 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006431 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006432
6433static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006434{
Glauber Costac8b2a362012-12-18 14:22:13 -08006435 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07006436
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006437 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006438
6439 for_each_node(node)
6440 free_mem_cgroup_per_zone_info(memcg, node);
6441
6442 free_percpu(memcg->stat);
6443
Glauber Costa3f134612012-05-29 15:07:11 -07006444 /*
6445 * We need to make sure that (at least for now), the jump label
6446 * destruction code runs outside of the cgroup lock. This is because
6447 * get_online_cpus(), which is called from the static_branch update,
6448 * can't be called inside the cgroup_lock. cpusets are the ones
6449 * enforcing this dependency, so if they ever change, we might as well.
6450 *
6451 * schedule_work() will guarantee this happens. Be careful if you need
6452 * to move this code around, and make sure it is outside
6453 * the cgroup_lock.
6454 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006455 disarm_static_keys(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006456 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006457}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006458
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006459/*
6460 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6461 */
Glauber Costae1aab162011-12-11 21:47:03 +00006462struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006463{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006464 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006465 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006466 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006467}
Glauber Costae1aab162011-12-11 21:47:03 +00006468EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006469
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006470static void __init mem_cgroup_soft_limit_tree_init(void)
6471{
6472 struct mem_cgroup_tree_per_node *rtpn;
6473 struct mem_cgroup_tree_per_zone *rtpz;
6474 int tmp, node, zone;
6475
6476 for_each_node(node) {
6477 tmp = node;
6478 if (!node_state(node, N_NORMAL_MEMORY))
6479 tmp = -1;
6480 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
6481 BUG_ON(!rtpn);
6482
6483 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6484
6485 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6486 rtpz = &rtpn->rb_tree_per_zone[zone];
6487 rtpz->rb_root = RB_ROOT;
6488 spin_lock_init(&rtpz->lock);
6489 }
6490 }
6491}
6492
Li Zefan0eb253e2009-01-15 13:51:25 -08006493static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006494mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006495{
Glauber Costad142e3e2013-02-22 16:34:52 -08006496 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006497 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006498 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006499
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006500 memcg = mem_cgroup_alloc();
6501 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006502 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006503
Bob Liu3ed28fa2012-01-12 17:19:04 -08006504 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006505 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006506 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006507
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006508 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006509 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006510 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006511 res_counter_init(&memcg->res, NULL);
6512 res_counter_init(&memcg->memsw, NULL);
6513 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006514 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006515
Glauber Costad142e3e2013-02-22 16:34:52 -08006516 memcg->last_scanned_node = MAX_NUMNODES;
6517 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006518 memcg->move_charge_at_immigrate = 0;
6519 mutex_init(&memcg->thresholds_lock);
6520 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006521 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05006522 INIT_LIST_HEAD(&memcg->event_list);
6523 spin_lock_init(&memcg->event_list_lock);
Glauber Costad142e3e2013-02-22 16:34:52 -08006524
6525 return &memcg->css;
6526
6527free_out:
6528 __mem_cgroup_free(memcg);
6529 return ERR_PTR(error);
6530}
6531
6532static int
Tejun Heoeb954192013-08-08 20:11:23 -04006533mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006534{
Tejun Heoeb954192013-08-08 20:11:23 -04006535 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6536 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006537
Li Zefan4219b2d2013-09-23 16:56:29 +08006538 if (css->cgroup->id > MEM_CGROUP_ID_MAX)
6539 return -ENOSPC;
6540
Tejun Heo63876982013-08-08 20:11:23 -04006541 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006542 return 0;
6543
Glauber Costa09998212013-02-22 16:34:55 -08006544 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006545
6546 memcg->use_hierarchy = parent->use_hierarchy;
6547 memcg->oom_kill_disable = parent->oom_kill_disable;
6548 memcg->swappiness = mem_cgroup_swappiness(parent);
6549
6550 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006551 res_counter_init(&memcg->res, &parent->res);
6552 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006553 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006554
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006555 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006556 * No need to take a reference to the parent because cgroup
6557 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006558 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006559 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006560 res_counter_init(&memcg->res, NULL);
6561 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006562 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006563 /*
6564 * Deeper hierachy with use_hierarchy == false doesn't make
6565 * much sense so let cgroup subsystem know about this
6566 * unfortunate state in our controller.
