blob: 9acfb165eb52eeff2271888ec61fe093c9015641 [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 Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000069#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080070#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080071
Balbir Singh8697d332008-02-07 00:13:59 -080072#include <asm/uaccess.h>
73
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070074#include <trace/events/vmscan.h>
75
Tejun Heo073219e2014-02-08 10:36:58 -050076struct cgroup_subsys memory_cgrp_subsys __read_mostly;
77EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080078
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 Heo56161632015-05-22 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 Thelenc4843a72015-05-22 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 Hockoe752eb62015-09-08 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 Heoad7fa852015-05-27 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 Heo9e10a132015-09-18 11:56:28 -0400438 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400439 memcg = root_mem_cgroup;
440
441 rcu_read_unlock();
442 return &memcg->css;
443}
444
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700445/**
446 * page_cgroup_ino - return inode number of the memcg a page is charged to
447 * @page: the page
448 *
449 * Look up the closest online ancestor of the memory cgroup @page is charged to
450 * and return its inode number or 0 if @page is not charged to any cgroup. It
451 * is safe to call this function without holding a reference to @page.
452 *
453 * Note, this function is inherently racy, because there is nothing to prevent
454 * the cgroup inode from getting torn down and potentially reallocated a moment
455 * after page_cgroup_ino() returns, so it only should be used by callers that
456 * do not care (such as procfs interfaces).
457 */
458ino_t page_cgroup_ino(struct page *page)
459{
460 struct mem_cgroup *memcg;
461 unsigned long ino = 0;
462
463 rcu_read_lock();
464 memcg = READ_ONCE(page->mem_cgroup);
465 while (memcg && !(memcg->css.flags & CSS_ONLINE))
466 memcg = parent_mem_cgroup(memcg);
467 if (memcg)
468 ino = cgroup_ino(memcg->css.cgroup);
469 rcu_read_unlock();
470 return ino;
471}
472
Balbir Singhf64c3f52009-09-23 15:56:37 -0700473static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700474mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700475{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700476 int nid = page_to_nid(page);
477 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700478
Jianyu Zhane2318752014-06-06 14:38:20 -0700479 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700480}
481
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700482static struct mem_cgroup_tree_per_zone *
483soft_limit_tree_node_zone(int nid, int zid)
484{
485 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
486}
487
488static struct mem_cgroup_tree_per_zone *
489soft_limit_tree_from_page(struct page *page)
490{
491 int nid = page_to_nid(page);
492 int zid = page_zonenum(page);
493
494 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
495}
496
Johannes Weinercf2c8122014-06-06 14:38:21 -0700497static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
498 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800499 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700500{
501 struct rb_node **p = &mctz->rb_root.rb_node;
502 struct rb_node *parent = NULL;
503 struct mem_cgroup_per_zone *mz_node;
504
505 if (mz->on_tree)
506 return;
507
508 mz->usage_in_excess = new_usage_in_excess;
509 if (!mz->usage_in_excess)
510 return;
511 while (*p) {
512 parent = *p;
513 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
514 tree_node);
515 if (mz->usage_in_excess < mz_node->usage_in_excess)
516 p = &(*p)->rb_left;
517 /*
518 * We can't avoid mem cgroups that are over their soft
519 * limit by the same amount
520 */
521 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
522 p = &(*p)->rb_right;
523 }
524 rb_link_node(&mz->tree_node, parent, p);
525 rb_insert_color(&mz->tree_node, &mctz->rb_root);
526 mz->on_tree = true;
527}
528
Johannes Weinercf2c8122014-06-06 14:38:21 -0700529static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
530 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700531{
532 if (!mz->on_tree)
533 return;
534 rb_erase(&mz->tree_node, &mctz->rb_root);
535 mz->on_tree = false;
536}
537
Johannes Weinercf2c8122014-06-06 14:38:21 -0700538static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
539 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700540{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700541 unsigned long flags;
542
543 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700544 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700545 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700546}
547
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800548static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
549{
550 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700551 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800552 unsigned long excess = 0;
553
554 if (nr_pages > soft_limit)
555 excess = nr_pages - soft_limit;
556
557 return excess;
558}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700559
560static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
561{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800562 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700563 struct mem_cgroup_per_zone *mz;
564 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700565
Jianyu Zhane2318752014-06-06 14:38:20 -0700566 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700567 /*
568 * Necessary to update all ancestors when hierarchy is used.
569 * because their event counter is not touched.
570 */
571 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700572 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800573 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700574 /*
575 * We have to update the tree if mz is on RB-tree or
576 * mem is over its softlimit.
577 */
578 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700579 unsigned long flags;
580
581 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700582 /* if on-tree, remove it */
583 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700584 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700585 /*
586 * Insert again. mz->usage_in_excess will be updated.
587 * If excess is 0, no tree ops.
588 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700589 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700590 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700591 }
592 }
593}
594
595static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
596{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700597 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700598 struct mem_cgroup_per_zone *mz;
599 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700600
Jianyu Zhane2318752014-06-06 14:38:20 -0700601 for_each_node(nid) {
602 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
603 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
604 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700605 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700606 }
607 }
608}
609
610static struct mem_cgroup_per_zone *
611__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
612{
613 struct rb_node *rightmost = NULL;
614 struct mem_cgroup_per_zone *mz;
615
616retry:
617 mz = NULL;
618 rightmost = rb_last(&mctz->rb_root);
619 if (!rightmost)
620 goto done; /* Nothing to reclaim from */
621
622 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
623 /*
624 * Remove the node now but someone else can add it back,
625 * we will to add it back at the end of reclaim to its correct
626 * position in the tree.
627 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700628 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800629 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400630 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700631 goto retry;
632done:
633 return mz;
634}
635
636static struct mem_cgroup_per_zone *
637mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
638{
639 struct mem_cgroup_per_zone *mz;
640
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700641 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700642 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700643 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700644 return mz;
645}
646
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700647/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700648 * Return page count for single (non recursive) @memcg.
649 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700650 * Implementation Note: reading percpu statistics for memcg.
651 *
652 * Both of vmstat[] and percpu_counter has threshold and do periodic
653 * synchronization to implement "quick" read. There are trade-off between
654 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700655 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700656 *
657 * But this _read() function is used for user interface now. The user accounts
658 * memory usage by memory cgroup and he _always_ requires exact value because
659 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
660 * have to visit all online cpus and make sum. So, for now, unnecessary
661 * synchronization is not implemented. (just implemented for cpu hotplug)
662 *
663 * If there are kernel internal actions which can make use of some not-exact
664 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700665 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700666 * implemented.
667 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700668static unsigned long
669mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800670{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700671 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800672 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800673
Greg Thelen484ebb32015-10-01 15:37:05 -0700674 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400675 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700676 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700677 /*
678 * Summing races with updates, so val may be negative. Avoid exposing
679 * transient negative values.
680 */
681 if (val < 0)
682 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800683 return val;
684}
685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700687 enum mem_cgroup_events_index idx)
688{
689 unsigned long val = 0;
690 int cpu;
691
Tejun Heo733a5722015-05-22 18:23:18 -0400692 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700693 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700694 return val;
695}
696
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700697static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700698 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700699 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800700{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700701 /*
702 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
703 * counted as CACHE even if it's on ANON LRU.
704 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700705 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700706 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700707 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800708 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700709 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700710 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700711
David Rientjesb070e652013-05-07 16:18:09 -0700712 if (PageTransHuge(page))
713 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
714 nr_pages);
715
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800716 /* pagein of a big page is an event. So, ignore page size */
717 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700718 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800719 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700720 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800721 nr_pages = -nr_pages; /* for event */
722 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800723
Johannes Weiner13114712012-05-29 15:07:07 -0700724 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800725}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800726
Jianyu Zhane2318752014-06-06 14:38:20 -0700727static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
728 int nid,
729 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700730{
Jianyu Zhane2318752014-06-06 14:38:20 -0700731 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700732 int zid;
733
Jianyu Zhane2318752014-06-06 14:38:20 -0700734 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700735
Jianyu Zhane2318752014-06-06 14:38:20 -0700736 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
737 struct mem_cgroup_per_zone *mz;
738 enum lru_list lru;
739
740 for_each_lru(lru) {
741 if (!(BIT(lru) & lru_mask))
742 continue;
743 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
744 nr += mz->lru_size[lru];
745 }
746 }
747 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700748}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700749
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700750static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700751 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800752{
Jianyu Zhane2318752014-06-06 14:38:20 -0700753 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700754 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800755
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800756 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700757 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
758 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800759}
760
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800761static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
762 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800763{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700764 unsigned long val, next;
765
Johannes Weiner13114712012-05-29 15:07:07 -0700766 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700767 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700768 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800769 if ((long)next - (long)val < 0) {
770 switch (target) {
771 case MEM_CGROUP_TARGET_THRESH:
772 next = val + THRESHOLDS_EVENTS_TARGET;
773 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700774 case MEM_CGROUP_TARGET_SOFTLIMIT:
775 next = val + SOFTLIMIT_EVENTS_TARGET;
776 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800777 case MEM_CGROUP_TARGET_NUMAINFO:
778 next = val + NUMAINFO_EVENTS_TARGET;
779 break;
780 default:
781 break;
782 }
783 __this_cpu_write(memcg->stat->targets[target], next);
784 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700785 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800786 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800787}
788
789/*
790 * Check events in order.
