blob: 7a67afa3c3dc65101fa46e38db2ef2c8c1cc37fd [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 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080065#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000066#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070067#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000068#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070078#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070079static struct mem_cgroup *root_mem_cgroup __read_mostly;
Tejun Heodbee2272015-05-22 17:13:20 -040080struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weiner21afa382015-02-11 15:26:36 -080082/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070083#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080084int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070086#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080087#endif
88
Johannes Weineraf7c4b02012-05-29 15:07:08 -070089static const char * const mem_cgroup_stat_names[] = {
90 "cache",
91 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070092 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070093 "mapped_file",
Greg Thelen7c9d3ff2015-05-22 17:13:16 -040094 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070095 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070096 "swap",
97};
98
Johannes Weineraf7c4b02012-05-29 15:07:08 -070099static const char * const mem_cgroup_events_names[] = {
100 "pgpgin",
101 "pgpgout",
102 "pgfault",
103 "pgmajfault",
104};
105
Sha Zhengju58cf1882013-02-22 16:32:05 -0800106static const char * const mem_cgroup_lru_names[] = {
107 "inactive_anon",
108 "active_anon",
109 "inactive_file",
110 "active_file",
111 "unevictable",
112};
113
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700114/*
115 * Per memcg event counter is incremented at every pagein/pageout. With THP,
116 * it will be incremated by the number of pages. This counter is used for
117 * for trigger some periodic events. This is straightforward and better
118 * than using jiffies etc. to handle periodic memcg event.
119 */
120enum mem_cgroup_events_target {
121 MEM_CGROUP_TARGET_THRESH,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700122 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700123 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700124 MEM_CGROUP_NTARGETS,
125};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800130struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700131 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weiner241994ed2015-02-11 15:26:06 -0800132 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner13114712012-05-29 15:07:07 -0700133 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700134 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800135};
136
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800137struct reclaim_iter {
138 struct mem_cgroup *position;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800139 /* scan generation, increased every round-trip */
140 unsigned int generation;
141};
142
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800143/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800144 * per-zone information in memory controller.
145 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800146struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800147 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700148 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800149
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800150 struct reclaim_iter iter[DEF_PRIORITY + 1];
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800151
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700152 struct rb_node tree_node; /* RB tree node */
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800153 unsigned long usage_in_excess;/* Set to the value by which */
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700154 /* the soft limit is exceeded*/
155 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700156 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700157 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800158};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800159
160struct mem_cgroup_per_node {
161 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
162};
163
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700164/*
165 * Cgroups above their limits are maintained in a RB-Tree, independent of
166 * their hierarchy representation
167 */
168
169struct mem_cgroup_tree_per_zone {
170 struct rb_root rb_root;
171 spinlock_t lock;
172};
173
174struct mem_cgroup_tree_per_node {
175 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
176};
177
178struct mem_cgroup_tree {
179 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
180};
181
182static struct mem_cgroup_tree soft_limit_tree __read_mostly;
183
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800184struct mem_cgroup_threshold {
185 struct eventfd_ctx *eventfd;
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800186 unsigned long threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800187};
188
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700189/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800190struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700191 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700192 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800193 /* Size of entries[] */
194 unsigned int size;
195 /* Array of thresholds */
196 struct mem_cgroup_threshold entries[0];
197};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700198
199struct mem_cgroup_thresholds {
200 /* Primary thresholds array */
201 struct mem_cgroup_threshold_ary *primary;
202 /*
203 * Spare threshold array.
204 * This is needed to make mem_cgroup_unregister_event() "never fail".
205 * It must be able to store at least primary->size - 1 entries.
206 */
207 struct mem_cgroup_threshold_ary *spare;
208};
209
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700210/* for OOM */
211struct mem_cgroup_eventfd_list {
212 struct list_head list;
213 struct eventfd_ctx *eventfd;
214};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800215
Tejun Heo79bd9812013-11-22 18:20:42 -0500216/*
217 * cgroup_event represents events which userspace want to receive.
218 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500219struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500220 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500221 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500222 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500223 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500224 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500225 * eventfd to signal userspace about the event.
226 */
227 struct eventfd_ctx *eventfd;
228 /*
229 * Each of these stored in a list by the cgroup.
230 */
231 struct list_head list;
232 /*
Tejun Heofba94802013-11-22 18:20:43 -0500233 * register_event() callback will be used to add new userspace
234 * waiter for changes related to this event. Use eventfd_signal()
235 * on eventfd to send notification to userspace.
236 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500237 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500238 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500239 /*
240 * unregister_event() callback will be called when userspace closes
241 * the eventfd or on cgroup removing. This callback must be set,
242 * if you want provide notification functionality.
243 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500244 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500245 struct eventfd_ctx *eventfd);
246 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500247 * All fields below needed to unregister event when
248 * userspace closes eventfd.
249 */
250 poll_table pt;
251 wait_queue_head_t *wqh;
252 wait_queue_t wait;
253 struct work_struct remove;
254};
255
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700256static void mem_cgroup_threshold(struct mem_cgroup *memcg);
257static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800258
Balbir Singhf64c3f52009-09-23 15:56:37 -0700259/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800260 * The memory controller data structure. The memory controller controls both
261 * page cache and RSS per cgroup. We would eventually like to provide
262 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
263 * to help the administrator determine what knobs to tune.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800264 */
265struct mem_cgroup {
266 struct cgroup_subsys_state css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800267
268 /* Accounted resources */
269 struct page_counter memory;
270 struct page_counter memsw;
271 struct page_counter kmem;
272
Johannes Weiner241994ed2015-02-11 15:26:06 -0800273 /* Normal memory consumption range */
274 unsigned long low;
275 unsigned long high;
276
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800277 unsigned long soft_limit;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700278
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700279 /* vmpressure notifications */
280 struct vmpressure vmpressure;
281
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -0700282 /* css_online() has been completed */
283 int initialized;
284
Li Zefan465939a2013-07-08 16:00:38 -0700285 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800286 * Should the accounting and control be hierarchical, per subtree?
287 */
288 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700289
290 bool oom_lock;
291 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700292 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700293
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700294 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700295 /* OOM-Killer disable */
296 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800297
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800298 /* protect arrays of thresholds */
299 struct mutex thresholds_lock;
300
301 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700302 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700303
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800304 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700305 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700306
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700307 /* For oom notifier event fd */
308 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700309
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800310 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800311 * Should we move charges of a task when a task is moved into this
312 * mem_cgroup ? And what type of charges should we move ?
313 */
Andrew Mortonf894ffa2013-09-12 15:13:35 -0700314 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800315 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700316 * set > 0 if pages under this cgroup are moving to other cgroup.
317 */
Johannes Weiner6de22612015-02-11 15:25:01 -0800318 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700319 /* taken only while moving_account > 0 */
Johannes Weiner6de22612015-02-11 15:25:01 -0800320 spinlock_t move_lock;
321 struct task_struct *move_lock_task;
322 unsigned long move_lock_flags;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700323 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800324 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800325 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700326 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700327 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000328
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700329#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700330 struct cg_proto tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000331#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800332#if defined(CONFIG_MEMCG_KMEM)
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800333 /* Index in the kmem_cache->memcg_params.memcg_caches array */
Glauber Costa2633d7a2012-12-18 14:22:34 -0800334 int kmemcg_id;
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -0800335 bool kmem_acct_activated;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800336 bool kmem_acct_active;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800337#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800338
339 int last_scanned_node;
340#if MAX_NUMNODES > 1
341 nodemask_t scan_nodes;
342 atomic_t numainfo_events;
343 atomic_t numainfo_updating;
344#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700345
Tejun Heo6b385782015-05-22 17:13:37 -0400346#ifdef CONFIG_CGROUP_WRITEBACK
347 struct list_head cgwb_list;
348#endif
349
Tejun Heofba94802013-11-22 18:20:43 -0500350 /* List of events which userspace want to receive */
351 struct list_head event_list;
352 spinlock_t event_list_lock;
353
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700354 struct mem_cgroup_per_node *nodeinfo[0];
355 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800356};
357
Glauber Costa510fc4e2012-12-18 14:21:47 -0800358#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovcb731d62015-02-12 14:58:54 -0800359bool memcg_kmem_is_active(struct mem_cgroup *memcg)
Glauber Costa7de37682012-12-18 14:22:07 -0800360{
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800361 return memcg->kmem_acct_active;
Glauber Costa7de37682012-12-18 14:22:07 -0800362}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800363#endif
364
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800365/* Stuffs for move charges at task migration. */
366/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800367 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800368 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800369#define MOVE_ANON 0x1U
370#define MOVE_FILE 0x2U
371#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800372
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800373/* "mc" and its members are protected by cgroup_mutex */
374static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800375 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800376 struct mem_cgroup *from;
377 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800378 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800379 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800380 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800381 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800382 struct task_struct *moving_task; /* a task moving charges */
383 wait_queue_head_t waitq; /* a waitq for other context */
384} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700385 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800386 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
387};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800388
Balbir Singh4e416952009-09-23 15:56:39 -0700389/*
390 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
391 * limit reclaim to prevent infinite loops, if they ever occur.
392 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700393#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700394#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700395
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800396enum charge_type {
397 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700398 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800399 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700400 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700401 NR_CHARGE_TYPE,
402};
403
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800404/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800405enum res_type {
406 _MEM,
407 _MEMSWAP,
408 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800409 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800410};
411
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700412#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
413#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800414#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700415/* Used for OOM nofiier */
416#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800417
Balbir Singh75822b42009-09-23 15:56:38 -0700418/*
Glauber Costa09998212013-02-22 16:34:55 -0800419 * The memcg_create_mutex will be held whenever a new cgroup is created.
420 * As a consequence, any change that needs to protect against new child cgroups
421 * appearing has to hold it as well.
422 */
423static DEFINE_MUTEX(memcg_create_mutex);
424
Wanpeng Lib2145142012-07-31 16:46:01 -0700425struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
426{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400427 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700428}
429
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700430/* Some nice accessors for the vmpressure. */
431struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
432{
433 if (!memcg)
434 memcg = root_mem_cgroup;
435 return &memcg->vmpressure;
436}
437
438struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
439{
440 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
441}
442
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700443static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
444{
445 return (memcg == root_mem_cgroup);
446}
447
Li Zefan4219b2d2013-09-23 16:56:29 +0800448/*
449 * We restrict the id in the range of [1, 65535], so it can fit into
450 * an unsigned short.
451 */
452#define MEM_CGROUP_ID_MAX USHRT_MAX
453
Li Zefan34c00c32013-09-23 16:56:01 +0800454static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
455{
Tejun Heo15a4c832014-05-04 15:09:14 -0400456 return memcg->css.id;
Li Zefan34c00c32013-09-23 16:56:01 +0800457}
458
Vladimir Davydovadbe4272015-04-15 16:13:00 -0700459/*
460 * A helper function to get mem_cgroup from ID. must be called under
461 * rcu_read_lock(). The caller is responsible for calling
462 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
463 * refcnt from swap can be called against removed memcg.)
464 */
Li Zefan34c00c32013-09-23 16:56:01 +0800465static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
466{
467 struct cgroup_subsys_state *css;
468
Tejun Heo7d699dd2014-05-04 15:09:13 -0400469 css = css_from_id(id, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800470 return mem_cgroup_from_css(css);
471}
472
Glauber Costae1aab162011-12-11 21:47:03 +0000473/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700474#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000475
Glauber Costae1aab162011-12-11 21:47:03 +0000476void sock_update_memcg(struct sock *sk)
477{
Glauber Costa376be5f2012-01-20 04:57:14 +0000478 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000479 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700480 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000481
482 BUG_ON(!sk->sk_prot->proto_cgroup);
483
Glauber Costaf3f511e2012-01-05 20:16:39 +0000484 /* Socket cloning can throw us here with sk_cgrp already
485 * filled. It won't however, necessarily happen from
486 * process context. So the test for root memcg given
487 * the current task's memcg won't help us in this case.
488 *
489 * Respecting the original socket's memcg is a better
490 * decision in this case.
491 */
492 if (sk->sk_cgrp) {
493 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700494 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000495 return;
496 }
497
Glauber Costae1aab162011-12-11 21:47:03 +0000498 rcu_read_lock();
499 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700500 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700501 if (!mem_cgroup_is_root(memcg) &&
Tejun Heoec903c02014-05-13 12:11:01 -0400502 memcg_proto_active(cg_proto) &&
503 css_tryget_online(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700504 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000505 }
506 rcu_read_unlock();
507 }
508}
509EXPORT_SYMBOL(sock_update_memcg);
510
511void sock_release_memcg(struct sock *sk)
512{
Glauber Costa376be5f2012-01-20 04:57:14 +0000513 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000514 struct mem_cgroup *memcg;
515 WARN_ON(!sk->sk_cgrp->memcg);
516 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700517 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000518 }
519}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000520
521struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
522{
523 if (!memcg || mem_cgroup_is_root(memcg))
524 return NULL;
525
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700526 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000527}
528EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000529
Glauber Costa3f134612012-05-29 15:07:11 -0700530#endif
531
Glauber Costaa8964b92012-12-18 14:22:09 -0800532#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800533/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800534 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800535 * The main reason for not using cgroup id for this:
536 * this works better in sparse environments, where we have a lot of memcgs,
537 * but only a few kmem-limited. Or also, if we have, for instance, 200
538 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
539 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800540 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800541 * The current size of the caches array is stored in memcg_nr_cache_ids. It
542 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800543 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800544static DEFINE_IDA(memcg_cache_ida);
545int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800546
Vladimir Davydov05257a12015-02-12 14:59:01 -0800547/* Protects memcg_nr_cache_ids */
548static DECLARE_RWSEM(memcg_cache_ids_sem);
549
550void memcg_get_cache_ids(void)
551{
552 down_read(&memcg_cache_ids_sem);
553}
554
555void memcg_put_cache_ids(void)
556{
557 up_read(&memcg_cache_ids_sem);
558}
559
Glauber Costa55007d82012-12-18 14:22:38 -0800560/*
561 * MIN_SIZE is different than 1, because we would like to avoid going through
562 * the alloc/free process all the time. In a small machine, 4 kmem-limited
563 * cgroups is a reasonable guess. In the future, it could be a parameter or
564 * tunable, but that is strictly not necessary.
565 *
Li Zefanb8627832013-09-23 16:56:47 +0800566 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800567 * this constant directly from cgroup, but it is understandable that this is
568 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800569 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800570 * increase ours as well if it increases.
571 */
572#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800573#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800574
Glauber Costad7f25f82012-12-18 14:22:40 -0800575/*
576 * A lot of the calls to the cache allocation functions are expected to be
577 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
578 * conditional to this static branch, we'll have to allow modules that does
579 * kmem_cache_alloc and the such to see this symbol as well
580 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800581struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800582EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800583
Glauber Costaa8964b92012-12-18 14:22:09 -0800584#endif /* CONFIG_MEMCG_KMEM */
585
Balbir Singhf64c3f52009-09-23 15:56:37 -0700586static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700587mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700588{
Jianyu Zhane2318752014-06-06 14:38:20 -0700589 int nid = zone_to_nid(zone);
590 int zid = zone_idx(zone);
591
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700592 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700593}
594
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700595struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100596{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700597 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100598}
599
Tejun Heoe4487942015-05-27 20:00:02 -0400600/**
601 * mem_cgroup_css_from_page - css of the memcg associated with a page
602 * @page: page of interest
603 *
604 * If memcg is bound to the default hierarchy, css of the memcg associated
605 * with @page is returned. The returned css remains associated with @page
606 * until it is released.
607 *
608 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
609 * is returned.
610 *
611 * XXX: The above description of behavior on the default hierarchy isn't
612 * strictly true yet as replace_page_cache_page() can modify the
613 * association before @page is released even on the default hierarchy;
614 * however, the current and planned usages don't mix the the two functions
615 * and replace_page_cache_page() will soon be updated to make the invariant
616 * actually true.
617 */
618struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
619{
620 struct mem_cgroup *memcg;
621
622 rcu_read_lock();
623
624 memcg = page->mem_cgroup;
625
626 if (!memcg || !cgroup_on_dfl(memcg->css.cgroup))
627 memcg = root_mem_cgroup;
628
629 rcu_read_unlock();
630 return &memcg->css;
631}
632
Balbir Singhf64c3f52009-09-23 15:56:37 -0700633static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700634mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700635{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700636 int nid = page_to_nid(page);
637 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700638
Jianyu Zhane2318752014-06-06 14:38:20 -0700639 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700640}
641
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700642static struct mem_cgroup_tree_per_zone *
643soft_limit_tree_node_zone(int nid, int zid)
644{
645 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
646}
647
648static struct mem_cgroup_tree_per_zone *
649soft_limit_tree_from_page(struct page *page)
650{
651 int nid = page_to_nid(page);
652 int zid = page_zonenum(page);
653
654 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
655}
656
Johannes Weinercf2c8122014-06-06 14:38:21 -0700657static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
658 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800659 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700660{
661 struct rb_node **p = &mctz->rb_root.rb_node;
662 struct rb_node *parent = NULL;
663 struct mem_cgroup_per_zone *mz_node;
664
665 if (mz->on_tree)
666 return;
667
668 mz->usage_in_excess = new_usage_in_excess;
669 if (!mz->usage_in_excess)
670 return;
671 while (*p) {
672 parent = *p;
673 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
674 tree_node);
675 if (mz->usage_in_excess < mz_node->usage_in_excess)
676 p = &(*p)->rb_left;
677 /*
678 * We can't avoid mem cgroups that are over their soft
679 * limit by the same amount
680 */
681 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
682 p = &(*p)->rb_right;
683 }
684 rb_link_node(&mz->tree_node, parent, p);
685 rb_insert_color(&mz->tree_node, &mctz->rb_root);
686 mz->on_tree = true;
687}
688
Johannes Weinercf2c8122014-06-06 14:38:21 -0700689static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
690 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700691{
692 if (!mz->on_tree)
693 return;
694 rb_erase(&mz->tree_node, &mctz->rb_root);
695 mz->on_tree = false;
696}
697
Johannes Weinercf2c8122014-06-06 14:38:21 -0700698static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
699 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700701 unsigned long flags;
702
703 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700704 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700705 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700706}
707
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800708static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
709{
710 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700711 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800712 unsigned long excess = 0;
713
714 if (nr_pages > soft_limit)
715 excess = nr_pages - soft_limit;
716
717 return excess;
718}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700719
720static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
721{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800722 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700723 struct mem_cgroup_per_zone *mz;
724 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700725
Jianyu Zhane2318752014-06-06 14:38:20 -0700726 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700727 /*
728 * Necessary to update all ancestors when hierarchy is used.
729 * because their event counter is not touched.
730 */
731 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700732 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800733 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700734 /*
735 * We have to update the tree if mz is on RB-tree or
736 * mem is over its softlimit.
737 */
738 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700739 unsigned long flags;
740
741 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700742 /* if on-tree, remove it */
743 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700744 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700745 /*
746 * Insert again. mz->usage_in_excess will be updated.
747 * If excess is 0, no tree ops.
748 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700749 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700750 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700751 }
752 }
753}
754
755static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
756{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700757 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700758 struct mem_cgroup_per_zone *mz;
759 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700760
Jianyu Zhane2318752014-06-06 14:38:20 -0700761 for_each_node(nid) {
762 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
763 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
764 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700765 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700766 }
767 }
768}
769
770static struct mem_cgroup_per_zone *
771__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
772{
773 struct rb_node *rightmost = NULL;
774 struct mem_cgroup_per_zone *mz;
775
776retry:
777 mz = NULL;
778 rightmost = rb_last(&mctz->rb_root);
779 if (!rightmost)
780 goto done; /* Nothing to reclaim from */
781
782 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
783 /*
784 * Remove the node now but someone else can add it back,
785 * we will to add it back at the end of reclaim to its correct
786 * position in the tree.
