blob: 29501f04056887297be694c315c7caf3adf666f5 [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
Tejun Heo073219e2014-02-08 10:36:58 -050069struct cgroup_subsys memory_cgrp_subsys __read_mostly;
70EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080071
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
Tejun Heo073219e2014-02-08 10:36:58 -0500541 css = css_from_id(id - 1, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800542 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 Hiroyukib24028572012-03-21 16:34:22 -0700924 /*
925 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
926 * counted as CACHE even if it's on ANON LRU.
927 */
928 if (anon)
929 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700930 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800931 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700932 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700933 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700934
David Rientjesb070e652013-05-07 16:18:09 -0700935 if (PageTransHuge(page))
936 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
937 nr_pages);
938
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800939 /* pagein of a big page is an event. So, ignore page size */
940 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700941 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800942 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700943 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800944 nr_pages = -nr_pages; /* for event */
945 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800946
Johannes Weiner13114712012-05-29 15:07:07 -0700947 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800948}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800949
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700950unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700951mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700952{
953 struct mem_cgroup_per_zone *mz;
954
955 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
956 return mz->lru_size[lru];
957}
958
959static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700960mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700961 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700962{
963 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700964 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700965 unsigned long ret = 0;
966
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700967 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700968
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700969 for_each_lru(lru) {
970 if (BIT(lru) & lru_mask)
971 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700972 }
973 return ret;
974}
975
976static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700977mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700978 int nid, unsigned int lru_mask)
979{
Ying Han889976d2011-05-26 16:25:33 -0700980 u64 total = 0;
981 int zid;
982
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700983 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700984 total += mem_cgroup_zone_nr_lru_pages(memcg,
985 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700986
Ying Han889976d2011-05-26 16:25:33 -0700987 return total;
988}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700989
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700990static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700991 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800992{
Ying Han889976d2011-05-26 16:25:33 -0700993 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800994 u64 total = 0;
995
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800996 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700997 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800998 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800999}
1000
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001001static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
1002 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001003{
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001004 unsigned long val, next;
1005
Johannes Weiner13114712012-05-29 15:07:07 -07001006 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -07001007 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001008 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001009 if ((long)next - (long)val < 0) {
1010 switch (target) {
1011 case MEM_CGROUP_TARGET_THRESH:
1012 next = val + THRESHOLDS_EVENTS_TARGET;
1013 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001014 case MEM_CGROUP_TARGET_SOFTLIMIT:
1015 next = val + SOFTLIMIT_EVENTS_TARGET;
1016 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001017 case MEM_CGROUP_TARGET_NUMAINFO:
1018 next = val + NUMAINFO_EVENTS_TARGET;
1019 break;
1020 default:
1021 break;
1022 }
1023 __this_cpu_write(memcg->stat->targets[target], next);
1024 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001025 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001026 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001027}
1028
1029/*
1030 * Check events in order.
1031 *
1032 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001033static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001034{
Steven Rostedt47994012011-11-02 13:38:33 -07001035 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001036 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001037 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1038 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001039 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001040 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001041
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001042 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1043 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001044#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001045 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1046 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001047#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001048 preempt_enable();
1049
1050 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001051 if (unlikely(do_softlimit))
1052 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001053#if MAX_NUMNODES > 1
1054 if (unlikely(do_numainfo))
1055 atomic_inc(&memcg->numainfo_events);
1056#endif
1057 } else
1058 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001059}
1060
Balbir Singhcf475ad2008-04-29 01:00:16 -07001061struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001062{
Balbir Singh31a78f22008-09-28 23:09:31 +01001063 /*
1064 * mm_update_next_owner() may clear mm->owner to NULL
1065 * if it races with swapoff, page migration, etc.
1066 * So this can be called with p == NULL.
1067 */
1068 if (unlikely(!p))
1069 return NULL;
1070
Tejun Heo073219e2014-02-08 10:36:58 -05001071 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001072}
1073
Johannes Weinerdf381972014-04-07 15:37:43 -07001074static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001075{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001076 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001077
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001078 rcu_read_lock();
1079 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001080 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1081 if (unlikely(!memcg))
Johannes Weinerdf381972014-04-07 15:37:43 -07001082 memcg = root_mem_cgroup;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001083 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001084 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001085 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001086}
1087
Michal Hocko16248d82013-04-29 15:07:19 -07001088/*
1089 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1090 * ref. count) or NULL if the whole root's subtree has been visited.
1091 *
1092 * helper function to be used by mem_cgroup_iter
1093 */
1094static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001095 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001096{
Tejun Heo492eb212013-08-08 20:11:25 -04001097 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001098
Tejun Heobd8815a2013-08-08 20:11:27 -04001099 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001100skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001101 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001102
1103 /*
1104 * Even if we found a group we have to make sure it is
1105 * alive. css && !memcg means that the groups should be
1106 * skipped and we should continue the tree walk.
1107 * last_visited css is safe to use because it is
1108 * protected by css_get and the tree walk is rcu safe.
Michal Hocko0eef6152014-01-23 15:53:37 -08001109 *
1110 * We do not take a reference on the root of the tree walk
1111 * because we might race with the root removal when it would
1112 * be the only node in the iterated hierarchy and mem_cgroup_iter
1113 * would end up in an endless loop because it expects that at
1114 * least one valid node will be returned. Root cannot disappear
1115 * because caller of the iterator should hold it already so
1116 * skipping css reference should be safe.
Michal Hocko16248d82013-04-29 15:07:19 -07001117 */
Tejun Heo492eb212013-08-08 20:11:25 -04001118 if (next_css) {
Hugh Dickinsce482252014-03-03 15:38:24 -08001119 if ((next_css == &root->css) ||
1120 ((next_css->flags & CSS_ONLINE) && css_tryget(next_css)))
Hugh Dickinsd8ad3052014-01-23 15:53:32 -08001121 return mem_cgroup_from_css(next_css);
Michal Hocko0eef6152014-01-23 15:53:37 -08001122
1123 prev_css = next_css;
1124 goto skip_node;
Michal Hocko16248d82013-04-29 15:07:19 -07001125 }
1126
1127 return NULL;
1128}
1129
Johannes Weiner519ebea2013-07-03 15:04:51 -07001130static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1131{
1132 /*
1133 * When a group in the hierarchy below root is destroyed, the
1134 * hierarchy iterator can no longer be trusted since it might
1135 * have pointed to the destroyed group. Invalidate it.
1136 */
1137 atomic_inc(&root->dead_count);
1138}
1139
1140static struct mem_cgroup *
1141mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1142 struct mem_cgroup *root,
1143 int *sequence)
1144{
1145 struct mem_cgroup *position = NULL;
1146 /*
1147 * A cgroup destruction happens in two stages: offlining and
1148 * release. They are separated by a RCU grace period.
1149 *
1150 * If the iterator is valid, we may still race with an
1151 * offlining. The RCU lock ensures the object won't be
1152 * released, tryget will fail if we lost the race.
1153 */
1154 *sequence = atomic_read(&root->dead_count);
1155 if (iter->last_dead_count == *sequence) {
1156 smp_rmb();
1157 position = iter->last_visited;
Michal Hockoecc736f2014-01-23 15:53:35 -08001158
1159 /*
1160 * We cannot take a reference to root because we might race
1161 * with root removal and returning NULL would end up in
1162 * an endless loop on the iterator user level when root
1163 * would be returned all the time.
1164 */
1165 if (position && position != root &&
1166 !css_tryget(&position->css))
Johannes Weiner519ebea2013-07-03 15:04:51 -07001167 position = NULL;
1168 }
1169 return position;
1170}
1171
1172static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1173 struct mem_cgroup *last_visited,
1174 struct mem_cgroup *new_position,
Michal Hockoecc736f2014-01-23 15:53:35 -08001175 struct mem_cgroup *root,
Johannes Weiner519ebea2013-07-03 15:04:51 -07001176 int sequence)
1177{
Michal Hockoecc736f2014-01-23 15:53:35 -08001178 /* root reference counting symmetric to mem_cgroup_iter_load */
1179 if (last_visited && last_visited != root)
Johannes Weiner519ebea2013-07-03 15:04:51 -07001180 css_put(&last_visited->css);
1181 /*
1182 * We store the sequence count from the time @last_visited was
1183 * loaded successfully instead of rereading it here so that we
1184 * don't lose destruction events in between. We could have
1185 * raced with the destruction of @new_position after all.
1186 */
1187 iter->last_visited = new_position;
1188 smp_wmb();
1189 iter->last_dead_count = sequence;
1190}
1191
Johannes Weiner56600482012-01-12 17:17:59 -08001192/**
1193 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1194 * @root: hierarchy root
1195 * @prev: previously returned memcg, NULL on first invocation
1196 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1197 *
1198 * Returns references to children of the hierarchy below @root, or
1199 * @root itself, or %NULL after a full round-trip.
1200 *
1201 * Caller must pass the return value in @prev on subsequent
1202 * invocations for reference counting, or use mem_cgroup_iter_break()
1203 * to cancel a hierarchy walk before the round-trip is complete.
1204 *
1205 * Reclaimers can specify a zone and a priority level in @reclaim to
1206 * divide up the memcgs in the hierarchy among all concurrent
1207 * reclaimers operating on the same zone and priority.
1208 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001209struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001210 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001211 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001212{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001213 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001214 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001215
Andrew Morton694fbc02013-09-24 15:27:37 -07001216 if (mem_cgroup_disabled())
1217 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001218
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001219 if (!root)
1220 root = root_mem_cgroup;
1221
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001222 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001223 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001224
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001225 if (!root->use_hierarchy && root != root_mem_cgroup) {
1226 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001227 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001228 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001229 }
1230
Michal Hocko542f85f2013-04-29 15:07:15 -07001231 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001232 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001233 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001234 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001235
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001236 if (reclaim) {
1237 int nid = zone_to_nid(reclaim->zone);
1238 int zid = zone_idx(reclaim->zone);
1239 struct mem_cgroup_per_zone *mz;
1240
1241 mz = mem_cgroup_zoneinfo(root, nid, zid);
1242 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001243 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001244 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001245 goto out_unlock;
1246 }
Michal Hocko5f578162013-04-29 15:07:17 -07001247
Johannes Weiner519ebea2013-07-03 15:04:51 -07001248 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001249 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001250
Andrew Morton694fbc02013-09-24 15:27:37 -07001251 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001252
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001253 if (reclaim) {
Michal Hockoecc736f2014-01-23 15:53:35 -08001254 mem_cgroup_iter_update(iter, last_visited, memcg, root,
1255 seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001256
Michal Hocko19f39402013-04-29 15:07:18 -07001257 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001258 iter->generation++;
1259 else if (!prev && memcg)
1260 reclaim->generation = iter->generation;
1261 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001262
Andrew Morton694fbc02013-09-24 15:27:37 -07001263 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001264 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001265 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001266out_unlock:
1267 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001268out_css_put:
1269 if (prev && prev != root)
1270 css_put(&prev->css);
1271
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001272 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001273}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001274
Johannes Weiner56600482012-01-12 17:17:59 -08001275/**
1276 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1277 * @root: hierarchy root
1278 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1279 */
1280void mem_cgroup_iter_break(struct mem_cgroup *root,
1281 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001282{
1283 if (!root)
1284 root = root_mem_cgroup;
1285 if (prev && prev != root)
1286 css_put(&prev->css);
1287}
1288
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001289/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001290 * Iteration constructs for visiting all cgroups (under a tree). If
1291 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1292 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001293 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001294#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001295 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001296 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001297 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001298
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001299#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001300 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001301 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001302 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001303
David Rientjes68ae5642012-12-12 13:51:57 -08001304void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001305{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001306 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001307
Ying Han456f9982011-05-26 16:25:38 -07001308 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001309 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1310 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001311 goto out;
1312
1313 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001314 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001315 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1316 break;
1317 case PGMAJFAULT:
1318 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001319 break;
1320 default:
1321 BUG();
1322 }
1323out:
1324 rcu_read_unlock();
1325}
David Rientjes68ae5642012-12-12 13:51:57 -08001326EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001327
Johannes Weiner925b7672012-01-12 17:18:15 -08001328/**
1329 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1330 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001331 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001332 *
1333 * Returns the lru list vector holding pages for the given @zone and
1334 * @mem. This can be the global zone lruvec, if the memory controller
1335 * is disabled.
1336 */
1337struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1338 struct mem_cgroup *memcg)
1339{
1340 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001341 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001342
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001343 if (mem_cgroup_disabled()) {
1344 lruvec = &zone->lruvec;
1345 goto out;
1346 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001347
1348 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001349 lruvec = &mz->lruvec;
1350out:
1351 /*
1352 * Since a node can be onlined after the mem_cgroup was created,
1353 * we have to be prepared to initialize lruvec->zone here;
1354 * and if offlined then reonlined, we need to reinitialize it.
1355 */
1356 if (unlikely(lruvec->zone != zone))
1357 lruvec->zone = zone;
1358 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001359}
1360
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001361/*
1362 * Following LRU functions are allowed to be used without PCG_LOCK.
1363 * Operations are called by routine of global LRU independently from memcg.
1364 * What we have to take care of here is validness of pc->mem_cgroup.
1365 *
1366 * Changes to pc->mem_cgroup happens when
1367 * 1. charge
1368 * 2. moving account
1369 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1370 * It is added to LRU before charge.
1371 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1372 * When moving account, the page is not on LRU. It's isolated.
1373 */
1374
Johannes Weiner925b7672012-01-12 17:18:15 -08001375/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001376 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001377 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001378 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001379 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001380struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001381{
1382 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001383 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001384 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001385 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001386
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001387 if (mem_cgroup_disabled()) {
1388 lruvec = &zone->lruvec;
1389 goto out;
1390 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001391
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001392 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001393 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001394
1395 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001396 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001397 * an uncharged page off lru does nothing to secure
1398 * its former mem_cgroup from sudden removal.
1399 *
1400 * Our caller holds lru_lock, and PageCgroupUsed is updated
1401 * under page_cgroup lock: between them, they make all uses
1402 * of pc->mem_cgroup safe.
1403 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001404 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001405 pc->mem_cgroup = memcg = root_mem_cgroup;
1406
Johannes Weiner925b7672012-01-12 17:18:15 -08001407 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001408 lruvec = &mz->lruvec;
1409out:
1410 /*
1411 * Since a node can be onlined after the mem_cgroup was created,
1412 * we have to be prepared to initialize lruvec->zone here;
1413 * and if offlined then reonlined, we need to reinitialize it.
1414 */
1415 if (unlikely(lruvec->zone != zone))
1416 lruvec->zone = zone;
1417 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001418}
1419
1420/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001421 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1422 * @lruvec: mem_cgroup per zone lru vector
1423 * @lru: index of lru list the page is sitting on
1424 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001425 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001426 * This function must be called when a page is added to or removed from an
1427 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001428 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001429void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1430 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001431{
1432 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001433 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001434
1435 if (mem_cgroup_disabled())
1436 return;
1437
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001438 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1439 lru_size = mz->lru_size + lru;
1440 *lru_size += nr_pages;
1441 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001442}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001443
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001444/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001445 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001446 * hierarchy subtree
1447 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001448bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1449 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001450{
Johannes Weiner91c637342012-05-29 15:06:24 -07001451 if (root_memcg == memcg)
1452 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001453 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001454 return false;
Li Zefanb47f77b2013-09-23 16:55:43 +08001455 return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001456}
1457
1458static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1459 struct mem_cgroup *memcg)
1460{
1461 bool ret;
1462
Johannes Weiner91c637342012-05-29 15:06:24 -07001463 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001464 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001465 rcu_read_unlock();
1466 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001467}
1468
David Rientjesffbdccf2013-07-03 15:01:23 -07001469bool task_in_mem_cgroup(struct task_struct *task,
1470 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001471{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001472 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001473 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001474 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001475
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001476 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001477 if (p) {
Johannes Weinerdf381972014-04-07 15:37:43 -07001478 curr = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001479 task_unlock(p);
1480 } else {
1481 /*
1482 * All threads may have already detached their mm's, but the oom
1483 * killer still needs to detect if they have already been oom
1484 * killed to prevent needlessly killing additional tasks.
1485 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001486 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001487 curr = mem_cgroup_from_task(task);
1488 if (curr)
1489 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001490 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001491 }
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001492 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001493 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001494 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001495 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1496 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001497 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001498 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001499 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001500 return ret;
1501}
1502
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001503int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001504{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001505 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001506 unsigned long inactive;
1507 unsigned long active;
1508 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001509
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001510 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1511 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001512
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001513 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1514 if (gb)
1515 inactive_ratio = int_sqrt(10 * gb);
1516 else
1517 inactive_ratio = 1;
1518
Johannes Weiner9b272972011-11-02 13:38:23 -07001519 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001520}
1521
Balbir Singh6d61ef42009-01-07 18:08:06 -08001522#define mem_cgroup_from_res_counter(counter, member) \
1523 container_of(counter, struct mem_cgroup, member)
1524
Johannes Weiner19942822011-02-01 15:52:43 -08001525/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001526 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001527 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001528 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001529 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001530 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001531 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001532static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001533{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001534 unsigned long long margin;
1535
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001536 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001537 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001538 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001539 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001540}
1541
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001542int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001543{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001544 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001545 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001546 return vm_swappiness;
1547
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001548 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001549}
1550
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001551/*
1552 * memcg->moving_account is used for checking possibility that some thread is
1553 * calling move_account(). When a thread on CPU-A starts moving pages under
1554 * a memcg, other threads should check memcg->moving_account under
1555 * rcu_read_lock(), like this:
1556 *
1557 * CPU-A CPU-B
1558 * rcu_read_lock()
1559 * memcg->moving_account+1 if (memcg->mocing_account)
1560 * take heavy locks.
1561 * synchronize_rcu() update something.
1562 * rcu_read_unlock()
1563 * start move here.
1564 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001565
1566/* for quick checking without looking up memcg */
1567atomic_t memcg_moving __read_mostly;
1568
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001569static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001570{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001571 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001572 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001573 synchronize_rcu();
1574}
1575
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001576static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001577{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001578 /*
1579 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1580 * We check NULL in callee rather than caller.
1581 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001582 if (memcg) {
1583 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001584 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001585 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001586}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001587
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001588/*
1589 * 2 routines for checking "mem" is under move_account() or not.
1590 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001591 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1592 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001593 * pc->mem_cgroup may be overwritten.
1594 *
1595 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1596 * under hierarchy of moving cgroups. This is for
1597 * waiting at hith-memory prressure caused by "move".
1598 */
1599
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001600static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001601{
1602 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001603 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001604}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001605
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001606static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001607{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001608 struct mem_cgroup *from;
1609 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001610 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001611 /*
1612 * Unlike task_move routines, we access mc.to, mc.from not under
1613 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1614 */
1615 spin_lock(&mc.lock);
1616 from = mc.from;
1617 to = mc.to;
1618 if (!from)
1619 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001620
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001621 ret = mem_cgroup_same_or_subtree(memcg, from)
1622 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001623unlock:
1624 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001625 return ret;
1626}
1627
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001628static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001629{
1630 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001631 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001632 DEFINE_WAIT(wait);
1633 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1634 /* moving charge context might have finished. */
1635 if (mc.moving_task)
1636 schedule();
1637 finish_wait(&mc.waitq, &wait);
1638 return true;
1639 }
1640 }
1641 return false;
1642}
1643
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001644/*
1645 * Take this lock when
1646 * - a code tries to modify page's memcg while it's USED.
1647 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001648 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001649 */
1650static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1651 unsigned long *flags)
1652{
1653 spin_lock_irqsave(&memcg->move_lock, *flags);
1654}
1655
1656static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1657 unsigned long *flags)
1658{
1659 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1660}
1661
Sha Zhengju58cf1882013-02-22 16:32:05 -08001662#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001663/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001664 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001665 * @memcg: The memory cgroup that went over limit
1666 * @p: Task that is going to be killed
1667 *
1668 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1669 * enabled
1670 */
1671void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1672{
Tejun Heoe61734c2014-02-12 09:29:50 -05001673 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001674 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001675 struct mem_cgroup *iter;
1676 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001677
Sha Zhengju58cf1882013-02-22 16:32:05 -08001678 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001679 return;
1680
Michal Hocko08088cb2014-02-25 15:01:44 -08001681 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001682 rcu_read_lock();
1683
Tejun Heoe61734c2014-02-12 09:29:50 -05001684 pr_info("Task in ");
1685 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1686 pr_info(" killed as a result of limit of ");
1687 pr_cont_cgroup_path(memcg->css.cgroup);
1688 pr_info("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001689
Balbir Singhe2224322009-04-02 16:57:39 -07001690 rcu_read_unlock();
1691
Andrew Mortond0451972013-02-22 16:32:06 -08001692 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001693 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1694 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1695 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001696 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001697 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1698 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1699 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001700 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001701 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1702 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1703 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001704
1705 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001706 pr_info("Memory cgroup stats for ");
1707 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001708 pr_cont(":");
1709
1710 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1711 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1712 continue;
1713 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1714 K(mem_cgroup_read_stat(iter, i)));
1715 }
1716
1717 for (i = 0; i < NR_LRU_LISTS; i++)
1718 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1719 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1720
1721 pr_cont("\n");
1722 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001723 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001724}
1725
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001726/*
1727 * This function returns the number of memcg under hierarchy tree. Returns
1728 * 1(self count) if no children.
1729 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001730static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001731{
1732 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001733 struct mem_cgroup *iter;
1734
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001735 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001736 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001737 return num;
1738}
1739
Balbir Singh6d61ef42009-01-07 18:08:06 -08001740/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001741 * Return the memory (and swap, if configured) limit for a memcg.