791 *
792 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700793static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800794{
795 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800796 if (unlikely(mem_cgroup_event_ratelimit(memcg,
797 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700798 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800799 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800800
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700801 do_softlimit = mem_cgroup_event_ratelimit(memcg,
802 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700803#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800804 do_numainfo = mem_cgroup_event_ratelimit(memcg,
805 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700806#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800807 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700808 if (unlikely(do_softlimit))
809 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800810#if MAX_NUMNODES > 1
811 if (unlikely(do_numainfo))
812 atomic_inc(&memcg->numainfo_events);
813#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700814 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800815}
816
Balbir Singhcf475ad2008-04-29 01:00:16 -0700817struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800818{
Balbir Singh31a78f22008-09-28 23:09:31 +0100819 /*
820 * mm_update_next_owner() may clear mm->owner to NULL
821 * if it races with swapoff, page migration, etc.
822 * So this can be called with p == NULL.
823 */
824 if (unlikely(!p))
825 return NULL;
826
Tejun Heo073219e2014-02-08 10:36:58 -0500827 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800828}
Michal Hocko33398cf2015-09-08 15:01:02 -0700829EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800830
Johannes Weinerdf381972014-04-07 15:37:43 -0700831static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800832{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700833 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700834
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800835 rcu_read_lock();
836 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700837 /*
838 * Page cache insertions can happen withou an
839 * actual mm context, e.g. during disk probing
840 * on boot, loopback IO, acct() writes etc.
841 */
842 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700843 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700844 else {
845 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
846 if (unlikely(!memcg))
847 memcg = root_mem_cgroup;
848 }
Tejun Heoec903c02014-05-13 12:11:01 -0400849 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800850 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700851 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800852}
853
Johannes Weiner56600482012-01-12 17:17:59 -0800854/**
855 * mem_cgroup_iter - iterate over memory cgroup hierarchy
856 * @root: hierarchy root
857 * @prev: previously returned memcg, NULL on first invocation
858 * @reclaim: cookie for shared reclaim walks, NULL for full walks
859 *
860 * Returns references to children of the hierarchy below @root, or
861 * @root itself, or %NULL after a full round-trip.
862 *
863 * Caller must pass the return value in @prev on subsequent
864 * invocations for reference counting, or use mem_cgroup_iter_break()
865 * to cancel a hierarchy walk before the round-trip is complete.
866 *
867 * Reclaimers can specify a zone and a priority level in @reclaim to
868 * divide up the memcgs in the hierarchy among all concurrent
869 * reclaimers operating on the same zone and priority.
870 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700871struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800872 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700873 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700874{
Michal Hocko33398cf2015-09-08 15:01:02 -0700875 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800876 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800877 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800878 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700879
Andrew Morton694fbc02013-09-24 15:27:37 -0700880 if (mem_cgroup_disabled())
881 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800882
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700883 if (!root)
884 root = root_mem_cgroup;
885
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800886 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800887 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800888
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800889 if (!root->use_hierarchy && root != root_mem_cgroup) {
890 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800891 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700892 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800893 }
894
Michal Hocko542f85f2013-04-29 15:07:15 -0700895 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800896
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800897 if (reclaim) {
898 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800899
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800900 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
901 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700902
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800903 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700904 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800905
906 do {
Jason Low4db0c3c2015-04-15 16:14:08 -0700907 pos = READ_ONCE(iter->position);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800908 /*
909 * A racing update may change the position and
910 * put the last reference, hence css_tryget(),
911 * or retry to see the updated position.
912 */
913 } while (pos && !css_tryget(&pos->css));
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800914 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800915
916 if (pos)
917 css = &pos->css;
918
919 for (;;) {
920 css = css_next_descendant_pre(css, &root->css);
921 if (!css) {
922 /*
923 * Reclaimers share the hierarchy walk, and a
924 * new one might jump in right at the end of
925 * the hierarchy - make sure they see at least
926 * one group and restart from the beginning.
927 */
928 if (!prev)
929 continue;
930 break;
931 }
932
933 /*
934 * Verify the css and acquire a reference. The root
935 * is provided by the caller, so we know it's alive
936 * and kicking, and don't take an extra reference.
937 */
938 memcg = mem_cgroup_from_css(css);
939
940 if (css == &root->css)
941 break;
942
Johannes Weinerb2052562014-12-10 15:42:48 -0800943 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800944 /*
945 * Make sure the memcg is initialized:
946 * mem_cgroup_css_online() orders the the
947 * initialization against setting the flag.
948 */
949 if (smp_load_acquire(&memcg->initialized))
950 break;
951
952 css_put(css);
953 }
954
955 memcg = NULL;
956 }
957
958 if (reclaim) {
959 if (cmpxchg(&iter->position, pos, memcg) == pos) {
960 if (memcg)
961 css_get(&memcg->css);
962 if (pos)
963 css_put(&pos->css);
964 }
965
966 /*
967 * pairs with css_tryget when dereferencing iter->position
968 * above.
969 */
970 if (pos)
971 css_put(&pos->css);
972
973 if (!memcg)
974 iter->generation++;
975 else if (!prev)
976 reclaim->generation = iter->generation;
977 }
978
Michal Hocko542f85f2013-04-29 15:07:15 -0700979out_unlock:
980 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800981out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700982 if (prev && prev != root)
983 css_put(&prev->css);
984
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800985 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700986}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800987
Johannes Weiner56600482012-01-12 17:17:59 -0800988/**
989 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
990 * @root: hierarchy root
991 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
992 */
993void mem_cgroup_iter_break(struct mem_cgroup *root,
994 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800995{
996 if (!root)
997 root = root_mem_cgroup;
998 if (prev && prev != root)
999 css_put(&prev->css);
1000}
1001
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001002/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001003 * Iteration constructs for visiting all cgroups (under a tree). If
1004 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1005 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001006 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001007#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001008 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001009 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001010 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001011
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001012#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001013 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001014 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001015 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001016
Johannes Weiner925b7672012-01-12 17:18:15 -08001017/**
1018 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1019 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001020 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001021 *
1022 * Returns the lru list vector holding pages for the given @zone and
1023 * @mem. This can be the global zone lruvec, if the memory controller
1024 * is disabled.
1025 */
1026struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1027 struct mem_cgroup *memcg)
1028{
1029 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001030 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001031
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001032 if (mem_cgroup_disabled()) {
1033 lruvec = &zone->lruvec;
1034 goto out;
1035 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001036
Jianyu Zhane2318752014-06-06 14:38:20 -07001037 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001038 lruvec = &mz->lruvec;
1039out:
1040 /*
1041 * Since a node can be onlined after the mem_cgroup was created,
1042 * we have to be prepared to initialize lruvec->zone here;
1043 * and if offlined then reonlined, we need to reinitialize it.
1044 */
1045 if (unlikely(lruvec->zone != zone))
1046 lruvec->zone = zone;
1047 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001048}
1049
Johannes Weiner925b7672012-01-12 17:18:15 -08001050/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001051 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001052 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001053 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001054 *
1055 * This function is only safe when following the LRU page isolation
1056 * and putback protocol: the LRU lock must be held, and the page must
1057 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001058 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001059struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001060{
1061 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001062 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001063 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001064
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001065 if (mem_cgroup_disabled()) {
1066 lruvec = &zone->lruvec;
1067 goto out;
1068 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001069
Johannes Weiner1306a852014-12-10 15:44:52 -08001070 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001071 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001072 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001073 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001074 */
Johannes Weiner29833312014-12-10 15:44:02 -08001075 if (!memcg)
1076 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001077
Jianyu Zhane2318752014-06-06 14:38:20 -07001078 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001079 lruvec = &mz->lruvec;
1080out:
1081 /*
1082 * Since a node can be onlined after the mem_cgroup was created,
1083 * we have to be prepared to initialize lruvec->zone here;
1084 * and if offlined then reonlined, we need to reinitialize it.
