blob: a93fac288cca4a2d68efa24a433d6042e89d1985 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heoc2644f52015-11-05 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
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070079#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070080static struct mem_cgroup *root_mem_cgroup __read_mostly;
Tejun Heodbee2272015-05-22 17:13:20 -040081struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080082
Johannes Weiner21afa382015-02-11 15:26:36 -080083/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070084#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080086#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070087#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088#endif
89
Johannes Weineraf7c4b02012-05-29 15:07:08 -070090static const char * const mem_cgroup_stat_names[] = {
91 "cache",
92 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070093 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070094 "mapped_file",
Greg Thelen7c9d3ff2015-05-22 17:13:16 -040095 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070096 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070097 "swap",
98};
99
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700100static const char * const mem_cgroup_events_names[] = {
101 "pgpgin",
102 "pgpgout",
103 "pgfault",
104 "pgmajfault",
105};
106
Sha Zhengju58cf1882013-02-22 16:32:05 -0800107static const char * const mem_cgroup_lru_names[] = {
108 "inactive_anon",
109 "active_anon",
110 "inactive_file",
111 "active_file",
112 "unevictable",
113};
114
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700115#define THRESHOLDS_EVENTS_TARGET 128
116#define SOFTLIMIT_EVENTS_TARGET 1024
117#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700118
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700119/*
120 * Cgroups above their limits are maintained in a RB-Tree, independent of
121 * their hierarchy representation
122 */
123
124struct mem_cgroup_tree_per_zone {
125 struct rb_root rb_root;
126 spinlock_t lock;
127};
128
129struct mem_cgroup_tree_per_node {
130 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
131};
132
133struct mem_cgroup_tree {
134 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
135};
136
137static struct mem_cgroup_tree soft_limit_tree __read_mostly;
138
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700139/* for OOM */
140struct mem_cgroup_eventfd_list {
141 struct list_head list;
142 struct eventfd_ctx *eventfd;
143};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800144
Tejun Heo79bd9812013-11-22 18:20:42 -0500145/*
146 * cgroup_event represents events which userspace want to receive.
147 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500148struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500149 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500150 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500151 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500152 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500153 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500154 * eventfd to signal userspace about the event.
155 */
156 struct eventfd_ctx *eventfd;
157 /*
158 * Each of these stored in a list by the cgroup.
159 */
160 struct list_head list;
161 /*
Tejun Heofba94802013-11-22 18:20:43 -0500162 * register_event() callback will be used to add new userspace
163 * waiter for changes related to this event. Use eventfd_signal()
164 * on eventfd to send notification to userspace.
165 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500166 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500167 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500168 /*
169 * unregister_event() callback will be called when userspace closes
170 * the eventfd or on cgroup removing. This callback must be set,
171 * if you want provide notification functionality.
172 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500173 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500174 struct eventfd_ctx *eventfd);
175 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500176 * All fields below needed to unregister event when
177 * userspace closes eventfd.
178 */
179 poll_table pt;
180 wait_queue_head_t *wqh;
181 wait_queue_t wait;
182 struct work_struct remove;
183};
184
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700185static void mem_cgroup_threshold(struct mem_cgroup *memcg);
186static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800187
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800188/* Stuffs for move charges at task migration. */
189/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800190 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800191 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800192#define MOVE_ANON 0x1U
193#define MOVE_FILE 0x2U
194#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800195
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800196/* "mc" and its members are protected by cgroup_mutex */
197static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800198 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800199 struct mem_cgroup *from;
200 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800202 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800203 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800204 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800205 struct task_struct *moving_task; /* a task moving charges */
206 wait_queue_head_t waitq; /* a waitq for other context */
207} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700208 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800209 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
210};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800211
Balbir Singh4e416952009-09-23 15:56:39 -0700212/*
213 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
214 * limit reclaim to prevent infinite loops, if they ever occur.
215 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700216#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700217#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700218
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800219enum charge_type {
220 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700221 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800222 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700223 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700224 NR_CHARGE_TYPE,
225};
226
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800227/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800228enum res_type {
229 _MEM,
230 _MEMSWAP,
231 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800232 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800233};
234
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700235#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
236#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800237#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700238/* Used for OOM nofiier */
239#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800240
Balbir Singh75822b42009-09-23 15:56:38 -0700241/*
Glauber Costa09998212013-02-22 16:34:55 -0800242 * The memcg_create_mutex will be held whenever a new cgroup is created.
243 * As a consequence, any change that needs to protect against new child cgroups
244 * appearing has to hold it as well.
245 */
246static DEFINE_MUTEX(memcg_create_mutex);
247
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700248/* Some nice accessors for the vmpressure. */
249struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
250{
251 if (!memcg)
252 memcg = root_mem_cgroup;
253 return &memcg->vmpressure;
254}
255
256struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
257{
258 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
259}
260
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700261static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
262{
263 return (memcg == root_mem_cgroup);
264}
265
Li Zefan4219b2d2013-09-23 16:56:29 +0800266/*
267 * We restrict the id in the range of [1, 65535], so it can fit into
268 * an unsigned short.
269 */
270#define MEM_CGROUP_ID_MAX USHRT_MAX
271
Li Zefan34c00c32013-09-23 16:56:01 +0800272static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
273{
Tejun Heo15a4c832014-05-04 15:09:14 -0400274 return memcg->css.id;
Li Zefan34c00c32013-09-23 16:56:01 +0800275}
276
Vladimir Davydovadbe4272015-04-15 16:13:00 -0700277/*
278 * A helper function to get mem_cgroup from ID. must be called under
279 * rcu_read_lock(). The caller is responsible for calling
280 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
281 * refcnt from swap can be called against removed memcg.)
282 */
Li Zefan34c00c32013-09-23 16:56:01 +0800283static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
284{
285 struct cgroup_subsys_state *css;
286
Tejun Heo7d699dd2014-05-04 15:09:13 -0400287 css = css_from_id(id, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800288 return mem_cgroup_from_css(css);
289}
290
Glauber Costae1aab162011-12-11 21:47:03 +0000291/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700292#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000293
Glauber Costae1aab162011-12-11 21:47:03 +0000294void sock_update_memcg(struct sock *sk)
295{
Glauber Costa376be5f2012-01-20 04:57:14 +0000296 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000297 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700298 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000299
300 BUG_ON(!sk->sk_prot->proto_cgroup);
301
Glauber Costaf3f511e2012-01-05 20:16:39 +0000302 /* Socket cloning can throw us here with sk_cgrp already
303 * filled. It won't however, necessarily happen from
304 * process context. So the test for root memcg given
305 * the current task's memcg won't help us in this case.
306 *
307 * Respecting the original socket's memcg is a better
308 * decision in this case.
309 */
310 if (sk->sk_cgrp) {
311 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700312 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000313 return;
314 }
315
Glauber Costae1aab162011-12-11 21:47:03 +0000316 rcu_read_lock();
317 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700318 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Michal Hocko6b34b852015-09-08 15:01:16 -0700319 if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) &&
Tejun Heoec903c02014-05-13 12:11:01 -0400320 css_tryget_online(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700321 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000322 }
323 rcu_read_unlock();
324 }
325}
326EXPORT_SYMBOL(sock_update_memcg);
327
328void sock_release_memcg(struct sock *sk)
329{
Glauber Costa376be5f2012-01-20 04:57:14 +0000330 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000331 struct mem_cgroup *memcg;
332 WARN_ON(!sk->sk_cgrp->memcg);
333 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700334 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000335 }
336}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000337
338struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
339{
340 if (!memcg || mem_cgroup_is_root(memcg))
341 return NULL;
342
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700343 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000344}
345EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000346
Glauber Costa3f134612012-05-29 15:07:11 -0700347#endif
348
Glauber Costaa8964b92012-12-18 14:22:09 -0800349#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800350/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800351 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800352 * The main reason for not using cgroup id for this:
353 * this works better in sparse environments, where we have a lot of memcgs,
354 * but only a few kmem-limited. Or also, if we have, for instance, 200
355 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
356 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800357 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800358 * The current size of the caches array is stored in memcg_nr_cache_ids. It
359 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800360 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800361static DEFINE_IDA(memcg_cache_ida);
362int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800363
Vladimir Davydov05257a12015-02-12 14:59:01 -0800364/* Protects memcg_nr_cache_ids */
365static DECLARE_RWSEM(memcg_cache_ids_sem);
366
367void memcg_get_cache_ids(void)
368{
369 down_read(&memcg_cache_ids_sem);
370}
371
372void memcg_put_cache_ids(void)
373{
374 up_read(&memcg_cache_ids_sem);
375}
376
Glauber Costa55007d82012-12-18 14:22:38 -0800377/*
378 * MIN_SIZE is different than 1, because we would like to avoid going through
379 * the alloc/free process all the time. In a small machine, 4 kmem-limited
380 * cgroups is a reasonable guess. In the future, it could be a parameter or
381 * tunable, but that is strictly not necessary.
382 *
Li Zefanb8627832013-09-23 16:56:47 +0800383 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800384 * this constant directly from cgroup, but it is understandable that this is
385 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800386 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800387 * increase ours as well if it increases.
388 */
389#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800390#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800391
Glauber Costad7f25f82012-12-18 14:22:40 -0800392/*
393 * A lot of the calls to the cache allocation functions are expected to be
394 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
395 * conditional to this static branch, we'll have to allow modules that does
396 * kmem_cache_alloc and the such to see this symbol as well
397 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800398struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800399EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800400
Glauber Costaa8964b92012-12-18 14:22:09 -0800401#endif /* CONFIG_MEMCG_KMEM */
402
Balbir Singhf64c3f52009-09-23 15:56:37 -0700403static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700404mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700405{
Jianyu Zhane2318752014-06-06 14:38:20 -0700406 int nid = zone_to_nid(zone);
407 int zid = zone_idx(zone);
408
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700409 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700410}
411
Tejun Heoe4487942015-05-27 20:00:02 -0400412/**
413 * mem_cgroup_css_from_page - css of the memcg associated with a page
414 * @page: page of interest
415 *
416 * If memcg is bound to the default hierarchy, css of the memcg associated
417 * with @page is returned. The returned css remains associated with @page
418 * until it is released.
419 *
420 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
421 * is returned.
422 *
423 * XXX: The above description of behavior on the default hierarchy isn't
424 * strictly true yet as replace_page_cache_page() can modify the
425 * association before @page is released even on the default hierarchy;
426 * however, the current and planned usages don't mix the the two functions
427 * and replace_page_cache_page() will soon be updated to make the invariant
428 * actually true.
429 */
430struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
431{
432 struct mem_cgroup *memcg;
433
434 rcu_read_lock();
435
436 memcg = page->mem_cgroup;
437
Tejun Heoded8e212015-09-18 11:56:28 -0400438 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoe4487942015-05-27 20:00:02 -0400439 memcg = root_mem_cgroup;
440
441 rcu_read_unlock();
442 return &memcg->css;
443}
444
Balbir Singhf64c3f52009-09-23 15:56:37 -0700445static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700446mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700447{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700448 int nid = page_to_nid(page);
449 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700450
Jianyu Zhane2318752014-06-06 14:38:20 -0700451 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700452}
453
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700454static struct mem_cgroup_tree_per_zone *
455soft_limit_tree_node_zone(int nid, int zid)
456{
457 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
458}
459
460static struct mem_cgroup_tree_per_zone *
461soft_limit_tree_from_page(struct page *page)
462{
463 int nid = page_to_nid(page);
464 int zid = page_zonenum(page);
465
466 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
467}
468
Johannes Weinercf2c8122014-06-06 14:38:21 -0700469static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
470 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800471 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472{
473 struct rb_node **p = &mctz->rb_root.rb_node;
474 struct rb_node *parent = NULL;
475 struct mem_cgroup_per_zone *mz_node;
476
477 if (mz->on_tree)
478 return;
479
480 mz->usage_in_excess = new_usage_in_excess;
481 if (!mz->usage_in_excess)
482 return;
483 while (*p) {
484 parent = *p;
485 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
486 tree_node);
487 if (mz->usage_in_excess < mz_node->usage_in_excess)
488 p = &(*p)->rb_left;
489 /*
490 * We can't avoid mem cgroups that are over their soft
491 * limit by the same amount
492 */
493 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
494 p = &(*p)->rb_right;
495 }
496 rb_link_node(&mz->tree_node, parent, p);
497 rb_insert_color(&mz->tree_node, &mctz->rb_root);
498 mz->on_tree = true;
499}
500
Johannes Weinercf2c8122014-06-06 14:38:21 -0700501static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
502 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700503{
504 if (!mz->on_tree)
505 return;
506 rb_erase(&mz->tree_node, &mctz->rb_root);
507 mz->on_tree = false;
508}
509
Johannes Weinercf2c8122014-06-06 14:38:21 -0700510static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
511 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700512{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700513 unsigned long flags;
514
515 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700516 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700517 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700518}
519
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800520static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
521{
522 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700523 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800524 unsigned long excess = 0;
525
526 if (nr_pages > soft_limit)
527 excess = nr_pages - soft_limit;
528
529 return excess;
530}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700531
532static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
533{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800534 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700535 struct mem_cgroup_per_zone *mz;
536 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700537
Jianyu Zhane2318752014-06-06 14:38:20 -0700538 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700539 /*
540 * Necessary to update all ancestors when hierarchy is used.
541 * because their event counter is not touched.
542 */
543 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700544 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800545 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700546 /*
547 * We have to update the tree if mz is on RB-tree or
548 * mem is over its softlimit.
549 */
550 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700551 unsigned long flags;
552
553 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700554 /* if on-tree, remove it */
555 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700556 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700557 /*
558 * Insert again. mz->usage_in_excess will be updated.
559 * If excess is 0, no tree ops.
560 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700561 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700562 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700563 }
564 }
565}
566
567static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
568{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700569 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700570 struct mem_cgroup_per_zone *mz;
571 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700572
Jianyu Zhane2318752014-06-06 14:38:20 -0700573 for_each_node(nid) {
574 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
575 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
576 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700577 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700578 }
579 }
580}
581
582static struct mem_cgroup_per_zone *
583__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
584{
585 struct rb_node *rightmost = NULL;
586 struct mem_cgroup_per_zone *mz;
587
588retry:
589 mz = NULL;
590 rightmost = rb_last(&mctz->rb_root);
591 if (!rightmost)
592 goto done; /* Nothing to reclaim from */
593
594 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
595 /*
596 * Remove the node now but someone else can add it back,
597 * we will to add it back at the end of reclaim to its correct
598 * position in the tree.
599 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700600 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800601 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400602 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700603 goto retry;
604done:
605 return mz;
606}
607
608static struct mem_cgroup_per_zone *
609mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
610{
611 struct mem_cgroup_per_zone *mz;
612
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700613 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700614 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700615 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700616 return mz;
617}
618
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700619/*
620 * Implementation Note: reading percpu statistics for memcg.
621 *
622 * Both of vmstat[] and percpu_counter has threshold and do periodic
623 * synchronization to implement "quick" read. There are trade-off between
624 * reading cost and precision of value. Then, we may have a chance to implement
625 * a periodic synchronizion of counter in memcg's counter.
626 *
627 * But this _read() function is used for user interface now. The user accounts
628 * memory usage by memory cgroup and he _always_ requires exact value because
629 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
630 * have to visit all online cpus and make sum. So, for now, unnecessary
631 * synchronization is not implemented. (just implemented for cpu hotplug)
632 *
633 * If there are kernel internal actions which can make use of some not-exact
634 * value, and reading all cpu value can be performance bottleneck in some
635 * common workload, threashold and synchonization as vmstat[] should be
636 * implemented.
637 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700638static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700639 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800640{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700641 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800642 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800643
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400644 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700645 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800646 return val;
647}
648
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700649static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700650 enum mem_cgroup_events_index idx)
651{
652 unsigned long val = 0;
653 int cpu;
654
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400655 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700656 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700657 return val;
658}
659
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700660static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700661 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700662 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800663{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700664 /*
665 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
666 * counted as CACHE even if it's on ANON LRU.
667 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700668 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700669 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700670 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800671 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700672 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700673 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700674
David Rientjesb070e652013-05-07 16:18:09 -0700675 if (PageTransHuge(page))
676 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
677 nr_pages);
678
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800679 /* pagein of a big page is an event. So, ignore page size */
680 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700681 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800682 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700683 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800684 nr_pages = -nr_pages; /* for event */
685 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800686
Johannes Weiner13114712012-05-29 15:07:07 -0700687 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800688}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800689
Jianyu Zhane2318752014-06-06 14:38:20 -0700690static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
691 int nid,
692 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700693{
Jianyu Zhane2318752014-06-06 14:38:20 -0700694 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700695 int zid;
696
Jianyu Zhane2318752014-06-06 14:38:20 -0700697 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700698
Jianyu Zhane2318752014-06-06 14:38:20 -0700699 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
700 struct mem_cgroup_per_zone *mz;
701 enum lru_list lru;
702
703 for_each_lru(lru) {
704 if (!(BIT(lru) & lru_mask))
705 continue;
706 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
707 nr += mz->lru_size[lru];
708 }
709 }
710 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700711}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700712
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700713static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700714 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800715{
Jianyu Zhane2318752014-06-06 14:38:20 -0700716 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700717 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800718
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800719 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700720 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
721 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800722}
723
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800724static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
725 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800726{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700727 unsigned long val, next;
728
Johannes Weiner13114712012-05-29 15:07:07 -0700729 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700730 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700731 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800732 if ((long)next - (long)val < 0) {
733 switch (target) {
734 case MEM_CGROUP_TARGET_THRESH:
735 next = val + THRESHOLDS_EVENTS_TARGET;
736 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700737 case MEM_CGROUP_TARGET_SOFTLIMIT:
738 next = val + SOFTLIMIT_EVENTS_TARGET;
739 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800740 case MEM_CGROUP_TARGET_NUMAINFO:
741 next = val + NUMAINFO_EVENTS_TARGET;
742 break;
743 default:
744 break;
745 }
746 __this_cpu_write(memcg->stat->targets[target], next);
747 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700748 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800749 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800750}
751
752/*
753 * Check events in order.
754 *
755 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700756static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800757{
758 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800759 if (unlikely(mem_cgroup_event_ratelimit(memcg,
760 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700761 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800762 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800763
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700764 do_softlimit = mem_cgroup_event_ratelimit(memcg,
765 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700766#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800767 do_numainfo = mem_cgroup_event_ratelimit(memcg,
768 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700769#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800770 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700771 if (unlikely(do_softlimit))
772 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800773#if MAX_NUMNODES > 1
774 if (unlikely(do_numainfo))
775 atomic_inc(&memcg->numainfo_events);
776#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700777 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800778}
779
Balbir Singhcf475ad2008-04-29 01:00:16 -0700780struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800781{
Balbir Singh31a78f22008-09-28 23:09:31 +0100782 /*
783 * mm_update_next_owner() may clear mm->owner to NULL
784 * if it races with swapoff, page migration, etc.
785 * So this can be called with p == NULL.
786 */
787 if (unlikely(!p))
788 return NULL;
789
Tejun Heo073219e2014-02-08 10:36:58 -0500790 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800791}
Michal Hocko1e977cc2015-09-08 15:01:02 -0700792EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800793
Johannes Weinerdf381972014-04-07 15:37:43 -0700794static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800795{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700796 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700797
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800798 rcu_read_lock();
799 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700800 /*
801 * Page cache insertions can happen withou an
802 * actual mm context, e.g. during disk probing
803 * on boot, loopback IO, acct() writes etc.
