blob: 23d8cf5a86c30c9407def299e3088895b7c00ebf [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 Mortonbb4cc1a82013-09-24 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>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000069#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080070#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080071
Balbir Singh8697d332008-02-07 00:13:59 -080072#include <asm/uaccess.h>
73
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070074#include <trace/events/vmscan.h>
75
Tejun Heo073219e2014-02-08 10:36:58 -050076struct cgroup_subsys memory_cgrp_subsys __read_mostly;
77EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080078
Johannes Weiner7d828602016-01-14 15:20:56 -080079struct mem_cgroup *root_mem_cgroup __read_mostly;
80
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070081#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080082
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080083/* Socket memory accounting disabled? */
84static bool cgroup_memory_nosocket;
85
Johannes Weiner21afa382015-02-11 15:26:36 -080086/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070087#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080089#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070090#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#endif
92
Johannes Weiner7941d212016-01-14 15:21:23 -080093/* Whether legacy memory+swap accounting is active */
94static bool do_memsw_account(void)
95{
96 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
97}
98
Johannes Weineraf7c4b02012-05-29 15:07:08 -070099static const char * const mem_cgroup_stat_names[] = {
100 "cache",
101 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700102 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700103 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400104 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700105 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700106 "swap",
107};
108
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700109static const char * const mem_cgroup_events_names[] = {
110 "pgpgin",
111 "pgpgout",
112 "pgfault",
113 "pgmajfault",
114};
115
Sha Zhengju58cf1882013-02-22 16:32:05 -0800116static const char * const mem_cgroup_lru_names[] = {
117 "inactive_anon",
118 "active_anon",
119 "inactive_file",
120 "active_file",
121 "unevictable",
122};
123
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700124#define THRESHOLDS_EVENTS_TARGET 128
125#define SOFTLIMIT_EVENTS_TARGET 1024
126#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700127
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700128/*
129 * Cgroups above their limits are maintained in a RB-Tree, independent of
130 * their hierarchy representation
131 */
132
133struct mem_cgroup_tree_per_zone {
134 struct rb_root rb_root;
135 spinlock_t lock;
136};
137
138struct mem_cgroup_tree_per_node {
139 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
140};
141
142struct mem_cgroup_tree {
143 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
144};
145
146static struct mem_cgroup_tree soft_limit_tree __read_mostly;
147
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700148/* for OOM */
149struct mem_cgroup_eventfd_list {
150 struct list_head list;
151 struct eventfd_ctx *eventfd;
152};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800153
Tejun Heo79bd9812013-11-22 18:20:42 -0500154/*
155 * cgroup_event represents events which userspace want to receive.
156 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500157struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500158 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500161 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500163 * eventfd to signal userspace about the event.
164 */
165 struct eventfd_ctx *eventfd;
166 /*
167 * Each of these stored in a list by the cgroup.
168 */
169 struct list_head list;
170 /*
Tejun Heofba94802013-11-22 18:20:43 -0500171 * register_event() callback will be used to add new userspace
172 * waiter for changes related to this event. Use eventfd_signal()
173 * on eventfd to send notification to userspace.
174 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500175 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500176 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500177 /*
178 * unregister_event() callback will be called when userspace closes
179 * the eventfd or on cgroup removing. This callback must be set,
180 * if you want provide notification functionality.
181 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500182 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500183 struct eventfd_ctx *eventfd);
184 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500185 * All fields below needed to unregister event when
186 * userspace closes eventfd.
187 */
188 poll_table pt;
189 wait_queue_head_t *wqh;
190 wait_queue_t wait;
191 struct work_struct remove;
192};
193
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700194static void mem_cgroup_threshold(struct mem_cgroup *memcg);
195static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800196
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800197/* Stuffs for move charges at task migration. */
198/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800199 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800200 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201#define MOVE_ANON 0x1U
202#define MOVE_FILE 0x2U
203#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800204
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800205/* "mc" and its members are protected by cgroup_mutex */
206static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800207 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800208 struct mem_cgroup *from;
209 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800210 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800211 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800212 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800213 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800214 struct task_struct *moving_task; /* a task moving charges */
215 wait_queue_head_t waitq; /* a waitq for other context */
216} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700217 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800218 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
219};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800220
Balbir Singh4e416952009-09-23 15:56:39 -0700221/*
222 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
223 * limit reclaim to prevent infinite loops, if they ever occur.
224 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700225#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700226#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700227
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800228enum charge_type {
229 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700230 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800231 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700232 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700233 NR_CHARGE_TYPE,
234};
235
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800236/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800237enum res_type {
238 _MEM,
239 _MEMSWAP,
240 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800241 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800242};
243
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
245#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800246#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700247/* Used for OOM nofiier */
248#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800249
Balbir Singh75822b42009-09-23 15:56:38 -0700250/*
Glauber Costa09998212013-02-22 16:34:55 -0800251 * The memcg_create_mutex will be held whenever a new cgroup is created.
252 * As a consequence, any change that needs to protect against new child cgroups
253 * appearing has to hold it as well.
254 */
255static DEFINE_MUTEX(memcg_create_mutex);
256
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700257/* Some nice accessors for the vmpressure. */
258struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
259{
260 if (!memcg)
261 memcg = root_mem_cgroup;
262 return &memcg->vmpressure;
263}
264
265struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
266{
267 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
268}
269
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700270static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
271{
272 return (memcg == root_mem_cgroup);
273}
274
Li Zefan4219b2d2013-09-23 16:56:29 +0800275/*
276 * We restrict the id in the range of [1, 65535], so it can fit into
277 * an unsigned short.
278 */
279#define MEM_CGROUP_ID_MAX USHRT_MAX
280
Li Zefan34c00c32013-09-23 16:56:01 +0800281static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
282{
Tejun Heo15a4c832014-05-04 15:09:14 -0400283 return memcg->css.id;
Li Zefan34c00c32013-09-23 16:56:01 +0800284}
285
Vladimir Davydovadbe4272015-04-15 16:13:00 -0700286/*
287 * A helper function to get mem_cgroup from ID. must be called under
288 * rcu_read_lock(). The caller is responsible for calling
289 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
290 * refcnt from swap can be called against removed memcg.)
291 */
Li Zefan34c00c32013-09-23 16:56:01 +0800292static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
293{
294 struct cgroup_subsys_state *css;
295
Tejun Heo7d699dd2014-05-04 15:09:13 -0400296 css = css_from_id(id, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800297 return mem_cgroup_from_css(css);
298}
299
Glauber Costaa8964b92012-12-18 14:22:09 -0800300#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800301/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800302 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800303 * The main reason for not using cgroup id for this:
304 * this works better in sparse environments, where we have a lot of memcgs,
305 * but only a few kmem-limited. Or also, if we have, for instance, 200
306 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
307 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800308 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800309 * The current size of the caches array is stored in memcg_nr_cache_ids. It
310 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800311 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800312static DEFINE_IDA(memcg_cache_ida);
313int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800314
Vladimir Davydov05257a12015-02-12 14:59:01 -0800315/* Protects memcg_nr_cache_ids */
316static DECLARE_RWSEM(memcg_cache_ids_sem);
317
318void memcg_get_cache_ids(void)
319{
320 down_read(&memcg_cache_ids_sem);
321}
322
323void memcg_put_cache_ids(void)
324{
325 up_read(&memcg_cache_ids_sem);
326}
327
Glauber Costa55007d82012-12-18 14:22:38 -0800328/*
329 * MIN_SIZE is different than 1, because we would like to avoid going through
330 * the alloc/free process all the time. In a small machine, 4 kmem-limited
331 * cgroups is a reasonable guess. In the future, it could be a parameter or
332 * tunable, but that is strictly not necessary.
333 *
Li Zefanb8627832013-09-23 16:56:47 +0800334 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800335 * this constant directly from cgroup, but it is understandable that this is
336 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800337 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800338 * increase ours as well if it increases.
339 */
340#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800341#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800342
Glauber Costad7f25f82012-12-18 14:22:40 -0800343/*
344 * A lot of the calls to the cache allocation functions are expected to be
345 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
346 * conditional to this static branch, we'll have to allow modules that does
347 * kmem_cache_alloc and the such to see this symbol as well
348 */
Johannes Weineref129472016-01-14 15:21:34 -0800349DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800350EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800351
Glauber Costaa8964b92012-12-18 14:22:09 -0800352#endif /* CONFIG_MEMCG_KMEM */
353
Balbir Singhf64c3f52009-09-23 15:56:37 -0700354static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700355mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700356{
Jianyu Zhane2318752014-06-06 14:38:20 -0700357 int nid = zone_to_nid(zone);
358 int zid = zone_idx(zone);
359
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700360 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700361}
362
Tejun Heoad7fa852015-05-27 20:00:02 -0400363/**
364 * mem_cgroup_css_from_page - css of the memcg associated with a page
365 * @page: page of interest
366 *
367 * If memcg is bound to the default hierarchy, css of the memcg associated
368 * with @page is returned. The returned css remains associated with @page
369 * until it is released.
370 *
371 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
372 * is returned.
373 *
374 * XXX: The above description of behavior on the default hierarchy isn't
375 * strictly true yet as replace_page_cache_page() can modify the
376 * association before @page is released even on the default hierarchy;
377 * however, the current and planned usages don't mix the the two functions
378 * and replace_page_cache_page() will soon be updated to make the invariant
379 * actually true.
380 */
381struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
382{
383 struct mem_cgroup *memcg;
384
Tejun Heoad7fa852015-05-27 20:00:02 -0400385 memcg = page->mem_cgroup;
386
Tejun Heo9e10a132015-09-18 11:56:28 -0400387 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400388 memcg = root_mem_cgroup;
389
Tejun Heoad7fa852015-05-27 20:00:02 -0400390 return &memcg->css;
391}
392
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700393/**
394 * page_cgroup_ino - return inode number of the memcg a page is charged to
395 * @page: the page
396 *
397 * Look up the closest online ancestor of the memory cgroup @page is charged to
398 * and return its inode number or 0 if @page is not charged to any cgroup. It
399 * is safe to call this function without holding a reference to @page.
400 *
401 * Note, this function is inherently racy, because there is nothing to prevent
402 * the cgroup inode from getting torn down and potentially reallocated a moment
403 * after page_cgroup_ino() returns, so it only should be used by callers that
404 * do not care (such as procfs interfaces).
405 */
406ino_t page_cgroup_ino(struct page *page)
407{
408 struct mem_cgroup *memcg;
409 unsigned long ino = 0;
410
411 rcu_read_lock();
412 memcg = READ_ONCE(page->mem_cgroup);
413 while (memcg && !(memcg->css.flags & CSS_ONLINE))
414 memcg = parent_mem_cgroup(memcg);
415 if (memcg)
416 ino = cgroup_ino(memcg->css.cgroup);
417 rcu_read_unlock();
418 return ino;
419}
420
Balbir Singhf64c3f52009-09-23 15:56:37 -0700421static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700422mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700423{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700424 int nid = page_to_nid(page);
425 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700426
Jianyu Zhane2318752014-06-06 14:38:20 -0700427 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700428}
429
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700430static struct mem_cgroup_tree_per_zone *
431soft_limit_tree_node_zone(int nid, int zid)
432{
433 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
434}
435
436static struct mem_cgroup_tree_per_zone *
437soft_limit_tree_from_page(struct page *page)
438{
439 int nid = page_to_nid(page);
440 int zid = page_zonenum(page);
441
442 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
443}
444
Johannes Weinercf2c8122014-06-06 14:38:21 -0700445static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
446 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800447 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700448{
449 struct rb_node **p = &mctz->rb_root.rb_node;
450 struct rb_node *parent = NULL;
451 struct mem_cgroup_per_zone *mz_node;
452
453 if (mz->on_tree)
454 return;
455
456 mz->usage_in_excess = new_usage_in_excess;
457 if (!mz->usage_in_excess)
458 return;
459 while (*p) {
460 parent = *p;
461 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
462 tree_node);
463 if (mz->usage_in_excess < mz_node->usage_in_excess)
464 p = &(*p)->rb_left;
465 /*
466 * We can't avoid mem cgroups that are over their soft
467 * limit by the same amount
468 */
469 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
470 p = &(*p)->rb_right;
471 }
472 rb_link_node(&mz->tree_node, parent, p);
473 rb_insert_color(&mz->tree_node, &mctz->rb_root);
474 mz->on_tree = true;
475}
476
Johannes Weinercf2c8122014-06-06 14:38:21 -0700477static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
478 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700479{
480 if (!mz->on_tree)
481 return;
482 rb_erase(&mz->tree_node, &mctz->rb_root);
483 mz->on_tree = false;
484}
485
Johannes Weinercf2c8122014-06-06 14:38:21 -0700486static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
487 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700488{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700489 unsigned long flags;
490
491 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700492 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700493 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700494}
495
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800496static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
497{
498 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700499 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800500 unsigned long excess = 0;
501
502 if (nr_pages > soft_limit)
503 excess = nr_pages - soft_limit;
504
505 return excess;
506}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700507
508static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
509{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800510 unsigned long excess;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700511 struct mem_cgroup_per_zone *mz;
512 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700513
Jianyu Zhane2318752014-06-06 14:38:20 -0700514 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700515 /*
516 * Necessary to update all ancestors when hierarchy is used.
517 * because their event counter is not touched.
518 */
519 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700520 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800521 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700522 /*
523 * We have to update the tree if mz is on RB-tree or
524 * mem is over its softlimit.
525 */
526 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700527 unsigned long flags;
528
529 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700530 /* if on-tree, remove it */
531 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700532 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700533 /*
534 * Insert again. mz->usage_in_excess will be updated.
535 * If excess is 0, no tree ops.
536 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700537 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700538 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700539 }
540 }
541}
542
543static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
544{
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700545 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700546 struct mem_cgroup_per_zone *mz;
547 int nid, zid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700548
Jianyu Zhane2318752014-06-06 14:38:20 -0700549 for_each_node(nid) {
550 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
551 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
552 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700553 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700554 }
555 }
556}
557
558static struct mem_cgroup_per_zone *
559__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
560{
561 struct rb_node *rightmost = NULL;
562 struct mem_cgroup_per_zone *mz;
563
564retry:
565 mz = NULL;
566 rightmost = rb_last(&mctz->rb_root);
567 if (!rightmost)
568 goto done; /* Nothing to reclaim from */
569
570 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
571 /*
572 * Remove the node now but someone else can add it back,
573 * we will to add it back at the end of reclaim to its correct
574 * position in the tree.
575 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700576 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800577 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400578 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700579 goto retry;
580done:
581 return mz;
582}
583
584static struct mem_cgroup_per_zone *
585mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
586{
587 struct mem_cgroup_per_zone *mz;
588
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700589 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700590 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700591 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700592 return mz;
593}
594
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700595/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700596 * Return page count for single (non recursive) @memcg.
597 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700598 * Implementation Note: reading percpu statistics for memcg.
599 *
600 * Both of vmstat[] and percpu_counter has threshold and do periodic
601 * synchronization to implement "quick" read. There are trade-off between
602 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700603 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700604 *
605 * But this _read() function is used for user interface now. The user accounts
606 * memory usage by memory cgroup and he _always_ requires exact value because
607 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
608 * have to visit all online cpus and make sum. So, for now, unnecessary
609 * synchronization is not implemented. (just implemented for cpu hotplug)
610 *
611 * If there are kernel internal actions which can make use of some not-exact
612 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700613 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700614 * implemented.
615 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700616static unsigned long
617mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800618{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700619 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800620 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800621
Greg Thelen484ebb32015-10-01 15:37:05 -0700622 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400623 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700624 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700625 /*
626 * Summing races with updates, so val may be negative. Avoid exposing
627 * transient negative values.
628 */
629 if (val < 0)
630 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800631 return val;
632}
633
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700634static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700635 enum mem_cgroup_events_index idx)
636{
637 unsigned long val = 0;
638 int cpu;
639
Tejun Heo733a5722015-05-22 18:23:18 -0400640 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700641 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700642 return val;
643}
644
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700645static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700646 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800647 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800648{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700649 /*
650 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
651 * counted as CACHE even if it's on ANON LRU.
652 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700653 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700654 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700655 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800656 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700657 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700658 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700659
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800660 if (compound) {
661 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700662 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
663 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800664 }
David Rientjesb070e652013-05-07 16:18:09 -0700665
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800666 /* pagein of a big page is an event. So, ignore page size */
667 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700668 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800669 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700670 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800671 nr_pages = -nr_pages; /* for event */
672 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800673
Johannes Weiner13114712012-05-29 15:07:07 -0700674 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800675}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800676
Jianyu Zhane2318752014-06-06 14:38:20 -0700677static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
678 int nid,
679 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700680{
Jianyu Zhane2318752014-06-06 14:38:20 -0700681 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700682 int zid;
683
Jianyu Zhane2318752014-06-06 14:38:20 -0700684 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700685
Jianyu Zhane2318752014-06-06 14:38:20 -0700686 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
687 struct mem_cgroup_per_zone *mz;
688 enum lru_list lru;
689
690 for_each_lru(lru) {
691 if (!(BIT(lru) & lru_mask))
692 continue;
693 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
694 nr += mz->lru_size[lru];
695 }
696 }
697 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700698}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700699
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700700static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700701 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800702{
Jianyu Zhane2318752014-06-06 14:38:20 -0700703 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700704 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800705
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800706 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700707 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
708 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800709}
710
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800711static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
712 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800713{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700714 unsigned long val, next;
715
Johannes Weiner13114712012-05-29 15:07:07 -0700716 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700717 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700718 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800719 if ((long)next - (long)val < 0) {
720 switch (target) {
721 case MEM_CGROUP_TARGET_THRESH:
722 next = val + THRESHOLDS_EVENTS_TARGET;
723 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700724 case MEM_CGROUP_TARGET_SOFTLIMIT:
725 next = val + SOFTLIMIT_EVENTS_TARGET;
726 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800727 case MEM_CGROUP_TARGET_NUMAINFO:
728 next = val + NUMAINFO_EVENTS_TARGET;
729 break;
730 default:
731 break;
732 }
733 __this_cpu_write(memcg->stat->targets[target], next);
734 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700735 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800736 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800737}
738
739/*
740 * Check events in order.