1085 */
1086 if (unlikely(lruvec->zone != zone))
1087 lruvec->zone = zone;
1088 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001089}
1090
1091/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001092 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1093 * @lruvec: mem_cgroup per zone lru vector
1094 * @lru: index of lru list the page is sitting on
1095 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001096 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001097 * This function must be called when a page is added to or removed from an
1098 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001099 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001100void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1101 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001102{
1103 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001104 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001105
1106 if (mem_cgroup_disabled())
1107 return;
1108
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001109 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1110 lru_size = mz->lru_size + lru;
1111 *lru_size += nr_pages;
1112 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001113}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001114
Johannes Weiner2314b422014-12-10 15:44:33 -08001115bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001116{
Johannes Weiner2314b422014-12-10 15:44:33 -08001117 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001118 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001119 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001120
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001121 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001122 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001123 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001124 task_unlock(p);
1125 } else {
1126 /*
1127 * All threads may have already detached their mm's, but the oom
1128 * killer still needs to detect if they have already been oom
1129 * killed to prevent needlessly killing additional tasks.
1130 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001131 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001132 task_memcg = mem_cgroup_from_task(task);
1133 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001134 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001135 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001136 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1137 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001138 return ret;
1139}
1140
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001141#define mem_cgroup_from_counter(counter, member) \
Balbir Singh6d61ef42009-01-07 18:08:06 -08001142 container_of(counter, struct mem_cgroup, member)
1143
Johannes Weiner19942822011-02-01 15:52:43 -08001144/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001145 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001146 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001147 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001148 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001149 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001150 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001151static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001152{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001153 unsigned long margin = 0;
1154 unsigned long count;
1155 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001156
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001157 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001158 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001159 if (count < limit)
1160 margin = limit - count;
1161
1162 if (do_swap_account) {
1163 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001164 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001165 if (count <= limit)
1166 margin = min(margin, limit - count);
1167 }
1168
1169 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001170}
1171
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001172/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001173 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001174 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001175 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1176 * moving cgroups. This is for waiting at high-memory pressure
1177 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001178 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001179static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001180{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001181 struct mem_cgroup *from;
1182 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001183 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001184 /*
1185 * Unlike task_move routines, we access mc.to, mc.from not under
1186 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1187 */
1188 spin_lock(&mc.lock);
1189 from = mc.from;
1190 to = mc.to;
1191 if (!from)
1192 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001193
Johannes Weiner2314b422014-12-10 15:44:33 -08001194 ret = mem_cgroup_is_descendant(from, memcg) ||
1195 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001196unlock:
1197 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001198 return ret;
1199}
1200
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001201static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001202{
1203 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001204 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001205 DEFINE_WAIT(wait);
1206 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1207 /* moving charge context might have finished. */
1208 if (mc.moving_task)
1209 schedule();
1210 finish_wait(&mc.waitq, &wait);
1211 return true;
1212 }
1213 }
1214 return false;
1215}
1216
Sha Zhengju58cf1882013-02-22 16:32:05 -08001217#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001218/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001219 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001220 * @memcg: The memory cgroup that went over limit
1221 * @p: Task that is going to be killed
1222 *
1223 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1224 * enabled
1225 */
1226void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1227{
Tejun Heoe61734c2014-02-12 09:29:50 -05001228 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001229 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001230 struct mem_cgroup *iter;
1231 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001232
Michal Hocko08088cb2014-02-25 15:01:44 -08001233 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001234 rcu_read_lock();
1235
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001236 if (p) {
1237 pr_info("Task in ");
1238 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1239 pr_cont(" killed as a result of limit of ");
1240 } else {
1241 pr_info("Memory limit reached of cgroup ");
1242 }
1243
Tejun Heoe61734c2014-02-12 09:29:50 -05001244 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001245 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001246
Balbir Singhe2224322009-04-02 16:57:39 -07001247 rcu_read_unlock();
1248
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001249 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1250 K((u64)page_counter_read(&memcg->memory)),
1251 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1252 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1253 K((u64)page_counter_read(&memcg->memsw)),
1254 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1255 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1256 K((u64)page_counter_read(&memcg->kmem)),
1257 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001258
1259 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001260 pr_info("Memory cgroup stats for ");
1261 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001262 pr_cont(":");
1263
1264 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1265 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1266 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001267 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001268 K(mem_cgroup_read_stat(iter, i)));
1269 }
1270
1271 for (i = 0; i < NR_LRU_LISTS; i++)
1272 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1273 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1274
1275 pr_cont("\n");
1276 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001277 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001278}
1279
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001280/*
1281 * This function returns the number of memcg under hierarchy tree. Returns
1282 * 1(self count) if no children.
1283 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001284static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001285{
1286 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001287 struct mem_cgroup *iter;
1288
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001289 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001290 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001291 return num;
1292}
1293
Balbir Singh6d61ef42009-01-07 18:08:06 -08001294/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001295 * Return the memory (and swap, if configured) limit for a memcg.
1296 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001297static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001298{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001299 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001300
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001301 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001302 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001303 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001304
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001305 memsw_limit = memcg->memsw.limit;
1306 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001307 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001308 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001309}
1310
David Rientjes19965462012-12-11 16:00:26 -08001311static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1312 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001313{
David Rientjes6e0fc462015-09-08 15:00:36 -07001314 struct oom_control oc = {
1315 .zonelist = NULL,
1316 .nodemask = NULL,
1317 .gfp_mask = gfp_mask,
1318 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001319 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001320 struct mem_cgroup *iter;
1321 unsigned long chosen_points = 0;
1322 unsigned long totalpages;
1323 unsigned int points = 0;
1324 struct task_struct *chosen = NULL;
1325
Johannes Weinerdc564012015-06-24 16:57:19 -07001326 mutex_lock(&oom_lock);
1327
David Rientjes876aafb2012-07-31 16:43:48 -07001328 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001329 * If current has a pending SIGKILL or is exiting, then automatically
1330 * select it. The goal is to allow it to allocate so that it may
1331 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001332 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001333 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001334 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001335 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001336 }
1337
David Rientjes6e0fc462015-09-08 15:00:36 -07001338 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001339 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001340 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001341 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001342 struct task_struct *task;
1343
Tejun Heo72ec7022013-08-08 20:11:26 -04001344 css_task_iter_start(&iter->css, &it);
1345 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001346 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001347 case OOM_SCAN_SELECT:
1348 if (chosen)
1349 put_task_struct(chosen);
1350 chosen = task;
1351 chosen_points = ULONG_MAX;
1352 get_task_struct(chosen);
1353 /* fall through */
1354 case OOM_SCAN_CONTINUE:
1355 continue;
1356 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001357 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001358 mem_cgroup_iter_break(memcg, iter);
1359 if (chosen)
1360 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001361 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001362 case OOM_SCAN_OK:
1363 break;
1364 };
1365 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001366 if (!points || points < chosen_points)
1367 continue;
1368 /* Prefer thread group leaders for display purposes */
1369 if (points == chosen_points &&
1370 thread_group_leader(chosen))
1371 continue;
1372
1373 if (chosen)
1374 put_task_struct(chosen);
1375 chosen = task;
1376 chosen_points = points;
1377 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001378 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001379 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001380 }
1381
Johannes Weinerdc564012015-06-24 16:57:19 -07001382 if (chosen) {
1383 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001384 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1385 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001386 }
1387unlock:
1388 mutex_unlock(&oom_lock);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001389}
1390
Michele Curtiae6e71d2014-12-12 16:56:35 -08001391#if MAX_NUMNODES > 1
1392
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001393/**
1394 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001395 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001396 * @nid: the node ID to be checked.
1397 * @noswap : specify true here if the user wants flle only information.
1398 *
1399 * This function returns whether the specified memcg contains any
1400 * reclaimable pages on a node. Returns true if there are any reclaimable
1401 * pages in the node.
1402 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001403static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001404 int nid, bool noswap)
1405{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001406 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001407 return true;
1408 if (noswap || !total_swap_pages)
1409 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001410 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001411 return true;
1412 return false;
1413
1414}
Ying Han889976d2011-05-26 16:25:33 -07001415
1416/*
1417 * Always updating the nodemask is not very good - even if we have an empty
1418 * list or the wrong list here, we can start from some node and traverse all
1419 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1420 *
1421 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001422static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001423{
1424 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001425 /*
1426 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1427 * pagein/pageout changes since the last update.