1742 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001743static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001744{
1745 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001746
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001747 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001748
David Rientjesa63d83f2010-08-09 17:19:46 -07001749 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001750 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001751 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001752 if (mem_cgroup_swappiness(memcg)) {
1753 u64 memsw;
1754
1755 limit += total_swap_pages << PAGE_SHIFT;
1756 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1757
1758 /*
1759 * If memsw is finite and limits the amount of swap space
1760 * available to this memcg, return that limit.
1761 */
1762 limit = min(limit, memsw);
1763 }
1764
1765 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001766}
1767
David Rientjes19965462012-12-11 16:00:26 -08001768static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1769 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001770{
1771 struct mem_cgroup *iter;
1772 unsigned long chosen_points = 0;
1773 unsigned long totalpages;
1774 unsigned int points = 0;
1775 struct task_struct *chosen = NULL;
1776
David Rientjes876aafb2012-07-31 16:43:48 -07001777 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001778 * If current has a pending SIGKILL or is exiting, then automatically
1779 * select it. The goal is to allow it to allocate so that it may
1780 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001781 */
David Rientjes465adcf2013-04-29 15:08:45 -07001782 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001783 set_thread_flag(TIF_MEMDIE);
1784 return;
1785 }
1786
1787 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001788 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1789 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001790 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001791 struct task_struct *task;
1792
Tejun Heo72ec7022013-08-08 20:11:26 -04001793 css_task_iter_start(&iter->css, &it);
1794 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001795 switch (oom_scan_process_thread(task, totalpages, NULL,
1796 false)) {
1797 case OOM_SCAN_SELECT:
1798 if (chosen)
1799 put_task_struct(chosen);
1800 chosen = task;
1801 chosen_points = ULONG_MAX;
1802 get_task_struct(chosen);
1803 /* fall through */
1804 case OOM_SCAN_CONTINUE:
1805 continue;
1806 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001807 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001808 mem_cgroup_iter_break(memcg, iter);
1809 if (chosen)
1810 put_task_struct(chosen);
1811 return;
1812 case OOM_SCAN_OK:
1813 break;
1814 };
1815 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001816 if (!points || points < chosen_points)
1817 continue;
1818 /* Prefer thread group leaders for display purposes */
1819 if (points == chosen_points &&
1820 thread_group_leader(chosen))
1821 continue;
1822
1823 if (chosen)
1824 put_task_struct(chosen);
1825 chosen = task;
1826 chosen_points = points;
1827 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001828 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001829 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001830 }
1831
1832 if (!chosen)
1833 return;
1834 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001835 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1836 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001837}
1838
Johannes Weiner56600482012-01-12 17:17:59 -08001839static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1840 gfp_t gfp_mask,
1841 unsigned long flags)
1842{
1843 unsigned long total = 0;
1844 bool noswap = false;
1845 int loop;
1846
1847 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1848 noswap = true;
1849 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1850 noswap = true;
1851
1852 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1853 if (loop)
1854 drain_all_stock_async(memcg);
1855 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1856 /*
1857 * Allow limit shrinkers, which are triggered directly
1858 * by userspace, to catch signals and stop reclaim
1859 * after minimal progress, regardless of the margin.
1860 */
1861 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1862 break;
1863 if (mem_cgroup_margin(memcg))
1864 break;
1865 /*
1866 * If nothing was reclaimed after two attempts, there
1867 * may be no reclaimable pages in this hierarchy.
1868 */
1869 if (loop && !total)
1870 break;
1871 }
1872 return total;
1873}
1874
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001875/**
1876 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001877 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001878 * @nid: the node ID to be checked.
1879 * @noswap : specify true here if the user wants flle only information.
1880 *
1881 * This function returns whether the specified memcg contains any
1882 * reclaimable pages on a node. Returns true if there are any reclaimable
1883 * pages in the node.
1884 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001885static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001886 int nid, bool noswap)
1887{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001888 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001889 return true;
1890 if (noswap || !total_swap_pages)
1891 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001893 return true;
1894 return false;
1895
1896}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001897#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001898
1899/*
1900 * Always updating the nodemask is not very good - even if we have an empty
1901 * list or the wrong list here, we can start from some node and traverse all
1902 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1903 *
1904 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001905static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001906{
1907 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001908 /*
1909 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1910 * pagein/pageout changes since the last update.
1911 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001912 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001913 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001914 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001915 return;
1916
Ying Han889976d2011-05-26 16:25:33 -07001917 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001918 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001919
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001920 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001921
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001922 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1923 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001924 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001925
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926 atomic_set(&memcg->numainfo_events, 0);
1927 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001928}
1929
1930/*
1931 * Selecting a node where we start reclaim from. Because what we need is just
1932 * reducing usage counter, start from anywhere is O,K. Considering
1933 * memory reclaim from current node, there are pros. and cons.
1934 *
1935 * Freeing memory from current node means freeing memory from a node which
1936 * we'll use or we've used. So, it may make LRU bad. And if several threads
1937 * hit limits, it will see a contention on a node. But freeing from remote
1938 * node means more costs for memory reclaim because of memory latency.
1939 *
1940 * Now, we use round-robin. Better algorithm is welcomed.
1941 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001942int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001943{
1944 int node;
1945
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001946 mem_cgroup_may_update_nodemask(memcg);
1947 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001948
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001949 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001950 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001951 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001952 /*
1953 * We call this when we hit limit, not when pages are added to LRU.
1954 * No LRU may hold pages because all pages are UNEVICTABLE or
1955 * memcg is too small and all pages are not on LRU. In that case,
1956 * we use curret node.
1957 */
1958 if (unlikely(node == MAX_NUMNODES))
1959 node = numa_node_id();
1960
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001961 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001962 return node;
1963}
1964
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001965/*
1966 * Check all nodes whether it contains reclaimable pages or not.
1967 * For quick scan, we make use of scan_nodes. This will allow us to skip
1968 * unused nodes. But scan_nodes is lazily updated and may not cotain
1969 * enough new information. We need to do double check.
1970 */
1971static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1972{
1973 int nid;
1974
1975 /*
1976 * quick check...making use of scan_node.
1977 * We can skip unused nodes.
1978 */
1979 if (!nodes_empty(memcg->scan_nodes)) {
1980 for (nid = first_node(memcg->scan_nodes);
1981 nid < MAX_NUMNODES;
1982 nid = next_node(nid, memcg->scan_nodes)) {
1983
1984 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1985 return true;
1986 }
1987 }
1988 /*
1989 * Check rest of nodes.
1990 */
1991 for_each_node_state(nid, N_MEMORY) {
1992 if (node_isset(nid, memcg->scan_nodes))
1993 continue;
1994 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1995 return true;
1996 }
1997 return false;
1998}
1999
Ying Han889976d2011-05-26 16:25:33 -07002000#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002001int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002002{
2003 return 0;
2004}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002005
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002006static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
2007{
2008 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
2009}
Ying Han889976d2011-05-26 16:25:33 -07002010#endif
2011
Andrew Morton0608f432013-09-24 15:27:41 -07002012static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2013 struct zone *zone,
2014 gfp_t gfp_mask,
2015 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002016{
Andrew Morton0608f432013-09-24 15:27:41 -07002017 struct mem_cgroup *victim = NULL;
2018 int total = 0;
2019 int loop = 0;
2020 unsigned long excess;
2021 unsigned long nr_scanned;
2022 struct mem_cgroup_reclaim_cookie reclaim = {
2023 .zone = zone,
2024 .priority = 0,
2025 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002026
Andrew Morton0608f432013-09-24 15:27:41 -07002027 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002028
Andrew Morton0608f432013-09-24 15:27:41 -07002029 while (1) {
2030 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
2031 if (!victim) {
2032 loop++;
2033 if (loop >= 2) {
2034 /*
2035 * If we have not been able to reclaim
2036 * anything, it might because there are
2037 * no reclaimable pages under this hierarchy
2038 */
2039 if (!total)
2040 break;
2041 /*
2042 * We want to do more targeted reclaim.
2043 * excess >> 2 is not to excessive so as to
2044 * reclaim too much, nor too less that we keep
2045 * coming back to reclaim from this cgroup
2046 */
2047 if (total >= (excess >> 2) ||
2048 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
2049 break;
2050 }
2051 continue;
2052 }
2053 if (!mem_cgroup_reclaimable(victim, false))
2054 continue;
2055 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2056 zone, &nr_scanned);
2057 *total_scanned += nr_scanned;
2058 if (!res_counter_soft_limit_excess(&root_memcg->res))
2059 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002060 }
Andrew Morton0608f432013-09-24 15:27:41 -07002061 mem_cgroup_iter_break(root_memcg, victim);
2062 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002063}
2064
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002065#ifdef CONFIG_LOCKDEP
2066static struct lockdep_map memcg_oom_lock_dep_map = {
2067 .name = "memcg_oom_lock",
2068};
2069#endif
2070
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002071static DEFINE_SPINLOCK(memcg_oom_lock);
2072
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002073/*
2074 * Check OOM-Killer is already running under our hierarchy.
2075 * If someone is running, return false.
2076 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002077static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002078{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002079 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002080
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002081 spin_lock(&memcg_oom_lock);
2082
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002083 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002084 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002085 /*
2086 * this subtree of our hierarchy is already locked
2087 * so we cannot give a lock.
2088 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002089 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002090 mem_cgroup_iter_break(memcg, iter);
2091 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002092 } else
2093 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002094 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002095
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002096 if (failed) {
2097 /*
2098 * OK, we failed to lock the whole subtree so we have
2099 * to clean up what we set up to the failing subtree
2100 */
2101 for_each_mem_cgroup_tree(iter, memcg) {
2102 if (iter == failed) {
2103 mem_cgroup_iter_break(memcg, iter);
2104 break;
2105 }
2106 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002107 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002108 } else
2109 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002110
2111 spin_unlock(&memcg_oom_lock);
2112
2113 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002114}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002115
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002116static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002117{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002118 struct mem_cgroup *iter;
2119
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002120 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002121 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002122 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002123 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002124 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002125}
2126
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002127static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002128{
2129 struct mem_cgroup *iter;
2130
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002131 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002132 atomic_inc(&iter->under_oom);
2133}
2134
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002136{
2137 struct mem_cgroup *iter;
2138
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002139 /*
2140 * When a new child is created while the hierarchy is under oom,
2141 * mem_cgroup_oom_lock() may not be called. We have to use
2142 * atomic_add_unless() here.
2143 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002144 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002145 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002146}
2147
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002148static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2149
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002150struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002151 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002152 wait_queue_t wait;
2153};
2154
2155static int memcg_oom_wake_function(wait_queue_t *wait,
2156 unsigned mode, int sync, void *arg)
2157{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002158 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2159 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002160 struct oom_wait_info *oom_wait_info;
2161
2162 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002163 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002164
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002165 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002166 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002167 * Then we can use css_is_ancestor without taking care of RCU.
2168 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002169 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2170 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002171 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002172 return autoremove_wake_function(wait, mode, sync, arg);
2173}
2174
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002175static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002176{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002177 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002178 /* for filtering, pass "memcg" as argument. */
2179 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002180}
2181
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002182static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002183{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002184 if (memcg && atomic_read(&memcg->under_oom))
2185 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002186}
2187
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002188static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002189{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002190 if (!current->memcg_oom.may_oom)
2191 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002192 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002193 * We are in the middle of the charge context here, so we
2194 * don't want to block when potentially sitting on a callstack
2195 * that holds all kinds of filesystem and mm locks.
2196 *
2197 * Also, the caller may handle a failed allocation gracefully
2198 * (like optional page cache readahead) and so an OOM killer
2199 * invocation might not even be necessary.
2200 *
2201 * That's why we don't do anything here except remember the
2202 * OOM context and then deal with it at the end of the page
2203 * fault when the stack is unwound, the locks are released,
2204 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002205 */
Johannes Weiner49426422013-10-16 13:46:59 -07002206 css_get(&memcg->css);
2207 current->memcg_oom.memcg = memcg;
2208 current->memcg_oom.gfp_mask = mask;
2209 current->memcg_oom.order = order;
2210}
2211
2212/**
2213 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2214 * @handle: actually kill/wait or just clean up the OOM state
2215 *
2216 * This has to be called at the end of a page fault if the memcg OOM
2217 * handler was enabled.
2218 *
2219 * Memcg supports userspace OOM handling where failed allocations must
2220 * sleep on a waitqueue until the userspace task resolves the
2221 * situation. Sleeping directly in the charge context with all kinds
2222 * of locks held is not a good idea, instead we remember an OOM state
2223 * in the task and mem_cgroup_oom_synchronize() has to be called at
2224 * the end of the page fault to complete the OOM handling.
2225 *
2226 * Returns %true if an ongoing memcg OOM situation was detected and
2227 * completed, %false otherwise.
2228 */
2229bool mem_cgroup_oom_synchronize(bool handle)
2230{
2231 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2232 struct oom_wait_info owait;
2233 bool locked;
2234
2235 /* OOM is global, do not handle */
2236 if (!memcg)
2237 return false;
2238
2239 if (!handle)
2240 goto cleanup;
2241
2242 owait.memcg = memcg;
2243 owait.wait.flags = 0;
2244 owait.wait.func = memcg_oom_wake_function;
2245 owait.wait.private = current;
2246 INIT_LIST_HEAD(&owait.wait.task_list);
2247
2248 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002249 mem_cgroup_mark_under_oom(memcg);
2250
2251 locked = mem_cgroup_oom_trylock(memcg);
2252
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002253 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002254 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002255
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002256 if (locked && !memcg->oom_kill_disable) {
2257 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002258 finish_wait(&memcg_oom_waitq, &owait.wait);
2259 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2260 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002261 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002262 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002263 mem_cgroup_unmark_under_oom(memcg);
2264 finish_wait(&memcg_oom_waitq, &owait.wait);
2265 }
2266
2267 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002268 mem_cgroup_oom_unlock(memcg);
2269 /*
2270 * There is no guarantee that an OOM-lock contender
2271 * sees the wakeups triggered by the OOM kill
2272 * uncharges. Wake any sleepers explicitely.
2273 */
2274 memcg_oom_recover(memcg);
2275 }
Johannes Weiner49426422013-10-16 13:46:59 -07002276cleanup:
2277 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002278 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002279 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002280}
2281
Balbir Singhd69b0422009-06-17 16:26:34 -07002282/*
2283 * Currently used to update mapped file statistics, but the routine can be
2284 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002285 *
2286 * Notes: Race condition
2287 *
2288 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2289 * it tends to be costly. But considering some conditions, we doesn't need
2290 * to do so _always_.
2291 *
2292 * Considering "charge", lock_page_cgroup() is not required because all
2293 * file-stat operations happen after a page is attached to radix-tree. There
2294 * are no race with "charge".
2295 *
2296 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2297 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2298 * if there are race with "uncharge". Statistics itself is properly handled
2299 * by flags.
2300 *
2301 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002302 * small, we check mm->moving_account and detect there are possibility of race
2303 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002304 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002305
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002306void __mem_cgroup_begin_update_page_stat(struct page *page,
2307 bool *locked, unsigned long *flags)
2308{
2309 struct mem_cgroup *memcg;
2310 struct page_cgroup *pc;
2311
2312 pc = lookup_page_cgroup(page);
2313again:
2314 memcg = pc->mem_cgroup;
2315 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2316 return;
2317 /*
2318 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002319 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002320 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002321 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002322 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002323 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002324 return;
2325
2326 move_lock_mem_cgroup(memcg, flags);
2327 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2328 move_unlock_mem_cgroup(memcg, flags);
2329 goto again;
2330 }
2331 *locked = true;
2332}
2333
2334void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2335{
2336 struct page_cgroup *pc = lookup_page_cgroup(page);
2337
2338 /*
2339 * It's guaranteed that pc->mem_cgroup never changes while
2340 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002341 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002342 */
2343 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2344}
2345
Greg Thelen2a7106f2011-01-13 15:47:37 -08002346void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002347 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002348{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002349 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002350 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002351 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002352
Johannes Weinercfa44942012-01-12 17:18:38 -08002353 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002354 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002355
Sha Zhengju658b72c2013-09-12 15:13:52 -07002356 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002357 memcg = pc->mem_cgroup;
2358 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002359 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002360
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002361 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002362}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002363
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002364/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002365 * size of first charge trial. "32" comes from vmscan.c's magic value.
2366 * TODO: maybe necessary to use big numbers in big irons.
2367 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002368#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369struct memcg_stock_pcp {
2370 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002371 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002372 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002373 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002374#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002375};
2376static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002377static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002378
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002379/**
2380 * consume_stock: Try to consume stocked charge on this cpu.
2381 * @memcg: memcg to consume from.
2382 * @nr_pages: how many pages to charge.
2383 *
2384 * The charges will only happen if @memcg matches the current cpu's memcg
2385 * stock, and at least @nr_pages are available in that stock. Failure to
2386 * service an allocation will refill the stock.
2387 *
2388 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002389 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002390static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002391{
2392 struct memcg_stock_pcp *stock;
2393 bool ret = true;
2394
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002395 if (nr_pages > CHARGE_BATCH)
2396 return false;
2397
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002398 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002399 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2400 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002401 else /* need to call res_counter_charge */
2402 ret = false;
2403 put_cpu_var(memcg_stock);
2404 return ret;
2405}
2406
2407/*
2408 * Returns stocks cached in percpu to res_counter and reset cached information.
2409 */
2410static void drain_stock(struct memcg_stock_pcp *stock)
2411{
2412 struct mem_cgroup *old = stock->cached;
2413
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002414 if (stock->nr_pages) {
2415 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2416
2417 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002418 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002419 res_counter_uncharge(&old->memsw, bytes);
2420 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002421 }
2422 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002423}
2424
2425/*
2426 * This must be called under preempt disabled or must be called by
2427 * a thread which is pinned to local cpu.
2428 */
2429static void drain_local_stock(struct work_struct *dummy)
2430{
2431 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2432 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002433 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002434}
2435
Michal Hockoe4777492013-02-22 16:35:40 -08002436static void __init memcg_stock_init(void)
2437{
2438 int cpu;
2439
2440 for_each_possible_cpu(cpu) {
2441 struct memcg_stock_pcp *stock =
2442 &per_cpu(memcg_stock, cpu);
2443 INIT_WORK(&stock->work, drain_local_stock);
2444 }
2445}
2446
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002447/*
2448 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002449 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002450 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002451static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002452{
2453 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2454
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002455 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002456 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002457 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002458 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002459 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002460 put_cpu_var(memcg_stock);
2461}
2462
2463/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002464 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002465 * of the hierarchy under it. sync flag says whether we should block
2466 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002467 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002468static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002469{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002470 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002471
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002472 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002473 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002474 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002475 for_each_online_cpu(cpu) {
2476 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002477 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002478
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002479 memcg = stock->cached;
2480 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002481 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002482 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002483 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002484 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2485 if (cpu == curcpu)
2486 drain_local_stock(&stock->work);
2487 else
2488 schedule_work_on(cpu, &stock->work);
2489 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002490 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002491 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002492
2493 if (!sync)
2494 goto out;
2495
2496 for_each_online_cpu(cpu) {
2497 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002498 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002499 flush_work(&stock->work);
2500 }
2501out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002502 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002503}
2504
2505/*
2506 * Tries to drain stocked charges in other cpus. This function is asynchronous
2507 * and just put a work per cpu for draining localy on each cpu. Caller can
2508 * expects some charges will be back to res_counter later but cannot wait for
2509 * it.
2510 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002511static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002512{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002513 /*
2514 * If someone calls draining, avoid adding more kworker runs.
2515 */
2516 if (!mutex_trylock(&percpu_charge_mutex))
2517 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002518 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002519 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002520}
2521
2522/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002523static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002524{
2525 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002526 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002527 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002528 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002529}
2530
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002531/*
2532 * This function drains percpu counter value from DEAD cpu and
2533 * move it to local cpu. Note that this function can be preempted.
2534 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002535static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002536{
2537 int i;
2538
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002539 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002540 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002541 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002542
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002543 per_cpu(memcg->stat->count[i], cpu) = 0;
2544 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002545 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002546 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002547 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002549 per_cpu(memcg->stat->events[i], cpu) = 0;
2550 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002551 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002552 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002553}
2554
Paul Gortmaker0db06282013-06-19 14:53:51 -04002555static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002556 unsigned long action,
2557 void *hcpu)
2558{
2559 int cpu = (unsigned long)hcpu;
2560 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002561 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002562
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002563 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002564 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002565
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002566 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002567 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002568
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002569 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002570 mem_cgroup_drain_pcp_counter(iter, cpu);
2571
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002572 stock = &per_cpu(memcg_stock, cpu);
2573 drain_stock(stock);
2574 return NOTIFY_OK;
2575}
2576
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002577
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002578/* See mem_cgroup_try_charge() for details */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002579enum {
2580 CHARGE_OK, /* success */
2581 CHARGE_RETRY, /* need to retry but retry is not bad */
2582 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2583 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002584};
2585
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002586static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002587 unsigned int nr_pages, unsigned int min_pages,
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002588 bool invoke_oom)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002589{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002590 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002591 struct mem_cgroup *mem_over_limit;
2592 struct res_counter *fail_res;
2593 unsigned long flags = 0;
2594 int ret;
2595
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002596 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002597
2598 if (likely(!ret)) {
2599 if (!do_swap_account)
2600 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002601 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002602 if (likely(!ret))
2603 return CHARGE_OK;
2604
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002605 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002606 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2607 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2608 } else
2609 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002610 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002611 * Never reclaim on behalf of optional batching, retry with a
2612 * single page instead.
2613 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002614 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002615 return CHARGE_RETRY;
2616
2617 if (!(gfp_mask & __GFP_WAIT))
2618 return CHARGE_WOULDBLOCK;
2619
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002620 if (gfp_mask & __GFP_NORETRY)
2621 return CHARGE_NOMEM;
2622
Johannes Weiner56600482012-01-12 17:17:59 -08002623 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002624 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002625 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002626 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002627 * Even though the limit is exceeded at this point, reclaim
2628 * may have been able to free some pages. Retry the charge
2629 * before killing the task.