741 *
742 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800744{
745 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800746 if (unlikely(mem_cgroup_event_ratelimit(memcg,
747 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700748 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800749 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800750
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700751 do_softlimit = mem_cgroup_event_ratelimit(memcg,
752 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700753#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800754 do_numainfo = mem_cgroup_event_ratelimit(memcg,
755 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700756#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800757 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700758 if (unlikely(do_softlimit))
759 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800760#if MAX_NUMNODES > 1
761 if (unlikely(do_numainfo))
762 atomic_inc(&memcg->numainfo_events);
763#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700764 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800765}
766
Balbir Singhcf475ad2008-04-29 01:00:16 -0700767struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800768{
Balbir Singh31a78f22008-09-28 23:09:31 +0100769 /*
770 * mm_update_next_owner() may clear mm->owner to NULL
771 * if it races with swapoff, page migration, etc.
772 * So this can be called with p == NULL.
773 */
774 if (unlikely(!p))
775 return NULL;
776
Tejun Heo073219e2014-02-08 10:36:58 -0500777 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800778}
Michal Hocko33398cf2015-09-08 15:01:02 -0700779EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800780
Johannes Weinerdf381972014-04-07 15:37:43 -0700781static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800782{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700783 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700784
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800785 rcu_read_lock();
786 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700787 /*
788 * Page cache insertions can happen withou an
789 * actual mm context, e.g. during disk probing
790 * on boot, loopback IO, acct() writes etc.
791 */
792 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700793 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700794 else {
795 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
796 if (unlikely(!memcg))
797 memcg = root_mem_cgroup;
798 }
Tejun Heoec903c02014-05-13 12:11:01 -0400799 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800800 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700801 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800802}
803
Johannes Weiner56600482012-01-12 17:17:59 -0800804/**
805 * mem_cgroup_iter - iterate over memory cgroup hierarchy
806 * @root: hierarchy root
807 * @prev: previously returned memcg, NULL on first invocation
808 * @reclaim: cookie for shared reclaim walks, NULL for full walks
809 *
810 * Returns references to children of the hierarchy below @root, or
811 * @root itself, or %NULL after a full round-trip.
812 *
813 * Caller must pass the return value in @prev on subsequent
814 * invocations for reference counting, or use mem_cgroup_iter_break()
815 * to cancel a hierarchy walk before the round-trip is complete.
816 *
817 * Reclaimers can specify a zone and a priority level in @reclaim to
818 * divide up the memcgs in the hierarchy among all concurrent
819 * reclaimers operating on the same zone and priority.
820 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700821struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800822 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700823 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700824{
Michal Hocko33398cf2015-09-08 15:01:02 -0700825 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800826 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800827 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800828 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700829
Andrew Morton694fbc02013-09-24 15:27:37 -0700830 if (mem_cgroup_disabled())
831 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800832
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700833 if (!root)
834 root = root_mem_cgroup;
835
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800836 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800837 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800838
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800839 if (!root->use_hierarchy && root != root_mem_cgroup) {
840 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800841 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700842 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800843 }
844
Michal Hocko542f85f2013-04-29 15:07:15 -0700845 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800846
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800847 if (reclaim) {
848 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800849
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800850 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
851 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700852
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800853 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700854 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800855
Vladimir Davydov6df38682015-12-29 14:54:10 -0800856 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700857 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800858 if (!pos || css_tryget(&pos->css))
859 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800860 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800861 * css reference reached zero, so iter->position will
862 * be cleared by ->css_released. However, we should not
863 * rely on this happening soon, because ->css_released
864 * is called from a work queue, and by busy-waiting we
865 * might block it. So we clear iter->position right
866 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800867 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800868 (void)cmpxchg(&iter->position, pos, NULL);
869 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800870 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800871
872 if (pos)
873 css = &pos->css;
874
875 for (;;) {
876 css = css_next_descendant_pre(css, &root->css);
877 if (!css) {
878 /*
879 * Reclaimers share the hierarchy walk, and a
880 * new one might jump in right at the end of
881 * the hierarchy - make sure they see at least
882 * one group and restart from the beginning.
883 */
884 if (!prev)
885 continue;
886 break;
887 }
888
889 /*
890 * Verify the css and acquire a reference. The root
891 * is provided by the caller, so we know it's alive
892 * and kicking, and don't take an extra reference.
893 */
894 memcg = mem_cgroup_from_css(css);
895
896 if (css == &root->css)
897 break;
898
Johannes Weinerb2052562014-12-10 15:42:48 -0800899 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800900 /*
901 * Make sure the memcg is initialized:
902 * mem_cgroup_css_online() orders the the
903 * initialization against setting the flag.
904 */
905 if (smp_load_acquire(&memcg->initialized))
906 break;
907
908 css_put(css);
909 }
910
911 memcg = NULL;
912 }
913
914 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800915 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800916 * The position could have already been updated by a competing
917 * thread, so check that the value hasn't changed since we read
918 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800919 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800920 (void)cmpxchg(&iter->position, pos, memcg);
921
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800922 if (pos)
923 css_put(&pos->css);
924
925 if (!memcg)
926 iter->generation++;
927 else if (!prev)
928 reclaim->generation = iter->generation;
929 }
930
Michal Hocko542f85f2013-04-29 15:07:15 -0700931out_unlock:
932 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800933out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700934 if (prev && prev != root)
935 css_put(&prev->css);
936
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800937 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700938}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800939
Johannes Weiner56600482012-01-12 17:17:59 -0800940/**
941 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
942 * @root: hierarchy root
943 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
944 */
945void mem_cgroup_iter_break(struct mem_cgroup *root,
946 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800947{
948 if (!root)
949 root = root_mem_cgroup;
950 if (prev && prev != root)
951 css_put(&prev->css);
952}
953
Vladimir Davydov6df38682015-12-29 14:54:10 -0800954static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
955{
956 struct mem_cgroup *memcg = dead_memcg;
957 struct mem_cgroup_reclaim_iter *iter;
958 struct mem_cgroup_per_zone *mz;
959 int nid, zid;
960 int i;
961
962 while ((memcg = parent_mem_cgroup(memcg))) {
963 for_each_node(nid) {
964 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
965 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
966 for (i = 0; i <= DEF_PRIORITY; i++) {
967 iter = &mz->iter[i];
968 cmpxchg(&iter->position,
969 dead_memcg, NULL);
970 }
971 }
972 }
973 }
974}
975
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700976/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800977 * Iteration constructs for visiting all cgroups (under a tree). If
978 * loops are exited prematurely (break), mem_cgroup_iter_break() must
979 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700980 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800981#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800982 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800983 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800984 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700985
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800986#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800987 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800988 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800989 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700990
Johannes Weiner925b7672012-01-12 17:18:15 -0800991/**
992 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
993 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700994 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800995 *
996 * Returns the lru list vector holding pages for the given @zone and
997 * @mem. This can be the global zone lruvec, if the memory controller
998 * is disabled.
999 */
1000struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1001 struct mem_cgroup *memcg)
1002{
1003 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001004 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001005
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001006 if (mem_cgroup_disabled()) {
1007 lruvec = &zone->lruvec;
1008 goto out;
1009 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001010
Jianyu Zhane2318752014-06-06 14:38:20 -07001011 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001012 lruvec = &mz->lruvec;
1013out:
1014 /*
1015 * Since a node can be onlined after the mem_cgroup was created,
1016 * we have to be prepared to initialize lruvec->zone here;
1017 * and if offlined then reonlined, we need to reinitialize it.
1018 */
1019 if (unlikely(lruvec->zone != zone))
1020 lruvec->zone = zone;
1021 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001022}
1023
Johannes Weiner925b7672012-01-12 17:18:15 -08001024/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001025 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001026 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001027 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001028 *
1029 * This function is only safe when following the LRU page isolation
1030 * and putback protocol: the LRU lock must be held, and the page must
1031 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001032 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001033struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001034{
1035 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001036 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001037 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001038
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001039 if (mem_cgroup_disabled()) {
1040 lruvec = &zone->lruvec;
1041 goto out;
1042 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001043
Johannes Weiner1306a852014-12-10 15:44:52 -08001044 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001045 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001046 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001047 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001048 */
Johannes Weiner29833312014-12-10 15:44:02 -08001049 if (!memcg)
1050 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001051
Jianyu Zhane2318752014-06-06 14:38:20 -07001052 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001053 lruvec = &mz->lruvec;
1054out:
1055 /*
1056 * Since a node can be onlined after the mem_cgroup was created,
1057 * we have to be prepared to initialize lruvec->zone here;
1058 * and if offlined then reonlined, we need to reinitialize it.
1059 */
1060 if (unlikely(lruvec->zone != zone))
1061 lruvec->zone = zone;
1062 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001063}
1064
1065/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001066 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1067 * @lruvec: mem_cgroup per zone lru vector
1068 * @lru: index of lru list the page is sitting on
1069 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001070 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001071 * This function must be called when a page is added to or removed from an
1072 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001073 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001074void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1075 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001076{
1077 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001078 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001079
1080 if (mem_cgroup_disabled())
1081 return;
1082
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001083 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1084 lru_size = mz->lru_size + lru;
1085 *lru_size += nr_pages;
1086 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001087}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001088
Johannes Weiner2314b422014-12-10 15:44:33 -08001089bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001090{
Johannes Weiner2314b422014-12-10 15:44:33 -08001091 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001092 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001093 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001094
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001095 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001096 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001097 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001098 task_unlock(p);
1099 } else {
1100 /*
1101 * All threads may have already detached their mm's, but the oom
1102 * killer still needs to detect if they have already been oom
1103 * killed to prevent needlessly killing additional tasks.
1104 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001105 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001106 task_memcg = mem_cgroup_from_task(task);
1107 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001108 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001109 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001110 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1111 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001112 return ret;
1113}
1114
Johannes Weiner19942822011-02-01 15:52:43 -08001115/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001116 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001117 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001118 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001119 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001120 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001121 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001122static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001123{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001124 unsigned long margin = 0;
1125 unsigned long count;
1126 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001127
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001128 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001129 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001130 if (count < limit)
1131 margin = limit - count;
1132
Johannes Weiner7941d212016-01-14 15:21:23 -08001133 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001134 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001135 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001136 if (count <= limit)
1137 margin = min(margin, limit - count);
1138 }
1139
1140 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001141}
1142
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001143/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001144 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001145 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001146 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1147 * moving cgroups. This is for waiting at high-memory pressure
1148 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001149 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001150static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001151{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001152 struct mem_cgroup *from;
1153 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001154 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001155 /*
1156 * Unlike task_move routines, we access mc.to, mc.from not under
1157 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1158 */
1159 spin_lock(&mc.lock);
1160 from = mc.from;
1161 to = mc.to;
1162 if (!from)
1163 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001164
Johannes Weiner2314b422014-12-10 15:44:33 -08001165 ret = mem_cgroup_is_descendant(from, memcg) ||
1166 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001167unlock:
1168 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001169 return ret;
1170}
1171
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001172static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001173{
1174 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001175 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001176 DEFINE_WAIT(wait);
1177 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1178 /* moving charge context might have finished. */
1179 if (mc.moving_task)
1180 schedule();
1181 finish_wait(&mc.waitq, &wait);
1182 return true;
1183 }
1184 }
1185 return false;
1186}
1187
Sha Zhengju58cf1882013-02-22 16:32:05 -08001188#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001189/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001190 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001191 * @memcg: The memory cgroup that went over limit
1192 * @p: Task that is going to be killed
1193 *
1194 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1195 * enabled
1196 */
1197void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1198{
Tejun Heoe61734c2014-02-12 09:29:50 -05001199 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001200 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001201 struct mem_cgroup *iter;
1202 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001203
Michal Hocko08088cb2014-02-25 15:01:44 -08001204 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001205 rcu_read_lock();
1206
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001207 if (p) {
1208 pr_info("Task in ");
1209 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1210 pr_cont(" killed as a result of limit of ");
1211 } else {
1212 pr_info("Memory limit reached of cgroup ");
1213 }
1214
Tejun Heoe61734c2014-02-12 09:29:50 -05001215 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001216 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001217
Balbir Singhe2224322009-04-02 16:57:39 -07001218 rcu_read_unlock();
1219
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001220 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1221 K((u64)page_counter_read(&memcg->memory)),
1222 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1223 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1224 K((u64)page_counter_read(&memcg->memsw)),
1225 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1226 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1227 K((u64)page_counter_read(&memcg->kmem)),
1228 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001229
1230 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001231 pr_info("Memory cgroup stats for ");
1232 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001233 pr_cont(":");
1234
1235 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08001236 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Sha Zhengju58cf1882013-02-22 16:32:05 -08001237 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001238 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001239 K(mem_cgroup_read_stat(iter, i)));
1240 }
1241
1242 for (i = 0; i < NR_LRU_LISTS; i++)
1243 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1244 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1245
1246 pr_cont("\n");
1247 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001248 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001249}
1250
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001251/*
1252 * This function returns the number of memcg under hierarchy tree. Returns
1253 * 1(self count) if no children.
1254 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001255static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001256{
1257 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001258 struct mem_cgroup *iter;
1259
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001260 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001261 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001262 return num;
1263}
1264
Balbir Singh6d61ef42009-01-07 18:08:06 -08001265/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001266 * Return the memory (and swap, if configured) limit for a memcg.
1267 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001268static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001269{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001270 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001271
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001272 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001273 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001274 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001275
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001276 memsw_limit = memcg->memsw.limit;
1277 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001278 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001279 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001280}
1281
David Rientjes19965462012-12-11 16:00:26 -08001282static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1283 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001284{
David Rientjes6e0fc462015-09-08 15:00:36 -07001285 struct oom_control oc = {
1286 .zonelist = NULL,
1287 .nodemask = NULL,
1288 .gfp_mask = gfp_mask,
1289 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001290 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001291 struct mem_cgroup *iter;
1292 unsigned long chosen_points = 0;
1293 unsigned long totalpages;
1294 unsigned int points = 0;
1295 struct task_struct *chosen = NULL;
1296
Johannes Weinerdc564012015-06-24 16:57:19 -07001297 mutex_lock(&oom_lock);
1298
David Rientjes876aafb2012-07-31 16:43:48 -07001299 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001300 * If current has a pending SIGKILL or is exiting, then automatically
1301 * select it. The goal is to allow it to allocate so that it may
1302 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001303 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001304 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001305 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001306 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001307 }
1308
David Rientjes6e0fc462015-09-08 15:00:36 -07001309 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001310 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001311 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001312 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001313 struct task_struct *task;
1314
Tejun Heo72ec7022013-08-08 20:11:26 -04001315 css_task_iter_start(&iter->css, &it);
1316 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001317 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001318 case OOM_SCAN_SELECT:
1319 if (chosen)
1320 put_task_struct(chosen);
1321 chosen = task;
1322 chosen_points = ULONG_MAX;
1323 get_task_struct(chosen);
1324 /* fall through */
1325 case OOM_SCAN_CONTINUE:
1326 continue;
1327 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001328 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001329 mem_cgroup_iter_break(memcg, iter);
1330 if (chosen)
1331 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001332 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001333 case OOM_SCAN_OK:
1334 break;
1335 };
1336 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001337 if (!points || points < chosen_points)
1338 continue;
1339 /* Prefer thread group leaders for display purposes */
1340 if (points == chosen_points &&
1341 thread_group_leader(chosen))
1342 continue;
1343
1344 if (chosen)
1345 put_task_struct(chosen);
1346 chosen = task;
1347 chosen_points = points;
1348 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001349 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001350 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001351 }
1352
Johannes Weinerdc564012015-06-24 16:57:19 -07001353 if (chosen) {
1354 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001355 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1356 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001357 }
1358unlock:
1359 mutex_unlock(&oom_lock);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001360}
1361
Michele Curtiae6e71d2014-12-12 16:56:35 -08001362#if MAX_NUMNODES > 1
1363
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001364/**
1365 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001366 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001367 * @nid: the node ID to be checked.
1368 * @noswap : specify true here if the user wants flle only information.
1369 *
1370 * This function returns whether the specified memcg contains any
1371 * reclaimable pages on a node. Returns true if there are any reclaimable
1372 * pages in the node.
1373 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001374static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001375 int nid, bool noswap)
1376{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001377 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001378 return true;
1379 if (noswap || !total_swap_pages)
1380 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001381 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001382 return true;
1383 return false;
1384
1385}
Ying Han889976d2011-05-26 16:25:33 -07001386
1387/*
1388 * Always updating the nodemask is not very good - even if we have an empty
1389 * list or the wrong list here, we can start from some node and traverse all
1390 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1391 *
1392 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001393static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001394{
1395 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001396 /*
1397 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1398 * pagein/pageout changes since the last update.
1399 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001400 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001401 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001402 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001403 return;
1404
Ying Han889976d2011-05-26 16:25:33 -07001405 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001406 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001407
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001408 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001409
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001410 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1411 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001412 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001413
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001414 atomic_set(&memcg->numainfo_events, 0);
1415 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001416}
1417
1418/*
1419 * Selecting a node where we start reclaim from. Because what we need is just
1420 * reducing usage counter, start from anywhere is O,K. Considering
1421 * memory reclaim from current node, there are pros. and cons.
1422 *
1423 * Freeing memory from current node means freeing memory from a node which
1424 * we'll use or we've used. So, it may make LRU bad. And if several threads
1425 * hit limits, it will see a contention on a node. But freeing from remote
1426 * node means more costs for memory reclaim because of memory latency.
1427 *
1428 * Now, we use round-robin. Better algorithm is welcomed.
1429 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001430int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001431{
1432 int node;
1433
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001434 mem_cgroup_may_update_nodemask(memcg);
1435 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001436
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001437 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001438 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001439 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001440 /*
1441 * We call this when we hit limit, not when pages are added to LRU.
1442 * No LRU may hold pages because all pages are UNEVICTABLE or
1443 * memcg is too small and all pages are not on LRU. In that case,
1444 * we use curret node.
1445 */
1446 if (unlikely(node == MAX_NUMNODES))
1447 node = numa_node_id();
1448
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001449 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001450 return node;
1451}
Ying Han889976d2011-05-26 16:25:33 -07001452#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001453int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001454{
1455 return 0;
1456}
1457#endif
1458
Andrew Morton0608f432013-09-24 15:27:41 -07001459static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1460 struct zone *zone,
1461 gfp_t gfp_mask,
1462 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001463{
Andrew Morton0608f432013-09-24 15:27:41 -07001464 struct mem_cgroup *victim = NULL;
1465 int total = 0;
1466 int loop = 0;
1467 unsigned long excess;
1468 unsigned long nr_scanned;
1469 struct mem_cgroup_reclaim_cookie reclaim = {
1470 .zone = zone,
1471 .priority = 0,
1472 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001473
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001474 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001475
Andrew Morton0608f432013-09-24 15:27:41 -07001476 while (1) {
1477 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1478 if (!victim) {
1479 loop++;
1480 if (loop >= 2) {
1481 /*
1482 * If we have not been able to reclaim
1483 * anything, it might because there are
1484 * no reclaimable pages under this hierarchy
1485 */
1486 if (!total)
1487 break;
1488 /*
1489 * We want to do more targeted reclaim.