1428 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001429 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001430 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001431 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001432 return;
1433
Ying Han889976d2011-05-26 16:25:33 -07001434 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001435 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001436
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001437 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001438
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001439 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1440 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001441 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001442
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001443 atomic_set(&memcg->numainfo_events, 0);
1444 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001445}
1446
1447/*
1448 * Selecting a node where we start reclaim from. Because what we need is just
1449 * reducing usage counter, start from anywhere is O,K. Considering
1450 * memory reclaim from current node, there are pros. and cons.
1451 *
1452 * Freeing memory from current node means freeing memory from a node which
1453 * we'll use or we've used. So, it may make LRU bad. And if several threads
1454 * hit limits, it will see a contention on a node. But freeing from remote
1455 * node means more costs for memory reclaim because of memory latency.
1456 *
1457 * Now, we use round-robin. Better algorithm is welcomed.
1458 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001459int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001460{
1461 int node;
1462
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001463 mem_cgroup_may_update_nodemask(memcg);
1464 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001465
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001466 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001467 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001468 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001469 /*
1470 * We call this when we hit limit, not when pages are added to LRU.
1471 * No LRU may hold pages because all pages are UNEVICTABLE or
1472 * memcg is too small and all pages are not on LRU. In that case,
1473 * we use curret node.
1474 */
1475 if (unlikely(node == MAX_NUMNODES))
1476 node = numa_node_id();
1477
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001478 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001479 return node;
1480}
Ying Han889976d2011-05-26 16:25:33 -07001481#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001482int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001483{
1484 return 0;
1485}
1486#endif
1487
Andrew Morton0608f432013-09-24 15:27:41 -07001488static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1489 struct zone *zone,
1490 gfp_t gfp_mask,
1491 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001492{
Andrew Morton0608f432013-09-24 15:27:41 -07001493 struct mem_cgroup *victim = NULL;
1494 int total = 0;
1495 int loop = 0;
1496 unsigned long excess;
1497 unsigned long nr_scanned;
1498 struct mem_cgroup_reclaim_cookie reclaim = {
1499 .zone = zone,
1500 .priority = 0,
1501 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001502
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001503 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001504
Andrew Morton0608f432013-09-24 15:27:41 -07001505 while (1) {
1506 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1507 if (!victim) {
1508 loop++;
1509 if (loop >= 2) {
1510 /*
1511 * If we have not been able to reclaim
1512 * anything, it might because there are
1513 * no reclaimable pages under this hierarchy
1514 */
1515 if (!total)
1516 break;
1517 /*
1518 * We want to do more targeted reclaim.
1519 * excess >> 2 is not to excessive so as to
1520 * reclaim too much, nor too less that we keep
1521 * coming back to reclaim from this cgroup
1522 */
1523 if (total >= (excess >> 2) ||
1524 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1525 break;
1526 }
1527 continue;
1528 }
Andrew Morton0608f432013-09-24 15:27:41 -07001529 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1530 zone, &nr_scanned);
1531 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001532 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001533 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001534 }
Andrew Morton0608f432013-09-24 15:27:41 -07001535 mem_cgroup_iter_break(root_memcg, victim);
1536 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001537}
1538
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001539#ifdef CONFIG_LOCKDEP
1540static struct lockdep_map memcg_oom_lock_dep_map = {
1541 .name = "memcg_oom_lock",
1542};
1543#endif
1544
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001545static DEFINE_SPINLOCK(memcg_oom_lock);
1546
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001547/*
1548 * Check OOM-Killer is already running under our hierarchy.
1549 * If someone is running, return false.
1550 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001551static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001552{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001553 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001554
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001555 spin_lock(&memcg_oom_lock);
1556
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001557 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001558 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001559 /*
1560 * this subtree of our hierarchy is already locked
1561 * so we cannot give a lock.
1562 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001563 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001564 mem_cgroup_iter_break(memcg, iter);
1565 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001566 } else
1567 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001568 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001569
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001570 if (failed) {
1571 /*
1572 * OK, we failed to lock the whole subtree so we have
1573 * to clean up what we set up to the failing subtree
1574 */
1575 for_each_mem_cgroup_tree(iter, memcg) {
1576 if (iter == failed) {
1577 mem_cgroup_iter_break(memcg, iter);
1578 break;
1579 }
1580 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001581 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001582 } else
1583 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001584
1585 spin_unlock(&memcg_oom_lock);
1586
1587 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001588}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001589
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001590static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001591{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001592 struct mem_cgroup *iter;
1593
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001594 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001595 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001596 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001597 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001598 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001599}
1600
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001601static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001602{
1603 struct mem_cgroup *iter;
1604
Tejun Heoc2b42d32015-06-24 16:58:23 -07001605 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001606 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001607 iter->under_oom++;
1608 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001609}
1610
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001611static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001612{
1613 struct mem_cgroup *iter;
1614
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001615 /*
1616 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001617 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001618 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001619 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001620 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001621 if (iter->under_oom > 0)
1622 iter->under_oom--;
1623 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001624}
1625
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001626static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1627
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001628struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001629 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001630 wait_queue_t wait;
1631};
1632
1633static int memcg_oom_wake_function(wait_queue_t *wait,
1634 unsigned mode, int sync, void *arg)
1635{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001636 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1637 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001638 struct oom_wait_info *oom_wait_info;
1639
1640 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001641 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001642
Johannes Weiner2314b422014-12-10 15:44:33 -08001643 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1644 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001645 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001646 return autoremove_wake_function(wait, mode, sync, arg);
1647}
1648
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001649static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001650{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001651 /*
1652 * For the following lockless ->under_oom test, the only required
1653 * guarantee is that it must see the state asserted by an OOM when
1654 * this function is called as a result of userland actions
1655 * triggered by the notification of the OOM. This is trivially
1656 * achieved by invoking mem_cgroup_mark_under_oom() before
1657 * triggering notification.
1658 */
1659 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001660 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001661}
1662
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001663static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001664{
Tejun Heo626ebc42015-11-05 18:46:09 -08001665 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001666 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001667 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001668 * We are in the middle of the charge context here, so we
1669 * don't want to block when potentially sitting on a callstack
1670 * that holds all kinds of filesystem and mm locks.
1671 *
1672 * Also, the caller may handle a failed allocation gracefully
1673 * (like optional page cache readahead) and so an OOM killer
1674 * invocation might not even be necessary.
1675 *
1676 * That's why we don't do anything here except remember the
1677 * OOM context and then deal with it at the end of the page
1678 * fault when the stack is unwound, the locks are released,
1679 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001680 */
Johannes Weiner49426422013-10-16 13:46:59 -07001681 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001682 current->memcg_in_oom = memcg;
1683 current->memcg_oom_gfp_mask = mask;
1684 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001685}
1686
1687/**
1688 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1689 * @handle: actually kill/wait or just clean up the OOM state
1690 *
1691 * This has to be called at the end of a page fault if the memcg OOM
1692 * handler was enabled.
1693 *
1694 * Memcg supports userspace OOM handling where failed allocations must
1695 * sleep on a waitqueue until the userspace task resolves the
1696 * situation. Sleeping directly in the charge context with all kinds
1697 * of locks held is not a good idea, instead we remember an OOM state
1698 * in the task and mem_cgroup_oom_synchronize() has to be called at
1699 * the end of the page fault to complete the OOM handling.
1700 *
1701 * Returns %true if an ongoing memcg OOM situation was detected and
1702 * completed, %false otherwise.
1703 */
1704bool mem_cgroup_oom_synchronize(bool handle)
1705{
Tejun Heo626ebc42015-11-05 18:46:09 -08001706 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001707 struct oom_wait_info owait;
1708 bool locked;
1709
1710 /* OOM is global, do not handle */
1711 if (!memcg)
1712 return false;
1713
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001714 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001715 goto cleanup;
1716
1717 owait.memcg = memcg;
1718 owait.wait.flags = 0;
1719 owait.wait.func = memcg_oom_wake_function;
1720 owait.wait.private = current;
1721 INIT_LIST_HEAD(&owait.wait.task_list);
1722
1723 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001724 mem_cgroup_mark_under_oom(memcg);
1725
1726 locked = mem_cgroup_oom_trylock(memcg);
1727
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001728 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001729 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001730
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001731 if (locked && !memcg->oom_kill_disable) {
1732 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001733 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001734 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1735 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001736 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001737 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001738 mem_cgroup_unmark_under_oom(memcg);
1739 finish_wait(&memcg_oom_waitq, &owait.wait);
1740 }
1741
1742 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001743 mem_cgroup_oom_unlock(memcg);
1744 /*
1745 * There is no guarantee that an OOM-lock contender
1746 * sees the wakeups triggered by the OOM kill
1747 * uncharges. Wake any sleepers explicitely.