2630 *
2631 * Only for regular pages, though: huge pages are rather
2632 * unlikely to succeed so close to the limit, and we fall back
2633 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002634 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002635 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002636 return CHARGE_RETRY;
2637
2638 /*
2639 * At task move, charge accounts can be doubly counted. So, it's
2640 * better to wait until the end of task_move if something is going on.
2641 */
2642 if (mem_cgroup_wait_acct_move(mem_over_limit))
2643 return CHARGE_RETRY;
2644
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002645 if (invoke_oom)
2646 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002647
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002648 return CHARGE_NOMEM;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002649}
2650
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002651/**
2652 * mem_cgroup_try_charge - try charging a memcg
2653 * @memcg: memcg to charge
2654 * @nr_pages: number of pages to charge
2655 * @oom: trigger OOM if reclaim fails
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002656 *
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002657 * Returns 0 if @memcg was charged successfully, -EINTR if the charge
2658 * was bypassed to root_mem_cgroup, and -ENOMEM if the charge failed.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002659 */
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002660static int mem_cgroup_try_charge(struct mem_cgroup *memcg,
2661 gfp_t gfp_mask,
2662 unsigned int nr_pages,
2663 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002664{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002665 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002666 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002667 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002668
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002669 if (mem_cgroup_is_root(memcg))
2670 goto done;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002671 /*
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002672 * Unlike in global OOM situations, memcg is not in a physical
2673 * memory shortage. Allow dying and OOM-killed tasks to
2674 * bypass the last charges so that they can exit quickly and
2675 * free their memory.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002676 */
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002677 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2678 fatal_signal_pending(current)))
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002679 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002680
Johannes Weiner49426422013-10-16 13:46:59 -07002681 if (unlikely(task_in_memcg_oom(current)))
Johannes Weiner1f14c1a2013-12-12 17:12:35 -08002682 goto nomem;
Johannes Weiner49426422013-10-16 13:46:59 -07002683
Johannes Weinera0d8b002013-12-12 17:12:20 -08002684 if (gfp_mask & __GFP_NOFAIL)
2685 oom = false;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002686again:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002687 if (consume_stock(memcg, nr_pages))
2688 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002689
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002690 do {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002691 bool invoke_oom = oom && !nr_oom_retries;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002692
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002693 /* If killed, bypass charge */
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002694 if (fatal_signal_pending(current))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002695 goto bypass;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002696
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002697 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
2698 nr_pages, invoke_oom);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002699 switch (ret) {
2700 case CHARGE_OK:
2701 break;
2702 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002703 batch = nr_pages;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002704 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002705 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002706 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002707 case CHARGE_NOMEM: /* OOM routine works */
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002708 if (!oom || invoke_oom)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002709 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002710 nr_oom_retries--;
2711 break;
Balbir Singh66e17072008-02-07 00:13:56 -08002712 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002713 } while (ret != CHARGE_OK);
2714
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002715 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002716 refill_stock(memcg, batch - nr_pages);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002717done:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002718 return 0;
2719nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002720 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002721 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002722bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002723 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002724}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002725
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002726/**
2727 * mem_cgroup_try_charge_mm - try charging a mm
2728 * @mm: mm_struct to charge
2729 * @nr_pages: number of pages to charge
2730 * @oom: trigger OOM if reclaim fails
2731 *
2732 * Returns the charged mem_cgroup associated with the given mm_struct or
2733 * NULL the charge failed.
2734 */
2735static struct mem_cgroup *mem_cgroup_try_charge_mm(struct mm_struct *mm,
2736 gfp_t gfp_mask,
2737 unsigned int nr_pages,
2738 bool oom)
2739
2740{
2741 struct mem_cgroup *memcg;
2742 int ret;
2743
2744 memcg = get_mem_cgroup_from_mm(mm);
2745 ret = mem_cgroup_try_charge(memcg, gfp_mask, nr_pages, oom);
2746 css_put(&memcg->css);
2747 if (ret == -EINTR)
2748 memcg = root_mem_cgroup;
2749 else if (ret)
2750 memcg = NULL;
2751
2752 return memcg;
2753}
2754
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002755/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002756 * Somemtimes we have to undo a charge we got by try_charge().
2757 * This function is for that and do uncharge, put css's refcnt.
2758 * gotten by try_charge().
2759 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002760static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002761 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002762{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002763 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002764 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002765
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002766 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002767 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002768 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002769 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002770}
2771
2772/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002773 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2774 * This is useful when moving usage to parent cgroup.
2775 */
2776static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2777 unsigned int nr_pages)
2778{
2779 unsigned long bytes = nr_pages * PAGE_SIZE;
2780
2781 if (mem_cgroup_is_root(memcg))
2782 return;
2783
2784 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2785 if (do_swap_account)
2786 res_counter_uncharge_until(&memcg->memsw,
2787 memcg->memsw.parent, bytes);
2788}
2789
2790/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002791 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002792 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2793 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2794 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002795 */
2796static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2797{
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002798 /* ID 0 is unused ID */
2799 if (!id)
2800 return NULL;
Li Zefan34c00c32013-09-23 16:56:01 +08002801 return mem_cgroup_from_id(id);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002802}
2803
Wu Fengguange42d9d52009-12-16 12:19:59 +01002804struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002805{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002806 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002807 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002808 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002809 swp_entry_t ent;
2810
Sasha Levin309381fea2014-01-23 15:52:54 -08002811 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002812
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002813 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002814 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002815 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002816 memcg = pc->mem_cgroup;
2817 if (memcg && !css_tryget(&memcg->css))
2818 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002819 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002820 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002821 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002822 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002823 memcg = mem_cgroup_lookup(id);
2824 if (memcg && !css_tryget(&memcg->css))
2825 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002826 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002827 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002828 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002829 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002830}
2831
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002832static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002833 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002834 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002835 enum charge_type ctype,
2836 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002837{
Johannes Weinerce587e62012-04-24 20:22:33 +02002838 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002839 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002840 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002841 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002842 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002843
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002844 lock_page_cgroup(pc);
Sasha Levin309381fea2014-01-23 15:52:54 -08002845 VM_BUG_ON_PAGE(PageCgroupUsed(pc), page);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002846 /*
2847 * we don't need page_cgroup_lock about tail pages, becase they are not
2848 * accessed by any other context at this point.
2849 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002850
2851 /*
2852 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2853 * may already be on some other mem_cgroup's LRU. Take care of it.
2854 */
2855 if (lrucare) {
2856 zone = page_zone(page);
2857 spin_lock_irq(&zone->lru_lock);
2858 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002859 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002860 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002861 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002862 was_on_lru = true;
2863 }
2864 }
2865
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002866 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002867 /*
2868 * We access a page_cgroup asynchronously without lock_page_cgroup().
2869 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2870 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2871 * before USED bit, we need memory barrier here.
2872 * See mem_cgroup_add_lru_list(), etc.
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002873 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002874 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002875 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002876
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002877 if (lrucare) {
2878 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002879 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Sasha Levin309381fea2014-01-23 15:52:54 -08002880 VM_BUG_ON_PAGE(PageLRU(page), page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002881 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002882 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002883 }
2884 spin_unlock_irq(&zone->lru_lock);
2885 }
2886
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002887 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002888 anon = true;
2889 else
2890 anon = false;
2891
David Rientjesb070e652013-05-07 16:18:09 -07002892 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002893 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002894
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002895 /*
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002896 * "charge_statistics" updated event counter. Then, check it.
2897 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2898 * if they exceeds softlimit.
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002899 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002900 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002901}
2902
Glauber Costa7cf27982012-12-18 14:22:55 -08002903static DEFINE_MUTEX(set_limit_mutex);
2904
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002905#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08002906static DEFINE_MUTEX(activate_kmem_mutex);
2907
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002908static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2909{
2910 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
Vladimir Davydov6de64be2014-01-23 15:53:08 -08002911 memcg_kmem_is_active(memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002912}
2913
Glauber Costa1f458cb2012-12-18 14:22:50 -08002914/*
2915 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2916 * in the memcg_cache_params struct.
2917 */
2918static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2919{
2920 struct kmem_cache *cachep;
2921
2922 VM_BUG_ON(p->is_root_cache);
2923 cachep = p->root_cache;
Qiang Huang7a67d7a2013-11-12 15:08:24 -08002924 return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg));
Glauber Costa1f458cb2012-12-18 14:22:50 -08002925}
2926
Glauber Costa749c5412012-12-18 14:23:01 -08002927#ifdef CONFIG_SLABINFO
Tejun Heo2da8ca82013-12-05 12:28:04 -05002928static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v)
Glauber Costa749c5412012-12-18 14:23:01 -08002929{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002930 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Glauber Costa749c5412012-12-18 14:23:01 -08002931 struct memcg_cache_params *params;
2932
2933 if (!memcg_can_account_kmem(memcg))
2934 return -EIO;
2935
2936 print_slabinfo_header(m);
2937
2938 mutex_lock(&memcg->slab_caches_mutex);
2939 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2940 cache_show(memcg_params_to_cache(params), m);
2941 mutex_unlock(&memcg->slab_caches_mutex);
2942
2943 return 0;
2944}
2945#endif
2946
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002947static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2948{
2949 struct res_counter *fail_res;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002950 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002951
2952 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2953 if (ret)
2954 return ret;
2955
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002956 ret = mem_cgroup_try_charge(memcg, gfp, size >> PAGE_SHIFT,
2957 oom_gfp_allowed(gfp));
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002958 if (ret == -EINTR) {
2959 /*
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002960 * mem_cgroup_try_charge() chosed to bypass to root due to
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002961 * OOM kill or fatal signal. Since our only options are to
2962 * either fail the allocation or charge it to this cgroup, do
2963 * it as a temporary condition. But we can't fail. From a
2964 * kmem/slab perspective, the cache has already been selected,
2965 * by mem_cgroup_kmem_get_cache(), so it is too late to change
2966 * our minds.
2967 *
2968 * This condition will only trigger if the task entered
2969 * memcg_charge_kmem in a sane state, but was OOM-killed during
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002970 * mem_cgroup_try_charge() above. Tasks that were already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002971 * dying when the allocation triggers should have been already
2972 * directed to the root cgroup in memcontrol.h
2973 */
2974 res_counter_charge_nofail(&memcg->res, size, &fail_res);
2975 if (do_swap_account)
2976 res_counter_charge_nofail(&memcg->memsw, size,
2977 &fail_res);
2978 ret = 0;
2979 } else if (ret)
2980 res_counter_uncharge(&memcg->kmem, size);
2981
2982 return ret;
2983}
2984
2985static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
2986{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002987 res_counter_uncharge(&memcg->res, size);
2988 if (do_swap_account)
2989 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08002990
2991 /* Not down to 0 */
2992 if (res_counter_uncharge(&memcg->kmem, size))
2993 return;
2994
Li Zefan10d5ebf2013-07-08 16:00:33 -07002995 /*
2996 * Releases a reference taken in kmem_cgroup_css_offline in case
2997 * this last uncharge is racing with the offlining code or it is
2998 * outliving the memcg existence.
2999 *
3000 * The memory barrier imposed by test&clear is paired with the
3001 * explicit one in memcg_kmem_mark_dead().
3002 */
Glauber Costa7de37682012-12-18 14:22:07 -08003003 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003004 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003005}
3006
Glauber Costa2633d7a2012-12-18 14:22:34 -08003007/*
3008 * helper for acessing a memcg's index. It will be used as an index in the
3009 * child cache array in kmem_cache, and also to derive its name. This function
3010 * will return -1 when this is not a kmem-limited memcg.
3011 */
3012int memcg_cache_id(struct mem_cgroup *memcg)
3013{
3014 return memcg ? memcg->kmemcg_id : -1;
3015}
3016
Glauber Costa55007d82012-12-18 14:22:38 -08003017static size_t memcg_caches_array_size(int num_groups)
3018{
3019 ssize_t size;
3020 if (num_groups <= 0)
3021 return 0;
3022
3023 size = 2 * num_groups;
3024 if (size < MEMCG_CACHES_MIN_SIZE)
3025 size = MEMCG_CACHES_MIN_SIZE;
3026 else if (size > MEMCG_CACHES_MAX_SIZE)
3027 size = MEMCG_CACHES_MAX_SIZE;
3028
3029 return size;
3030}
3031
3032/*
3033 * We should update the current array size iff all caches updates succeed. This
3034 * can only be done from the slab side. The slab mutex needs to be held when
3035 * calling this.
3036 */
3037void memcg_update_array_size(int num)
3038{
3039 if (num > memcg_limited_groups_array_size)
3040 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3041}
3042
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003043static void kmem_cache_destroy_work_func(struct work_struct *w);
3044
Glauber Costa55007d82012-12-18 14:22:38 -08003045int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3046{
3047 struct memcg_cache_params *cur_params = s->memcg_params;
3048
Qiang Huangf35c3a82013-11-12 15:08:22 -08003049 VM_BUG_ON(!is_root_cache(s));
Glauber Costa55007d82012-12-18 14:22:38 -08003050
3051 if (num_groups > memcg_limited_groups_array_size) {
3052 int i;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003053 struct memcg_cache_params *new_params;
Glauber Costa55007d82012-12-18 14:22:38 -08003054 ssize_t size = memcg_caches_array_size(num_groups);
3055
3056 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003057 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003058
Vladimir Davydovf8570262014-01-23 15:53:06 -08003059 new_params = kzalloc(size, GFP_KERNEL);
3060 if (!new_params)
Glauber Costa55007d82012-12-18 14:22:38 -08003061 return -ENOMEM;
Glauber Costa55007d82012-12-18 14:22:38 -08003062
Vladimir Davydovf8570262014-01-23 15:53:06 -08003063 new_params->is_root_cache = true;
Glauber Costa55007d82012-12-18 14:22:38 -08003064
3065 /*
3066 * There is the chance it will be bigger than
3067 * memcg_limited_groups_array_size, if we failed an allocation
3068 * in a cache, in which case all caches updated before it, will
3069 * have a bigger array.
3070 *
3071 * But if that is the case, the data after
3072 * memcg_limited_groups_array_size is certainly unused
3073 */
3074 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3075 if (!cur_params->memcg_caches[i])
3076 continue;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003077 new_params->memcg_caches[i] =
Glauber Costa55007d82012-12-18 14:22:38 -08003078 cur_params->memcg_caches[i];
3079 }
3080
3081 /*
3082 * Ideally, we would wait until all caches succeed, and only
3083 * then free the old one. But this is not worth the extra
3084 * pointer per-cache we'd have to have for this.
3085 *
3086 * It is not a big deal if some caches are left with a size
3087 * bigger than the others. And all updates will reset this
3088 * anyway.
3089 */
Vladimir Davydovf8570262014-01-23 15:53:06 -08003090 rcu_assign_pointer(s->memcg_params, new_params);
3091 if (cur_params)
3092 kfree_rcu(cur_params, rcu_head);
Glauber Costa55007d82012-12-18 14:22:38 -08003093 }
3094 return 0;
3095}
3096
Vladimir Davydov5722d092014-04-07 15:39:24 -07003097char *memcg_create_cache_name(struct mem_cgroup *memcg,
3098 struct kmem_cache *root_cache)
3099{
3100 static char *buf = NULL;
3101
3102 /*
3103 * We need a mutex here to protect the shared buffer. Since this is
3104 * expected to be called only on cache creation, we can employ the
3105 * slab_mutex for that purpose.
3106 */
3107 lockdep_assert_held(&slab_mutex);
3108
3109 if (!buf) {
3110 buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
3111 if (!buf)
3112 return NULL;
3113 }
3114
3115 cgroup_name(memcg->css.cgroup, buf, NAME_MAX + 1);
3116 return kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
3117 memcg_cache_id(memcg), buf);
3118}
3119
Vladimir Davydov363a0442014-01-23 15:52:56 -08003120int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s,
3121 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003122{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003123 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003124
3125 if (!memcg_kmem_enabled())
3126 return 0;
3127
Andrey Vagin90c7a792013-09-11 14:22:18 -07003128 if (!memcg) {
3129 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003130 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003131 } else
3132 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003133
Glauber Costa2633d7a2012-12-18 14:22:34 -08003134 s->memcg_params = kzalloc(size, GFP_KERNEL);
3135 if (!s->memcg_params)
3136 return -ENOMEM;
3137
Glauber Costa943a4512012-12-18 14:23:03 -08003138 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003139 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003140 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003141 INIT_WORK(&s->memcg_params->destroy,
3142 kmem_cache_destroy_work_func);
Vladimir Davydov051dd462014-04-07 15:39:27 -07003143 css_get(&memcg->css);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003144 } else
3145 s->memcg_params->is_root_cache = true;
3146
Glauber Costa2633d7a2012-12-18 14:22:34 -08003147 return 0;
3148}
3149
Vladimir Davydov363a0442014-01-23 15:52:56 -08003150void memcg_free_cache_params(struct kmem_cache *s)
3151{
Vladimir Davydov051dd462014-04-07 15:39:27 -07003152 if (!s->memcg_params)
3153 return;
3154 if (!s->memcg_params->is_root_cache)
3155 css_put(&s->memcg_params->memcg->css);
Vladimir Davydov363a0442014-01-23 15:52:56 -08003156 kfree(s->memcg_params);
3157}
3158
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003159void memcg_register_cache(struct kmem_cache *s)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003160{
Glauber Costad7f25f82012-12-18 14:22:40 -08003161 struct kmem_cache *root;
3162 struct mem_cgroup *memcg;
3163 int id;
3164
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003165 if (is_root_cache(s))
Glauber Costad7f25f82012-12-18 14:22:40 -08003166 return;
3167
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003168 /*
3169 * Holding the slab_mutex assures nobody will touch the memcg_caches
3170 * array while we are modifying it.
3171 */
3172 lockdep_assert_held(&slab_mutex);
3173
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003174 root = s->memcg_params->root_cache;
3175 memcg = s->memcg_params->memcg;
3176 id = memcg_cache_id(memcg);
3177
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003178 /*
Vladimir Davydov959c8962014-01-23 15:52:59 -08003179 * Since readers won't lock (see cache_from_memcg_idx()), we need a
3180 * barrier here to ensure nobody will see the kmem_cache partially
3181 * initialized.
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003182 */
Vladimir Davydov959c8962014-01-23 15:52:59 -08003183 smp_wmb();
3184
Vladimir Davydov96403da2014-01-23 15:53:01 -08003185 /*
3186 * Initialize the pointer to this cache in its parent's memcg_params
3187 * before adding it to the memcg_slab_caches list, otherwise we can
3188 * fail to convert memcg_params_to_cache() while traversing the list.
3189 */
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003190 VM_BUG_ON(root->memcg_params->memcg_caches[id]);
Vladimir Davydov959c8962014-01-23 15:52:59 -08003191 root->memcg_params->memcg_caches[id] = s;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003192
3193 mutex_lock(&memcg->slab_caches_mutex);
3194 list_add(&s->memcg_params->list, &memcg->memcg_slab_caches);
3195 mutex_unlock(&memcg->slab_caches_mutex);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003196}
3197
3198void memcg_unregister_cache(struct kmem_cache *s)
3199{
3200 struct kmem_cache *root;
3201 struct mem_cgroup *memcg;
3202 int id;
3203
3204 if (is_root_cache(s))
3205 return;
Glauber Costad7f25f82012-12-18 14:22:40 -08003206
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003207 /*
3208 * Holding the slab_mutex assures nobody will touch the memcg_caches
3209 * array while we are modifying it.
3210 */
3211 lockdep_assert_held(&slab_mutex);
3212
Glauber Costad7f25f82012-12-18 14:22:40 -08003213 root = s->memcg_params->root_cache;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003214 memcg = s->memcg_params->memcg;
3215 id = memcg_cache_id(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08003216
3217 mutex_lock(&memcg->slab_caches_mutex);
3218 list_del(&s->memcg_params->list);
3219 mutex_unlock(&memcg->slab_caches_mutex);
3220
Vladimir Davydov96403da2014-01-23 15:53:01 -08003221 /*
3222 * Clear the pointer to this cache in its parent's memcg_params only
3223 * after removing it from the memcg_slab_caches list, otherwise we can
3224 * fail to convert memcg_params_to_cache() while traversing the list.
3225 */
Vladimir Davydov051dd462014-04-07 15:39:27 -07003226 VM_BUG_ON(root->memcg_params->memcg_caches[id] != s);
Vladimir Davydov96403da2014-01-23 15:53:01 -08003227 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003228}
3229
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003230/*
3231 * During the creation a new cache, we need to disable our accounting mechanism
3232 * altogether. This is true even if we are not creating, but rather just
3233 * enqueing new caches to be created.
3234 *
3235 * This is because that process will trigger allocations; some visible, like
3236 * explicit kmallocs to auxiliary data structures, name strings and internal
3237 * cache structures; some well concealed, like INIT_WORK() that can allocate
3238 * objects during debug.
3239 *
3240 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3241 * to it. This may not be a bounded recursion: since the first cache creation
3242 * failed to complete (waiting on the allocation), we'll just try to create the
3243 * cache again, failing at the same point.
3244 *
3245 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3246 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3247 * inside the following two functions.
3248 */
3249static inline void memcg_stop_kmem_account(void)
3250{
3251 VM_BUG_ON(!current->mm);
3252 current->memcg_kmem_skip_account++;
3253}
3254
3255static inline void memcg_resume_kmem_account(void)
3256{
3257 VM_BUG_ON(!current->mm);
3258 current->memcg_kmem_skip_account--;
3259}
3260
Glauber Costa1f458cb2012-12-18 14:22:50 -08003261static void kmem_cache_destroy_work_func(struct work_struct *w)
3262{
3263 struct kmem_cache *cachep;
3264 struct memcg_cache_params *p;
3265
3266 p = container_of(w, struct memcg_cache_params, destroy);
3267
3268 cachep = memcg_params_to_cache(p);
3269
Glauber Costa22933152012-12-18 14:22:59 -08003270 /*
3271 * If we get down to 0 after shrink, we could delete right away.