804 */
805 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700806 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700807 else {
808 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
809 if (unlikely(!memcg))
810 memcg = root_mem_cgroup;
811 }
Tejun Heoec903c02014-05-13 12:11:01 -0400812 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800813 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700814 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800815}
816
Johannes Weiner56600482012-01-12 17:17:59 -0800817/**
818 * mem_cgroup_iter - iterate over memory cgroup hierarchy
819 * @root: hierarchy root
820 * @prev: previously returned memcg, NULL on first invocation
821 * @reclaim: cookie for shared reclaim walks, NULL for full walks
822 *
823 * Returns references to children of the hierarchy below @root, or
824 * @root itself, or %NULL after a full round-trip.
825 *
826 * Caller must pass the return value in @prev on subsequent
827 * invocations for reference counting, or use mem_cgroup_iter_break()
828 * to cancel a hierarchy walk before the round-trip is complete.
829 *
830 * Reclaimers can specify a zone and a priority level in @reclaim to
831 * divide up the memcgs in the hierarchy among all concurrent
832 * reclaimers operating on the same zone and priority.
833 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700834struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800835 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700836 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700837{
Michal Hocko1e977cc2015-09-08 15:01:02 -0700838 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800839 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800840 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800841 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700842
Andrew Morton694fbc02013-09-24 15:27:37 -0700843 if (mem_cgroup_disabled())
844 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800845
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700846 if (!root)
847 root = root_mem_cgroup;
848
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800849 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800850 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800851
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800852 if (!root->use_hierarchy && root != root_mem_cgroup) {
853 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800854 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700855 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800856 }
857
Michal Hocko542f85f2013-04-29 15:07:15 -0700858 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800859
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800860 if (reclaim) {
861 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800862
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800863 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
864 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700865
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800866 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700867 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800868
869 do {
Jason Low4db0c3c2015-04-15 16:14:08 -0700870 pos = READ_ONCE(iter->position);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800871 /*
872 * A racing update may change the position and
873 * put the last reference, hence css_tryget(),
874 * or retry to see the updated position.
875 */
876 } while (pos && !css_tryget(&pos->css));
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800877 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800878
879 if (pos)
880 css = &pos->css;
881
882 for (;;) {
883 css = css_next_descendant_pre(css, &root->css);
884 if (!css) {
885 /*
886 * Reclaimers share the hierarchy walk, and a
887 * new one might jump in right at the end of
888 * the hierarchy - make sure they see at least
889 * one group and restart from the beginning.
890 */
891 if (!prev)
892 continue;
893 break;
894 }
895
896 /*
897 * Verify the css and acquire a reference. The root
898 * is provided by the caller, so we know it's alive
899 * and kicking, and don't take an extra reference.
900 */
901 memcg = mem_cgroup_from_css(css);
902
903 if (css == &root->css)
904 break;
905
Johannes Weinerb2052562014-12-10 15:42:48 -0800906 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800907 /*
908 * Make sure the memcg is initialized:
909 * mem_cgroup_css_online() orders the the
910 * initialization against setting the flag.
911 */
912 if (smp_load_acquire(&memcg->initialized))
913 break;
914
915 css_put(css);
916 }
917
918 memcg = NULL;
919 }
920
921 if (reclaim) {
922 if (cmpxchg(&iter->position, pos, memcg) == pos) {
923 if (memcg)
924 css_get(&memcg->css);
925 if (pos)
926 css_put(&pos->css);
927 }
928
929 /*
930 * pairs with css_tryget when dereferencing iter->position
931 * above.
932 */
933 if (pos)
934 css_put(&pos->css);
935
936 if (!memcg)
937 iter->generation++;
938 else if (!prev)
939 reclaim->generation = iter->generation;
940 }
941
Michal Hocko542f85f2013-04-29 15:07:15 -0700942out_unlock:
943 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800944out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700945 if (prev && prev != root)
946 css_put(&prev->css);
947
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800948 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700949}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800950
Johannes Weiner56600482012-01-12 17:17:59 -0800951/**
952 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
953 * @root: hierarchy root
954 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
955 */
956void mem_cgroup_iter_break(struct mem_cgroup *root,
957 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800958{
959 if (!root)
960 root = root_mem_cgroup;
961 if (prev && prev != root)
962 css_put(&prev->css);
963}
964
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700965/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800966 * Iteration constructs for visiting all cgroups (under a tree). If
967 * loops are exited prematurely (break), mem_cgroup_iter_break() must
968 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700969 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800970#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800971 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800972 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800973 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700974
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800975#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800976 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800977 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800978 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700979
Johannes Weiner925b7672012-01-12 17:18:15 -0800980/**
981 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
982 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700983 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800984 *
985 * Returns the lru list vector holding pages for the given @zone and
986 * @mem. This can be the global zone lruvec, if the memory controller
987 * is disabled.
988 */
989struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
990 struct mem_cgroup *memcg)
991{
992 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800993 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800994
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800995 if (mem_cgroup_disabled()) {
996 lruvec = &zone->lruvec;
997 goto out;
998 }
Johannes Weiner925b7672012-01-12 17:18:15 -0800999
Jianyu Zhane2318752014-06-06 14:38:20 -07001000 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001001 lruvec = &mz->lruvec;
1002out:
1003 /*
1004 * Since a node can be onlined after the mem_cgroup was created,
1005 * we have to be prepared to initialize lruvec->zone here;
1006 * and if offlined then reonlined, we need to reinitialize it.
1007 */
1008 if (unlikely(lruvec->zone != zone))
1009 lruvec->zone = zone;
1010 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001011}
1012
Johannes Weiner925b7672012-01-12 17:18:15 -08001013/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001014 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001015 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001016 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001017 *
1018 * This function is only safe when following the LRU page isolation
1019 * and putback protocol: the LRU lock must be held, and the page must
1020 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001021 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001022struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001023{
1024 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001025 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001026 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001027
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001028 if (mem_cgroup_disabled()) {
1029 lruvec = &zone->lruvec;
1030 goto out;
1031 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001032
Johannes Weiner1306a852014-12-10 15:44:52 -08001033 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001034 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001035 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001036 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001037 */
Johannes Weiner29833312014-12-10 15:44:02 -08001038 if (!memcg)
1039 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001040
Jianyu Zhane2318752014-06-06 14:38:20 -07001041 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001042 lruvec = &mz->lruvec;
1043out:
1044 /*
1045 * Since a node can be onlined after the mem_cgroup was created,
1046 * we have to be prepared to initialize lruvec->zone here;
1047 * and if offlined then reonlined, we need to reinitialize it.
1048 */
1049 if (unlikely(lruvec->zone != zone))
1050 lruvec->zone = zone;
1051 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001052}
1053
1054/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001055 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1056 * @lruvec: mem_cgroup per zone lru vector
1057 * @lru: index of lru list the page is sitting on
1058 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001059 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001060 * This function must be called when a page is added to or removed from an
1061 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001062 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001063void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1064 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001065{
1066 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001067 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001068
1069 if (mem_cgroup_disabled())
1070 return;
1071
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001072 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1073 lru_size = mz->lru_size + lru;
1074 *lru_size += nr_pages;
1075 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001076}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001077
Johannes Weiner2314b422014-12-10 15:44:33 -08001078bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001079{
Johannes Weiner2314b422014-12-10 15:44:33 -08001080 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001081 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001082 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001083
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001084 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001085 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001086 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001087 task_unlock(p);
1088 } else {
1089 /*
1090 * All threads may have already detached their mm's, but the oom
1091 * killer still needs to detect if they have already been oom
1092 * killed to prevent needlessly killing additional tasks.
1093 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001094 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001095 task_memcg = mem_cgroup_from_task(task);
1096 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001097 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001098 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001099 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1100 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001101 return ret;
1102}
1103
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001104#define mem_cgroup_from_counter(counter, member) \
Balbir Singh6d61ef42009-01-07 18:08:06 -08001105 container_of(counter, struct mem_cgroup, member)
1106
Johannes Weiner19942822011-02-01 15:52:43 -08001107/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001108 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001109 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001110 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001111 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001112 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001113 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001114static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001115{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001116 unsigned long margin = 0;
1117 unsigned long count;
1118 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001119
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001120 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001121 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001122 if (count < limit)
1123 margin = limit - count;
1124
1125 if (do_swap_account) {
1126 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001127 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001128 if (count <= limit)
1129 margin = min(margin, limit - count);
1130 }
1131
1132 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001133}
1134
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001135/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001136 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001137 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001138 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1139 * moving cgroups. This is for waiting at high-memory pressure
1140 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001141 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001142static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001143{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001144 struct mem_cgroup *from;
1145 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001146 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001147 /*
1148 * Unlike task_move routines, we access mc.to, mc.from not under
1149 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1150 */
1151 spin_lock(&mc.lock);
1152 from = mc.from;
1153 to = mc.to;
1154 if (!from)
1155 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001156
Johannes Weiner2314b422014-12-10 15:44:33 -08001157 ret = mem_cgroup_is_descendant(from, memcg) ||
1158 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001159unlock:
1160 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001161 return ret;
1162}
1163
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001164static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001165{
1166 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001167 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001168 DEFINE_WAIT(wait);
1169 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1170 /* moving charge context might have finished. */
1171 if (mc.moving_task)
1172 schedule();
1173 finish_wait(&mc.waitq, &wait);
1174 return true;
1175 }
1176 }
1177 return false;
1178}
1179
Sha Zhengju58cf1882013-02-22 16:32:05 -08001180#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001181/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001182 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001183 * @memcg: The memory cgroup that went over limit
1184 * @p: Task that is going to be killed
1185 *
1186 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1187 * enabled
1188 */
1189void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1190{
Tejun Heoe61734c2014-02-12 09:29:50 -05001191 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001192 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001193 struct mem_cgroup *iter;
1194 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001195
Michal Hocko08088cb2014-02-25 15:01:44 -08001196 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001197 rcu_read_lock();
1198
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001199 if (p) {
1200 pr_info("Task in ");
1201 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1202 pr_cont(" killed as a result of limit of ");
1203 } else {
1204 pr_info("Memory limit reached of cgroup ");
1205 }
1206
Tejun Heoe61734c2014-02-12 09:29:50 -05001207 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001208 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001209
Balbir Singhe2224322009-04-02 16:57:39 -07001210 rcu_read_unlock();
1211
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001212 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1213 K((u64)page_counter_read(&memcg->memory)),
1214 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1215 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1216 K((u64)page_counter_read(&memcg->memsw)),
1217 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1218 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1219 K((u64)page_counter_read(&memcg->kmem)),
1220 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001221
1222 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001223 pr_info("Memory cgroup stats for ");
1224 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001225 pr_cont(":");
1226
1227 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1228 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1229 continue;
1230 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1231 K(mem_cgroup_read_stat(iter, i)));
1232 }
1233
1234 for (i = 0; i < NR_LRU_LISTS; i++)
1235 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1236 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1237
1238 pr_cont("\n");
1239 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001240 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001241}
1242
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001243/*
1244 * This function returns the number of memcg under hierarchy tree. Returns
1245 * 1(self count) if no children.
1246 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001247static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001248{
1249 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001250 struct mem_cgroup *iter;
1251
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001252 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001253 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001254 return num;
1255}
1256
Balbir Singh6d61ef42009-01-07 18:08:06 -08001257/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001258 * Return the memory (and swap, if configured) limit for a memcg.
1259 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001260static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001261{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001262 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001263
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001264 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001265 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001266 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001267
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001268 memsw_limit = memcg->memsw.limit;
1269 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001270 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001271 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001272}
1273
David Rientjes19965462012-12-11 16:00:26 -08001274static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1275 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001276{
1277 struct mem_cgroup *iter;
1278 unsigned long chosen_points = 0;
1279 unsigned long totalpages;
1280 unsigned int points = 0;
1281 struct task_struct *chosen = NULL;
1282
David Rientjes876aafb2012-07-31 16:43:48 -07001283 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001284 * If current has a pending SIGKILL or is exiting, then automatically
1285 * select it. The goal is to allow it to allocate so that it may
1286 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001287 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001288 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Michal Hocko49550b62015-02-11 15:26:12 -08001289 mark_tsk_oom_victim(current);
David Rientjes876aafb2012-07-31 16:43:48 -07001290 return;
1291 }
1292
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001293 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001294 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001295 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001296 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001297 struct task_struct *task;
1298
Tejun Heo72ec7022013-08-08 20:11:26 -04001299 css_task_iter_start(&iter->css, &it);
1300 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001301 switch (oom_scan_process_thread(task, totalpages, NULL,
1302 false)) {
1303 case OOM_SCAN_SELECT:
1304 if (chosen)
1305 put_task_struct(chosen);
1306 chosen = task;
1307 chosen_points = ULONG_MAX;
1308 get_task_struct(chosen);
1309 /* fall through */
1310 case OOM_SCAN_CONTINUE:
1311 continue;
1312 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001313 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001314 mem_cgroup_iter_break(memcg, iter);
1315 if (chosen)
1316 put_task_struct(chosen);
1317 return;
1318 case OOM_SCAN_OK:
1319 break;
1320 };
1321 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001322 if (!points || points < chosen_points)
1323 continue;
1324 /* Prefer thread group leaders for display purposes */
1325 if (points == chosen_points &&
1326 thread_group_leader(chosen))
1327 continue;
1328
1329 if (chosen)
1330 put_task_struct(chosen);
1331 chosen = task;
1332 chosen_points = points;
1333 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001334 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001335 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001336 }
1337
1338 if (!chosen)
1339 return;
1340 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001341 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1342 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001343}
1344
Michele Curtiae6e71d2014-12-12 16:56:35 -08001345#if MAX_NUMNODES > 1
1346
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001347/**
1348 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001349 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001350 * @nid: the node ID to be checked.
1351 * @noswap : specify true here if the user wants flle only information.
1352 *
1353 * This function returns whether the specified memcg contains any
1354 * reclaimable pages on a node. Returns true if there are any reclaimable
1355 * pages in the node.
1356 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001357static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001358 int nid, bool noswap)
1359{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001360 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001361 return true;
1362 if (noswap || !total_swap_pages)
1363 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001364 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001365 return true;
1366 return false;
1367
1368}
Ying Han889976d2011-05-26 16:25:33 -07001369
1370/*
1371 * Always updating the nodemask is not very good - even if we have an empty
1372 * list or the wrong list here, we can start from some node and traverse all
1373 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1374 *
1375 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001376static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001377{
1378 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001379 /*
1380 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1381 * pagein/pageout changes since the last update.
1382 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001383 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001384 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001385 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001386 return;
1387
Ying Han889976d2011-05-26 16:25:33 -07001388 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001389 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001390
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001391 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001392
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001393 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1394 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001395 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001396
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001397 atomic_set(&memcg->numainfo_events, 0);
1398 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001399}
1400
1401/*
1402 * Selecting a node where we start reclaim from. Because what we need is just
1403 * reducing usage counter, start from anywhere is O,K. Considering
1404 * memory reclaim from current node, there are pros. and cons.
1405 *
1406 * Freeing memory from current node means freeing memory from a node which
1407 * we'll use or we've used. So, it may make LRU bad. And if several threads
1408 * hit limits, it will see a contention on a node. But freeing from remote
1409 * node means more costs for memory reclaim because of memory latency.
1410 *
1411 * Now, we use round-robin. Better algorithm is welcomed.
1412 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001413int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001414{
1415 int node;
1416
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001417 mem_cgroup_may_update_nodemask(memcg);
1418 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001419
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001420 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001421 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001422 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001423 /*
1424 * We call this when we hit limit, not when pages are added to LRU.
1425 * No LRU may hold pages because all pages are UNEVICTABLE or
1426 * memcg is too small and all pages are not on LRU. In that case,
1427 * we use curret node.
1428 */
1429 if (unlikely(node == MAX_NUMNODES))
1430 node = numa_node_id();
1431
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001432 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001433 return node;
1434}
Ying Han889976d2011-05-26 16:25:33 -07001435#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001436int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001437{
1438 return 0;
1439}
1440#endif
1441
Andrew Morton0608f432013-09-24 15:27:41 -07001442static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1443 struct zone *zone,
1444 gfp_t gfp_mask,
1445 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001446{
Andrew Morton0608f432013-09-24 15:27:41 -07001447 struct mem_cgroup *victim = NULL;
1448 int total = 0;
1449 int loop = 0;
1450 unsigned long excess;
1451 unsigned long nr_scanned;
1452 struct mem_cgroup_reclaim_cookie reclaim = {
1453 .zone = zone,
1454 .priority = 0,
1455 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001456
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001457 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001458
Andrew Morton0608f432013-09-24 15:27:41 -07001459 while (1) {
1460 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1461 if (!victim) {
1462 loop++;
1463 if (loop >= 2) {
1464 /*
1465 * If we have not been able to reclaim
1466 * anything, it might because there are
1467 * no reclaimable pages under this hierarchy
1468 */
1469 if (!total)
1470 break;
1471 /*
1472 * We want to do more targeted reclaim.
1473 * excess >> 2 is not to excessive so as to
1474 * reclaim too much, nor too less that we keep
1475 * coming back to reclaim from this cgroup
1476 */
1477 if (total >= (excess >> 2) ||
1478 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1479 break;
1480 }
1481 continue;
1482 }
Andrew Morton0608f432013-09-24 15:27:41 -07001483 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1484 zone, &nr_scanned);
1485 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001486 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001487 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001488 }
Andrew Morton0608f432013-09-24 15:27:41 -07001489 mem_cgroup_iter_break(root_memcg, victim);
1490 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001491}
1492
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001493#ifdef CONFIG_LOCKDEP
1494static struct lockdep_map memcg_oom_lock_dep_map = {
1495 .name = "memcg_oom_lock",
1496};
1497#endif
1498
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001499static DEFINE_SPINLOCK(memcg_oom_lock);
1500
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001501/*
1502 * Check OOM-Killer is already running under our hierarchy.
1503 * If someone is running, return false.
1504 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001505static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001506{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001507 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001508
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001509 spin_lock(&memcg_oom_lock);
1510
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001511 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001512 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001513 /*
1514 * this subtree of our hierarchy is already locked
1515 * so we cannot give a lock.