1748 */
1749 memcg_oom_recover(memcg);
1750 }
Johannes Weiner49426422013-10-16 13:46:59 -07001751cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001752 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001753 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001754 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001755}
1756
Johannes Weinerd7365e72014-10-29 14:50:48 -07001757/**
1758 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1759 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001760 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001761 * This function must mark the beginning of an accounted page state
1762 * change to prevent double accounting when the page is concurrently
1763 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001764 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001765 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001766 * if (TestClearPageState(page))
1767 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001768 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001769 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001770struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001771{
1772 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001773 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001774
Johannes Weiner6de22612015-02-11 15:25:01 -08001775 /*
1776 * The RCU lock is held throughout the transaction. The fast
1777 * path can get away without acquiring the memcg->move_lock
1778 * because page moving starts with an RCU grace period.
1779 *
1780 * The RCU lock also protects the memcg from being freed when
1781 * the page state that is going to change is the only thing
1782 * preventing the page from being uncharged.
1783 * E.g. end-writeback clearing PageWriteback(), which allows
1784 * migration to go ahead and uncharge the page before the
1785 * account transaction might be complete.
1786 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001787 rcu_read_lock();
1788
1789 if (mem_cgroup_disabled())
1790 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001791again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001792 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001793 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07001794 return NULL;
1795
Qiang Huangbdcbb652014-06-04 16:08:21 -07001796 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07001797 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001798
Johannes Weiner6de22612015-02-11 15:25:01 -08001799 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001800 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001801 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001802 goto again;
1803 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001804
1805 /*
1806 * When charge migration first begins, we can have locked and
1807 * unlocked page stat updates happening concurrently. Track
1808 * the task who has the lock for mem_cgroup_end_page_stat().
1809 */
1810 memcg->move_lock_task = current;
1811 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001812
1813 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001814}
Greg Thelenc4843a72015-05-22 17:13:16 -04001815EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001816
Johannes Weinerd7365e72014-10-29 14:50:48 -07001817/**
1818 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1819 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07001820 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001821void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001822{
Johannes Weiner6de22612015-02-11 15:25:01 -08001823 if (memcg && memcg->move_lock_task == current) {
1824 unsigned long flags = memcg->move_lock_flags;
1825
1826 memcg->move_lock_task = NULL;
1827 memcg->move_lock_flags = 0;
1828
1829 spin_unlock_irqrestore(&memcg->move_lock, flags);
1830 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001831
Johannes Weinerd7365e72014-10-29 14:50:48 -07001832 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001833}
Greg Thelenc4843a72015-05-22 17:13:16 -04001834EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001835
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001836/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001837 * size of first charge trial. "32" comes from vmscan.c's magic value.
1838 * TODO: maybe necessary to use big numbers in big irons.
1839 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001840#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001841struct memcg_stock_pcp {
1842 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001843 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001844 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001845 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001846#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001847};
1848static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001849static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001850
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001851/**
1852 * consume_stock: Try to consume stocked charge on this cpu.
1853 * @memcg: memcg to consume from.
1854 * @nr_pages: how many pages to charge.
1855 *
1856 * The charges will only happen if @memcg matches the current cpu's memcg
1857 * stock, and at least @nr_pages are available in that stock. Failure to
1858 * service an allocation will refill the stock.
1859 *
1860 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001861 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001862static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001863{
1864 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001865 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001866
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001867 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001868 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001869
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001870 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001871 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001872 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001873 ret = true;
1874 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001875 put_cpu_var(memcg_stock);
1876 return ret;
1877}
1878
1879/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001880 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001881 */
1882static void drain_stock(struct memcg_stock_pcp *stock)
1883{
1884 struct mem_cgroup *old = stock->cached;
1885
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001886 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001887 page_counter_uncharge(&old->memory, stock->nr_pages);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001888 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001889 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001890 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001891 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001892 }
1893 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001894}
1895
1896/*
1897 * This must be called under preempt disabled or must be called by
1898 * a thread which is pinned to local cpu.
1899 */
1900static void drain_local_stock(struct work_struct *dummy)
1901{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001902 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001903 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001904 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001905}
1906
1907/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001908 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001909 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001910 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001911static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001912{
1913 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1914
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001915 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001916 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001917 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001918 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001919 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001920 put_cpu_var(memcg_stock);
1921}
1922
1923/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001924 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001925 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001926 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001927static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001928{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001929 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001930
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001931 /* If someone's already draining, avoid adding running more workers. */
1932 if (!mutex_trylock(&percpu_charge_mutex))
1933 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001934 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001935 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001936 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001937 for_each_online_cpu(cpu) {
1938 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001939 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001940
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001941 memcg = stock->cached;
1942 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001943 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001944 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001945 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001946 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1947 if (cpu == curcpu)
1948 drain_local_stock(&stock->work);
1949 else
1950 schedule_work_on(cpu, &stock->work);
1951 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001952 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001953 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001954 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001955 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001956}
1957
Paul Gortmaker0db06282013-06-19 14:53:51 -04001958static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001959 unsigned long action,
1960 void *hcpu)
1961{
1962 int cpu = (unsigned long)hcpu;
1963 struct memcg_stock_pcp *stock;
1964
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001965 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001966 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001967
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001968 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001969 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001970
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001971 stock = &per_cpu(memcg_stock, cpu);
1972 drain_stock(stock);
1973 return NOTIFY_OK;
1974}
1975
Tejun Heob23afb92015-11-05 18:46:11 -08001976/*
1977 * Scheduled by try_charge() to be executed from the userland return path
1978 * and reclaims memory over the high limit.
1979 */
1980void mem_cgroup_handle_over_high(void)
1981{
1982 unsigned int nr_pages = current->memcg_nr_pages_over_high;
1983 struct mem_cgroup *memcg, *pos;
1984
1985 if (likely(!nr_pages))
1986 return;
1987
1988 pos = memcg = get_mem_cgroup_from_mm(current->mm);
1989
1990 do {
1991 if (page_counter_read(&pos->memory) <= pos->high)
1992 continue;
1993 mem_cgroup_events(pos, MEMCG_HIGH, 1);
1994 try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true);
1995 } while ((pos = parent_mem_cgroup(pos)));
1996
1997 css_put(&memcg->css);
1998 current->memcg_nr_pages_over_high = 0;
1999}
2000
Johannes Weiner00501b52014-08-08 14:19:20 -07002001static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2002 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002003{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002004 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07002005 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002006 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002007 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002008 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002009 bool may_swap = true;
2010 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002011
Johannes Weinerce00a962014-09-05 08:43:57 -04002012 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08002013 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002014retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002015 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002016 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002017
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002018 if (!do_swap_account ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002019 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2020 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002021 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002022 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002023 page_counter_uncharge(&memcg->memsw, batch);
2024 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002025 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002026 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002027 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002028 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002029
Johannes Weiner6539cc02014-08-06 16:05:42 -07002030 if (batch > nr_pages) {
2031 batch = nr_pages;
2032 goto retry;
2033 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002034
Johannes Weiner06b078f2014-08-06 16:05:44 -07002035 /*
2036 * Unlike in global OOM situations, memcg is not in a physical
2037 * memory shortage. Allow dying and OOM-killed tasks to
2038 * bypass the last charges so that they can exit quickly and
2039 * free their memory.
2040 */
2041 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2042 fatal_signal_pending(current) ||
2043 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08002044 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002045
2046 if (unlikely(task_in_memcg_oom(current)))
2047 goto nomem;
2048
Mel Gormand0164ad2015-11-06 16:28:21 -08002049 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002050 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002051
Johannes Weiner241994e2015-02-11 15:26:06 -08002052 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2053
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002054 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2055 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002056
Johannes Weiner61e02c72014-08-06 16:08:16 -07002057 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002058 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002059
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002060 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002061 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002062 drained = true;
2063 goto retry;
2064 }
2065
Johannes Weiner28c34c22014-08-06 16:05:47 -07002066 if (gfp_mask & __GFP_NORETRY)
2067 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002068 /*
2069 * Even though the limit is exceeded at this point, reclaim
2070 * may have been able to free some pages. Retry the charge
2071 * before killing the task.
2072 *
2073 * Only for regular pages, though: huge pages are rather
2074 * unlikely to succeed so close to the limit, and we fall back
2075 * to regular pages anyway in case of failure.
2076 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002077 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002078 goto retry;
2079 /*
2080 * At task move, charge accounts can be doubly counted. So, it's
2081 * better to wait until the end of task_move if something is going on.