787 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700788 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800789 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400790 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700791 goto retry;
792done:
793 return mz;
794}
795
796static struct mem_cgroup_per_zone *
797mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
798{
799 struct mem_cgroup_per_zone *mz;
800
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700801 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700802 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700803 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700804 return mz;
805}
806
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700807/*
808 * Implementation Note: reading percpu statistics for memcg.
809 *
810 * Both of vmstat[] and percpu_counter has threshold and do periodic
811 * synchronization to implement "quick" read. There are trade-off between
812 * reading cost and precision of value. Then, we may have a chance to implement
813 * a periodic synchronizion of counter in memcg's counter.
814 *
815 * But this _read() function is used for user interface now. The user accounts
816 * memory usage by memory cgroup and he _always_ requires exact value because
817 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
818 * have to visit all online cpus and make sum. So, for now, unnecessary
819 * synchronization is not implemented. (just implemented for cpu hotplug)
820 *
821 * If there are kernel internal actions which can make use of some not-exact
822 * value, and reading all cpu value can be performance bottleneck in some
823 * common workload, threashold and synchonization as vmstat[] should be
824 * implemented.
825 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700826static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700827 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800828{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700829 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800830 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800831
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400832 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700833 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800834 return val;
835}
836
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700837static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700838 enum mem_cgroup_events_index idx)
839{
840 unsigned long val = 0;
841 int cpu;
842
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400843 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700844 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700845 return val;
846}
847
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700848static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700849 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700850 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800851{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700852 /*
853 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
854 * counted as CACHE even if it's on ANON LRU.
855 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700856 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700857 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700858 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800859 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700860 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700861 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700862
David Rientjesb070e652013-05-07 16:18:09 -0700863 if (PageTransHuge(page))
864 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
865 nr_pages);
866
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800867 /* pagein of a big page is an event. So, ignore page size */
868 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700869 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800870 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700871 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800872 nr_pages = -nr_pages; /* for event */
873 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800874
Johannes Weiner13114712012-05-29 15:07:07 -0700875 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800876}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800877
Jianyu Zhane2318752014-06-06 14:38:20 -0700878unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700879{
880 struct mem_cgroup_per_zone *mz;
881
882 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
883 return mz->lru_size[lru];
884}
885
Jianyu Zhane2318752014-06-06 14:38:20 -0700886static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
887 int nid,
888 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700889{
Jianyu Zhane2318752014-06-06 14:38:20 -0700890 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700891 int zid;
892
Jianyu Zhane2318752014-06-06 14:38:20 -0700893 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700894
Jianyu Zhane2318752014-06-06 14:38:20 -0700895 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
896 struct mem_cgroup_per_zone *mz;
897 enum lru_list lru;
898
899 for_each_lru(lru) {
900 if (!(BIT(lru) & lru_mask))
901 continue;
902 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
903 nr += mz->lru_size[lru];
904 }
905 }
906 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700907}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700908
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700909static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700910 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800911{
Jianyu Zhane2318752014-06-06 14:38:20 -0700912 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700913 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800914
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800915 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700916 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
917 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800918}
919
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800920static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
921 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800922{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700923 unsigned long val, next;
924
Johannes Weiner13114712012-05-29 15:07:07 -0700925 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700926 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700927 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800928 if ((long)next - (long)val < 0) {
929 switch (target) {
930 case MEM_CGROUP_TARGET_THRESH:
931 next = val + THRESHOLDS_EVENTS_TARGET;
932 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700933 case MEM_CGROUP_TARGET_SOFTLIMIT:
934 next = val + SOFTLIMIT_EVENTS_TARGET;
935 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800936 case MEM_CGROUP_TARGET_NUMAINFO:
937 next = val + NUMAINFO_EVENTS_TARGET;
938 break;
939 default:
940 break;
941 }
942 __this_cpu_write(memcg->stat->targets[target], next);
943 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700944 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800945 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800946}
947
948/*
949 * Check events in order.
950 *
951 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700952static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800953{
954 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800955 if (unlikely(mem_cgroup_event_ratelimit(memcg,
956 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700957 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800958 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800959
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700960 do_softlimit = mem_cgroup_event_ratelimit(memcg,
961 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700962#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800963 do_numainfo = mem_cgroup_event_ratelimit(memcg,
964 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700965#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800966 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700967 if (unlikely(do_softlimit))
968 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800969#if MAX_NUMNODES > 1
970 if (unlikely(do_numainfo))
971 atomic_inc(&memcg->numainfo_events);
972#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700973 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800974}
975
Balbir Singhcf475ad2008-04-29 01:00:16 -0700976struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800977{
Balbir Singh31a78f22008-09-28 23:09:31 +0100978 /*
979 * mm_update_next_owner() may clear mm->owner to NULL
980 * if it races with swapoff, page migration, etc.
981 * So this can be called with p == NULL.
982 */
983 if (unlikely(!p))
984 return NULL;
985
Tejun Heo073219e2014-02-08 10:36:58 -0500986 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800987}
988
Johannes Weinerdf381972014-04-07 15:37:43 -0700989static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800990{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700991 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700992
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800993 rcu_read_lock();
994 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700995 /*
996 * Page cache insertions can happen withou an
997 * actual mm context, e.g. during disk probing
998 * on boot, loopback IO, acct() writes etc.
999 */
1000 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -07001001 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -07001002 else {
1003 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1004 if (unlikely(!memcg))
1005 memcg = root_mem_cgroup;
1006 }
Tejun Heoec903c02014-05-13 12:11:01 -04001007 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001008 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001009 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001010}
1011
Johannes Weiner56600482012-01-12 17:17:59 -08001012/**
1013 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1014 * @root: hierarchy root
1015 * @prev: previously returned memcg, NULL on first invocation
1016 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1017 *
1018 * Returns references to children of the hierarchy below @root, or
1019 * @root itself, or %NULL after a full round-trip.
1020 *
1021 * Caller must pass the return value in @prev on subsequent
1022 * invocations for reference counting, or use mem_cgroup_iter_break()
1023 * to cancel a hierarchy walk before the round-trip is complete.
1024 *
1025 * Reclaimers can specify a zone and a priority level in @reclaim to
1026 * divide up the memcgs in the hierarchy among all concurrent
1027 * reclaimers operating on the same zone and priority.
1028 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001029struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001030 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001031 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001032{
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001033 struct reclaim_iter *uninitialized_var(iter);
1034 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001035 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001036 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001037
Andrew Morton694fbc02013-09-24 15:27:37 -07001038 if (mem_cgroup_disabled())
1039 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001040
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001041 if (!root)
1042 root = root_mem_cgroup;
1043
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001044 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001045 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001046
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001047 if (!root->use_hierarchy && root != root_mem_cgroup) {
1048 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001049 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -07001050 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001051 }
1052
Michal Hocko542f85f2013-04-29 15:07:15 -07001053 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001054
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001055 if (reclaim) {
1056 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001057
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001058 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
1059 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -07001060
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001061 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -07001062 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001063
1064 do {
Jason Low4db0c3c2015-04-15 16:14:08 -07001065 pos = READ_ONCE(iter->position);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001066 /*
1067 * A racing update may change the position and
1068 * put the last reference, hence css_tryget(),
1069 * or retry to see the updated position.
1070 */
1071 } while (pos && !css_tryget(&pos->css));
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001072 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001073
1074 if (pos)
1075 css = &pos->css;
1076
1077 for (;;) {
1078 css = css_next_descendant_pre(css, &root->css);
1079 if (!css) {
1080 /*
1081 * Reclaimers share the hierarchy walk, and a
1082 * new one might jump in right at the end of
1083 * the hierarchy - make sure they see at least
1084 * one group and restart from the beginning.
1085 */
1086 if (!prev)
1087 continue;
1088 break;
1089 }
1090
1091 /*
1092 * Verify the css and acquire a reference. The root
1093 * is provided by the caller, so we know it's alive
1094 * and kicking, and don't take an extra reference.
1095 */
1096 memcg = mem_cgroup_from_css(css);
1097
1098 if (css == &root->css)
1099 break;
1100
Johannes Weinerb2052562014-12-10 15:42:48 -08001101 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001102 /*
1103 * Make sure the memcg is initialized:
1104 * mem_cgroup_css_online() orders the the
1105 * initialization against setting the flag.
1106 */
1107 if (smp_load_acquire(&memcg->initialized))
1108 break;
1109
1110 css_put(css);
1111 }
1112
1113 memcg = NULL;
1114 }
1115
1116 if (reclaim) {
1117 if (cmpxchg(&iter->position, pos, memcg) == pos) {
1118 if (memcg)
1119 css_get(&memcg->css);
1120 if (pos)
1121 css_put(&pos->css);
1122 }
1123
1124 /*
1125 * pairs with css_tryget when dereferencing iter->position
1126 * above.
1127 */
1128 if (pos)
1129 css_put(&pos->css);
1130
1131 if (!memcg)
1132 iter->generation++;
1133 else if (!prev)
1134 reclaim->generation = iter->generation;
1135 }
1136
Michal Hocko542f85f2013-04-29 15:07:15 -07001137out_unlock:
1138 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001139out:
Michal Hockoc40046f2013-04-29 15:07:14 -07001140 if (prev && prev != root)
1141 css_put(&prev->css);
1142
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001143 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001144}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001145
Johannes Weiner56600482012-01-12 17:17:59 -08001146/**
1147 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1148 * @root: hierarchy root
1149 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1150 */
1151void mem_cgroup_iter_break(struct mem_cgroup *root,
1152 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001153{
1154 if (!root)
1155 root = root_mem_cgroup;
1156 if (prev && prev != root)
1157 css_put(&prev->css);
1158}
1159
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001160/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001161 * Iteration constructs for visiting all cgroups (under a tree). If
1162 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1163 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001164 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001165#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001166 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001167 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001168 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001169
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001170#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001171 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001172 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001173 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001174
David Rientjes68ae5642012-12-12 13:51:57 -08001175void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001176{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001177 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001178
Ying Han456f9982011-05-26 16:25:38 -07001179 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001180 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1181 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001182 goto out;
1183
1184 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001185 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001186 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1187 break;
1188 case PGMAJFAULT:
1189 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001190 break;
1191 default:
1192 BUG();
1193 }
1194out:
1195 rcu_read_unlock();
1196}
David Rientjes68ae5642012-12-12 13:51:57 -08001197EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001198
Johannes Weiner925b7672012-01-12 17:18:15 -08001199/**
1200 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1201 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001202 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001203 *
1204 * Returns the lru list vector holding pages for the given @zone and
1205 * @mem. This can be the global zone lruvec, if the memory controller
1206 * is disabled.
1207 */
1208struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1209 struct mem_cgroup *memcg)
1210{
1211 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001212 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001213
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001214 if (mem_cgroup_disabled()) {
1215 lruvec = &zone->lruvec;
1216 goto out;
1217 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001218
Jianyu Zhane2318752014-06-06 14:38:20 -07001219 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001220 lruvec = &mz->lruvec;
1221out:
1222 /*
1223 * Since a node can be onlined after the mem_cgroup was created,
1224 * we have to be prepared to initialize lruvec->zone here;
1225 * and if offlined then reonlined, we need to reinitialize it.
1226 */
1227 if (unlikely(lruvec->zone != zone))
1228 lruvec->zone = zone;
1229 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001230}
1231
Johannes Weiner925b7672012-01-12 17:18:15 -08001232/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001233 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001234 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001235 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001236 *
1237 * This function is only safe when following the LRU page isolation
1238 * and putback protocol: the LRU lock must be held, and the page must
1239 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001240 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001241struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001242{
1243 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001244 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001245 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001246
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001247 if (mem_cgroup_disabled()) {
1248 lruvec = &zone->lruvec;
1249 goto out;
1250 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001251
Johannes Weiner1306a852014-12-10 15:44:52 -08001252 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001253 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001254 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001255 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001256 */
Johannes Weiner29833312014-12-10 15:44:02 -08001257 if (!memcg)
1258 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001259
Jianyu Zhane2318752014-06-06 14:38:20 -07001260 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001261 lruvec = &mz->lruvec;
1262out:
1263 /*
1264 * Since a node can be onlined after the mem_cgroup was created,
1265 * we have to be prepared to initialize lruvec->zone here;
1266 * and if offlined then reonlined, we need to reinitialize it.
1267 */
1268 if (unlikely(lruvec->zone != zone))
1269 lruvec->zone = zone;
1270 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001271}
1272
1273/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001274 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1275 * @lruvec: mem_cgroup per zone lru vector
1276 * @lru: index of lru list the page is sitting on
1277 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001278 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001279 * This function must be called when a page is added to or removed from an
1280 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001281 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001282void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1283 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001284{
1285 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001286 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001287
1288 if (mem_cgroup_disabled())
1289 return;
1290
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001291 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1292 lru_size = mz->lru_size + lru;
1293 *lru_size += nr_pages;
1294 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001295}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001296
Johannes Weiner2314b422014-12-10 15:44:33 -08001297bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root)
Michal Hocko3e920412011-07-26 16:08:29 -07001298{
Johannes Weiner2314b422014-12-10 15:44:33 -08001299 if (root == memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001300 return true;
Johannes Weiner2314b422014-12-10 15:44:33 -08001301 if (!root->use_hierarchy)
Johannes Weiner91c637342012-05-29 15:06:24 -07001302 return false;
Johannes Weiner2314b422014-12-10 15:44:33 -08001303 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001304}
1305
Johannes Weiner2314b422014-12-10 15:44:33 -08001306bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001307{
Johannes Weiner2314b422014-12-10 15:44:33 -08001308 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001309 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001310 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001311
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001312 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001313 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001314 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001315 task_unlock(p);
1316 } else {
1317 /*
1318 * All threads may have already detached their mm's, but the oom
1319 * killer still needs to detect if they have already been oom
1320 * killed to prevent needlessly killing additional tasks.
1321 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001322 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001323 task_memcg = mem_cgroup_from_task(task);
1324 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001325 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001326 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001327 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1328 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001329 return ret;
1330}
1331
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001332int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001333{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001334 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001335 unsigned long inactive;
1336 unsigned long active;
1337 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001338
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001339 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1340 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001341
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001342 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1343 if (gb)
1344 inactive_ratio = int_sqrt(10 * gb);
1345 else
1346 inactive_ratio = 1;
1347
Johannes Weiner9b272972011-11-02 13:38:23 -07001348 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001349}
1350
Vladimir Davydov90cbc252015-02-11 15:25:55 -08001351bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
1352{
1353 struct mem_cgroup_per_zone *mz;
1354 struct mem_cgroup *memcg;
1355
1356 if (mem_cgroup_disabled())
1357 return true;
1358
1359 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1360 memcg = mz->memcg;
1361
1362 return !!(memcg->css.flags & CSS_ONLINE);
1363}
1364
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001365#define mem_cgroup_from_counter(counter, member) \
Balbir Singh6d61ef42009-01-07 18:08:06 -08001366 container_of(counter, struct mem_cgroup, member)
1367
Johannes Weiner19942822011-02-01 15:52:43 -08001368/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001369 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001370 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001371 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001372 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001373 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001374 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001375static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001376{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001377 unsigned long margin = 0;
1378 unsigned long count;
1379 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001380
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001381 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001382 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001383 if (count < limit)
1384 margin = limit - count;
1385
1386 if (do_swap_account) {
1387 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001388 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001389 if (count <= limit)
1390 margin = min(margin, limit - count);
1391 }
1392
1393 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001394}
1395
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001396int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001397{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001398 /* root ? */
Linus Torvalds14208b02014-06-09 15:03:33 -07001399 if (mem_cgroup_disabled() || !memcg->css.parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001400 return vm_swappiness;
1401
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001402 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001403}
1404
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001405/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001406 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001407 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001408 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1409 * moving cgroups. This is for waiting at high-memory pressure
1410 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001411 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001412static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001413{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001414 struct mem_cgroup *from;
1415 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001416 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001417 /*
1418 * Unlike task_move routines, we access mc.to, mc.from not under
1419 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1420 */
1421 spin_lock(&mc.lock);
1422 from = mc.from;
1423 to = mc.to;
1424 if (!from)
1425 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001426
Johannes Weiner2314b422014-12-10 15:44:33 -08001427 ret = mem_cgroup_is_descendant(from, memcg) ||
1428 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001429unlock:
1430 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001431 return ret;
1432}
1433
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001434static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001435{
1436 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001437 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001438 DEFINE_WAIT(wait);
1439 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1440 /* moving charge context might have finished. */
1441 if (mc.moving_task)
1442 schedule();
1443 finish_wait(&mc.waitq, &wait);
1444 return true;
1445 }
1446 }
1447 return false;
1448}
1449
Sha Zhengju58cf1882013-02-22 16:32:05 -08001450#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001451/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001452 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001453 * @memcg: The memory cgroup that went over limit
1454 * @p: Task that is going to be killed
1455 *
1456 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1457 * enabled
1458 */
1459void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1460{
Tejun Heoe61734c2014-02-12 09:29:50 -05001461 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001462 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001463 struct mem_cgroup *iter;
1464 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001465
Michal Hocko08088cb2014-02-25 15:01:44 -08001466 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001467 rcu_read_lock();
1468
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001469 if (p) {
1470 pr_info("Task in ");
1471 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1472 pr_cont(" killed as a result of limit of ");
1473 } else {
1474 pr_info("Memory limit reached of cgroup ");
1475 }
1476
Tejun Heoe61734c2014-02-12 09:29:50 -05001477 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001478 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001479
Balbir Singhe2224322009-04-02 16:57:39 -07001480 rcu_read_unlock();
1481
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001482 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1483 K((u64)page_counter_read(&memcg->memory)),
1484 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1485 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1486 K((u64)page_counter_read(&memcg->memsw)),
1487 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1488 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1489 K((u64)page_counter_read(&memcg->kmem)),
1490 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001491
1492 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001493 pr_info("Memory cgroup stats for ");
1494 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001495 pr_cont(":");
1496
1497 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1498 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1499 continue;
1500 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1501 K(mem_cgroup_read_stat(iter, i)));
1502 }
1503
1504 for (i = 0; i < NR_LRU_LISTS; i++)
1505 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1506 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1507
1508 pr_cont("\n");
1509 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001510 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001511}
1512
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001513/*
1514 * This function returns the number of memcg under hierarchy tree. Returns
1515 * 1(self count) if no children.
1516 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001517static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001518{
1519 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001520 struct mem_cgroup *iter;
1521
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001522 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001523 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001524 return num;
1525}
1526
Balbir Singh6d61ef42009-01-07 18:08:06 -08001527/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001528 * Return the memory (and swap, if configured) limit for a memcg.