6567 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006568 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006569 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006570 }
Glauber Costa09998212013-02-22 16:34:55 -08006571 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08006572
6573 return memcg_init_kmem(memcg, &mem_cgroup_subsys);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006574}
6575
Michal Hocko5f578162013-04-29 15:07:17 -07006576/*
6577 * Announce all parents that a group from their hierarchy is gone.
6578 */
6579static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6580{
6581 struct mem_cgroup *parent = memcg;
6582
6583 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006584 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006585
6586 /*
6587 * if the root memcg is not hierarchical we have to check it
6588 * explicitely.
6589 */
6590 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006591 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006592}
6593
Tejun Heoeb954192013-08-08 20:11:23 -04006594static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006595{
Tejun Heoeb954192013-08-08 20:11:23 -04006596 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006597 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05006598
6599 /*
6600 * Unregister events and notify userspace.
6601 * Notify userspace about cgroup removing only after rmdir of cgroup
6602 * directory to avoid race between userspace and kernelspace.
6603 */
Tejun Heofba94802013-11-22 18:20:43 -05006604 spin_lock(&memcg->event_list_lock);
6605 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05006606 list_del_init(&event->list);
6607 schedule_work(&event->remove);
6608 }
Tejun Heofba94802013-11-22 18:20:43 -05006609 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006610
Li Zefan10d5ebf2013-07-08 16:00:33 -07006611 kmem_cgroup_css_offline(memcg);
6612
Michal Hocko5f578162013-04-29 15:07:17 -07006613 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006614 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006615 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006616 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006617}
6618
Tejun Heoeb954192013-08-08 20:11:23 -04006619static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006620{
Tejun Heoeb954192013-08-08 20:11:23 -04006621 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner96f1c582013-12-12 17:12:34 -08006622 /*
6623 * XXX: css_offline() would be where we should reparent all
6624 * memory to prepare the cgroup for destruction. However,
6625 * memcg does not do css_tryget() and res_counter charging
6626 * under the same RCU lock region, which means that charging
6627 * could race with offlining. Offlining only happens to
6628 * cgroups with no tasks in them but charges can show up
6629 * without any tasks from the swapin path when the target
6630 * memcg is looked up from the swapout record and not from the
6631 * current task as it usually is. A race like this can leak
6632 * charges and put pages with stale cgroup pointers into
6633 * circulation:
6634 *
6635 * #0 #1
6636 * lookup_swap_cgroup_id()
6637 * rcu_read_lock()
6638 * mem_cgroup_lookup()
6639 * css_tryget()
6640 * rcu_read_unlock()
6641 * disable css_tryget()
6642 * call_rcu()
6643 * offline_css()
6644 * reparent_charges()
6645 * res_counter_charge()
6646 * css_put()
6647 * css_free()
6648 * pc->mem_cgroup = dead memcg
6649 * add page to lru
6650 *
6651 * The bulk of the charges are still moved in offline_css() to
6652 * avoid pinning a lot of pages in case a long-term reference
6653 * like a swapout record is deferring the css_free() to long
6654 * after offlining. But this makes sure we catch any charges
6655 * made after offlining:
6656 */
6657 mem_cgroup_reparent_charges(memcg);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006658
Li Zefan10d5ebf2013-07-08 16:00:33 -07006659 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006660 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006661}
6662
Daisuke Nishimura02491442010-03-10 15:22:17 -08006663#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006664/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006665#define PRECHARGE_COUNT_AT_ONCE 256
6666static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006667{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006668 int ret = 0;
6669 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006670 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006671
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006672 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006673 mc.precharge += count;
6674 /* we don't need css_get for root */
6675 return ret;
6676 }
6677 /* try to charge at once */
6678 if (count > 1) {
6679 struct res_counter *dummy;
6680 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006681 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006682 * by cgroup_lock_live_cgroup() that it is not removed and we
6683 * are still under the same cgroup_mutex. So we can postpone
6684 * css_get().