2082 */
2083 if (mem_cgroup_wait_acct_move(mem_over_limit))
2084 goto retry;
2085
Johannes Weiner9b130612014-08-06 16:05:51 -07002086 if (nr_retries--)
2087 goto retry;
2088
Johannes Weiner06b078f2014-08-06 16:05:44 -07002089 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002090 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002091
Johannes Weiner6539cc02014-08-06 16:05:42 -07002092 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002093 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002094
Johannes Weiner241994e2015-02-11 15:26:06 -08002095 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2096
Jerome Marchand3608de02015-11-05 18:47:29 -08002097 mem_cgroup_oom(mem_over_limit, gfp_mask,
2098 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002099nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002100 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002101 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002102force:
2103 /*
2104 * The allocation either can't fail or will lead to more memory
2105 * being freed very soon. Allow memory usage go over the limit
2106 * temporarily by force charging it.
2107 */
2108 page_counter_charge(&memcg->memory, nr_pages);
2109 if (do_swap_account)
2110 page_counter_charge(&memcg->memsw, nr_pages);
2111 css_get_many(&memcg->css, nr_pages);
2112
2113 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002114
2115done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002116 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002117 if (batch > nr_pages)
2118 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002119
Johannes Weiner241994e2015-02-11 15:26:06 -08002120 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002121 * If the hierarchy is above the normal consumption range, schedule
2122 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002123 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002124 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2125 * not recorded as it most likely matches current's and won't
2126 * change in the meantime. As high limit is checked again before
2127 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002128 */
2129 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002130 if (page_counter_read(&memcg->memory) > memcg->high) {
2131 current->memcg_nr_pages_over_high += nr_pages;
2132 set_notify_resume(current);
2133 break;
2134 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002135 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002136
2137 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002138}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002139
Johannes Weiner00501b52014-08-08 14:19:20 -07002140static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002141{
Johannes Weinerce00a962014-09-05 08:43:57 -04002142 if (mem_cgroup_is_root(memcg))
2143 return;
2144
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002145 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner05b84302014-08-06 16:05:59 -07002146 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002147 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002148
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002149 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002150}
2151
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002152static void lock_page_lru(struct page *page, int *isolated)
2153{
2154 struct zone *zone = page_zone(page);
2155
2156 spin_lock_irq(&zone->lru_lock);
2157 if (PageLRU(page)) {
2158 struct lruvec *lruvec;
2159
2160 lruvec = mem_cgroup_page_lruvec(page, zone);
2161 ClearPageLRU(page);
2162 del_page_from_lru_list(page, lruvec, page_lru(page));
2163 *isolated = 1;
2164 } else
2165 *isolated = 0;
2166}
2167
2168static void unlock_page_lru(struct page *page, int isolated)
2169{
2170 struct zone *zone = page_zone(page);
2171
2172 if (isolated) {
2173 struct lruvec *lruvec;
2174
2175 lruvec = mem_cgroup_page_lruvec(page, zone);
2176 VM_BUG_ON_PAGE(PageLRU(page), page);
2177 SetPageLRU(page);
2178 add_page_to_lru_list(page, lruvec, page_lru(page));
2179 }
2180 spin_unlock_irq(&zone->lru_lock);
2181}
2182
Johannes Weiner00501b52014-08-08 14:19:20 -07002183static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002184 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002185{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002186 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002187
Johannes Weiner1306a852014-12-10 15:44:52 -08002188 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002189
2190 /*
2191 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2192 * may already be on some other mem_cgroup's LRU. Take care of it.
2193 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002194 if (lrucare)
2195 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002196
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002197 /*
2198 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002199 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002200 *
2201 * - the page is uncharged
2202 *
2203 * - the page is off-LRU
2204 *
2205 * - an anonymous fault has exclusive page access, except for
2206 * a locked page table
2207 *
2208 * - a page cache insertion, a swapin fault, or a migration
2209 * have the page locked
2210 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002211 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002212
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002213 if (lrucare)
2214 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002215}
2216
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002217#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002218static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002219{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002220 int id, size;
2221 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002222
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002223 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002224 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2225 if (id < 0)
2226 return id;
2227
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002228 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002229 return id;
2230
2231 /*
2232 * There's no space for the new id in memcg_caches arrays,
2233 * so we have to grow them.
2234 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002235 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002236
2237 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002238 if (size < MEMCG_CACHES_MIN_SIZE)
2239 size = MEMCG_CACHES_MIN_SIZE;
2240 else if (size > MEMCG_CACHES_MAX_SIZE)
2241 size = MEMCG_CACHES_MAX_SIZE;
2242
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002243 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002244 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002245 err = memcg_update_all_list_lrus(size);
2246 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002247 memcg_nr_cache_ids = size;
2248
2249 up_write(&memcg_cache_ids_sem);
2250
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002251 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002252 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002253 return err;
2254 }
2255 return id;
2256}
2257
2258static void memcg_free_cache_id(int id)
2259{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002260 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002261}
2262
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002263struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002264 struct mem_cgroup *memcg;
2265 struct kmem_cache *cachep;
2266 struct work_struct work;
2267};
2268
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002269static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002270{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002271 struct memcg_kmem_cache_create_work *cw =
2272 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002273 struct mem_cgroup *memcg = cw->memcg;
2274 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002275
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002276 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002277
Vladimir Davydov5722d092014-04-07 15:39:24 -07002278 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002279 kfree(cw);
2280}
2281
2282/*
2283 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002284 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002285static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2286 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002287{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002288 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002289
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002290 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002291 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002292 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002293
2294 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002295
2296 cw->memcg = memcg;
2297 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002298 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002299
Glauber Costad7f25f82012-12-18 14:22:40 -08002300 schedule_work(&cw->work);
2301}
2302
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002303static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2304 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002305{
2306 /*
2307 * We need to stop accounting when we kmalloc, because if the
2308 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002309 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002310 *
2311 * However, it is better to enclose the whole function. Depending on
2312 * the debugging options enabled, INIT_WORK(), for instance, can
2313 * trigger an allocation. This too, will make us recurse. Because at
2314 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2315 * the safest choice is to do it like this, wrapping the whole function.
2316 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002317 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002318 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002319 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002320}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002321
Glauber Costad7f25f82012-12-18 14:22:40 -08002322/*
2323 * Return the kmem_cache we're supposed to use for a slab allocation.
2324 * We try to use the current memcg's version of the cache.
2325 *
2326 * If the cache does not exist yet, if we are the first user of it,
2327 * we either create it immediately, if possible, or create it asynchronously
2328 * in a workqueue.
2329 * In the latter case, we will let the current allocation go through with
2330 * the original cache.
2331 *
2332 * Can't be called in interrupt context or from kernel threads.
2333 * This function needs to be called with rcu_read_lock() held.
2334 */
Zhang Zhen056b7cc2014-12-12 16:55:38 -08002335struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002336{
2337 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002338 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002339 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002340
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002341 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002342
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002343 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002344 return cachep;
2345
Vladimir Davydov8135be52014-12-12 16:56:38 -08002346 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002347 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002348 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002349 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002350
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002351 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002352 if (likely(memcg_cachep))
2353 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002354
2355 /*
2356 * If we are in a safe context (can wait, and not in interrupt
2357 * context), we could be be predictable and return right away.
2358 * This would guarantee that the allocation being performed
2359 * already belongs in the new cache.
2360 *
2361 * However, there are some clashes that can arrive from locking.