1490 * excess >> 2 is not to excessive so as to
1491 * reclaim too much, nor too less that we keep
1492 * coming back to reclaim from this cgroup
1493 */
1494 if (total >= (excess >> 2) ||
1495 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1496 break;
1497 }
1498 continue;
1499 }
Andrew Morton0608f432013-09-24 15:27:41 -07001500 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1501 zone, &nr_scanned);
1502 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001503 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001504 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001505 }
Andrew Morton0608f432013-09-24 15:27:41 -07001506 mem_cgroup_iter_break(root_memcg, victim);
1507 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001508}
1509
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001510#ifdef CONFIG_LOCKDEP
1511static struct lockdep_map memcg_oom_lock_dep_map = {
1512 .name = "memcg_oom_lock",
1513};
1514#endif
1515
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001516static DEFINE_SPINLOCK(memcg_oom_lock);
1517
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001518/*
1519 * Check OOM-Killer is already running under our hierarchy.
1520 * If someone is running, return false.
1521 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001522static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001523{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001524 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001525
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001526 spin_lock(&memcg_oom_lock);
1527
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001528 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001529 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001530 /*
1531 * this subtree of our hierarchy is already locked
1532 * so we cannot give a lock.
1533 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001534 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001535 mem_cgroup_iter_break(memcg, iter);
1536 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001537 } else
1538 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001539 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001540
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001541 if (failed) {
1542 /*
1543 * OK, we failed to lock the whole subtree so we have
1544 * to clean up what we set up to the failing subtree
1545 */
1546 for_each_mem_cgroup_tree(iter, memcg) {
1547 if (iter == failed) {
1548 mem_cgroup_iter_break(memcg, iter);
1549 break;
1550 }
1551 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001552 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001553 } else
1554 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001555
1556 spin_unlock(&memcg_oom_lock);
1557
1558 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001559}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001560
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001561static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001562{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001563 struct mem_cgroup *iter;
1564
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001565 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001566 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001567 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001568 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001569 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001570}
1571
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001572static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001573{
1574 struct mem_cgroup *iter;
1575
Tejun Heoc2b42d32015-06-24 16:58:23 -07001576 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001577 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001578 iter->under_oom++;
1579 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001580}
1581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001582static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001583{
1584 struct mem_cgroup *iter;
1585
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001586 /*
1587 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001588 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001589 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001590 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001591 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001592 if (iter->under_oom > 0)
1593 iter->under_oom--;
1594 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001595}
1596
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001597static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1598
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001599struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001600 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001601 wait_queue_t wait;
1602};
1603
1604static int memcg_oom_wake_function(wait_queue_t *wait,
1605 unsigned mode, int sync, void *arg)
1606{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001607 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1608 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001609 struct oom_wait_info *oom_wait_info;
1610
1611 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001612 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001613
Johannes Weiner2314b422014-12-10 15:44:33 -08001614 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1615 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001616 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001617 return autoremove_wake_function(wait, mode, sync, arg);
1618}
1619
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001620static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001621{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001622 /*
1623 * For the following lockless ->under_oom test, the only required
1624 * guarantee is that it must see the state asserted by an OOM when
1625 * this function is called as a result of userland actions
1626 * triggered by the notification of the OOM. This is trivially
1627 * achieved by invoking mem_cgroup_mark_under_oom() before
1628 * triggering notification.
1629 */
1630 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001631 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001632}
1633
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001634static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001635{
Tejun Heo626ebc42015-11-05 18:46:09 -08001636 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001637 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001638 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001639 * We are in the middle of the charge context here, so we
1640 * don't want to block when potentially sitting on a callstack
1641 * that holds all kinds of filesystem and mm locks.
1642 *
1643 * Also, the caller may handle a failed allocation gracefully
1644 * (like optional page cache readahead) and so an OOM killer
1645 * invocation might not even be necessary.
1646 *
1647 * That's why we don't do anything here except remember the
1648 * OOM context and then deal with it at the end of the page
1649 * fault when the stack is unwound, the locks are released,
1650 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001651 */
Johannes Weiner49426422013-10-16 13:46:59 -07001652 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001653 current->memcg_in_oom = memcg;
1654 current->memcg_oom_gfp_mask = mask;
1655 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001656}
1657
1658/**
1659 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1660 * @handle: actually kill/wait or just clean up the OOM state
1661 *
1662 * This has to be called at the end of a page fault if the memcg OOM
1663 * handler was enabled.
1664 *
1665 * Memcg supports userspace OOM handling where failed allocations must
1666 * sleep on a waitqueue until the userspace task resolves the
1667 * situation. Sleeping directly in the charge context with all kinds
1668 * of locks held is not a good idea, instead we remember an OOM state
1669 * in the task and mem_cgroup_oom_synchronize() has to be called at
1670 * the end of the page fault to complete the OOM handling.
1671 *
1672 * Returns %true if an ongoing memcg OOM situation was detected and
1673 * completed, %false otherwise.
1674 */
1675bool mem_cgroup_oom_synchronize(bool handle)
1676{
Tejun Heo626ebc42015-11-05 18:46:09 -08001677 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001678 struct oom_wait_info owait;
1679 bool locked;
1680
1681 /* OOM is global, do not handle */
1682 if (!memcg)
1683 return false;
1684
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001685 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001686 goto cleanup;
1687
1688 owait.memcg = memcg;
1689 owait.wait.flags = 0;
1690 owait.wait.func = memcg_oom_wake_function;
1691 owait.wait.private = current;
1692 INIT_LIST_HEAD(&owait.wait.task_list);
1693
1694 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001695 mem_cgroup_mark_under_oom(memcg);
1696
1697 locked = mem_cgroup_oom_trylock(memcg);
1698
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001699 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001700 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001701
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001702 if (locked && !memcg->oom_kill_disable) {
1703 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001704 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001705 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1706 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001707 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001708 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001709 mem_cgroup_unmark_under_oom(memcg);
1710 finish_wait(&memcg_oom_waitq, &owait.wait);
1711 }
1712
1713 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001714 mem_cgroup_oom_unlock(memcg);
1715 /*
1716 * There is no guarantee that an OOM-lock contender
1717 * sees the wakeups triggered by the OOM kill
1718 * uncharges. Wake any sleepers explicitely.
1719 */
1720 memcg_oom_recover(memcg);
1721 }
Johannes Weiner49426422013-10-16 13:46:59 -07001722cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001723 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001724 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001725 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001726}
1727
Johannes Weinerd7365e72014-10-29 14:50:48 -07001728/**
1729 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1730 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001731 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001732 * This function must mark the beginning of an accounted page state
1733 * change to prevent double accounting when the page is concurrently
1734 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001735 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001736 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001737 * if (TestClearPageState(page))
1738 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001739 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001740 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001741struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001742{
1743 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001744 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001745
Johannes Weiner6de22612015-02-11 15:25:01 -08001746 /*
1747 * The RCU lock is held throughout the transaction. The fast
1748 * path can get away without acquiring the memcg->move_lock
1749 * because page moving starts with an RCU grace period.
1750 *
1751 * The RCU lock also protects the memcg from being freed when
1752 * the page state that is going to change is the only thing
1753 * preventing the page from being uncharged.
1754 * E.g. end-writeback clearing PageWriteback(), which allows
1755 * migration to go ahead and uncharge the page before the
1756 * account transaction might be complete.
1757 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001758 rcu_read_lock();
1759
1760 if (mem_cgroup_disabled())
1761 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001762again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001763 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001764 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07001765 return NULL;
1766
Qiang Huangbdcbb652014-06-04 16:08:21 -07001767 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07001768 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001769
Johannes Weiner6de22612015-02-11 15:25:01 -08001770 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001771 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001772 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001773 goto again;
1774 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001775
1776 /*
1777 * When charge migration first begins, we can have locked and
1778 * unlocked page stat updates happening concurrently. Track
1779 * the task who has the lock for mem_cgroup_end_page_stat().
1780 */
1781 memcg->move_lock_task = current;
1782 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001783
1784 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001785}
Greg Thelenc4843a72015-05-22 17:13:16 -04001786EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001787
Johannes Weinerd7365e72014-10-29 14:50:48 -07001788/**
1789 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1790 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07001791 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001792void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001793{
Johannes Weiner6de22612015-02-11 15:25:01 -08001794 if (memcg && memcg->move_lock_task == current) {
1795 unsigned long flags = memcg->move_lock_flags;
1796
1797 memcg->move_lock_task = NULL;
1798 memcg->move_lock_flags = 0;
1799
1800 spin_unlock_irqrestore(&memcg->move_lock, flags);
1801 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001802
Johannes Weinerd7365e72014-10-29 14:50:48 -07001803 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001804}
Greg Thelenc4843a72015-05-22 17:13:16 -04001805EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001806
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001807/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001808 * size of first charge trial. "32" comes from vmscan.c's magic value.
1809 * TODO: maybe necessary to use big numbers in big irons.
1810 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001811#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001812struct memcg_stock_pcp {
1813 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001814 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001816 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001817#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818};
1819static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001820static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001821
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001822/**
1823 * consume_stock: Try to consume stocked charge on this cpu.
1824 * @memcg: memcg to consume from.
1825 * @nr_pages: how many pages to charge.
1826 *
1827 * The charges will only happen if @memcg matches the current cpu's memcg
1828 * stock, and at least @nr_pages are available in that stock. Failure to
1829 * service an allocation will refill the stock.
1830 *
1831 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001832 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001833static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834{
1835 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001836 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001837
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001838 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001839 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001840
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001841 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001842 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001843 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001844 ret = true;
1845 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001846 put_cpu_var(memcg_stock);
1847 return ret;
1848}
1849
1850/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001851 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001852 */
1853static void drain_stock(struct memcg_stock_pcp *stock)
1854{
1855 struct mem_cgroup *old = stock->cached;
1856
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001857 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001858 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001859 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001860 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001861 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001862 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001863 }
1864 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001865}
1866
1867/*
1868 * This must be called under preempt disabled or must be called by
1869 * a thread which is pinned to local cpu.
1870 */
1871static void drain_local_stock(struct work_struct *dummy)
1872{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001873 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001874 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001875 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001876}
1877
1878/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001879 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001880 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001881 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001882static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001883{
1884 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1885
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001886 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001887 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001888 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001889 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001890 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001891 put_cpu_var(memcg_stock);
1892}
1893
1894/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001895 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001896 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001897 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001898static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001899{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001900 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001901
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001902 /* If someone's already draining, avoid adding running more workers. */
1903 if (!mutex_trylock(&percpu_charge_mutex))
1904 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001905 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001906 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001907 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001908 for_each_online_cpu(cpu) {
1909 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001910 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001911
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001912 memcg = stock->cached;
1913 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001914 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001915 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001916 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001917 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1918 if (cpu == curcpu)
1919 drain_local_stock(&stock->work);
1920 else
1921 schedule_work_on(cpu, &stock->work);
1922 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001923 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001924 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001925 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001926 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001927}
1928
Paul Gortmaker0db06282013-06-19 14:53:51 -04001929static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001930 unsigned long action,
1931 void *hcpu)
1932{
1933 int cpu = (unsigned long)hcpu;
1934 struct memcg_stock_pcp *stock;
1935
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001936 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001937 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001938
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001939 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001940 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001941
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001942 stock = &per_cpu(memcg_stock, cpu);
1943 drain_stock(stock);
1944 return NOTIFY_OK;
1945}
1946
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001947static void reclaim_high(struct mem_cgroup *memcg,
1948 unsigned int nr_pages,
1949 gfp_t gfp_mask)
1950{
1951 do {
1952 if (page_counter_read(&memcg->memory) <= memcg->high)
1953 continue;
1954 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1955 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1956 } while ((memcg = parent_mem_cgroup(memcg)));
1957}
1958
1959static void high_work_func(struct work_struct *work)
1960{
1961 struct mem_cgroup *memcg;
1962
1963 memcg = container_of(work, struct mem_cgroup, high_work);
1964 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1965}
1966
Tejun Heob23afb92015-11-05 18:46:11 -08001967/*
1968 * Scheduled by try_charge() to be executed from the userland return path
1969 * and reclaims memory over the high limit.
1970 */
1971void mem_cgroup_handle_over_high(void)
1972{
1973 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001974 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001975
1976 if (likely(!nr_pages))
1977 return;
1978
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001979 memcg = get_mem_cgroup_from_mm(current->mm);
1980 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001981 css_put(&memcg->css);
1982 current->memcg_nr_pages_over_high = 0;
1983}
1984
Johannes Weiner00501b52014-08-08 14:19:20 -07001985static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1986 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001987{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001988 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001989 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001990 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001991 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001992 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001993 bool may_swap = true;
1994 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001995
Johannes Weinerce00a962014-09-05 08:43:57 -04001996 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001997 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001998retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001999 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002000 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002001
Johannes Weiner7941d212016-01-14 15:21:23 -08002002 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002003 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2004 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002005 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08002006 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002007 page_counter_uncharge(&memcg->memsw, batch);
2008 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002009 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002010 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002011 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002012 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002013
Johannes Weiner6539cc02014-08-06 16:05:42 -07002014 if (batch > nr_pages) {
2015 batch = nr_pages;
2016 goto retry;
2017 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002018
Johannes Weiner06b078f2014-08-06 16:05:44 -07002019 /*
2020 * Unlike in global OOM situations, memcg is not in a physical
2021 * memory shortage. Allow dying and OOM-killed tasks to
2022 * bypass the last charges so that they can exit quickly and
2023 * free their memory.
2024 */
2025 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2026 fatal_signal_pending(current) ||
2027 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08002028 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002029
2030 if (unlikely(task_in_memcg_oom(current)))
2031 goto nomem;
2032
Mel Gormand0164ad2015-11-06 16:28:21 -08002033 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002034 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002035
Johannes Weiner241994ed2015-02-11 15:26:06 -08002036 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2037
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002038 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2039 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002040
Johannes Weiner61e02c72014-08-06 16:08:16 -07002041 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002042 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002043
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002044 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002045 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002046 drained = true;
2047 goto retry;
2048 }
2049
Johannes Weiner28c34c22014-08-06 16:05:47 -07002050 if (gfp_mask & __GFP_NORETRY)
2051 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002052 /*
2053 * Even though the limit is exceeded at this point, reclaim
2054 * may have been able to free some pages. Retry the charge
2055 * before killing the task.
2056 *
2057 * Only for regular pages, though: huge pages are rather
2058 * unlikely to succeed so close to the limit, and we fall back
2059 * to regular pages anyway in case of failure.
2060 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002061 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002062 goto retry;
2063 /*
2064 * At task move, charge accounts can be doubly counted. So, it's
2065 * better to wait until the end of task_move if something is going on.
2066 */
2067 if (mem_cgroup_wait_acct_move(mem_over_limit))
2068 goto retry;
2069
Johannes Weiner9b130612014-08-06 16:05:51 -07002070 if (nr_retries--)
2071 goto retry;
2072
Johannes Weiner06b078f2014-08-06 16:05:44 -07002073 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002074 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002075
Johannes Weiner6539cc02014-08-06 16:05:42 -07002076 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002077 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002078
Johannes Weiner241994ed2015-02-11 15:26:06 -08002079 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2080
Jerome Marchand3608de02015-11-05 18:47:29 -08002081 mem_cgroup_oom(mem_over_limit, gfp_mask,
2082 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002083nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002084 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002085 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002086force:
2087 /*
2088 * The allocation either can't fail or will lead to more memory
2089 * being freed very soon. Allow memory usage go over the limit
2090 * temporarily by force charging it.
2091 */
2092 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002093 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002094 page_counter_charge(&memcg->memsw, nr_pages);
2095 css_get_many(&memcg->css, nr_pages);
2096
2097 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002098
2099done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002100 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002101 if (batch > nr_pages)
2102 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002103
Johannes Weiner241994ed2015-02-11 15:26:06 -08002104 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002105 * If the hierarchy is above the normal consumption range, schedule
2106 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002107 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002108 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2109 * not recorded as it most likely matches current's and won't
2110 * change in the meantime. As high limit is checked again before
2111 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002112 */
2113 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002114 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002115 /* Don't bother a random interrupted task */
2116 if (in_interrupt()) {
2117 schedule_work(&memcg->high_work);
2118 break;
2119 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002120 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002121 set_notify_resume(current);
2122 break;
2123 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002124 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002125
2126 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002127}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002128
Johannes Weiner00501b52014-08-08 14:19:20 -07002129static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002130{
Johannes Weinerce00a962014-09-05 08:43:57 -04002131 if (mem_cgroup_is_root(memcg))
2132 return;
2133
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002134 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002135 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002136 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002137
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002138 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002139}
2140
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002141static void lock_page_lru(struct page *page, int *isolated)
2142{
2143 struct zone *zone = page_zone(page);
2144
2145 spin_lock_irq(&zone->lru_lock);
2146 if (PageLRU(page)) {
2147 struct lruvec *lruvec;
2148
2149 lruvec = mem_cgroup_page_lruvec(page, zone);
2150 ClearPageLRU(page);
2151 del_page_from_lru_list(page, lruvec, page_lru(page));
2152 *isolated = 1;
2153 } else
2154 *isolated = 0;
2155}
2156
2157static void unlock_page_lru(struct page *page, int isolated)
2158{
2159 struct zone *zone = page_zone(page);
2160
2161 if (isolated) {
2162 struct lruvec *lruvec;
2163
2164 lruvec = mem_cgroup_page_lruvec(page, zone);
2165 VM_BUG_ON_PAGE(PageLRU(page), page);
2166 SetPageLRU(page);
2167 add_page_to_lru_list(page, lruvec, page_lru(page));
2168 }
2169 spin_unlock_irq(&zone->lru_lock);
2170}
2171
Johannes Weiner00501b52014-08-08 14:19:20 -07002172static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002173 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002174{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002175 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002176
Johannes Weiner1306a852014-12-10 15:44:52 -08002177 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002178
2179 /*
2180 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2181 * may already be on some other mem_cgroup's LRU. Take care of it.
2182 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002183 if (lrucare)
2184 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002185
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002186 /*
2187 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002188 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002189 *
2190 * - the page is uncharged
2191 *
2192 * - the page is off-LRU
2193 *
2194 * - an anonymous fault has exclusive page access, except for
2195 * a locked page table
2196 *
2197 * - a page cache insertion, a swapin fault, or a migration
2198 * have the page locked
2199 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002200 page->mem_cgroup = memcg;
Hugh Dickins3be912772008-02-07 00:14:19 -08002201
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002202 if (lrucare)
2203 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002204}
2205
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002206#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002207static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002208{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002209 int id, size;
2210 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002211
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002212 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002213 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2214 if (id < 0)
2215 return id;
2216
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002217 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002218 return id;
2219
2220 /*
2221 * There's no space for the new id in memcg_caches arrays,
2222 * so we have to grow them.