3272 * However, memcg_release_pages() already puts us back in the workqueue
3273 * in that case. If we proceed deleting, we'll get a dangling
3274 * reference, and removing the object from the workqueue in that case
3275 * is unnecessary complication. We are not a fast path.
3276 *
3277 * Note that this case is fundamentally different from racing with
3278 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3279 * kmem_cache_shrink, not only we would be reinserting a dead cache
3280 * into the queue, but doing so from inside the worker racing to
3281 * destroy it.
3282 *
3283 * So if we aren't down to zero, we'll just schedule a worker and try
3284 * again
3285 */
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003286 if (atomic_read(&cachep->memcg_params->nr_pages) != 0)
Glauber Costa22933152012-12-18 14:22:59 -08003287 kmem_cache_shrink(cachep);
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003288 else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003289 kmem_cache_destroy(cachep);
3290}
3291
3292void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3293{
3294 if (!cachep->memcg_params->dead)
3295 return;
3296
3297 /*
Glauber Costa22933152012-12-18 14:22:59 -08003298 * There are many ways in which we can get here.
3299 *
3300 * We can get to a memory-pressure situation while the delayed work is
3301 * still pending to run. The vmscan shrinkers can then release all
3302 * cache memory and get us to destruction. If this is the case, we'll
3303 * be executed twice, which is a bug (the second time will execute over
3304 * bogus data). In this case, cancelling the work should be fine.
3305 *
3306 * But we can also get here from the worker itself, if
3307 * kmem_cache_shrink is enough to shake all the remaining objects and
3308 * get the page count to 0. In this case, we'll deadlock if we try to
3309 * cancel the work (the worker runs with an internal lock held, which
3310 * is the same lock we would hold for cancel_work_sync().)
3311 *
3312 * Since we can't possibly know who got us here, just refrain from
3313 * running if there is already work pending
3314 */
3315 if (work_pending(&cachep->memcg_params->destroy))
3316 return;
3317 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003318 * We have to defer the actual destroying to a workqueue, because
3319 * we might currently be in a context that cannot sleep.
3320 */
3321 schedule_work(&cachep->memcg_params->destroy);
3322}
3323
Vladimir Davydovb8529902014-04-07 15:39:28 -07003324int __kmem_cache_destroy_memcg_children(struct kmem_cache *s)
Glauber Costa7cf27982012-12-18 14:22:55 -08003325{
3326 struct kmem_cache *c;
Vladimir Davydovb8529902014-04-07 15:39:28 -07003327 int i, failed = 0;
Glauber Costa7cf27982012-12-18 14:22:55 -08003328
3329 /*
3330 * If the cache is being destroyed, we trust that there is no one else
3331 * requesting objects from it. Even if there are, the sanity checks in
3332 * kmem_cache_destroy should caught this ill-case.
3333 *
3334 * Still, we don't want anyone else freeing memcg_caches under our
3335 * noses, which can happen if a new memcg comes to life. As usual,
Vladimir Davydovd6441632014-01-23 15:53:09 -08003336 * we'll take the activate_kmem_mutex to protect ourselves against
3337 * this.
Glauber Costa7cf27982012-12-18 14:22:55 -08003338 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08003339 mutex_lock(&activate_kmem_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003340 for_each_memcg_cache_index(i) {
3341 c = cache_from_memcg_idx(s, i);
Glauber Costa7cf27982012-12-18 14:22:55 -08003342 if (!c)
3343 continue;
3344
3345 /*
3346 * We will now manually delete the caches, so to avoid races
3347 * we need to cancel all pending destruction workers and
3348 * proceed with destruction ourselves.
3349 *
3350 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3351 * and that could spawn the workers again: it is likely that
3352 * the cache still have active pages until this very moment.
3353 * This would lead us back to mem_cgroup_destroy_cache.
3354 *
3355 * But that will not execute at all if the "dead" flag is not
3356 * set, so flip it down to guarantee we are in control.
3357 */
3358 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003359 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003360 kmem_cache_destroy(c);
Vladimir Davydovb8529902014-04-07 15:39:28 -07003361
3362 if (cache_from_memcg_idx(s, i))
3363 failed++;
Glauber Costa7cf27982012-12-18 14:22:55 -08003364 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08003365 mutex_unlock(&activate_kmem_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -07003366 return failed;
Glauber Costa7cf27982012-12-18 14:22:55 -08003367}
3368
Glauber Costa1f458cb2012-12-18 14:22:50 -08003369static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3370{
3371 struct kmem_cache *cachep;
3372 struct memcg_cache_params *params;
3373
3374 if (!memcg_kmem_is_active(memcg))
3375 return;
3376
3377 mutex_lock(&memcg->slab_caches_mutex);
3378 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3379 cachep = memcg_params_to_cache(params);
3380 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003381 schedule_work(&cachep->memcg_params->destroy);
3382 }
3383 mutex_unlock(&memcg->slab_caches_mutex);
3384}
3385
Vladimir Davydov5722d092014-04-07 15:39:24 -07003386struct create_work {
3387 struct mem_cgroup *memcg;
3388 struct kmem_cache *cachep;
3389 struct work_struct work;
3390};
3391
Glauber Costad7f25f82012-12-18 14:22:40 -08003392static void memcg_create_cache_work_func(struct work_struct *w)
3393{
Vladimir Davydov5722d092014-04-07 15:39:24 -07003394 struct create_work *cw = container_of(w, struct create_work, work);
3395 struct mem_cgroup *memcg = cw->memcg;
3396 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003397
Vladimir Davydov794b1242014-04-07 15:39:26 -07003398 kmem_cache_create_memcg(memcg, cachep);
Vladimir Davydov5722d092014-04-07 15:39:24 -07003399 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003400 kfree(cw);
3401}
3402
3403/*
3404 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003405 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003406static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3407 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003408{
3409 struct create_work *cw;
3410
3411 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003412 if (cw == NULL) {
3413 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003414 return;
3415 }
3416
3417 cw->memcg = memcg;
3418 cw->cachep = cachep;
3419
3420 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3421 schedule_work(&cw->work);
3422}
3423
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003424static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3425 struct kmem_cache *cachep)
3426{
3427 /*
3428 * We need to stop accounting when we kmalloc, because if the
3429 * corresponding kmalloc cache is not yet created, the first allocation
3430 * in __memcg_create_cache_enqueue will recurse.
3431 *
3432 * However, it is better to enclose the whole function. Depending on
3433 * the debugging options enabled, INIT_WORK(), for instance, can
3434 * trigger an allocation. This too, will make us recurse. Because at
3435 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3436 * the safest choice is to do it like this, wrapping the whole function.
3437 */
3438 memcg_stop_kmem_account();
3439 __memcg_create_cache_enqueue(memcg, cachep);
3440 memcg_resume_kmem_account();
3441}
Glauber Costad7f25f82012-12-18 14:22:40 -08003442/*
3443 * Return the kmem_cache we're supposed to use for a slab allocation.
3444 * We try to use the current memcg's version of the cache.
3445 *
3446 * If the cache does not exist yet, if we are the first user of it,
3447 * we either create it immediately, if possible, or create it asynchronously
3448 * in a workqueue.
3449 * In the latter case, we will let the current allocation go through with
3450 * the original cache.
3451 *
3452 * Can't be called in interrupt context or from kernel threads.
3453 * This function needs to be called with rcu_read_lock() held.
3454 */
3455struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3456 gfp_t gfp)
3457{
3458 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08003459 struct kmem_cache *memcg_cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003460
3461 VM_BUG_ON(!cachep->memcg_params);
3462 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3463
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003464 if (!current->mm || current->memcg_kmem_skip_account)
3465 return cachep;
3466
Glauber Costad7f25f82012-12-18 14:22:40 -08003467 rcu_read_lock();
3468 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003469
3470 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003471 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003472
Vladimir Davydov959c8962014-01-23 15:52:59 -08003473 memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg));
3474 if (likely(memcg_cachep)) {
3475 cachep = memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07003476 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003477 }
3478
Li Zefanca0dde92013-04-29 15:08:57 -07003479 /* The corresponding put will be done in the workqueue. */
3480 if (!css_tryget(&memcg->css))
3481 goto out;
3482 rcu_read_unlock();
3483
3484 /*
3485 * If we are in a safe context (can wait, and not in interrupt
3486 * context), we could be be predictable and return right away.
3487 * This would guarantee that the allocation being performed
3488 * already belongs in the new cache.
3489 *
3490 * However, there are some clashes that can arrive from locking.
3491 * For instance, because we acquire the slab_mutex while doing
3492 * kmem_cache_dup, this means no further allocation could happen
3493 * with the slab_mutex held.
3494 *
3495 * Also, because cache creation issue get_online_cpus(), this
3496 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3497 * that ends up reversed during cpu hotplug. (cpuset allocates
3498 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3499 * better to defer everything.
3500 */
3501 memcg_create_cache_enqueue(memcg, cachep);
3502 return cachep;
3503out:
3504 rcu_read_unlock();
3505 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003506}
3507EXPORT_SYMBOL(__memcg_kmem_get_cache);
3508
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003509/*
3510 * We need to verify if the allocation against current->mm->owner's memcg is
3511 * possible for the given order. But the page is not allocated yet, so we'll
3512 * need a further commit step to do the final arrangements.
3513 *
3514 * It is possible for the task to switch cgroups in this mean time, so at
3515 * commit time, we can't rely on task conversion any longer. We'll then use
3516 * the handle argument to return to the caller which cgroup we should commit
3517 * against. We could also return the memcg directly and avoid the pointer
3518 * passing, but a boolean return value gives better semantics considering
3519 * the compiled-out case as well.
3520 *
3521 * Returning true means the allocation is possible.
3522 */
3523bool
3524__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3525{
3526 struct mem_cgroup *memcg;
3527 int ret;
3528
3529 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003530
3531 /*
3532 * Disabling accounting is only relevant for some specific memcg
3533 * internal allocations. Therefore we would initially not have such
3534 * check here, since direct calls to the page allocator that are marked
3535 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3536 * concerned with cache allocations, and by having this test at
3537 * memcg_kmem_get_cache, we are already able to relay the allocation to
3538 * the root cache and bypass the memcg cache altogether.
3539 *
3540 * There is one exception, though: the SLUB allocator does not create
3541 * large order caches, but rather service large kmallocs directly from
3542 * the page allocator. Therefore, the following sequence when backed by
3543 * the SLUB allocator:
3544 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003545 * memcg_stop_kmem_account();
3546 * kmalloc(<large_number>)
3547 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003548 *
3549 * would effectively ignore the fact that we should skip accounting,
3550 * since it will drive us directly to this function without passing
3551 * through the cache selector memcg_kmem_get_cache. Such large
3552 * allocations are extremely rare but can happen, for instance, for the
3553 * cache arrays. We bring this test here.
3554 */
3555 if (!current->mm || current->memcg_kmem_skip_account)
3556 return true;
3557
Johannes Weinerdf381972014-04-07 15:37:43 -07003558 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003559
3560 if (!memcg_can_account_kmem(memcg)) {
3561 css_put(&memcg->css);
3562 return true;
3563 }
3564
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003565 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3566 if (!ret)
3567 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003568
3569 css_put(&memcg->css);
3570 return (ret == 0);
3571}
3572
3573void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3574 int order)
3575{
3576 struct page_cgroup *pc;
3577
3578 VM_BUG_ON(mem_cgroup_is_root(memcg));
3579
3580 /* The page allocation failed. Revert */
3581 if (!page) {
3582 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003583 return;
3584 }
3585
3586 pc = lookup_page_cgroup(page);
3587 lock_page_cgroup(pc);
3588 pc->mem_cgroup = memcg;
3589 SetPageCgroupUsed(pc);
3590 unlock_page_cgroup(pc);
3591}
3592
3593void __memcg_kmem_uncharge_pages(struct page *page, int order)
3594{
3595 struct mem_cgroup *memcg = NULL;
3596 struct page_cgroup *pc;
3597
3598
3599 pc = lookup_page_cgroup(page);
3600 /*
3601 * Fast unlocked return. Theoretically might have changed, have to
3602 * check again after locking.
3603 */
3604 if (!PageCgroupUsed(pc))
3605 return;
3606
3607 lock_page_cgroup(pc);
3608 if (PageCgroupUsed(pc)) {
3609 memcg = pc->mem_cgroup;
3610 ClearPageCgroupUsed(pc);
3611 }
3612 unlock_page_cgroup(pc);
3613
3614 /*
3615 * We trust that only if there is a memcg associated with the page, it
3616 * is a valid allocation
3617 */
3618 if (!memcg)
3619 return;
3620
Sasha Levin309381fea2014-01-23 15:52:54 -08003621 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003622 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003623}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003624#else
3625static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3626{
3627}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003628#endif /* CONFIG_MEMCG_KMEM */
3629
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003630#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3631
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003632#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003633/*
3634 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003635 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3636 * charge/uncharge will be never happen and move_account() is done under
3637 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003638 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003639void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003640{
3641 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003642 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003643 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003644 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003645
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003646 if (mem_cgroup_disabled())
3647 return;
David Rientjesb070e652013-05-07 16:18:09 -07003648
3649 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003650 for (i = 1; i < HPAGE_PMD_NR; i++) {
3651 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003652 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003653 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003654 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3655 }
David Rientjesb070e652013-05-07 16:18:09 -07003656 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3657 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003658}
Hugh Dickins12d27102012-01-12 17:19:52 -08003659#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003660
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003661/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003662 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003663 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003664 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003665 * @pc: page_cgroup of the page.
3666 * @from: mem_cgroup which the page is moved from.
3667 * @to: mem_cgroup which the page is moved to. @from != @to.
3668 *
3669 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003670 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003671 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003672 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003673 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3674 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003675 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003676static int mem_cgroup_move_account(struct page *page,
3677 unsigned int nr_pages,
3678 struct page_cgroup *pc,
3679 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003680 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003681{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003682 unsigned long flags;
3683 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003684 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003685
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003686 VM_BUG_ON(from == to);
Sasha Levin309381fea2014-01-23 15:52:54 -08003687 VM_BUG_ON_PAGE(PageLRU(page), page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003688 /*
3689 * The page is isolated from LRU. So, collapse function
3690 * will not handle this page. But page splitting can happen.
3691 * Do this check under compound_page_lock(). The caller should
3692 * hold it.
3693 */
3694 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003695 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003696 goto out;
3697
3698 lock_page_cgroup(pc);
3699
3700 ret = -EINVAL;
3701 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3702 goto unlock;
3703
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003704 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003705
Johannes Weiner59d1d252014-04-07 15:37:40 -07003706 if (!anon && page_mapped(page)) {
3707 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3708 nr_pages);
3709 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3710 nr_pages);
3711 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003712
Johannes Weiner59d1d252014-04-07 15:37:40 -07003713 if (PageWriteback(page)) {
3714 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3715 nr_pages);
3716 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3717 nr_pages);
3718 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003719
David Rientjesb070e652013-05-07 16:18:09 -07003720 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003721
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003722 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003723 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003724 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003725 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003726 ret = 0;
3727unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003728 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003729 /*
3730 * check events
3731 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003732 memcg_check_events(to, page);
3733 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003734out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003735 return ret;
3736}
3737
Michal Hocko2ef37d32012-10-26 13:37:30 +02003738/**
3739 * mem_cgroup_move_parent - moves page to the parent group
3740 * @page: the page to move
3741 * @pc: page_cgroup of the page
3742 * @child: page's cgroup
3743 *
3744 * move charges to its parent or the root cgroup if the group has no
3745 * parent (aka use_hierarchy==0).
3746 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3747 * mem_cgroup_move_account fails) the failure is always temporary and
3748 * it signals a race with a page removal/uncharge or migration. In the
3749 * first case the page is on the way out and it will vanish from the LRU
3750 * on the next attempt and the call should be retried later.
3751 * Isolation from the LRU fails only if page has been isolated from
3752 * the LRU since we looked at it and that usually means either global
3753 * reclaim or migration going on. The page will either get back to the
3754 * LRU or vanish.
3755 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3756 * (!PageCgroupUsed) or moved to a different group. The page will
3757 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003758 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003759static int mem_cgroup_move_parent(struct page *page,
3760 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003761 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003762{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003763 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003764 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003765 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003766 int ret;
3767
Michal Hockod8423012012-10-26 13:37:29 +02003768 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003769
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003770 ret = -EBUSY;
3771 if (!get_page_unless_zero(page))
3772 goto out;
3773 if (isolate_lru_page(page))
3774 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003775
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003776 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003777
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003778 parent = parent_mem_cgroup(child);
3779 /*
3780 * If no parent, move charges to root cgroup.
3781 */
3782 if (!parent)
3783 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003784
Michal Hocko2ef37d32012-10-26 13:37:30 +02003785 if (nr_pages > 1) {
Sasha Levin309381fea2014-01-23 15:52:54 -08003786 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003787 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003788 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003789
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003790 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003791 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003792 if (!ret)
3793 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003794
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003795 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003796 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003797 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003798put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003799 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003800out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003801 return ret;
3802}
3803
Michal Hockod715ae02014-04-07 15:37:46 -07003804int mem_cgroup_charge_anon(struct page *page,
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003805 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003806{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003807 unsigned int nr_pages = 1;
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003808 struct mem_cgroup *memcg;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003809 bool oom = true;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003810
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003811 if (mem_cgroup_disabled())
3812 return 0;
3813
3814 VM_BUG_ON_PAGE(page_mapped(page), page);
3815 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page);
3816 VM_BUG_ON(!mm);
3817
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003818 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003819 nr_pages <<= compound_order(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08003820 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Johannes Weiner8493ae42011-02-01 15:52:44 -08003821 /*
3822 * Never OOM-kill a process for a huge page. The
3823 * fault handler will fall back to regular pages.
3824 */
3825 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003826 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003827
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003828 memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, nr_pages, oom);
3829 if (!memcg)
3830 return -ENOMEM;
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003831 __mem_cgroup_commit_charge(memcg, page, nr_pages,
3832 MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003833 return 0;
3834}
3835
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003836/*
3837 * While swap-in, try_charge -> commit or cancel, the page is locked.
3838 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003839 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003840 * "commit()" or removed by "cancel()"
3841 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003842static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3843 struct page *page,
3844 gfp_t mask,
3845 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003846{
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003847 struct mem_cgroup *memcg = NULL;
Johannes Weiner90deb782012-07-31 16:45:47 -07003848 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003849 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003850
Johannes Weiner90deb782012-07-31 16:45:47 -07003851 pc = lookup_page_cgroup(page);
3852 /*
3853 * Every swap fault against a single page tries to charge the
3854 * page, bail as early as possible. shmem_unuse() encounters
3855 * already charged pages, too. The USED bit is protected by
3856 * the page lock, which serializes swap cache removal, which
3857 * in turn serializes uncharging.
3858 */
3859 if (PageCgroupUsed(pc))
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003860 goto out;
3861 if (do_swap_account)
3862 memcg = try_get_mem_cgroup_from_page(page);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003863 if (!memcg)
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003864 memcg = get_mem_cgroup_from_mm(mm);
3865 ret = mem_cgroup_try_charge(memcg, mask, 1, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003866 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003867 if (ret == -EINTR)
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003868 memcg = root_mem_cgroup;
3869 else if (ret)
3870 return ret;
3871out:
3872 *memcgp = memcg;
3873 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003874}
3875
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003876int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
3877 gfp_t gfp_mask, struct mem_cgroup **memcgp)
3878{
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003879 if (mem_cgroup_disabled()) {
3880 *memcgp = NULL;
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003881 return 0;
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003882 }
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003883 /*
3884 * A racing thread's fault, or swapoff, may have already
3885 * updated the pte, and even removed page from swap cache: in
3886 * those cases unuse_pte()'s pte_same() test will fail; but
3887 * there's also a KSM case which does need to charge the page.
3888 */
3889 if (!PageSwapCache(page)) {
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003890 struct mem_cgroup *memcg;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003891
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003892 memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1, true);
3893 if (!memcg)
3894 return -ENOMEM;
3895 *memcgp = memcg;
3896 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003897 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003898 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
3899}
3900
Johannes Weiner827a03d2012-07-31 16:45:36 -07003901void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
3902{
3903 if (mem_cgroup_disabled())
3904 return;
3905 if (!memcg)
3906 return;
3907 __mem_cgroup_cancel_charge(memcg, 1);
3908}
3909
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003910static void
Johannes Weiner72835c82012-01-12 17:18:32 -08003911__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003912 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003913{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003914 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003915 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08003916 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003917 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07003918
Johannes Weinerce587e62012-04-24 20:22:33 +02003919 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003920 /*
3921 * Now swap is on-memory. This means this page may be
3922 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003923 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
3924 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
3925 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003926 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003927 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003928 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07003929 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003930 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003931}
3932
Johannes Weiner72835c82012-01-12 17:18:32 -08003933void mem_cgroup_commit_charge_swapin(struct page *page,
3934 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003935{
Johannes Weiner72835c82012-01-12 17:18:32 -08003936 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003937 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003938}
3939
Michal Hockod715ae02014-04-07 15:37:46 -07003940int mem_cgroup_charge_file(struct page *page, struct mm_struct *mm,
Johannes Weiner827a03d2012-07-31 16:45:36 -07003941 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003942{
Johannes Weiner827a03d2012-07-31 16:45:36 -07003943 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003944 struct mem_cgroup *memcg;
Johannes Weiner827a03d2012-07-31 16:45:36 -07003945 int ret;
3946
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003947 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07003948 return 0;
3949 if (PageCompound(page))
3950 return 0;
3951
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003952 if (PageSwapCache(page)) { /* shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003953 ret = __mem_cgroup_try_charge_swapin(mm, page,
3954 gfp_mask, &memcg);
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003955 if (ret)
3956 return ret;
3957 __mem_cgroup_commit_charge_swapin(page, memcg, type);
3958 return 0;
Johannes Weiner827a03d2012-07-31 16:45:36 -07003959 }
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07003960
3961 /*
3962 * Page cache insertions can happen without an actual mm
3963 * context, e.g. during disk probing on boot.