1516 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001517 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001518 mem_cgroup_iter_break(memcg, iter);
1519 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001520 } else
1521 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001522 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001523
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001524 if (failed) {
1525 /*
1526 * OK, we failed to lock the whole subtree so we have
1527 * to clean up what we set up to the failing subtree
1528 */
1529 for_each_mem_cgroup_tree(iter, memcg) {
1530 if (iter == failed) {
1531 mem_cgroup_iter_break(memcg, iter);
1532 break;
1533 }
1534 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001535 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001536 } else
1537 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001538
1539 spin_unlock(&memcg_oom_lock);
1540
1541 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001542}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001543
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001544static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001545{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001546 struct mem_cgroup *iter;
1547
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001548 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001549 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001550 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001551 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001552 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001553}
1554
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001555static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001556{
1557 struct mem_cgroup *iter;
1558
Tejun Heo47088812015-06-24 16:58:23 -07001559 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001560 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heo47088812015-06-24 16:58:23 -07001561 iter->under_oom++;
1562 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001563}
1564
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001565static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001566{
1567 struct mem_cgroup *iter;
1568
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001569 /*
1570 * When a new child is created while the hierarchy is under oom,
Tejun Heo47088812015-06-24 16:58:23 -07001571 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001572 */
Tejun Heo47088812015-06-24 16:58:23 -07001573 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001574 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heo47088812015-06-24 16:58:23 -07001575 if (iter->under_oom > 0)
1576 iter->under_oom--;
1577 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001578}
1579
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001580static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1581
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001582struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001583 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001584 wait_queue_t wait;
1585};
1586
1587static int memcg_oom_wake_function(wait_queue_t *wait,
1588 unsigned mode, int sync, void *arg)
1589{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001590 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1591 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001592 struct oom_wait_info *oom_wait_info;
1593
1594 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001595 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001596
Johannes Weiner2314b422014-12-10 15:44:33 -08001597 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1598 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001599 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001600 return autoremove_wake_function(wait, mode, sync, arg);
1601}
1602
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001603static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001604{
Tejun Heo47088812015-06-24 16:58:23 -07001605 /*
1606 * For the following lockless ->under_oom test, the only required
1607 * guarantee is that it must see the state asserted by an OOM when
1608 * this function is called as a result of userland actions
1609 * triggered by the notification of the OOM. This is trivially
1610 * achieved by invoking mem_cgroup_mark_under_oom() before
1611 * triggering notification.
1612 */
1613 if (memcg && memcg->under_oom)
Tejun Heo55c3f9b2015-06-24 16:58:21 -07001614 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001615}
1616
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001617static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001618{
Tejun Heobd5d9d62015-11-05 18:46:09 -08001619 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001620 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001621 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001622 * We are in the middle of the charge context here, so we
1623 * don't want to block when potentially sitting on a callstack
1624 * that holds all kinds of filesystem and mm locks.
1625 *
1626 * Also, the caller may handle a failed allocation gracefully
1627 * (like optional page cache readahead) and so an OOM killer
1628 * invocation might not even be necessary.
1629 *
1630 * That's why we don't do anything here except remember the
1631 * OOM context and then deal with it at the end of the page
1632 * fault when the stack is unwound, the locks are released,
1633 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001634 */
Johannes Weiner49426422013-10-16 13:46:59 -07001635 css_get(&memcg->css);
Tejun Heobd5d9d62015-11-05 18:46:09 -08001636 current->memcg_in_oom = memcg;
1637 current->memcg_oom_gfp_mask = mask;
1638 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001639}
1640
1641/**
1642 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1643 * @handle: actually kill/wait or just clean up the OOM state
1644 *
1645 * This has to be called at the end of a page fault if the memcg OOM
1646 * handler was enabled.
1647 *
1648 * Memcg supports userspace OOM handling where failed allocations must
1649 * sleep on a waitqueue until the userspace task resolves the
1650 * situation. Sleeping directly in the charge context with all kinds
1651 * of locks held is not a good idea, instead we remember an OOM state
1652 * in the task and mem_cgroup_oom_synchronize() has to be called at
1653 * the end of the page fault to complete the OOM handling.
1654 *
1655 * Returns %true if an ongoing memcg OOM situation was detected and
1656 * completed, %false otherwise.
1657 */
1658bool mem_cgroup_oom_synchronize(bool handle)
1659{
Tejun Heobd5d9d62015-11-05 18:46:09 -08001660 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001661 struct oom_wait_info owait;
1662 bool locked;
1663
1664 /* OOM is global, do not handle */
1665 if (!memcg)
1666 return false;
1667
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001668 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001669 goto cleanup;
1670
1671 owait.memcg = memcg;
1672 owait.wait.flags = 0;
1673 owait.wait.func = memcg_oom_wake_function;
1674 owait.wait.private = current;
1675 INIT_LIST_HEAD(&owait.wait.task_list);
1676
1677 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001678 mem_cgroup_mark_under_oom(memcg);
1679
1680 locked = mem_cgroup_oom_trylock(memcg);
1681
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001682 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001683 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001684
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001685 if (locked && !memcg->oom_kill_disable) {
1686 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001687 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heobd5d9d62015-11-05 18:46:09 -08001688 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1689 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001690 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001691 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001692 mem_cgroup_unmark_under_oom(memcg);
1693 finish_wait(&memcg_oom_waitq, &owait.wait);
1694 }
1695
1696 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001697 mem_cgroup_oom_unlock(memcg);
1698 /*
1699 * There is no guarantee that an OOM-lock contender
1700 * sees the wakeups triggered by the OOM kill
1701 * uncharges. Wake any sleepers explicitely.
1702 */
1703 memcg_oom_recover(memcg);
1704 }
Johannes Weiner49426422013-10-16 13:46:59 -07001705cleanup:
Tejun Heobd5d9d62015-11-05 18:46:09 -08001706 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001707 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001708 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001709}
1710
Johannes Weinerd7365e72014-10-29 14:50:48 -07001711/**
1712 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1713 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001714 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001715 * This function must mark the beginning of an accounted page state
1716 * change to prevent double accounting when the page is concurrently
1717 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001718 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001719 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001720 * if (TestClearPageState(page))
1721 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001722 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001723 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001724struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001725{
1726 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001727 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001728
Johannes Weiner6de22612015-02-11 15:25:01 -08001729 /*
1730 * The RCU lock is held throughout the transaction. The fast
1731 * path can get away without acquiring the memcg->move_lock
1732 * because page moving starts with an RCU grace period.
1733 *
1734 * The RCU lock also protects the memcg from being freed when
1735 * the page state that is going to change is the only thing
1736 * preventing the page from being uncharged.
1737 * E.g. end-writeback clearing PageWriteback(), which allows
1738 * migration to go ahead and uncharge the page before the
1739 * account transaction might be complete.
1740 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001741 rcu_read_lock();
1742
1743 if (mem_cgroup_disabled())
1744 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001745again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001746 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001747 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07001748 return NULL;
1749
Qiang Huangbdcbb652014-06-04 16:08:21 -07001750 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07001751 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001752
Johannes Weiner6de22612015-02-11 15:25:01 -08001753 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001754 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001755 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001756 goto again;
1757 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001758
1759 /*
1760 * When charge migration first begins, we can have locked and
1761 * unlocked page stat updates happening concurrently. Track
1762 * the task who has the lock for mem_cgroup_end_page_stat().
1763 */
1764 memcg->move_lock_task = current;
1765 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001766
1767 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001768}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04001769EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001770
Johannes Weinerd7365e72014-10-29 14:50:48 -07001771/**
1772 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1773 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07001774 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001775void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001776{
Johannes Weiner6de22612015-02-11 15:25:01 -08001777 if (memcg && memcg->move_lock_task == current) {
1778 unsigned long flags = memcg->move_lock_flags;
1779
1780 memcg->move_lock_task = NULL;
1781 memcg->move_lock_flags = 0;
1782
1783 spin_unlock_irqrestore(&memcg->move_lock, flags);
1784 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001785
Johannes Weinerd7365e72014-10-29 14:50:48 -07001786 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001787}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04001788EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001789
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001790/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001791 * size of first charge trial. "32" comes from vmscan.c's magic value.
1792 * TODO: maybe necessary to use big numbers in big irons.
1793 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001794#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001795struct memcg_stock_pcp {
1796 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001797 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001798 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001799 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001800#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001801};
1802static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001803static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001804
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001805/**
1806 * consume_stock: Try to consume stocked charge on this cpu.
1807 * @memcg: memcg to consume from.
1808 * @nr_pages: how many pages to charge.
1809 *
1810 * The charges will only happen if @memcg matches the current cpu's memcg
1811 * stock, and at least @nr_pages are available in that stock. Failure to
1812 * service an allocation will refill the stock.
1813 *
1814 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001816static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001817{
1818 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001819 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001820
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001821 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001822 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001823
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001824 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001825 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001826 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001827 ret = true;
1828 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001829 put_cpu_var(memcg_stock);
1830 return ret;
1831}
1832
1833/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001834 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001835 */
1836static void drain_stock(struct memcg_stock_pcp *stock)
1837{
1838 struct mem_cgroup *old = stock->cached;
1839
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001840 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001841 page_counter_uncharge(&old->memory, stock->nr_pages);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001842 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001843 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001844 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001845 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001846 }
1847 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001848}
1849
1850/*
1851 * This must be called under preempt disabled or must be called by
1852 * a thread which is pinned to local cpu.
1853 */
1854static void drain_local_stock(struct work_struct *dummy)
1855{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001856 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001857 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001858 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001859}
1860
1861/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001862 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001863 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001864 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001865static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001866{
1867 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1868
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001869 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001870 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001871 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001872 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001873 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001874 put_cpu_var(memcg_stock);
1875}
1876
1877/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001878 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001879 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001880 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001881static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001882{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001883 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001884
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001885 /* If someone's already draining, avoid adding running more workers. */
1886 if (!mutex_trylock(&percpu_charge_mutex))
1887 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001888 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001889 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001890 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001891 for_each_online_cpu(cpu) {
1892 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001893 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001894
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001895 memcg = stock->cached;
1896 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001897 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001898 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001899 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001900 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1901 if (cpu == curcpu)
1902 drain_local_stock(&stock->work);
1903 else
1904 schedule_work_on(cpu, &stock->work);
1905 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001906 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001907 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001908 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001909 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001910}
1911
Paul Gortmaker0db06282013-06-19 14:53:51 -04001912static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001913 unsigned long action,
1914 void *hcpu)
1915{
1916 int cpu = (unsigned long)hcpu;
1917 struct memcg_stock_pcp *stock;
1918
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001919 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001920 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001921
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001922 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001923 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001924
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001925 stock = &per_cpu(memcg_stock, cpu);
1926 drain_stock(stock);
1927 return NOTIFY_OK;
1928}
1929
Tejun Heoc2644f52015-11-05 18:46:11 -08001930/*
1931 * Scheduled by try_charge() to be executed from the userland return path
1932 * and reclaims memory over the high limit.
1933 */
1934void mem_cgroup_handle_over_high(void)
1935{
1936 unsigned int nr_pages = current->memcg_nr_pages_over_high;
1937 struct mem_cgroup *memcg, *pos;
1938
1939 if (likely(!nr_pages))
1940 return;
1941
1942 pos = memcg = get_mem_cgroup_from_mm(current->mm);
1943
1944 do {
1945 if (page_counter_read(&pos->memory) <= pos->high)
1946 continue;
1947 mem_cgroup_events(pos, MEMCG_HIGH, 1);
1948 try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true);
1949 } while ((pos = parent_mem_cgroup(pos)));
1950
1951 css_put(&memcg->css);
1952 current->memcg_nr_pages_over_high = 0;
1953}
1954
Johannes Weiner00501b52014-08-08 14:19:20 -07001955static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1956 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001957{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001958 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001959 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001960 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001961 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001962 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001963 bool may_swap = true;
1964 bool drained = false;
Johannes Weiner05b84302014-08-06 16:05:59 -07001965 int ret = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001966
Johannes Weinerce00a962014-09-05 08:43:57 -04001967 if (mem_cgroup_is_root(memcg))
1968 goto done;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001969retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001970 if (consume_stock(memcg, nr_pages))
1971 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001972
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001973 if (!do_swap_account ||
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001974 !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1975 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001976 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001977 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001978 page_counter_uncharge(&memcg->memsw, batch);
1979 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001980 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001981 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001982 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001983 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001984
Johannes Weiner6539cc02014-08-06 16:05:42 -07001985 if (batch > nr_pages) {
1986 batch = nr_pages;
1987 goto retry;
1988 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001989
Johannes Weiner06b078f2014-08-06 16:05:44 -07001990 /*
1991 * Unlike in global OOM situations, memcg is not in a physical
1992 * memory shortage. Allow dying and OOM-killed tasks to
1993 * bypass the last charges so that they can exit quickly and
1994 * free their memory.
1995 */
1996 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1997 fatal_signal_pending(current) ||
1998 current->flags & PF_EXITING))
1999 goto bypass;
2000
2001 if (unlikely(task_in_memcg_oom(current)))
2002 goto nomem;
2003
Johannes Weiner6539cc02014-08-06 16:05:42 -07002004 if (!(gfp_mask & __GFP_WAIT))
2005 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002006
Johannes Weiner241994ed2015-02-11 15:26:06 -08002007 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2008
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002009 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2010 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002011
Johannes Weiner61e02c72014-08-06 16:08:16 -07002012 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002013 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002014
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002015 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002016 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002017 drained = true;
2018 goto retry;
2019 }
2020
Johannes Weiner28c34c22014-08-06 16:05:47 -07002021 if (gfp_mask & __GFP_NORETRY)
2022 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002023 /*
2024 * Even though the limit is exceeded at this point, reclaim
2025 * may have been able to free some pages. Retry the charge
2026 * before killing the task.
2027 *
2028 * Only for regular pages, though: huge pages are rather
2029 * unlikely to succeed so close to the limit, and we fall back
2030 * to regular pages anyway in case of failure.
2031 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002032 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002033 goto retry;
2034 /*
2035 * At task move, charge accounts can be doubly counted. So, it's
2036 * better to wait until the end of task_move if something is going on.
2037 */
2038 if (mem_cgroup_wait_acct_move(mem_over_limit))
2039 goto retry;
2040
Johannes Weiner9b130612014-08-06 16:05:51 -07002041 if (nr_retries--)
2042 goto retry;
2043
Johannes Weiner06b078f2014-08-06 16:05:44 -07002044 if (gfp_mask & __GFP_NOFAIL)
2045 goto bypass;
2046
Johannes Weiner6539cc02014-08-06 16:05:42 -07002047 if (fatal_signal_pending(current))
2048 goto bypass;
2049
Johannes Weiner241994ed2015-02-11 15:26:06 -08002050 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2051
Johannes Weiner61e02c72014-08-06 16:08:16 -07002052 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002053nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002054 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002055 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002056bypass:
Johannes Weinerce00a962014-09-05 08:43:57 -04002057 return -EINTR;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002058
2059done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002060 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002061 if (batch > nr_pages)
2062 refill_stock(memcg, batch - nr_pages);
Tejun Heoc2644f52015-11-05 18:46:11 -08002063
Johannes Weiner241994ed2015-02-11 15:26:06 -08002064 /*
Tejun Heoc2644f52015-11-05 18:46:11 -08002065 * If the hierarchy is above the normal consumption range, schedule
2066 * reclaim on returning to userland. We can perform reclaim here
2067 * if __GFP_WAIT but let's always punt for simplicity and so that
2068 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2069 * not recorded as it most likely matches current's and won't
2070 * change in the meantime. As high limit is checked again before
2071 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002072 */
2073 do {
Tejun Heoc2644f52015-11-05 18:46:11 -08002074 if (page_counter_read(&memcg->memory) > memcg->high) {
2075 current->memcg_nr_pages_over_high += nr_pages;
2076 set_notify_resume(current);
2077 break;
2078 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002079 } while ((memcg = parent_mem_cgroup(memcg)));
Johannes Weiner6539cc02014-08-06 16:05:42 -07002080done:
Johannes Weiner05b84302014-08-06 16:05:59 -07002081 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002082}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002083
Johannes Weiner00501b52014-08-08 14:19:20 -07002084static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002085{
Johannes Weinerce00a962014-09-05 08:43:57 -04002086 if (mem_cgroup_is_root(memcg))
2087 return;
2088
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002089 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner05b84302014-08-06 16:05:59 -07002090 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002091 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002092
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002093 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002094}
2095
2096/*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002097 * try_get_mem_cgroup_from_page - look up page's memcg association
2098 * @page: the page
2099 *
2100 * Look up, get a css reference, and return the memcg that owns @page.
2101 *
2102 * The page must be locked to prevent racing with swap-in and page
2103 * cache charges. If coming from an unlocked page table, the caller
2104 * must ensure the page is on the LRU or this can race with charging.
2105 */
Wu Fengguange42d9d52009-12-16 12:19:59 +01002106struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002107{
Johannes Weiner29833312014-12-10 15:44:02 -08002108 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002109 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002110 swp_entry_t ent;
2111
Sasha Levin309381fea2014-01-23 15:52:54 -08002112 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002113
Johannes Weiner1306a852014-12-10 15:44:52 -08002114 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002115 if (memcg) {
2116 if (!css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002117 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002118 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002119 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002120 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002121 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07002122 memcg = mem_cgroup_from_id(id);
Tejun Heoec903c02014-05-13 12:11:01 -04002123 if (memcg && !css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002124 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002125 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002126 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002127 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002128}
2129
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002130static void lock_page_lru(struct page *page, int *isolated)
2131{
2132 struct zone *zone = page_zone(page);
2133
2134 spin_lock_irq(&zone->lru_lock);
2135 if (PageLRU(page)) {
2136 struct lruvec *lruvec;
2137
2138 lruvec = mem_cgroup_page_lruvec(page, zone);
2139 ClearPageLRU(page);
2140 del_page_from_lru_list(page, lruvec, page_lru(page));
2141 *isolated = 1;
2142 } else
2143 *isolated = 0;
2144}
2145
2146static void unlock_page_lru(struct page *page, int isolated)
2147{
2148 struct zone *zone = page_zone(page);
2149
2150 if (isolated) {
2151 struct lruvec *lruvec;
2152
2153 lruvec = mem_cgroup_page_lruvec(page, zone);
2154 VM_BUG_ON_PAGE(PageLRU(page), page);
2155 SetPageLRU(page);
2156 add_page_to_lru_list(page, lruvec, page_lru(page));
2157 }
2158 spin_unlock_irq(&zone->lru_lock);
2159}
2160
Johannes Weiner00501b52014-08-08 14:19:20 -07002161static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002162 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002163{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002164 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002165
Johannes Weiner1306a852014-12-10 15:44:52 -08002166 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002167
2168 /*
2169 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2170 * may already be on some other mem_cgroup's LRU. Take care of it.
2171 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002172 if (lrucare)
2173 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002174
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002175 /*
2176 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002177 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002178 *
2179 * - the page is uncharged
2180 *
2181 * - the page is off-LRU
2182 *
2183 * - an anonymous fault has exclusive page access, except for
2184 * a locked page table
2185 *
2186 * - a page cache insertion, a swapin fault, or a migration
2187 * have the page locked
2188 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002189 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002190
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002191 if (lrucare)
2192 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002193}
2194
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002195#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002196int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2197 unsigned long nr_pages)
Glauber Costa749c5412012-12-18 14:23:01 -08002198{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002199 struct page_counter *counter;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002200 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002201
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002202 ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2203 if (ret < 0)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002204 return ret;
2205
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002206 ret = try_charge(memcg, gfp, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002207 if (ret == -EINTR) {
2208 /*
Johannes Weiner00501b52014-08-08 14:19:20 -07002209 * try_charge() chose to bypass to root due to OOM kill or
2210 * fatal signal. Since our only options are to either fail
2211 * the allocation or charge it to this cgroup, do it as a
2212 * temporary condition. But we can't fail. From a kmem/slab
2213 * perspective, the cache has already been selected, by
2214 * mem_cgroup_kmem_get_cache(), so it is too late to change
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002215 * our minds.