2223 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002224 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002225
2226 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002227 if (size < MEMCG_CACHES_MIN_SIZE)
2228 size = MEMCG_CACHES_MIN_SIZE;
2229 else if (size > MEMCG_CACHES_MAX_SIZE)
2230 size = MEMCG_CACHES_MAX_SIZE;
2231
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002232 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002233 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002234 err = memcg_update_all_list_lrus(size);
2235 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002236 memcg_nr_cache_ids = size;
2237
2238 up_write(&memcg_cache_ids_sem);
2239
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002240 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002241 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002242 return err;
2243 }
2244 return id;
2245}
2246
2247static void memcg_free_cache_id(int id)
2248{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002249 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002250}
2251
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002252struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002253 struct mem_cgroup *memcg;
2254 struct kmem_cache *cachep;
2255 struct work_struct work;
2256};
2257
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002258static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002259{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002260 struct memcg_kmem_cache_create_work *cw =
2261 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002262 struct mem_cgroup *memcg = cw->memcg;
2263 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002264
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002265 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002266
Vladimir Davydov5722d092014-04-07 15:39:24 -07002267 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002268 kfree(cw);
2269}
2270
2271/*
2272 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002273 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002274static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2275 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002276{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002277 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002278
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002279 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002280 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002281 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002282
2283 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002284
2285 cw->memcg = memcg;
2286 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002287 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002288
Glauber Costad7f25f82012-12-18 14:22:40 -08002289 schedule_work(&cw->work);
2290}
2291
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002292static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2293 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002294{
2295 /*
2296 * We need to stop accounting when we kmalloc, because if the
2297 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002298 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002299 *
2300 * However, it is better to enclose the whole function. Depending on
2301 * the debugging options enabled, INIT_WORK(), for instance, can
2302 * trigger an allocation. This too, will make us recurse. Because at
2303 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2304 * the safest choice is to do it like this, wrapping the whole function.
2305 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002306 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002307 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002308 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002309}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002310
Glauber Costad7f25f82012-12-18 14:22:40 -08002311/*
2312 * Return the kmem_cache we're supposed to use for a slab allocation.
2313 * We try to use the current memcg's version of the cache.
2314 *
2315 * If the cache does not exist yet, if we are the first user of it,
2316 * we either create it immediately, if possible, or create it asynchronously
2317 * in a workqueue.
2318 * In the latter case, we will let the current allocation go through with
2319 * the original cache.
2320 *
2321 * Can't be called in interrupt context or from kernel threads.
2322 * This function needs to be called with rcu_read_lock() held.
2323 */
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002324struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
Glauber Costad7f25f82012-12-18 14:22:40 -08002325{
2326 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002327 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002328 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002329
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002330 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002331
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002332 if (cachep->flags & SLAB_ACCOUNT)
2333 gfp |= __GFP_ACCOUNT;
2334
2335 if (!(gfp & __GFP_ACCOUNT))
2336 return cachep;
2337
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002338 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002339 return cachep;
2340
Vladimir Davydov8135be52014-12-12 16:56:38 -08002341 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002342 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002343 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002344 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002345
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002346 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002347 if (likely(memcg_cachep))
2348 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002349
2350 /*
2351 * If we are in a safe context (can wait, and not in interrupt
2352 * context), we could be be predictable and return right away.
2353 * This would guarantee that the allocation being performed
2354 * already belongs in the new cache.
2355 *
2356 * However, there are some clashes that can arrive from locking.
2357 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002358 * memcg_create_kmem_cache, this means no further allocation
2359 * could happen with the slab_mutex held. So it's better to
2360 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002361 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002362 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002363out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002364 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002365 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002366}
Glauber Costad7f25f82012-12-18 14:22:40 -08002367
Vladimir Davydov8135be52014-12-12 16:56:38 -08002368void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2369{
2370 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002371 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002372}
2373
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002374int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2375 struct mem_cgroup *memcg)
2376{
2377 unsigned int nr_pages = 1 << order;
2378 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002379 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002380
2381 if (!memcg_kmem_is_active(memcg))
2382 return 0;
2383
Johannes Weiner6071ca52015-11-05 18:50:26 -08002384 if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter))
2385 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002386
2387 ret = try_charge(memcg, gfp, nr_pages);
2388 if (ret) {
2389 page_counter_uncharge(&memcg->kmem, nr_pages);
2390 return ret;
2391 }
2392
2393 page->mem_cgroup = memcg;
2394
2395 return 0;
2396}
2397
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002398int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002399{
2400 struct mem_cgroup *memcg;
2401 int ret;
2402
Johannes Weinerdf381972014-04-07 15:37:43 -07002403 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002404 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002405 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002406 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002407}
2408
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002409void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002410{
Johannes Weiner1306a852014-12-10 15:44:52 -08002411 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002412 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002413
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002414 if (!memcg)
2415 return;
2416
Sasha Levin309381fea2014-01-23 15:52:54 -08002417 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002418
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002419 page_counter_uncharge(&memcg->kmem, nr_pages);
2420 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002421 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002422 page_counter_uncharge(&memcg->memsw, nr_pages);
2423
Johannes Weiner1306a852014-12-10 15:44:52 -08002424 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002425 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002426}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002427#endif /* CONFIG_MEMCG_KMEM */
2428
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002429#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2430
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002431/*
2432 * Because tail pages are not marked as "used", set it. We're under
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08002433 * zone->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002434 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002435void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002436{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002437 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002438
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002439 if (mem_cgroup_disabled())
2440 return;
David Rientjesb070e652013-05-07 16:18:09 -07002441
Johannes Weiner29833312014-12-10 15:44:02 -08002442 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002443 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002444
Johannes Weiner1306a852014-12-10 15:44:52 -08002445 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002446 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002447}
Hugh Dickins12d27102012-01-12 17:19:52 -08002448#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002449
Andrew Mortonc255a452012-07-31 16:43:02 -07002450#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002451static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2452 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002453{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002454 int val = (charge) ? 1 : -1;
2455 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002456}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002457
2458/**
2459 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2460 * @entry: swap entry to be moved
2461 * @from: mem_cgroup which the entry is moved from
2462 * @to: mem_cgroup which the entry is moved to
2463 *
2464 * It succeeds only when the swap_cgroup's record for this entry is the same
2465 * as the mem_cgroup's id of @from.
2466 *
2467 * Returns 0 on success, -EINVAL on failure.
2468 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002469 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002470 * both res and memsw, and called css_get().
2471 */
2472static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002473 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002474{
2475 unsigned short old_id, new_id;
2476
Li Zefan34c00c32013-09-23 16:56:01 +08002477 old_id = mem_cgroup_id(from);
2478 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002479
2480 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002481 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002482 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002483 return 0;
2484 }
2485 return -EINVAL;
2486}
2487#else
2488static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002489 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002490{
2491 return -EINVAL;
2492}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002493#endif
2494
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002495static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002496
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002497static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002498 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002499{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002500 unsigned long curusage;
2501 unsigned long oldusage;
2502 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002503 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002504 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002505
2506 /*
2507 * For keeping hierarchical_reclaim simple, how long we should retry
2508 * is depends on callers. We set our retry-count to be function
2509 * of # of children which we should visit in this loop.
2510 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002511 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2512 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002513
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002514 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002515
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002516 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002517 if (signal_pending(current)) {
2518 ret = -EINTR;
2519 break;
2520 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002521
2522 mutex_lock(&memcg_limit_mutex);
2523 if (limit > memcg->memsw.limit) {
2524 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002525 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002526 break;
2527 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002528 if (limit > memcg->memory.limit)
2529 enlarge = true;
2530 ret = page_counter_limit(&memcg->memory, limit);
2531 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002532
2533 if (!ret)
2534 break;
2535
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002536 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2537
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002538 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002539 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002540 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002541 retry_count--;
2542 else
2543 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002544 } while (retry_count);
2545
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002546 if (!ret && enlarge)
2547 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002548
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002549 return ret;
2550}
2551
Li Zefan338c8432009-06-17 16:27:15 -07002552static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002553 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002554{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002555 unsigned long curusage;
2556 unsigned long oldusage;
2557 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002558 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002559 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002560
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002561 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002562 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2563 mem_cgroup_count_children(memcg);
2564
2565 oldusage = page_counter_read(&memcg->memsw);
2566
2567 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002568 if (signal_pending(current)) {
2569 ret = -EINTR;
2570 break;
2571 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002572
2573 mutex_lock(&memcg_limit_mutex);
2574 if (limit < memcg->memory.limit) {
2575 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002576 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002577 break;
2578 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002579 if (limit > memcg->memsw.limit)
2580 enlarge = true;
2581 ret = page_counter_limit(&memcg->memsw, limit);
2582 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002583
2584 if (!ret)
2585 break;
2586
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002587 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2588
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002589 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002590 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002591 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002592 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002593 else
2594 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002595 } while (retry_count);
2596
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002597 if (!ret && enlarge)
2598 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002599
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002600 return ret;
2601}
2602
Andrew Morton0608f432013-09-24 15:27:41 -07002603unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2604 gfp_t gfp_mask,
2605 unsigned long *total_scanned)
2606{
2607 unsigned long nr_reclaimed = 0;
2608 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2609 unsigned long reclaimed;
2610 int loop = 0;
2611 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002612 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002613 unsigned long nr_scanned;
2614
2615 if (order > 0)
2616 return 0;
2617
2618 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2619 /*
2620 * This loop can run a while, specially if mem_cgroup's continuously
2621 * keep exceeding their soft limit and putting the system under
2622 * pressure
2623 */
2624 do {
2625 if (next_mz)
2626 mz = next_mz;
2627 else
2628 mz = mem_cgroup_largest_soft_limit_node(mctz);
2629 if (!mz)
2630 break;
2631
2632 nr_scanned = 0;
2633 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2634 gfp_mask, &nr_scanned);
2635 nr_reclaimed += reclaimed;
2636 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002637 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002638 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002639
2640 /*
2641 * If we failed to reclaim anything from this memory cgroup
2642 * it is time to move on to the next cgroup
2643 */
2644 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002645 if (!reclaimed)
2646 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2647
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002648 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002649 /*
2650 * One school of thought says that we should not add
2651 * back the node to the tree if reclaim returns 0.
2652 * But our reclaim could return 0, simply because due
2653 * to priority we are exposing a smaller subset of
2654 * memory to reclaim from. Consider this as a longer
2655 * term TODO.
2656 */
2657 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002658 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002659 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002660 css_put(&mz->memcg->css);
2661 loop++;
2662 /*
2663 * Could not reclaim anything and there are no more
2664 * mem cgroups to try or we seem to be looping without
2665 * reclaiming anything.
2666 */
2667 if (!nr_reclaimed &&
2668 (next_mz == NULL ||
2669 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2670 break;
2671 } while (!nr_reclaimed);
2672 if (next_mz)
2673 css_put(&next_mz->memcg->css);
2674 return nr_reclaimed;
2675}
2676
Tejun Heoea280e72014-05-16 13:22:48 -04002677/*
2678 * Test whether @memcg has children, dead or alive. Note that this
2679 * function doesn't care whether @memcg has use_hierarchy enabled and
2680 * returns %true if there are child csses according to the cgroup
2681 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2682 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002683static inline bool memcg_has_children(struct mem_cgroup *memcg)
2684{
Tejun Heoea280e72014-05-16 13:22:48 -04002685 bool ret;
2686
Johannes Weiner696ac172013-10-31 16:34:15 -07002687 /*
Tejun Heoea280e72014-05-16 13:22:48 -04002688 * The lock does not prevent addition or deletion of children, but
2689 * it prevents a new child from being initialized based on this
2690 * parent in css_online(), so it's enough to decide whether
2691 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07002692 */
Tejun Heoea280e72014-05-16 13:22:48 -04002693 lockdep_assert_held(&memcg_create_mutex);
2694
2695 rcu_read_lock();
2696 ret = css_next_child(NULL, &memcg->css);
2697 rcu_read_unlock();
2698 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002699}
2700
2701/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002702 * Reclaims as many pages from the given memcg as possible and moves
2703 * the rest to the parent.
2704 *
2705 * Caller is responsible for holding css reference for memcg.
2706 */
2707static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2708{
2709 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002710
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002711 /* we call try-to-free pages for make this cgroup empty */
2712 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002713 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002714 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002715 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002716
Michal Hockoc26251f2012-10-26 13:37:28 +02002717 if (signal_pending(current))
2718 return -EINTR;
2719
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002720 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2721 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002722 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002723 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002724 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002725 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002726 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002727
2728 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002729
2730 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002731}
2732
Tejun Heo6770c642014-05-13 12:16:21 -04002733static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2734 char *buf, size_t nbytes,
2735 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002736{
Tejun Heo6770c642014-05-13 12:16:21 -04002737 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002738
Michal Hockod8423012012-10-26 13:37:29 +02002739 if (mem_cgroup_is_root(memcg))
2740 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002741 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002742}
2743
Tejun Heo182446d2013-08-08 20:11:24 -04002744static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2745 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002746{
Tejun Heo182446d2013-08-08 20:11:24 -04002747 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002748}
2749
Tejun Heo182446d2013-08-08 20:11:24 -04002750static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2751 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002752{
2753 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002754 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002755 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002756
Glauber Costa09998212013-02-22 16:34:55 -08002757 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07002758
2759 if (memcg->use_hierarchy == val)
2760 goto out;
2761
Balbir Singh18f59ea2009-01-07 18:08:07 -08002762 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002763 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002764 * in the child subtrees. If it is unset, then the change can
2765 * occur, provided the current cgroup has no children.
2766 *
2767 * For the root cgroup, parent_mem is NULL, we allow value to be
2768 * set if there are no children.
2769 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002770 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002771 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002772 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002773 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002774 else
2775 retval = -EBUSY;
2776 } else
2777 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002778
2779out:
Glauber Costa09998212013-02-22 16:34:55 -08002780 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002781
2782 return retval;
2783}
2784
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002785static unsigned long tree_stat(struct mem_cgroup *memcg,
2786 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04002787{
2788 struct mem_cgroup *iter;
Greg Thelen484ebb32015-10-01 15:37:05 -07002789 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002790
Johannes Weinerce00a962014-09-05 08:43:57 -04002791 for_each_mem_cgroup_tree(iter, memcg)
2792 val += mem_cgroup_read_stat(iter, idx);
2793
Johannes Weinerce00a962014-09-05 08:43:57 -04002794 return val;
2795}
2796
Andrew Morton6f646152015-11-06 16:28:58 -08002797static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002798{
Michal Hockoc12176d2015-11-05 18:50:29 -08002799 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002800
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002801 if (mem_cgroup_is_root(memcg)) {
2802 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
2803 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
2804 if (swap)
2805 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
2806 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002807 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002808 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002809 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002810 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002811 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002812 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002813}
2814
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002815enum {
2816 RES_USAGE,
2817 RES_LIMIT,
2818 RES_MAX_USAGE,
2819 RES_FAILCNT,
2820 RES_SOFT_LIMIT,
2821};
Johannes Weinerce00a962014-09-05 08:43:57 -04002822
Tejun Heo791badb2013-12-05 12:28:02 -05002823static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002824 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002825{
Tejun Heo182446d2013-08-08 20:11:24 -04002826 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002827 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002828
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002829 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002830 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002831 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002832 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002833 case _MEMSWAP:
2834 counter = &memcg->memsw;
2835 break;
2836 case _KMEM:
2837 counter = &memcg->kmem;
2838 break;
2839 default:
2840 BUG();
2841 }
2842
2843 switch (MEMFILE_ATTR(cft->private)) {
2844 case RES_USAGE:
2845 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002846 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002847 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002848 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002849 return (u64)page_counter_read(counter) * PAGE_SIZE;
2850 case RES_LIMIT:
2851 return (u64)counter->limit * PAGE_SIZE;
2852 case RES_MAX_USAGE:
2853 return (u64)counter->watermark * PAGE_SIZE;
2854 case RES_FAILCNT:
2855 return counter->failcnt;
2856 case RES_SOFT_LIMIT:
2857 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002858 default:
2859 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002860 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002861}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002862
Glauber Costa510fc4e2012-12-18 14:21:47 -08002863#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002864static int memcg_activate_kmem(struct mem_cgroup *memcg,
2865 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002866{
2867 int err = 0;
2868 int memcg_id;
2869
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002870 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002871 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002872 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002873
2874 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08002875 * For simplicity, we won't allow this to be disabled. It also can't
2876 * be changed if the cgroup has children already, or if tasks had
2877 * already joined.
2878 *
2879 * If tasks join before we set the limit, a person looking at
2880 * kmem.usage_in_bytes will have no way to determine when it took
2881 * place, which makes the value quite meaningless.
2882 *
2883 * After it first became limited, changes in the value of the limit are
2884 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08002885 */
Glauber Costa09998212013-02-22 16:34:55 -08002886 mutex_lock(&memcg_create_mutex);
Tejun Heo27bd4db2015-10-15 16:41:50 -04002887 if (cgroup_is_populated(memcg->css.cgroup) ||
Tejun Heoea280e72014-05-16 13:22:48 -04002888 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08002889 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08002890 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002891 if (err)
2892 goto out;
2893
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002894 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08002895 if (memcg_id < 0) {
2896 err = memcg_id;
2897 goto out;
2898 }
2899
Vladimir Davydovd6441632014-01-23 15:53:09 -08002900 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002901 * We couldn't have accounted to this cgroup, because it hasn't got
2902 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08002903 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002904 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002905 VM_BUG_ON(err);
2906
Johannes Weineref129472016-01-14 15:21:34 -08002907 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002908 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002909 * A memory cgroup is considered kmem-active as soon as it gets
2910 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002911 * guarantee no one starts accounting before all call sites are
2912 * patched.
2913 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002914 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002915 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002916 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002917out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08002918 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002919}
2920
Vladimir Davydovd6441632014-01-23 15:53:09 -08002921static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002922 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002923{
2924 int ret;
2925
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002926 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002927 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002928 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002929 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002930 ret = page_counter_limit(&memcg->kmem, limit);
2931 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002932 return ret;
2933}
2934
Glauber Costa55007d82012-12-18 14:22:38 -08002935static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08002936{
Glauber Costa55007d82012-12-18 14:22:38 -08002937 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002938 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002939
Glauber Costa510fc4e2012-12-18 14:21:47 -08002940 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002941 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08002942
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002943 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08002944 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08002945 * If the parent cgroup is not kmem-active now, it cannot be activated
2946 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08002947 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08002948 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002949 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
2950 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08002951 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002952}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002953#else
2954static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002955 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002956{
2957 return -EINVAL;
2958}
Hugh Dickins6d0439902013-02-22 16:35:50 -08002959#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08002960
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002961/*
2962 * The user of this function is...