1529 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001530static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001531{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001532 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001533
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001534 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001535 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001536 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001537
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001538 memsw_limit = memcg->memsw.limit;
1539 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001540 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001541 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001542}
1543
David Rientjes19965462012-12-11 16:00:26 -08001544static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1545 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001546{
1547 struct mem_cgroup *iter;
1548 unsigned long chosen_points = 0;
1549 unsigned long totalpages;
1550 unsigned int points = 0;
1551 struct task_struct *chosen = NULL;
1552
David Rientjes876aafb2012-07-31 16:43:48 -07001553 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001554 * If current has a pending SIGKILL or is exiting, then automatically
1555 * select it. The goal is to allow it to allocate so that it may
1556 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001557 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001558 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Michal Hocko49550b62015-02-11 15:26:12 -08001559 mark_tsk_oom_victim(current);
David Rientjes876aafb2012-07-31 16:43:48 -07001560 return;
1561 }
1562
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001563 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001564 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001565 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001566 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001567 struct task_struct *task;
1568
Tejun Heo72ec7022013-08-08 20:11:26 -04001569 css_task_iter_start(&iter->css, &it);
1570 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001571 switch (oom_scan_process_thread(task, totalpages, NULL,
1572 false)) {
1573 case OOM_SCAN_SELECT:
1574 if (chosen)
1575 put_task_struct(chosen);
1576 chosen = task;
1577 chosen_points = ULONG_MAX;
1578 get_task_struct(chosen);
1579 /* fall through */
1580 case OOM_SCAN_CONTINUE:
1581 continue;
1582 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001583 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001584 mem_cgroup_iter_break(memcg, iter);
1585 if (chosen)
1586 put_task_struct(chosen);
1587 return;
1588 case OOM_SCAN_OK:
1589 break;
1590 };
1591 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001592 if (!points || points < chosen_points)
1593 continue;
1594 /* Prefer thread group leaders for display purposes */
1595 if (points == chosen_points &&
1596 thread_group_leader(chosen))
1597 continue;
1598
1599 if (chosen)
1600 put_task_struct(chosen);
1601 chosen = task;
1602 chosen_points = points;
1603 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001604 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001605 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001606 }
1607
1608 if (!chosen)
1609 return;
1610 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001611 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1612 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001613}
1614
Michele Curtiae6e71d2014-12-12 16:56:35 -08001615#if MAX_NUMNODES > 1
1616
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001617/**
1618 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001619 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001620 * @nid: the node ID to be checked.
1621 * @noswap : specify true here if the user wants flle only information.
1622 *
1623 * This function returns whether the specified memcg contains any
1624 * reclaimable pages on a node. Returns true if there are any reclaimable
1625 * pages in the node.
1626 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001628 int nid, bool noswap)
1629{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001630 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001631 return true;
1632 if (noswap || !total_swap_pages)
1633 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001634 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001635 return true;
1636 return false;
1637
1638}
Ying Han889976d2011-05-26 16:25:33 -07001639
1640/*
1641 * Always updating the nodemask is not very good - even if we have an empty
1642 * list or the wrong list here, we can start from some node and traverse all
1643 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1644 *
1645 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001646static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001647{
1648 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001649 /*
1650 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1651 * pagein/pageout changes since the last update.
1652 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001653 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001654 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001655 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001656 return;
1657
Ying Han889976d2011-05-26 16:25:33 -07001658 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001659 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001660
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001661 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001662
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001663 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1664 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001665 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001666
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001667 atomic_set(&memcg->numainfo_events, 0);
1668 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001669}
1670
1671/*
1672 * Selecting a node where we start reclaim from. Because what we need is just
1673 * reducing usage counter, start from anywhere is O,K. Considering
1674 * memory reclaim from current node, there are pros. and cons.
1675 *
1676 * Freeing memory from current node means freeing memory from a node which
1677 * we'll use or we've used. So, it may make LRU bad. And if several threads
1678 * hit limits, it will see a contention on a node. But freeing from remote
1679 * node means more costs for memory reclaim because of memory latency.
1680 *
1681 * Now, we use round-robin. Better algorithm is welcomed.
1682 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001683int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001684{
1685 int node;
1686
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001687 mem_cgroup_may_update_nodemask(memcg);
1688 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001689
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001690 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001691 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001692 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001693 /*
1694 * We call this when we hit limit, not when pages are added to LRU.
1695 * No LRU may hold pages because all pages are UNEVICTABLE or
1696 * memcg is too small and all pages are not on LRU. In that case,
1697 * we use curret node.
1698 */
1699 if (unlikely(node == MAX_NUMNODES))
1700 node = numa_node_id();
1701
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001702 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001703 return node;
1704}
Ying Han889976d2011-05-26 16:25:33 -07001705#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001706int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001707{
1708 return 0;
1709}
1710#endif
1711
Andrew Morton0608f432013-09-24 15:27:41 -07001712static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1713 struct zone *zone,
1714 gfp_t gfp_mask,
1715 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001716{
Andrew Morton0608f432013-09-24 15:27:41 -07001717 struct mem_cgroup *victim = NULL;
1718 int total = 0;
1719 int loop = 0;
1720 unsigned long excess;
1721 unsigned long nr_scanned;
1722 struct mem_cgroup_reclaim_cookie reclaim = {
1723 .zone = zone,
1724 .priority = 0,
1725 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001726
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001727 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001728
Andrew Morton0608f432013-09-24 15:27:41 -07001729 while (1) {
1730 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1731 if (!victim) {
1732 loop++;
1733 if (loop >= 2) {
1734 /*
1735 * If we have not been able to reclaim
1736 * anything, it might because there are
1737 * no reclaimable pages under this hierarchy
1738 */
1739 if (!total)
1740 break;
1741 /*
1742 * We want to do more targeted reclaim.
1743 * excess >> 2 is not to excessive so as to
1744 * reclaim too much, nor too less that we keep
1745 * coming back to reclaim from this cgroup
1746 */
1747 if (total >= (excess >> 2) ||
1748 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1749 break;
1750 }
1751 continue;
1752 }
Andrew Morton0608f432013-09-24 15:27:41 -07001753 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1754 zone, &nr_scanned);
1755 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001756 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001757 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001758 }
Andrew Morton0608f432013-09-24 15:27:41 -07001759 mem_cgroup_iter_break(root_memcg, victim);
1760 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001761}
1762
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001763#ifdef CONFIG_LOCKDEP
1764static struct lockdep_map memcg_oom_lock_dep_map = {
1765 .name = "memcg_oom_lock",
1766};
1767#endif
1768
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001769static DEFINE_SPINLOCK(memcg_oom_lock);
1770
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001771/*
1772 * Check OOM-Killer is already running under our hierarchy.
1773 * If someone is running, return false.
1774 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001775static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001776{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001777 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001778
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001779 spin_lock(&memcg_oom_lock);
1780
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001781 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001782 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001783 /*
1784 * this subtree of our hierarchy is already locked
1785 * so we cannot give a lock.
1786 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001787 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001788 mem_cgroup_iter_break(memcg, iter);
1789 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001790 } else
1791 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001792 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001793
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001794 if (failed) {
1795 /*
1796 * OK, we failed to lock the whole subtree so we have
1797 * to clean up what we set up to the failing subtree
1798 */
1799 for_each_mem_cgroup_tree(iter, memcg) {
1800 if (iter == failed) {
1801 mem_cgroup_iter_break(memcg, iter);
1802 break;
1803 }
1804 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001805 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001806 } else
1807 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001808
1809 spin_unlock(&memcg_oom_lock);
1810
1811 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001812}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001813
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001814static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001815{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001816 struct mem_cgroup *iter;
1817
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001818 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001819 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001820 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001821 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001822 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001823}
1824
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001825static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001826{
1827 struct mem_cgroup *iter;
1828
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001829 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001830 atomic_inc(&iter->under_oom);
1831}
1832
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001833static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001834{
1835 struct mem_cgroup *iter;
1836
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001837 /*
1838 * When a new child is created while the hierarchy is under oom,
1839 * mem_cgroup_oom_lock() may not be called. We have to use
1840 * atomic_add_unless() here.
1841 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001842 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001843 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001844}
1845
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001846static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1847
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001848struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001849 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001850 wait_queue_t wait;
1851};
1852
1853static int memcg_oom_wake_function(wait_queue_t *wait,
1854 unsigned mode, int sync, void *arg)
1855{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001856 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1857 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001858 struct oom_wait_info *oom_wait_info;
1859
1860 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001861 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001862
Johannes Weiner2314b422014-12-10 15:44:33 -08001863 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1864 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001865 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001866 return autoremove_wake_function(wait, mode, sync, arg);
1867}
1868
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001869static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001870{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001871 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001872 /* for filtering, pass "memcg" as argument. */
1873 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001874}
1875
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001876static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001877{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001878 if (memcg && atomic_read(&memcg->under_oom))
1879 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001880}
1881
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001882static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001883{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001884 if (!current->memcg_oom.may_oom)
1885 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001886 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001887 * We are in the middle of the charge context here, so we
1888 * don't want to block when potentially sitting on a callstack
1889 * that holds all kinds of filesystem and mm locks.
1890 *
1891 * Also, the caller may handle a failed allocation gracefully
1892 * (like optional page cache readahead) and so an OOM killer
1893 * invocation might not even be necessary.
1894 *
1895 * That's why we don't do anything here except remember the
1896 * OOM context and then deal with it at the end of the page
1897 * fault when the stack is unwound, the locks are released,
1898 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001899 */
Johannes Weiner49426422013-10-16 13:46:59 -07001900 css_get(&memcg->css);
1901 current->memcg_oom.memcg = memcg;
1902 current->memcg_oom.gfp_mask = mask;
1903 current->memcg_oom.order = order;
1904}
1905
1906/**
1907 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1908 * @handle: actually kill/wait or just clean up the OOM state
1909 *
1910 * This has to be called at the end of a page fault if the memcg OOM
1911 * handler was enabled.
1912 *
1913 * Memcg supports userspace OOM handling where failed allocations must
1914 * sleep on a waitqueue until the userspace task resolves the
1915 * situation. Sleeping directly in the charge context with all kinds
1916 * of locks held is not a good idea, instead we remember an OOM state
1917 * in the task and mem_cgroup_oom_synchronize() has to be called at
1918 * the end of the page fault to complete the OOM handling.
1919 *
1920 * Returns %true if an ongoing memcg OOM situation was detected and
1921 * completed, %false otherwise.
1922 */
1923bool mem_cgroup_oom_synchronize(bool handle)
1924{
1925 struct mem_cgroup *memcg = current->memcg_oom.memcg;
1926 struct oom_wait_info owait;
1927 bool locked;
1928
1929 /* OOM is global, do not handle */
1930 if (!memcg)
1931 return false;
1932
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001933 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001934 goto cleanup;
1935
1936 owait.memcg = memcg;
1937 owait.wait.flags = 0;
1938 owait.wait.func = memcg_oom_wake_function;
1939 owait.wait.private = current;
1940 INIT_LIST_HEAD(&owait.wait.task_list);
1941
1942 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001943 mem_cgroup_mark_under_oom(memcg);
1944
1945 locked = mem_cgroup_oom_trylock(memcg);
1946
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001947 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001948 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001949
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001950 if (locked && !memcg->oom_kill_disable) {
1951 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001952 finish_wait(&memcg_oom_waitq, &owait.wait);
1953 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
1954 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001955 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001956 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001957 mem_cgroup_unmark_under_oom(memcg);
1958 finish_wait(&memcg_oom_waitq, &owait.wait);
1959 }
1960
1961 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001962 mem_cgroup_oom_unlock(memcg);
1963 /*
1964 * There is no guarantee that an OOM-lock contender
1965 * sees the wakeups triggered by the OOM kill
1966 * uncharges. Wake any sleepers explicitely.
1967 */
1968 memcg_oom_recover(memcg);
1969 }
Johannes Weiner49426422013-10-16 13:46:59 -07001970cleanup:
1971 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001972 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001973 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001974}
1975
Johannes Weinerd7365e72014-10-29 14:50:48 -07001976/**
1977 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1978 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001979 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001980 * This function must mark the beginning of an accounted page state
1981 * change to prevent double accounting when the page is concurrently
1982 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001983 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001984 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001985 * if (TestClearPageState(page))
1986 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001987 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001988 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001989struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001990{
1991 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001992 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001993
Johannes Weiner6de22612015-02-11 15:25:01 -08001994 /*
1995 * The RCU lock is held throughout the transaction. The fast
1996 * path can get away without acquiring the memcg->move_lock
1997 * because page moving starts with an RCU grace period.
1998 *
1999 * The RCU lock also protects the memcg from being freed when
2000 * the page state that is going to change is the only thing
2001 * preventing the page from being uncharged.
2002 * E.g. end-writeback clearing PageWriteback(), which allows
2003 * migration to go ahead and uncharge the page before the
2004 * account transaction might be complete.
2005 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07002006 rcu_read_lock();
2007
2008 if (mem_cgroup_disabled())
2009 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002010again:
Johannes Weiner1306a852014-12-10 15:44:52 -08002011 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002012 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07002013 return NULL;
2014
Qiang Huangbdcbb652014-06-04 16:08:21 -07002015 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07002016 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002017
Johannes Weiner6de22612015-02-11 15:25:01 -08002018 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08002019 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08002020 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002021 goto again;
2022 }
Johannes Weiner6de22612015-02-11 15:25:01 -08002023
2024 /*
2025 * When charge migration first begins, we can have locked and
2026 * unlocked page stat updates happening concurrently. Track
2027 * the task who has the lock for mem_cgroup_end_page_stat().
2028 */
2029 memcg->move_lock_task = current;
2030 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002031
2032 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002033}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04002034EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002035
Johannes Weinerd7365e72014-10-29 14:50:48 -07002036/**
2037 * mem_cgroup_end_page_stat - finish a page state statistics transaction
2038 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07002039 */
Johannes Weiner6de22612015-02-11 15:25:01 -08002040void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002041{
Johannes Weiner6de22612015-02-11 15:25:01 -08002042 if (memcg && memcg->move_lock_task == current) {
2043 unsigned long flags = memcg->move_lock_flags;
2044
2045 memcg->move_lock_task = NULL;
2046 memcg->move_lock_flags = 0;
2047
2048 spin_unlock_irqrestore(&memcg->move_lock, flags);
2049 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002050
Johannes Weinerd7365e72014-10-29 14:50:48 -07002051 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002052}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04002053EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002054
Johannes Weinerd7365e72014-10-29 14:50:48 -07002055/**
2056 * mem_cgroup_update_page_stat - update page state statistics
2057 * @memcg: memcg to account against
2058 * @idx: page state item to account
2059 * @val: number of pages (positive or negative)
2060 *
2061 * See mem_cgroup_begin_page_stat() for locking requirements.
2062 */
2063void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
Sha Zhengju68b48762013-09-12 15:13:50 -07002064 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002065{
Sha Zhengju658b72c2013-09-12 15:13:52 -07002066 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002067
Johannes Weinerd7365e72014-10-29 14:50:48 -07002068 if (memcg)
2069 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002070}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002071
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002072/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002073 * size of first charge trial. "32" comes from vmscan.c's magic value.
2074 * TODO: maybe necessary to use big numbers in big irons.
2075 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002076#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002077struct memcg_stock_pcp {
2078 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002079 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002080 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002081 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002082#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002083};
2084static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002085static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002086
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002087/**
2088 * consume_stock: Try to consume stocked charge on this cpu.
2089 * @memcg: memcg to consume from.
2090 * @nr_pages: how many pages to charge.
2091 *
2092 * The charges will only happen if @memcg matches the current cpu's memcg
2093 * stock, and at least @nr_pages are available in that stock. Failure to
2094 * service an allocation will refill the stock.
2095 *
2096 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002097 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002098static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002099{
2100 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002101 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002102
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002103 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002104 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002105
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002106 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002107 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002108 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002109 ret = true;
2110 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002111 put_cpu_var(memcg_stock);
2112 return ret;
2113}
2114
2115/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002116 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002117 */
2118static void drain_stock(struct memcg_stock_pcp *stock)
2119{
2120 struct mem_cgroup *old = stock->cached;
2121
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002122 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002123 page_counter_uncharge(&old->memory, stock->nr_pages);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002124 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002125 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002126 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002127 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002128 }
2129 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002130}
2131
2132/*
2133 * This must be called under preempt disabled or must be called by
2134 * a thread which is pinned to local cpu.
2135 */
2136static void drain_local_stock(struct work_struct *dummy)
2137{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002138 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002139 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002140 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002141}
2142
2143/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002144 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002145 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002146 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002147static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002148{
2149 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2150
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002151 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002152 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002153 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002154 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002155 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002156 put_cpu_var(memcg_stock);
2157}
2158
2159/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002160 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002161 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002162 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002163static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002164{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002165 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002166
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002167 /* If someone's already draining, avoid adding running more workers. */
2168 if (!mutex_trylock(&percpu_charge_mutex))
2169 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002170 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002171 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002172 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002173 for_each_online_cpu(cpu) {
2174 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002175 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002176
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002177 memcg = stock->cached;
2178 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002179 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08002180 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002181 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002182 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2183 if (cpu == curcpu)
2184 drain_local_stock(&stock->work);
2185 else
2186 schedule_work_on(cpu, &stock->work);
2187 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002188 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002189 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002190 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02002191 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002192}
2193
Paul Gortmaker0db06282013-06-19 14:53:51 -04002194static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002195 unsigned long action,
2196 void *hcpu)
2197{
2198 int cpu = (unsigned long)hcpu;
2199 struct memcg_stock_pcp *stock;
2200
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002201 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002202 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002203
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002204 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002205 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002206
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002207 stock = &per_cpu(memcg_stock, cpu);
2208 drain_stock(stock);
2209 return NOTIFY_OK;
2210}
2211
Johannes Weiner00501b52014-08-08 14:19:20 -07002212static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2213 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002214{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002215 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07002216 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002217 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002218 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002219 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002220 bool may_swap = true;
2221 bool drained = false;
Johannes Weiner05b84302014-08-06 16:05:59 -07002222 int ret = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002223
Johannes Weinerce00a962014-09-05 08:43:57 -04002224 if (mem_cgroup_is_root(memcg))
2225 goto done;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002226retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002227 if (consume_stock(memcg, nr_pages))
2228 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002229
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002230 if (!do_swap_account ||
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002231 !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2232 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002233 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002234 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002235 page_counter_uncharge(&memcg->memsw, batch);
2236 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002237 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002238 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002239 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002240 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002241
Johannes Weiner6539cc02014-08-06 16:05:42 -07002242 if (batch > nr_pages) {
2243 batch = nr_pages;
2244 goto retry;
2245 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002246
Johannes Weiner06b078f2014-08-06 16:05:44 -07002247 /*
2248 * Unlike in global OOM situations, memcg is not in a physical
2249 * memory shortage. Allow dying and OOM-killed tasks to
2250 * bypass the last charges so that they can exit quickly and
2251 * free their memory.
2252 */
2253 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2254 fatal_signal_pending(current) ||
2255 current->flags & PF_EXITING))
2256 goto bypass;
2257
2258 if (unlikely(task_in_memcg_oom(current)))
2259 goto nomem;
2260
Johannes Weiner6539cc02014-08-06 16:05:42 -07002261 if (!(gfp_mask & __GFP_WAIT))
2262 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002263
Johannes Weiner241994ed2015-02-11 15:26:06 -08002264 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2265
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002266 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2267 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002268
Johannes Weiner61e02c72014-08-06 16:08:16 -07002269 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002270 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002271
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002272 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002273 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002274 drained = true;
2275 goto retry;
2276 }
2277
Johannes Weiner28c34c22014-08-06 16:05:47 -07002278 if (gfp_mask & __GFP_NORETRY)
2279 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002280 /*
2281 * Even though the limit is exceeded at this point, reclaim
2282 * may have been able to free some pages. Retry the charge
2283 * before killing the task.
2284 *
2285 * Only for regular pages, though: huge pages are rather
2286 * unlikely to succeed so close to the limit, and we fall back
2287 * to regular pages anyway in case of failure.
2288 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002289 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002290 goto retry;
2291 /*
2292 * At task move, charge accounts can be doubly counted. So, it's
2293 * better to wait until the end of task_move if something is going on.