2362 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002363 * memcg_create_kmem_cache, this means no further allocation
2364 * could happen with the slab_mutex held. So it's better to
2365 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002366 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002367 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002368out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002369 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002370 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002371}
Glauber Costad7f25f82012-12-18 14:22:40 -08002372
Vladimir Davydov8135be52014-12-12 16:56:38 -08002373void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2374{
2375 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002376 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002377}
2378
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002379int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2380 struct mem_cgroup *memcg)
2381{
2382 unsigned int nr_pages = 1 << order;
2383 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002384 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002385
2386 if (!memcg_kmem_is_active(memcg))
2387 return 0;
2388
Johannes Weiner6071ca52015-11-05 18:50:26 -08002389 if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter))
2390 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002391
2392 ret = try_charge(memcg, gfp, nr_pages);
2393 if (ret) {
2394 page_counter_uncharge(&memcg->kmem, nr_pages);
2395 return ret;
2396 }
2397
2398 page->mem_cgroup = memcg;
2399
2400 return 0;
2401}
2402
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002403int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002404{
2405 struct mem_cgroup *memcg;
2406 int ret;
2407
Johannes Weinerdf381972014-04-07 15:37:43 -07002408 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002409 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002410 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002411 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002412}
2413
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002414void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002415{
Johannes Weiner1306a852014-12-10 15:44:52 -08002416 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002417 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002418
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002419 if (!memcg)
2420 return;
2421
Sasha Levin309381fea2014-01-23 15:52:54 -08002422 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002423
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002424 page_counter_uncharge(&memcg->kmem, nr_pages);
2425 page_counter_uncharge(&memcg->memory, nr_pages);
2426 if (do_swap_account)
2427 page_counter_uncharge(&memcg->memsw, nr_pages);
2428
Johannes Weiner1306a852014-12-10 15:44:52 -08002429 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002430 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002431}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002432#endif /* CONFIG_MEMCG_KMEM */
2433
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002434#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2435
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002436/*
2437 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002438 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2439 * charge/uncharge will be never happen and move_account() is done under
2440 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002441 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002442void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002443{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002444 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002445
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002446 if (mem_cgroup_disabled())
2447 return;
David Rientjesb070e652013-05-07 16:18:09 -07002448
Johannes Weiner29833312014-12-10 15:44:02 -08002449 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002450 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002451
Johannes Weiner1306a852014-12-10 15:44:52 -08002452 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002453 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002454}
Hugh Dickins12d27102012-01-12 17:19:52 -08002455#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002456
Andrew Mortonc255a452012-07-31 16:43:02 -07002457#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002458static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2459 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002460{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002461 int val = (charge) ? 1 : -1;
2462 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002463}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002464
2465/**
2466 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2467 * @entry: swap entry to be moved
2468 * @from: mem_cgroup which the entry is moved from
2469 * @to: mem_cgroup which the entry is moved to
2470 *
2471 * It succeeds only when the swap_cgroup's record for this entry is the same
2472 * as the mem_cgroup's id of @from.
2473 *
2474 * Returns 0 on success, -EINVAL on failure.
2475 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002476 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002477 * both res and memsw, and called css_get().
2478 */
2479static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002480 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002481{
2482 unsigned short old_id, new_id;
2483
Li Zefan34c00c32013-09-23 16:56:01 +08002484 old_id = mem_cgroup_id(from);
2485 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002486
2487 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002488 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002489 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002490 return 0;
2491 }
2492 return -EINVAL;
2493}
2494#else
2495static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002496 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002497{
2498 return -EINVAL;
2499}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002500#endif
2501
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002502static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002503
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002504static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002506{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002507 unsigned long curusage;
2508 unsigned long oldusage;
2509 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002510 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002511 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002512
2513 /*
2514 * For keeping hierarchical_reclaim simple, how long we should retry
2515 * is depends on callers. We set our retry-count to be function
2516 * of # of children which we should visit in this loop.
2517 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002518 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2519 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002520
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002521 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002522
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002523 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002524 if (signal_pending(current)) {
2525 ret = -EINTR;
2526 break;
2527 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002528
2529 mutex_lock(&memcg_limit_mutex);
2530 if (limit > memcg->memsw.limit) {
2531 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002532 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002533 break;
2534 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002535 if (limit > memcg->memory.limit)
2536 enlarge = true;
2537 ret = page_counter_limit(&memcg->memory, limit);
2538 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002539
2540 if (!ret)
2541 break;
2542
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002543 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2544
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002545 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002546 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002547 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002548 retry_count--;
2549 else
2550 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002551 } while (retry_count);
2552
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002553 if (!ret && enlarge)
2554 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002555
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002556 return ret;
2557}
2558
Li Zefan338c8432009-06-17 16:27:15 -07002559static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002560 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002561{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002562 unsigned long curusage;
2563 unsigned long oldusage;
2564 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002565 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002566 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002567
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002568 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002569 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2570 mem_cgroup_count_children(memcg);
2571
2572 oldusage = page_counter_read(&memcg->memsw);
2573
2574 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002575 if (signal_pending(current)) {
2576 ret = -EINTR;
2577 break;
2578 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002579
2580 mutex_lock(&memcg_limit_mutex);
2581 if (limit < memcg->memory.limit) {
2582 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002583 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002584 break;
2585 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002586 if (limit > memcg->memsw.limit)
2587 enlarge = true;
2588 ret = page_counter_limit(&memcg->memsw, limit);
2589 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002590
2591 if (!ret)
2592 break;
2593
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002594 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2595
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002596 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002597 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002598 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002599 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002600 else
2601 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002602 } while (retry_count);
2603
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002604 if (!ret && enlarge)
2605 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002606
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002607 return ret;
2608}
2609
Andrew Morton0608f432013-09-24 15:27:41 -07002610unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2611 gfp_t gfp_mask,
2612 unsigned long *total_scanned)
2613{
2614 unsigned long nr_reclaimed = 0;
2615 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2616 unsigned long reclaimed;
2617 int loop = 0;
2618 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002619 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002620 unsigned long nr_scanned;
2621
2622 if (order > 0)
2623 return 0;
2624
2625 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2626 /*
2627 * This loop can run a while, specially if mem_cgroup's continuously
2628 * keep exceeding their soft limit and putting the system under
2629 * pressure
2630 */
2631 do {
2632 if (next_mz)
2633 mz = next_mz;
2634 else
2635 mz = mem_cgroup_largest_soft_limit_node(mctz);
2636 if (!mz)
2637 break;
2638
2639 nr_scanned = 0;
2640 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2641 gfp_mask, &nr_scanned);
2642 nr_reclaimed += reclaimed;
2643 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002644 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002645 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002646
2647 /*
2648 * If we failed to reclaim anything from this memory cgroup
2649 * it is time to move on to the next cgroup
2650 */
2651 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002652 if (!reclaimed)
2653 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2654
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002655 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002656 /*
2657 * One school of thought says that we should not add
2658 * back the node to the tree if reclaim returns 0.
2659 * But our reclaim could return 0, simply because due
2660 * to priority we are exposing a smaller subset of
2661 * memory to reclaim from. Consider this as a longer
2662 * term TODO.
2663 */
2664 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002665 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002666 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002667 css_put(&mz->memcg->css);
2668 loop++;
2669 /*
2670 * Could not reclaim anything and there are no more
2671 * mem cgroups to try or we seem to be looping without
2672 * reclaiming anything.
2673 */
2674 if (!nr_reclaimed &&
2675 (next_mz == NULL ||
2676 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2677 break;
2678 } while (!nr_reclaimed);
2679 if (next_mz)
2680 css_put(&next_mz->memcg->css);
2681 return nr_reclaimed;
2682}
2683
Tejun Heoea280e72014-05-16 13:22:48 -04002684/*
2685 * Test whether @memcg has children, dead or alive. Note that this
2686 * function doesn't care whether @memcg has use_hierarchy enabled and
2687 * returns %true if there are child csses according to the cgroup
2688 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2689 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002690static inline bool memcg_has_children(struct mem_cgroup *memcg)
2691{
Tejun Heoea280e72014-05-16 13:22:48 -04002692 bool ret;
2693
Johannes Weiner696ac172013-10-31 16:34:15 -07002694 /*
Tejun Heoea280e72014-05-16 13:22:48 -04002695 * The lock does not prevent addition or deletion of children, but
2696 * it prevents a new child from being initialized based on this
2697 * parent in css_online(), so it's enough to decide whether
2698 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07002699 */
Tejun Heoea280e72014-05-16 13:22:48 -04002700 lockdep_assert_held(&memcg_create_mutex);
2701
2702 rcu_read_lock();
2703 ret = css_next_child(NULL, &memcg->css);
2704 rcu_read_unlock();
2705 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002706}
2707
2708/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002709 * Reclaims as many pages from the given memcg as possible and moves
2710 * the rest to the parent.
2711 *
2712 * Caller is responsible for holding css reference for memcg.
2713 */
2714static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2715{
2716 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002717
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002718 /* we call try-to-free pages for make this cgroup empty */
2719 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002720 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002721 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002722 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002723
Michal Hockoc26251f2012-10-26 13:37:28 +02002724 if (signal_pending(current))
2725 return -EINTR;
2726
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002727 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2728 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002729 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002730 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002731 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002732 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002733 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002734
2735 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002736
2737 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002738}
2739
Tejun Heo6770c642014-05-13 12:16:21 -04002740static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2741 char *buf, size_t nbytes,
2742 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002743{
Tejun Heo6770c642014-05-13 12:16:21 -04002744 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002745
Michal Hockod8423012012-10-26 13:37:29 +02002746 if (mem_cgroup_is_root(memcg))
2747 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002748 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002749}
2750
Tejun Heo182446d2013-08-08 20:11:24 -04002751static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2752 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002753{
Tejun Heo182446d2013-08-08 20:11:24 -04002754 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002755}
2756
Tejun Heo182446d2013-08-08 20:11:24 -04002757static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2758 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002759{
2760 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002761 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002762 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002763
Glauber Costa09998212013-02-22 16:34:55 -08002764 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07002765
2766 if (memcg->use_hierarchy == val)
2767 goto out;
2768
Balbir Singh18f59ea2009-01-07 18:08:07 -08002769 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002770 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002771 * in the child subtrees. If it is unset, then the change can
2772 * occur, provided the current cgroup has no children.