2963 * RES_LIMIT.
2964 */
Tejun Heo451af502014-05-13 12:16:21 -04002965static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2966 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002967{
Tejun Heo451af502014-05-13 12:16:21 -04002968 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002969 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002970 int ret;
2971
Tejun Heo451af502014-05-13 12:16:21 -04002972 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002973 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002974 if (ret)
2975 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002976
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002977 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002978 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002979 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2980 ret = -EINVAL;
2981 break;
2982 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002983 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2984 case _MEM:
2985 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002986 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002987 case _MEMSWAP:
2988 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2989 break;
2990 case _KMEM:
2991 ret = memcg_update_kmem_limit(memcg, nr_pages);
2992 break;
2993 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002994 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07002995 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002996 memcg->soft_limit = nr_pages;
2997 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002998 break;
2999 }
Tejun Heo451af502014-05-13 12:16:21 -04003000 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003001}
3002
Tejun Heo6770c642014-05-13 12:16:21 -04003003static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3004 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003005{
Tejun Heo6770c642014-05-13 12:16:21 -04003006 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003008
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003009 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3010 case _MEM:
3011 counter = &memcg->memory;
3012 break;
3013 case _MEMSWAP:
3014 counter = &memcg->memsw;
3015 break;
3016 case _KMEM:
3017 counter = &memcg->kmem;
3018 break;
3019 default:
3020 BUG();
3021 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003022
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003023 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003024 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003025 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003026 break;
3027 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003028 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003029 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003030 default:
3031 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003032 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003033
Tejun Heo6770c642014-05-13 12:16:21 -04003034 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003035}
3036
Tejun Heo182446d2013-08-08 20:11:24 -04003037static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003038 struct cftype *cft)
3039{
Tejun Heo182446d2013-08-08 20:11:24 -04003040 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003041}
3042
Daisuke Nishimura02491442010-03-10 15:22:17 -08003043#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003044static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003045 struct cftype *cft, u64 val)
3046{
Tejun Heo182446d2013-08-08 20:11:24 -04003047 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003048
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003049 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003050 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003051
Glauber Costaee5e8472013-02-22 16:34:50 -08003052 /*
3053 * No kind of locking is needed in here, because ->can_attach() will
3054 * check this value once in the beginning of the process, and then carry
3055 * on with stale data. This means that changes to this value will only
3056 * affect task migrations starting after the change.
3057 */
3058 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003059 return 0;
3060}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003061#else
Tejun Heo182446d2013-08-08 20:11:24 -04003062static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003063 struct cftype *cft, u64 val)
3064{
3065 return -ENOSYS;
3066}
3067#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068
Ying Han406eb0c2011-05-26 16:25:37 -07003069#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003070static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003071{
Greg Thelen25485de2013-11-12 15:07:40 -08003072 struct numa_stat {
3073 const char *name;
3074 unsigned int lru_mask;
3075 };
3076
3077 static const struct numa_stat stats[] = {
3078 { "total", LRU_ALL },
3079 { "file", LRU_ALL_FILE },
3080 { "anon", LRU_ALL_ANON },
3081 { "unevictable", BIT(LRU_UNEVICTABLE) },
3082 };
3083 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003084 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003085 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003086 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003087
Greg Thelen25485de2013-11-12 15:07:40 -08003088 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3089 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3090 seq_printf(m, "%s=%lu", stat->name, nr);
3091 for_each_node_state(nid, N_MEMORY) {
3092 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3093 stat->lru_mask);
3094 seq_printf(m, " N%d=%lu", nid, nr);
3095 }
3096 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003097 }
Ying Han406eb0c2011-05-26 16:25:37 -07003098
Ying Han071aee12013-11-12 15:07:41 -08003099 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3100 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003101
Ying Han071aee12013-11-12 15:07:41 -08003102 nr = 0;
3103 for_each_mem_cgroup_tree(iter, memcg)
3104 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3105 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3106 for_each_node_state(nid, N_MEMORY) {
3107 nr = 0;
3108 for_each_mem_cgroup_tree(iter, memcg)
3109 nr += mem_cgroup_node_nr_lru_pages(
3110 iter, nid, stat->lru_mask);
3111 seq_printf(m, " N%d=%lu", nid, nr);
3112 }
3113 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003114 }
Ying Han406eb0c2011-05-26 16:25:37 -07003115
Ying Han406eb0c2011-05-26 16:25:37 -07003116 return 0;
3117}
3118#endif /* CONFIG_NUMA */
3119
Tejun Heo2da8ca82013-12-05 12:28:04 -05003120static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003121{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003122 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003123 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003124 struct mem_cgroup *mi;
3125 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003126
Greg Thelen0ca44b12015-02-11 15:25:58 -08003127 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3128 MEM_CGROUP_STAT_NSTATS);
3129 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3130 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003131 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3132
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003133 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003134 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003135 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003136 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003137 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003138 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003139
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003140 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3141 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3142 mem_cgroup_read_events(memcg, i));
3143
3144 for (i = 0; i < NR_LRU_LISTS; i++)
3145 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3146 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3147
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003148 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003149 memory = memsw = PAGE_COUNTER_MAX;
3150 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3151 memory = min(memory, mi->memory.limit);
3152 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003153 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003154 seq_printf(m, "hierarchical_memory_limit %llu\n",
3155 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003156 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003157 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3158 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003159
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003160 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003161 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003162
Johannes Weiner7941d212016-01-14 15:21:23 -08003163 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003164 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003165 for_each_mem_cgroup_tree(mi, memcg)
3166 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003167 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003168 }
3169
3170 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3171 unsigned long long val = 0;
3172
3173 for_each_mem_cgroup_tree(mi, memcg)
3174 val += mem_cgroup_read_events(mi, i);
3175 seq_printf(m, "total_%s %llu\n",
3176 mem_cgroup_events_names[i], val);
3177 }
3178
3179 for (i = 0; i < NR_LRU_LISTS; i++) {
3180 unsigned long long val = 0;
3181
3182 for_each_mem_cgroup_tree(mi, memcg)
3183 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3184 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003185 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003186
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003187#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003188 {
3189 int nid, zid;
3190 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003191 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003192 unsigned long recent_rotated[2] = {0, 0};
3193 unsigned long recent_scanned[2] = {0, 0};
3194
3195 for_each_online_node(nid)
3196 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003197 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003198 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003199
Hugh Dickins89abfab2012-05-29 15:06:53 -07003200 recent_rotated[0] += rstat->recent_rotated[0];
3201 recent_rotated[1] += rstat->recent_rotated[1];
3202 recent_scanned[0] += rstat->recent_scanned[0];
3203 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003204 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003205 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3206 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3207 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3208 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003209 }
3210#endif
3211
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003212 return 0;
3213}
3214
Tejun Heo182446d2013-08-08 20:11:24 -04003215static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3216 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003217{
Tejun Heo182446d2013-08-08 20:11:24 -04003218 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003219
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003220 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003221}
3222
Tejun Heo182446d2013-08-08 20:11:24 -04003223static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3224 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003225{
Tejun Heo182446d2013-08-08 20:11:24 -04003226 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003227
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003228 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003229 return -EINVAL;
3230
Linus Torvalds14208b02014-06-09 15:03:33 -07003231 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003232 memcg->swappiness = val;
3233 else
3234 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003235
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003236 return 0;
3237}
3238
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003239static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3240{
3241 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003242 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003243 int i;
3244
3245 rcu_read_lock();
3246 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003247 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003248 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003249 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003250
3251 if (!t)
3252 goto unlock;
3253
Johannes Weinerce00a962014-09-05 08:43:57 -04003254 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003255
3256 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003257 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003258 * If it's not true, a threshold was crossed after last
3259 * call of __mem_cgroup_threshold().
3260 */
Phil Carmody5407a562010-05-26 14:42:42 -07003261 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003262
3263 /*
3264 * Iterate backward over array of thresholds starting from
3265 * current_threshold and check if a threshold is crossed.
3266 * If none of thresholds below usage is crossed, we read
3267 * only one element of the array here.
3268 */
3269 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3270 eventfd_signal(t->entries[i].eventfd, 1);
3271
3272 /* i = current_threshold + 1 */
3273 i++;
3274
3275 /*
3276 * Iterate forward over array of thresholds starting from
3277 * current_threshold+1 and check if a threshold is crossed.
3278 * If none of thresholds above usage is crossed, we read
3279 * only one element of the array here.
3280 */
3281 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3282 eventfd_signal(t->entries[i].eventfd, 1);
3283
3284 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003285 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003286unlock:
3287 rcu_read_unlock();
3288}
3289
3290static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3291{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003292 while (memcg) {
3293 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003294 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003295 __mem_cgroup_threshold(memcg, true);
3296
3297 memcg = parent_mem_cgroup(memcg);
3298 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003299}
3300
3301static int compare_thresholds(const void *a, const void *b)
3302{
3303 const struct mem_cgroup_threshold *_a = a;
3304 const struct mem_cgroup_threshold *_b = b;
3305
Greg Thelen2bff24a2013-09-11 14:23:08 -07003306 if (_a->threshold > _b->threshold)
3307 return 1;
3308
3309 if (_a->threshold < _b->threshold)
3310 return -1;
3311
3312 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003313}
3314
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003315static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003316{
3317 struct mem_cgroup_eventfd_list *ev;
3318
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003319 spin_lock(&memcg_oom_lock);
3320
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003321 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003322 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003323
3324 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003325 return 0;
3326}
3327
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003328static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003329{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003330 struct mem_cgroup *iter;
3331
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003332 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003333 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003334}
3335
Tejun Heo59b6f872013-11-22 18:20:43 -05003336static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003337 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003338{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003339 struct mem_cgroup_thresholds *thresholds;
3340 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003341 unsigned long threshold;
3342 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003343 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003344
Johannes Weiner650c5e52015-02-11 15:26:03 -08003345 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003346 if (ret)
3347 return ret;
3348
3349 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003350
Johannes Weiner05b84302014-08-06 16:05:59 -07003351 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003352 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003353 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003354 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003355 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003356 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003357 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003358 BUG();
3359
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003360 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003361 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003362 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3363
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003364 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003365
3366 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003367 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003368 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003369 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003370 ret = -ENOMEM;
3371 goto unlock;
3372 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003373 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003374
3375 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003376 if (thresholds->primary) {
3377 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003379 }
3380
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003382 new->entries[size - 1].eventfd = eventfd;
3383 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003384
3385 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003387 compare_thresholds, NULL);
3388
3389 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003390 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003392 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003393 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003394 * new->current_threshold will not be used until
3395 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003396 * it here.
3397 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003399 } else
3400 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003401 }
3402
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 /* Free old spare buffer and save old primary buffer as spare */
3404 kfree(thresholds->spare);
3405 thresholds->spare = thresholds->primary;
3406
3407 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003409 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003410 synchronize_rcu();
3411
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412unlock:
3413 mutex_unlock(&memcg->thresholds_lock);
3414
3415 return ret;
3416}
3417
Tejun Heo59b6f872013-11-22 18:20:43 -05003418static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003419 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003420{
Tejun Heo59b6f872013-11-22 18:20:43 -05003421 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003422}
3423
Tejun Heo59b6f872013-11-22 18:20:43 -05003424static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003425 struct eventfd_ctx *eventfd, const char *args)
3426{
Tejun Heo59b6f872013-11-22 18:20:43 -05003427 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003428}
3429
Tejun Heo59b6f872013-11-22 18:20:43 -05003430static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003431 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003432{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003433 struct mem_cgroup_thresholds *thresholds;
3434 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003435 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003436 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003437
3438 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003439
3440 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003441 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003442 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003443 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003444 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003445 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003446 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003447 BUG();
3448
Anton Vorontsov371528c2012-02-24 05:14:46 +04003449 if (!thresholds->primary)
3450 goto unlock;
3451
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003452 /* Check if a threshold crossed before removing */
3453 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3454
3455 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003456 size = 0;
3457 for (i = 0; i < thresholds->primary->size; i++) {
3458 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003459 size++;
3460 }
3461
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003462 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003463
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003464 /* Set thresholds array to NULL if we don't have thresholds */
3465 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003466 kfree(new);
3467 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003468 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003469 }
3470
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003471 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003472
3473 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003474 new->current_threshold = -1;
3475 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3476 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003477 continue;
3478
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003479 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003480 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003481 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003482 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003483 * until rcu_assign_pointer(), so it's safe to increment
3484 * it here.
3485 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003486 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003487 }
3488 j++;
3489 }
3490
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003491swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003492 /* Swap primary and spare array */
3493 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003494 /* If all events are unregistered, free the spare array */
3495 if (!new) {
3496 kfree(thresholds->spare);
3497 thresholds->spare = NULL;
3498 }
3499
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003500 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003501
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003502 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04003504unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003505 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003507
Tejun Heo59b6f872013-11-22 18:20:43 -05003508static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003509 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003510{
Tejun Heo59b6f872013-11-22 18:20:43 -05003511 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003512}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003513
Tejun Heo59b6f872013-11-22 18:20:43 -05003514static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003515 struct eventfd_ctx *eventfd)
3516{
Tejun Heo59b6f872013-11-22 18:20:43 -05003517 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003518}
3519
Tejun Heo59b6f872013-11-22 18:20:43 -05003520static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003521 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003522{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003523 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003524
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003525 event = kmalloc(sizeof(*event), GFP_KERNEL);
3526 if (!event)
3527 return -ENOMEM;
3528
Michal Hocko1af8efe2011-07-26 16:08:24 -07003529 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530
3531 event->eventfd = eventfd;
3532 list_add(&event->list, &memcg->oom_notify);
3533
3534 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003535 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003536 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003537 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003538
3539 return 0;
3540}
3541
Tejun Heo59b6f872013-11-22 18:20:43 -05003542static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003543 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003544{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003546
Michal Hocko1af8efe2011-07-26 16:08:24 -07003547 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003549 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003550 if (ev->eventfd == eventfd) {
3551 list_del(&ev->list);
3552 kfree(ev);
3553 }
3554 }
3555
Michal Hocko1af8efe2011-07-26 16:08:24 -07003556 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003557}
3558
Tejun Heo2da8ca82013-12-05 12:28:04 -05003559static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003560{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003561 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003562
Tejun Heo791badb2013-12-05 12:28:02 -05003563 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003564 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003565 return 0;
3566}
3567
Tejun Heo182446d2013-08-08 20:11:24 -04003568static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003569 struct cftype *cft, u64 val)
3570{
Tejun Heo182446d2013-08-08 20:11:24 -04003571 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003572
3573 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003574 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003575 return -EINVAL;
3576
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003577 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003578 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003579 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003580
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003581 return 0;
3582}
3583
Andrew Mortonc255a452012-07-31 16:43:02 -07003584#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003585static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003586{
Glauber Costa55007d82012-12-18 14:22:38 -08003587 int ret;
3588
Glauber Costa55007d82012-12-18 14:22:38 -08003589 ret = memcg_propagate_kmem(memcg);
3590 if (ret)
3591 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003592
Johannes Weinerbaac50b2016-01-14 15:21:17 -08003593 return tcp_init_cgroup(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003594}
Glauber Costae5671df2011-12-11 21:47:01 +00003595
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003596static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3597{
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003598 struct cgroup_subsys_state *css;
3599 struct mem_cgroup *parent, *child;
3600 int kmemcg_id;
3601
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003602 if (!memcg->kmem_acct_active)
3603 return;
3604
3605 /*
3606 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3607 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3608 * guarantees no cache will be created for this cgroup after we are
3609 * done (see memcg_create_kmem_cache()).
3610 */
3611 memcg->kmem_acct_active = false;
3612
3613 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003614
3615 kmemcg_id = memcg->kmemcg_id;
3616 BUG_ON(kmemcg_id < 0);
3617
3618 parent = parent_mem_cgroup(memcg);
3619 if (!parent)
3620 parent = root_mem_cgroup;
3621
3622 /*
3623 * Change kmemcg_id of this cgroup and all its descendants to the
3624 * parent's id, and then move all entries from this cgroup's list_lrus
3625 * to ones of the parent. After we have finished, all list_lrus
3626 * corresponding to this cgroup are guaranteed to remain empty. The
3627 * ordering is imposed by list_lru_node->lock taken by
3628 * memcg_drain_all_list_lrus().