3964 */
3965 if (unlikely(!mm))
3966 memcg = root_mem_cgroup;
3967 else {
3968 memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1, true);
3969 if (!memcg)
3970 return -ENOMEM;
3971 }
3972 __mem_cgroup_commit_charge(memcg, page, 1, type, false);
3973 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003974}
3975
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003976static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003977 unsigned int nr_pages,
3978 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003979{
3980 struct memcg_batch_info *batch = NULL;
3981 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003982
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003983 /* If swapout, usage of swap doesn't decrease */
3984 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
3985 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003986
3987 batch = &current->memcg_batch;
3988 /*
3989 * In usual, we do css_get() when we remember memcg pointer.
3990 * But in this case, we keep res->usage until end of a series of
3991 * uncharges. Then, it's ok to ignore memcg's refcnt.
3992 */
3993 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003994 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003995 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003996 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003997 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003998 * the same cgroup and we have chance to coalesce uncharges.
3999 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4000 * because we want to do uncharge as soon as possible.
4001 */
4002
4003 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4004 goto direct_uncharge;
4005
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004006 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004007 goto direct_uncharge;
4008
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004009 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004010 * In typical case, batch->memcg == mem. This means we can
4011 * merge a series of uncharges to an uncharge of res_counter.
4012 * If not, we uncharge res_counter ony by one.
4013 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004014 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004015 goto direct_uncharge;
4016 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004017 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004018 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004019 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004020 return;
4021direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004022 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004023 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004024 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4025 if (unlikely(batch->memcg != memcg))
4026 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004027}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004028
Balbir Singh8697d332008-02-07 00:13:59 -08004029/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004030 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004031 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004032static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004033__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4034 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004035{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004036 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004037 unsigned int nr_pages = 1;
4038 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004039 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004040
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004041 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004042 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004043
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004044 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004045 nr_pages <<= compound_order(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08004046 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004047 }
Balbir Singh8697d332008-02-07 00:13:59 -08004048 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004049 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004050 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004051 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004052 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004053 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004054
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004055 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004056
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004057 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004058
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004059 if (!PageCgroupUsed(pc))
4060 goto unlock_out;
4061
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004062 anon = PageAnon(page);
4063
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004064 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004065 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004066 /*
4067 * Generally PageAnon tells if it's the anon statistics to be
4068 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4069 * used before page reached the stage of being marked PageAnon.
4070 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004071 anon = true;
4072 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004073 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004074 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004075 if (page_mapped(page))
4076 goto unlock_out;
4077 /*
4078 * Pages under migration may not be uncharged. But
4079 * end_migration() /must/ be the one uncharging the
4080 * unused post-migration page and so it has to call
4081 * here with the migration bit still set. See the
4082 * res_counter handling below.
4083 */
4084 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004085 goto unlock_out;
4086 break;
4087 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4088 if (!PageAnon(page)) { /* Shared memory */
4089 if (page->mapping && !page_is_file_cache(page))
4090 goto unlock_out;
4091 } else if (page_mapped(page)) /* Anon */
4092 goto unlock_out;
4093 break;
4094 default:
4095 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004096 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004097
David Rientjesb070e652013-05-07 16:18:09 -07004098 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004099
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004100 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004101 /*
4102 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4103 * freed from LRU. This is safe because uncharged page is expected not
4104 * to be reused (freed soon). Exception is SwapCache, it's handled by
4105 * special functions.
4106 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004107
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004108 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004109 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004110 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004111 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004112 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004113 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004114 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004115 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004116 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004117 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004118 /*
4119 * Migration does not charge the res_counter for the
4120 * replacement page, so leave it alone when phasing out the
4121 * page that is unused after the migration.
4122 */
4123 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004124 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004125
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004126 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004127
4128unlock_out:
4129 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004130 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004131}
4132
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004133void mem_cgroup_uncharge_page(struct page *page)
4134{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004135 /* early check. */
4136 if (page_mapped(page))
4137 return;
Sasha Levin309381fea2014-01-23 15:52:54 -08004138 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page);
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004139 /*
4140 * If the page is in swap cache, uncharge should be deferred
4141 * to the swap path, which also properly accounts swap usage
4142 * and handles memcg lifetime.
4143 *
4144 * Note that this check is not stable and reclaim may add the
4145 * page to swap cache at any time after this. However, if the
4146 * page is not in swap cache by the time page->mapcount hits
4147 * 0, there won't be any page table references to the swap
4148 * slot, and reclaim will free it and not actually write the
4149 * page to disk.
4150 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004151 if (PageSwapCache(page))
4152 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004153 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004154}
4155
4156void mem_cgroup_uncharge_cache_page(struct page *page)
4157{
Sasha Levin309381fea2014-01-23 15:52:54 -08004158 VM_BUG_ON_PAGE(page_mapped(page), page);
4159 VM_BUG_ON_PAGE(page->mapping, page);
Johannes Weiner0030f532012-07-31 16:45:25 -07004160 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004161}
4162
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004163/*
4164 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4165 * In that cases, pages are freed continuously and we can expect pages
4166 * are in the same memcg. All these calls itself limits the number of
4167 * pages freed at once, then uncharge_start/end() is called properly.
4168 * This may be called prural(2) times in a context,
4169 */
4170
4171void mem_cgroup_uncharge_start(void)
4172{
4173 current->memcg_batch.do_batch++;
4174 /* We can do nest. */
4175 if (current->memcg_batch.do_batch == 1) {
4176 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004177 current->memcg_batch.nr_pages = 0;
4178 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004179 }
4180}
4181
4182void mem_cgroup_uncharge_end(void)
4183{
4184 struct memcg_batch_info *batch = &current->memcg_batch;
4185
4186 if (!batch->do_batch)
4187 return;
4188
4189 batch->do_batch--;
4190 if (batch->do_batch) /* If stacked, do nothing. */
4191 return;
4192
4193 if (!batch->memcg)
4194 return;
4195 /*
4196 * This "batch->memcg" is valid without any css_get/put etc...
4197 * bacause we hide charges behind us.
4198 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004199 if (batch->nr_pages)
4200 res_counter_uncharge(&batch->memcg->res,
4201 batch->nr_pages * PAGE_SIZE);
4202 if (batch->memsw_nr_pages)
4203 res_counter_uncharge(&batch->memcg->memsw,
4204 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004205 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004206 /* forget this pointer (for sanity check) */
4207 batch->memcg = NULL;
4208}
4209
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004210#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004211/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004212 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004213 * memcg information is recorded to swap_cgroup of "ent"
4214 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004215void
4216mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004217{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004218 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004219 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004220
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004221 if (!swapout) /* this was a swap cache but the swap is unused ! */
4222 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4223
Johannes Weiner0030f532012-07-31 16:45:25 -07004224 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004225
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004226 /*
4227 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004228 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004229 */
4230 if (do_swap_account && swapout && memcg)
Li Zefan34c00c32013-09-23 16:56:01 +08004231 swap_cgroup_record(ent, mem_cgroup_id(memcg));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004232}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004233#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004234
Andrew Mortonc255a452012-07-31 16:43:02 -07004235#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004236/*
4237 * called from swap_entry_free(). remove record in swap_cgroup and
4238 * uncharge "memsw" account.
4239 */
4240void mem_cgroup_uncharge_swap(swp_entry_t ent)
4241{
4242 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004243 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004244
4245 if (!do_swap_account)
4246 return;
4247
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004248 id = swap_cgroup_record(ent, 0);
4249 rcu_read_lock();
4250 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004251 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004252 /*
4253 * We uncharge this because swap is freed.
4254 * This memcg can be obsolete one. We avoid calling css_tryget
4255 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004256 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004257 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004258 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004259 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004260 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004261 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004262}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004263
4264/**
4265 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4266 * @entry: swap entry to be moved
4267 * @from: mem_cgroup which the entry is moved from
4268 * @to: mem_cgroup which the entry is moved to
4269 *
4270 * It succeeds only when the swap_cgroup's record for this entry is the same
4271 * as the mem_cgroup's id of @from.
4272 *
4273 * Returns 0 on success, -EINVAL on failure.
4274 *
4275 * The caller must have charged to @to, IOW, called res_counter_charge() about
4276 * both res and memsw, and called css_get().
4277 */
4278static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004279 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004280{
4281 unsigned short old_id, new_id;
4282
Li Zefan34c00c32013-09-23 16:56:01 +08004283 old_id = mem_cgroup_id(from);
4284 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004285
4286 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004287 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004288 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004289 /*
4290 * This function is only called from task migration context now.
4291 * It postpones res_counter and refcount handling till the end
4292 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004293 * improvement. But we cannot postpone css_get(to) because if
4294 * the process that has been moved to @to does swap-in, the
4295 * refcount of @to might be decreased to 0.
4296 *
4297 * We are in attach() phase, so the cgroup is guaranteed to be
4298 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004299 */
Li Zefan40503772013-07-08 16:00:34 -07004300 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004301 return 0;
4302 }
4303 return -EINVAL;
4304}
4305#else
4306static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004307 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004308{
4309 return -EINVAL;
4310}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004311#endif
4312
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004313/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004314 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4315 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004316 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004317void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4318 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004319{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004320 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004321 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004322 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004323 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004324
Johannes Weiner72835c82012-01-12 17:18:32 -08004325 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004326
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004327 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004328 return;
Balbir Singh40779602008-04-04 14:29:59 -07004329
Mel Gormanb32967f2012-11-19 12:35:47 +00004330 if (PageTransHuge(page))
4331 nr_pages <<= compound_order(page);
4332
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004333 pc = lookup_page_cgroup(page);
4334 lock_page_cgroup(pc);
4335 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004336 memcg = pc->mem_cgroup;
4337 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004338 /*
4339 * At migrating an anonymous page, its mapcount goes down
4340 * to 0 and uncharge() will be called. But, even if it's fully
4341 * unmapped, migration may fail and this page has to be
4342 * charged again. We set MIGRATION flag here and delay uncharge
4343 * until end_migration() is called
4344 *
4345 * Corner Case Thinking
4346 * A)
4347 * When the old page was mapped as Anon and it's unmap-and-freed
4348 * while migration was ongoing.
4349 * If unmap finds the old page, uncharge() of it will be delayed
4350 * until end_migration(). If unmap finds a new page, it's
4351 * uncharged when it make mapcount to be 1->0. If unmap code
4352 * finds swap_migration_entry, the new page will not be mapped
4353 * and end_migration() will find it(mapcount==0).
4354 *
4355 * B)
4356 * When the old page was mapped but migraion fails, the kernel
4357 * remaps it. A charge for it is kept by MIGRATION flag even
4358 * if mapcount goes down to 0. We can do remap successfully
4359 * without charging it again.
4360 *
4361 * C)
4362 * The "old" page is under lock_page() until the end of
4363 * migration, so, the old page itself will not be swapped-out.
4364 * If the new page is swapped out before end_migraton, our
4365 * hook to usual swap-out path will catch the event.
4366 */
4367 if (PageAnon(page))
4368 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004369 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004370 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004371 /*
4372 * If the page is not charged at this point,
4373 * we return here.
4374 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004375 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004376 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004377
Johannes Weiner72835c82012-01-12 17:18:32 -08004378 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004379 /*
4380 * We charge new page before it's used/mapped. So, even if unlock_page()
4381 * is called before end_migration, we can catch all events on this new
4382 * page. In the case new page is migrated but not remapped, new page's
4383 * mapcount will be finally 0 and we call uncharge in end_migration().
4384 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004385 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004386 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004387 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004388 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004389 /*
4390 * The page is committed to the memcg, but it's not actually
4391 * charged to the res_counter since we plan on replacing the
4392 * old one and only one page is going to be left afterwards.
4393 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004394 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004395}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004396
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004397/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004398void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004399 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004400{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004401 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004402 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004403 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004404
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004405 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004406 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004407
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004408 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004409 used = oldpage;
4410 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004411 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004412 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004413 unused = oldpage;
4414 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004415 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004416 __mem_cgroup_uncharge_common(unused,
4417 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4418 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4419 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004420 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004421 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004422 * We disallowed uncharge of pages under migration because mapcount
4423 * of the page goes down to zero, temporarly.
4424 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004425 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004426 pc = lookup_page_cgroup(oldpage);
4427 lock_page_cgroup(pc);
4428 ClearPageCgroupMigration(pc);
4429 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004430
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004431 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004432 * If a page is a file cache, radix-tree replacement is very atomic
4433 * and we can skip this check. When it was an Anon page, its mapcount
4434 * goes down to 0. But because we added MIGRATION flage, it's not
4435 * uncharged yet. There are several case but page->mapcount check
4436 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4437 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004438 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004439 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004440 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004441}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004442
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004443/*
4444 * At replace page cache, newpage is not under any memcg but it's on
4445 * LRU. So, this function doesn't touch res_counter but handles LRU
4446 * in correct way. Both pages are locked so we cannot race with uncharge.
4447 */
4448void mem_cgroup_replace_page_cache(struct page *oldpage,
4449 struct page *newpage)
4450{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004451 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004452 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004453 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004454
4455 if (mem_cgroup_disabled())
4456 return;
4457
4458 pc = lookup_page_cgroup(oldpage);
4459 /* fix accounting on old pages */
4460 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004461 if (PageCgroupUsed(pc)) {
4462 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004463 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004464 ClearPageCgroupUsed(pc);
4465 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004466 unlock_page_cgroup(pc);
4467
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004468 /*
4469 * When called from shmem_replace_page(), in some cases the
4470 * oldpage has already been charged, and in some cases not.
4471 */
4472 if (!memcg)
4473 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004474 /*
4475 * Even if newpage->mapping was NULL before starting replacement,
4476 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4477 * LRU while we overwrite pc->mem_cgroup.
4478 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004479 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004480}
4481
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004482#ifdef CONFIG_DEBUG_VM
4483static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4484{
4485 struct page_cgroup *pc;
4486
4487 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004488 /*
4489 * Can be NULL while feeding pages into the page allocator for
4490 * the first time, i.e. during boot or memory hotplug;
4491 * or when mem_cgroup_disabled().
4492 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004493 if (likely(pc) && PageCgroupUsed(pc))
4494 return pc;
4495 return NULL;
4496}
4497
4498bool mem_cgroup_bad_page_check(struct page *page)
4499{
4500 if (mem_cgroup_disabled())
4501 return false;
4502
4503 return lookup_page_cgroup_used(page) != NULL;
4504}
4505
4506void mem_cgroup_print_bad_page(struct page *page)
4507{
4508 struct page_cgroup *pc;
4509
4510 pc = lookup_page_cgroup_used(page);
4511 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004512 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4513 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004514 }
4515}
4516#endif
4517
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004518static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004519 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004520{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004521 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004522 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004523 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004524 int children = mem_cgroup_count_children(memcg);
4525 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004526 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004527
4528 /*
4529 * For keeping hierarchical_reclaim simple, how long we should retry
4530 * is depends on callers. We set our retry-count to be function
4531 * of # of children which we should visit in this loop.
4532 */
4533 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4534
4535 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004536
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004537 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004538 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004539 if (signal_pending(current)) {
4540 ret = -EINTR;
4541 break;
4542 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004543 /*
4544 * Rather than hide all in some function, I do this in
4545 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004546 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004547 */
4548 mutex_lock(&set_limit_mutex);
4549 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4550 if (memswlimit < val) {
4551 ret = -EINVAL;
4552 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004553 break;
4554 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004555
4556 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4557 if (memlimit < val)
4558 enlarge = 1;
4559
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004560 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004561 if (!ret) {
4562 if (memswlimit == val)
4563 memcg->memsw_is_minimum = true;
4564 else
4565 memcg->memsw_is_minimum = false;
4566 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004567 mutex_unlock(&set_limit_mutex);
4568
4569 if (!ret)
4570 break;
4571
Johannes Weiner56600482012-01-12 17:17:59 -08004572 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4573 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004574 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4575 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004576 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004577 retry_count--;
4578 else
4579 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004580 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004581 if (!ret && enlarge)
4582 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004583
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004584 return ret;
4585}
4586
Li Zefan338c8432009-06-17 16:27:15 -07004587static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4588 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004589{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004590 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004591 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004592 int children = mem_cgroup_count_children(memcg);
4593 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004594 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004595
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004596 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004597 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004598 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004599 while (retry_count) {
4600 if (signal_pending(current)) {
4601 ret = -EINTR;
4602 break;
4603 }
4604 /*
4605 * Rather than hide all in some function, I do this in
4606 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004607 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004608 */
4609 mutex_lock(&set_limit_mutex);
4610 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4611 if (memlimit > val) {
4612 ret = -EINVAL;
4613 mutex_unlock(&set_limit_mutex);
4614 break;
4615 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004616 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4617 if (memswlimit < val)
4618 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004619 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004620 if (!ret) {
4621 if (memlimit == val)
4622 memcg->memsw_is_minimum = true;
4623 else
4624 memcg->memsw_is_minimum = false;
4625 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004626 mutex_unlock(&set_limit_mutex);
4627
4628 if (!ret)
4629 break;
4630
Johannes Weiner56600482012-01-12 17:17:59 -08004631 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4632 MEM_CGROUP_RECLAIM_NOSWAP |
4633 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004634 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004635 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004636 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004637 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004638 else
4639 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004640 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004641 if (!ret && enlarge)
4642 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004643 return ret;
4644}
4645
Andrew Morton0608f432013-09-24 15:27:41 -07004646unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
4647 gfp_t gfp_mask,
4648 unsigned long *total_scanned)
4649{
4650 unsigned long nr_reclaimed = 0;
4651 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4652 unsigned long reclaimed;
4653 int loop = 0;
4654 struct mem_cgroup_tree_per_zone *mctz;
4655 unsigned long long excess;
4656 unsigned long nr_scanned;
4657
4658 if (order > 0)
4659 return 0;
4660
4661 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
4662 /*
4663 * This loop can run a while, specially if mem_cgroup's continuously
4664 * keep exceeding their soft limit and putting the system under
4665 * pressure
4666 */
4667 do {
4668 if (next_mz)
4669 mz = next_mz;
4670 else
4671 mz = mem_cgroup_largest_soft_limit_node(mctz);
4672 if (!mz)
4673 break;
4674
4675 nr_scanned = 0;
4676 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
4677 gfp_mask, &nr_scanned);
4678 nr_reclaimed += reclaimed;
4679 *total_scanned += nr_scanned;
4680 spin_lock(&mctz->lock);
4681
4682 /*
4683 * If we failed to reclaim anything from this memory cgroup
4684 * it is time to move on to the next cgroup
4685 */
4686 next_mz = NULL;
4687 if (!reclaimed) {
4688 do {
4689 /*
4690 * Loop until we find yet another one.
4691 *
4692 * By the time we get the soft_limit lock
4693 * again, someone might have aded the
4694 * group back on the RB tree. Iterate to
4695 * make sure we get a different mem.
4696 * mem_cgroup_largest_soft_limit_node returns
4697 * NULL if no other cgroup is present on
4698 * the tree
4699 */
4700 next_mz =
4701 __mem_cgroup_largest_soft_limit_node(mctz);
4702 if (next_mz == mz)
4703 css_put(&next_mz->memcg->css);
4704 else /* next_mz == NULL or other memcg */
4705 break;
4706 } while (1);
4707 }
4708 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4709 excess = res_counter_soft_limit_excess(&mz->memcg->res);
4710 /*
4711 * One school of thought says that we should not add
4712 * back the node to the tree if reclaim returns 0.
4713 * But our reclaim could return 0, simply because due
4714 * to priority we are exposing a smaller subset of
4715 * memory to reclaim from. Consider this as a longer
4716 * term TODO.
4717 */
4718 /* If excess == 0, no tree ops */
4719 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
4720 spin_unlock(&mctz->lock);
4721 css_put(&mz->memcg->css);
4722 loop++;
4723 /*
4724 * Could not reclaim anything and there are no more
4725 * mem cgroups to try or we seem to be looping without
4726 * reclaiming anything.
4727 */
4728 if (!nr_reclaimed &&
4729 (next_mz == NULL ||
4730 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4731 break;
4732 } while (!nr_reclaimed);
4733 if (next_mz)
4734 css_put(&next_mz->memcg->css);
4735 return nr_reclaimed;
4736}
4737
Michal Hocko2ef37d32012-10-26 13:37:30 +02004738/**
4739 * mem_cgroup_force_empty_list - clears LRU of a group
4740 * @memcg: group to clear
4741 * @node: NUMA node
4742 * @zid: zone id
4743 * @lru: lru to to clear
4744 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004745 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004746 * reclaim the pages page themselves - pages are moved to the parent (or root)
4747 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004748 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004749static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004750 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004751{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004752 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004753 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004754 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004755 struct page *busy;
4756 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004757
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004758 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004759 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4760 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004761
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004762 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004763 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004764 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004765 struct page *page;
4766
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004767 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004768 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004769 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004770 break;
4771 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004772 page = list_entry(list->prev, struct page, lru);
4773 if (busy == page) {
4774 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004775 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004776 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004777 continue;
4778 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004779 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004780
Johannes Weiner925b7672012-01-12 17:18:15 -08004781 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004782
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004783 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004784 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004785 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004786 cond_resched();
4787 } else
4788 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004789 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004790}
4791
4792/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004793 * make mem_cgroup's charge to be 0 if there is no task by moving
4794 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004795 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004796 *
4797 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004798 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004799static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004800{
Michal Hockoc26251f2012-10-26 13:37:28 +02004801 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004802 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004803
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004804 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004805 /* This is for making all *used* pages to be on LRU. */
4806 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004807 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004808 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004809 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004810 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004811 enum lru_list lru;
4812 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004813 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004814 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004815 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004816 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004817 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004818 mem_cgroup_end_move(memcg);
4819 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004820 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004821
Michal Hocko2ef37d32012-10-26 13:37:30 +02004822 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004823 * Kernel memory may not necessarily be trackable to a specific
4824 * process. So they are not migrated, and therefore we can't
4825 * expect their value to drop to 0 here.