2294 */
2295 if (mem_cgroup_wait_acct_move(mem_over_limit))
2296 goto retry;
2297
Johannes Weiner9b130612014-08-06 16:05:51 -07002298 if (nr_retries--)
2299 goto retry;
2300
Johannes Weiner06b078f2014-08-06 16:05:44 -07002301 if (gfp_mask & __GFP_NOFAIL)
2302 goto bypass;
2303
Johannes Weiner6539cc02014-08-06 16:05:42 -07002304 if (fatal_signal_pending(current))
2305 goto bypass;
2306
Johannes Weiner241994ed2015-02-11 15:26:06 -08002307 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2308
Johannes Weiner61e02c72014-08-06 16:08:16 -07002309 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002310nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002311 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002312 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002313bypass:
Johannes Weinerce00a962014-09-05 08:43:57 -04002314 return -EINTR;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002315
2316done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002317 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002318 if (batch > nr_pages)
2319 refill_stock(memcg, batch - nr_pages);
Vladimir Davydov7d638092015-06-10 11:14:46 -07002320 if (!(gfp_mask & __GFP_WAIT))
2321 goto done;
Johannes Weiner241994ed2015-02-11 15:26:06 -08002322 /*
2323 * If the hierarchy is above the normal consumption range,
2324 * make the charging task trim their excess contribution.
2325 */
2326 do {
2327 if (page_counter_read(&memcg->memory) <= memcg->high)
2328 continue;
2329 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
2330 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
2331 } while ((memcg = parent_mem_cgroup(memcg)));
Johannes Weiner6539cc02014-08-06 16:05:42 -07002332done:
Johannes Weiner05b84302014-08-06 16:05:59 -07002333 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002334}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002335
Johannes Weiner00501b52014-08-08 14:19:20 -07002336static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002337{
Johannes Weinerce00a962014-09-05 08:43:57 -04002338 if (mem_cgroup_is_root(memcg))
2339 return;
2340
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002341 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner05b84302014-08-06 16:05:59 -07002342 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002343 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002344
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002345 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002346}
2347
2348/*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002349 * try_get_mem_cgroup_from_page - look up page's memcg association
2350 * @page: the page
2351 *
2352 * Look up, get a css reference, and return the memcg that owns @page.
2353 *
2354 * The page must be locked to prevent racing with swap-in and page
2355 * cache charges. If coming from an unlocked page table, the caller
2356 * must ensure the page is on the LRU or this can race with charging.
2357 */
Wu Fengguange42d9d52009-12-16 12:19:59 +01002358struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002359{
Johannes Weiner29833312014-12-10 15:44:02 -08002360 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002361 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002362 swp_entry_t ent;
2363
Sasha Levin309381fea2014-01-23 15:52:54 -08002364 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002365
Johannes Weiner1306a852014-12-10 15:44:52 -08002366 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002367 if (memcg) {
2368 if (!css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002369 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002370 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002371 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002372 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002373 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07002374 memcg = mem_cgroup_from_id(id);
Tejun Heoec903c02014-05-13 12:11:01 -04002375 if (memcg && !css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002376 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002377 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002378 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002379 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002380}
2381
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002382static void lock_page_lru(struct page *page, int *isolated)
2383{
2384 struct zone *zone = page_zone(page);
2385
2386 spin_lock_irq(&zone->lru_lock);
2387 if (PageLRU(page)) {
2388 struct lruvec *lruvec;
2389
2390 lruvec = mem_cgroup_page_lruvec(page, zone);
2391 ClearPageLRU(page);
2392 del_page_from_lru_list(page, lruvec, page_lru(page));
2393 *isolated = 1;
2394 } else
2395 *isolated = 0;
2396}
2397
2398static void unlock_page_lru(struct page *page, int isolated)
2399{
2400 struct zone *zone = page_zone(page);
2401
2402 if (isolated) {
2403 struct lruvec *lruvec;
2404
2405 lruvec = mem_cgroup_page_lruvec(page, zone);
2406 VM_BUG_ON_PAGE(PageLRU(page), page);
2407 SetPageLRU(page);
2408 add_page_to_lru_list(page, lruvec, page_lru(page));
2409 }
2410 spin_unlock_irq(&zone->lru_lock);
2411}
2412
Johannes Weiner00501b52014-08-08 14:19:20 -07002413static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002414 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002415{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002416 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002417
Johannes Weiner1306a852014-12-10 15:44:52 -08002418 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002419
2420 /*
2421 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2422 * may already be on some other mem_cgroup's LRU. Take care of it.
2423 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002424 if (lrucare)
2425 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002426
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002427 /*
2428 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002429 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002430 *
2431 * - the page is uncharged
2432 *
2433 * - the page is off-LRU
2434 *
2435 * - an anonymous fault has exclusive page access, except for
2436 * a locked page table
2437 *
2438 * - a page cache insertion, a swapin fault, or a migration
2439 * have the page locked
2440 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002441 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002442
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002443 if (lrucare)
2444 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002445}
2446
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002447#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002448int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2449 unsigned long nr_pages)
Glauber Costa749c5412012-12-18 14:23:01 -08002450{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002451 struct page_counter *counter;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002452 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002453
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002454 ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2455 if (ret < 0)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002456 return ret;
2457
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002458 ret = try_charge(memcg, gfp, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002459 if (ret == -EINTR) {
2460 /*
Johannes Weiner00501b52014-08-08 14:19:20 -07002461 * try_charge() chose to bypass to root due to OOM kill or
2462 * fatal signal. Since our only options are to either fail
2463 * the allocation or charge it to this cgroup, do it as a
2464 * temporary condition. But we can't fail. From a kmem/slab
2465 * perspective, the cache has already been selected, by
2466 * mem_cgroup_kmem_get_cache(), so it is too late to change
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002467 * our minds.
2468 *
2469 * This condition will only trigger if the task entered
Johannes Weiner00501b52014-08-08 14:19:20 -07002470 * memcg_charge_kmem in a sane state, but was OOM-killed
2471 * during try_charge() above. Tasks that were already dying
2472 * when the allocation triggers should have been already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002473 * directed to the root cgroup in memcontrol.h
2474 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002475 page_counter_charge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002476 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002477 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002478 css_get_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002479 ret = 0;
2480 } else if (ret)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002481 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002482
2483 return ret;
2484}
2485
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002486void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002487{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002488 page_counter_uncharge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002489 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002490 page_counter_uncharge(&memcg->memsw, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002491
Johannes Weiner64f21992014-12-10 15:42:45 -08002492 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002493
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002494 css_put_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002495}
2496
Glauber Costa2633d7a2012-12-18 14:22:34 -08002497/*
2498 * helper for acessing a memcg's index. It will be used as an index in the
2499 * child cache array in kmem_cache, and also to derive its name. This function
2500 * will return -1 when this is not a kmem-limited memcg.
2501 */
2502int memcg_cache_id(struct mem_cgroup *memcg)
2503{
2504 return memcg ? memcg->kmemcg_id : -1;
2505}
2506
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002507static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002508{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002509 int id, size;
2510 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002511
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002512 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002513 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2514 if (id < 0)
2515 return id;
2516
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002517 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002518 return id;
2519
2520 /*
2521 * There's no space for the new id in memcg_caches arrays,
2522 * so we have to grow them.
2523 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002524 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002525
2526 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002527 if (size < MEMCG_CACHES_MIN_SIZE)
2528 size = MEMCG_CACHES_MIN_SIZE;
2529 else if (size > MEMCG_CACHES_MAX_SIZE)
2530 size = MEMCG_CACHES_MAX_SIZE;
2531
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002532 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002533 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002534 err = memcg_update_all_list_lrus(size);
2535 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002536 memcg_nr_cache_ids = size;
2537
2538 up_write(&memcg_cache_ids_sem);
2539
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002540 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002541 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002542 return err;
2543 }
2544 return id;
2545}
2546
2547static void memcg_free_cache_id(int id)
2548{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002549 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002550}
2551
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002552struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002553 struct mem_cgroup *memcg;
2554 struct kmem_cache *cachep;
2555 struct work_struct work;
2556};
2557
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002558static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002559{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002560 struct memcg_kmem_cache_create_work *cw =
2561 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002562 struct mem_cgroup *memcg = cw->memcg;
2563 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002564
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002565 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002566
Vladimir Davydov5722d092014-04-07 15:39:24 -07002567 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002568 kfree(cw);
2569}
2570
2571/*
2572 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002573 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002574static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2575 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002576{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002577 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002578
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002579 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002580 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002581 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002582
2583 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002584
2585 cw->memcg = memcg;
2586 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002587 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002588
Glauber Costad7f25f82012-12-18 14:22:40 -08002589 schedule_work(&cw->work);
2590}
2591
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002592static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2593 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002594{
2595 /*
2596 * We need to stop accounting when we kmalloc, because if the
2597 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002598 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002599 *
2600 * However, it is better to enclose the whole function. Depending on
2601 * the debugging options enabled, INIT_WORK(), for instance, can
2602 * trigger an allocation. This too, will make us recurse. Because at
2603 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2604 * the safest choice is to do it like this, wrapping the whole function.
2605 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002606 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002607 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002608 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002609}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002610
Glauber Costad7f25f82012-12-18 14:22:40 -08002611/*
2612 * Return the kmem_cache we're supposed to use for a slab allocation.
2613 * We try to use the current memcg's version of the cache.
2614 *
2615 * If the cache does not exist yet, if we are the first user of it,
2616 * we either create it immediately, if possible, or create it asynchronously
2617 * in a workqueue.
2618 * In the latter case, we will let the current allocation go through with
2619 * the original cache.
2620 *
2621 * Can't be called in interrupt context or from kernel threads.
2622 * This function needs to be called with rcu_read_lock() held.
2623 */
Zhang Zhen056b7cc2014-12-12 16:55:38 -08002624struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002625{
2626 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002627 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002628 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002629
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002630 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002631
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002632 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002633 return cachep;
2634
Vladimir Davydov8135be52014-12-12 16:56:38 -08002635 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002636 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002637 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002638 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002639
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002640 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002641 if (likely(memcg_cachep))
2642 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002643
2644 /*
2645 * If we are in a safe context (can wait, and not in interrupt
2646 * context), we could be be predictable and return right away.
2647 * This would guarantee that the allocation being performed
2648 * already belongs in the new cache.
2649 *
2650 * However, there are some clashes that can arrive from locking.
2651 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002652 * memcg_create_kmem_cache, this means no further allocation
2653 * could happen with the slab_mutex held. So it's better to
2654 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002655 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002656 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002657out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002658 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002659 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002660}
Glauber Costad7f25f82012-12-18 14:22:40 -08002661
Vladimir Davydov8135be52014-12-12 16:56:38 -08002662void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2663{
2664 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002665 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002666}
2667
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002668/*
2669 * We need to verify if the allocation against current->mm->owner's memcg is
2670 * possible for the given order. But the page is not allocated yet, so we'll
2671 * need a further commit step to do the final arrangements.
2672 *
2673 * It is possible for the task to switch cgroups in this mean time, so at
2674 * commit time, we can't rely on task conversion any longer. We'll then use
2675 * the handle argument to return to the caller which cgroup we should commit
2676 * against. We could also return the memcg directly and avoid the pointer
2677 * passing, but a boolean return value gives better semantics considering
2678 * the compiled-out case as well.
2679 *
2680 * Returning true means the allocation is possible.
2681 */
2682bool
2683__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2684{
2685 struct mem_cgroup *memcg;
2686 int ret;
2687
2688 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07002689
Johannes Weinerdf381972014-04-07 15:37:43 -07002690 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002691
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07002692 if (!memcg_kmem_is_active(memcg)) {
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002693 css_put(&memcg->css);
2694 return true;
2695 }
2696
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002697 ret = memcg_charge_kmem(memcg, gfp, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002698 if (!ret)
2699 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002700
2701 css_put(&memcg->css);
2702 return (ret == 0);
2703}
2704
2705void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2706 int order)
2707{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002708 VM_BUG_ON(mem_cgroup_is_root(memcg));
2709
2710 /* The page allocation failed. Revert */
2711 if (!page) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002712 memcg_uncharge_kmem(memcg, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002713 return;
2714 }
Johannes Weiner1306a852014-12-10 15:44:52 -08002715 page->mem_cgroup = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002716}
2717
2718void __memcg_kmem_uncharge_pages(struct page *page, int order)
2719{
Johannes Weiner1306a852014-12-10 15:44:52 -08002720 struct mem_cgroup *memcg = page->mem_cgroup;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002721
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002722 if (!memcg)
2723 return;
2724
Sasha Levin309381fea2014-01-23 15:52:54 -08002725 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002726
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002727 memcg_uncharge_kmem(memcg, 1 << order);
Johannes Weiner1306a852014-12-10 15:44:52 -08002728 page->mem_cgroup = NULL;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002729}
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002730
2731struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2732{
2733 struct mem_cgroup *memcg = NULL;
2734 struct kmem_cache *cachep;
2735 struct page *page;
2736
2737 page = virt_to_head_page(ptr);
2738 if (PageSlab(page)) {
2739 cachep = page->slab_cache;
2740 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002741 memcg = cachep->memcg_params.memcg;
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002742 } else
2743 /* page allocated by alloc_kmem_pages */
2744 memcg = page->mem_cgroup;
2745
2746 return memcg;
2747}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002748#endif /* CONFIG_MEMCG_KMEM */
2749
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002750#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2751
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002752/*
2753 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002754 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2755 * charge/uncharge will be never happen and move_account() is done under
2756 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002757 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002758void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002759{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002760 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002761
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002762 if (mem_cgroup_disabled())
2763 return;
David Rientjesb070e652013-05-07 16:18:09 -07002764
Johannes Weiner29833312014-12-10 15:44:02 -08002765 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002766 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002767
Johannes Weiner1306a852014-12-10 15:44:52 -08002768 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002769 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002770}
Hugh Dickins12d27102012-01-12 17:19:52 -08002771#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002772
Andrew Mortonc255a452012-07-31 16:43:02 -07002773#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002774static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2775 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002776{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002777 int val = (charge) ? 1 : -1;
2778 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002779}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002780
2781/**
2782 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2783 * @entry: swap entry to be moved
2784 * @from: mem_cgroup which the entry is moved from
2785 * @to: mem_cgroup which the entry is moved to
2786 *
2787 * It succeeds only when the swap_cgroup's record for this entry is the same
2788 * as the mem_cgroup's id of @from.
2789 *
2790 * Returns 0 on success, -EINVAL on failure.
2791 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002793 * both res and memsw, and called css_get().
2794 */
2795static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002796 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002797{
2798 unsigned short old_id, new_id;
2799
Li Zefan34c00c32013-09-23 16:56:01 +08002800 old_id = mem_cgroup_id(from);
2801 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002802
2803 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002804 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002805 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002806 return 0;
2807 }
2808 return -EINVAL;
2809}
2810#else
2811static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002812 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002813{
2814 return -EINVAL;
2815}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002816#endif
2817
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002818static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002819
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002820static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002821 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002822{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002823 unsigned long curusage;
2824 unsigned long oldusage;
2825 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002826 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002827 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002828
2829 /*
2830 * For keeping hierarchical_reclaim simple, how long we should retry
2831 * is depends on callers. We set our retry-count to be function
2832 * of # of children which we should visit in this loop.
2833 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002834 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2835 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002836
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002837 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002838
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002839 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002840 if (signal_pending(current)) {
2841 ret = -EINTR;
2842 break;
2843 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002844
2845 mutex_lock(&memcg_limit_mutex);
2846 if (limit > memcg->memsw.limit) {
2847 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002848 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002849 break;
2850 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002851 if (limit > memcg->memory.limit)
2852 enlarge = true;
2853 ret = page_counter_limit(&memcg->memory, limit);
2854 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002855
2856 if (!ret)
2857 break;
2858
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002859 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2860
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002861 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002862 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002863 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002864 retry_count--;
2865 else
2866 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002867 } while (retry_count);
2868
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002869 if (!ret && enlarge)
2870 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002871
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002872 return ret;
2873}
2874
Li Zefan338c8432009-06-17 16:27:15 -07002875static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002876 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002877{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002878 unsigned long curusage;
2879 unsigned long oldusage;
2880 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002881 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002882 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002883
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002884 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002885 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2886 mem_cgroup_count_children(memcg);
2887
2888 oldusage = page_counter_read(&memcg->memsw);
2889
2890 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002891 if (signal_pending(current)) {
2892 ret = -EINTR;
2893 break;
2894 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002895
2896 mutex_lock(&memcg_limit_mutex);
2897 if (limit < memcg->memory.limit) {
2898 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002899 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002900 break;
2901 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002902 if (limit > memcg->memsw.limit)
2903 enlarge = true;
2904 ret = page_counter_limit(&memcg->memsw, limit);
2905 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002906
2907 if (!ret)
2908 break;
2909
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002910 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2911
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002912 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002913 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002914 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002915 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002916 else
2917 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002918 } while (retry_count);
2919
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002920 if (!ret && enlarge)
2921 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002922
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002923 return ret;
2924}
2925
Andrew Morton0608f432013-09-24 15:27:41 -07002926unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2927 gfp_t gfp_mask,
2928 unsigned long *total_scanned)
2929{
2930 unsigned long nr_reclaimed = 0;
2931 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2932 unsigned long reclaimed;
2933 int loop = 0;
2934 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002935 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002936 unsigned long nr_scanned;
2937
2938 if (order > 0)
2939 return 0;
2940
2941 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2942 /*
2943 * This loop can run a while, specially if mem_cgroup's continuously
2944 * keep exceeding their soft limit and putting the system under
2945 * pressure
2946 */
2947 do {
2948 if (next_mz)
2949 mz = next_mz;
2950 else
2951 mz = mem_cgroup_largest_soft_limit_node(mctz);
2952 if (!mz)
2953 break;
2954
2955 nr_scanned = 0;
2956 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2957 gfp_mask, &nr_scanned);
2958 nr_reclaimed += reclaimed;
2959 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002960 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002961 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002962
2963 /*
2964 * If we failed to reclaim anything from this memory cgroup
2965 * it is time to move on to the next cgroup
2966 */
2967 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002968 if (!reclaimed)
2969 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2970
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002971 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002972 /*
2973 * One school of thought says that we should not add
2974 * back the node to the tree if reclaim returns 0.
2975 * But our reclaim could return 0, simply because due
2976 * to priority we are exposing a smaller subset of
2977 * memory to reclaim from. Consider this as a longer
2978 * term TODO.
2979 */
2980 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002981 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002982 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002983 css_put(&mz->memcg->css);
2984 loop++;
2985 /*
2986 * Could not reclaim anything and there are no more
2987 * mem cgroups to try or we seem to be looping without
2988 * reclaiming anything.