6685 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006686 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006687 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006688 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006689 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006690 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006691 goto one_by_one;
6692 }
6693 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006694 return ret;
6695 }
6696one_by_one:
6697 /* fall back to one by one charge */
6698 while (count--) {
6699 if (signal_pending(current)) {
6700 ret = -EINTR;
6701 break;
6702 }
6703 if (!batch_count--) {
6704 batch_count = PRECHARGE_COUNT_AT_ONCE;
6705 cond_resched();
6706 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006707 ret = __mem_cgroup_try_charge(NULL,
6708 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006709 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006710 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006711 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006712 mc.precharge++;
6713 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006714 return ret;
6715}
6716
6717/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006718 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006719 * @vma: the vma the pte to be checked belongs
6720 * @addr: the address corresponding to the pte to be checked
6721 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006722 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006723 *
6724 * Returns
6725 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6726 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6727 * move charge. if @target is not NULL, the page is stored in target->page
6728 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006729 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6730 * target for charge migration. if @target is not NULL, the entry is stored
6731 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006732 *
6733 * Called with pte lock held.
6734 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006735union mc_target {
6736 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006737 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006738};
6739
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006740enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006741 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006742 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006743 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006744};
6745
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006746static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6747 unsigned long addr, pte_t ptent)
6748{
6749 struct page *page = vm_normal_page(vma, addr, ptent);
6750
6751 if (!page || !page_mapped(page))
6752 return NULL;
6753 if (PageAnon(page)) {
6754 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006755 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006756 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006757 } else if (!move_file())
6758 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006759 return NULL;
6760 if (!get_page_unless_zero(page))
6761 return NULL;
6762
6763 return page;
6764}
6765
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006766#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006767static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6768 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6769{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006770 struct page *page = NULL;
6771 swp_entry_t ent = pte_to_swp_entry(ptent);
6772
6773 if (!move_anon() || non_swap_entry(ent))
6774 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006775 /*
6776 * Because lookup_swap_cache() updates some statistics counter,
6777 * we call find_get_page() with swapper_space directly.
6778 */
Shaohua Li33806f02013-02-22 16:34:37 -08006779 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006780 if (do_swap_account)
6781 entry->val = ent.val;
6782
6783 return page;
6784}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006785#else
6786static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6787 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6788{
6789 return NULL;
6790}
6791#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006792
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006793static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6794 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6795{
6796 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006797 struct address_space *mapping;
6798 pgoff_t pgoff;
6799
6800 if (!vma->vm_file) /* anonymous vma */
6801 return NULL;
6802 if (!move_file())
6803 return NULL;
6804
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006805 mapping = vma->vm_file->f_mapping;
6806 if (pte_none(ptent))
6807 pgoff = linear_page_index(vma, addr);
6808 else /* pte_file(ptent) is true */
6809 pgoff = pte_to_pgoff(ptent);
6810
6811 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006812 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006813
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006814#ifdef CONFIG_SWAP
6815 /* shmem/tmpfs may report page out on swap: account for that too. */
6816 if (radix_tree_exceptional_entry(page)) {
6817 swp_entry_t swap = radix_to_swp_entry(page);
6818 if (do_swap_account)
6819 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006820 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006821 }
6822#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006823 return page;
6824}
6825
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006826static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006827 unsigned long addr, pte_t ptent, union mc_target *target)
6828{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006829 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006830 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006831 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006832 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006833
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006834 if (pte_present(ptent))
6835 page = mc_handle_present_pte(vma, addr, ptent);
6836 else if (is_swap_pte(ptent))
6837 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006838 else if (pte_none(ptent) || pte_file(ptent))
6839 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006840
6841 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006842 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006843 if (page) {
6844 pc = lookup_page_cgroup(page);
6845 /*
6846 * Do only loose check w/o page_cgroup lock.
6847 * mem_cgroup_move_account() checks the pc is valid or not under
6848 * the lock.
6849 */
6850 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6851 ret = MC_TARGET_PAGE;
6852 if (target)
6853 target->page = page;
6854 }
6855 if (!ret || !target)
6856 put_page(page);
6857 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006858 /* There is a swap entry and a page doesn't exist or isn't charged */
6859 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08006860 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006861 ret = MC_TARGET_SWAP;
6862 if (target)
6863 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006864 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006865 return ret;
6866}
6867
Naoya Horiguchi12724852012-03-21 16:34:28 -07006868#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6869/*
6870 * We don't consider swapping or file mapped pages because THP does not
6871 * support them for now.