2216 *
2217 * This condition will only trigger if the task entered
Johannes Weiner00501b52014-08-08 14:19:20 -07002218 * memcg_charge_kmem in a sane state, but was OOM-killed
2219 * during try_charge() above. Tasks that were already dying
2220 * when the allocation triggers should have been already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002221 * directed to the root cgroup in memcontrol.h
2222 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002223 page_counter_charge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002224 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002225 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002226 css_get_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002227 ret = 0;
2228 } else if (ret)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002229 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002230
2231 return ret;
2232}
2233
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002234void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002235{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002236 page_counter_uncharge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002237 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002238 page_counter_uncharge(&memcg->memsw, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002239
Johannes Weiner64f21992014-12-10 15:42:45 -08002240 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002241
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002242 css_put_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002243}
2244
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002245static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002246{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002247 int id, size;
2248 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002249
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002250 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002251 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2252 if (id < 0)
2253 return id;
2254
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002255 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002256 return id;
2257
2258 /*
2259 * There's no space for the new id in memcg_caches arrays,
2260 * so we have to grow them.
2261 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002262 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002263
2264 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002265 if (size < MEMCG_CACHES_MIN_SIZE)
2266 size = MEMCG_CACHES_MIN_SIZE;
2267 else if (size > MEMCG_CACHES_MAX_SIZE)
2268 size = MEMCG_CACHES_MAX_SIZE;
2269
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002270 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002271 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002272 err = memcg_update_all_list_lrus(size);
2273 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002274 memcg_nr_cache_ids = size;
2275
2276 up_write(&memcg_cache_ids_sem);
2277
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002278 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002279 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002280 return err;
2281 }
2282 return id;
2283}
2284
2285static void memcg_free_cache_id(int id)
2286{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002287 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002288}
2289
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002290struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002291 struct mem_cgroup *memcg;
2292 struct kmem_cache *cachep;
2293 struct work_struct work;
2294};
2295
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002296static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002297{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002298 struct memcg_kmem_cache_create_work *cw =
2299 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002300 struct mem_cgroup *memcg = cw->memcg;
2301 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002302
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002303 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002304
Vladimir Davydov5722d092014-04-07 15:39:24 -07002305 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002306 kfree(cw);
2307}
2308
2309/*
2310 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002311 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002312static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2313 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002314{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002315 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002316
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002317 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002318 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002319 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002320
2321 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002322
2323 cw->memcg = memcg;
2324 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002325 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002326
Glauber Costad7f25f82012-12-18 14:22:40 -08002327 schedule_work(&cw->work);
2328}
2329
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002330static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2331 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002332{
2333 /*
2334 * We need to stop accounting when we kmalloc, because if the
2335 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002336 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002337 *
2338 * However, it is better to enclose the whole function. Depending on
2339 * the debugging options enabled, INIT_WORK(), for instance, can
2340 * trigger an allocation. This too, will make us recurse. Because at
2341 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2342 * the safest choice is to do it like this, wrapping the whole function.
2343 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002344 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002345 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002346 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002347}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002348
Glauber Costad7f25f82012-12-18 14:22:40 -08002349/*
2350 * Return the kmem_cache we're supposed to use for a slab allocation.
2351 * We try to use the current memcg's version of the cache.
2352 *
2353 * If the cache does not exist yet, if we are the first user of it,
2354 * we either create it immediately, if possible, or create it asynchronously
2355 * in a workqueue.
2356 * In the latter case, we will let the current allocation go through with
2357 * the original cache.
2358 *
2359 * Can't be called in interrupt context or from kernel threads.
2360 * This function needs to be called with rcu_read_lock() held.
2361 */
Zhang Zhen056b7cc2014-12-12 16:55:38 -08002362struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002363{
2364 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002365 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002366 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002367
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002368 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002369
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002370 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002371 return cachep;
2372
Vladimir Davydov8135be52014-12-12 16:56:38 -08002373 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002374 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002375 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002376 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002377
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002378 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002379 if (likely(memcg_cachep))
2380 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002381
2382 /*
2383 * If we are in a safe context (can wait, and not in interrupt
2384 * context), we could be be predictable and return right away.
2385 * This would guarantee that the allocation being performed
2386 * already belongs in the new cache.
2387 *
2388 * However, there are some clashes that can arrive from locking.
2389 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002390 * memcg_create_kmem_cache, this means no further allocation
2391 * could happen with the slab_mutex held. So it's better to
2392 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002393 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002394 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002395out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002396 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002397 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002398}
Glauber Costad7f25f82012-12-18 14:22:40 -08002399
Vladimir Davydov8135be52014-12-12 16:56:38 -08002400void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2401{
2402 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002403 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002404}
2405
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002406/*
2407 * We need to verify if the allocation against current->mm->owner's memcg is
2408 * possible for the given order. But the page is not allocated yet, so we'll
2409 * need a further commit step to do the final arrangements.
2410 *
2411 * It is possible for the task to switch cgroups in this mean time, so at
2412 * commit time, we can't rely on task conversion any longer. We'll then use
2413 * the handle argument to return to the caller which cgroup we should commit
2414 * against. We could also return the memcg directly and avoid the pointer
2415 * passing, but a boolean return value gives better semantics considering
2416 * the compiled-out case as well.
2417 *
2418 * Returning true means the allocation is possible.
2419 */
2420bool
2421__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2422{
2423 struct mem_cgroup *memcg;
2424 int ret;
2425
2426 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07002427
Johannes Weinerdf381972014-04-07 15:37:43 -07002428 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002429
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07002430 if (!memcg_kmem_is_active(memcg)) {
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002431 css_put(&memcg->css);
2432 return true;
2433 }
2434
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002435 ret = memcg_charge_kmem(memcg, gfp, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002436 if (!ret)
2437 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002438
2439 css_put(&memcg->css);
2440 return (ret == 0);
2441}
2442
2443void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2444 int order)
2445{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002446 VM_BUG_ON(mem_cgroup_is_root(memcg));
2447
2448 /* The page allocation failed. Revert */
2449 if (!page) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002450 memcg_uncharge_kmem(memcg, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002451 return;
2452 }
Johannes Weiner1306a852014-12-10 15:44:52 -08002453 page->mem_cgroup = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002454}
2455
2456void __memcg_kmem_uncharge_pages(struct page *page, int order)
2457{
Johannes Weiner1306a852014-12-10 15:44:52 -08002458 struct mem_cgroup *memcg = page->mem_cgroup;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002459
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002460 if (!memcg)
2461 return;
2462
Sasha Levin309381fea2014-01-23 15:52:54 -08002463 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002464
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002465 memcg_uncharge_kmem(memcg, 1 << order);
Johannes Weiner1306a852014-12-10 15:44:52 -08002466 page->mem_cgroup = NULL;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002467}
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002468
2469struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2470{
2471 struct mem_cgroup *memcg = NULL;
2472 struct kmem_cache *cachep;
2473 struct page *page;
2474
2475 page = virt_to_head_page(ptr);
2476 if (PageSlab(page)) {
2477 cachep = page->slab_cache;
2478 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002479 memcg = cachep->memcg_params.memcg;
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002480 } else
2481 /* page allocated by alloc_kmem_pages */
2482 memcg = page->mem_cgroup;
2483
2484 return memcg;
2485}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002486#endif /* CONFIG_MEMCG_KMEM */
2487
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002488#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2489
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002490/*
2491 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002492 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2493 * charge/uncharge will be never happen and move_account() is done under
2494 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002495 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002496void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002497{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002498 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002499
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002500 if (mem_cgroup_disabled())
2501 return;
David Rientjesb070e652013-05-07 16:18:09 -07002502
Johannes Weiner29833312014-12-10 15:44:02 -08002503 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002504 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002505
Johannes Weiner1306a852014-12-10 15:44:52 -08002506 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002507 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002508}
Hugh Dickins12d27102012-01-12 17:19:52 -08002509#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002510
Andrew Mortonc255a452012-07-31 16:43:02 -07002511#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002512static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2513 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002514{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002515 int val = (charge) ? 1 : -1;
2516 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002517}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002518
2519/**
2520 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2521 * @entry: swap entry to be moved
2522 * @from: mem_cgroup which the entry is moved from
2523 * @to: mem_cgroup which the entry is moved to
2524 *
2525 * It succeeds only when the swap_cgroup's record for this entry is the same
2526 * as the mem_cgroup's id of @from.
2527 *
2528 * Returns 0 on success, -EINVAL on failure.
2529 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002530 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002531 * both res and memsw, and called css_get().
2532 */
2533static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002534 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002535{
2536 unsigned short old_id, new_id;
2537
Li Zefan34c00c32013-09-23 16:56:01 +08002538 old_id = mem_cgroup_id(from);
2539 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002540
2541 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002542 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002543 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002544 return 0;
2545 }
2546 return -EINVAL;
2547}
2548#else
2549static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002550 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002551{
2552 return -EINVAL;
2553}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002554#endif
2555
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002556static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002557
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002558static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002559 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002560{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002561 unsigned long curusage;
2562 unsigned long oldusage;
2563 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002564 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002565 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002566
2567 /*
2568 * For keeping hierarchical_reclaim simple, how long we should retry
2569 * is depends on callers. We set our retry-count to be function
2570 * of # of children which we should visit in this loop.
2571 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002572 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2573 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002574
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002575 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002576
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002577 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002578 if (signal_pending(current)) {
2579 ret = -EINTR;
2580 break;
2581 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002582
2583 mutex_lock(&memcg_limit_mutex);
2584 if (limit > memcg->memsw.limit) {
2585 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002586 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002587 break;
2588 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002589 if (limit > memcg->memory.limit)
2590 enlarge = true;
2591 ret = page_counter_limit(&memcg->memory, limit);
2592 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002593
2594 if (!ret)
2595 break;
2596
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002597 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2598
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002599 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002600 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002601 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002602 retry_count--;
2603 else
2604 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002605 } while (retry_count);
2606
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002607 if (!ret && enlarge)
2608 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002609
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002610 return ret;
2611}
2612
Li Zefan338c8432009-06-17 16:27:15 -07002613static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002614 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002615{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002616 unsigned long curusage;
2617 unsigned long oldusage;
2618 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002619 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002620 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002621
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002622 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002623 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2624 mem_cgroup_count_children(memcg);
2625
2626 oldusage = page_counter_read(&memcg->memsw);
2627
2628 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002629 if (signal_pending(current)) {
2630 ret = -EINTR;
2631 break;
2632 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002633
2634 mutex_lock(&memcg_limit_mutex);
2635 if (limit < memcg->memory.limit) {
2636 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002637 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002638 break;
2639 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002640 if (limit > memcg->memsw.limit)
2641 enlarge = true;
2642 ret = page_counter_limit(&memcg->memsw, limit);
2643 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002644
2645 if (!ret)
2646 break;
2647
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002648 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2649
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002650 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002651 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002652 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002653 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002654 else
2655 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002656 } while (retry_count);
2657
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002658 if (!ret && enlarge)
2659 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002660
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002661 return ret;
2662}
2663
Andrew Morton0608f432013-09-24 15:27:41 -07002664unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2665 gfp_t gfp_mask,
2666 unsigned long *total_scanned)
2667{
2668 unsigned long nr_reclaimed = 0;
2669 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2670 unsigned long reclaimed;
2671 int loop = 0;
2672 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002673 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002674 unsigned long nr_scanned;
2675
2676 if (order > 0)
2677 return 0;
2678
2679 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2680 /*
2681 * This loop can run a while, specially if mem_cgroup's continuously
2682 * keep exceeding their soft limit and putting the system under
2683 * pressure
2684 */
2685 do {
2686 if (next_mz)
2687 mz = next_mz;
2688 else
2689 mz = mem_cgroup_largest_soft_limit_node(mctz);
2690 if (!mz)
2691 break;
2692
2693 nr_scanned = 0;
2694 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2695 gfp_mask, &nr_scanned);
2696 nr_reclaimed += reclaimed;
2697 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002698 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002699 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002700
2701 /*
2702 * If we failed to reclaim anything from this memory cgroup
2703 * it is time to move on to the next cgroup
2704 */
2705 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002706 if (!reclaimed)
2707 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2708
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002709 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002710 /*
2711 * One school of thought says that we should not add
2712 * back the node to the tree if reclaim returns 0.
2713 * But our reclaim could return 0, simply because due
2714 * to priority we are exposing a smaller subset of
2715 * memory to reclaim from. Consider this as a longer
2716 * term TODO.
2717 */
2718 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002719 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002720 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002721 css_put(&mz->memcg->css);
2722 loop++;
2723 /*
2724 * Could not reclaim anything and there are no more
2725 * mem cgroups to try or we seem to be looping without
2726 * reclaiming anything.
2727 */
2728 if (!nr_reclaimed &&
2729 (next_mz == NULL ||
2730 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2731 break;
2732 } while (!nr_reclaimed);
2733 if (next_mz)
2734 css_put(&next_mz->memcg->css);
2735 return nr_reclaimed;
2736}
2737
Tejun Heoea280e72014-05-16 13:22:48 -04002738/*
2739 * Test whether @memcg has children, dead or alive. Note that this
2740 * function doesn't care whether @memcg has use_hierarchy enabled and
2741 * returns %true if there are child csses according to the cgroup
2742 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2743 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002744static inline bool memcg_has_children(struct mem_cgroup *memcg)
2745{
Tejun Heoea280e72014-05-16 13:22:48 -04002746 bool ret;
2747
Johannes Weiner696ac172013-10-31 16:34:15 -07002748 /*
Tejun Heoea280e72014-05-16 13:22:48 -04002749 * The lock does not prevent addition or deletion of children, but
2750 * it prevents a new child from being initialized based on this
2751 * parent in css_online(), so it's enough to decide whether
2752 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07002753 */
Tejun Heoea280e72014-05-16 13:22:48 -04002754 lockdep_assert_held(&memcg_create_mutex);
2755
2756 rcu_read_lock();
2757 ret = css_next_child(NULL, &memcg->css);
2758 rcu_read_unlock();
2759 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002760}
2761
2762/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002763 * Reclaims as many pages from the given memcg as possible and moves
2764 * the rest to the parent.
2765 *
2766 * Caller is responsible for holding css reference for memcg.
2767 */
2768static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2769{
2770 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002771
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002772 /* we call try-to-free pages for make this cgroup empty */
2773 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002774 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002775 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002776 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002777
Michal Hockoc26251f2012-10-26 13:37:28 +02002778 if (signal_pending(current))
2779 return -EINTR;
2780
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002781 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2782 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002783 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002784 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002785 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002786 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002787 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002788
2789 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002790
2791 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002792}
2793
Tejun Heo6770c642014-05-13 12:16:21 -04002794static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2795 char *buf, size_t nbytes,
2796 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002797{
Tejun Heo6770c642014-05-13 12:16:21 -04002798 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002799
Michal Hockod8423012012-10-26 13:37:29 +02002800 if (mem_cgroup_is_root(memcg))
2801 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002802 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002803}
2804
Tejun Heo182446d2013-08-08 20:11:24 -04002805static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2806 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002807{
Tejun Heo182446d2013-08-08 20:11:24 -04002808 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002809}
2810
Tejun Heo182446d2013-08-08 20:11:24 -04002811static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2812 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002813{
2814 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002815 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002816 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002817
Glauber Costa09998212013-02-22 16:34:55 -08002818 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07002819
2820 if (memcg->use_hierarchy == val)
2821 goto out;
2822
Balbir Singh18f59ea2009-01-07 18:08:07 -08002823 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002824 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002825 * in the child subtrees. If it is unset, then the change can
2826 * occur, provided the current cgroup has no children.
2827 *
2828 * For the root cgroup, parent_mem is NULL, we allow value to be
2829 * set if there are no children.
2830 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002831 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002832 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002833 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002834 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002835 else
2836 retval = -EBUSY;
2837 } else
2838 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002839
2840out:
Glauber Costa09998212013-02-22 16:34:55 -08002841 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002842
2843 return retval;
2844}
2845
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002846static unsigned long tree_stat(struct mem_cgroup *memcg,
2847 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04002848{
2849 struct mem_cgroup *iter;
2850 long val = 0;
2851
2852 /* Per-cpu values can be negative, use a signed accumulator */
2853 for_each_mem_cgroup_tree(iter, memcg)
2854 val += mem_cgroup_read_stat(iter, idx);
2855
2856 if (val < 0) /* race ? */
2857 val = 0;
2858 return val;
2859}
2860
2861static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2862{
2863 u64 val;
2864
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002865 if (mem_cgroup_is_root(memcg)) {
2866 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
2867 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
2868 if (swap)
2869 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
2870 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002871 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002872 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002873 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002874 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002875 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002876 return val << PAGE_SHIFT;
2877}
2878
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002879enum {
2880 RES_USAGE,
2881 RES_LIMIT,
2882 RES_MAX_USAGE,
2883 RES_FAILCNT,
2884 RES_SOFT_LIMIT,
2885};
Johannes Weinerce00a962014-09-05 08:43:57 -04002886
Tejun Heo791badb2013-12-05 12:28:02 -05002887static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002888 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002889{
Tejun Heo182446d2013-08-08 20:11:24 -04002890 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002891 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002892
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002893 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002894 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002895 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002896 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002897 case _MEMSWAP:
2898 counter = &memcg->memsw;
2899 break;
2900 case _KMEM:
2901 counter = &memcg->kmem;
2902 break;
2903 default:
2904 BUG();
2905 }
2906
2907 switch (MEMFILE_ATTR(cft->private)) {
2908 case RES_USAGE:
2909 if (counter == &memcg->memory)
2910 return mem_cgroup_usage(memcg, false);
2911 if (counter == &memcg->memsw)
2912 return mem_cgroup_usage(memcg, true);
2913 return (u64)page_counter_read(counter) * PAGE_SIZE;
2914 case RES_LIMIT:
2915 return (u64)counter->limit * PAGE_SIZE;
2916 case RES_MAX_USAGE:
2917 return (u64)counter->watermark * PAGE_SIZE;
2918 case RES_FAILCNT:
2919 return counter->failcnt;
2920 case RES_SOFT_LIMIT:
2921 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002922 default:
2923 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002924 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002925}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002926
Glauber Costa510fc4e2012-12-18 14:21:47 -08002927#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002928static int memcg_activate_kmem(struct mem_cgroup *memcg,
2929 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002930{
2931 int err = 0;
2932 int memcg_id;
2933
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002934 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08002935 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002936 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002937
2938 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08002939 * For simplicity, we won't allow this to be disabled. It also can't
2940 * be changed if the cgroup has children already, or if tasks had
2941 * already joined.
2942 *
2943 * If tasks join before we set the limit, a person looking at
2944 * kmem.usage_in_bytes will have no way to determine when it took
2945 * place, which makes the value quite meaningless.
2946 *
2947 * After it first became limited, changes in the value of the limit are
2948 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08002949 */
Glauber Costa09998212013-02-22 16:34:55 -08002950 mutex_lock(&memcg_create_mutex);
Tejun Heo86d57552015-10-15 16:41:50 -04002951 if (cgroup_is_populated(memcg->css.cgroup) ||
Tejun Heoea280e72014-05-16 13:22:48 -04002952 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08002953 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08002954 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002955 if (err)
2956 goto out;
2957
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002958 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08002959 if (memcg_id < 0) {
2960 err = memcg_id;
2961 goto out;
2962 }
2963
Vladimir Davydovd6441632014-01-23 15:53:09 -08002964 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002965 * We couldn't have accounted to this cgroup, because it hasn't got
2966 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08002967 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002968 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002969 VM_BUG_ON(err);
2970
2971 static_key_slow_inc(&memcg_kmem_enabled_key);
2972 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002973 * A memory cgroup is considered kmem-active as soon as it gets
2974 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002975 * guarantee no one starts accounting before all call sites are
2976 * patched.