2989 */
2990 if (!nr_reclaimed &&
2991 (next_mz == NULL ||
2992 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2993 break;
2994 } while (!nr_reclaimed);
2995 if (next_mz)
2996 css_put(&next_mz->memcg->css);
2997 return nr_reclaimed;
2998}
2999
Tejun Heoea280e72014-05-16 13:22:48 -04003000/*
3001 * Test whether @memcg has children, dead or alive. Note that this
3002 * function doesn't care whether @memcg has use_hierarchy enabled and
3003 * returns %true if there are child csses according to the cgroup
3004 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
3005 */
Glauber Costab5f99b52013-02-22 16:34:53 -08003006static inline bool memcg_has_children(struct mem_cgroup *memcg)
3007{
Tejun Heoea280e72014-05-16 13:22:48 -04003008 bool ret;
3009
Johannes Weiner696ac172013-10-31 16:34:15 -07003010 /*
Tejun Heoea280e72014-05-16 13:22:48 -04003011 * The lock does not prevent addition or deletion of children, but
3012 * it prevents a new child from being initialized based on this
3013 * parent in css_online(), so it's enough to decide whether
3014 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07003015 */
Tejun Heoea280e72014-05-16 13:22:48 -04003016 lockdep_assert_held(&memcg_create_mutex);
3017
3018 rcu_read_lock();
3019 ret = css_next_child(NULL, &memcg->css);
3020 rcu_read_unlock();
3021 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08003022}
3023
3024/*
Michal Hockoc26251f2012-10-26 13:37:28 +02003025 * Reclaims as many pages from the given memcg as possible and moves
3026 * the rest to the parent.
3027 *
3028 * Caller is responsible for holding css reference for memcg.
3029 */
3030static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3031{
3032 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003033
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003034 /* we call try-to-free pages for make this cgroup empty */
3035 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003036 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003037 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003038 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003039
Michal Hockoc26251f2012-10-26 13:37:28 +02003040 if (signal_pending(current))
3041 return -EINTR;
3042
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003043 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3044 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003045 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003046 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003047 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003048 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003049 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003050
3051 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003052
3053 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003054}
3055
Tejun Heo6770c642014-05-13 12:16:21 -04003056static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3057 char *buf, size_t nbytes,
3058 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003059{
Tejun Heo6770c642014-05-13 12:16:21 -04003060 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003061
Michal Hockod8423012012-10-26 13:37:29 +02003062 if (mem_cgroup_is_root(memcg))
3063 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003064 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003065}
3066
Tejun Heo182446d2013-08-08 20:11:24 -04003067static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3068 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003069{
Tejun Heo182446d2013-08-08 20:11:24 -04003070 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003071}
3072
Tejun Heo182446d2013-08-08 20:11:24 -04003073static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3074 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003075{
3076 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04003077 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04003078 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003079
Glauber Costa09998212013-02-22 16:34:55 -08003080 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07003081
3082 if (memcg->use_hierarchy == val)
3083 goto out;
3084
Balbir Singh18f59ea2009-01-07 18:08:07 -08003085 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003086 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003087 * in the child subtrees. If it is unset, then the change can
3088 * occur, provided the current cgroup has no children.
3089 *
3090 * For the root cgroup, parent_mem is NULL, we allow value to be
3091 * set if there are no children.
3092 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003093 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003094 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04003095 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003096 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003097 else
3098 retval = -EBUSY;
3099 } else
3100 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07003101
3102out:
Glauber Costa09998212013-02-22 16:34:55 -08003103 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003104
3105 return retval;
3106}
3107
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003108static unsigned long tree_stat(struct mem_cgroup *memcg,
3109 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04003110{
3111 struct mem_cgroup *iter;
3112 long val = 0;
3113
3114 /* Per-cpu values can be negative, use a signed accumulator */
3115 for_each_mem_cgroup_tree(iter, memcg)
3116 val += mem_cgroup_read_stat(iter, idx);
3117
3118 if (val < 0) /* race ? */
3119 val = 0;
3120 return val;
3121}
3122
3123static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3124{
3125 u64 val;
3126
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003127 if (mem_cgroup_is_root(memcg)) {
3128 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
3129 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
3130 if (swap)
3131 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
3132 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04003133 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003134 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04003135 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003136 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04003137 }
Johannes Weinerce00a962014-09-05 08:43:57 -04003138 return val << PAGE_SHIFT;
3139}
3140
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003141enum {
3142 RES_USAGE,
3143 RES_LIMIT,
3144 RES_MAX_USAGE,
3145 RES_FAILCNT,
3146 RES_SOFT_LIMIT,
3147};
Johannes Weinerce00a962014-09-05 08:43:57 -04003148
Tejun Heo791badb2013-12-05 12:28:02 -05003149static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07003150 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003151{
Tejun Heo182446d2013-08-08 20:11:24 -04003152 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003153 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07003154
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003155 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003156 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003157 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003158 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003159 case _MEMSWAP:
3160 counter = &memcg->memsw;
3161 break;
3162 case _KMEM:
3163 counter = &memcg->kmem;
3164 break;
3165 default:
3166 BUG();
3167 }
3168
3169 switch (MEMFILE_ATTR(cft->private)) {
3170 case RES_USAGE:
3171 if (counter == &memcg->memory)
3172 return mem_cgroup_usage(memcg, false);
3173 if (counter == &memcg->memsw)
3174 return mem_cgroup_usage(memcg, true);
3175 return (u64)page_counter_read(counter) * PAGE_SIZE;
3176 case RES_LIMIT:
3177 return (u64)counter->limit * PAGE_SIZE;
3178 case RES_MAX_USAGE:
3179 return (u64)counter->watermark * PAGE_SIZE;
3180 case RES_FAILCNT:
3181 return counter->failcnt;
3182 case RES_SOFT_LIMIT:
3183 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003184 default:
3185 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003186 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003187}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003188
Glauber Costa510fc4e2012-12-18 14:21:47 -08003189#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08003190static int memcg_activate_kmem(struct mem_cgroup *memcg,
3191 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003192{
3193 int err = 0;
3194 int memcg_id;
3195
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003196 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003197 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003198 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003199
3200 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08003201 * For simplicity, we won't allow this to be disabled. It also can't
3202 * be changed if the cgroup has children already, or if tasks had
3203 * already joined.
3204 *
3205 * If tasks join before we set the limit, a person looking at
3206 * kmem.usage_in_bytes will have no way to determine when it took
3207 * place, which makes the value quite meaningless.
3208 *
3209 * After it first became limited, changes in the value of the limit are
3210 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08003211 */
Glauber Costa09998212013-02-22 16:34:55 -08003212 mutex_lock(&memcg_create_mutex);
Tejun Heoea280e72014-05-16 13:22:48 -04003213 if (cgroup_has_tasks(memcg->css.cgroup) ||
3214 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08003215 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08003216 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003217 if (err)
3218 goto out;
3219
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003220 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08003221 if (memcg_id < 0) {
3222 err = memcg_id;
3223 goto out;
3224 }
3225
Vladimir Davydovd6441632014-01-23 15:53:09 -08003226 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003227 * We couldn't have accounted to this cgroup, because it hasn't got
3228 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08003229 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003230 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003231 VM_BUG_ON(err);
3232
3233 static_key_slow_inc(&memcg_kmem_enabled_key);
3234 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003235 * A memory cgroup is considered kmem-active as soon as it gets
3236 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08003237 * guarantee no one starts accounting before all call sites are
3238 * patched.
3239 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003240 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003241 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003242 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003243out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08003244 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003245}
3246
Vladimir Davydovd6441632014-01-23 15:53:09 -08003247static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003248 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003249{
3250 int ret;
3251
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003252 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003253 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003254 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003255 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003256 ret = page_counter_limit(&memcg->kmem, limit);
3257 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003258 return ret;
3259}
3260
Glauber Costa55007d82012-12-18 14:22:38 -08003261static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08003262{
Glauber Costa55007d82012-12-18 14:22:38 -08003263 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003264 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003265
Glauber Costa510fc4e2012-12-18 14:21:47 -08003266 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003267 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08003268
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08003269 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08003270 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08003271 * If the parent cgroup is not kmem-active now, it cannot be activated
3272 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08003273 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08003274 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08003275 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
3276 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08003277 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003278}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003279#else
3280static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003281 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003282{
3283 return -EINVAL;
3284}
Hugh Dickins6d0439902013-02-22 16:35:50 -08003285#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08003286
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003287/*
3288 * The user of this function is...
3289 * RES_LIMIT.
3290 */
Tejun Heo451af502014-05-13 12:16:21 -04003291static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3292 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003293{
Tejun Heo451af502014-05-13 12:16:21 -04003294 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003295 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003296 int ret;
3297
Tejun Heo451af502014-05-13 12:16:21 -04003298 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003299 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003300 if (ret)
3301 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003302
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003303 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003304 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003305 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3306 ret = -EINVAL;
3307 break;
3308 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003309 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3310 case _MEM:
3311 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003312 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003313 case _MEMSWAP:
3314 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3315 break;
3316 case _KMEM:
3317 ret = memcg_update_kmem_limit(memcg, nr_pages);
3318 break;
3319 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003320 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003321 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003322 memcg->soft_limit = nr_pages;
3323 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003324 break;
3325 }
Tejun Heo451af502014-05-13 12:16:21 -04003326 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003327}
3328
Tejun Heo6770c642014-05-13 12:16:21 -04003329static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3330 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003331{
Tejun Heo6770c642014-05-13 12:16:21 -04003332 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003333 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003334
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003335 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3336 case _MEM:
3337 counter = &memcg->memory;
3338 break;
3339 case _MEMSWAP:
3340 counter = &memcg->memsw;
3341 break;
3342 case _KMEM:
3343 counter = &memcg->kmem;
3344 break;
3345 default:
3346 BUG();
3347 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003348
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003349 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003350 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003351 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003352 break;
3353 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003354 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003355 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003356 default:
3357 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003358 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003359
Tejun Heo6770c642014-05-13 12:16:21 -04003360 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003361}
3362
Tejun Heo182446d2013-08-08 20:11:24 -04003363static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003364 struct cftype *cft)
3365{
Tejun Heo182446d2013-08-08 20:11:24 -04003366 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003367}
3368
Daisuke Nishimura02491442010-03-10 15:22:17 -08003369#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003370static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003371 struct cftype *cft, u64 val)
3372{
Tejun Heo182446d2013-08-08 20:11:24 -04003373 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003374
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003375 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003376 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003377
Glauber Costaee5e8472013-02-22 16:34:50 -08003378 /*
3379 * No kind of locking is needed in here, because ->can_attach() will
3380 * check this value once in the beginning of the process, and then carry
3381 * on with stale data. This means that changes to this value will only
3382 * affect task migrations starting after the change.
3383 */
3384 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003385 return 0;
3386}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003387#else
Tejun Heo182446d2013-08-08 20:11:24 -04003388static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003389 struct cftype *cft, u64 val)
3390{
3391 return -ENOSYS;
3392}
3393#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003394
Ying Han406eb0c2011-05-26 16:25:37 -07003395#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003396static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003397{
Greg Thelen25485de2013-11-12 15:07:40 -08003398 struct numa_stat {
3399 const char *name;
3400 unsigned int lru_mask;
3401 };
3402
3403 static const struct numa_stat stats[] = {
3404 { "total", LRU_ALL },
3405 { "file", LRU_ALL_FILE },
3406 { "anon", LRU_ALL_ANON },
3407 { "unevictable", BIT(LRU_UNEVICTABLE) },
3408 };
3409 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003410 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003411 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003412 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003413
Greg Thelen25485de2013-11-12 15:07:40 -08003414 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3415 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3416 seq_printf(m, "%s=%lu", stat->name, nr);
3417 for_each_node_state(nid, N_MEMORY) {
3418 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3419 stat->lru_mask);
3420 seq_printf(m, " N%d=%lu", nid, nr);
3421 }
3422 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003423 }
Ying Han406eb0c2011-05-26 16:25:37 -07003424
Ying Han071aee12013-11-12 15:07:41 -08003425 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3426 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003427
Ying Han071aee12013-11-12 15:07:41 -08003428 nr = 0;
3429 for_each_mem_cgroup_tree(iter, memcg)
3430 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3431 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3432 for_each_node_state(nid, N_MEMORY) {
3433 nr = 0;
3434 for_each_mem_cgroup_tree(iter, memcg)
3435 nr += mem_cgroup_node_nr_lru_pages(
3436 iter, nid, stat->lru_mask);
3437 seq_printf(m, " N%d=%lu", nid, nr);
3438 }
3439 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003440 }
Ying Han406eb0c2011-05-26 16:25:37 -07003441
Ying Han406eb0c2011-05-26 16:25:37 -07003442 return 0;
3443}
3444#endif /* CONFIG_NUMA */
3445
Tejun Heo2da8ca82013-12-05 12:28:04 -05003446static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003447{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003448 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003449 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003450 struct mem_cgroup *mi;
3451 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003452
Greg Thelen0ca44b12015-02-11 15:25:58 -08003453 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3454 MEM_CGROUP_STAT_NSTATS);
3455 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3456 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003457 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3458
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003459 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003460 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003461 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003462 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
3463 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003464 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003465
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003466 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3467 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3468 mem_cgroup_read_events(memcg, i));
3469
3470 for (i = 0; i < NR_LRU_LISTS; i++)
3471 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3472 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3473
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003474 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003475 memory = memsw = PAGE_COUNTER_MAX;
3476 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3477 memory = min(memory, mi->memory.limit);
3478 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003479 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003480 seq_printf(m, "hierarchical_memory_limit %llu\n",
3481 (u64)memory * PAGE_SIZE);
3482 if (do_swap_account)
3483 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3484 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003485
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003486 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3487 long long val = 0;
3488
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003489 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003490 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003491 for_each_mem_cgroup_tree(mi, memcg)
3492 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3493 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
3494 }
3495
3496 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3497 unsigned long long val = 0;
3498
3499 for_each_mem_cgroup_tree(mi, memcg)
3500 val += mem_cgroup_read_events(mi, i);
3501 seq_printf(m, "total_%s %llu\n",
3502 mem_cgroup_events_names[i], val);
3503 }
3504
3505 for (i = 0; i < NR_LRU_LISTS; i++) {
3506 unsigned long long val = 0;
3507
3508 for_each_mem_cgroup_tree(mi, memcg)
3509 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3510 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003511 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003512
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003513#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003514 {
3515 int nid, zid;
3516 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003517 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003518 unsigned long recent_rotated[2] = {0, 0};
3519 unsigned long recent_scanned[2] = {0, 0};
3520
3521 for_each_online_node(nid)
3522 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003523 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003524 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003525
Hugh Dickins89abfab2012-05-29 15:06:53 -07003526 recent_rotated[0] += rstat->recent_rotated[0];
3527 recent_rotated[1] += rstat->recent_rotated[1];
3528 recent_scanned[0] += rstat->recent_scanned[0];
3529 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003530 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003531 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3532 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3533 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3534 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003535 }
3536#endif
3537
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003538 return 0;
3539}
3540
Tejun Heo182446d2013-08-08 20:11:24 -04003541static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3542 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003543{
Tejun Heo182446d2013-08-08 20:11:24 -04003544 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003545
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003546 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003547}
3548
Tejun Heo182446d2013-08-08 20:11:24 -04003549static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3550 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003551{
Tejun Heo182446d2013-08-08 20:11:24 -04003552 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003553
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003554 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003555 return -EINVAL;
3556
Linus Torvalds14208b02014-06-09 15:03:33 -07003557 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003558 memcg->swappiness = val;
3559 else
3560 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003561
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003562 return 0;
3563}
3564
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003565static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3566{
3567 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003568 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003569 int i;
3570
3571 rcu_read_lock();
3572 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003573 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003574 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003575 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003576
3577 if (!t)
3578 goto unlock;
3579
Johannes Weinerce00a962014-09-05 08:43:57 -04003580 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003581
3582 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003583 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003584 * If it's not true, a threshold was crossed after last
3585 * call of __mem_cgroup_threshold().
3586 */
Phil Carmody5407a562010-05-26 14:42:42 -07003587 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003588
3589 /*
3590 * Iterate backward over array of thresholds starting from
3591 * current_threshold and check if a threshold is crossed.
3592 * If none of thresholds below usage is crossed, we read
3593 * only one element of the array here.
3594 */
3595 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3596 eventfd_signal(t->entries[i].eventfd, 1);
3597
3598 /* i = current_threshold + 1 */
3599 i++;
3600
3601 /*
3602 * Iterate forward over array of thresholds starting from
3603 * current_threshold+1 and check if a threshold is crossed.
3604 * If none of thresholds above usage is crossed, we read
3605 * only one element of the array here.
3606 */
3607 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3608 eventfd_signal(t->entries[i].eventfd, 1);
3609
3610 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003611 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003612unlock:
3613 rcu_read_unlock();
3614}
3615
3616static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3617{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003618 while (memcg) {
3619 __mem_cgroup_threshold(memcg, false);
3620 if (do_swap_account)
3621 __mem_cgroup_threshold(memcg, true);
3622
3623 memcg = parent_mem_cgroup(memcg);
3624 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003625}
3626
3627static int compare_thresholds(const void *a, const void *b)
3628{
3629 const struct mem_cgroup_threshold *_a = a;
3630 const struct mem_cgroup_threshold *_b = b;
3631
Greg Thelen2bff24a2013-09-11 14:23:08 -07003632 if (_a->threshold > _b->threshold)
3633 return 1;
3634
3635 if (_a->threshold < _b->threshold)
3636 return -1;
3637
3638 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003639}
3640
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003641static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003642{
3643 struct mem_cgroup_eventfd_list *ev;
3644
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003645 spin_lock(&memcg_oom_lock);
3646
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003647 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003648 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003649
3650 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003651 return 0;
3652}
3653
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003654static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003655{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003656 struct mem_cgroup *iter;
3657
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003658 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003659 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003660}
3661
Tejun Heo59b6f872013-11-22 18:20:43 -05003662static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003663 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003664{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003665 struct mem_cgroup_thresholds *thresholds;
3666 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003667 unsigned long threshold;
3668 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003669 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003670
Johannes Weiner650c5e52015-02-11 15:26:03 -08003671 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003672 if (ret)
3673 return ret;
3674
3675 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003676
Johannes Weiner05b84302014-08-06 16:05:59 -07003677 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003678 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003679 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003680 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003681 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003682 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003683 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003684 BUG();
3685
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003686 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003687 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003688 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3689
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003690 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003691
3692 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003693 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003694 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003695 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003696 ret = -ENOMEM;
3697 goto unlock;
3698 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003699 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003700
3701 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003702 if (thresholds->primary) {
3703 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003704 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003705 }
3706
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003707 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003708 new->entries[size - 1].eventfd = eventfd;
3709 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003710
3711 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003712 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003713 compare_thresholds, NULL);
3714
3715 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003716 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003717 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003718 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003719 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003720 * new->current_threshold will not be used until
3721 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003722 * it here.
3723 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003724 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003725 } else
3726 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003727 }
3728
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003729 /* Free old spare buffer and save old primary buffer as spare */
3730 kfree(thresholds->spare);
3731 thresholds->spare = thresholds->primary;
3732
3733 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003734
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003735 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003736 synchronize_rcu();
3737
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003738unlock:
3739 mutex_unlock(&memcg->thresholds_lock);
3740
3741 return ret;
3742}
3743
Tejun Heo59b6f872013-11-22 18:20:43 -05003744static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003745 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003746{
Tejun Heo59b6f872013-11-22 18:20:43 -05003747 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003748}
3749
Tejun Heo59b6f872013-11-22 18:20:43 -05003750static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003751 struct eventfd_ctx *eventfd, const char *args)
3752{
Tejun Heo59b6f872013-11-22 18:20:43 -05003753 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003754}
3755
Tejun Heo59b6f872013-11-22 18:20:43 -05003756static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003757 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003758{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003759 struct mem_cgroup_thresholds *thresholds;
3760 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003761 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003762 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003763
3764 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003765
3766 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003767 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003768 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003769 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003770 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003771 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003772 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003773 BUG();
3774
Anton Vorontsov371528c2012-02-24 05:14:46 +04003775 if (!thresholds->primary)
3776 goto unlock;
3777
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003778 /* Check if a threshold crossed before removing */
3779 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3780
3781 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003782 size = 0;
3783 for (i = 0; i < thresholds->primary->size; i++) {
3784 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003785 size++;
3786 }
3787
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003788 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003789
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003790 /* Set thresholds array to NULL if we don't have thresholds */
3791 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003792 kfree(new);
3793 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003794 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003795 }
3796
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003797 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003798
3799 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003800 new->current_threshold = -1;
3801 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3802 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003803 continue;
3804
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003805 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003806 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003807 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003808 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003809 * until rcu_assign_pointer(), so it's safe to increment
3810 * it here.