6872 * Caller should make sure that pmd_trans_huge(pmd) is true.
6873 */
6874static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6875 unsigned long addr, pmd_t pmd, union mc_target *target)
6876{
6877 struct page *page = NULL;
6878 struct page_cgroup *pc;
6879 enum mc_target_type ret = MC_TARGET_NONE;
6880
6881 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08006882 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006883 if (!move_anon())
6884 return ret;
6885 pc = lookup_page_cgroup(page);
6886 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6887 ret = MC_TARGET_PAGE;
6888 if (target) {
6889 get_page(page);
6890 target->page = page;
6891 }
6892 }
6893 return ret;
6894}
6895#else
6896static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6897 unsigned long addr, pmd_t pmd, union mc_target *target)
6898{
6899 return MC_TARGET_NONE;
6900}
6901#endif
6902
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006903static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6904 unsigned long addr, unsigned long end,
6905 struct mm_walk *walk)
6906{
6907 struct vm_area_struct *vma = walk->private;
6908 pte_t *pte;
6909 spinlock_t *ptl;
6910
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006911 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006912 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6913 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006914 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006915 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006916 }
Dave Hansen03319322011-03-22 16:32:56 -07006917
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006918 if (pmd_trans_unstable(pmd))
6919 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006920 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6921 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006922 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006923 mc.precharge++; /* increment precharge temporarily */
6924 pte_unmap_unlock(pte - 1, ptl);
6925 cond_resched();
6926
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006927 return 0;
6928}
6929
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006930static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6931{
6932 unsigned long precharge;
6933 struct vm_area_struct *vma;
6934
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006935 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006936 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6937 struct mm_walk mem_cgroup_count_precharge_walk = {
6938 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6939 .mm = mm,
6940 .private = vma,
6941 };
6942 if (is_vm_hugetlb_page(vma))
6943 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006944 walk_page_range(vma->vm_start, vma->vm_end,
6945 &mem_cgroup_count_precharge_walk);
6946 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006947 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006948
6949 precharge = mc.precharge;
6950 mc.precharge = 0;
6951
6952 return precharge;
6953}
6954
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006955static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6956{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006957 unsigned long precharge = mem_cgroup_count_precharge(mm);
6958
6959 VM_BUG_ON(mc.moving_task);
6960 mc.moving_task = current;
6961 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006962}
6963
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006964/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6965static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006966{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006967 struct mem_cgroup *from = mc.from;
6968 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006969 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006970
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006971 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006972 if (mc.precharge) {
6973 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6974 mc.precharge = 0;
6975 }
6976 /*
6977 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6978 * we must uncharge here.
6979 */
6980 if (mc.moved_charge) {
6981 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6982 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006983 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006984 /* we must fixup refcnts and charges */
6985 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006986 /* uncharge swap account from the old cgroup */
6987 if (!mem_cgroup_is_root(mc.from))
6988 res_counter_uncharge(&mc.from->memsw,
6989 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006990
6991 for (i = 0; i < mc.moved_swap; i++)
6992 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006993
6994 if (!mem_cgroup_is_root(mc.to)) {
6995 /*
6996 * we charged both to->res and to->memsw, so we should
6997 * uncharge to->res.
6998 */
6999 res_counter_uncharge(&mc.to->res,
7000 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007001 }
Li Zefan40503772013-07-08 16:00:34 -07007002 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007003 mc.moved_swap = 0;
7004 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007005 memcg_oom_recover(from);
7006 memcg_oom_recover(to);
7007 wake_up_all(&mc.waitq);
7008}
7009
7010static void mem_cgroup_clear_mc(void)
7011{
7012 struct mem_cgroup *from = mc.from;
7013
7014 /*
7015 * we must clear moving_task before waking up waiters at the end of
7016 * task migration.