2977 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002978 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08002979 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002980 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002981out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08002982 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002983}
2984
Vladimir Davydovd6441632014-01-23 15:53:09 -08002985static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002986 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002987{
2988 int ret;
2989
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002990 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002991 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002992 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002993 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002994 ret = page_counter_limit(&memcg->kmem, limit);
2995 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002996 return ret;
2997}
2998
Glauber Costa55007d82012-12-18 14:22:38 -08002999static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08003000{
Glauber Costa55007d82012-12-18 14:22:38 -08003001 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003002 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003003
Glauber Costa510fc4e2012-12-18 14:21:47 -08003004 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003005 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08003006
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08003007 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08003008 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08003009 * If the parent cgroup is not kmem-active now, it cannot be activated
3010 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08003011 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08003012 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08003013 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
3014 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08003015 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003016}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003017#else
3018static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003019 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003020{
3021 return -EINVAL;
3022}
Hugh Dickins6d0439902013-02-22 16:35:50 -08003023#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08003024
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003025/*
3026 * The user of this function is...
3027 * RES_LIMIT.
3028 */
Tejun Heo451af502014-05-13 12:16:21 -04003029static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3030 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003031{
Tejun Heo451af502014-05-13 12:16:21 -04003032 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003033 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003034 int ret;
3035
Tejun Heo451af502014-05-13 12:16:21 -04003036 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003037 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003038 if (ret)
3039 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003040
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003041 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003042 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003043 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3044 ret = -EINVAL;
3045 break;
3046 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003047 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3048 case _MEM:
3049 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003050 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003051 case _MEMSWAP:
3052 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3053 break;
3054 case _KMEM:
3055 ret = memcg_update_kmem_limit(memcg, nr_pages);
3056 break;
3057 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003058 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003059 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003060 memcg->soft_limit = nr_pages;
3061 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003062 break;
3063 }
Tejun Heo451af502014-05-13 12:16:21 -04003064 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003065}
3066
Tejun Heo6770c642014-05-13 12:16:21 -04003067static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3068 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003069{
Tejun Heo6770c642014-05-13 12:16:21 -04003070 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003071 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003072
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003073 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3074 case _MEM:
3075 counter = &memcg->memory;
3076 break;
3077 case _MEMSWAP:
3078 counter = &memcg->memsw;
3079 break;
3080 case _KMEM:
3081 counter = &memcg->kmem;
3082 break;
3083 default:
3084 BUG();
3085 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003086
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003087 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003088 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003089 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003090 break;
3091 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003092 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003093 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003094 default:
3095 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003096 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003097
Tejun Heo6770c642014-05-13 12:16:21 -04003098 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003099}
3100
Tejun Heo182446d2013-08-08 20:11:24 -04003101static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003102 struct cftype *cft)
3103{
Tejun Heo182446d2013-08-08 20:11:24 -04003104 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003105}
3106
Daisuke Nishimura02491442010-03-10 15:22:17 -08003107#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003108static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003109 struct cftype *cft, u64 val)
3110{
Tejun Heo182446d2013-08-08 20:11:24 -04003111 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003112
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003113 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003114 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003115
Glauber Costaee5e8472013-02-22 16:34:50 -08003116 /*
3117 * No kind of locking is needed in here, because ->can_attach() will
3118 * check this value once in the beginning of the process, and then carry
3119 * on with stale data. This means that changes to this value will only
3120 * affect task migrations starting after the change.
3121 */
3122 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003123 return 0;
3124}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003125#else
Tejun Heo182446d2013-08-08 20:11:24 -04003126static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003127 struct cftype *cft, u64 val)
3128{
3129 return -ENOSYS;
3130}
3131#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003132
Ying Han406eb0c2011-05-26 16:25:37 -07003133#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003134static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003135{
Greg Thelen25485de2013-11-12 15:07:40 -08003136 struct numa_stat {
3137 const char *name;
3138 unsigned int lru_mask;
3139 };
3140
3141 static const struct numa_stat stats[] = {
3142 { "total", LRU_ALL },
3143 { "file", LRU_ALL_FILE },
3144 { "anon", LRU_ALL_ANON },
3145 { "unevictable", BIT(LRU_UNEVICTABLE) },
3146 };
3147 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003148 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003149 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003150 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003151
Greg Thelen25485de2013-11-12 15:07:40 -08003152 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3153 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3154 seq_printf(m, "%s=%lu", stat->name, nr);
3155 for_each_node_state(nid, N_MEMORY) {
3156 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3157 stat->lru_mask);
3158 seq_printf(m, " N%d=%lu", nid, nr);
3159 }
3160 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003161 }
Ying Han406eb0c2011-05-26 16:25:37 -07003162
Ying Han071aee12013-11-12 15:07:41 -08003163 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3164 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003165
Ying Han071aee12013-11-12 15:07:41 -08003166 nr = 0;
3167 for_each_mem_cgroup_tree(iter, memcg)
3168 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3169 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3170 for_each_node_state(nid, N_MEMORY) {
3171 nr = 0;
3172 for_each_mem_cgroup_tree(iter, memcg)
3173 nr += mem_cgroup_node_nr_lru_pages(
3174 iter, nid, stat->lru_mask);
3175 seq_printf(m, " N%d=%lu", nid, nr);
3176 }
3177 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003178 }
Ying Han406eb0c2011-05-26 16:25:37 -07003179
Ying Han406eb0c2011-05-26 16:25:37 -07003180 return 0;
3181}
3182#endif /* CONFIG_NUMA */
3183
Tejun Heo2da8ca82013-12-05 12:28:04 -05003184static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003185{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003186 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003187 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003188 struct mem_cgroup *mi;
3189 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003190
Greg Thelen0ca44b12015-02-11 15:25:58 -08003191 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3192 MEM_CGROUP_STAT_NSTATS);
3193 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3194 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003195 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3196
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003197 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003198 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003199 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003200 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
3201 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003202 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003203
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003204 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3205 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3206 mem_cgroup_read_events(memcg, i));
3207
3208 for (i = 0; i < NR_LRU_LISTS; i++)
3209 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3210 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3211
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003212 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003213 memory = memsw = PAGE_COUNTER_MAX;
3214 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3215 memory = min(memory, mi->memory.limit);
3216 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003217 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003218 seq_printf(m, "hierarchical_memory_limit %llu\n",
3219 (u64)memory * PAGE_SIZE);
3220 if (do_swap_account)
3221 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3222 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003223
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003224 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3225 long long val = 0;
3226
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003227 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003228 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003229 for_each_mem_cgroup_tree(mi, memcg)
3230 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3231 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
3232 }
3233
3234 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3235 unsigned long long val = 0;
3236
3237 for_each_mem_cgroup_tree(mi, memcg)
3238 val += mem_cgroup_read_events(mi, i);
3239 seq_printf(m, "total_%s %llu\n",
3240 mem_cgroup_events_names[i], val);
3241 }
3242
3243 for (i = 0; i < NR_LRU_LISTS; i++) {
3244 unsigned long long val = 0;
3245
3246 for_each_mem_cgroup_tree(mi, memcg)
3247 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3248 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003249 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003250
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003251#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003252 {
3253 int nid, zid;
3254 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003255 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003256 unsigned long recent_rotated[2] = {0, 0};
3257 unsigned long recent_scanned[2] = {0, 0};
3258
3259 for_each_online_node(nid)
3260 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003261 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003262 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003263
Hugh Dickins89abfab2012-05-29 15:06:53 -07003264 recent_rotated[0] += rstat->recent_rotated[0];
3265 recent_rotated[1] += rstat->recent_rotated[1];
3266 recent_scanned[0] += rstat->recent_scanned[0];
3267 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003268 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003269 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3270 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3271 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3272 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003273 }
3274#endif
3275
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003276 return 0;
3277}
3278
Tejun Heo182446d2013-08-08 20:11:24 -04003279static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3280 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003281{
Tejun Heo182446d2013-08-08 20:11:24 -04003282 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003283
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003284 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003285}
3286
Tejun Heo182446d2013-08-08 20:11:24 -04003287static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3288 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003289{
Tejun Heo182446d2013-08-08 20:11:24 -04003290 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003291
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003292 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003293 return -EINVAL;
3294
Linus Torvalds14208b02014-06-09 15:03:33 -07003295 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003296 memcg->swappiness = val;
3297 else
3298 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003299
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003300 return 0;
3301}
3302
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003303static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3304{
3305 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003306 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003307 int i;
3308
3309 rcu_read_lock();
3310 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003311 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003312 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003313 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003314
3315 if (!t)
3316 goto unlock;
3317
Johannes Weinerce00a962014-09-05 08:43:57 -04003318 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003319
3320 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003321 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003322 * If it's not true, a threshold was crossed after last
3323 * call of __mem_cgroup_threshold().
3324 */
Phil Carmody5407a562010-05-26 14:42:42 -07003325 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003326
3327 /*
3328 * Iterate backward over array of thresholds starting from
3329 * current_threshold and check if a threshold is crossed.
3330 * If none of thresholds below usage is crossed, we read
3331 * only one element of the array here.
3332 */
3333 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3334 eventfd_signal(t->entries[i].eventfd, 1);
3335
3336 /* i = current_threshold + 1 */
3337 i++;
3338
3339 /*
3340 * Iterate forward over array of thresholds starting from
3341 * current_threshold+1 and check if a threshold is crossed.
3342 * If none of thresholds above usage is crossed, we read
3343 * only one element of the array here.
3344 */
3345 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3346 eventfd_signal(t->entries[i].eventfd, 1);
3347
3348 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003349 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003350unlock:
3351 rcu_read_unlock();
3352}
3353
3354static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3355{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003356 while (memcg) {
3357 __mem_cgroup_threshold(memcg, false);
3358 if (do_swap_account)
3359 __mem_cgroup_threshold(memcg, true);
3360
3361 memcg = parent_mem_cgroup(memcg);
3362 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003363}
3364
3365static int compare_thresholds(const void *a, const void *b)
3366{
3367 const struct mem_cgroup_threshold *_a = a;
3368 const struct mem_cgroup_threshold *_b = b;
3369
Greg Thelen2bff24a2013-09-11 14:23:08 -07003370 if (_a->threshold > _b->threshold)
3371 return 1;
3372
3373 if (_a->threshold < _b->threshold)
3374 return -1;
3375
3376 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003377}
3378
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003379static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003380{
3381 struct mem_cgroup_eventfd_list *ev;
3382
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003383 spin_lock(&memcg_oom_lock);
3384
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003385 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003386 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003387
3388 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003389 return 0;
3390}
3391
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003392static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003393{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003394 struct mem_cgroup *iter;
3395
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003396 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003397 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003398}
3399
Tejun Heo59b6f872013-11-22 18:20:43 -05003400static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003401 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003402{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 struct mem_cgroup_thresholds *thresholds;
3404 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003405 unsigned long threshold;
3406 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003407 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
Johannes Weiner650c5e52015-02-11 15:26:03 -08003409 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003410 if (ret)
3411 return ret;
Shaohua Li2105f9a2015-10-15 15:28:29 -07003412 threshold <<= PAGE_SHIFT;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003413
3414 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003415
Johannes Weiner05b84302014-08-06 16:05:59 -07003416 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003417 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003418 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003419 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003420 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003421 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003422 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003423 BUG();
3424
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003425 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003426 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3428
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003429 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003430
3431 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003432 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003433 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003434 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003435 ret = -ENOMEM;
3436 goto unlock;
3437 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003438 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003439
3440 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003441 if (thresholds->primary) {
3442 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003443 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003444 }
3445
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003446 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003447 new->entries[size - 1].eventfd = eventfd;
3448 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003449
3450 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003451 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003452 compare_thresholds, NULL);
3453
3454 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003455 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003456 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003457 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003458 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003459 * new->current_threshold will not be used until
3460 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003461 * it here.
3462 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003463 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003464 } else
3465 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466 }
3467
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003468 /* Free old spare buffer and save old primary buffer as spare */
3469 kfree(thresholds->spare);
3470 thresholds->spare = thresholds->primary;
3471
3472 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003473
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003474 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003475 synchronize_rcu();
3476
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003477unlock:
3478 mutex_unlock(&memcg->thresholds_lock);
3479
3480 return ret;
3481}
3482
Tejun Heo59b6f872013-11-22 18:20:43 -05003483static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003484 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003485{
Tejun Heo59b6f872013-11-22 18:20:43 -05003486 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003487}
3488
Tejun Heo59b6f872013-11-22 18:20:43 -05003489static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003490 struct eventfd_ctx *eventfd, const char *args)
3491{
Tejun Heo59b6f872013-11-22 18:20:43 -05003492 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003493}
3494
Tejun Heo59b6f872013-11-22 18:20:43 -05003495static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003496 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003497{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003498 struct mem_cgroup_thresholds *thresholds;
3499 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003500 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003501 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003502
3503 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003504
3505 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003506 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003507 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003508 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003509 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003510 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003511 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003512 BUG();
3513
Anton Vorontsov371528c2012-02-24 05:14:46 +04003514 if (!thresholds->primary)
3515 goto unlock;
3516
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003517 /* Check if a threshold crossed before removing */
3518 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3519
3520 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003521 size = 0;
3522 for (i = 0; i < thresholds->primary->size; i++) {
3523 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003524 size++;
3525 }
3526
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003527 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003528
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003529 /* Set thresholds array to NULL if we don't have thresholds */
3530 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003531 kfree(new);
3532 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003533 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003534 }
3535
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003536 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003537
3538 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003539 new->current_threshold = -1;
3540 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3541 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003542 continue;
3543
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003544 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003545 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003546 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003547 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003548 * until rcu_assign_pointer(), so it's safe to increment
3549 * it here.
3550 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003551 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003552 }
3553 j++;
3554 }
3555
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003556swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003557 /* Swap primary and spare array */
3558 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003559
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003560 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003561
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003562 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003563 synchronize_rcu();
Martijn Coenen2c641f52016-01-15 16:57:49 -08003564
3565 /* If all events are unregistered, free the spare array */
3566 if (!new) {
3567 kfree(thresholds->spare);
3568 thresholds->spare = NULL;
3569 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003570unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003571 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003572}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003573
Tejun Heo59b6f872013-11-22 18:20:43 -05003574static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003575 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003576{
Tejun Heo59b6f872013-11-22 18:20:43 -05003577 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003578}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003579
Tejun Heo59b6f872013-11-22 18:20:43 -05003580static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003581 struct eventfd_ctx *eventfd)
3582{
Tejun Heo59b6f872013-11-22 18:20:43 -05003583 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003584}
3585
Tejun Heo59b6f872013-11-22 18:20:43 -05003586static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003587 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003588{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003589 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003590
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003591 event = kmalloc(sizeof(*event), GFP_KERNEL);
3592 if (!event)
3593 return -ENOMEM;
3594
Michal Hocko1af8efe2011-07-26 16:08:24 -07003595 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003596
3597 event->eventfd = eventfd;
3598 list_add(&event->list, &memcg->oom_notify);
3599
3600 /* already in OOM ? */
Tejun Heo47088812015-06-24 16:58:23 -07003601 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003602 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003603 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003604
3605 return 0;
3606}
3607
Tejun Heo59b6f872013-11-22 18:20:43 -05003608static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003609 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003610{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003611 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003612
Michal Hocko1af8efe2011-07-26 16:08:24 -07003613 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003614
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003615 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003616 if (ev->eventfd == eventfd) {
3617 list_del(&ev->list);
3618 kfree(ev);
3619 }
3620 }
3621
Michal Hocko1af8efe2011-07-26 16:08:24 -07003622 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003623}
3624
Tejun Heo2da8ca82013-12-05 12:28:04 -05003625static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003626{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003627 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003628
Tejun Heo791badb2013-12-05 12:28:02 -05003629 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heo47088812015-06-24 16:58:23 -07003630 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003631 return 0;
3632}
3633
Tejun Heo182446d2013-08-08 20:11:24 -04003634static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003635 struct cftype *cft, u64 val)
3636{
Tejun Heo182446d2013-08-08 20:11:24 -04003637 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003638
3639 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003640 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003641 return -EINVAL;
3642
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003643 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003644 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003645 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003646
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003647 return 0;
3648}
3649
Andrew Mortonc255a452012-07-31 16:43:02 -07003650#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003651static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003652{
Glauber Costa55007d82012-12-18 14:22:38 -08003653 int ret;
3654
Glauber Costa55007d82012-12-18 14:22:38 -08003655 ret = memcg_propagate_kmem(memcg);
3656 if (ret)
3657 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003658
Glauber Costa1d62e432012-04-09 19:36:33 -03003659 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003660}
Glauber Costae5671df2011-12-11 21:47:01 +00003661
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003662static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3663{
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003664 struct cgroup_subsys_state *css;
3665 struct mem_cgroup *parent, *child;
3666 int kmemcg_id;
3667
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003668 if (!memcg->kmem_acct_active)
3669 return;
3670
3671 /*
3672 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3673 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3674 * guarantees no cache will be created for this cgroup after we are
3675 * done (see memcg_create_kmem_cache()).
3676 */
3677 memcg->kmem_acct_active = false;
3678
3679 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003680
3681 kmemcg_id = memcg->kmemcg_id;
3682 BUG_ON(kmemcg_id < 0);
3683
3684 parent = parent_mem_cgroup(memcg);
3685 if (!parent)
3686 parent = root_mem_cgroup;
3687
3688 /*
3689 * Change kmemcg_id of this cgroup and all its descendants to the
3690 * parent's id, and then move all entries from this cgroup's list_lrus
3691 * to ones of the parent. After we have finished, all list_lrus
3692 * corresponding to this cgroup are guaranteed to remain empty. The
3693 * ordering is imposed by list_lru_node->lock taken by
3694 * memcg_drain_all_list_lrus().