3811 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003812 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003813 }
3814 j++;
3815 }
3816
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003817swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003818 /* Swap primary and spare array */
3819 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003820 /* If all events are unregistered, free the spare array */
3821 if (!new) {
3822 kfree(thresholds->spare);
3823 thresholds->spare = NULL;
3824 }
3825
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003826 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003827
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003828 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003829 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04003830unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003831 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003832}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003833
Tejun Heo59b6f872013-11-22 18:20:43 -05003834static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003835 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003836{
Tejun Heo59b6f872013-11-22 18:20:43 -05003837 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003838}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003839
Tejun Heo59b6f872013-11-22 18:20:43 -05003840static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003841 struct eventfd_ctx *eventfd)
3842{
Tejun Heo59b6f872013-11-22 18:20:43 -05003843 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003844}
3845
Tejun Heo59b6f872013-11-22 18:20:43 -05003846static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003847 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003848{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003849 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003850
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003851 event = kmalloc(sizeof(*event), GFP_KERNEL);
3852 if (!event)
3853 return -ENOMEM;
3854
Michal Hocko1af8efe2011-07-26 16:08:24 -07003855 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003856
3857 event->eventfd = eventfd;
3858 list_add(&event->list, &memcg->oom_notify);
3859
3860 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07003861 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003862 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003863 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003864
3865 return 0;
3866}
3867
Tejun Heo59b6f872013-11-22 18:20:43 -05003868static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003869 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003870{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003871 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003872
Michal Hocko1af8efe2011-07-26 16:08:24 -07003873 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003874
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003875 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003876 if (ev->eventfd == eventfd) {
3877 list_del(&ev->list);
3878 kfree(ev);
3879 }
3880 }
3881
Michal Hocko1af8efe2011-07-26 16:08:24 -07003882 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003883}
3884
Tejun Heo2da8ca82013-12-05 12:28:04 -05003885static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003886{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003887 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003888
Tejun Heo791badb2013-12-05 12:28:02 -05003889 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3890 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003891 return 0;
3892}
3893
Tejun Heo182446d2013-08-08 20:11:24 -04003894static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003895 struct cftype *cft, u64 val)
3896{
Tejun Heo182446d2013-08-08 20:11:24 -04003897 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003898
3899 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003900 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003901 return -EINVAL;
3902
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003903 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003904 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003905 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003906
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003907 return 0;
3908}
3909
Andrew Mortonc255a452012-07-31 16:43:02 -07003910#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003911static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003912{
Glauber Costa55007d82012-12-18 14:22:38 -08003913 int ret;
3914
Glauber Costa55007d82012-12-18 14:22:38 -08003915 ret = memcg_propagate_kmem(memcg);
3916 if (ret)
3917 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003918
Glauber Costa1d62e432012-04-09 19:36:33 -03003919 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003920}
Glauber Costae5671df2011-12-11 21:47:01 +00003921
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003922static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3923{
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003924 struct cgroup_subsys_state *css;
3925 struct mem_cgroup *parent, *child;
3926 int kmemcg_id;
3927
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003928 if (!memcg->kmem_acct_active)
3929 return;
3930
3931 /*
3932 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3933 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3934 * guarantees no cache will be created for this cgroup after we are
3935 * done (see memcg_create_kmem_cache()).
3936 */
3937 memcg->kmem_acct_active = false;
3938
3939 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003940
3941 kmemcg_id = memcg->kmemcg_id;
3942 BUG_ON(kmemcg_id < 0);
3943
3944 parent = parent_mem_cgroup(memcg);
3945 if (!parent)
3946 parent = root_mem_cgroup;
3947
3948 /*
3949 * Change kmemcg_id of this cgroup and all its descendants to the
3950 * parent's id, and then move all entries from this cgroup's list_lrus
3951 * to ones of the parent. After we have finished, all list_lrus
3952 * corresponding to this cgroup are guaranteed to remain empty. The
3953 * ordering is imposed by list_lru_node->lock taken by
3954 * memcg_drain_all_list_lrus().
3955 */
3956 css_for_each_descendant_pre(css, &memcg->css) {
3957 child = mem_cgroup_from_css(css);
3958 BUG_ON(child->kmemcg_id != kmemcg_id);
3959 child->kmemcg_id = parent->kmemcg_id;
3960 if (!memcg->use_hierarchy)
3961 break;
3962 }
3963 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3964
3965 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003966}
3967
Li Zefan10d5ebf2013-07-08 16:00:33 -07003968static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003969{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003970 if (memcg->kmem_acct_activated) {
3971 memcg_destroy_kmem_caches(memcg);
3972 static_key_slow_dec(&memcg_kmem_enabled_key);
3973 WARN_ON(page_counter_read(&memcg->kmem));
3974 }
Glauber Costa1d62e432012-04-09 19:36:33 -03003975 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003976}
Glauber Costae5671df2011-12-11 21:47:01 +00003977#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003978static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003979{
3980 return 0;
3981}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003982
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003983static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3984{
3985}
3986
Li Zefan10d5ebf2013-07-08 16:00:33 -07003987static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3988{
3989}
Glauber Costae5671df2011-12-11 21:47:01 +00003990#endif
3991
Tejun Heo6b385782015-05-22 17:13:37 -04003992#ifdef CONFIG_CGROUP_WRITEBACK
3993
3994struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3995{
3996 return &memcg->cgwb_list;
3997}
3998
3999#endif /* CONFIG_CGROUP_WRITEBACK */
4000
Tejun Heo79bd9812013-11-22 18:20:42 -05004001/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004002 * DO NOT USE IN NEW FILES.
4003 *
4004 * "cgroup.event_control" implementation.
4005 *
4006 * This is way over-engineered. It tries to support fully configurable
4007 * events for each user. Such level of flexibility is completely
4008 * unnecessary especially in the light of the planned unified hierarchy.
4009 *
4010 * Please deprecate this and replace with something simpler if at all
4011 * possible.
4012 */
4013
4014/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004015 * Unregister event and free resources.
4016 *
4017 * Gets called from workqueue.
4018 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004019static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004020{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004021 struct mem_cgroup_event *event =
4022 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004023 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004024
4025 remove_wait_queue(event->wqh, &event->wait);
4026
Tejun Heo59b6f872013-11-22 18:20:43 -05004027 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004028
4029 /* Notify userspace the event is going away. */
4030 eventfd_signal(event->eventfd, 1);
4031
4032 eventfd_ctx_put(event->eventfd);
4033 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004034 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004035}
4036
4037/*
4038 * Gets called on POLLHUP on eventfd when user closes it.
4039 *
4040 * Called with wqh->lock held and interrupts disabled.
4041 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004042static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
4043 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004044{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004045 struct mem_cgroup_event *event =
4046 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004047 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004048 unsigned long flags = (unsigned long)key;
4049
4050 if (flags & POLLHUP) {
4051 /*
4052 * If the event has been detached at cgroup removal, we
4053 * can simply return knowing the other side will cleanup
4054 * for us.
4055 *
4056 * We can't race against event freeing since the other
4057 * side will require wqh->lock via remove_wait_queue(),
4058 * which we hold.
4059 */
Tejun Heofba94802013-11-22 18:20:43 -05004060 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004061 if (!list_empty(&event->list)) {
4062 list_del_init(&event->list);
4063 /*
4064 * We are in atomic context, but cgroup_event_remove()
4065 * may sleep, so we have to call it in workqueue.
4066 */
4067 schedule_work(&event->remove);
4068 }
Tejun Heofba94802013-11-22 18:20:43 -05004069 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004070 }
4071
4072 return 0;
4073}
4074
Tejun Heo3bc942f2013-11-22 18:20:44 -05004075static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004076 wait_queue_head_t *wqh, poll_table *pt)
4077{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004078 struct mem_cgroup_event *event =
4079 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004080
4081 event->wqh = wqh;
4082 add_wait_queue(wqh, &event->wait);
4083}
4084
4085/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004086 * DO NOT USE IN NEW FILES.
4087 *
Tejun Heo79bd9812013-11-22 18:20:42 -05004088 * Parse input and register new cgroup event handler.
4089 *
4090 * Input must be in format '<event_fd> <control_fd> <args>'.
4091 * Interpretation of args is defined by control file implementation.
4092 */
Tejun Heo451af502014-05-13 12:16:21 -04004093static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4094 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05004095{
Tejun Heo451af502014-05-13 12:16:21 -04004096 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05004097 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004098 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05004099 struct cgroup_subsys_state *cfile_css;
4100 unsigned int efd, cfd;
4101 struct fd efile;
4102 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05004103 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05004104 char *endp;
4105 int ret;
4106
Tejun Heo451af502014-05-13 12:16:21 -04004107 buf = strstrip(buf);
4108
4109 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004110 if (*endp != ' ')
4111 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004112 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004113
Tejun Heo451af502014-05-13 12:16:21 -04004114 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004115 if ((*endp != ' ') && (*endp != '\0'))
4116 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004117 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004118
4119 event = kzalloc(sizeof(*event), GFP_KERNEL);
4120 if (!event)
4121 return -ENOMEM;
4122
Tejun Heo59b6f872013-11-22 18:20:43 -05004123 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004124 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004125 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4126 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4127 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05004128
4129 efile = fdget(efd);
4130 if (!efile.file) {
4131 ret = -EBADF;
4132 goto out_kfree;
4133 }
4134
4135 event->eventfd = eventfd_ctx_fileget(efile.file);
4136 if (IS_ERR(event->eventfd)) {
4137 ret = PTR_ERR(event->eventfd);
4138 goto out_put_efile;
4139 }
4140
4141 cfile = fdget(cfd);
4142 if (!cfile.file) {
4143 ret = -EBADF;
4144 goto out_put_eventfd;
4145 }
4146
4147 /* the process need read permission on control file */
4148 /* AV: shouldn't we check that it's been opened for read instead? */
4149 ret = inode_permission(file_inode(cfile.file), MAY_READ);
4150 if (ret < 0)
4151 goto out_put_cfile;
4152
Tejun Heo79bd9812013-11-22 18:20:42 -05004153 /*
Tejun Heofba94802013-11-22 18:20:43 -05004154 * Determine the event callbacks and set them in @event. This used
4155 * to be done via struct cftype but cgroup core no longer knows
4156 * about these events. The following is crude but the whole thing
4157 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05004158 *
4159 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05004160 */
Al Virob5830432014-10-31 01:22:04 -04004161 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05004162
4163 if (!strcmp(name, "memory.usage_in_bytes")) {
4164 event->register_event = mem_cgroup_usage_register_event;
4165 event->unregister_event = mem_cgroup_usage_unregister_event;
4166 } else if (!strcmp(name, "memory.oom_control")) {
4167 event->register_event = mem_cgroup_oom_register_event;
4168 event->unregister_event = mem_cgroup_oom_unregister_event;
4169 } else if (!strcmp(name, "memory.pressure_level")) {
4170 event->register_event = vmpressure_register_event;
4171 event->unregister_event = vmpressure_unregister_event;
4172 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05004173 event->register_event = memsw_cgroup_usage_register_event;
4174 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05004175 } else {
4176 ret = -EINVAL;
4177 goto out_put_cfile;
4178 }
4179
4180 /*
Tejun Heob5557c42013-11-22 18:20:42 -05004181 * Verify @cfile should belong to @css. Also, remaining events are
4182 * automatically removed on cgroup destruction but the removal is
4183 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05004184 */
Al Virob5830432014-10-31 01:22:04 -04004185 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04004186 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05004187 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05004188 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05004189 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05004190 if (cfile_css != css) {
4191 css_put(cfile_css);
4192 goto out_put_cfile;
4193 }
Tejun Heo79bd9812013-11-22 18:20:42 -05004194
Tejun Heo451af502014-05-13 12:16:21 -04004195 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05004196 if (ret)
4197 goto out_put_css;
4198
4199 efile.file->f_op->poll(efile.file, &event->pt);
4200
Tejun Heofba94802013-11-22 18:20:43 -05004201 spin_lock(&memcg->event_list_lock);
4202 list_add(&event->list, &memcg->event_list);
4203 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004204
4205 fdput(cfile);
4206 fdput(efile);
4207
Tejun Heo451af502014-05-13 12:16:21 -04004208 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004209
4210out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004211 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004212out_put_cfile:
4213 fdput(cfile);
4214out_put_eventfd:
4215 eventfd_ctx_put(event->eventfd);
4216out_put_efile:
4217 fdput(efile);
4218out_kfree:
4219 kfree(event);
4220
4221 return ret;
4222}
4223
Johannes Weiner241994ed2015-02-11 15:26:06 -08004224static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004225 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004226 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004227 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004228 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004229 },
4230 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004231 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004232 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004233 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004234 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004235 },
4236 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004237 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004238 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004239 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004240 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004241 },
4242 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004243 .name = "soft_limit_in_bytes",
4244 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004245 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004246 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004247 },
4248 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004249 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004250 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004251 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004252 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004253 },
Balbir Singh8697d332008-02-07 00:13:59 -08004254 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004255 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004256 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004257 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004258 {
4259 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004260 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004261 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004262 {
4263 .name = "use_hierarchy",
4264 .write_u64 = mem_cgroup_hierarchy_write,
4265 .read_u64 = mem_cgroup_hierarchy_read,
4266 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004267 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004268 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004269 .write = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05004270 .flags = CFTYPE_NO_PREFIX,
4271 .mode = S_IWUGO,
4272 },
4273 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004274 .name = "swappiness",
4275 .read_u64 = mem_cgroup_swappiness_read,
4276 .write_u64 = mem_cgroup_swappiness_write,
4277 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004278 {
4279 .name = "move_charge_at_immigrate",
4280 .read_u64 = mem_cgroup_move_charge_read,
4281 .write_u64 = mem_cgroup_move_charge_write,
4282 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004283 {
4284 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004285 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004286 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004287 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4288 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004289 {
4290 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004291 },
Ying Han406eb0c2011-05-26 16:25:37 -07004292#ifdef CONFIG_NUMA
4293 {
4294 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004295 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004296 },
4297#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004298#ifdef CONFIG_MEMCG_KMEM
4299 {
4300 .name = "kmem.limit_in_bytes",
4301 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004302 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004303 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004304 },
4305 {
4306 .name = "kmem.usage_in_bytes",
4307 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004308 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004309 },
4310 {
4311 .name = "kmem.failcnt",
4312 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004313 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004314 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004315 },
4316 {
4317 .name = "kmem.max_usage_in_bytes",
4318 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004319 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004320 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004321 },
Glauber Costa749c5412012-12-18 14:23:01 -08004322#ifdef CONFIG_SLABINFO
4323 {
4324 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004325 .seq_start = slab_start,
4326 .seq_next = slab_next,
4327 .seq_stop = slab_stop,
4328 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004329 },
4330#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004331#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004332 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004333};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004335static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004336{
4337 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004338 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004339 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004340 /*
4341 * This routine is called against possible nodes.
4342 * But it's BUG to call kmalloc() against offline node.
4343 *
4344 * TODO: this routine can waste much memory for nodes which will
4345 * never be onlined. It's better to use memory hotplug callback
4346 * function.
4347 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004348 if (!node_state(node, N_NORMAL_MEMORY))
4349 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004350 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004351 if (!pn)
4352 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004353
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004354 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4355 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004356 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004357 mz->usage_in_excess = 0;
4358 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004359 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004360 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004361 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004362 return 0;
4363}
4364
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004365static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004366{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004367 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004368}
4369
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004370static struct mem_cgroup *mem_cgroup_alloc(void)
4371{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004372 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004373 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004374
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004375 size = sizeof(struct mem_cgroup);
4376 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004377
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004378 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004379 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004380 return NULL;
4381
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004382 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4383 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004384 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004385 spin_lock_init(&memcg->pcp_counter_lock);
4386 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004387
4388out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004389 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004390 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004391}
4392
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004393/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004394 * At destroying mem_cgroup, references from swap_cgroup can remain.
4395 * (scanning all at force_empty is too costly...)
4396 *
4397 * Instead of clearing all references at force_empty, we remember
4398 * the number of reference from swap_cgroup and free mem_cgroup when
4399 * it goes down to 0.
4400 *
4401 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004402 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004403
4404static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004405{
Glauber Costac8b2a362012-12-18 14:22:13 -08004406 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004407
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004408 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004409
4410 for_each_node(node)
4411 free_mem_cgroup_per_zone_info(memcg, node);
4412
4413 free_percpu(memcg->stat);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004414 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004415}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004416
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004417/*
4418 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4419 */
Glauber Costae1aab162011-12-11 21:47:03 +00004420struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004421{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004422 if (!memcg->memory.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004423 return NULL;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004424 return mem_cgroup_from_counter(memcg->memory.parent, memory);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004425}
Glauber Costae1aab162011-12-11 21:47:03 +00004426EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004427
Li Zefan0eb253e2009-01-15 13:51:25 -08004428static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004429mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004430{
Glauber Costad142e3e2013-02-22 16:34:52 -08004431 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004432 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004433 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004434
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004435 memcg = mem_cgroup_alloc();
4436 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004437 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004438
Bob Liu3ed28fa2012-01-12 17:19:04 -08004439 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004440 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004441 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004442
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004443 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004444 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004445 root_mem_cgroup = memcg;
Tejun Heodbee2272015-05-22 17:13:20 -04004446 mem_cgroup_root_css = &memcg->css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004447 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004448 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004449 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004450 page_counter_init(&memcg->memsw, NULL);
4451 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004452 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004453
Glauber Costad142e3e2013-02-22 16:34:52 -08004454 memcg->last_scanned_node = MAX_NUMNODES;
4455 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004456 memcg->move_charge_at_immigrate = 0;
4457 mutex_init(&memcg->thresholds_lock);
4458 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004459 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004460 INIT_LIST_HEAD(&memcg->event_list);
4461 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004462#ifdef CONFIG_MEMCG_KMEM
4463 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004464#endif
Tejun Heo6b385782015-05-22 17:13:37 -04004465#ifdef CONFIG_CGROUP_WRITEBACK
4466 INIT_LIST_HEAD(&memcg->cgwb_list);
4467#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004468 return &memcg->css;
4469
4470free_out:
4471 __mem_cgroup_free(memcg);
4472 return ERR_PTR(error);
4473}
4474
4475static int
Tejun Heoeb954192013-08-08 20:11:23 -04004476mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004477{
Tejun Heoeb954192013-08-08 20:11:23 -04004478 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004479 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004480 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004481
Tejun Heo15a4c832014-05-04 15:09:14 -04004482 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004483 return -ENOSPC;
4484
Tejun Heo63876982013-08-08 20:11:23 -04004485 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004486 return 0;
4487
Glauber Costa09998212013-02-22 16:34:55 -08004488 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004489
4490 memcg->use_hierarchy = parent->use_hierarchy;
4491 memcg->oom_kill_disable = parent->oom_kill_disable;
4492 memcg->swappiness = mem_cgroup_swappiness(parent);
4493
4494 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004495 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004496 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004497 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004498 page_counter_init(&memcg->memsw, &parent->memsw);
4499 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004500
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004501 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004502 * No need to take a reference to the parent because cgroup
4503 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004504 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004505 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004506 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004507 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004508 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004509 page_counter_init(&memcg->memsw, NULL);
4510 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004511 /*
4512 * Deeper hierachy with use_hierarchy == false doesn't make
4513 * much sense so let cgroup subsystem know about this
4514 * unfortunate state in our controller.