7017 */
7018 mc.moving_task = NULL;
7019 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007020 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007021 mc.from = NULL;
7022 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007023 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07007024 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007025}
7026
Tejun Heoeb954192013-08-08 20:11:23 -04007027static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007028 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007029{
Tejun Heo2f7ee562011-12-12 18:12:21 -08007030 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007031 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04007032 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08007033 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007034
Glauber Costaee5e8472013-02-22 16:34:50 -08007035 /*
7036 * We are now commited to this value whatever it is. Changes in this
7037 * tunable will only affect upcoming migrations, not the current one.
7038 * So we need to save it, and keep it going.
7039 */
7040 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
7041 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007042 struct mm_struct *mm;
7043 struct mem_cgroup *from = mem_cgroup_from_task(p);
7044
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07007045 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007046
7047 mm = get_task_mm(p);
7048 if (!mm)
7049 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007050 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007051 if (mm->owner == p) {
7052 VM_BUG_ON(mc.from);
7053 VM_BUG_ON(mc.to);
7054 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007055 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007056 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07007057 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007058 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007059 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07007060 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08007061 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007062 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007063 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007064
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007065 ret = mem_cgroup_precharge_mc(mm);
7066 if (ret)
7067 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007068 }
7069 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007070 }
7071 return ret;
7072}
7073
Tejun Heoeb954192013-08-08 20:11:23 -04007074static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007075 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007076{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007077 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007078}
7079
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007080static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
7081 unsigned long addr, unsigned long end,
7082 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007083{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007084 int ret = 0;
7085 struct vm_area_struct *vma = walk->private;
7086 pte_t *pte;
7087 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07007088 enum mc_target_type target_type;
7089 union mc_target target;
7090 struct page *page;
7091 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007092
Naoya Horiguchi12724852012-03-21 16:34:28 -07007093 /*
7094 * We don't take compound_lock() here but no race with splitting thp
7095 * happens because:
7096 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
7097 * under splitting, which means there's no concurrent thp split,
7098 * - if another thread runs into split_huge_page() just after we
7099 * entered this if-block, the thread must wait for page table lock
7100 * to be unlocked in __split_huge_page_splitting(), where the main
7101 * part of thp split is not executed yet.
7102 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007103 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07007104 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007105 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07007106 return 0;
7107 }
7108 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
7109 if (target_type == MC_TARGET_PAGE) {
7110 page = target.page;
7111 if (!isolate_lru_page(page)) {
7112 pc = lookup_page_cgroup(page);
7113 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07007114 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07007115 mc.precharge -= HPAGE_PMD_NR;
7116 mc.moved_charge += HPAGE_PMD_NR;
7117 }
7118 putback_lru_page(page);
7119 }
7120 put_page(page);
7121 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007122 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07007123 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07007124 }
7125
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07007126 if (pmd_trans_unstable(pmd))
7127 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007128retry:
7129 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
7130 for (; addr != end; addr += PAGE_SIZE) {
7131 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08007132 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007133
7134 if (!mc.precharge)
7135 break;
7136
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007137 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007138 case MC_TARGET_PAGE:
7139 page = target.page;
7140 if (isolate_lru_page(page))
7141 goto put;
7142 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07007143 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07007144 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007145 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007146 /* we uncharge from mc.from later. */
7147 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007148 }
7149 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007150put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007151 put_page(page);
7152 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08007153 case MC_TARGET_SWAP:
7154 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07007155 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08007156 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007157 /* we fixup refcnts and charges later. */
7158 mc.moved_swap++;
7159 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08007160 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007161 default:
7162 break;
7163 }
7164 }
7165 pte_unmap_unlock(pte - 1, ptl);
7166 cond_resched();
7167
7168 if (addr != end) {
7169 /*
7170 * We have consumed all precharges we got in can_attach().
7171 * We try charge one by one, but don't do any additional
7172 * charges to mc.to if we have failed in charge once in attach()
7173 * phase.
7174 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007175 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007176 if (!ret)
7177 goto retry;
7178 }
7179
7180 return ret;
7181}
7182
7183static void mem_cgroup_move_charge(struct mm_struct *mm)
7184{
7185 struct vm_area_struct *vma;
7186
7187 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007188retry:
7189 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
7190 /*
7191 * Someone who are holding the mmap_sem might be waiting in
7192 * waitq. So we cancel all extra charges, wake up all waiters,
7193 * and retry. Because we cancel precharges, we might not be able
7194 * to move enough charges, but moving charge is a best-effort
7195 * feature anyway, so it wouldn't be a big problem.