3629 */
3630 css_for_each_descendant_pre(css, &memcg->css) {
3631 child = mem_cgroup_from_css(css);
3632 BUG_ON(child->kmemcg_id != kmemcg_id);
3633 child->kmemcg_id = parent->kmemcg_id;
3634 if (!memcg->use_hierarchy)
3635 break;
3636 }
3637 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3638
3639 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003640}
3641
Li Zefan10d5ebf2013-07-08 16:00:33 -07003642static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003643{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003644 if (memcg->kmem_acct_activated) {
3645 memcg_destroy_kmem_caches(memcg);
Johannes Weineref129472016-01-14 15:21:34 -08003646 static_branch_dec(&memcg_kmem_enabled_key);
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003647 WARN_ON(page_counter_read(&memcg->kmem));
3648 }
Johannes Weinerbaac50b2016-01-14 15:21:17 -08003649 tcp_destroy_cgroup(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003650}
Glauber Costae5671df2011-12-11 21:47:01 +00003651#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003652static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003653{
3654 return 0;
3655}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003656
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003657static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3658{
3659}
3660
Li Zefan10d5ebf2013-07-08 16:00:33 -07003661static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3662{
3663}
Glauber Costae5671df2011-12-11 21:47:01 +00003664#endif
3665
Tejun Heo52ebea72015-05-22 17:13:37 -04003666#ifdef CONFIG_CGROUP_WRITEBACK
3667
3668struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3669{
3670 return &memcg->cgwb_list;
3671}
3672
Tejun Heo841710a2015-05-22 18:23:33 -04003673static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3674{
3675 return wb_domain_init(&memcg->cgwb_domain, gfp);
3676}
3677
3678static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3679{
3680 wb_domain_exit(&memcg->cgwb_domain);
3681}
3682
Tejun Heo2529bb32015-05-22 18:23:34 -04003683static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3684{
3685 wb_domain_size_changed(&memcg->cgwb_domain);
3686}
3687
Tejun Heo841710a2015-05-22 18:23:33 -04003688struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3689{
3690 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3691
3692 if (!memcg->css.parent)
3693 return NULL;
3694
3695 return &memcg->cgwb_domain;
3696}
3697
Tejun Heoc2aa7232015-05-22 18:23:35 -04003698/**
3699 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3700 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003701 * @pfilepages: out parameter for number of file pages
3702 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003703 * @pdirty: out parameter for number of dirty pages
3704 * @pwriteback: out parameter for number of pages under writeback
3705 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003706 * Determine the numbers of file, headroom, dirty, and writeback pages in
3707 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3708 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003709 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003710 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3711 * headroom is calculated as the lowest headroom of itself and the
3712 * ancestors. Note that this doesn't consider the actual amount of
3713 * available memory in the system. The caller should further cap
3714 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003715 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003716void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3717 unsigned long *pheadroom, unsigned long *pdirty,
3718 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003719{
3720 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3721 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003722
3723 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3724
3725 /* this should eventually include NR_UNSTABLE_NFS */
3726 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003727 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3728 (1 << LRU_ACTIVE_FILE));
3729 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003730
Tejun Heoc2aa7232015-05-22 18:23:35 -04003731 while ((parent = parent_mem_cgroup(memcg))) {
3732 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3733 unsigned long used = page_counter_read(&memcg->memory);
3734
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003735 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003736 memcg = parent;
3737 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003738}
3739
Tejun Heo841710a2015-05-22 18:23:33 -04003740#else /* CONFIG_CGROUP_WRITEBACK */
3741
3742static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3743{
3744 return 0;
3745}
3746
3747static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3748{
3749}
3750
Tejun Heo2529bb32015-05-22 18:23:34 -04003751static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3752{
3753}
3754
Tejun Heo52ebea72015-05-22 17:13:37 -04003755#endif /* CONFIG_CGROUP_WRITEBACK */
3756
Tejun Heo79bd9812013-11-22 18:20:42 -05003757/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003758 * DO NOT USE IN NEW FILES.
3759 *
3760 * "cgroup.event_control" implementation.
3761 *
3762 * This is way over-engineered. It tries to support fully configurable
3763 * events for each user. Such level of flexibility is completely
3764 * unnecessary especially in the light of the planned unified hierarchy.
3765 *
3766 * Please deprecate this and replace with something simpler if at all
3767 * possible.
3768 */
3769
3770/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003771 * Unregister event and free resources.
3772 *
3773 * Gets called from workqueue.
3774 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003775static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003776{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003777 struct mem_cgroup_event *event =
3778 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003779 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003780
3781 remove_wait_queue(event->wqh, &event->wait);
3782
Tejun Heo59b6f872013-11-22 18:20:43 -05003783 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003784
3785 /* Notify userspace the event is going away. */
3786 eventfd_signal(event->eventfd, 1);
3787
3788 eventfd_ctx_put(event->eventfd);
3789 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003790 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003791}
3792
3793/*
3794 * Gets called on POLLHUP on eventfd when user closes it.
3795 *
3796 * Called with wqh->lock held and interrupts disabled.
3797 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003798static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3799 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003800{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003801 struct mem_cgroup_event *event =
3802 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003803 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003804 unsigned long flags = (unsigned long)key;
3805
3806 if (flags & POLLHUP) {
3807 /*
3808 * If the event has been detached at cgroup removal, we
3809 * can simply return knowing the other side will cleanup
3810 * for us.
3811 *
3812 * We can't race against event freeing since the other
3813 * side will require wqh->lock via remove_wait_queue(),
3814 * which we hold.
3815 */
Tejun Heofba94802013-11-22 18:20:43 -05003816 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003817 if (!list_empty(&event->list)) {
3818 list_del_init(&event->list);
3819 /*
3820 * We are in atomic context, but cgroup_event_remove()
3821 * may sleep, so we have to call it in workqueue.
3822 */
3823 schedule_work(&event->remove);
3824 }
Tejun Heofba94802013-11-22 18:20:43 -05003825 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003826 }
3827
3828 return 0;
3829}
3830
Tejun Heo3bc942f2013-11-22 18:20:44 -05003831static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003832 wait_queue_head_t *wqh, poll_table *pt)
3833{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003834 struct mem_cgroup_event *event =
3835 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003836
3837 event->wqh = wqh;
3838 add_wait_queue(wqh, &event->wait);
3839}
3840
3841/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003842 * DO NOT USE IN NEW FILES.
3843 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003844 * Parse input and register new cgroup event handler.
3845 *
3846 * Input must be in format '<event_fd> <control_fd> <args>'.
3847 * Interpretation of args is defined by control file implementation.
3848 */
Tejun Heo451af502014-05-13 12:16:21 -04003849static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3850 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003851{
Tejun Heo451af502014-05-13 12:16:21 -04003852 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003853 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003854 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003855 struct cgroup_subsys_state *cfile_css;
3856 unsigned int efd, cfd;
3857 struct fd efile;
3858 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003859 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003860 char *endp;
3861 int ret;
3862
Tejun Heo451af502014-05-13 12:16:21 -04003863 buf = strstrip(buf);
3864
3865 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003866 if (*endp != ' ')
3867 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003868 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003869
Tejun Heo451af502014-05-13 12:16:21 -04003870 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003871 if ((*endp != ' ') && (*endp != '\0'))
3872 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003873 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003874
3875 event = kzalloc(sizeof(*event), GFP_KERNEL);
3876 if (!event)
3877 return -ENOMEM;
3878
Tejun Heo59b6f872013-11-22 18:20:43 -05003879 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003880 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003881 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3882 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3883 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003884
3885 efile = fdget(efd);
3886 if (!efile.file) {
3887 ret = -EBADF;
3888 goto out_kfree;
3889 }
3890
3891 event->eventfd = eventfd_ctx_fileget(efile.file);
3892 if (IS_ERR(event->eventfd)) {
3893 ret = PTR_ERR(event->eventfd);
3894 goto out_put_efile;
3895 }
3896
3897 cfile = fdget(cfd);
3898 if (!cfile.file) {
3899 ret = -EBADF;
3900 goto out_put_eventfd;
3901 }
3902
3903 /* the process need read permission on control file */
3904 /* AV: shouldn't we check that it's been opened for read instead? */
3905 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3906 if (ret < 0)
3907 goto out_put_cfile;
3908
Tejun Heo79bd9812013-11-22 18:20:42 -05003909 /*
Tejun Heofba94802013-11-22 18:20:43 -05003910 * Determine the event callbacks and set them in @event. This used
3911 * to be done via struct cftype but cgroup core no longer knows
3912 * about these events. The following is crude but the whole thing
3913 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003914 *
3915 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003916 */
Al Virob5830432014-10-31 01:22:04 -04003917 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003918
3919 if (!strcmp(name, "memory.usage_in_bytes")) {
3920 event->register_event = mem_cgroup_usage_register_event;
3921 event->unregister_event = mem_cgroup_usage_unregister_event;
3922 } else if (!strcmp(name, "memory.oom_control")) {
3923 event->register_event = mem_cgroup_oom_register_event;
3924 event->unregister_event = mem_cgroup_oom_unregister_event;
3925 } else if (!strcmp(name, "memory.pressure_level")) {
3926 event->register_event = vmpressure_register_event;
3927 event->unregister_event = vmpressure_unregister_event;
3928 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003929 event->register_event = memsw_cgroup_usage_register_event;
3930 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003931 } else {
3932 ret = -EINVAL;
3933 goto out_put_cfile;
3934 }
3935
3936 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003937 * Verify @cfile should belong to @css. Also, remaining events are
3938 * automatically removed on cgroup destruction but the removal is
3939 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003940 */
Al Virob5830432014-10-31 01:22:04 -04003941 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003942 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003943 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003944 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003945 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003946 if (cfile_css != css) {
3947 css_put(cfile_css);
3948 goto out_put_cfile;
3949 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003950
Tejun Heo451af502014-05-13 12:16:21 -04003951 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003952 if (ret)
3953 goto out_put_css;
3954
3955 efile.file->f_op->poll(efile.file, &event->pt);
3956
Tejun Heofba94802013-11-22 18:20:43 -05003957 spin_lock(&memcg->event_list_lock);
3958 list_add(&event->list, &memcg->event_list);
3959 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003960
3961 fdput(cfile);
3962 fdput(efile);
3963
Tejun Heo451af502014-05-13 12:16:21 -04003964 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003965
3966out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003967 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003968out_put_cfile:
3969 fdput(cfile);
3970out_put_eventfd:
3971 eventfd_ctx_put(event->eventfd);
3972out_put_efile:
3973 fdput(efile);
3974out_kfree:
3975 kfree(event);
3976
3977 return ret;
3978}
3979
Johannes Weiner241994ed2015-02-11 15:26:06 -08003980static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003981 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003982 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003983 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003984 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003985 },
3986 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003987 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003988 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003989 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003990 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003991 },
3992 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003993 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003994 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003995 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003996 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003997 },
3998 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003999 .name = "soft_limit_in_bytes",
4000 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004001 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004002 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004003 },
4004 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004005 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004006 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004007 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004008 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004009 },
Balbir Singh8697d332008-02-07 00:13:59 -08004010 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004011 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004012 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004013 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004014 {
4015 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004016 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004017 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004018 {
4019 .name = "use_hierarchy",
4020 .write_u64 = mem_cgroup_hierarchy_write,
4021 .read_u64 = mem_cgroup_hierarchy_read,
4022 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004023 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004024 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004025 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04004026 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004027 },
4028 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004029 .name = "swappiness",
4030 .read_u64 = mem_cgroup_swappiness_read,
4031 .write_u64 = mem_cgroup_swappiness_write,
4032 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004033 {
4034 .name = "move_charge_at_immigrate",
4035 .read_u64 = mem_cgroup_move_charge_read,
4036 .write_u64 = mem_cgroup_move_charge_write,
4037 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004038 {
4039 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004040 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004041 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004042 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4043 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004044 {
4045 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004046 },
Ying Han406eb0c2011-05-26 16:25:37 -07004047#ifdef CONFIG_NUMA
4048 {
4049 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004050 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004051 },
4052#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004053#ifdef CONFIG_MEMCG_KMEM
4054 {
4055 .name = "kmem.limit_in_bytes",
4056 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004057 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004058 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004059 },
4060 {
4061 .name = "kmem.usage_in_bytes",
4062 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004063 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004064 },
4065 {
4066 .name = "kmem.failcnt",
4067 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004068 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004069 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004070 },
4071 {
4072 .name = "kmem.max_usage_in_bytes",
4073 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004074 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004075 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004076 },
Glauber Costa749c5412012-12-18 14:23:01 -08004077#ifdef CONFIG_SLABINFO
4078 {
4079 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004080 .seq_start = slab_start,
4081 .seq_next = slab_next,
4082 .seq_stop = slab_stop,
4083 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004084 },
4085#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004086#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004087 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004088};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004089
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004090static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004091{
4092 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004093 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004094 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004095 /*
4096 * This routine is called against possible nodes.
4097 * But it's BUG to call kmalloc() against offline node.
4098 *
4099 * TODO: this routine can waste much memory for nodes which will
4100 * never be onlined. It's better to use memory hotplug callback
4101 * function.
4102 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004103 if (!node_state(node, N_NORMAL_MEMORY))
4104 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004105 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004106 if (!pn)
4107 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004108
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004109 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4110 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004111 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07004112 mz->usage_in_excess = 0;
4113 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004114 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004115 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004116 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004117 return 0;
4118}
4119
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004120static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004121{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004122 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004123}
4124
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004125static struct mem_cgroup *mem_cgroup_alloc(void)
4126{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004127 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004128 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004129
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004130 size = sizeof(struct mem_cgroup);
4131 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004132
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004133 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004134 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004135 return NULL;
4136
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004137 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4138 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004139 goto out_free;
Tejun Heo841710a2015-05-22 18:23:33 -04004140
4141 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4142 goto out_free_stat;
4143
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004144 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004145
Tejun Heo841710a2015-05-22 18:23:33 -04004146out_free_stat:
4147 free_percpu(memcg->stat);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004148out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004149 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004150 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004151}
4152
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004153/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004154 * At destroying mem_cgroup, references from swap_cgroup can remain.
4155 * (scanning all at force_empty is too costly...)
4156 *
4157 * Instead of clearing all references at force_empty, we remember
4158 * the number of reference from swap_cgroup and free mem_cgroup when
4159 * it goes down to 0.
4160 *
4161 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004162 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004163
4164static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004165{
Glauber Costac8b2a362012-12-18 14:22:13 -08004166 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004167
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004168 cancel_work_sync(&memcg->high_work);
4169
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07004170 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004171
4172 for_each_node(node)
4173 free_mem_cgroup_per_zone_info(memcg, node);
4174
4175 free_percpu(memcg->stat);
Tejun Heo841710a2015-05-22 18:23:33 -04004176 memcg_wb_domain_exit(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004177 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004178}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004179
Li Zefan0eb253e2009-01-15 13:51:25 -08004180static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004181mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004182{
Glauber Costad142e3e2013-02-22 16:34:52 -08004183 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004184 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004185 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004186
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004187 memcg = mem_cgroup_alloc();
4188 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004189 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004190
Bob Liu3ed28fa2012-01-12 17:19:04 -08004191 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004192 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004193 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004194
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004195 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004196 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004197 root_mem_cgroup = memcg;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004198 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004199 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004200 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004201 page_counter_init(&memcg->memsw, NULL);
4202 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004203 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004204
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004205 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004206 memcg->last_scanned_node = MAX_NUMNODES;
4207 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004208 memcg->move_charge_at_immigrate = 0;
4209 mutex_init(&memcg->thresholds_lock);
4210 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004211 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004212 INIT_LIST_HEAD(&memcg->event_list);
4213 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004214#ifdef CONFIG_MEMCG_KMEM
4215 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004216#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004217#ifdef CONFIG_CGROUP_WRITEBACK
4218 INIT_LIST_HEAD(&memcg->cgwb_list);
4219#endif
Johannes Weiner8e8ae642016-01-14 15:21:32 -08004220#ifdef CONFIG_INET
4221 memcg->socket_pressure = jiffies;
4222#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004223 return &memcg->css;
4224
4225free_out:
4226 __mem_cgroup_free(memcg);
4227 return ERR_PTR(error);
4228}
4229
4230static int
Tejun Heoeb954192013-08-08 20:11:23 -04004231mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004232{
Tejun Heoeb954192013-08-08 20:11:23 -04004233 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004234 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004235 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004236
Tejun Heo15a4c832014-05-04 15:09:14 -04004237 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004238 return -ENOSPC;
4239
Tejun Heo63876982013-08-08 20:11:23 -04004240 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004241 return 0;
4242
Glauber Costa09998212013-02-22 16:34:55 -08004243 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004244
4245 memcg->use_hierarchy = parent->use_hierarchy;
4246 memcg->oom_kill_disable = parent->oom_kill_disable;
4247 memcg->swappiness = mem_cgroup_swappiness(parent);
4248
4249 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004250 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004251 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004252 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004253 page_counter_init(&memcg->memsw, &parent->memsw);
4254 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004255
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004256 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004257 * No need to take a reference to the parent because cgroup
4258 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004259 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004260 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004261 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004262 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004263 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004264 page_counter_init(&memcg->memsw, NULL);
4265 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004266 /*
4267 * Deeper hierachy with use_hierarchy == false doesn't make
4268 * much sense so let cgroup subsystem know about this
4269 * unfortunate state in our controller.
4270 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004271 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004272 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004273 }
Glauber Costa09998212013-02-22 16:34:55 -08004274 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004275
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004276 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4277 if (ret)
4278 return ret;
4279
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004280#ifdef CONFIG_INET
4281 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004282 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004283#endif
4284
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004285 /*
4286 * Make sure the memcg is initialized: mem_cgroup_iter()
4287 * orders reading memcg->initialized against its callers
4288 * reading the memcg members.
4289 */
4290 smp_store_release(&memcg->initialized, 1);
4291
4292 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004293}
4294
Tejun Heoeb954192013-08-08 20:11:23 -04004295static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004296{
Tejun Heoeb954192013-08-08 20:11:23 -04004297 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004298 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004299
4300 /*
4301 * Unregister events and notify userspace.
4302 * Notify userspace about cgroup removing only after rmdir of cgroup
4303 * directory to avoid race between userspace and kernelspace.
4304 */
Tejun Heofba94802013-11-22 18:20:43 -05004305 spin_lock(&memcg->event_list_lock);
4306 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004307 list_del_init(&event->list);
4308 schedule_work(&event->remove);
4309 }
Tejun Heofba94802013-11-22 18:20:43 -05004310 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004311
Michal Hocko33cb8762013-07-31 13:53:51 -07004312 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004313
4314 memcg_deactivate_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004315
4316 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004317}
4318
Vladimir Davydov6df38682015-12-29 14:54:10 -08004319static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4320{
4321 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4322
4323 invalidate_reclaim_iterators(memcg);
4324}
4325
Tejun Heoeb954192013-08-08 20:11:23 -04004326static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004327{
Tejun Heoeb954192013-08-08 20:11:23 -04004328 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004329
Li Zefan10d5ebf2013-07-08 16:00:33 -07004330 memcg_destroy_kmem(memcg);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004331#ifdef CONFIG_INET
4332 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004333 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004334#endif
Li Zefan465939a2013-07-08 16:00:38 -07004335 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004336}
4337
Tejun Heo1ced9532014-07-08 18:02:57 -04004338/**
4339 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4340 * @css: the target css
4341 *
4342 * Reset the states of the mem_cgroup associated with @css. This is
4343 * invoked when the userland requests disabling on the default hierarchy
4344 * but the memcg is pinned through dependency. The memcg should stop
4345 * applying policies and should revert to the vanilla state as it may be
4346 * made visible again.
4347 *
4348 * The current implementation only resets the essential configurations.
4349 * This needs to be expanded to cover all the visible parts.
4350 */
4351static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4352{
4353 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4354
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004355 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4356 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4357 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004358 memcg->low = 0;
4359 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004360 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004361 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004362}
4363
Daisuke Nishimura02491442010-03-10 15:22:17 -08004364#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004365/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004366static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004367{
Johannes Weiner05b84302014-08-06 16:05:59 -07004368 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004369
Mel Gormand0164ad2015-11-06 16:28:21 -08004370 /* Try a single bulk charge without reclaim first, kswapd may wake */
4371 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004372 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004373 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004374 return ret;
4375 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004376
4377 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004378 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004379 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004380 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004381 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004382 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004383 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004384 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004385 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004386}
4387
4388/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004389 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004390 * @vma: the vma the pte to be checked belongs
4391 * @addr: the address corresponding to the pte to be checked
4392 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004393 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004394 *
4395 * Returns
4396 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4397 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4398 * move charge. if @target is not NULL, the page is stored in target->page
4399 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004400 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4401 * target for charge migration. if @target is not NULL, the entry is stored
4402 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004403 *
4404 * Called with pte lock held.