4515 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004516 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004517 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004518 }
Glauber Costa09998212013-02-22 16:34:55 -08004519 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004520
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004521 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4522 if (ret)
4523 return ret;
4524
4525 /*
4526 * Make sure the memcg is initialized: mem_cgroup_iter()
4527 * orders reading memcg->initialized against its callers
4528 * reading the memcg members.
4529 */
4530 smp_store_release(&memcg->initialized, 1);
4531
4532 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004533}
4534
Tejun Heoeb954192013-08-08 20:11:23 -04004535static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004536{
Tejun Heoeb954192013-08-08 20:11:23 -04004537 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004538 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004539
4540 /*
4541 * Unregister events and notify userspace.
4542 * Notify userspace about cgroup removing only after rmdir of cgroup
4543 * directory to avoid race between userspace and kernelspace.
4544 */
Tejun Heofba94802013-11-22 18:20:43 -05004545 spin_lock(&memcg->event_list_lock);
4546 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004547 list_del_init(&event->list);
4548 schedule_work(&event->remove);
4549 }
Tejun Heofba94802013-11-22 18:20:43 -05004550 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004551
Michal Hocko33cb8762013-07-31 13:53:51 -07004552 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004553
4554 memcg_deactivate_kmem(memcg);
Tejun Heo6b385782015-05-22 17:13:37 -04004555
4556 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004557}
4558
Tejun Heoeb954192013-08-08 20:11:23 -04004559static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004560{
Tejun Heoeb954192013-08-08 20:11:23 -04004561 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004562
Li Zefan10d5ebf2013-07-08 16:00:33 -07004563 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07004564 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004565}
4566
Tejun Heo1ced9532014-07-08 18:02:57 -04004567/**
4568 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4569 * @css: the target css
4570 *
4571 * Reset the states of the mem_cgroup associated with @css. This is
4572 * invoked when the userland requests disabling on the default hierarchy
4573 * but the memcg is pinned through dependency. The memcg should stop
4574 * applying policies and should revert to the vanilla state as it may be
4575 * made visible again.
4576 *
4577 * The current implementation only resets the essential configurations.
4578 * This needs to be expanded to cover all the visible parts.
4579 */
4580static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4581{
4582 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4583
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004584 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4585 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4586 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004587 memcg->low = 0;
4588 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004589 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo1ced9532014-07-08 18:02:57 -04004590}
4591
Daisuke Nishimura02491442010-03-10 15:22:17 -08004592#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004593/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004594static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004595{
Johannes Weiner05b84302014-08-06 16:05:59 -07004596 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004597
4598 /* Try a single bulk charge without reclaim first */
Johannes Weiner00501b52014-08-08 14:19:20 -07004599 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004600 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004601 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004602 return ret;
4603 }
Johannes Weiner692e7c42014-08-06 16:05:57 -07004604 if (ret == -EINTR) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004605 cancel_charge(root_mem_cgroup, count);
Johannes Weiner692e7c42014-08-06 16:05:57 -07004606 return ret;
4607 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004608
4609 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004610 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004611 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
Johannes Weiner9476db92014-08-06 16:05:55 -07004612 /*
4613 * In case of failure, any residual charges against
4614 * mc.to will be dropped by mem_cgroup_clear_mc()
Johannes Weiner692e7c42014-08-06 16:05:57 -07004615 * later on. However, cancel any charges that are
4616 * bypassed to root right away or they'll be lost.
Johannes Weiner9476db92014-08-06 16:05:55 -07004617 */
Johannes Weiner692e7c42014-08-06 16:05:57 -07004618 if (ret == -EINTR)
Johannes Weiner00501b52014-08-08 14:19:20 -07004619 cancel_charge(root_mem_cgroup, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004620 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004621 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004622 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004623 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004624 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004625 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004626}
4627
4628/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004629 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004630 * @vma: the vma the pte to be checked belongs
4631 * @addr: the address corresponding to the pte to be checked
4632 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004633 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004634 *
4635 * Returns
4636 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4637 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4638 * move charge. if @target is not NULL, the page is stored in target->page
4639 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004640 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4641 * target for charge migration. if @target is not NULL, the entry is stored
4642 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004643 *
4644 * Called with pte lock held.
4645 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004646union mc_target {
4647 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004648 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004649};
4650
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004651enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004652 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004653 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004654 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004655};
4656
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004657static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4658 unsigned long addr, pte_t ptent)
4659{
4660 struct page *page = vm_normal_page(vma, addr, ptent);
4661
4662 if (!page || !page_mapped(page))
4663 return NULL;
4664 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004665 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004666 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004667 } else {
4668 if (!(mc.flags & MOVE_FILE))
4669 return NULL;
4670 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004671 if (!get_page_unless_zero(page))
4672 return NULL;
4673
4674 return page;
4675}
4676
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004677#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004678static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4679 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4680{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004681 struct page *page = NULL;
4682 swp_entry_t ent = pte_to_swp_entry(ptent);
4683
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004684 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004685 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004686 /*
4687 * Because lookup_swap_cache() updates some statistics counter,
4688 * we call find_get_page() with swapper_space directly.
4689 */
Shaohua Li33806f02013-02-22 16:34:37 -08004690 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004691 if (do_swap_account)
4692 entry->val = ent.val;
4693
4694 return page;
4695}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004696#else
4697static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4698 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4699{
4700 return NULL;
4701}
4702#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004703
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004704static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4705 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4706{
4707 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004708 struct address_space *mapping;
4709 pgoff_t pgoff;
4710
4711 if (!vma->vm_file) /* anonymous vma */
4712 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004713 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004714 return NULL;
4715
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004716 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004717 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004718
4719 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004720#ifdef CONFIG_SWAP
4721 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004722 if (shmem_mapping(mapping)) {
4723 page = find_get_entry(mapping, pgoff);
4724 if (radix_tree_exceptional_entry(page)) {
4725 swp_entry_t swp = radix_to_swp_entry(page);
4726 if (do_swap_account)
4727 *entry = swp;
4728 page = find_get_page(swap_address_space(swp), swp.val);
4729 }
4730 } else
4731 page = find_get_page(mapping, pgoff);
4732#else
4733 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004734#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004735 return page;
4736}
4737
Chen Gangb1b0dea2015-04-14 15:47:35 -07004738/**
4739 * mem_cgroup_move_account - move account of the page
4740 * @page: the page
4741 * @nr_pages: number of regular pages (>1 for huge pages)
4742 * @from: mem_cgroup which the page is moved from.
4743 * @to: mem_cgroup which the page is moved to. @from != @to.
4744 *
4745 * The caller must confirm following.
4746 * - page is not on LRU (isolate_page() is useful.)
4747 * - compound_lock is held when nr_pages > 1
4748 *
4749 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4750 * from old cgroup.
4751 */
4752static int mem_cgroup_move_account(struct page *page,
4753 unsigned int nr_pages,
4754 struct mem_cgroup *from,
4755 struct mem_cgroup *to)
4756{
4757 unsigned long flags;
4758 int ret;
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004759 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004760
4761 VM_BUG_ON(from == to);
4762 VM_BUG_ON_PAGE(PageLRU(page), page);
4763 /*
4764 * The page is isolated from LRU. So, collapse function
4765 * will not handle this page. But page splitting can happen.
4766 * Do this check under compound_page_lock(). The caller should
4767 * hold it.
4768 */
4769 ret = -EBUSY;
4770 if (nr_pages > 1 && !PageTransHuge(page))
4771 goto out;
4772
4773 /*
4774 * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
4775 * of its source page while we change it: page migration takes
4776 * both pages off the LRU, but page cache replacement doesn't.
4777 */
4778 if (!trylock_page(page))
4779 goto out;
4780
4781 ret = -EINVAL;
4782 if (page->mem_cgroup != from)
4783 goto out_unlock;
4784
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004785 anon = PageAnon(page);
4786
Chen Gangb1b0dea2015-04-14 15:47:35 -07004787 spin_lock_irqsave(&from->move_lock, flags);
4788
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004789 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004790 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4791 nr_pages);
4792 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4793 nr_pages);
4794 }
4795
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004796 /*
4797 * move_lock grabbed above and caller set from->moving_account, so
4798 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4799 * So mapping should be stable for dirty pages.
4800 */
4801 if (!anon && PageDirty(page)) {
4802 struct address_space *mapping = page_mapping(page);
4803
4804 if (mapping_cap_account_dirty(mapping)) {
4805 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4806 nr_pages);
4807 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4808 nr_pages);
4809 }
4810 }
4811
Chen Gangb1b0dea2015-04-14 15:47:35 -07004812 if (PageWriteback(page)) {
4813 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4814 nr_pages);
4815 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4816 nr_pages);
4817 }
4818
4819 /*
4820 * It is safe to change page->mem_cgroup here because the page
4821 * is referenced, charged, and isolated - we can't race with
4822 * uncharging, charging, migration, or LRU putback.
4823 */
4824
4825 /* caller should have done css_get */
4826 page->mem_cgroup = to;
4827 spin_unlock_irqrestore(&from->move_lock, flags);
4828
4829 ret = 0;
4830
4831 local_irq_disable();
4832 mem_cgroup_charge_statistics(to, page, nr_pages);
4833 memcg_check_events(to, page);
4834 mem_cgroup_charge_statistics(from, page, -nr_pages);
4835 memcg_check_events(from, page);
4836 local_irq_enable();
4837out_unlock:
4838 unlock_page(page);
4839out:
4840 return ret;
4841}
4842
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004843static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004844 unsigned long addr, pte_t ptent, union mc_target *target)
4845{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004846 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004847 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004848 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004849
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004850 if (pte_present(ptent))
4851 page = mc_handle_present_pte(vma, addr, ptent);
4852 else if (is_swap_pte(ptent))
4853 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004854 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004855 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004856
4857 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004858 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004859 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004860 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004861 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004862 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004863 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004864 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004865 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004866 ret = MC_TARGET_PAGE;
4867 if (target)
4868 target->page = page;
4869 }
4870 if (!ret || !target)
4871 put_page(page);
4872 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004873 /* There is a swap entry and a page doesn't exist or isn't charged */
4874 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004875 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004876 ret = MC_TARGET_SWAP;
4877 if (target)
4878 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004879 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004880 return ret;
4881}
4882
Naoya Horiguchi12724852012-03-21 16:34:28 -07004883#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4884/*
4885 * We don't consider swapping or file mapped pages because THP does not
4886 * support them for now.
4887 * Caller should make sure that pmd_trans_huge(pmd) is true.
4888 */
4889static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4890 unsigned long addr, pmd_t pmd, union mc_target *target)
4891{
4892 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004893 enum mc_target_type ret = MC_TARGET_NONE;
4894
4895 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004896 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004897 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004898 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004899 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004900 ret = MC_TARGET_PAGE;
4901 if (target) {
4902 get_page(page);
4903 target->page = page;
4904 }
4905 }
4906 return ret;
4907}
4908#else
4909static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4910 unsigned long addr, pmd_t pmd, union mc_target *target)
4911{
4912 return MC_TARGET_NONE;
4913}
4914#endif
4915
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004916static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4917 unsigned long addr, unsigned long end,
4918 struct mm_walk *walk)
4919{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004920 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004921 pte_t *pte;
4922 spinlock_t *ptl;
4923
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004924 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004925 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4926 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004927 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004928 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004929 }
Dave Hansen03319322011-03-22 16:32:56 -07004930
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004931 if (pmd_trans_unstable(pmd))
4932 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004933 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4934 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004935 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004936 mc.precharge++; /* increment precharge temporarily */
4937 pte_unmap_unlock(pte - 1, ptl);
4938 cond_resched();
4939
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004940 return 0;
4941}
4942
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004943static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4944{
4945 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004946
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004947 struct mm_walk mem_cgroup_count_precharge_walk = {
4948 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4949 .mm = mm,
4950 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004951 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004952 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004953 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004954
4955 precharge = mc.precharge;
4956 mc.precharge = 0;
4957
4958 return precharge;
4959}
4960
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004961static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4962{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004963 unsigned long precharge = mem_cgroup_count_precharge(mm);
4964
4965 VM_BUG_ON(mc.moving_task);
4966 mc.moving_task = current;
4967 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004968}
4969
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004970/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4971static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004972{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004973 struct mem_cgroup *from = mc.from;
4974 struct mem_cgroup *to = mc.to;
4975
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004976 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004977 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004978 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004979 mc.precharge = 0;
4980 }
4981 /*
4982 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4983 * we must uncharge here.
4984 */
4985 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004986 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004987 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004988 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004989 /* we must fixup refcnts and charges */
4990 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004991 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004992 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004993 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004994
Johannes Weiner05b84302014-08-06 16:05:59 -07004995 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004996 * we charged both to->memory and to->memsw, so we
4997 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004998 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004999 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005000 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005001
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005002 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005003
Li Zefan40503772013-07-08 16:00:34 -07005004 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005005 mc.moved_swap = 0;
5006 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005007 memcg_oom_recover(from);
5008 memcg_oom_recover(to);
5009 wake_up_all(&mc.waitq);
5010}
5011
5012static void mem_cgroup_clear_mc(void)
5013{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005014 /*
5015 * we must clear moving_task before waking up waiters at the end of
5016 * task migration.
5017 */
5018 mc.moving_task = NULL;
5019 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005020 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005021 mc.from = NULL;
5022 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005023 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005024}
5025
Tejun Heoeb954192013-08-08 20:11:23 -04005026static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005027 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005028{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005029 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005030 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04005031 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005032 unsigned long move_flags;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005033
Glauber Costaee5e8472013-02-22 16:34:50 -08005034 /*
5035 * We are now commited to this value whatever it is. Changes in this
5036 * tunable will only affect upcoming migrations, not the current one.
5037 * So we need to save it, and keep it going.
5038 */
Jason Low4db0c3c2015-04-15 16:14:08 -07005039 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005040 if (move_flags) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005041 struct mm_struct *mm;
5042 struct mem_cgroup *from = mem_cgroup_from_task(p);
5043
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005044 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005045
5046 mm = get_task_mm(p);
5047 if (!mm)
5048 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005049 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005050 if (mm->owner == p) {
5051 VM_BUG_ON(mc.from);
5052 VM_BUG_ON(mc.to);
5053 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005054 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005055 VM_BUG_ON(mc.moved_swap);
Johannes Weiner247b1442014-12-10 15:44:11 -08005056
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005057 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005058 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005059 mc.to = memcg;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005060 mc.flags = move_flags;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005061 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005062 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005063
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005064 ret = mem_cgroup_precharge_mc(mm);
5065 if (ret)
5066 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005067 }
5068 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005069 }
5070 return ret;
5071}
5072
Tejun Heoeb954192013-08-08 20:11:23 -04005073static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005074 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005075{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08005076 if (mc.to)
5077 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005078}
5079
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005080static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5081 unsigned long addr, unsigned long end,
5082 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005083{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005084 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005085 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005086 pte_t *pte;
5087 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005088 enum mc_target_type target_type;
5089 union mc_target target;
5090 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005091
Naoya Horiguchi12724852012-03-21 16:34:28 -07005092 /*
5093 * We don't take compound_lock() here but no race with splitting thp
5094 * happens because:
5095 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5096 * under splitting, which means there's no concurrent thp split,
5097 * - if another thread runs into split_huge_page() just after we
5098 * entered this if-block, the thread must wait for page table lock
5099 * to be unlocked in __split_huge_page_splitting(), where the main
5100 * part of thp split is not executed yet.
5101 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005102 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005103 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005104 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07005105 return 0;
5106 }
5107 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5108 if (target_type == MC_TARGET_PAGE) {
5109 page = target.page;
5110 if (!isolate_lru_page(page)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005111 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
Johannes Weiner1306a852014-12-10 15:44:52 -08005112 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005113 mc.precharge -= HPAGE_PMD_NR;
5114 mc.moved_charge += HPAGE_PMD_NR;
5115 }
5116 putback_lru_page(page);
5117 }
5118 put_page(page);
5119 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005120 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005121 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005122 }
5123
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005124 if (pmd_trans_unstable(pmd))
5125 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005126retry:
5127 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5128 for (; addr != end; addr += PAGE_SIZE) {
5129 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005130 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005131
5132 if (!mc.precharge)
5133 break;
5134
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005135 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005136 case MC_TARGET_PAGE:
5137 page = target.page;
5138 if (isolate_lru_page(page))
5139 goto put;
Johannes Weiner1306a852014-12-10 15:44:52 -08005140 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005141 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005142 /* we uncharge from mc.from later. */
5143 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005144 }
5145 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005146put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005147 put_page(page);
5148 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005149 case MC_TARGET_SWAP:
5150 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07005151 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005152 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005153 /* we fixup refcnts and charges later. */
5154 mc.moved_swap++;
5155 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005156 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005157 default:
5158 break;
5159 }
5160 }
5161 pte_unmap_unlock(pte - 1, ptl);
5162 cond_resched();
5163
5164 if (addr != end) {
5165 /*
5166 * We have consumed all precharges we got in can_attach().
5167 * We try charge one by one, but don't do any additional
5168 * charges to mc.to if we have failed in charge once in attach()
5169 * phase.
5170 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005171 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005172 if (!ret)
5173 goto retry;
5174 }
5175
5176 return ret;
5177}
5178
5179static void mem_cgroup_move_charge(struct mm_struct *mm)
5180{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005181 struct mm_walk mem_cgroup_move_charge_walk = {
5182 .pmd_entry = mem_cgroup_move_charge_pte_range,
5183 .mm = mm,
5184 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005185
5186 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08005187 /*
5188 * Signal mem_cgroup_begin_page_stat() to take the memcg's
5189 * move_lock while we're moving its pages to another memcg.
5190 * Then wait for already started RCU-only updates to finish.
5191 */
5192 atomic_inc(&mc.from->moving_account);
5193 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005194retry:
5195 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5196 /*
5197 * Someone who are holding the mmap_sem might be waiting in
5198 * waitq. So we cancel all extra charges, wake up all waiters,
5199 * and retry. Because we cancel precharges, we might not be able
5200 * to move enough charges, but moving charge is a best-effort
5201 * feature anyway, so it wouldn't be a big problem.
5202 */
5203 __mem_cgroup_clear_mc();
5204 cond_resched();
5205 goto retry;
5206 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005207 /*
5208 * When we have consumed all precharges and failed in doing
5209 * additional charge, the page walk just aborts.
5210 */
5211 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005212 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005213 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005214}
5215
Tejun Heoeb954192013-08-08 20:11:23 -04005216static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005217 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005218{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005219 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005220 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005221
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005222 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005223 if (mc.to)
5224 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005225 mmput(mm);
5226 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005227 if (mc.to)
5228 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005229}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005230#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04005231static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005232 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005233{
5234 return 0;
5235}
Tejun Heoeb954192013-08-08 20:11:23 -04005236static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005237 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005238{
5239}
Tejun Heoeb954192013-08-08 20:11:23 -04005240static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005241 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005242{
5243}
5244#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005245
Tejun Heof00baae2013-04-15 13:41:15 -07005246/*
5247 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005248 * to verify whether we're attached to the default hierarchy on each mount
5249 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005250 */
Tejun Heoeb954192013-08-08 20:11:23 -04005251static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005252{
5253 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005254 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005255 * guarantees that @root doesn't have any children, so turning it
5256 * on for the root memcg is enough.