4826 * Having res filled up with kmem only is enough.
4827 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004828 * This is a safety check because mem_cgroup_force_empty_list
4829 * could have raced with mem_cgroup_replace_page_cache callers
4830 * so the lru seemed empty but the page could have been added
4831 * right after the check. RES_USAGE should be safe as we always
4832 * charge before adding to the LRU.
4833 */
Glauber Costabea207c2012-12-18 14:22:11 -08004834 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4835 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4836 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004837}
4838
Glauber Costab5f99b52013-02-22 16:34:53 -08004839static inline bool memcg_has_children(struct mem_cgroup *memcg)
4840{
Johannes Weiner696ac172013-10-31 16:34:15 -07004841 lockdep_assert_held(&memcg_create_mutex);
4842 /*
4843 * The lock does not prevent addition or deletion to the list
4844 * of children, but it prevents a new child from being
4845 * initialized based on this parent in css_online(), so it's
4846 * enough to decide whether hierarchically inherited
4847 * attributes can still be changed or not.
4848 */
4849 return memcg->use_hierarchy &&
4850 !list_empty(&memcg->css.cgroup->children);
Glauber Costab5f99b52013-02-22 16:34:53 -08004851}
4852
4853/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004854 * Reclaims as many pages from the given memcg as possible and moves
4855 * the rest to the parent.
4856 *
4857 * Caller is responsible for holding css reference for memcg.
4858 */
4859static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4860{
4861 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4862 struct cgroup *cgrp = memcg->css.cgroup;
4863
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004864 /* returns EBUSY if there is a task or if we come here twice. */
Tejun Heo07bc3562014-02-13 06:58:39 -05004865 if (cgroup_has_tasks(cgrp) || !list_empty(&cgrp->children))
Michal Hockoc26251f2012-10-26 13:37:28 +02004866 return -EBUSY;
4867
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004868 /* we call try-to-free pages for make this cgroup empty */
4869 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004870 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004871 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004872 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004873
Michal Hockoc26251f2012-10-26 13:37:28 +02004874 if (signal_pending(current))
4875 return -EINTR;
4876
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004877 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07004878 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004879 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004880 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004881 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02004882 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004883 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004884
4885 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004886 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02004887 mem_cgroup_reparent_charges(memcg);
4888
4889 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004890}
4891
Tejun Heo182446d2013-08-08 20:11:24 -04004892static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
4893 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004894{
Tejun Heo182446d2013-08-08 20:11:24 -04004895 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02004896
Michal Hockod8423012012-10-26 13:37:29 +02004897 if (mem_cgroup_is_root(memcg))
4898 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07004899 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004900}
4901
Tejun Heo182446d2013-08-08 20:11:24 -04004902static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
4903 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004904{
Tejun Heo182446d2013-08-08 20:11:24 -04004905 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004906}
4907
Tejun Heo182446d2013-08-08 20:11:24 -04004908static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
4909 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004910{
4911 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04004912 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04004913 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08004914
Glauber Costa09998212013-02-22 16:34:55 -08004915 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07004916
4917 if (memcg->use_hierarchy == val)
4918 goto out;
4919
Balbir Singh18f59ea2009-01-07 18:08:07 -08004920 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02004921 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08004922 * in the child subtrees. If it is unset, then the change can
4923 * occur, provided the current cgroup has no children.
4924 *
4925 * For the root cgroup, parent_mem is NULL, we allow value to be
4926 * set if there are no children.
4927 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004928 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08004929 (val == 1 || val == 0)) {
Johannes Weiner696ac172013-10-31 16:34:15 -07004930 if (list_empty(&memcg->css.cgroup->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004931 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004932 else
4933 retval = -EBUSY;
4934 } else
4935 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07004936
4937out:
Glauber Costa09998212013-02-22 16:34:55 -08004938 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004939
4940 return retval;
4941}
4942
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004943
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004944static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004945 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004946{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004947 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004948 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004949
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004950 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004951 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004952 val += mem_cgroup_read_stat(iter, idx);
4953
4954 if (val < 0) /* race ? */
4955 val = 0;
4956 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004957}
4958
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004959static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004960{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004961 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004962
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004963 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004964 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00004965 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004966 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00004967 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004968 }
4969
David Rientjesb070e652013-05-07 16:18:09 -07004970 /*
4971 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
4972 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
4973 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004974 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
4975 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004976
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004977 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004978 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004979
4980 return val << PAGE_SHIFT;
4981}
4982
Tejun Heo791badb2013-12-05 12:28:02 -05004983static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
4984 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004985{
Tejun Heo182446d2013-08-08 20:11:24 -04004986 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004987 u64 val;
Tejun Heo791badb2013-12-05 12:28:02 -05004988 int name;
Glauber Costa86ae53e2012-12-18 14:21:45 -08004989 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004990
4991 type = MEMFILE_TYPE(cft->private);
4992 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07004993
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004994 switch (type) {
4995 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004996 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004997 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004998 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004999 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005000 break;
5001 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005002 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005003 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005004 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005005 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005006 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005007 case _KMEM:
5008 val = res_counter_read_u64(&memcg->kmem, name);
5009 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005010 default:
5011 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005012 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005013
Tejun Heo791badb2013-12-05 12:28:02 -05005014 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005015}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005016
Glauber Costa510fc4e2012-12-18 14:21:47 -08005017#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08005018/* should be called with activate_kmem_mutex held */
5019static int __memcg_activate_kmem(struct mem_cgroup *memcg,
5020 unsigned long long limit)
5021{
5022 int err = 0;
5023 int memcg_id;
5024
5025 if (memcg_kmem_is_active(memcg))
5026 return 0;
5027
5028 /*
5029 * We are going to allocate memory for data shared by all memory
5030 * cgroups so let's stop accounting here.
5031 */
5032 memcg_stop_kmem_account();
5033
Glauber Costa510fc4e2012-12-18 14:21:47 -08005034 /*
5035 * For simplicity, we won't allow this to be disabled. It also can't
5036 * be changed if the cgroup has children already, or if tasks had
5037 * already joined.
5038 *
5039 * If tasks join before we set the limit, a person looking at
5040 * kmem.usage_in_bytes will have no way to determine when it took
5041 * place, which makes the value quite meaningless.
5042 *
5043 * After it first became limited, changes in the value of the limit are
5044 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005045 */
Glauber Costa09998212013-02-22 16:34:55 -08005046 mutex_lock(&memcg_create_mutex);
Tejun Heo07bc3562014-02-13 06:58:39 -05005047 if (cgroup_has_tasks(memcg->css.cgroup) || memcg_has_children(memcg))
Vladimir Davydovd6441632014-01-23 15:53:09 -08005048 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08005049 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005050 if (err)
5051 goto out;
5052
5053 memcg_id = ida_simple_get(&kmem_limited_groups,
5054 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
5055 if (memcg_id < 0) {
5056 err = memcg_id;
5057 goto out;
5058 }
5059
5060 /*
5061 * Make sure we have enough space for this cgroup in each root cache's
5062 * memcg_params.
5063 */
5064 err = memcg_update_all_caches(memcg_id + 1);
5065 if (err)
5066 goto out_rmid;
5067
5068 memcg->kmemcg_id = memcg_id;
5069 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
5070 mutex_init(&memcg->slab_caches_mutex);
5071
5072 /*
5073 * We couldn't have accounted to this cgroup, because it hasn't got the
5074 * active bit set yet, so this should succeed.
5075 */
5076 err = res_counter_set_limit(&memcg->kmem, limit);
5077 VM_BUG_ON(err);
5078
5079 static_key_slow_inc(&memcg_kmem_enabled_key);
5080 /*
5081 * Setting the active bit after enabling static branching will
5082 * guarantee no one starts accounting before all call sites are
5083 * patched.
5084 */
5085 memcg_kmem_set_active(memcg);
5086out:
5087 memcg_resume_kmem_account();
5088 return err;
5089
5090out_rmid:
5091 ida_simple_remove(&kmem_limited_groups, memcg_id);
5092 goto out;
5093}
5094
5095static int memcg_activate_kmem(struct mem_cgroup *memcg,
5096 unsigned long long limit)
5097{
5098 int ret;
5099
5100 mutex_lock(&activate_kmem_mutex);
5101 ret = __memcg_activate_kmem(memcg, limit);
5102 mutex_unlock(&activate_kmem_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005103 return ret;
5104}
5105
Vladimir Davydovd6441632014-01-23 15:53:09 -08005106static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5107 unsigned long long val)
5108{
5109 int ret;
5110
5111 if (!memcg_kmem_is_active(memcg))
5112 ret = memcg_activate_kmem(memcg, val);
5113 else
5114 ret = res_counter_set_limit(&memcg->kmem, val);
5115 return ret;
5116}
5117
Glauber Costa55007d82012-12-18 14:22:38 -08005118static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005119{
Glauber Costa55007d82012-12-18 14:22:38 -08005120 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005121 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005122
Glauber Costa510fc4e2012-12-18 14:21:47 -08005123 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005124 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08005125
Vladimir Davydovd6441632014-01-23 15:53:09 -08005126 mutex_lock(&activate_kmem_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08005127 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08005128 * If the parent cgroup is not kmem-active now, it cannot be activated
5129 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08005130 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08005131 if (memcg_kmem_is_active(parent))
5132 ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX);
5133 mutex_unlock(&activate_kmem_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005134 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005135}
Vladimir Davydovd6441632014-01-23 15:53:09 -08005136#else
5137static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5138 unsigned long long val)
5139{
5140 return -EINVAL;
5141}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005142#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005143
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005144/*
5145 * The user of this function is...
5146 * RES_LIMIT.
5147 */
Tejun Heo182446d2013-08-08 20:11:24 -04005148static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Tejun Heo4d3bb512014-03-19 10:23:54 -04005149 char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005150{
Tejun Heo182446d2013-08-08 20:11:24 -04005151 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005152 enum res_type type;
5153 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005154 unsigned long long val;
5155 int ret;
5156
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005157 type = MEMFILE_TYPE(cft->private);
5158 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005159
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005160 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005161 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005162 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5163 ret = -EINVAL;
5164 break;
5165 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005166 /* This function does all necessary parse...reuse it */
5167 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005168 if (ret)
5169 break;
5170 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005171 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005172 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005173 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005174 else if (type == _KMEM)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005175 ret = memcg_update_kmem_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005176 else
5177 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005178 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005179 case RES_SOFT_LIMIT:
5180 ret = res_counter_memparse_write_strategy(buffer, &val);
5181 if (ret)
5182 break;
5183 /*
5184 * For memsw, soft limits are hard to implement in terms
5185 * of semantics, for now, we support soft limits for
5186 * control without swap
5187 */
5188 if (type == _MEM)
5189 ret = res_counter_set_soft_limit(&memcg->res, val);
5190 else
5191 ret = -EINVAL;
5192 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005193 default:
5194 ret = -EINVAL; /* should be BUG() ? */
5195 break;
5196 }
5197 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005198}
5199
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005200static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5201 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5202{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005203 unsigned long long min_limit, min_memsw_limit, tmp;
5204
5205 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5206 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005207 if (!memcg->use_hierarchy)
5208 goto out;
5209
Tejun Heo63876982013-08-08 20:11:23 -04005210 while (css_parent(&memcg->css)) {
5211 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005212 if (!memcg->use_hierarchy)
5213 break;
5214 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5215 min_limit = min(min_limit, tmp);
5216 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5217 min_memsw_limit = min(min_memsw_limit, tmp);
5218 }
5219out:
5220 *mem_limit = min_limit;
5221 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005222}
5223
Tejun Heo182446d2013-08-08 20:11:24 -04005224static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005225{
Tejun Heo182446d2013-08-08 20:11:24 -04005226 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005227 int name;
5228 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005229
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005230 type = MEMFILE_TYPE(event);
5231 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005232
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005233 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005234 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005235 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005236 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005237 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005238 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005239 else if (type == _KMEM)
5240 res_counter_reset_max(&memcg->kmem);
5241 else
5242 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005243 break;
5244 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005245 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005246 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005247 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005248 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005249 else if (type == _KMEM)
5250 res_counter_reset_failcnt(&memcg->kmem);
5251 else
5252 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005253 break;
5254 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005255
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005256 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005257}
5258
Tejun Heo182446d2013-08-08 20:11:24 -04005259static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005260 struct cftype *cft)
5261{
Tejun Heo182446d2013-08-08 20:11:24 -04005262 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005263}
5264
Daisuke Nishimura02491442010-03-10 15:22:17 -08005265#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005266static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005267 struct cftype *cft, u64 val)
5268{
Tejun Heo182446d2013-08-08 20:11:24 -04005269 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005270
5271 if (val >= (1 << NR_MOVE_TYPE))
5272 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005273
Glauber Costaee5e8472013-02-22 16:34:50 -08005274 /*
5275 * No kind of locking is needed in here, because ->can_attach() will
5276 * check this value once in the beginning of the process, and then carry
5277 * on with stale data. This means that changes to this value will only
5278 * affect task migrations starting after the change.
5279 */
5280 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005281 return 0;
5282}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005283#else
Tejun Heo182446d2013-08-08 20:11:24 -04005284static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005285 struct cftype *cft, u64 val)
5286{
5287 return -ENOSYS;
5288}
5289#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005290
Ying Han406eb0c2011-05-26 16:25:37 -07005291#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05005292static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07005293{
Greg Thelen25485de2013-11-12 15:07:40 -08005294 struct numa_stat {
5295 const char *name;
5296 unsigned int lru_mask;
5297 };
5298
5299 static const struct numa_stat stats[] = {
5300 { "total", LRU_ALL },
5301 { "file", LRU_ALL_FILE },
5302 { "anon", LRU_ALL_ANON },
5303 { "unevictable", BIT(LRU_UNEVICTABLE) },
5304 };
5305 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07005306 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08005307 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05005308 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07005309
Greg Thelen25485de2013-11-12 15:07:40 -08005310 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5311 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
5312 seq_printf(m, "%s=%lu", stat->name, nr);
5313 for_each_node_state(nid, N_MEMORY) {
5314 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5315 stat->lru_mask);
5316 seq_printf(m, " N%d=%lu", nid, nr);
5317 }
5318 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005319 }
Ying Han406eb0c2011-05-26 16:25:37 -07005320
Ying Han071aee12013-11-12 15:07:41 -08005321 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5322 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07005323
Ying Han071aee12013-11-12 15:07:41 -08005324 nr = 0;
5325 for_each_mem_cgroup_tree(iter, memcg)
5326 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
5327 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
5328 for_each_node_state(nid, N_MEMORY) {
5329 nr = 0;
5330 for_each_mem_cgroup_tree(iter, memcg)
5331 nr += mem_cgroup_node_nr_lru_pages(
5332 iter, nid, stat->lru_mask);
5333 seq_printf(m, " N%d=%lu", nid, nr);
5334 }
5335 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005336 }
Ying Han406eb0c2011-05-26 16:25:37 -07005337
Ying Han406eb0c2011-05-26 16:25:37 -07005338 return 0;
5339}
5340#endif /* CONFIG_NUMA */
5341
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005342static inline void mem_cgroup_lru_names_not_uptodate(void)
5343{
5344 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5345}
5346
Tejun Heo2da8ca82013-12-05 12:28:04 -05005347static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005348{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005349 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005350 struct mem_cgroup *mi;
5351 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005352
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005353 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005354 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005355 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005356 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5357 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005358 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005359
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005360 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5361 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5362 mem_cgroup_read_events(memcg, i));
5363
5364 for (i = 0; i < NR_LRU_LISTS; i++)
5365 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5366 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5367
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005368 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005369 {
5370 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005371 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005372 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005373 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005374 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5375 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005376 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005377
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005378 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5379 long long val = 0;
5380
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005381 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005382 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005383 for_each_mem_cgroup_tree(mi, memcg)
5384 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5385 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5386 }
5387
5388 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5389 unsigned long long val = 0;
5390
5391 for_each_mem_cgroup_tree(mi, memcg)
5392 val += mem_cgroup_read_events(mi, i);
5393 seq_printf(m, "total_%s %llu\n",
5394 mem_cgroup_events_names[i], val);
5395 }
5396
5397 for (i = 0; i < NR_LRU_LISTS; i++) {
5398 unsigned long long val = 0;
5399
5400 for_each_mem_cgroup_tree(mi, memcg)
5401 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5402 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005403 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005404
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005405#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005406 {
5407 int nid, zid;
5408 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005409 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005410 unsigned long recent_rotated[2] = {0, 0};
5411 unsigned long recent_scanned[2] = {0, 0};
5412
5413 for_each_online_node(nid)
5414 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005415 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005416 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005417
Hugh Dickins89abfab2012-05-29 15:06:53 -07005418 recent_rotated[0] += rstat->recent_rotated[0];
5419 recent_rotated[1] += rstat->recent_rotated[1];
5420 recent_scanned[0] += rstat->recent_scanned[0];
5421 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005422 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005423 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5424 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5425 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5426 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005427 }
5428#endif
5429
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005430 return 0;
5431}
5432
Tejun Heo182446d2013-08-08 20:11:24 -04005433static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5434 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005435{
Tejun Heo182446d2013-08-08 20:11:24 -04005436 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005437
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005438 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005439}
5440
Tejun Heo182446d2013-08-08 20:11:24 -04005441static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5442 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005443{
Tejun Heo182446d2013-08-08 20:11:24 -04005444 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005445 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005446
Tejun Heo63876982013-08-08 20:11:23 -04005447 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005448 return -EINVAL;
5449
Glauber Costa09998212013-02-22 16:34:55 -08005450 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005451
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005452 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005453 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005454 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005455 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005456 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005457
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005458 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005459
Glauber Costa09998212013-02-22 16:34:55 -08005460 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005461
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005462 return 0;
5463}
5464
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005465static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5466{
5467 struct mem_cgroup_threshold_ary *t;
5468 u64 usage;
5469 int i;
5470
5471 rcu_read_lock();
5472 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005473 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005474 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005475 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005476
5477 if (!t)
5478 goto unlock;
5479
5480 usage = mem_cgroup_usage(memcg, swap);
5481
5482 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005483 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005484 * If it's not true, a threshold was crossed after last
5485 * call of __mem_cgroup_threshold().
5486 */
Phil Carmody5407a562010-05-26 14:42:42 -07005487 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005488
5489 /*
5490 * Iterate backward over array of thresholds starting from
5491 * current_threshold and check if a threshold is crossed.
5492 * If none of thresholds below usage is crossed, we read
5493 * only one element of the array here.
5494 */
5495 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5496 eventfd_signal(t->entries[i].eventfd, 1);
5497
5498 /* i = current_threshold + 1 */
5499 i++;
5500
5501 /*
5502 * Iterate forward over array of thresholds starting from
5503 * current_threshold+1 and check if a threshold is crossed.
5504 * If none of thresholds above usage is crossed, we read
5505 * only one element of the array here.
5506 */
5507 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5508 eventfd_signal(t->entries[i].eventfd, 1);
5509
5510 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005511 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005512unlock:
5513 rcu_read_unlock();
5514}
5515
5516static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5517{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005518 while (memcg) {
5519 __mem_cgroup_threshold(memcg, false);
5520 if (do_swap_account)
5521 __mem_cgroup_threshold(memcg, true);
5522
5523 memcg = parent_mem_cgroup(memcg);
5524 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005525}
5526
5527static int compare_thresholds(const void *a, const void *b)
5528{
5529 const struct mem_cgroup_threshold *_a = a;
5530 const struct mem_cgroup_threshold *_b = b;
5531
Greg Thelen2bff24a2013-09-11 14:23:08 -07005532 if (_a->threshold > _b->threshold)
5533 return 1;
5534
5535 if (_a->threshold < _b->threshold)
5536 return -1;
5537
5538 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005539}
5540
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005541static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005542{
5543 struct mem_cgroup_eventfd_list *ev;
5544
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005545 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005546 eventfd_signal(ev->eventfd, 1);
5547 return 0;
5548}
5549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005550static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005551{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005552 struct mem_cgroup *iter;
5553
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005554 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005555 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005556}
5557
Tejun Heo59b6f872013-11-22 18:20:43 -05005558static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005559 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005560{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005561 struct mem_cgroup_thresholds *thresholds;
5562 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005563 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005564 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005565
5566 ret = res_counter_memparse_write_strategy(args, &threshold);
5567 if (ret)
5568 return ret;
5569
5570 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005571
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005572 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005573 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005574 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005575 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005576 else
5577 BUG();
5578
5579 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5580
5581 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005582 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005583 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5584
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005585 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005586
5587 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005588 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005589 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005590 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005591 ret = -ENOMEM;
5592 goto unlock;
5593 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005594 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005595
5596 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005597 if (thresholds->primary) {
5598 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005599 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005600 }
5601
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005602 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005603 new->entries[size - 1].eventfd = eventfd;
5604 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005605
5606 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005607 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005608 compare_thresholds, NULL);
5609
5610 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005611 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005612 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005613 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005614 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005615 * new->current_threshold will not be used until
5616 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005617 * it here.