3695 */
3696 css_for_each_descendant_pre(css, &memcg->css) {
3697 child = mem_cgroup_from_css(css);
3698 BUG_ON(child->kmemcg_id != kmemcg_id);
3699 child->kmemcg_id = parent->kmemcg_id;
3700 if (!memcg->use_hierarchy)
3701 break;
3702 }
3703 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3704
3705 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003706}
3707
Li Zefan10d5ebf2013-07-08 16:00:33 -07003708static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003709{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003710 if (memcg->kmem_acct_activated) {
3711 memcg_destroy_kmem_caches(memcg);
3712 static_key_slow_dec(&memcg_kmem_enabled_key);
3713 WARN_ON(page_counter_read(&memcg->kmem));
3714 }
Glauber Costa1d62e432012-04-09 19:36:33 -03003715 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003716}
Glauber Costae5671df2011-12-11 21:47:01 +00003717#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003718static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003719{
3720 return 0;
3721}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003722
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003723static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3724{
3725}
3726
Li Zefan10d5ebf2013-07-08 16:00:33 -07003727static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3728{
3729}
Glauber Costae5671df2011-12-11 21:47:01 +00003730#endif
3731
Tejun Heo6b385782015-05-22 17:13:37 -04003732#ifdef CONFIG_CGROUP_WRITEBACK
3733
3734struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3735{
3736 return &memcg->cgwb_list;
3737}
3738
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003739static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3740{
3741 return wb_domain_init(&memcg->cgwb_domain, gfp);
3742}
3743
3744static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3745{
3746 wb_domain_exit(&memcg->cgwb_domain);
3747}
3748
Tejun Heoe65e8b62015-05-22 18:23:34 -04003749static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3750{
3751 wb_domain_size_changed(&memcg->cgwb_domain);
3752}
3753
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003754struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3755{
3756 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3757
3758 if (!memcg->css.parent)
3759 return NULL;
3760
3761 return &memcg->cgwb_domain;
3762}
3763
Tejun Heo6a139c82015-05-22 18:23:35 -04003764/**
3765 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3766 * @wb: bdi_writeback in question
Tejun Heo8debf2d2015-09-29 13:04:26 -04003767 * @pfilepages: out parameter for number of file pages
3768 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heo6a139c82015-05-22 18:23:35 -04003769 * @pdirty: out parameter for number of dirty pages
3770 * @pwriteback: out parameter for number of pages under writeback
3771 *
Tejun Heo8debf2d2015-09-29 13:04:26 -04003772 * Determine the numbers of file, headroom, dirty, and writeback pages in
3773 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3774 * is a bit more involved.
Tejun Heo6a139c82015-05-22 18:23:35 -04003775 *
Tejun Heo8debf2d2015-09-29 13:04:26 -04003776 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3777 * headroom is calculated as the lowest headroom of itself and the
3778 * ancestors. Note that this doesn't consider the actual amount of
3779 * available memory in the system. The caller should further cap
3780 * *@pheadroom accordingly.
Tejun Heo6a139c82015-05-22 18:23:35 -04003781 */
Tejun Heo8debf2d2015-09-29 13:04:26 -04003782void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3783 unsigned long *pheadroom, unsigned long *pdirty,
3784 unsigned long *pwriteback)
Tejun Heo6a139c82015-05-22 18:23:35 -04003785{
3786 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3787 struct mem_cgroup *parent;
Tejun Heo6a139c82015-05-22 18:23:35 -04003788
3789 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3790
3791 /* this should eventually include NR_UNSTABLE_NFS */
3792 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heo8debf2d2015-09-29 13:04:26 -04003793 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3794 (1 << LRU_ACTIVE_FILE));
3795 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heo6a139c82015-05-22 18:23:35 -04003796
Tejun Heo6a139c82015-05-22 18:23:35 -04003797 while ((parent = parent_mem_cgroup(memcg))) {
3798 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3799 unsigned long used = page_counter_read(&memcg->memory);
3800
Tejun Heo8debf2d2015-09-29 13:04:26 -04003801 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heo6a139c82015-05-22 18:23:35 -04003802 memcg = parent;
3803 }
Tejun Heo6a139c82015-05-22 18:23:35 -04003804}
3805
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003806#else /* CONFIG_CGROUP_WRITEBACK */
3807
3808static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3809{
3810 return 0;
3811}
3812
3813static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3814{
3815}
3816
Tejun Heoe65e8b62015-05-22 18:23:34 -04003817static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3818{
3819}
3820
Tejun Heo6b385782015-05-22 17:13:37 -04003821#endif /* CONFIG_CGROUP_WRITEBACK */
3822
Tejun Heo79bd9812013-11-22 18:20:42 -05003823/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003824 * DO NOT USE IN NEW FILES.
3825 *
3826 * "cgroup.event_control" implementation.
3827 *
3828 * This is way over-engineered. It tries to support fully configurable
3829 * events for each user. Such level of flexibility is completely
3830 * unnecessary especially in the light of the planned unified hierarchy.
3831 *
3832 * Please deprecate this and replace with something simpler if at all
3833 * possible.
3834 */
3835
3836/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003837 * Unregister event and free resources.
3838 *
3839 * Gets called from workqueue.
3840 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003841static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003842{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003843 struct mem_cgroup_event *event =
3844 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003845 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003846
3847 remove_wait_queue(event->wqh, &event->wait);
3848
Tejun Heo59b6f872013-11-22 18:20:43 -05003849 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003850
3851 /* Notify userspace the event is going away. */
3852 eventfd_signal(event->eventfd, 1);
3853
3854 eventfd_ctx_put(event->eventfd);
3855 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003856 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003857}
3858
3859/*
3860 * Gets called on POLLHUP on eventfd when user closes it.
3861 *
3862 * Called with wqh->lock held and interrupts disabled.
3863 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003864static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3865 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003866{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003867 struct mem_cgroup_event *event =
3868 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003869 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003870 unsigned long flags = (unsigned long)key;
3871
3872 if (flags & POLLHUP) {
3873 /*
3874 * If the event has been detached at cgroup removal, we
3875 * can simply return knowing the other side will cleanup
3876 * for us.
3877 *
3878 * We can't race against event freeing since the other
3879 * side will require wqh->lock via remove_wait_queue(),
3880 * which we hold.
3881 */
Tejun Heofba94802013-11-22 18:20:43 -05003882 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003883 if (!list_empty(&event->list)) {
3884 list_del_init(&event->list);
3885 /*
3886 * We are in atomic context, but cgroup_event_remove()
3887 * may sleep, so we have to call it in workqueue.
3888 */
3889 schedule_work(&event->remove);
3890 }
Tejun Heofba94802013-11-22 18:20:43 -05003891 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003892 }
3893
3894 return 0;
3895}
3896
Tejun Heo3bc942f2013-11-22 18:20:44 -05003897static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003898 wait_queue_head_t *wqh, poll_table *pt)
3899{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003900 struct mem_cgroup_event *event =
3901 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003902
3903 event->wqh = wqh;
3904 add_wait_queue(wqh, &event->wait);
3905}
3906
3907/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003908 * DO NOT USE IN NEW FILES.
3909 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003910 * Parse input and register new cgroup event handler.
3911 *
3912 * Input must be in format '<event_fd> <control_fd> <args>'.
3913 * Interpretation of args is defined by control file implementation.
3914 */
Tejun Heo451af502014-05-13 12:16:21 -04003915static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3916 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003917{
Tejun Heo451af502014-05-13 12:16:21 -04003918 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003919 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003920 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003921 struct cgroup_subsys_state *cfile_css;
3922 unsigned int efd, cfd;
3923 struct fd efile;
3924 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003925 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003926 char *endp;
3927 int ret;
3928
Tejun Heo451af502014-05-13 12:16:21 -04003929 buf = strstrip(buf);
3930
3931 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003932 if (*endp != ' ')
3933 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003934 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003935
Tejun Heo451af502014-05-13 12:16:21 -04003936 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003937 if ((*endp != ' ') && (*endp != '\0'))
3938 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003939 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003940
3941 event = kzalloc(sizeof(*event), GFP_KERNEL);
3942 if (!event)
3943 return -ENOMEM;
3944
Tejun Heo59b6f872013-11-22 18:20:43 -05003945 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003946 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003947 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3948 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3949 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003950
3951 efile = fdget(efd);
3952 if (!efile.file) {
3953 ret = -EBADF;
3954 goto out_kfree;
3955 }
3956
3957 event->eventfd = eventfd_ctx_fileget(efile.file);
3958 if (IS_ERR(event->eventfd)) {
3959 ret = PTR_ERR(event->eventfd);
3960 goto out_put_efile;
3961 }
3962
3963 cfile = fdget(cfd);
3964 if (!cfile.file) {
3965 ret = -EBADF;
3966 goto out_put_eventfd;
3967 }
3968
3969 /* the process need read permission on control file */
3970 /* AV: shouldn't we check that it's been opened for read instead? */
3971 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3972 if (ret < 0)
3973 goto out_put_cfile;
3974
Tejun Heo79bd9812013-11-22 18:20:42 -05003975 /*
Tejun Heofba94802013-11-22 18:20:43 -05003976 * Determine the event callbacks and set them in @event. This used
3977 * to be done via struct cftype but cgroup core no longer knows
3978 * about these events. The following is crude but the whole thing
3979 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003980 *
3981 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003982 */
Al Virob5830432014-10-31 01:22:04 -04003983 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003984
3985 if (!strcmp(name, "memory.usage_in_bytes")) {
3986 event->register_event = mem_cgroup_usage_register_event;
3987 event->unregister_event = mem_cgroup_usage_unregister_event;
3988 } else if (!strcmp(name, "memory.oom_control")) {
3989 event->register_event = mem_cgroup_oom_register_event;
3990 event->unregister_event = mem_cgroup_oom_unregister_event;
3991 } else if (!strcmp(name, "memory.pressure_level")) {
3992 event->register_event = vmpressure_register_event;
3993 event->unregister_event = vmpressure_unregister_event;
3994 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003995 event->register_event = memsw_cgroup_usage_register_event;
3996 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003997 } else {
3998 ret = -EINVAL;
3999 goto out_put_cfile;
4000 }
4001
4002 /*
Tejun Heob5557c42013-11-22 18:20:42 -05004003 * Verify @cfile should belong to @css. Also, remaining events are
4004 * automatically removed on cgroup destruction but the removal is
4005 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05004006 */
Al Virob5830432014-10-31 01:22:04 -04004007 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04004008 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05004009 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05004010 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05004011 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05004012 if (cfile_css != css) {
4013 css_put(cfile_css);
4014 goto out_put_cfile;
4015 }
Tejun Heo79bd9812013-11-22 18:20:42 -05004016
Tejun Heo451af502014-05-13 12:16:21 -04004017 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05004018 if (ret)
4019 goto out_put_css;
4020
4021 efile.file->f_op->poll(efile.file, &event->pt);
4022
Tejun Heofba94802013-11-22 18:20:43 -05004023 spin_lock(&memcg->event_list_lock);
4024 list_add(&event->list, &memcg->event_list);
4025 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004026
4027 fdput(cfile);
4028 fdput(efile);
4029
Tejun Heo451af502014-05-13 12:16:21 -04004030 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004031
4032out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004033 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004034out_put_cfile:
4035 fdput(cfile);
4036out_put_eventfd:
4037 eventfd_ctx_put(event->eventfd);
4038out_put_efile:
4039 fdput(efile);
4040out_kfree:
4041 kfree(event);
4042
4043 return ret;
4044}
4045
Johannes Weiner241994ed2015-02-11 15:26:06 -08004046static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004047 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004048 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004049 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004050 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004051 },
4052 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004053 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004054 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004055 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004056 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004057 },
4058 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004059 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004060 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004061 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004062 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004063 },
4064 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004065 .name = "soft_limit_in_bytes",
4066 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004067 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004068 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004069 },
4070 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004071 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004072 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004073 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004074 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004075 },
Balbir Singh8697d332008-02-07 00:13:59 -08004076 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004077 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004078 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004079 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004080 {
4081 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004082 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004083 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004084 {
4085 .name = "use_hierarchy",
4086 .write_u64 = mem_cgroup_hierarchy_write,
4087 .read_u64 = mem_cgroup_hierarchy_read,
4088 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004089 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004090 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004091 .write = memcg_write_event_control,
Tejun Heo6c106f52015-09-18 17:54:23 -04004092 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004093 },
4094 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004095 .name = "swappiness",
4096 .read_u64 = mem_cgroup_swappiness_read,
4097 .write_u64 = mem_cgroup_swappiness_write,
4098 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004099 {
4100 .name = "move_charge_at_immigrate",
4101 .read_u64 = mem_cgroup_move_charge_read,
4102 .write_u64 = mem_cgroup_move_charge_write,
4103 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004104 {
4105 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004106 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004107 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004108 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4109 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004110 {
4111 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004112 },
Ying Han406eb0c2011-05-26 16:25:37 -07004113#ifdef CONFIG_NUMA
4114 {
4115 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004116 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004117 },
4118#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004119#ifdef CONFIG_MEMCG_KMEM
4120 {
4121 .name = "kmem.limit_in_bytes",
4122 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004123 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004124 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004125 },
4126 {
4127 .name = "kmem.usage_in_bytes",
4128 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004129 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004130 },
4131 {
4132 .name = "kmem.failcnt",
4133 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004134 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004135 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004136 },
4137 {
4138 .name = "kmem.max_usage_in_bytes",
4139 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004140 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004141 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004142 },
Glauber Costa749c5412012-12-18 14:23:01 -08004143#ifdef CONFIG_SLABINFO
4144 {
4145 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004146 .seq_start = slab_start,
4147 .seq_next = slab_next,
4148 .seq_stop = slab_stop,
4149 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004150 },
4151#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004152#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004153 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004154};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004155
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004156static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004157{
4158 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004159 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004160 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004161 /*
4162 * This routine is called against possible nodes.
4163 * But it's BUG to call kmalloc() against offline node.
4164 *
4165 * TODO: this routine can waste much memory for nodes which will
4166 * never be onlined. It's better to use memory hotplug callback
4167 * function.
4168 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004169 if (!node_state(node, N_NORMAL_MEMORY))
4170 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004171 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004172 if (!pn)
4173 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004174
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004175 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4176 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004177 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004178 mz->usage_in_excess = 0;
4179 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004180 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004181 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004182 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004183 return 0;
4184}
4185
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004186static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004187{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004188 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004189}
4190
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004191static struct mem_cgroup *mem_cgroup_alloc(void)
4192{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004193 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004194 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004195
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004196 size = sizeof(struct mem_cgroup);
4197 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004198
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004199 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004200 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004201 return NULL;
4202
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004203 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4204 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004205 goto out_free;
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004206
4207 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4208 goto out_free_stat;
4209
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004210 spin_lock_init(&memcg->pcp_counter_lock);
4211 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004212
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004213out_free_stat:
4214 free_percpu(memcg->stat);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004215out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004216 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004217 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004218}
4219
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004220/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004221 * At destroying mem_cgroup, references from swap_cgroup can remain.
4222 * (scanning all at force_empty is too costly...)
4223 *
4224 * Instead of clearing all references at force_empty, we remember
4225 * the number of reference from swap_cgroup and free mem_cgroup when
4226 * it goes down to 0.
4227 *
4228 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004229 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004230
4231static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004232{
Glauber Costac8b2a362012-12-18 14:22:13 -08004233 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004234
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004235 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004236
4237 for_each_node(node)
4238 free_mem_cgroup_per_zone_info(memcg, node);
4239
4240 free_percpu(memcg->stat);
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004241 memcg_wb_domain_exit(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004242 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004243}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004244
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004245/*
4246 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4247 */
Glauber Costae1aab162011-12-11 21:47:03 +00004248struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004249{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004250 if (!memcg->memory.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004251 return NULL;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004252 return mem_cgroup_from_counter(memcg->memory.parent, memory);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004253}
Glauber Costae1aab162011-12-11 21:47:03 +00004254EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004255
Li Zefan0eb253e2009-01-15 13:51:25 -08004256static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004257mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004258{
Glauber Costad142e3e2013-02-22 16:34:52 -08004259 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004260 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004261 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004262
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004263 memcg = mem_cgroup_alloc();
4264 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004265 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004266
Bob Liu3ed28fa2012-01-12 17:19:04 -08004267 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004268 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004269 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004270
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004271 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004272 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004273 root_mem_cgroup = memcg;
Tejun Heodbee2272015-05-22 17:13:20 -04004274 mem_cgroup_root_css = &memcg->css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004275 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004276 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004277 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004278 page_counter_init(&memcg->memsw, NULL);
4279 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004280 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004281
Glauber Costad142e3e2013-02-22 16:34:52 -08004282 memcg->last_scanned_node = MAX_NUMNODES;
4283 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004284 memcg->move_charge_at_immigrate = 0;
4285 mutex_init(&memcg->thresholds_lock);
4286 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004287 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004288 INIT_LIST_HEAD(&memcg->event_list);
4289 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004290#ifdef CONFIG_MEMCG_KMEM
4291 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004292#endif
Tejun Heo6b385782015-05-22 17:13:37 -04004293#ifdef CONFIG_CGROUP_WRITEBACK
4294 INIT_LIST_HEAD(&memcg->cgwb_list);
4295#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004296 return &memcg->css;
4297
4298free_out:
4299 __mem_cgroup_free(memcg);
4300 return ERR_PTR(error);
4301}
4302
4303static int
Tejun Heoeb954192013-08-08 20:11:23 -04004304mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004305{
Tejun Heoeb954192013-08-08 20:11:23 -04004306 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004307 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004308 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004309
Tejun Heo15a4c832014-05-04 15:09:14 -04004310 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004311 return -ENOSPC;
4312
Tejun Heo63876982013-08-08 20:11:23 -04004313 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004314 return 0;
4315
Glauber Costa09998212013-02-22 16:34:55 -08004316 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004317
4318 memcg->use_hierarchy = parent->use_hierarchy;
4319 memcg->oom_kill_disable = parent->oom_kill_disable;
4320 memcg->swappiness = mem_cgroup_swappiness(parent);
4321
4322 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004323 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004324 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004325 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004326 page_counter_init(&memcg->memsw, &parent->memsw);
4327 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004328
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004329 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004330 * No need to take a reference to the parent because cgroup
4331 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004332 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004333 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004334 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004335 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004336 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004337 page_counter_init(&memcg->memsw, NULL);
4338 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004339 /*
4340 * Deeper hierachy with use_hierarchy == false doesn't make
4341 * much sense so let cgroup subsystem know about this
4342 * unfortunate state in our controller.
4343 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004344 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004345 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004346 }
Glauber Costa09998212013-02-22 16:34:55 -08004347 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004348
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004349 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4350 if (ret)
4351 return ret;
4352
4353 /*
4354 * Make sure the memcg is initialized: mem_cgroup_iter()
4355 * orders reading memcg->initialized against its callers
4356 * reading the memcg members.
4357 */
4358 smp_store_release(&memcg->initialized, 1);
4359
4360 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004361}
4362
Tejun Heoeb954192013-08-08 20:11:23 -04004363static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004364{
Tejun Heoeb954192013-08-08 20:11:23 -04004365 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004366 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004367
4368 /*
4369 * Unregister events and notify userspace.
4370 * Notify userspace about cgroup removing only after rmdir of cgroup
4371 * directory to avoid race between userspace and kernelspace.
4372 */
Tejun Heofba94802013-11-22 18:20:43 -05004373 spin_lock(&memcg->event_list_lock);
4374 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004375 list_del_init(&event->list);
4376 schedule_work(&event->remove);
4377 }
Tejun Heofba94802013-11-22 18:20:43 -05004378 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004379
Michal Hocko33cb8762013-07-31 13:53:51 -07004380 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004381
4382 memcg_deactivate_kmem(memcg);
Tejun Heo6b385782015-05-22 17:13:37 -04004383
4384 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004385}
4386
Tejun Heoeb954192013-08-08 20:11:23 -04004387static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004388{
Tejun Heoeb954192013-08-08 20:11:23 -04004389 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004390
Li Zefan10d5ebf2013-07-08 16:00:33 -07004391 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07004392 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004393}
4394
Tejun Heo1ced9532014-07-08 18:02:57 -04004395/**
4396 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4397 * @css: the target css
4398 *
4399 * Reset the states of the mem_cgroup associated with @css. This is
4400 * invoked when the userland requests disabling on the default hierarchy
4401 * but the memcg is pinned through dependency. The memcg should stop
4402 * applying policies and should revert to the vanilla state as it may be
4403 * made visible again.