5257 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04005258 if (cgroup_on_dfl(root_css->cgroup))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005259 root_mem_cgroup->use_hierarchy = true;
5260 else
5261 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005262}
5263
Johannes Weiner241994ed2015-02-11 15:26:06 -08005264static u64 memory_current_read(struct cgroup_subsys_state *css,
5265 struct cftype *cft)
5266{
5267 return mem_cgroup_usage(mem_cgroup_from_css(css), false);
5268}
5269
5270static int memory_low_show(struct seq_file *m, void *v)
5271{
5272 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005273 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005274
5275 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005276 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005277 else
5278 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5279
5280 return 0;
5281}
5282
5283static ssize_t memory_low_write(struct kernfs_open_file *of,
5284 char *buf, size_t nbytes, loff_t off)
5285{
5286 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5287 unsigned long low;
5288 int err;
5289
5290 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005291 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005292 if (err)
5293 return err;
5294
5295 memcg->low = low;
5296
5297 return nbytes;
5298}
5299
5300static int memory_high_show(struct seq_file *m, void *v)
5301{
5302 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005303 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005304
5305 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005306 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005307 else
5308 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5309
5310 return 0;
5311}
5312
5313static ssize_t memory_high_write(struct kernfs_open_file *of,
5314 char *buf, size_t nbytes, loff_t off)
5315{
5316 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5317 unsigned long high;
5318 int err;
5319
5320 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005321 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005322 if (err)
5323 return err;
5324
5325 memcg->high = high;
5326
5327 return nbytes;
5328}
5329
5330static int memory_max_show(struct seq_file *m, void *v)
5331{
5332 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005333 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005334
5335 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005336 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005337 else
5338 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5339
5340 return 0;
5341}
5342
5343static ssize_t memory_max_write(struct kernfs_open_file *of,
5344 char *buf, size_t nbytes, loff_t off)
5345{
5346 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5347 unsigned long max;
5348 int err;
5349
5350 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005351 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005352 if (err)
5353 return err;
5354
5355 err = mem_cgroup_resize_limit(memcg, max);
5356 if (err)
5357 return err;
5358
5359 return nbytes;
5360}
5361
5362static int memory_events_show(struct seq_file *m, void *v)
5363{
5364 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5365
5366 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5367 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5368 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5369 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5370
5371 return 0;
5372}
5373
5374static struct cftype memory_files[] = {
5375 {
5376 .name = "current",
5377 .read_u64 = memory_current_read,
5378 },
5379 {
5380 .name = "low",
5381 .flags = CFTYPE_NOT_ON_ROOT,
5382 .seq_show = memory_low_show,
5383 .write = memory_low_write,
5384 },
5385 {
5386 .name = "high",
5387 .flags = CFTYPE_NOT_ON_ROOT,
5388 .seq_show = memory_high_show,
5389 .write = memory_high_write,
5390 },
5391 {
5392 .name = "max",
5393 .flags = CFTYPE_NOT_ON_ROOT,
5394 .seq_show = memory_max_show,
5395 .write = memory_max_write,
5396 },
5397 {
5398 .name = "events",
5399 .flags = CFTYPE_NOT_ON_ROOT,
5400 .seq_show = memory_events_show,
5401 },
5402 { } /* terminate */
5403};
5404
Tejun Heo073219e2014-02-08 10:36:58 -05005405struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005406 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005407 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005408 .css_offline = mem_cgroup_css_offline,
5409 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005410 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005411 .can_attach = mem_cgroup_can_attach,
5412 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005413 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005414 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005415 .dfl_cftypes = memory_files,
5416 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005417 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005418};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005419
Johannes Weiner241994ed2015-02-11 15:26:06 -08005420/**
5421 * mem_cgroup_events - count memory events against a cgroup
5422 * @memcg: the memory cgroup
5423 * @idx: the event index
5424 * @nr: the number of events to account for
5425 */
5426void mem_cgroup_events(struct mem_cgroup *memcg,
5427 enum mem_cgroup_events_index idx,
5428 unsigned int nr)
5429{
5430 this_cpu_add(memcg->stat->events[idx], nr);
5431}
5432
5433/**
5434 * mem_cgroup_low - check if memory consumption is below the normal range
5435 * @root: the highest ancestor to consider
5436 * @memcg: the memory cgroup to check
5437 *
5438 * Returns %true if memory consumption of @memcg, and that of all
5439 * configurable ancestors up to @root, is below the normal range.
5440 */
5441bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5442{
5443 if (mem_cgroup_disabled())
5444 return false;
5445
5446 /*
5447 * The toplevel group doesn't have a configurable range, so
5448 * it's never low when looked at directly, and it is not
5449 * considered an ancestor when assessing the hierarchy.
5450 */
5451
5452 if (memcg == root_mem_cgroup)
5453 return false;
5454
Michal Hocko4e54ded2015-02-27 15:51:46 -08005455 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005456 return false;
5457
5458 while (memcg != root) {
5459 memcg = parent_mem_cgroup(memcg);
5460
5461 if (memcg == root_mem_cgroup)
5462 break;
5463
Michal Hocko4e54ded2015-02-27 15:51:46 -08005464 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005465 return false;
5466 }
5467 return true;
5468}
5469
Johannes Weiner00501b52014-08-08 14:19:20 -07005470/**
5471 * mem_cgroup_try_charge - try charging a page
5472 * @page: page to charge
5473 * @mm: mm context of the victim
5474 * @gfp_mask: reclaim mode
5475 * @memcgp: charged memcg return
5476 *
5477 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5478 * pages according to @gfp_mask if necessary.
5479 *
5480 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5481 * Otherwise, an error code is returned.
5482 *
5483 * After page->mapping has been set up, the caller must finalize the
5484 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5485 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5486 */
5487int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5488 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5489{
5490 struct mem_cgroup *memcg = NULL;
5491 unsigned int nr_pages = 1;
5492 int ret = 0;
5493
5494 if (mem_cgroup_disabled())
5495 goto out;
5496
5497 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005498 /*
5499 * Every swap fault against a single page tries to charge the
5500 * page, bail as early as possible. shmem_unuse() encounters
5501 * already charged pages, too. The USED bit is protected by
5502 * the page lock, which serializes swap cache removal, which
5503 * in turn serializes uncharging.
5504 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005505 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005506 goto out;
5507 }
5508
5509 if (PageTransHuge(page)) {
5510 nr_pages <<= compound_order(page);
5511 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5512 }
5513
5514 if (do_swap_account && PageSwapCache(page))
5515 memcg = try_get_mem_cgroup_from_page(page);
5516 if (!memcg)
5517 memcg = get_mem_cgroup_from_mm(mm);
5518
5519 ret = try_charge(memcg, gfp_mask, nr_pages);
5520
5521 css_put(&memcg->css);
5522
5523 if (ret == -EINTR) {
5524 memcg = root_mem_cgroup;
5525 ret = 0;
5526 }
5527out:
5528 *memcgp = memcg;
5529 return ret;
5530}
5531
5532/**
5533 * mem_cgroup_commit_charge - commit a page charge
5534 * @page: page to charge
5535 * @memcg: memcg to charge the page to
5536 * @lrucare: page might be on LRU already
5537 *
5538 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5539 * after page->mapping has been set up. This must happen atomically
5540 * as part of the page instantiation, i.e. under the page table lock
5541 * for anonymous pages, under the page lock for page and swap cache.
5542 *
5543 * In addition, the page must not be on the LRU during the commit, to
5544 * prevent racing with task migration. If it might be, use @lrucare.
5545 *
5546 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5547 */
5548void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5549 bool lrucare)
5550{
5551 unsigned int nr_pages = 1;
5552
5553 VM_BUG_ON_PAGE(!page->mapping, page);
5554 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5555
5556 if (mem_cgroup_disabled())
5557 return;
5558 /*
5559 * Swap faults will attempt to charge the same page multiple
5560 * times. But reuse_swap_page() might have removed the page
5561 * from swapcache already, so we can't check PageSwapCache().
5562 */
5563 if (!memcg)
5564 return;
5565
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005566 commit_charge(page, memcg, lrucare);
5567
Johannes Weiner00501b52014-08-08 14:19:20 -07005568 if (PageTransHuge(page)) {
5569 nr_pages <<= compound_order(page);
5570 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5571 }
5572
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005573 local_irq_disable();
5574 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5575 memcg_check_events(memcg, page);
5576 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005577
5578 if (do_swap_account && PageSwapCache(page)) {
5579 swp_entry_t entry = { .val = page_private(page) };
5580 /*
5581 * The swap entry might not get freed for a long time,
5582 * let's not wait for it. The page already received a
5583 * memory+swap charge, drop the swap entry duplicate.
5584 */
5585 mem_cgroup_uncharge_swap(entry);
5586 }
5587}
5588
5589/**
5590 * mem_cgroup_cancel_charge - cancel a page charge
5591 * @page: page to charge
5592 * @memcg: memcg to charge the page to
5593 *
5594 * Cancel a charge transaction started by mem_cgroup_try_charge().
5595 */
5596void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5597{
5598 unsigned int nr_pages = 1;
5599
5600 if (mem_cgroup_disabled())
5601 return;
5602 /*
5603 * Swap faults will attempt to charge the same page multiple
5604 * times. But reuse_swap_page() might have removed the page
5605 * from swapcache already, so we can't check PageSwapCache().
5606 */
5607 if (!memcg)
5608 return;
5609
5610 if (PageTransHuge(page)) {
5611 nr_pages <<= compound_order(page);
5612 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5613 }
5614
5615 cancel_charge(memcg, nr_pages);
5616}
5617
Johannes Weiner747db952014-08-08 14:19:24 -07005618static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005619 unsigned long nr_anon, unsigned long nr_file,
5620 unsigned long nr_huge, struct page *dummy_page)
5621{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005622 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005623 unsigned long flags;
5624
Johannes Weinerce00a962014-09-05 08:43:57 -04005625 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005626 page_counter_uncharge(&memcg->memory, nr_pages);
5627 if (do_swap_account)
5628 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005629 memcg_oom_recover(memcg);
5630 }
Johannes Weiner747db952014-08-08 14:19:24 -07005631
5632 local_irq_save(flags);
5633 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5634 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5635 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5636 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005637 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005638 memcg_check_events(memcg, dummy_page);
5639 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005640
5641 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005642 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005643}
5644
5645static void uncharge_list(struct list_head *page_list)
5646{
5647 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005648 unsigned long nr_anon = 0;
5649 unsigned long nr_file = 0;
5650 unsigned long nr_huge = 0;
5651 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005652 struct list_head *next;
5653 struct page *page;
5654
5655 next = page_list->next;
5656 do {
5657 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005658
5659 page = list_entry(next, struct page, lru);
5660 next = page->lru.next;
5661
5662 VM_BUG_ON_PAGE(PageLRU(page), page);
5663 VM_BUG_ON_PAGE(page_count(page), page);
5664
Johannes Weiner1306a852014-12-10 15:44:52 -08005665 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005666 continue;
5667
5668 /*
5669 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005670 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005671 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005672 */
5673
Johannes Weiner1306a852014-12-10 15:44:52 -08005674 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005675 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005676 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5677 nr_huge, page);
5678 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005679 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005680 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005681 }
5682
5683 if (PageTransHuge(page)) {
5684 nr_pages <<= compound_order(page);
5685 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5686 nr_huge += nr_pages;
5687 }
5688
5689 if (PageAnon(page))
5690 nr_anon += nr_pages;
5691 else
5692 nr_file += nr_pages;
5693
Johannes Weiner1306a852014-12-10 15:44:52 -08005694 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005695
5696 pgpgout++;
5697 } while (next != page_list);
5698
5699 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005700 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5701 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005702}
5703
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005704/**
5705 * mem_cgroup_uncharge - uncharge a page
5706 * @page: page to uncharge
5707 *
5708 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5709 * mem_cgroup_commit_charge().
5710 */
5711void mem_cgroup_uncharge(struct page *page)
5712{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005713 if (mem_cgroup_disabled())
5714 return;
5715
Johannes Weiner747db952014-08-08 14:19:24 -07005716 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005717 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005718 return;
5719
Johannes Weiner747db952014-08-08 14:19:24 -07005720 INIT_LIST_HEAD(&page->lru);
5721 uncharge_list(&page->lru);
5722}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005723
Johannes Weiner747db952014-08-08 14:19:24 -07005724/**
5725 * mem_cgroup_uncharge_list - uncharge a list of page
5726 * @page_list: list of pages to uncharge
5727 *
5728 * Uncharge a list of pages previously charged with
5729 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5730 */
5731void mem_cgroup_uncharge_list(struct list_head *page_list)
5732{
5733 if (mem_cgroup_disabled())
5734 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005735
Johannes Weiner747db952014-08-08 14:19:24 -07005736 if (!list_empty(page_list))
5737 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005738}
5739
5740/**
5741 * mem_cgroup_migrate - migrate a charge to another page
5742 * @oldpage: currently charged page
5743 * @newpage: page to transfer the charge to
Michal Hockof5e03a42015-02-05 12:25:14 -08005744 * @lrucare: either or both pages might be on the LRU already
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005745 *
5746 * Migrate the charge from @oldpage to @newpage.
5747 *
5748 * Both pages must be locked, @newpage->mapping must be set up.
5749 */
5750void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5751 bool lrucare)
5752{
Johannes Weiner29833312014-12-10 15:44:02 -08005753 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005754 int isolated;
5755
5756 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5757 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5758 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5759 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5760 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005761 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5762 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005763
5764 if (mem_cgroup_disabled())
5765 return;
5766
5767 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005768 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005769 return;
5770
Johannes Weiner7d5e3242014-12-10 15:43:46 -08005771 /*
5772 * Swapcache readahead pages can get migrated before being
5773 * charged, and migration from compaction can happen to an
5774 * uncharged page when the PFN walker finds a page that
5775 * reclaim just put back on the LRU but has not released yet.
5776 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005777 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005778 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005779 return;
5780
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005781 if (lrucare)
5782 lock_page_lru(oldpage, &isolated);
5783
Johannes Weiner1306a852014-12-10 15:44:52 -08005784 oldpage->mem_cgroup = NULL;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005785
5786 if (lrucare)
5787 unlock_page_lru(oldpage, isolated);
5788
Johannes Weiner29833312014-12-10 15:44:02 -08005789 commit_charge(newpage, memcg, lrucare);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005790}
5791
Michal Hocko2d110852013-02-22 16:34:43 -08005792/*
Michal Hocko10813122013-02-22 16:35:41 -08005793 * subsys_initcall() for memory controller.
5794 *
5795 * Some parts like hotcpu_notifier() have to be initialized from this context
5796 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5797 * everything that doesn't depend on a specific mem_cgroup structure should
5798 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005799 */
5800static int __init mem_cgroup_init(void)
5801{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005802 int cpu, node;
5803
Michal Hocko2d110852013-02-22 16:34:43 -08005804 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005805
5806 for_each_possible_cpu(cpu)
5807 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5808 drain_local_stock);
5809
5810 for_each_node(node) {
5811 struct mem_cgroup_tree_per_node *rtpn;
5812 int zone;
5813
5814 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5815 node_online(node) ? node : NUMA_NO_NODE);
5816
5817 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5818 struct mem_cgroup_tree_per_zone *rtpz;
5819
5820 rtpz = &rtpn->rb_tree_per_zone[zone];
5821 rtpz->rb_root = RB_ROOT;
5822 spin_lock_init(&rtpz->lock);
5823 }
5824 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5825 }
5826
Michal Hocko2d110852013-02-22 16:34:43 -08005827 return 0;
5828}
5829subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005830
5831#ifdef CONFIG_MEMCG_SWAP
5832/**
5833 * mem_cgroup_swapout - transfer a memsw charge to swap
5834 * @page: page whose memsw charge to transfer
5835 * @entry: swap entry to move the charge to
5836 *
5837 * Transfer the memsw charge of @page to @entry.
5838 */
5839void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5840{
5841 struct mem_cgroup *memcg;
5842 unsigned short oldid;
5843
5844 VM_BUG_ON_PAGE(PageLRU(page), page);
5845 VM_BUG_ON_PAGE(page_count(page), page);
5846
5847 if (!do_swap_account)
5848 return;
5849
5850 memcg = page->mem_cgroup;
5851
5852 /* Readahead page, never charged */
5853 if (!memcg)
5854 return;
5855
5856 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5857 VM_BUG_ON_PAGE(oldid, page);
5858 mem_cgroup_swap_statistics(memcg, true);
5859
5860 page->mem_cgroup = NULL;
5861
5862 if (!mem_cgroup_is_root(memcg))
5863 page_counter_uncharge(&memcg->memory, 1);
5864
Johannes Weinerf3717632015-06-10 11:14:54 -07005865 /* Caller disabled preemption with mapping->tree_lock */
Johannes Weiner21afa382015-02-11 15:26:36 -08005866 mem_cgroup_charge_statistics(memcg, page, -1);
5867 memcg_check_events(memcg, page);
5868}
5869
5870/**
5871 * mem_cgroup_uncharge_swap - uncharge a swap entry
5872 * @entry: swap entry to uncharge
5873 *
5874 * Drop the memsw charge associated with @entry.
5875 */
5876void mem_cgroup_uncharge_swap(swp_entry_t entry)
5877{
5878 struct mem_cgroup *memcg;
5879 unsigned short id;
5880
5881 if (!do_swap_account)
5882 return;
5883
5884 id = swap_cgroup_record(entry, 0);
5885 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005886 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005887 if (memcg) {
5888 if (!mem_cgroup_is_root(memcg))
5889 page_counter_uncharge(&memcg->memsw, 1);
5890 mem_cgroup_swap_statistics(memcg, false);
5891 css_put(&memcg->css);
5892 }
5893 rcu_read_unlock();
5894}
5895
5896/* for remember boot option*/
5897#ifdef CONFIG_MEMCG_SWAP_ENABLED
5898static int really_do_swap_account __initdata = 1;
5899#else
5900static int really_do_swap_account __initdata;
5901#endif
5902
5903static int __init enable_swap_account(char *s)
5904{
5905 if (!strcmp(s, "1"))
5906 really_do_swap_account = 1;
5907 else if (!strcmp(s, "0"))
5908 really_do_swap_account = 0;
5909 return 1;
5910}
5911__setup("swapaccount=", enable_swap_account);
5912
5913static struct cftype memsw_cgroup_files[] = {
5914 {
5915 .name = "memsw.usage_in_bytes",
5916 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5917 .read_u64 = mem_cgroup_read_u64,
5918 },
5919 {
5920 .name = "memsw.max_usage_in_bytes",
5921 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5922 .write = mem_cgroup_reset,
5923 .read_u64 = mem_cgroup_read_u64,
5924 },
5925 {
5926 .name = "memsw.limit_in_bytes",
5927 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5928 .write = mem_cgroup_write,
5929 .read_u64 = mem_cgroup_read_u64,
5930 },
5931 {
5932 .name = "memsw.failcnt",
5933 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5934 .write = mem_cgroup_reset,
5935 .read_u64 = mem_cgroup_read_u64,
5936 },
5937 { }, /* terminate */
5938};
5939
5940static int __init mem_cgroup_swap_init(void)
5941{
5942 if (!mem_cgroup_disabled() && really_do_swap_account) {
5943 do_swap_account = 1;
5944 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5945 memsw_cgroup_files));
5946 }
5947 return 0;
5948}
5949subsys_initcall(mem_cgroup_swap_init);
5950
5951#endif /* CONFIG_MEMCG_SWAP */