5618 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005619 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005620 } else
5621 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005622 }
5623
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005624 /* Free old spare buffer and save old primary buffer as spare */
5625 kfree(thresholds->spare);
5626 thresholds->spare = thresholds->primary;
5627
5628 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005629
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005630 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005631 synchronize_rcu();
5632
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005633unlock:
5634 mutex_unlock(&memcg->thresholds_lock);
5635
5636 return ret;
5637}
5638
Tejun Heo59b6f872013-11-22 18:20:43 -05005639static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005640 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005641{
Tejun Heo59b6f872013-11-22 18:20:43 -05005642 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005643}
5644
Tejun Heo59b6f872013-11-22 18:20:43 -05005645static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005646 struct eventfd_ctx *eventfd, const char *args)
5647{
Tejun Heo59b6f872013-11-22 18:20:43 -05005648 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005649}
5650
Tejun Heo59b6f872013-11-22 18:20:43 -05005651static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005652 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005653{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005654 struct mem_cgroup_thresholds *thresholds;
5655 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005656 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005657 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005658
5659 mutex_lock(&memcg->thresholds_lock);
5660 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005661 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005662 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005663 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005664 else
5665 BUG();
5666
Anton Vorontsov371528c2012-02-24 05:14:46 +04005667 if (!thresholds->primary)
5668 goto unlock;
5669
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005670 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5671
5672 /* Check if a threshold crossed before removing */
5673 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5674
5675 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005676 size = 0;
5677 for (i = 0; i < thresholds->primary->size; i++) {
5678 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005679 size++;
5680 }
5681
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005682 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005683
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005684 /* Set thresholds array to NULL if we don't have thresholds */
5685 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005686 kfree(new);
5687 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005688 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005689 }
5690
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005691 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005692
5693 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005694 new->current_threshold = -1;
5695 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5696 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005697 continue;
5698
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005699 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005700 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005701 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005702 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005703 * until rcu_assign_pointer(), so it's safe to increment
5704 * it here.
5705 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005706 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005707 }
5708 j++;
5709 }
5710
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005711swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005712 /* Swap primary and spare array */
5713 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005714 /* If all events are unregistered, free the spare array */
5715 if (!new) {
5716 kfree(thresholds->spare);
5717 thresholds->spare = NULL;
5718 }
5719
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005720 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005721
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005722 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005723 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005724unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005725 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005726}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005727
Tejun Heo59b6f872013-11-22 18:20:43 -05005728static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005729 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005730{
Tejun Heo59b6f872013-11-22 18:20:43 -05005731 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005732}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005733
Tejun Heo59b6f872013-11-22 18:20:43 -05005734static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005735 struct eventfd_ctx *eventfd)
5736{
Tejun Heo59b6f872013-11-22 18:20:43 -05005737 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005738}
5739
Tejun Heo59b6f872013-11-22 18:20:43 -05005740static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005741 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005742{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005743 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005744
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005745 event = kmalloc(sizeof(*event), GFP_KERNEL);
5746 if (!event)
5747 return -ENOMEM;
5748
Michal Hocko1af8efe2011-07-26 16:08:24 -07005749 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005750
5751 event->eventfd = eventfd;
5752 list_add(&event->list, &memcg->oom_notify);
5753
5754 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005755 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005756 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005757 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005758
5759 return 0;
5760}
5761
Tejun Heo59b6f872013-11-22 18:20:43 -05005762static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005763 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005764{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005765 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005766
Michal Hocko1af8efe2011-07-26 16:08:24 -07005767 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005768
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005769 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005770 if (ev->eventfd == eventfd) {
5771 list_del(&ev->list);
5772 kfree(ev);
5773 }
5774 }
5775
Michal Hocko1af8efe2011-07-26 16:08:24 -07005776 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005777}
5778
Tejun Heo2da8ca82013-12-05 12:28:04 -05005779static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005780{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005781 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005782
Tejun Heo791badb2013-12-05 12:28:02 -05005783 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
5784 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005785 return 0;
5786}
5787
Tejun Heo182446d2013-08-08 20:11:24 -04005788static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005789 struct cftype *cft, u64 val)
5790{
Tejun Heo182446d2013-08-08 20:11:24 -04005791 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005792 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005793
5794 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005795 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005796 return -EINVAL;
5797
Glauber Costa09998212013-02-22 16:34:55 -08005798 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005799 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005800 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005801 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005802 return -EINVAL;
5803 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005804 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005805 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005806 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005807 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005808 return 0;
5809}
5810
Andrew Mortonc255a452012-07-31 16:43:02 -07005811#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005812static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005813{
Glauber Costa55007d82012-12-18 14:22:38 -08005814 int ret;
5815
Glauber Costa2633d7a2012-12-18 14:22:34 -08005816 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005817 ret = memcg_propagate_kmem(memcg);
5818 if (ret)
5819 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005820
Glauber Costa1d62e432012-04-09 19:36:33 -03005821 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005822}
Glauber Costae5671df2011-12-11 21:47:01 +00005823
Li Zefan10d5ebf2013-07-08 16:00:33 -07005824static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005825{
Glauber Costa1d62e432012-04-09 19:36:33 -03005826 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005827}
5828
5829static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5830{
5831 if (!memcg_kmem_is_active(memcg))
5832 return;
5833
5834 /*
5835 * kmem charges can outlive the cgroup. In the case of slab
5836 * pages, for instance, a page contain objects from various
5837 * processes. As we prevent from taking a reference for every
5838 * such allocation we have to be careful when doing uncharge
5839 * (see memcg_uncharge_kmem) and here during offlining.
5840 *
5841 * The idea is that that only the _last_ uncharge which sees
5842 * the dead memcg will drop the last reference. An additional
5843 * reference is taken here before the group is marked dead
5844 * which is then paired with css_put during uncharge resp. here.
5845 *
5846 * Although this might sound strange as this path is called from
5847 * css_offline() when the referencemight have dropped down to 0
5848 * and shouldn't be incremented anymore (css_tryget would fail)
5849 * we do not have other options because of the kmem allocations
5850 * lifetime.
5851 */
5852 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005853
5854 memcg_kmem_mark_dead(memcg);
5855
5856 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5857 return;
5858
Glauber Costa7de37682012-12-18 14:22:07 -08005859 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005860 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005861}
Glauber Costae5671df2011-12-11 21:47:01 +00005862#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005863static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005864{
5865 return 0;
5866}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005867
Li Zefan10d5ebf2013-07-08 16:00:33 -07005868static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5869{
5870}
5871
5872static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005873{
5874}
Glauber Costae5671df2011-12-11 21:47:01 +00005875#endif
5876
Tejun Heo79bd9812013-11-22 18:20:42 -05005877/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05005878 * DO NOT USE IN NEW FILES.
5879 *
5880 * "cgroup.event_control" implementation.
5881 *
5882 * This is way over-engineered. It tries to support fully configurable
5883 * events for each user. Such level of flexibility is completely
5884 * unnecessary especially in the light of the planned unified hierarchy.
5885 *
5886 * Please deprecate this and replace with something simpler if at all
5887 * possible.
5888 */
5889
5890/*
Tejun Heo79bd9812013-11-22 18:20:42 -05005891 * Unregister event and free resources.
5892 *
5893 * Gets called from workqueue.
5894 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05005895static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05005896{
Tejun Heo3bc942f2013-11-22 18:20:44 -05005897 struct mem_cgroup_event *event =
5898 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05005899 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005900
5901 remove_wait_queue(event->wqh, &event->wait);
5902
Tejun Heo59b6f872013-11-22 18:20:43 -05005903 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05005904
5905 /* Notify userspace the event is going away. */
5906 eventfd_signal(event->eventfd, 1);
5907
5908 eventfd_ctx_put(event->eventfd);
5909 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05005910 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05005911}
5912
5913/*
5914 * Gets called on POLLHUP on eventfd when user closes it.
5915 *
5916 * Called with wqh->lock held and interrupts disabled.
5917 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05005918static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
5919 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05005920{
Tejun Heo3bc942f2013-11-22 18:20:44 -05005921 struct mem_cgroup_event *event =
5922 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05005923 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005924 unsigned long flags = (unsigned long)key;
5925
5926 if (flags & POLLHUP) {
5927 /*
5928 * If the event has been detached at cgroup removal, we
5929 * can simply return knowing the other side will cleanup
5930 * for us.
5931 *
5932 * We can't race against event freeing since the other
5933 * side will require wqh->lock via remove_wait_queue(),
5934 * which we hold.
5935 */
Tejun Heofba94802013-11-22 18:20:43 -05005936 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05005937 if (!list_empty(&event->list)) {
5938 list_del_init(&event->list);
5939 /*
5940 * We are in atomic context, but cgroup_event_remove()
5941 * may sleep, so we have to call it in workqueue.
5942 */
5943 schedule_work(&event->remove);
5944 }
Tejun Heofba94802013-11-22 18:20:43 -05005945 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05005946 }
5947
5948 return 0;
5949}
5950
Tejun Heo3bc942f2013-11-22 18:20:44 -05005951static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05005952 wait_queue_head_t *wqh, poll_table *pt)
5953{
Tejun Heo3bc942f2013-11-22 18:20:44 -05005954 struct mem_cgroup_event *event =
5955 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05005956
5957 event->wqh = wqh;
5958 add_wait_queue(wqh, &event->wait);
5959}
5960
5961/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05005962 * DO NOT USE IN NEW FILES.
5963 *
Tejun Heo79bd9812013-11-22 18:20:42 -05005964 * Parse input and register new cgroup event handler.
5965 *
5966 * Input must be in format '<event_fd> <control_fd> <args>'.
5967 * Interpretation of args is defined by control file implementation.
5968 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05005969static int memcg_write_event_control(struct cgroup_subsys_state *css,
Tejun Heo4d3bb512014-03-19 10:23:54 -04005970 struct cftype *cft, char *buffer)
Tejun Heo79bd9812013-11-22 18:20:42 -05005971{
Tejun Heofba94802013-11-22 18:20:43 -05005972 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005973 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05005974 struct cgroup_subsys_state *cfile_css;
5975 unsigned int efd, cfd;
5976 struct fd efile;
5977 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05005978 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05005979 char *endp;
5980 int ret;
5981
5982 efd = simple_strtoul(buffer, &endp, 10);
5983 if (*endp != ' ')
5984 return -EINVAL;
5985 buffer = endp + 1;
5986
5987 cfd = simple_strtoul(buffer, &endp, 10);
5988 if ((*endp != ' ') && (*endp != '\0'))
5989 return -EINVAL;
5990 buffer = endp + 1;
5991
5992 event = kzalloc(sizeof(*event), GFP_KERNEL);
5993 if (!event)
5994 return -ENOMEM;
5995
Tejun Heo59b6f872013-11-22 18:20:43 -05005996 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005997 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005998 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
5999 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
6000 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05006001
6002 efile = fdget(efd);
6003 if (!efile.file) {
6004 ret = -EBADF;
6005 goto out_kfree;
6006 }
6007
6008 event->eventfd = eventfd_ctx_fileget(efile.file);
6009 if (IS_ERR(event->eventfd)) {
6010 ret = PTR_ERR(event->eventfd);
6011 goto out_put_efile;
6012 }
6013
6014 cfile = fdget(cfd);
6015 if (!cfile.file) {
6016 ret = -EBADF;
6017 goto out_put_eventfd;
6018 }
6019
6020 /* the process need read permission on control file */
6021 /* AV: shouldn't we check that it's been opened for read instead? */
6022 ret = inode_permission(file_inode(cfile.file), MAY_READ);
6023 if (ret < 0)
6024 goto out_put_cfile;
6025
Tejun Heo79bd9812013-11-22 18:20:42 -05006026 /*
Tejun Heofba94802013-11-22 18:20:43 -05006027 * Determine the event callbacks and set them in @event. This used
6028 * to be done via struct cftype but cgroup core no longer knows
6029 * about these events. The following is crude but the whole thing
6030 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05006031 *
6032 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05006033 */
6034 name = cfile.file->f_dentry->d_name.name;
6035
6036 if (!strcmp(name, "memory.usage_in_bytes")) {
6037 event->register_event = mem_cgroup_usage_register_event;
6038 event->unregister_event = mem_cgroup_usage_unregister_event;
6039 } else if (!strcmp(name, "memory.oom_control")) {
6040 event->register_event = mem_cgroup_oom_register_event;
6041 event->unregister_event = mem_cgroup_oom_unregister_event;
6042 } else if (!strcmp(name, "memory.pressure_level")) {
6043 event->register_event = vmpressure_register_event;
6044 event->unregister_event = vmpressure_unregister_event;
6045 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05006046 event->register_event = memsw_cgroup_usage_register_event;
6047 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05006048 } else {
6049 ret = -EINVAL;
6050 goto out_put_cfile;
6051 }
6052
6053 /*
Tejun Heob5557c42013-11-22 18:20:42 -05006054 * Verify @cfile should belong to @css. Also, remaining events are
6055 * automatically removed on cgroup destruction but the removal is
6056 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05006057 */
Tejun Heo5a17f542014-02-11 11:52:47 -05006058 cfile_css = css_tryget_from_dir(cfile.file->f_dentry->d_parent,
6059 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05006060 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05006061 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05006062 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05006063 if (cfile_css != css) {
6064 css_put(cfile_css);
6065 goto out_put_cfile;
6066 }
Tejun Heo79bd9812013-11-22 18:20:42 -05006067
Tejun Heo59b6f872013-11-22 18:20:43 -05006068 ret = event->register_event(memcg, event->eventfd, buffer);
Tejun Heo79bd9812013-11-22 18:20:42 -05006069 if (ret)
6070 goto out_put_css;
6071
6072 efile.file->f_op->poll(efile.file, &event->pt);
6073
Tejun Heofba94802013-11-22 18:20:43 -05006074 spin_lock(&memcg->event_list_lock);
6075 list_add(&event->list, &memcg->event_list);
6076 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006077
6078 fdput(cfile);
6079 fdput(efile);
6080
6081 return 0;
6082
6083out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05006084 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05006085out_put_cfile:
6086 fdput(cfile);
6087out_put_eventfd:
6088 eventfd_ctx_put(event->eventfd);
6089out_put_efile:
6090 fdput(efile);
6091out_kfree:
6092 kfree(event);
6093
6094 return ret;
6095}
6096
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006097static struct cftype mem_cgroup_files[] = {
6098 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006099 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006100 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006101 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006102 },
6103 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006104 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006105 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006106 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006107 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006108 },
6109 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006110 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006111 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07006112 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006113 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006114 },
6115 {
Balbir Singh296c81d2009-09-23 15:56:36 -07006116 .name = "soft_limit_in_bytes",
6117 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
6118 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006119 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07006120 },
6121 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006122 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006123 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006124 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006125 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006126 },
Balbir Singh8697d332008-02-07 00:13:59 -08006127 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006128 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006129 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006130 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08006131 {
6132 .name = "force_empty",
6133 .trigger = mem_cgroup_force_empty_write,
6134 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08006135 {
6136 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07006137 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08006138 .write_u64 = mem_cgroup_hierarchy_write,
6139 .read_u64 = mem_cgroup_hierarchy_read,
6140 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006141 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05006142 .name = "cgroup.event_control", /* XXX: for compat */
6143 .write_string = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05006144 .flags = CFTYPE_NO_PREFIX,
6145 .mode = S_IWUGO,
6146 },
6147 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006148 .name = "swappiness",
6149 .read_u64 = mem_cgroup_swappiness_read,
6150 .write_u64 = mem_cgroup_swappiness_write,
6151 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006152 {
6153 .name = "move_charge_at_immigrate",
6154 .read_u64 = mem_cgroup_move_charge_read,
6155 .write_u64 = mem_cgroup_move_charge_write,
6156 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006157 {
6158 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006159 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006160 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006161 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6162 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006163 {
6164 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006165 },
Ying Han406eb0c2011-05-26 16:25:37 -07006166#ifdef CONFIG_NUMA
6167 {
6168 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006169 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006170 },
6171#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006172#ifdef CONFIG_MEMCG_KMEM
6173 {
6174 .name = "kmem.limit_in_bytes",
6175 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6176 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006177 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006178 },
6179 {
6180 .name = "kmem.usage_in_bytes",
6181 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006182 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006183 },
6184 {
6185 .name = "kmem.failcnt",
6186 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6187 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006188 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006189 },
6190 {
6191 .name = "kmem.max_usage_in_bytes",
6192 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6193 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006194 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006195 },
Glauber Costa749c5412012-12-18 14:23:01 -08006196#ifdef CONFIG_SLABINFO
6197 {
6198 .name = "kmem.slabinfo",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006199 .seq_show = mem_cgroup_slabinfo_read,
Glauber Costa749c5412012-12-18 14:23:01 -08006200 },
6201#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006202#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006203 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006204};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006205
Michal Hocko2d110852013-02-22 16:34:43 -08006206#ifdef CONFIG_MEMCG_SWAP
6207static struct cftype memsw_cgroup_files[] = {
6208 {
6209 .name = "memsw.usage_in_bytes",
6210 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006211 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006212 },
6213 {
6214 .name = "memsw.max_usage_in_bytes",
6215 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6216 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006217 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006218 },
6219 {
6220 .name = "memsw.limit_in_bytes",
6221 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6222 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006223 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006224 },
6225 {
6226 .name = "memsw.failcnt",
6227 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6228 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006229 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006230 },
6231 { }, /* terminate */
6232};
6233#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006234static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006235{
6236 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006237 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006238 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006239 /*
6240 * This routine is called against possible nodes.
6241 * But it's BUG to call kmalloc() against offline node.
6242 *
6243 * TODO: this routine can waste much memory for nodes which will
6244 * never be onlined. It's better to use memory hotplug callback
6245 * function.
6246 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006247 if (!node_state(node, N_NORMAL_MEMORY))
6248 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006249 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006250 if (!pn)
6251 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006252
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006253 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6254 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006255 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006256 mz->usage_in_excess = 0;
6257 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006258 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006259 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006260 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006261 return 0;
6262}
6263
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006264static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006265{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006266 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006267}
6268
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006269static struct mem_cgroup *mem_cgroup_alloc(void)
6270{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006271 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006272 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006273
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006274 size = sizeof(struct mem_cgroup);
6275 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006276
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006277 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006278 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006279 return NULL;
6280
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006281 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6282 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006283 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006284 spin_lock_init(&memcg->pcp_counter_lock);
6285 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006286
6287out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006288 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006289 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006290}
6291
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006292/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006293 * At destroying mem_cgroup, references from swap_cgroup can remain.
6294 * (scanning all at force_empty is too costly...)
6295 *
6296 * Instead of clearing all references at force_empty, we remember
6297 * the number of reference from swap_cgroup and free mem_cgroup when
6298 * it goes down to 0.
6299 *
6300 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006301 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006302
6303static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006304{
Glauber Costac8b2a362012-12-18 14:22:13 -08006305 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07006306
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006307 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006308
6309 for_each_node(node)
6310 free_mem_cgroup_per_zone_info(memcg, node);
6311
6312 free_percpu(memcg->stat);
6313
Glauber Costa3f134612012-05-29 15:07:11 -07006314 /*
6315 * We need to make sure that (at least for now), the jump label
6316 * destruction code runs outside of the cgroup lock. This is because
6317 * get_online_cpus(), which is called from the static_branch update,
6318 * can't be called inside the cgroup_lock. cpusets are the ones
6319 * enforcing this dependency, so if they ever change, we might as well.
6320 *
6321 * schedule_work() will guarantee this happens. Be careful if you need
6322 * to move this code around, and make sure it is outside
6323 * the cgroup_lock.
6324 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006325 disarm_static_keys(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006326 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006327}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006328
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006329/*
6330 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6331 */
Glauber Costae1aab162011-12-11 21:47:03 +00006332struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006333{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006334 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006335 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006336 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006337}
Glauber Costae1aab162011-12-11 21:47:03 +00006338EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006339
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006340static void __init mem_cgroup_soft_limit_tree_init(void)
6341{
6342 struct mem_cgroup_tree_per_node *rtpn;
6343 struct mem_cgroup_tree_per_zone *rtpz;
6344 int tmp, node, zone;
6345
6346 for_each_node(node) {
6347 tmp = node;
6348 if (!node_state(node, N_NORMAL_MEMORY))
6349 tmp = -1;
6350 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
6351 BUG_ON(!rtpn);
6352
6353 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6354
6355 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6356 rtpz = &rtpn->rb_tree_per_zone[zone];
6357 rtpz->rb_root = RB_ROOT;
6358 spin_lock_init(&rtpz->lock);
6359 }
6360 }
6361}
6362
Li Zefan0eb253e2009-01-15 13:51:25 -08006363static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006364mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006365{
Glauber Costad142e3e2013-02-22 16:34:52 -08006366 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006367 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006368 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006369
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006370 memcg = mem_cgroup_alloc();
6371 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006372 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006373
Bob Liu3ed28fa2012-01-12 17:19:04 -08006374 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006375 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006376 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006377
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006378 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006379 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006380 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006381 res_counter_init(&memcg->res, NULL);
6382 res_counter_init(&memcg->memsw, NULL);
6383 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006384 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006385
Glauber Costad142e3e2013-02-22 16:34:52 -08006386 memcg->last_scanned_node = MAX_NUMNODES;
6387 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006388 memcg->move_charge_at_immigrate = 0;
6389 mutex_init(&memcg->thresholds_lock);
6390 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006391 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05006392 INIT_LIST_HEAD(&memcg->event_list);
6393 spin_lock_init(&memcg->event_list_lock);
Glauber Costad142e3e2013-02-22 16:34:52 -08006394
6395 return &memcg->css;
6396
6397free_out:
6398 __mem_cgroup_free(memcg);
6399 return ERR_PTR(error);
6400}
6401
6402static int
Tejun Heoeb954192013-08-08 20:11:23 -04006403mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006404{
Tejun Heoeb954192013-08-08 20:11:23 -04006405 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6406 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006407
Li Zefan4219b2d2013-09-23 16:56:29 +08006408 if (css->cgroup->id > MEM_CGROUP_ID_MAX)
6409 return -ENOSPC;
6410
Tejun Heo63876982013-08-08 20:11:23 -04006411 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006412 return 0;
6413
Glauber Costa09998212013-02-22 16:34:55 -08006414 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006415
6416 memcg->use_hierarchy = parent->use_hierarchy;
6417 memcg->oom_kill_disable = parent->oom_kill_disable;
6418 memcg->swappiness = mem_cgroup_swappiness(parent);
6419
6420 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006421 res_counter_init(&memcg->res, &parent->res);
6422 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006423 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006424
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006425 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006426 * No need to take a reference to the parent because cgroup
6427 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006428 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006429 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006430 res_counter_init(&memcg->res, NULL);
6431 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006432 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006433 /*
6434 * Deeper hierachy with use_hierarchy == false doesn't make
6435 * much sense so let cgroup subsystem know about this
6436 * unfortunate state in our controller.