4405 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004406union mc_target {
4407 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004408 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004409};
4410
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004411enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004412 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004413 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004414 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004415};
4416
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004417static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4418 unsigned long addr, pte_t ptent)
4419{
4420 struct page *page = vm_normal_page(vma, addr, ptent);
4421
4422 if (!page || !page_mapped(page))
4423 return NULL;
4424 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004425 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004426 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004427 } else {
4428 if (!(mc.flags & MOVE_FILE))
4429 return NULL;
4430 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004431 if (!get_page_unless_zero(page))
4432 return NULL;
4433
4434 return page;
4435}
4436
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004437#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004438static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4439 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4440{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004441 struct page *page = NULL;
4442 swp_entry_t ent = pte_to_swp_entry(ptent);
4443
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004444 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004445 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004446 /*
4447 * Because lookup_swap_cache() updates some statistics counter,
4448 * we call find_get_page() with swapper_space directly.
4449 */
Shaohua Li33806f02013-02-22 16:34:37 -08004450 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004451 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004452 entry->val = ent.val;
4453
4454 return page;
4455}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004456#else
4457static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4458 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4459{
4460 return NULL;
4461}
4462#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004463
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004464static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4465 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4466{
4467 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004468 struct address_space *mapping;
4469 pgoff_t pgoff;
4470
4471 if (!vma->vm_file) /* anonymous vma */
4472 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004473 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004474 return NULL;
4475
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004476 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004477 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004478
4479 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004480#ifdef CONFIG_SWAP
4481 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004482 if (shmem_mapping(mapping)) {
4483 page = find_get_entry(mapping, pgoff);
4484 if (radix_tree_exceptional_entry(page)) {
4485 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004486 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004487 *entry = swp;
4488 page = find_get_page(swap_address_space(swp), swp.val);
4489 }
4490 } else
4491 page = find_get_page(mapping, pgoff);
4492#else
4493 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004494#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004495 return page;
4496}
4497
Chen Gangb1b0dea2015-04-14 15:47:35 -07004498/**
4499 * mem_cgroup_move_account - move account of the page
4500 * @page: the page
4501 * @nr_pages: number of regular pages (>1 for huge pages)
4502 * @from: mem_cgroup which the page is moved from.
4503 * @to: mem_cgroup which the page is moved to. @from != @to.
4504 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004505 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004506 *
4507 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4508 * from old cgroup.
4509 */
4510static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004511 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004512 struct mem_cgroup *from,
4513 struct mem_cgroup *to)
4514{
4515 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004516 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004517 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004518 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004519
4520 VM_BUG_ON(from == to);
4521 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004522 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004523
4524 /*
Hugh Dickins45637ba2015-11-05 18:49:40 -08004525 * Prevent mem_cgroup_replace_page() from looking at
4526 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004527 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004528 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004529 if (!trylock_page(page))
4530 goto out;
4531
4532 ret = -EINVAL;
4533 if (page->mem_cgroup != from)
4534 goto out_unlock;
4535
Greg Thelenc4843a72015-05-22 17:13:16 -04004536 anon = PageAnon(page);
4537
Chen Gangb1b0dea2015-04-14 15:47:35 -07004538 spin_lock_irqsave(&from->move_lock, flags);
4539
Greg Thelenc4843a72015-05-22 17:13:16 -04004540 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004541 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4542 nr_pages);
4543 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4544 nr_pages);
4545 }
4546
Greg Thelenc4843a72015-05-22 17:13:16 -04004547 /*
4548 * move_lock grabbed above and caller set from->moving_account, so
4549 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4550 * So mapping should be stable for dirty pages.
4551 */
4552 if (!anon && PageDirty(page)) {
4553 struct address_space *mapping = page_mapping(page);
4554
4555 if (mapping_cap_account_dirty(mapping)) {
4556 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4557 nr_pages);
4558 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4559 nr_pages);
4560 }
4561 }
4562
Chen Gangb1b0dea2015-04-14 15:47:35 -07004563 if (PageWriteback(page)) {
4564 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4565 nr_pages);
4566 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4567 nr_pages);
4568 }
4569
4570 /*
4571 * It is safe to change page->mem_cgroup here because the page
4572 * is referenced, charged, and isolated - we can't race with
4573 * uncharging, charging, migration, or LRU putback.
4574 */
4575
4576 /* caller should have done css_get */
4577 page->mem_cgroup = to;
4578 spin_unlock_irqrestore(&from->move_lock, flags);
4579
4580 ret = 0;
4581
4582 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004583 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004584 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004585 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004586 memcg_check_events(from, page);
4587 local_irq_enable();
4588out_unlock:
4589 unlock_page(page);
4590out:
4591 return ret;
4592}
4593
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004594static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004595 unsigned long addr, pte_t ptent, union mc_target *target)
4596{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004597 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004598 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004599 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004600
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004601 if (pte_present(ptent))
4602 page = mc_handle_present_pte(vma, addr, ptent);
4603 else if (is_swap_pte(ptent))
4604 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004605 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004606 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004607
4608 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004609 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004610 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004611 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004612 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004613 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004614 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004615 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004616 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004617 ret = MC_TARGET_PAGE;
4618 if (target)
4619 target->page = page;
4620 }
4621 if (!ret || !target)
4622 put_page(page);
4623 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004624 /* There is a swap entry and a page doesn't exist or isn't charged */
4625 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004626 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004627 ret = MC_TARGET_SWAP;
4628 if (target)
4629 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004630 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004631 return ret;
4632}
4633
Naoya Horiguchi12724852012-03-21 16:34:28 -07004634#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4635/*
4636 * We don't consider swapping or file mapped pages because THP does not
4637 * support them for now.
4638 * Caller should make sure that pmd_trans_huge(pmd) is true.
4639 */
4640static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4641 unsigned long addr, pmd_t pmd, union mc_target *target)
4642{
4643 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004644 enum mc_target_type ret = MC_TARGET_NONE;
4645
4646 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004647 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004648 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004649 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004650 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004651 ret = MC_TARGET_PAGE;
4652 if (target) {
4653 get_page(page);
4654 target->page = page;
4655 }
4656 }
4657 return ret;
4658}
4659#else
4660static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4661 unsigned long addr, pmd_t pmd, union mc_target *target)
4662{
4663 return MC_TARGET_NONE;
4664}
4665#endif
4666
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004667static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4668 unsigned long addr, unsigned long end,
4669 struct mm_walk *walk)
4670{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004671 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004672 pte_t *pte;
4673 spinlock_t *ptl;
4674
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -08004675 if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004676 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4677 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004678 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004679 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004680 }
Dave Hansen03319322011-03-22 16:32:56 -07004681
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004682 if (pmd_trans_unstable(pmd))
4683 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004684 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4685 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004686 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004687 mc.precharge++; /* increment precharge temporarily */
4688 pte_unmap_unlock(pte - 1, ptl);
4689 cond_resched();
4690
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004691 return 0;
4692}
4693
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004694static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4695{
4696 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004697
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004698 struct mm_walk mem_cgroup_count_precharge_walk = {
4699 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4700 .mm = mm,
4701 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004702 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004703 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004704 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004705
4706 precharge = mc.precharge;
4707 mc.precharge = 0;
4708
4709 return precharge;
4710}
4711
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004712static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4713{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004714 unsigned long precharge = mem_cgroup_count_precharge(mm);
4715
4716 VM_BUG_ON(mc.moving_task);
4717 mc.moving_task = current;
4718 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004719}
4720
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004721/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4722static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004723{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004724 struct mem_cgroup *from = mc.from;
4725 struct mem_cgroup *to = mc.to;
4726
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004727 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004728 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004729 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004730 mc.precharge = 0;
4731 }
4732 /*
4733 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4734 * we must uncharge here.
4735 */
4736 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004737 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004738 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004739 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004740 /* we must fixup refcnts and charges */
4741 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004742 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004743 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004744 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004745
Johannes Weiner05b84302014-08-06 16:05:59 -07004746 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004747 * we charged both to->memory and to->memsw, so we
4748 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004749 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004750 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004751 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004752
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004753 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004754
Li Zefan40503772013-07-08 16:00:34 -07004755 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004756 mc.moved_swap = 0;
4757 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004758 memcg_oom_recover(from);
4759 memcg_oom_recover(to);
4760 wake_up_all(&mc.waitq);
4761}
4762
4763static void mem_cgroup_clear_mc(void)
4764{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004765 /*
4766 * we must clear moving_task before waking up waiters at the end of
4767 * task migration.
4768 */
4769 mc.moving_task = NULL;
4770 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004771 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004772 mc.from = NULL;
4773 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004774 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004775}
4776
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004777static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004778{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004779 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004780 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004781 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004782 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004783 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004784 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004785 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004786
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004787 /* charge immigration isn't supported on the default hierarchy */
4788 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004789 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004790
Tejun Heo4530edd2015-09-11 15:00:19 -04004791 /*
4792 * Multi-process migrations only happen on the default hierarchy
4793 * where charge immigration is not used. Perform charge
4794 * immigration if @tset contains a leader and whine if there are
4795 * multiple.
4796 */
4797 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004798 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004799 WARN_ON_ONCE(p);
4800 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004801 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004802 }
4803 if (!p)
4804 return 0;
4805
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004806 /*
4807 * We are now commited to this value whatever it is. Changes in this
4808 * tunable will only affect upcoming migrations, not the current one.
4809 * So we need to save it, and keep it going.
4810 */
4811 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4812 if (!move_flags)
4813 return 0;
4814
Tejun Heo9f2115f2015-09-08 15:01:10 -07004815 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004816
Tejun Heo9f2115f2015-09-08 15:01:10 -07004817 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004818
Tejun Heo9f2115f2015-09-08 15:01:10 -07004819 mm = get_task_mm(p);
4820 if (!mm)
4821 return 0;
4822 /* We move charges only when we move a owner of the mm */
4823 if (mm->owner == p) {
4824 VM_BUG_ON(mc.from);
4825 VM_BUG_ON(mc.to);
4826 VM_BUG_ON(mc.precharge);
4827 VM_BUG_ON(mc.moved_charge);
4828 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004829
Tejun Heo9f2115f2015-09-08 15:01:10 -07004830 spin_lock(&mc.lock);
4831 mc.from = from;
4832 mc.to = memcg;
4833 mc.flags = move_flags;
4834 spin_unlock(&mc.lock);
4835 /* We set mc.moving_task later */
4836
4837 ret = mem_cgroup_precharge_mc(mm);
4838 if (ret)
4839 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004840 }
Tejun Heo9f2115f2015-09-08 15:01:10 -07004841 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004842 return ret;
4843}
4844
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004845static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004846{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004847 if (mc.to)
4848 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004849}
4850
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004851static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4852 unsigned long addr, unsigned long end,
4853 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004854{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004855 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004856 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004857 pte_t *pte;
4858 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004859 enum mc_target_type target_type;
4860 union mc_target target;
4861 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004862
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -08004863 if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004864 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004865 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004866 return 0;
4867 }
4868 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4869 if (target_type == MC_TARGET_PAGE) {
4870 page = target.page;
4871 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004872 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004873 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004874 mc.precharge -= HPAGE_PMD_NR;
4875 mc.moved_charge += HPAGE_PMD_NR;
4876 }
4877 putback_lru_page(page);
4878 }
4879 put_page(page);
4880 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004881 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004882 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004883 }
4884
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004885 if (pmd_trans_unstable(pmd))
4886 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004887retry:
4888 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4889 for (; addr != end; addr += PAGE_SIZE) {
4890 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004891 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004892
4893 if (!mc.precharge)
4894 break;
4895
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004896 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004897 case MC_TARGET_PAGE:
4898 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004899 /*
4900 * We can have a part of the split pmd here. Moving it
4901 * can be done but it would be too convoluted so simply
4902 * ignore such a partial THP and keep it in original
4903 * memcg. There should be somebody mapping the head.
4904 */
4905 if (PageTransCompound(page))
4906 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004907 if (isolate_lru_page(page))
4908 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004909 if (!mem_cgroup_move_account(page, false,
4910 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004911 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004912 /* we uncharge from mc.from later. */
4913 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004914 }
4915 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004916put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004917 put_page(page);
4918 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004919 case MC_TARGET_SWAP:
4920 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004921 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004922 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004923 /* we fixup refcnts and charges later. */
4924 mc.moved_swap++;
4925 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004926 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004927 default:
4928 break;
4929 }
4930 }
4931 pte_unmap_unlock(pte - 1, ptl);
4932 cond_resched();
4933
4934 if (addr != end) {
4935 /*
4936 * We have consumed all precharges we got in can_attach().
4937 * We try charge one by one, but don't do any additional
4938 * charges to mc.to if we have failed in charge once in attach()
4939 * phase.
4940 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004941 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004942 if (!ret)
4943 goto retry;
4944 }
4945
4946 return ret;
4947}
4948
4949static void mem_cgroup_move_charge(struct mm_struct *mm)
4950{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004951 struct mm_walk mem_cgroup_move_charge_walk = {
4952 .pmd_entry = mem_cgroup_move_charge_pte_range,
4953 .mm = mm,
4954 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004955
4956 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004957 /*
4958 * Signal mem_cgroup_begin_page_stat() to take the memcg's
4959 * move_lock while we're moving its pages to another memcg.
4960 * Then wait for already started RCU-only updates to finish.
4961 */
4962 atomic_inc(&mc.from->moving_account);
4963 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004964retry:
4965 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4966 /*
4967 * Someone who are holding the mmap_sem might be waiting in
4968 * waitq. So we cancel all extra charges, wake up all waiters,
4969 * and retry. Because we cancel precharges, we might not be able
4970 * to move enough charges, but moving charge is a best-effort
4971 * feature anyway, so it wouldn't be a big problem.
4972 */
4973 __mem_cgroup_clear_mc();
4974 cond_resched();
4975 goto retry;
4976 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004977 /*
4978 * When we have consumed all precharges and failed in doing
4979 * additional charge, the page walk just aborts.
4980 */
4981 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004982 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004983 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004984}
4985
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004986static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08004987{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004988 struct cgroup_subsys_state *css;
4989 struct task_struct *p = cgroup_taskset_first(tset, &css);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004990 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004991
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004992 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004993 if (mc.to)
4994 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004995 mmput(mm);
4996 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004997 if (mc.to)
4998 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08004999}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005000#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005001static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005002{
5003 return 0;
5004}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005005static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005006{
5007}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005008static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005009{
5010}
5011#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005012
Tejun Heof00baae2013-04-15 13:41:15 -07005013/*
5014 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005015 * to verify whether we're attached to the default hierarchy on each mount
5016 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005017 */
Tejun Heoeb954192013-08-08 20:11:23 -04005018static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005019{
5020 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005021 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005022 * guarantees that @root doesn't have any children, so turning it
5023 * on for the root memcg is enough.
5024 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005025 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07005026 root_mem_cgroup->use_hierarchy = true;
5027 else
5028 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005029}
5030
Johannes Weiner241994ed2015-02-11 15:26:06 -08005031static u64 memory_current_read(struct cgroup_subsys_state *css,
5032 struct cftype *cft)
5033{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005034 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5035
5036 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005037}
5038
5039static int memory_low_show(struct seq_file *m, void *v)
5040{
5041 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005042 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005043
5044 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005045 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005046 else
5047 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5048
5049 return 0;
5050}
5051
5052static ssize_t memory_low_write(struct kernfs_open_file *of,
5053 char *buf, size_t nbytes, loff_t off)
5054{
5055 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5056 unsigned long low;
5057 int err;
5058
5059 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005060 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005061 if (err)
5062 return err;
5063
5064 memcg->low = low;
5065
5066 return nbytes;
5067}
5068
5069static int memory_high_show(struct seq_file *m, void *v)
5070{
5071 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005072 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005073
5074 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005075 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005076 else
5077 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5078
5079 return 0;
5080}
5081
5082static ssize_t memory_high_write(struct kernfs_open_file *of,
5083 char *buf, size_t nbytes, loff_t off)
5084{
5085 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5086 unsigned long high;
5087 int err;
5088
5089 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005090 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005091 if (err)
5092 return err;
5093
5094 memcg->high = high;
5095
Tejun Heo2529bb32015-05-22 18:23:34 -04005096 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005097 return nbytes;
5098}
5099
5100static int memory_max_show(struct seq_file *m, void *v)
5101{
5102 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005103 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005104
5105 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005106 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005107 else
5108 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5109
5110 return 0;
5111}
5112
5113static ssize_t memory_max_write(struct kernfs_open_file *of,
5114 char *buf, size_t nbytes, loff_t off)
5115{
5116 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5117 unsigned long max;
5118 int err;
5119
5120 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005121 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005122 if (err)
5123 return err;
5124
5125 err = mem_cgroup_resize_limit(memcg, max);
5126 if (err)
5127 return err;
5128
Tejun Heo2529bb32015-05-22 18:23:34 -04005129 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005130 return nbytes;
5131}
5132
5133static int memory_events_show(struct seq_file *m, void *v)
5134{
5135 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5136
5137 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5138 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5139 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5140 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5141
5142 return 0;
5143}
5144
5145static struct cftype memory_files[] = {
5146 {
5147 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005148 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005149 .read_u64 = memory_current_read,
5150 },
5151 {
5152 .name = "low",
5153 .flags = CFTYPE_NOT_ON_ROOT,
5154 .seq_show = memory_low_show,
5155 .write = memory_low_write,
5156 },
5157 {
5158 .name = "high",
5159 .flags = CFTYPE_NOT_ON_ROOT,
5160 .seq_show = memory_high_show,
5161 .write = memory_high_write,
5162 },
5163 {
5164 .name = "max",
5165 .flags = CFTYPE_NOT_ON_ROOT,
5166 .seq_show = memory_max_show,
5167 .write = memory_max_write,
5168 },
5169 {
5170 .name = "events",
5171 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005172 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005173 .seq_show = memory_events_show,
5174 },
5175 { } /* terminate */
5176};
5177
Tejun Heo073219e2014-02-08 10:36:58 -05005178struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005179 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005180 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005181 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005182 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005183 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005184 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005185 .can_attach = mem_cgroup_can_attach,
5186 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005187 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005188 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005189 .dfl_cftypes = memory_files,
5190 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005191 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005192};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005193
Johannes Weiner241994ed2015-02-11 15:26:06 -08005194/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005195 * mem_cgroup_low - check if memory consumption is below the normal range
5196 * @root: the highest ancestor to consider
5197 * @memcg: the memory cgroup to check
5198 *
5199 * Returns %true if memory consumption of @memcg, and that of all
5200 * configurable ancestors up to @root, is below the normal range.