4404 *
4405 * The current implementation only resets the essential configurations.
4406 * This needs to be expanded to cover all the visible parts.
4407 */
4408static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4409{
4410 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4411
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004412 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4413 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4414 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004415 memcg->low = 0;
4416 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004417 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heoe65e8b62015-05-22 18:23:34 -04004418 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004419}
4420
Daisuke Nishimura02491442010-03-10 15:22:17 -08004421#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004422/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004423static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004424{
Johannes Weiner05b84302014-08-06 16:05:59 -07004425 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004426
4427 /* Try a single bulk charge without reclaim first */
Johannes Weiner00501b52014-08-08 14:19:20 -07004428 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004429 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004430 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004431 return ret;
4432 }
Johannes Weiner692e7c42014-08-06 16:05:57 -07004433 if (ret == -EINTR) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004434 cancel_charge(root_mem_cgroup, count);
Johannes Weiner692e7c42014-08-06 16:05:57 -07004435 return ret;
4436 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004437
4438 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004439 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004440 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
Johannes Weiner9476db92014-08-06 16:05:55 -07004441 /*
4442 * In case of failure, any residual charges against
4443 * mc.to will be dropped by mem_cgroup_clear_mc()
Johannes Weiner692e7c42014-08-06 16:05:57 -07004444 * later on. However, cancel any charges that are
4445 * bypassed to root right away or they'll be lost.
Johannes Weiner9476db92014-08-06 16:05:55 -07004446 */
Johannes Weiner692e7c42014-08-06 16:05:57 -07004447 if (ret == -EINTR)
Johannes Weiner00501b52014-08-08 14:19:20 -07004448 cancel_charge(root_mem_cgroup, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004449 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004450 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004451 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004452 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004453 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004454 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004455}
4456
4457/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004458 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004459 * @vma: the vma the pte to be checked belongs
4460 * @addr: the address corresponding to the pte to be checked
4461 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004462 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004463 *
4464 * Returns
4465 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4466 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4467 * move charge. if @target is not NULL, the page is stored in target->page
4468 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004469 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4470 * target for charge migration. if @target is not NULL, the entry is stored
4471 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004472 *
4473 * Called with pte lock held.
4474 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004475union mc_target {
4476 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004477 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004478};
4479
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004480enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004481 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004482 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004483 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004484};
4485
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004486static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4487 unsigned long addr, pte_t ptent)
4488{
4489 struct page *page = vm_normal_page(vma, addr, ptent);
4490
4491 if (!page || !page_mapped(page))
4492 return NULL;
4493 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004494 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004495 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004496 } else {
4497 if (!(mc.flags & MOVE_FILE))
4498 return NULL;
4499 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004500 if (!get_page_unless_zero(page))
4501 return NULL;
4502
4503 return page;
4504}
4505
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004506#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004507static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4508 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4509{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004510 struct page *page = NULL;
4511 swp_entry_t ent = pte_to_swp_entry(ptent);
4512
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004513 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004514 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004515 /*
4516 * Because lookup_swap_cache() updates some statistics counter,
4517 * we call find_get_page() with swapper_space directly.
4518 */
Shaohua Li33806f02013-02-22 16:34:37 -08004519 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004520 if (do_swap_account)
4521 entry->val = ent.val;
4522
4523 return page;
4524}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004525#else
4526static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4527 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4528{
4529 return NULL;
4530}
4531#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004532
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004533static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4534 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4535{
4536 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004537 struct address_space *mapping;
4538 pgoff_t pgoff;
4539
4540 if (!vma->vm_file) /* anonymous vma */
4541 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004542 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004543 return NULL;
4544
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004545 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004546 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004547
4548 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004549#ifdef CONFIG_SWAP
4550 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004551 if (shmem_mapping(mapping)) {
4552 page = find_get_entry(mapping, pgoff);
4553 if (radix_tree_exceptional_entry(page)) {
4554 swp_entry_t swp = radix_to_swp_entry(page);
4555 if (do_swap_account)
4556 *entry = swp;
4557 page = find_get_page(swap_address_space(swp), swp.val);
4558 }
4559 } else
4560 page = find_get_page(mapping, pgoff);
4561#else
4562 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004563#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004564 return page;
4565}
4566
Chen Gangb1b0dea2015-04-14 15:47:35 -07004567/**
4568 * mem_cgroup_move_account - move account of the page
4569 * @page: the page
4570 * @nr_pages: number of regular pages (>1 for huge pages)
4571 * @from: mem_cgroup which the page is moved from.
4572 * @to: mem_cgroup which the page is moved to. @from != @to.
4573 *
4574 * The caller must confirm following.
4575 * - page is not on LRU (isolate_page() is useful.)
4576 * - compound_lock is held when nr_pages > 1
4577 *
4578 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4579 * from old cgroup.
4580 */
4581static int mem_cgroup_move_account(struct page *page,
4582 unsigned int nr_pages,
4583 struct mem_cgroup *from,
4584 struct mem_cgroup *to)
4585{
4586 unsigned long flags;
4587 int ret;
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004588 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004589
4590 VM_BUG_ON(from == to);
4591 VM_BUG_ON_PAGE(PageLRU(page), page);
4592 /*
4593 * The page is isolated from LRU. So, collapse function
4594 * will not handle this page. But page splitting can happen.
4595 * Do this check under compound_page_lock(). The caller should
4596 * hold it.
4597 */
4598 ret = -EBUSY;
4599 if (nr_pages > 1 && !PageTransHuge(page))
4600 goto out;
4601
4602 /*
4603 * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
4604 * of its source page while we change it: page migration takes
4605 * both pages off the LRU, but page cache replacement doesn't.
4606 */
4607 if (!trylock_page(page))
4608 goto out;
4609
4610 ret = -EINVAL;
4611 if (page->mem_cgroup != from)
4612 goto out_unlock;
4613
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004614 anon = PageAnon(page);
4615
Chen Gangb1b0dea2015-04-14 15:47:35 -07004616 spin_lock_irqsave(&from->move_lock, flags);
4617
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004618 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004619 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4620 nr_pages);
4621 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4622 nr_pages);
4623 }
4624
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004625 /*
4626 * move_lock grabbed above and caller set from->moving_account, so
4627 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4628 * So mapping should be stable for dirty pages.
4629 */
4630 if (!anon && PageDirty(page)) {
4631 struct address_space *mapping = page_mapping(page);
4632
4633 if (mapping_cap_account_dirty(mapping)) {
4634 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4635 nr_pages);
4636 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4637 nr_pages);
4638 }
4639 }
4640
Chen Gangb1b0dea2015-04-14 15:47:35 -07004641 if (PageWriteback(page)) {
4642 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4643 nr_pages);
4644 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4645 nr_pages);
4646 }
4647
4648 /*
4649 * It is safe to change page->mem_cgroup here because the page
4650 * is referenced, charged, and isolated - we can't race with
4651 * uncharging, charging, migration, or LRU putback.
4652 */
4653
4654 /* caller should have done css_get */
4655 page->mem_cgroup = to;
4656 spin_unlock_irqrestore(&from->move_lock, flags);
4657
4658 ret = 0;
4659
4660 local_irq_disable();
4661 mem_cgroup_charge_statistics(to, page, nr_pages);
4662 memcg_check_events(to, page);
4663 mem_cgroup_charge_statistics(from, page, -nr_pages);
4664 memcg_check_events(from, page);
4665 local_irq_enable();
4666out_unlock:
4667 unlock_page(page);
4668out:
4669 return ret;
4670}
4671
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004672static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004673 unsigned long addr, pte_t ptent, union mc_target *target)
4674{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004675 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004676 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004677 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004678
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004679 if (pte_present(ptent))
4680 page = mc_handle_present_pte(vma, addr, ptent);
4681 else if (is_swap_pte(ptent))
4682 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004683 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004684 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004685
4686 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004687 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004688 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004689 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004690 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004691 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004692 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004693 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004694 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004695 ret = MC_TARGET_PAGE;
4696 if (target)
4697 target->page = page;
4698 }
4699 if (!ret || !target)
4700 put_page(page);
4701 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004702 /* There is a swap entry and a page doesn't exist or isn't charged */
4703 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004704 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004705 ret = MC_TARGET_SWAP;
4706 if (target)
4707 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004708 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004709 return ret;
4710}
4711
Naoya Horiguchi12724852012-03-21 16:34:28 -07004712#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4713/*
4714 * We don't consider swapping or file mapped pages because THP does not
4715 * support them for now.
4716 * Caller should make sure that pmd_trans_huge(pmd) is true.
4717 */
4718static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4719 unsigned long addr, pmd_t pmd, union mc_target *target)
4720{
4721 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004722 enum mc_target_type ret = MC_TARGET_NONE;
4723
4724 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004725 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004726 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004727 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004728 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004729 ret = MC_TARGET_PAGE;
4730 if (target) {
4731 get_page(page);
4732 target->page = page;
4733 }
4734 }
4735 return ret;
4736}
4737#else
4738static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4739 unsigned long addr, pmd_t pmd, union mc_target *target)
4740{
4741 return MC_TARGET_NONE;
4742}
4743#endif
4744
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004745static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4746 unsigned long addr, unsigned long end,
4747 struct mm_walk *walk)
4748{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004749 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004750 pte_t *pte;
4751 spinlock_t *ptl;
4752
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004753 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004754 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4755 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004756 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004757 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004758 }
Dave Hansen03319322011-03-22 16:32:56 -07004759
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004760 if (pmd_trans_unstable(pmd))
4761 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004762 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4763 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004764 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004765 mc.precharge++; /* increment precharge temporarily */
4766 pte_unmap_unlock(pte - 1, ptl);
4767 cond_resched();
4768
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004769 return 0;
4770}
4771
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004772static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4773{
4774 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004775
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004776 struct mm_walk mem_cgroup_count_precharge_walk = {
4777 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4778 .mm = mm,
4779 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004780 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004781 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004782 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004783
4784 precharge = mc.precharge;
4785 mc.precharge = 0;
4786
4787 return precharge;
4788}
4789
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004790static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4791{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004792 unsigned long precharge = mem_cgroup_count_precharge(mm);
4793
4794 VM_BUG_ON(mc.moving_task);
4795 mc.moving_task = current;
4796 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004797}
4798
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004799/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4800static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004801{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004802 struct mem_cgroup *from = mc.from;
4803 struct mem_cgroup *to = mc.to;
4804
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004805 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004806 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004807 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004808 mc.precharge = 0;
4809 }
4810 /*
4811 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4812 * we must uncharge here.
4813 */
4814 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004815 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004816 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004817 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004818 /* we must fixup refcnts and charges */
4819 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004820 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004821 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004822 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004823
Johannes Weiner05b84302014-08-06 16:05:59 -07004824 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004825 * we charged both to->memory and to->memsw, so we
4826 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004827 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004828 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004829 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004830
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004831 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004832
Li Zefan40503772013-07-08 16:00:34 -07004833 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004834 mc.moved_swap = 0;
4835 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004836 memcg_oom_recover(from);
4837 memcg_oom_recover(to);
4838 wake_up_all(&mc.waitq);
4839}
4840
4841static void mem_cgroup_clear_mc(void)
4842{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004843 /*
4844 * we must clear moving_task before waking up waiters at the end of
4845 * task migration.
4846 */
4847 mc.moving_task = NULL;
4848 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004849 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004850 mc.from = NULL;
4851 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004852 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004853}
4854
Tejun Heo041bdb52015-12-03 10:18:21 -05004855static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004856{
Tejun Heo041bdb52015-12-03 10:18:21 -05004857 struct cgroup_subsys_state *css;
4858 struct mem_cgroup *memcg;
Tejun Heo67d6b192015-09-08 15:01:10 -07004859 struct mem_cgroup *from;
Tejun Heo2a910252015-09-11 15:00:19 -04004860 struct task_struct *leader, *p;
Tejun Heo67d6b192015-09-08 15:01:10 -07004861 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004862 unsigned long move_flags;
Tejun Heo67d6b192015-09-08 15:01:10 -07004863 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004864
Tejun Heo041bdb52015-12-03 10:18:21 -05004865 /* charge immigration isn't supported on the default hierarchy */
4866 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo67d6b192015-09-08 15:01:10 -07004867 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004868
Tejun Heo2a910252015-09-11 15:00:19 -04004869 /*
4870 * Multi-process migrations only happen on the default hierarchy
4871 * where charge immigration is not used. Perform charge
4872 * immigration if @tset contains a leader and whine if there are
4873 * multiple.
4874 */
4875 p = NULL;
Tejun Heo041bdb52015-12-03 10:18:21 -05004876 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo2a910252015-09-11 15:00:19 -04004877 WARN_ON_ONCE(p);
4878 p = leader;
Tejun Heo041bdb52015-12-03 10:18:21 -05004879 memcg = mem_cgroup_from_css(css);
Tejun Heo2a910252015-09-11 15:00:19 -04004880 }
4881 if (!p)
4882 return 0;
4883
Tejun Heo041bdb52015-12-03 10:18:21 -05004884 /*
4885 * We are now commited to this value whatever it is. Changes in this
4886 * tunable will only affect upcoming migrations, not the current one.
4887 * So we need to save it, and keep it going.
4888 */
4889 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4890 if (!move_flags)
4891 return 0;
4892
Tejun Heo67d6b192015-09-08 15:01:10 -07004893 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004894
Tejun Heo67d6b192015-09-08 15:01:10 -07004895 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004896
Tejun Heo67d6b192015-09-08 15:01:10 -07004897 mm = get_task_mm(p);
4898 if (!mm)
4899 return 0;
4900 /* We move charges only when we move a owner of the mm */
4901 if (mm->owner == p) {
4902 VM_BUG_ON(mc.from);
4903 VM_BUG_ON(mc.to);
4904 VM_BUG_ON(mc.precharge);
4905 VM_BUG_ON(mc.moved_charge);
4906 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004907
Tejun Heo67d6b192015-09-08 15:01:10 -07004908 spin_lock(&mc.lock);
4909 mc.from = from;
4910 mc.to = memcg;
4911 mc.flags = move_flags;
4912 spin_unlock(&mc.lock);
4913 /* We set mc.moving_task later */
4914
4915 ret = mem_cgroup_precharge_mc(mm);
4916 if (ret)
4917 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004918 }
Tejun Heo67d6b192015-09-08 15:01:10 -07004919 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004920 return ret;
4921}
4922
Tejun Heo041bdb52015-12-03 10:18:21 -05004923static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004924{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004925 if (mc.to)
4926 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004927}
4928
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004929static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4930 unsigned long addr, unsigned long end,
4931 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004932{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004933 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004934 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004935 pte_t *pte;
4936 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004937 enum mc_target_type target_type;
4938 union mc_target target;
4939 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004940
Naoya Horiguchi12724852012-03-21 16:34:28 -07004941 /*
4942 * We don't take compound_lock() here but no race with splitting thp
4943 * happens because:
4944 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
4945 * under splitting, which means there's no concurrent thp split,
4946 * - if another thread runs into split_huge_page() just after we
4947 * entered this if-block, the thread must wait for page table lock
4948 * to be unlocked in __split_huge_page_splitting(), where the main
4949 * part of thp split is not executed yet.
4950 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004951 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004952 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004953 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004954 return 0;
4955 }
4956 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4957 if (target_type == MC_TARGET_PAGE) {
4958 page = target.page;
4959 if (!isolate_lru_page(page)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004960 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
Johannes Weiner1306a852014-12-10 15:44:52 -08004961 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004962 mc.precharge -= HPAGE_PMD_NR;
4963 mc.moved_charge += HPAGE_PMD_NR;
4964 }
4965 putback_lru_page(page);
4966 }
4967 put_page(page);
4968 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004969 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004970 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004971 }
4972
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004973 if (pmd_trans_unstable(pmd))
4974 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004975retry:
4976 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4977 for (; addr != end; addr += PAGE_SIZE) {
4978 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004979 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004980
4981 if (!mc.precharge)
4982 break;
4983
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004984 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004985 case MC_TARGET_PAGE:
4986 page = target.page;
4987 if (isolate_lru_page(page))
4988 goto put;
Johannes Weiner1306a852014-12-10 15:44:52 -08004989 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004990 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004991 /* we uncharge from mc.from later. */
4992 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004993 }
4994 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004995put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004996 put_page(page);
4997 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004998 case MC_TARGET_SWAP:
4999 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07005000 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005001 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005002 /* we fixup refcnts and charges later. */
5003 mc.moved_swap++;
5004 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005005 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005006 default:
5007 break;
5008 }
5009 }
5010 pte_unmap_unlock(pte - 1, ptl);
5011 cond_resched();
5012
5013 if (addr != end) {
5014 /*
5015 * We have consumed all precharges we got in can_attach().
5016 * We try charge one by one, but don't do any additional
5017 * charges to mc.to if we have failed in charge once in attach()
5018 * phase.
5019 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005020 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005021 if (!ret)
5022 goto retry;
5023 }
5024
5025 return ret;
5026}
5027
5028static void mem_cgroup_move_charge(struct mm_struct *mm)
5029{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005030 struct mm_walk mem_cgroup_move_charge_walk = {
5031 .pmd_entry = mem_cgroup_move_charge_pte_range,
5032 .mm = mm,
5033 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005034
5035 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08005036 /*
5037 * Signal mem_cgroup_begin_page_stat() to take the memcg's
5038 * move_lock while we're moving its pages to another memcg.
5039 * Then wait for already started RCU-only updates to finish.
5040 */
5041 atomic_inc(&mc.from->moving_account);
5042 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005043retry:
5044 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5045 /*
5046 * Someone who are holding the mmap_sem might be waiting in
5047 * waitq. So we cancel all extra charges, wake up all waiters,
5048 * and retry. Because we cancel precharges, we might not be able
5049 * to move enough charges, but moving charge is a best-effort
5050 * feature anyway, so it wouldn't be a big problem.
5051 */
5052 __mem_cgroup_clear_mc();
5053 cond_resched();
5054 goto retry;
5055 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005056 /*
5057 * When we have consumed all precharges and failed in doing
5058 * additional charge, the page walk just aborts.
5059 */
5060 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005061 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005062 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005063}
5064
Tejun Heo041bdb52015-12-03 10:18:21 -05005065static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005066{
Tejun Heo041bdb52015-12-03 10:18:21 -05005067 struct cgroup_subsys_state *css;
5068 struct task_struct *p = cgroup_taskset_first(tset, &css);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005069 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005070
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005071 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005072 if (mc.to)
5073 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005074 mmput(mm);
5075 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005076 if (mc.to)
5077 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005078}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005079#else /* !CONFIG_MMU */
Tejun Heo041bdb52015-12-03 10:18:21 -05005080static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005081{
5082 return 0;
5083}
Tejun Heo041bdb52015-12-03 10:18:21 -05005084static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005085{
5086}
Tejun Heo041bdb52015-12-03 10:18:21 -05005087static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005088{
5089}
5090#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005091
Tejun Heof00baae2013-04-15 13:41:15 -07005092/*
5093 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005094 * to verify whether we're attached to the default hierarchy on each mount
5095 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005096 */
Tejun Heoeb954192013-08-08 20:11:23 -04005097static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005098{
5099 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005100 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005101 * guarantees that @root doesn't have any children, so turning it
5102 * on for the root memcg is enough.