7196 */
7197 __mem_cgroup_clear_mc();
7198 cond_resched();
7199 goto retry;
7200 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007201 for (vma = mm->mmap; vma; vma = vma->vm_next) {
7202 int ret;
7203 struct mm_walk mem_cgroup_move_charge_walk = {
7204 .pmd_entry = mem_cgroup_move_charge_pte_range,
7205 .mm = mm,
7206 .private = vma,
7207 };
7208 if (is_vm_hugetlb_page(vma))
7209 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007210 ret = walk_page_range(vma->vm_start, vma->vm_end,
7211 &mem_cgroup_move_charge_walk);
7212 if (ret)
7213 /*
7214 * means we have consumed all precharges and failed in
7215 * doing additional charge. Just abandon here.
7216 */
7217 break;
7218 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007219 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007220}
7221
Tejun Heoeb954192013-08-08 20:11:23 -04007222static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007223 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08007224{
Tejun Heo2f7ee562011-12-12 18:12:21 -08007225 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007226 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007227
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007228 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007229 if (mc.to)
7230 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007231 mmput(mm);
7232 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007233 if (mc.to)
7234 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08007235}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007236#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04007237static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007238 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007239{
7240 return 0;
7241}
Tejun Heoeb954192013-08-08 20:11:23 -04007242static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007243 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007244{
7245}
Tejun Heoeb954192013-08-08 20:11:23 -04007246static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007247 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007248{
7249}
7250#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08007251
Tejun Heof00baae2013-04-15 13:41:15 -07007252/*
7253 * Cgroup retains root cgroups across [un]mount cycles making it necessary
7254 * to verify sane_behavior flag on each mount attempt.
7255 */
Tejun Heoeb954192013-08-08 20:11:23 -04007256static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07007257{
7258 /*
7259 * use_hierarchy is forced with sane_behavior. cgroup core
7260 * guarantees that @root doesn't have any children, so turning it
7261 * on for the root memcg is enough.
7262 */
Tejun Heoeb954192013-08-08 20:11:23 -04007263 if (cgroup_sane_behavior(root_css->cgroup))
7264 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07007265}
7266
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007267struct cgroup_subsys mem_cgroup_subsys = {
7268 .name = "memory",
7269 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08007270 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08007271 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08007272 .css_offline = mem_cgroup_css_offline,
7273 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007274 .can_attach = mem_cgroup_can_attach,
7275 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08007276 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07007277 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07007278 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08007279 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007280};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007281
Andrew Mortonc255a452012-07-31 16:43:02 -07007282#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08007283static int __init enable_swap_account(char *s)
7284{
Michal Hockoa2c89902011-05-24 17:12:50 -07007285 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007286 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07007287 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007288 really_do_swap_account = 0;
7289 return 1;
7290}
Michal Hockoa2c89902011-05-24 17:12:50 -07007291__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007292
Michal Hocko2d110852013-02-22 16:34:43 -08007293static void __init memsw_file_init(void)
7294{
Michal Hocko6acc8b02013-02-22 16:34:45 -08007295 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08007296}
Michal Hocko6acc8b02013-02-22 16:34:45 -08007297
7298static void __init enable_swap_cgroup(void)
7299{
7300 if (!mem_cgroup_disabled() && really_do_swap_account) {
7301 do_swap_account = 1;
7302 memsw_file_init();
7303 }
7304}
7305
Michal Hocko2d110852013-02-22 16:34:43 -08007306#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08007307static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08007308{
7309}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007310#endif
Michal Hocko2d110852013-02-22 16:34:43 -08007311
7312/*
Michal Hocko10813122013-02-22 16:35:41 -08007313 * subsys_initcall() for memory controller.
7314 *
7315 * Some parts like hotcpu_notifier() have to be initialized from this context
7316 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7317 * everything that doesn't depend on a specific mem_cgroup structure should
7318 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007319 */
7320static int __init mem_cgroup_init(void)
7321{
7322 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007323 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07007324 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007325 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007326 return 0;
7327}
7328subsys_initcall(mem_cgroup_init);