6437 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006438 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05006439 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006440 }
Glauber Costa09998212013-02-22 16:34:55 -08006441 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08006442
Tejun Heo073219e2014-02-08 10:36:58 -05006443 return memcg_init_kmem(memcg, &memory_cgrp_subsys);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006444}
6445
Michal Hocko5f578162013-04-29 15:07:17 -07006446/*
6447 * Announce all parents that a group from their hierarchy is gone.
6448 */
6449static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6450{
6451 struct mem_cgroup *parent = memcg;
6452
6453 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006454 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006455
6456 /*
6457 * if the root memcg is not hierarchical we have to check it
6458 * explicitely.
6459 */
6460 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006461 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006462}
6463
Tejun Heoeb954192013-08-08 20:11:23 -04006464static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006465{
Tejun Heoeb954192013-08-08 20:11:23 -04006466 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006467 struct mem_cgroup_event *event, *tmp;
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08006468 struct cgroup_subsys_state *iter;
Tejun Heo79bd9812013-11-22 18:20:42 -05006469
6470 /*
6471 * Unregister events and notify userspace.
6472 * Notify userspace about cgroup removing only after rmdir of cgroup
6473 * directory to avoid race between userspace and kernelspace.
6474 */
Tejun Heofba94802013-11-22 18:20:43 -05006475 spin_lock(&memcg->event_list_lock);
6476 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05006477 list_del_init(&event->list);
6478 schedule_work(&event->remove);
6479 }
Tejun Heofba94802013-11-22 18:20:43 -05006480 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006481
Li Zefan10d5ebf2013-07-08 16:00:33 -07006482 kmem_cgroup_css_offline(memcg);
6483
Michal Hocko5f578162013-04-29 15:07:17 -07006484 mem_cgroup_invalidate_reclaim_iterators(memcg);
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08006485
6486 /*
6487 * This requires that offlining is serialized. Right now that is
6488 * guaranteed because css_killed_work_fn() holds the cgroup_mutex.
6489 */
6490 css_for_each_descendant_post(iter, css)
6491 mem_cgroup_reparent_charges(mem_cgroup_from_css(iter));
6492
Glauber Costa1f458cb2012-12-18 14:22:50 -08006493 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006494 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006495}
6496
Tejun Heoeb954192013-08-08 20:11:23 -04006497static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006498{
Tejun Heoeb954192013-08-08 20:11:23 -04006499 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner96f1c582013-12-12 17:12:34 -08006500 /*
6501 * XXX: css_offline() would be where we should reparent all
6502 * memory to prepare the cgroup for destruction. However,
6503 * memcg does not do css_tryget() and res_counter charging
6504 * under the same RCU lock region, which means that charging
6505 * could race with offlining. Offlining only happens to
6506 * cgroups with no tasks in them but charges can show up
6507 * without any tasks from the swapin path when the target
6508 * memcg is looked up from the swapout record and not from the
6509 * current task as it usually is. A race like this can leak
6510 * charges and put pages with stale cgroup pointers into
6511 * circulation:
6512 *
6513 * #0 #1
6514 * lookup_swap_cgroup_id()
6515 * rcu_read_lock()
6516 * mem_cgroup_lookup()
6517 * css_tryget()
6518 * rcu_read_unlock()
6519 * disable css_tryget()
6520 * call_rcu()
6521 * offline_css()
6522 * reparent_charges()
6523 * res_counter_charge()
6524 * css_put()
6525 * css_free()
6526 * pc->mem_cgroup = dead memcg
6527 * add page to lru
6528 *
6529 * The bulk of the charges are still moved in offline_css() to
6530 * avoid pinning a lot of pages in case a long-term reference
6531 * like a swapout record is deferring the css_free() to long
6532 * after offlining. But this makes sure we catch any charges
6533 * made after offlining:
6534 */
6535 mem_cgroup_reparent_charges(memcg);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006536
Li Zefan10d5ebf2013-07-08 16:00:33 -07006537 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006538 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006539}
6540
Daisuke Nishimura02491442010-03-10 15:22:17 -08006541#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006542/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006543#define PRECHARGE_COUNT_AT_ONCE 256
6544static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006545{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006546 int ret = 0;
6547 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006548 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006550 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006551 mc.precharge += count;
6552 /* we don't need css_get for root */
6553 return ret;
6554 }
6555 /* try to charge at once */
6556 if (count > 1) {
6557 struct res_counter *dummy;
6558 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006559 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006560 * by cgroup_lock_live_cgroup() that it is not removed and we
6561 * are still under the same cgroup_mutex. So we can postpone
6562 * css_get().
6563 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006564 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006565 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006566 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006567 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006568 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006569 goto one_by_one;
6570 }
6571 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006572 return ret;
6573 }
6574one_by_one:
6575 /* fall back to one by one charge */
6576 while (count--) {
6577 if (signal_pending(current)) {
6578 ret = -EINTR;
6579 break;
6580 }
6581 if (!batch_count--) {
6582 batch_count = PRECHARGE_COUNT_AT_ONCE;
6583 cond_resched();
6584 }
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07006585 ret = mem_cgroup_try_charge(memcg, GFP_KERNEL, 1, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006586 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006587 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006588 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006589 mc.precharge++;
6590 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006591 return ret;
6592}
6593
6594/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006595 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006596 * @vma: the vma the pte to be checked belongs
6597 * @addr: the address corresponding to the pte to be checked
6598 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006599 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006600 *
6601 * Returns
6602 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6603 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6604 * move charge. if @target is not NULL, the page is stored in target->page
6605 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006606 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6607 * target for charge migration. if @target is not NULL, the entry is stored
6608 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006609 *
6610 * Called with pte lock held.
6611 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006612union mc_target {
6613 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006614 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006615};
6616
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006617enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006618 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006619 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006620 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006621};
6622
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006623static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6624 unsigned long addr, pte_t ptent)
6625{
6626 struct page *page = vm_normal_page(vma, addr, ptent);
6627
6628 if (!page || !page_mapped(page))
6629 return NULL;
6630 if (PageAnon(page)) {
6631 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006632 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006633 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006634 } else if (!move_file())
6635 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006636 return NULL;
6637 if (!get_page_unless_zero(page))
6638 return NULL;
6639
6640 return page;
6641}
6642
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006643#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006644static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6645 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6646{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006647 struct page *page = NULL;
6648 swp_entry_t ent = pte_to_swp_entry(ptent);
6649
6650 if (!move_anon() || non_swap_entry(ent))
6651 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006652 /*
6653 * Because lookup_swap_cache() updates some statistics counter,
6654 * we call find_get_page() with swapper_space directly.
6655 */
Shaohua Li33806f02013-02-22 16:34:37 -08006656 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006657 if (do_swap_account)
6658 entry->val = ent.val;
6659
6660 return page;
6661}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006662#else
6663static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6664 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6665{
6666 return NULL;
6667}
6668#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006669
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006670static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6671 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6672{
6673 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006674 struct address_space *mapping;
6675 pgoff_t pgoff;
6676
6677 if (!vma->vm_file) /* anonymous vma */
6678 return NULL;
6679 if (!move_file())
6680 return NULL;
6681
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006682 mapping = vma->vm_file->f_mapping;
6683 if (pte_none(ptent))
6684 pgoff = linear_page_index(vma, addr);
6685 else /* pte_file(ptent) is true */
6686 pgoff = pte_to_pgoff(ptent);
6687
6688 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006689 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006690
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006691#ifdef CONFIG_SWAP
6692 /* shmem/tmpfs may report page out on swap: account for that too. */
6693 if (radix_tree_exceptional_entry(page)) {
6694 swp_entry_t swap = radix_to_swp_entry(page);
6695 if (do_swap_account)
6696 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006697 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006698 }
6699#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006700 return page;
6701}
6702
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006703static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006704 unsigned long addr, pte_t ptent, union mc_target *target)
6705{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006706 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006707 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006708 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006709 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006710
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006711 if (pte_present(ptent))
6712 page = mc_handle_present_pte(vma, addr, ptent);
6713 else if (is_swap_pte(ptent))
6714 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006715 else if (pte_none(ptent) || pte_file(ptent))
6716 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006717
6718 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006719 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006720 if (page) {
6721 pc = lookup_page_cgroup(page);
6722 /*
6723 * Do only loose check w/o page_cgroup lock.
6724 * mem_cgroup_move_account() checks the pc is valid or not under
6725 * the lock.
6726 */
6727 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6728 ret = MC_TARGET_PAGE;
6729 if (target)
6730 target->page = page;
6731 }
6732 if (!ret || !target)
6733 put_page(page);
6734 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006735 /* There is a swap entry and a page doesn't exist or isn't charged */
6736 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08006737 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006738 ret = MC_TARGET_SWAP;
6739 if (target)
6740 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006741 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006742 return ret;
6743}
6744
Naoya Horiguchi12724852012-03-21 16:34:28 -07006745#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6746/*
6747 * We don't consider swapping or file mapped pages because THP does not
6748 * support them for now.
6749 * Caller should make sure that pmd_trans_huge(pmd) is true.
6750 */
6751static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6752 unsigned long addr, pmd_t pmd, union mc_target *target)
6753{
6754 struct page *page = NULL;
6755 struct page_cgroup *pc;
6756 enum mc_target_type ret = MC_TARGET_NONE;
6757
6758 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08006759 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006760 if (!move_anon())
6761 return ret;
6762 pc = lookup_page_cgroup(page);
6763 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6764 ret = MC_TARGET_PAGE;
6765 if (target) {
6766 get_page(page);
6767 target->page = page;
6768 }
6769 }
6770 return ret;
6771}
6772#else
6773static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6774 unsigned long addr, pmd_t pmd, union mc_target *target)
6775{
6776 return MC_TARGET_NONE;
6777}
6778#endif
6779
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006780static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6781 unsigned long addr, unsigned long end,
6782 struct mm_walk *walk)
6783{
6784 struct vm_area_struct *vma = walk->private;
6785 pte_t *pte;
6786 spinlock_t *ptl;
6787
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006788 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006789 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6790 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006791 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006792 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006793 }
Dave Hansen03319322011-03-22 16:32:56 -07006794
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006795 if (pmd_trans_unstable(pmd))
6796 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006797 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6798 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006799 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006800 mc.precharge++; /* increment precharge temporarily */
6801 pte_unmap_unlock(pte - 1, ptl);
6802 cond_resched();
6803
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006804 return 0;
6805}
6806
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006807static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6808{
6809 unsigned long precharge;
6810 struct vm_area_struct *vma;
6811
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006812 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006813 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6814 struct mm_walk mem_cgroup_count_precharge_walk = {
6815 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6816 .mm = mm,
6817 .private = vma,
6818 };
6819 if (is_vm_hugetlb_page(vma))
6820 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006821 walk_page_range(vma->vm_start, vma->vm_end,
6822 &mem_cgroup_count_precharge_walk);
6823 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006824 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006825
6826 precharge = mc.precharge;
6827 mc.precharge = 0;
6828
6829 return precharge;
6830}
6831
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006832static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6833{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006834 unsigned long precharge = mem_cgroup_count_precharge(mm);
6835
6836 VM_BUG_ON(mc.moving_task);
6837 mc.moving_task = current;
6838 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006839}
6840
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006841/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6842static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006843{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006844 struct mem_cgroup *from = mc.from;
6845 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006846 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006847
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006848 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006849 if (mc.precharge) {
6850 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6851 mc.precharge = 0;
6852 }
6853 /*
6854 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6855 * we must uncharge here.
6856 */
6857 if (mc.moved_charge) {
6858 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6859 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006860 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006861 /* we must fixup refcnts and charges */
6862 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006863 /* uncharge swap account from the old cgroup */
6864 if (!mem_cgroup_is_root(mc.from))
6865 res_counter_uncharge(&mc.from->memsw,
6866 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006867
6868 for (i = 0; i < mc.moved_swap; i++)
6869 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006870
6871 if (!mem_cgroup_is_root(mc.to)) {
6872 /*
6873 * we charged both to->res and to->memsw, so we should
6874 * uncharge to->res.
6875 */
6876 res_counter_uncharge(&mc.to->res,
6877 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006878 }
Li Zefan40503772013-07-08 16:00:34 -07006879 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006880 mc.moved_swap = 0;
6881 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006882 memcg_oom_recover(from);
6883 memcg_oom_recover(to);
6884 wake_up_all(&mc.waitq);
6885}
6886
6887static void mem_cgroup_clear_mc(void)
6888{
6889 struct mem_cgroup *from = mc.from;
6890
6891 /*
6892 * we must clear moving_task before waking up waiters at the end of
6893 * task migration.
6894 */
6895 mc.moving_task = NULL;
6896 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006897 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006898 mc.from = NULL;
6899 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006900 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006901 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006902}
6903
Tejun Heoeb954192013-08-08 20:11:23 -04006904static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006905 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006906{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006907 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006908 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04006909 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08006910 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006911
Glauber Costaee5e8472013-02-22 16:34:50 -08006912 /*
6913 * We are now commited to this value whatever it is. Changes in this
6914 * tunable will only affect upcoming migrations, not the current one.
6915 * So we need to save it, and keep it going.
6916 */
6917 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6918 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006919 struct mm_struct *mm;
6920 struct mem_cgroup *from = mem_cgroup_from_task(p);
6921
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006922 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006923
6924 mm = get_task_mm(p);
6925 if (!mm)
6926 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006927 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006928 if (mm->owner == p) {
6929 VM_BUG_ON(mc.from);
6930 VM_BUG_ON(mc.to);
6931 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006932 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006933 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006934 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006935 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006936 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006937 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006938 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006939 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006940 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006941
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006942 ret = mem_cgroup_precharge_mc(mm);
6943 if (ret)
6944 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006945 }
6946 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006947 }
6948 return ret;
6949}
6950
Tejun Heoeb954192013-08-08 20:11:23 -04006951static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006952 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006953{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006954 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006955}
6956
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006957static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6958 unsigned long addr, unsigned long end,
6959 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006960{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006961 int ret = 0;
6962 struct vm_area_struct *vma = walk->private;
6963 pte_t *pte;
6964 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006965 enum mc_target_type target_type;
6966 union mc_target target;
6967 struct page *page;
6968 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006969
Naoya Horiguchi12724852012-03-21 16:34:28 -07006970 /*
6971 * We don't take compound_lock() here but no race with splitting thp
6972 * happens because:
6973 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6974 * under splitting, which means there's no concurrent thp split,
6975 * - if another thread runs into split_huge_page() just after we
6976 * entered this if-block, the thread must wait for page table lock
6977 * to be unlocked in __split_huge_page_splitting(), where the main
6978 * part of thp split is not executed yet.
6979 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006980 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006981 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006982 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006983 return 0;
6984 }
6985 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6986 if (target_type == MC_TARGET_PAGE) {
6987 page = target.page;
6988 if (!isolate_lru_page(page)) {
6989 pc = lookup_page_cgroup(page);
6990 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006991 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006992 mc.precharge -= HPAGE_PMD_NR;
6993 mc.moved_charge += HPAGE_PMD_NR;
6994 }
6995 putback_lru_page(page);
6996 }
6997 put_page(page);
6998 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006999 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07007000 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07007001 }
7002
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07007003 if (pmd_trans_unstable(pmd))
7004 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007005retry:
7006 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
7007 for (; addr != end; addr += PAGE_SIZE) {
7008 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08007009 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007010
7011 if (!mc.precharge)
7012 break;
7013
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007014 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007015 case MC_TARGET_PAGE:
7016 page = target.page;
7017 if (isolate_lru_page(page))
7018 goto put;
7019 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07007020 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07007021 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007022 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007023 /* we uncharge from mc.from later. */
7024 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007025 }
7026 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007027put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007028 put_page(page);
7029 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08007030 case MC_TARGET_SWAP:
7031 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07007032 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08007033 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007034 /* we fixup refcnts and charges later. */
7035 mc.moved_swap++;
7036 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08007037 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007038 default:
7039 break;
7040 }
7041 }
7042 pte_unmap_unlock(pte - 1, ptl);
7043 cond_resched();
7044
7045 if (addr != end) {
7046 /*
7047 * We have consumed all precharges we got in can_attach().
7048 * We try charge one by one, but don't do any additional
7049 * charges to mc.to if we have failed in charge once in attach()
7050 * phase.
7051 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007052 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007053 if (!ret)
7054 goto retry;
7055 }
7056
7057 return ret;
7058}
7059
7060static void mem_cgroup_move_charge(struct mm_struct *mm)
7061{
7062 struct vm_area_struct *vma;
7063
7064 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007065retry:
7066 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
7067 /*
7068 * Someone who are holding the mmap_sem might be waiting in
7069 * waitq. So we cancel all extra charges, wake up all waiters,
7070 * and retry. Because we cancel precharges, we might not be able
7071 * to move enough charges, but moving charge is a best-effort
7072 * feature anyway, so it wouldn't be a big problem.
7073 */
7074 __mem_cgroup_clear_mc();
7075 cond_resched();
7076 goto retry;
7077 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007078 for (vma = mm->mmap; vma; vma = vma->vm_next) {
7079 int ret;
7080 struct mm_walk mem_cgroup_move_charge_walk = {
7081 .pmd_entry = mem_cgroup_move_charge_pte_range,
7082 .mm = mm,
7083 .private = vma,
7084 };
7085 if (is_vm_hugetlb_page(vma))
7086 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007087 ret = walk_page_range(vma->vm_start, vma->vm_end,
7088 &mem_cgroup_move_charge_walk);
7089 if (ret)
7090 /*
7091 * means we have consumed all precharges and failed in
7092 * doing additional charge. Just abandon here.
7093 */
7094 break;
7095 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007096 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007097}
7098
Tejun Heoeb954192013-08-08 20:11:23 -04007099static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007100 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08007101{
Tejun Heo2f7ee562011-12-12 18:12:21 -08007102 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007103 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007104
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007105 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007106 if (mc.to)
7107 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007108 mmput(mm);
7109 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007110 if (mc.to)
7111 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08007112}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007113#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04007114static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007115 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007116{
7117 return 0;
7118}
Tejun Heoeb954192013-08-08 20:11:23 -04007119static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007120 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007121{
7122}
Tejun Heoeb954192013-08-08 20:11:23 -04007123static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007124 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007125{
7126}
7127#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08007128
Tejun Heof00baae2013-04-15 13:41:15 -07007129/*
7130 * Cgroup retains root cgroups across [un]mount cycles making it necessary
7131 * to verify sane_behavior flag on each mount attempt.
7132 */
Tejun Heoeb954192013-08-08 20:11:23 -04007133static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07007134{
7135 /*
7136 * use_hierarchy is forced with sane_behavior. cgroup core
7137 * guarantees that @root doesn't have any children, so turning it
7138 * on for the root memcg is enough.
7139 */
Tejun Heoeb954192013-08-08 20:11:23 -04007140 if (cgroup_sane_behavior(root_css->cgroup))
7141 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07007142}
7143
Tejun Heo073219e2014-02-08 10:36:58 -05007144struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08007145 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08007146 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08007147 .css_offline = mem_cgroup_css_offline,
7148 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007149 .can_attach = mem_cgroup_can_attach,
7150 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08007151 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07007152 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07007153 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08007154 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007155};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007156
Andrew Mortonc255a452012-07-31 16:43:02 -07007157#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08007158static int __init enable_swap_account(char *s)
7159{
Michal Hockoa2c89902011-05-24 17:12:50 -07007160 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007161 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07007162 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007163 really_do_swap_account = 0;
7164 return 1;
7165}
Michal Hockoa2c89902011-05-24 17:12:50 -07007166__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007167
Michal Hocko2d110852013-02-22 16:34:43 -08007168static void __init memsw_file_init(void)
7169{
Tejun Heo073219e2014-02-08 10:36:58 -05007170 WARN_ON(cgroup_add_cftypes(&memory_cgrp_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08007171}
Michal Hocko6acc8b02013-02-22 16:34:45 -08007172
7173static void __init enable_swap_cgroup(void)
7174{
7175 if (!mem_cgroup_disabled() && really_do_swap_account) {
7176 do_swap_account = 1;
7177 memsw_file_init();
7178 }
7179}
7180
Michal Hocko2d110852013-02-22 16:34:43 -08007181#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08007182static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08007183{
7184}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007185#endif
Michal Hocko2d110852013-02-22 16:34:43 -08007186
7187/*
Michal Hocko10813122013-02-22 16:35:41 -08007188 * subsys_initcall() for memory controller.
7189 *
7190 * Some parts like hotcpu_notifier() have to be initialized from this context
7191 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7192 * everything that doesn't depend on a specific mem_cgroup structure should
7193 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007194 */
7195static int __init mem_cgroup_init(void)
7196{
7197 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007198 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07007199 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007200 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007201 return 0;
7202}
7203subsys_initcall(mem_cgroup_init);