5103 */
Tejun Heoded8e212015-09-18 11:56:28 -04005104 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005105 root_mem_cgroup->use_hierarchy = true;
5106 else
5107 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005108}
5109
Johannes Weiner241994ed2015-02-11 15:26:06 -08005110static u64 memory_current_read(struct cgroup_subsys_state *css,
5111 struct cftype *cft)
5112{
Johannes Weiner277767b2015-11-05 18:50:23 -08005113 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5114
5115 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005116}
5117
5118static int memory_low_show(struct seq_file *m, void *v)
5119{
5120 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005121 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005122
5123 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005124 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005125 else
5126 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5127
5128 return 0;
5129}
5130
5131static ssize_t memory_low_write(struct kernfs_open_file *of,
5132 char *buf, size_t nbytes, loff_t off)
5133{
5134 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5135 unsigned long low;
5136 int err;
5137
5138 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005139 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005140 if (err)
5141 return err;
5142
5143 memcg->low = low;
5144
5145 return nbytes;
5146}
5147
5148static int memory_high_show(struct seq_file *m, void *v)
5149{
5150 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005151 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005152
5153 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005154 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005155 else
5156 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5157
5158 return 0;
5159}
5160
5161static ssize_t memory_high_write(struct kernfs_open_file *of,
5162 char *buf, size_t nbytes, loff_t off)
5163{
5164 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5165 unsigned long high;
5166 int err;
5167
5168 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005169 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005170 if (err)
5171 return err;
5172
5173 memcg->high = high;
5174
Tejun Heoe65e8b62015-05-22 18:23:34 -04005175 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005176 return nbytes;
5177}
5178
5179static int memory_max_show(struct seq_file *m, void *v)
5180{
5181 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005182 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005183
5184 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005185 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005186 else
5187 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5188
5189 return 0;
5190}
5191
5192static ssize_t memory_max_write(struct kernfs_open_file *of,
5193 char *buf, size_t nbytes, loff_t off)
5194{
5195 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5196 unsigned long max;
5197 int err;
5198
5199 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005200 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005201 if (err)
5202 return err;
5203
5204 err = mem_cgroup_resize_limit(memcg, max);
5205 if (err)
5206 return err;
5207
Tejun Heoe65e8b62015-05-22 18:23:34 -04005208 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005209 return nbytes;
5210}
5211
5212static int memory_events_show(struct seq_file *m, void *v)
5213{
5214 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5215
5216 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5217 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5218 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5219 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5220
5221 return 0;
5222}
5223
5224static struct cftype memory_files[] = {
5225 {
5226 .name = "current",
Johannes Weiner277767b2015-11-05 18:50:23 -08005227 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005228 .read_u64 = memory_current_read,
5229 },
5230 {
5231 .name = "low",
5232 .flags = CFTYPE_NOT_ON_ROOT,
5233 .seq_show = memory_low_show,
5234 .write = memory_low_write,
5235 },
5236 {
5237 .name = "high",
5238 .flags = CFTYPE_NOT_ON_ROOT,
5239 .seq_show = memory_high_show,
5240 .write = memory_high_write,
5241 },
5242 {
5243 .name = "max",
5244 .flags = CFTYPE_NOT_ON_ROOT,
5245 .seq_show = memory_max_show,
5246 .write = memory_max_write,
5247 },
5248 {
5249 .name = "events",
5250 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo6ae3c192015-09-18 18:01:59 -04005251 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005252 .seq_show = memory_events_show,
5253 },
5254 { } /* terminate */
5255};
5256
Tejun Heo073219e2014-02-08 10:36:58 -05005257struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005258 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005259 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005260 .css_offline = mem_cgroup_css_offline,
5261 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005262 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005263 .can_attach = mem_cgroup_can_attach,
5264 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005265 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005266 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005267 .dfl_cftypes = memory_files,
5268 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005269 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005270};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005271
Johannes Weiner241994ed2015-02-11 15:26:06 -08005272/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005273 * mem_cgroup_low - check if memory consumption is below the normal range
5274 * @root: the highest ancestor to consider
5275 * @memcg: the memory cgroup to check
5276 *
5277 * Returns %true if memory consumption of @memcg, and that of all
5278 * configurable ancestors up to @root, is below the normal range.
5279 */
5280bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5281{
5282 if (mem_cgroup_disabled())
5283 return false;
5284
5285 /*
5286 * The toplevel group doesn't have a configurable range, so
5287 * it's never low when looked at directly, and it is not
5288 * considered an ancestor when assessing the hierarchy.
5289 */
5290
5291 if (memcg == root_mem_cgroup)
5292 return false;
5293
Michal Hocko4e54ded2015-02-27 15:51:46 -08005294 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005295 return false;
5296
5297 while (memcg != root) {
5298 memcg = parent_mem_cgroup(memcg);
5299
5300 if (memcg == root_mem_cgroup)
5301 break;
5302
Michal Hocko4e54ded2015-02-27 15:51:46 -08005303 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005304 return false;
5305 }
5306 return true;
5307}
5308
Johannes Weiner00501b52014-08-08 14:19:20 -07005309/**
5310 * mem_cgroup_try_charge - try charging a page
5311 * @page: page to charge
5312 * @mm: mm context of the victim
5313 * @gfp_mask: reclaim mode
5314 * @memcgp: charged memcg return
5315 *
5316 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5317 * pages according to @gfp_mask if necessary.
5318 *
5319 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5320 * Otherwise, an error code is returned.
5321 *
5322 * After page->mapping has been set up, the caller must finalize the
5323 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5324 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5325 */
5326int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5327 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5328{
5329 struct mem_cgroup *memcg = NULL;
5330 unsigned int nr_pages = 1;
5331 int ret = 0;
5332
5333 if (mem_cgroup_disabled())
5334 goto out;
5335
5336 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005337 /*
5338 * Every swap fault against a single page tries to charge the
5339 * page, bail as early as possible. shmem_unuse() encounters
5340 * already charged pages, too. The USED bit is protected by
5341 * the page lock, which serializes swap cache removal, which
5342 * in turn serializes uncharging.
5343 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005344 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005345 goto out;
5346 }
5347
5348 if (PageTransHuge(page)) {
5349 nr_pages <<= compound_order(page);
5350 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5351 }
5352
5353 if (do_swap_account && PageSwapCache(page))
5354 memcg = try_get_mem_cgroup_from_page(page);
5355 if (!memcg)
5356 memcg = get_mem_cgroup_from_mm(mm);
5357
5358 ret = try_charge(memcg, gfp_mask, nr_pages);
5359
5360 css_put(&memcg->css);
5361
5362 if (ret == -EINTR) {
5363 memcg = root_mem_cgroup;
5364 ret = 0;
5365 }
5366out:
5367 *memcgp = memcg;
5368 return ret;
5369}
5370
5371/**
5372 * mem_cgroup_commit_charge - commit a page charge
5373 * @page: page to charge
5374 * @memcg: memcg to charge the page to
5375 * @lrucare: page might be on LRU already
5376 *
5377 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5378 * after page->mapping has been set up. This must happen atomically
5379 * as part of the page instantiation, i.e. under the page table lock
5380 * for anonymous pages, under the page lock for page and swap cache.
5381 *
5382 * In addition, the page must not be on the LRU during the commit, to
5383 * prevent racing with task migration. If it might be, use @lrucare.
5384 *
5385 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5386 */
5387void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5388 bool lrucare)
5389{
5390 unsigned int nr_pages = 1;
5391
5392 VM_BUG_ON_PAGE(!page->mapping, page);
5393 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5394
5395 if (mem_cgroup_disabled())
5396 return;
5397 /*
5398 * Swap faults will attempt to charge the same page multiple
5399 * times. But reuse_swap_page() might have removed the page
5400 * from swapcache already, so we can't check PageSwapCache().
5401 */
5402 if (!memcg)
5403 return;
5404
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005405 commit_charge(page, memcg, lrucare);
5406
Johannes Weiner00501b52014-08-08 14:19:20 -07005407 if (PageTransHuge(page)) {
5408 nr_pages <<= compound_order(page);
5409 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5410 }
5411
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005412 local_irq_disable();
5413 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5414 memcg_check_events(memcg, page);
5415 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005416
5417 if (do_swap_account && PageSwapCache(page)) {
5418 swp_entry_t entry = { .val = page_private(page) };
5419 /*
5420 * The swap entry might not get freed for a long time,
5421 * let's not wait for it. The page already received a
5422 * memory+swap charge, drop the swap entry duplicate.
5423 */
5424 mem_cgroup_uncharge_swap(entry);
5425 }
5426}
5427
5428/**
5429 * mem_cgroup_cancel_charge - cancel a page charge
5430 * @page: page to charge
5431 * @memcg: memcg to charge the page to
5432 *
5433 * Cancel a charge transaction started by mem_cgroup_try_charge().
5434 */
5435void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5436{
5437 unsigned int nr_pages = 1;
5438
5439 if (mem_cgroup_disabled())
5440 return;
5441 /*
5442 * Swap faults will attempt to charge the same page multiple
5443 * times. But reuse_swap_page() might have removed the page
5444 * from swapcache already, so we can't check PageSwapCache().
5445 */
5446 if (!memcg)
5447 return;
5448
5449 if (PageTransHuge(page)) {
5450 nr_pages <<= compound_order(page);
5451 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5452 }
5453
5454 cancel_charge(memcg, nr_pages);
5455}
5456
Johannes Weiner747db952014-08-08 14:19:24 -07005457static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005458 unsigned long nr_anon, unsigned long nr_file,
5459 unsigned long nr_huge, struct page *dummy_page)
5460{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005461 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005462 unsigned long flags;
5463
Johannes Weinerce00a962014-09-05 08:43:57 -04005464 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005465 page_counter_uncharge(&memcg->memory, nr_pages);
5466 if (do_swap_account)
5467 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005468 memcg_oom_recover(memcg);
5469 }
Johannes Weiner747db952014-08-08 14:19:24 -07005470
5471 local_irq_save(flags);
5472 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5473 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5474 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5475 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005476 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005477 memcg_check_events(memcg, dummy_page);
5478 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005479
5480 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005481 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005482}
5483
5484static void uncharge_list(struct list_head *page_list)
5485{
5486 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005487 unsigned long nr_anon = 0;
5488 unsigned long nr_file = 0;
5489 unsigned long nr_huge = 0;
5490 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005491 struct list_head *next;
5492 struct page *page;
5493
5494 next = page_list->next;
5495 do {
5496 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005497
5498 page = list_entry(next, struct page, lru);
5499 next = page->lru.next;
5500
5501 VM_BUG_ON_PAGE(PageLRU(page), page);
5502 VM_BUG_ON_PAGE(page_count(page), page);
5503
Johannes Weiner1306a852014-12-10 15:44:52 -08005504 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005505 continue;
5506
5507 /*
5508 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005509 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005510 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005511 */
5512
Johannes Weiner1306a852014-12-10 15:44:52 -08005513 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005514 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005515 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5516 nr_huge, page);
5517 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005518 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005519 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005520 }
5521
5522 if (PageTransHuge(page)) {
5523 nr_pages <<= compound_order(page);
5524 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5525 nr_huge += nr_pages;
5526 }
5527
5528 if (PageAnon(page))
5529 nr_anon += nr_pages;
5530 else
5531 nr_file += nr_pages;
5532
Johannes Weiner1306a852014-12-10 15:44:52 -08005533 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005534
5535 pgpgout++;
5536 } while (next != page_list);
5537
5538 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005539 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5540 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005541}
5542
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005543/**
5544 * mem_cgroup_uncharge - uncharge a page
5545 * @page: page to uncharge
5546 *
5547 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5548 * mem_cgroup_commit_charge().
5549 */
5550void mem_cgroup_uncharge(struct page *page)
5551{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005552 if (mem_cgroup_disabled())
5553 return;
5554
Johannes Weiner747db952014-08-08 14:19:24 -07005555 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005556 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005557 return;
5558
Johannes Weiner747db952014-08-08 14:19:24 -07005559 INIT_LIST_HEAD(&page->lru);
5560 uncharge_list(&page->lru);
5561}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005562
Johannes Weiner747db952014-08-08 14:19:24 -07005563/**
5564 * mem_cgroup_uncharge_list - uncharge a list of page
5565 * @page_list: list of pages to uncharge
5566 *
5567 * Uncharge a list of pages previously charged with
5568 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5569 */
5570void mem_cgroup_uncharge_list(struct list_head *page_list)
5571{
5572 if (mem_cgroup_disabled())
5573 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005574
Johannes Weiner747db952014-08-08 14:19:24 -07005575 if (!list_empty(page_list))
5576 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005577}
5578
5579/**
5580 * mem_cgroup_migrate - migrate a charge to another page
5581 * @oldpage: currently charged page
5582 * @newpage: page to transfer the charge to
Michal Hockof5e03a42015-02-05 12:25:14 -08005583 * @lrucare: either or both pages might be on the LRU already
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005584 *
5585 * Migrate the charge from @oldpage to @newpage.
5586 *
5587 * Both pages must be locked, @newpage->mapping must be set up.
5588 */
5589void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5590 bool lrucare)
5591{
Johannes Weiner29833312014-12-10 15:44:02 -08005592 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005593 int isolated;
5594
5595 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5596 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5597 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5598 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5599 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005600 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5601 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005602
5603 if (mem_cgroup_disabled())
5604 return;
5605
5606 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005607 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005608 return;
5609
Johannes Weiner7d5e3242014-12-10 15:43:46 -08005610 /*
5611 * Swapcache readahead pages can get migrated before being
5612 * charged, and migration from compaction can happen to an
5613 * uncharged page when the PFN walker finds a page that
5614 * reclaim just put back on the LRU but has not released yet.
5615 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005616 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005617 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005618 return;
5619
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005620 if (lrucare)
5621 lock_page_lru(oldpage, &isolated);
5622
Johannes Weiner1306a852014-12-10 15:44:52 -08005623 oldpage->mem_cgroup = NULL;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005624
5625 if (lrucare)
5626 unlock_page_lru(oldpage, isolated);
5627
Johannes Weiner29833312014-12-10 15:44:02 -08005628 commit_charge(newpage, memcg, lrucare);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005629}
5630
Michal Hocko2d110852013-02-22 16:34:43 -08005631/*
Michal Hocko10813122013-02-22 16:35:41 -08005632 * subsys_initcall() for memory controller.
5633 *
5634 * Some parts like hotcpu_notifier() have to be initialized from this context
5635 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5636 * everything that doesn't depend on a specific mem_cgroup structure should
5637 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005638 */
5639static int __init mem_cgroup_init(void)
5640{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005641 int cpu, node;
5642
Michal Hocko2d110852013-02-22 16:34:43 -08005643 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005644
5645 for_each_possible_cpu(cpu)
5646 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5647 drain_local_stock);
5648
5649 for_each_node(node) {
5650 struct mem_cgroup_tree_per_node *rtpn;
5651 int zone;
5652
5653 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5654 node_online(node) ? node : NUMA_NO_NODE);
5655
5656 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5657 struct mem_cgroup_tree_per_zone *rtpz;
5658
5659 rtpz = &rtpn->rb_tree_per_zone[zone];
5660 rtpz->rb_root = RB_ROOT;
5661 spin_lock_init(&rtpz->lock);
5662 }
5663 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5664 }
5665
Michal Hocko2d110852013-02-22 16:34:43 -08005666 return 0;
5667}
5668subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005669
5670#ifdef CONFIG_MEMCG_SWAP
5671/**
5672 * mem_cgroup_swapout - transfer a memsw charge to swap
5673 * @page: page whose memsw charge to transfer
5674 * @entry: swap entry to move the charge to
5675 *
5676 * Transfer the memsw charge of @page to @entry.
5677 */
5678void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5679{
5680 struct mem_cgroup *memcg;
5681 unsigned short oldid;
5682
5683 VM_BUG_ON_PAGE(PageLRU(page), page);
5684 VM_BUG_ON_PAGE(page_count(page), page);
5685
5686 if (!do_swap_account)
5687 return;
5688
5689 memcg = page->mem_cgroup;
5690
5691 /* Readahead page, never charged */
5692 if (!memcg)
5693 return;
5694
5695 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5696 VM_BUG_ON_PAGE(oldid, page);
5697 mem_cgroup_swap_statistics(memcg, true);
5698
5699 page->mem_cgroup = NULL;
5700
5701 if (!mem_cgroup_is_root(memcg))
5702 page_counter_uncharge(&memcg->memory, 1);
5703
Johannes Weinerf3717632015-06-10 11:14:54 -07005704 /* Caller disabled preemption with mapping->tree_lock */
Johannes Weiner21afa382015-02-11 15:26:36 -08005705 mem_cgroup_charge_statistics(memcg, page, -1);
5706 memcg_check_events(memcg, page);
5707}
5708
5709/**
5710 * mem_cgroup_uncharge_swap - uncharge a swap entry
5711 * @entry: swap entry to uncharge
5712 *
5713 * Drop the memsw charge associated with @entry.
5714 */
5715void mem_cgroup_uncharge_swap(swp_entry_t entry)
5716{
5717 struct mem_cgroup *memcg;
5718 unsigned short id;
5719
5720 if (!do_swap_account)
5721 return;
5722
5723 id = swap_cgroup_record(entry, 0);
5724 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005725 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005726 if (memcg) {
5727 if (!mem_cgroup_is_root(memcg))
5728 page_counter_uncharge(&memcg->memsw, 1);
5729 mem_cgroup_swap_statistics(memcg, false);
5730 css_put(&memcg->css);
5731 }
5732 rcu_read_unlock();
5733}
5734
5735/* for remember boot option*/
5736#ifdef CONFIG_MEMCG_SWAP_ENABLED
5737static int really_do_swap_account __initdata = 1;
5738#else
5739static int really_do_swap_account __initdata;
5740#endif
5741
5742static int __init enable_swap_account(char *s)
5743{
5744 if (!strcmp(s, "1"))
5745 really_do_swap_account = 1;
5746 else if (!strcmp(s, "0"))
5747 really_do_swap_account = 0;
5748 return 1;
5749}
5750__setup("swapaccount=", enable_swap_account);
5751
5752static struct cftype memsw_cgroup_files[] = {
5753 {
5754 .name = "memsw.usage_in_bytes",
5755 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5756 .read_u64 = mem_cgroup_read_u64,
5757 },
5758 {
5759 .name = "memsw.max_usage_in_bytes",
5760 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5761 .write = mem_cgroup_reset,
5762 .read_u64 = mem_cgroup_read_u64,
5763 },
5764 {
5765 .name = "memsw.limit_in_bytes",
5766 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5767 .write = mem_cgroup_write,
5768 .read_u64 = mem_cgroup_read_u64,
5769 },
5770 {
5771 .name = "memsw.failcnt",
5772 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5773 .write = mem_cgroup_reset,
5774 .read_u64 = mem_cgroup_read_u64,
5775 },
5776 { }, /* terminate */
5777};
5778
5779static int __init mem_cgroup_swap_init(void)
5780{
5781 if (!mem_cgroup_disabled() && really_do_swap_account) {
5782 do_swap_account = 1;
5783 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5784 memsw_cgroup_files));
5785 }
5786 return 0;
5787}
5788subsys_initcall(mem_cgroup_swap_init);
5789
5790#endif /* CONFIG_MEMCG_SWAP */