5201 */
5202bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5203{
5204 if (mem_cgroup_disabled())
5205 return false;
5206
5207 /*
5208 * The toplevel group doesn't have a configurable range, so
5209 * it's never low when looked at directly, and it is not
5210 * considered an ancestor when assessing the hierarchy.
5211 */
5212
5213 if (memcg == root_mem_cgroup)
5214 return false;
5215
Michal Hocko4e54ded2015-02-27 15:51:46 -08005216 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005217 return false;
5218
5219 while (memcg != root) {
5220 memcg = parent_mem_cgroup(memcg);
5221
5222 if (memcg == root_mem_cgroup)
5223 break;
5224
Michal Hocko4e54ded2015-02-27 15:51:46 -08005225 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005226 return false;
5227 }
5228 return true;
5229}
5230
Johannes Weiner00501b52014-08-08 14:19:20 -07005231/**
5232 * mem_cgroup_try_charge - try charging a page
5233 * @page: page to charge
5234 * @mm: mm context of the victim
5235 * @gfp_mask: reclaim mode
5236 * @memcgp: charged memcg return
5237 *
5238 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5239 * pages according to @gfp_mask if necessary.
5240 *
5241 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5242 * Otherwise, an error code is returned.
5243 *
5244 * After page->mapping has been set up, the caller must finalize the
5245 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5246 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5247 */
5248int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005249 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5250 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005251{
5252 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005253 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005254 int ret = 0;
5255
5256 if (mem_cgroup_disabled())
5257 goto out;
5258
5259 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005260 /*
5261 * Every swap fault against a single page tries to charge the
5262 * page, bail as early as possible. shmem_unuse() encounters
5263 * already charged pages, too. The USED bit is protected by
5264 * the page lock, which serializes swap cache removal, which
5265 * in turn serializes uncharging.
5266 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005267 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005268 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005269 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005270
Johannes Weiner7941d212016-01-14 15:21:23 -08005271 if (do_memsw_account()) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005272 swp_entry_t ent = { .val = page_private(page), };
5273 unsigned short id = lookup_swap_cgroup_id(ent);
5274
5275 rcu_read_lock();
5276 memcg = mem_cgroup_from_id(id);
5277 if (memcg && !css_tryget_online(&memcg->css))
5278 memcg = NULL;
5279 rcu_read_unlock();
5280 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005281 }
5282
Johannes Weiner00501b52014-08-08 14:19:20 -07005283 if (!memcg)
5284 memcg = get_mem_cgroup_from_mm(mm);
5285
5286 ret = try_charge(memcg, gfp_mask, nr_pages);
5287
5288 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005289out:
5290 *memcgp = memcg;
5291 return ret;
5292}
5293
5294/**
5295 * mem_cgroup_commit_charge - commit a page charge
5296 * @page: page to charge
5297 * @memcg: memcg to charge the page to
5298 * @lrucare: page might be on LRU already
5299 *
5300 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5301 * after page->mapping has been set up. This must happen atomically
5302 * as part of the page instantiation, i.e. under the page table lock
5303 * for anonymous pages, under the page lock for page and swap cache.
5304 *
5305 * In addition, the page must not be on the LRU during the commit, to
5306 * prevent racing with task migration. If it might be, use @lrucare.
5307 *
5308 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5309 */
5310void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005311 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005312{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005313 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005314
5315 VM_BUG_ON_PAGE(!page->mapping, page);
5316 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5317
5318 if (mem_cgroup_disabled())
5319 return;
5320 /*
5321 * Swap faults will attempt to charge the same page multiple
5322 * times. But reuse_swap_page() might have removed the page
5323 * from swapcache already, so we can't check PageSwapCache().
5324 */
5325 if (!memcg)
5326 return;
5327
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005328 commit_charge(page, memcg, lrucare);
5329
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005330 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005331 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005332 memcg_check_events(memcg, page);
5333 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005334
Johannes Weiner7941d212016-01-14 15:21:23 -08005335 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005336 swp_entry_t entry = { .val = page_private(page) };
5337 /*
5338 * The swap entry might not get freed for a long time,
5339 * let's not wait for it. The page already received a
5340 * memory+swap charge, drop the swap entry duplicate.
5341 */
5342 mem_cgroup_uncharge_swap(entry);
5343 }
5344}
5345
5346/**
5347 * mem_cgroup_cancel_charge - cancel a page charge
5348 * @page: page to charge
5349 * @memcg: memcg to charge the page to
5350 *
5351 * Cancel a charge transaction started by mem_cgroup_try_charge().
5352 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005353void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5354 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005355{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005356 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005357
5358 if (mem_cgroup_disabled())
5359 return;
5360 /*
5361 * Swap faults will attempt to charge the same page multiple
5362 * times. But reuse_swap_page() might have removed the page
5363 * from swapcache already, so we can't check PageSwapCache().
5364 */
5365 if (!memcg)
5366 return;
5367
Johannes Weiner00501b52014-08-08 14:19:20 -07005368 cancel_charge(memcg, nr_pages);
5369}
5370
Johannes Weiner747db952014-08-08 14:19:24 -07005371static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005372 unsigned long nr_anon, unsigned long nr_file,
5373 unsigned long nr_huge, struct page *dummy_page)
5374{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005375 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005376 unsigned long flags;
5377
Johannes Weinerce00a962014-09-05 08:43:57 -04005378 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005379 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005380 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005381 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005382 memcg_oom_recover(memcg);
5383 }
Johannes Weiner747db952014-08-08 14:19:24 -07005384
5385 local_irq_save(flags);
5386 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5387 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5388 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5389 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005390 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005391 memcg_check_events(memcg, dummy_page);
5392 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005393
5394 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005395 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005396}
5397
5398static void uncharge_list(struct list_head *page_list)
5399{
5400 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005401 unsigned long nr_anon = 0;
5402 unsigned long nr_file = 0;
5403 unsigned long nr_huge = 0;
5404 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005405 struct list_head *next;
5406 struct page *page;
5407
5408 next = page_list->next;
5409 do {
5410 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005411
5412 page = list_entry(next, struct page, lru);
5413 next = page->lru.next;
5414
5415 VM_BUG_ON_PAGE(PageLRU(page), page);
5416 VM_BUG_ON_PAGE(page_count(page), page);
5417
Johannes Weiner1306a852014-12-10 15:44:52 -08005418 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005419 continue;
5420
5421 /*
5422 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005423 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005424 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005425 */
5426
Johannes Weiner1306a852014-12-10 15:44:52 -08005427 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005428 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005429 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5430 nr_huge, page);
5431 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005432 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005433 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005434 }
5435
5436 if (PageTransHuge(page)) {
5437 nr_pages <<= compound_order(page);
5438 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5439 nr_huge += nr_pages;
5440 }
5441
5442 if (PageAnon(page))
5443 nr_anon += nr_pages;
5444 else
5445 nr_file += nr_pages;
5446
Johannes Weiner1306a852014-12-10 15:44:52 -08005447 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005448
5449 pgpgout++;
5450 } while (next != page_list);
5451
5452 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005453 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5454 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005455}
5456
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005457/**
5458 * mem_cgroup_uncharge - uncharge a page
5459 * @page: page to uncharge
5460 *
5461 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5462 * mem_cgroup_commit_charge().
5463 */
5464void mem_cgroup_uncharge(struct page *page)
5465{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005466 if (mem_cgroup_disabled())
5467 return;
5468
Johannes Weiner747db952014-08-08 14:19:24 -07005469 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005470 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005471 return;
5472
Johannes Weiner747db952014-08-08 14:19:24 -07005473 INIT_LIST_HEAD(&page->lru);
5474 uncharge_list(&page->lru);
5475}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005476
Johannes Weiner747db952014-08-08 14:19:24 -07005477/**
5478 * mem_cgroup_uncharge_list - uncharge a list of page
5479 * @page_list: list of pages to uncharge
5480 *
5481 * Uncharge a list of pages previously charged with
5482 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5483 */
5484void mem_cgroup_uncharge_list(struct list_head *page_list)
5485{
5486 if (mem_cgroup_disabled())
5487 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005488
Johannes Weiner747db952014-08-08 14:19:24 -07005489 if (!list_empty(page_list))
5490 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005491}
5492
5493/**
Hugh Dickins45637ba2015-11-05 18:49:40 -08005494 * mem_cgroup_replace_page - migrate a charge to another page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005495 * @oldpage: currently charged page
5496 * @newpage: page to transfer the charge to
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005497 *
5498 * Migrate the charge from @oldpage to @newpage.
5499 *
5500 * Both pages must be locked, @newpage->mapping must be set up.
Hugh Dickins25be6a62015-12-11 13:40:40 -08005501 * Either or both pages might be on the LRU already.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005502 */
Hugh Dickins45637ba2015-11-05 18:49:40 -08005503void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005504{
Johannes Weiner29833312014-12-10 15:44:02 -08005505 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005506 int isolated;
5507
5508 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5509 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005510 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005511 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5512 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005513
5514 if (mem_cgroup_disabled())
5515 return;
5516
5517 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005518 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005519 return;
5520
Hugh Dickins45637ba2015-11-05 18:49:40 -08005521 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005522 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005523 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005524 return;
5525
Hugh Dickins45637ba2015-11-05 18:49:40 -08005526 lock_page_lru(oldpage, &isolated);
Johannes Weiner1306a852014-12-10 15:44:52 -08005527 oldpage->mem_cgroup = NULL;
Hugh Dickins45637ba2015-11-05 18:49:40 -08005528 unlock_page_lru(oldpage, isolated);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005529
Hugh Dickins45637ba2015-11-05 18:49:40 -08005530 commit_charge(newpage, memcg, true);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005531}
5532
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005533#ifdef CONFIG_INET
Johannes Weiner11092082016-01-14 15:21:26 -08005534
Johannes Weineref129472016-01-14 15:21:34 -08005535DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005536EXPORT_SYMBOL(memcg_sockets_enabled_key);
5537
5538void sock_update_memcg(struct sock *sk)
5539{
5540 struct mem_cgroup *memcg;
5541
5542 /* Socket cloning can throw us here with sk_cgrp already
5543 * filled. It won't however, necessarily happen from
5544 * process context. So the test for root memcg given
5545 * the current task's memcg won't help us in this case.
5546 *
5547 * Respecting the original socket's memcg is a better
5548 * decision in this case.
5549 */
5550 if (sk->sk_memcg) {
5551 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5552 css_get(&sk->sk_memcg->css);
5553 return;
5554 }
5555
5556 rcu_read_lock();
5557 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005558 if (memcg == root_mem_cgroup)
5559 goto out;
5560#ifdef CONFIG_MEMCG_KMEM
5561 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcp_mem.active)
5562 goto out;
5563#endif
5564 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005565 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005566out:
Johannes Weiner11092082016-01-14 15:21:26 -08005567 rcu_read_unlock();
5568}
5569EXPORT_SYMBOL(sock_update_memcg);
5570
5571void sock_release_memcg(struct sock *sk)
5572{
5573 WARN_ON(!sk->sk_memcg);
5574 css_put(&sk->sk_memcg->css);
5575}
5576
5577/**
5578 * mem_cgroup_charge_skmem - charge socket memory
5579 * @memcg: memcg to charge
5580 * @nr_pages: number of pages to charge
5581 *
5582 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5583 * @memcg's configured limit, %false if the charge had to be forced.
5584 */
5585bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5586{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005587 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005588
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005589#ifdef CONFIG_MEMCG_KMEM
5590 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5591 struct page_counter *counter;
5592
5593 if (page_counter_try_charge(&memcg->tcp_mem.memory_allocated,
5594 nr_pages, &counter)) {
5595 memcg->tcp_mem.memory_pressure = 0;
5596 return true;
5597 }
5598 page_counter_charge(&memcg->tcp_mem.memory_allocated, nr_pages);
5599 memcg->tcp_mem.memory_pressure = 1;
5600 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005601 }
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005602#endif
5603 /* Don't block in the packet receive path */
5604 if (in_softirq())
5605 gfp_mask = GFP_NOWAIT;
5606
5607 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5608 return true;
5609
5610 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005611 return false;
5612}
5613
5614/**
5615 * mem_cgroup_uncharge_skmem - uncharge socket memory
5616 * @memcg - memcg to uncharge
5617 * @nr_pages - number of pages to uncharge
5618 */
5619void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5620{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005621#ifdef CONFIG_MEMCG_KMEM
5622 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5623 page_counter_uncharge(&memcg->tcp_mem.memory_allocated,
5624 nr_pages);
5625 return;
5626 }
5627#endif
5628 page_counter_uncharge(&memcg->memory, nr_pages);
5629 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005630}
5631
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005632#endif /* CONFIG_INET */
5633
5634static int __init cgroup_memory(char *s)
5635{
5636 char *token;
5637
5638 while ((token = strsep(&s, ",")) != NULL) {
5639 if (!*token)
5640 continue;
5641 if (!strcmp(token, "nosocket"))
5642 cgroup_memory_nosocket = true;
5643 }
5644 return 0;
5645}
5646__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005647
Michal Hocko2d110852013-02-22 16:34:43 -08005648/*
Michal Hocko10813122013-02-22 16:35:41 -08005649 * subsys_initcall() for memory controller.
5650 *
5651 * Some parts like hotcpu_notifier() have to be initialized from this context
5652 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5653 * everything that doesn't depend on a specific mem_cgroup structure should
5654 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005655 */
5656static int __init mem_cgroup_init(void)
5657{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005658 int cpu, node;
5659
Michal Hocko2d110852013-02-22 16:34:43 -08005660 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005661
5662 for_each_possible_cpu(cpu)
5663 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5664 drain_local_stock);
5665
5666 for_each_node(node) {
5667 struct mem_cgroup_tree_per_node *rtpn;
5668 int zone;
5669
5670 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5671 node_online(node) ? node : NUMA_NO_NODE);
5672
5673 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5674 struct mem_cgroup_tree_per_zone *rtpz;
5675
5676 rtpz = &rtpn->rb_tree_per_zone[zone];
5677 rtpz->rb_root = RB_ROOT;
5678 spin_lock_init(&rtpz->lock);
5679 }
5680 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5681 }
5682
Michal Hocko2d110852013-02-22 16:34:43 -08005683 return 0;
5684}
5685subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005686
5687#ifdef CONFIG_MEMCG_SWAP
5688/**
5689 * mem_cgroup_swapout - transfer a memsw charge to swap
5690 * @page: page whose memsw charge to transfer
5691 * @entry: swap entry to move the charge to
5692 *
5693 * Transfer the memsw charge of @page to @entry.
5694 */
5695void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5696{
5697 struct mem_cgroup *memcg;
5698 unsigned short oldid;
5699
5700 VM_BUG_ON_PAGE(PageLRU(page), page);
5701 VM_BUG_ON_PAGE(page_count(page), page);
5702
Johannes Weiner7941d212016-01-14 15:21:23 -08005703 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005704 return;
5705
5706 memcg = page->mem_cgroup;
5707
5708 /* Readahead page, never charged */
5709 if (!memcg)
5710 return;
5711
5712 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5713 VM_BUG_ON_PAGE(oldid, page);
5714 mem_cgroup_swap_statistics(memcg, true);
5715
5716 page->mem_cgroup = NULL;
5717
5718 if (!mem_cgroup_is_root(memcg))
5719 page_counter_uncharge(&memcg->memory, 1);
5720
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005721 /*
5722 * Interrupts should be disabled here because the caller holds the
5723 * mapping->tree_lock lock which is taken with interrupts-off. It is
5724 * important here to have the interrupts disabled because it is the
5725 * only synchronisation we have for udpating the per-CPU variables.
5726 */
5727 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005728 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005729 memcg_check_events(memcg, page);
5730}
5731
5732/**
5733 * mem_cgroup_uncharge_swap - uncharge a swap entry
5734 * @entry: swap entry to uncharge
5735 *
5736 * Drop the memsw charge associated with @entry.
5737 */
5738void mem_cgroup_uncharge_swap(swp_entry_t entry)
5739{
5740 struct mem_cgroup *memcg;
5741 unsigned short id;
5742
Johannes Weiner7941d212016-01-14 15:21:23 -08005743 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005744 return;
5745
5746 id = swap_cgroup_record(entry, 0);
5747 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005748 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005749 if (memcg) {
5750 if (!mem_cgroup_is_root(memcg))
5751 page_counter_uncharge(&memcg->memsw, 1);
5752 mem_cgroup_swap_statistics(memcg, false);
5753 css_put(&memcg->css);
5754 }
5755 rcu_read_unlock();
5756}
5757
5758/* for remember boot option*/
5759#ifdef CONFIG_MEMCG_SWAP_ENABLED
5760static int really_do_swap_account __initdata = 1;
5761#else
5762static int really_do_swap_account __initdata;
5763#endif
5764
5765static int __init enable_swap_account(char *s)
5766{
5767 if (!strcmp(s, "1"))
5768 really_do_swap_account = 1;
5769 else if (!strcmp(s, "0"))
5770 really_do_swap_account = 0;
5771 return 1;
5772}
5773__setup("swapaccount=", enable_swap_account);
5774
5775static struct cftype memsw_cgroup_files[] = {
5776 {
5777 .name = "memsw.usage_in_bytes",
5778 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5779 .read_u64 = mem_cgroup_read_u64,
5780 },
5781 {
5782 .name = "memsw.max_usage_in_bytes",
5783 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5784 .write = mem_cgroup_reset,
5785 .read_u64 = mem_cgroup_read_u64,
5786 },
5787 {
5788 .name = "memsw.limit_in_bytes",
5789 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5790 .write = mem_cgroup_write,
5791 .read_u64 = mem_cgroup_read_u64,
5792 },
5793 {
5794 .name = "memsw.failcnt",
5795 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5796 .write = mem_cgroup_reset,
5797 .read_u64 = mem_cgroup_read_u64,
5798 },
5799 { }, /* terminate */
5800};
5801
5802static int __init mem_cgroup_swap_init(void)
5803{
5804 if (!mem_cgroup_disabled() && really_do_swap_account) {
5805 do_swap_account = 1;
5806 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5807 memsw_cgroup_files));
5808 }
5809 return 0;
5810}
5811subsys_initcall(mem_cgroup_swap_init);
5812
5813#endif /* CONFIG_MEMCG_SWAP */