2773 *
2774 * For the root cgroup, parent_mem is NULL, we allow value to be
2775 * set if there are no children.
2776 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002777 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002778 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002779 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002780 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002781 else
2782 retval = -EBUSY;
2783 } else
2784 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002785
2786out:
Glauber Costa09998212013-02-22 16:34:55 -08002787 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002788
2789 return retval;
2790}
2791
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792static unsigned long tree_stat(struct mem_cgroup *memcg,
2793 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04002794{
2795 struct mem_cgroup *iter;
Greg Thelen484ebb32015-10-01 15:37:05 -07002796 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002797
Johannes Weinerce00a962014-09-05 08:43:57 -04002798 for_each_mem_cgroup_tree(iter, memcg)
2799 val += mem_cgroup_read_stat(iter, idx);
2800
Johannes Weinerce00a962014-09-05 08:43:57 -04002801 return val;
2802}
2803
Andrew Morton6f646152015-11-06 16:28:58 -08002804static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002805{
Michal Hockoc12176d2015-11-05 18:50:29 -08002806 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002807
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002808 if (mem_cgroup_is_root(memcg)) {
2809 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
2810 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
2811 if (swap)
2812 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
2813 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002814 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002815 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002816 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002817 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002818 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002819 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002820}
2821
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002822enum {
2823 RES_USAGE,
2824 RES_LIMIT,
2825 RES_MAX_USAGE,
2826 RES_FAILCNT,
2827 RES_SOFT_LIMIT,
2828};
Johannes Weinerce00a962014-09-05 08:43:57 -04002829
Tejun Heo791badb2013-12-05 12:28:02 -05002830static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002831 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002832{
Tejun Heo182446d2013-08-08 20:11:24 -04002833 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002834 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002835
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002836 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002837 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002838 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002839 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002840 case _MEMSWAP:
2841 counter = &memcg->memsw;
2842 break;
2843 case _KMEM:
2844 counter = &memcg->kmem;
2845 break;
2846 default:
2847 BUG();
2848 }
2849
2850 switch (MEMFILE_ATTR(cft->private)) {
2851 case RES_USAGE:
2852 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002853 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002854 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002855 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002856 return (u64)page_counter_read(counter) * PAGE_SIZE;
2857 case RES_LIMIT:
2858 return (u64)counter->limit * PAGE_SIZE;
2859 case RES_MAX_USAGE:
2860 return (u64)counter->watermark * PAGE_SIZE;
2861 case RES_FAILCNT:
2862 return counter->failcnt;
2863 case RES_SOFT_LIMIT:
2864 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002865 default:
2866 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002867 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002868}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002869
Glauber Costa510fc4e2012-12-18 14:21:47 -08002870#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002871static int memcg_activate_kmem(struct mem_cgroup *memcg,
2872 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002873{
2874 int err = 0;
2875 int memcg_id;
2876
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002877 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002878 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002879 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002880
2881 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08002882 * For simplicity, we won't allow this to be disabled. It also can't
2883 * be changed if the cgroup has children already, or if tasks had
2884 * already joined.
2885 *
2886 * If tasks join before we set the limit, a person looking at
2887 * kmem.usage_in_bytes will have no way to determine when it took
2888 * place, which makes the value quite meaningless.
2889 *
2890 * After it first became limited, changes in the value of the limit are
2891 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08002892 */
Glauber Costa09998212013-02-22 16:34:55 -08002893 mutex_lock(&memcg_create_mutex);
Tejun Heo27bd4db2015-10-15 16:41:50 -04002894 if (cgroup_is_populated(memcg->css.cgroup) ||
Tejun Heoea280e72014-05-16 13:22:48 -04002895 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08002896 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08002897 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002898 if (err)
2899 goto out;
2900
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002901 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08002902 if (memcg_id < 0) {
2903 err = memcg_id;
2904 goto out;
2905 }
2906
Vladimir Davydovd6441632014-01-23 15:53:09 -08002907 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002908 * We couldn't have accounted to this cgroup, because it hasn't got
2909 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08002910 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002911 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002912 VM_BUG_ON(err);
2913
2914 static_key_slow_inc(&memcg_kmem_enabled_key);
2915 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002916 * A memory cgroup is considered kmem-active as soon as it gets
2917 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002918 * guarantee no one starts accounting before all call sites are
2919 * patched.
2920 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002921 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002922 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002923 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002924out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08002925 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002926}
2927
Vladimir Davydovd6441632014-01-23 15:53:09 -08002928static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002929 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002930{
2931 int ret;
2932
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002933 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002934 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002935 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002936 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002937 ret = page_counter_limit(&memcg->kmem, limit);
2938 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002939 return ret;
2940}
2941
Glauber Costa55007d82012-12-18 14:22:38 -08002942static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08002943{
Glauber Costa55007d82012-12-18 14:22:38 -08002944 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002945 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002946
Glauber Costa510fc4e2012-12-18 14:21:47 -08002947 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002948 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08002949
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002950 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08002951 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08002952 * If the parent cgroup is not kmem-active now, it cannot be activated
2953 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08002954 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08002955 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002956 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
2957 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08002958 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002959}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002960#else
2961static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002962 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002963{
2964 return -EINVAL;
2965}
Hugh Dickins6d0439902013-02-22 16:35:50 -08002966#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08002967
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002968/*
2969 * The user of this function is...
2970 * RES_LIMIT.
2971 */
Tejun Heo451af502014-05-13 12:16:21 -04002972static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2973 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002974{
Tejun Heo451af502014-05-13 12:16:21 -04002975 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002976 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002977 int ret;
2978
Tejun Heo451af502014-05-13 12:16:21 -04002979 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002980 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002981 if (ret)
2982 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002983
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002984 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002985 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002986 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2987 ret = -EINVAL;
2988 break;
2989 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002990 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2991 case _MEM:
2992 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002993 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002994 case _MEMSWAP:
2995 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2996 break;
2997 case _KMEM:
2998 ret = memcg_update_kmem_limit(memcg, nr_pages);
2999 break;
3000 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003001 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003002 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003003 memcg->soft_limit = nr_pages;
3004 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003005 break;
3006 }
Tejun Heo451af502014-05-13 12:16:21 -04003007 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003008}
3009
Tejun Heo6770c642014-05-13 12:16:21 -04003010static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3011 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003012{
Tejun Heo6770c642014-05-13 12:16:21 -04003013 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003014 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003015
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003016 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3017 case _MEM:
3018 counter = &memcg->memory;
3019 break;
3020 case _MEMSWAP:
3021 counter = &memcg->memsw;
3022 break;
3023 case _KMEM:
3024 counter = &memcg->kmem;
3025 break;
3026 default:
3027 BUG();
3028 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003029
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003030 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003031 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003032 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003033 break;
3034 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003035 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003036 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003037 default:
3038 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003039 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003040
Tejun Heo6770c642014-05-13 12:16:21 -04003041 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003042}
3043
Tejun Heo182446d2013-08-08 20:11:24 -04003044static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003045 struct cftype *cft)
3046{
Tejun Heo182446d2013-08-08 20:11:24 -04003047 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003048}
3049
Daisuke Nishimura02491442010-03-10 15:22:17 -08003050#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003051static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003052 struct cftype *cft, u64 val)
3053{
Tejun Heo182446d2013-08-08 20:11:24 -04003054 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003055
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003056 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003057 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058
Glauber Costaee5e8472013-02-22 16:34:50 -08003059 /*
3060 * No kind of locking is needed in here, because ->can_attach() will
3061 * check this value once in the beginning of the process, and then carry
3062 * on with stale data. This means that changes to this value will only
3063 * affect task migrations starting after the change.
3064 */
3065 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003066 return 0;
3067}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003068#else
Tejun Heo182446d2013-08-08 20:11:24 -04003069static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003070 struct cftype *cft, u64 val)
3071{
3072 return -ENOSYS;
3073}
3074#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003075
Ying Han406eb0c2011-05-26 16:25:37 -07003076#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003077static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003078{
Greg Thelen25485de2013-11-12 15:07:40 -08003079 struct numa_stat {
3080 const char *name;
3081 unsigned int lru_mask;
3082 };
3083
3084 static const struct numa_stat stats[] = {
3085 { "total", LRU_ALL },
3086 { "file", LRU_ALL_FILE },
3087 { "anon", LRU_ALL_ANON },
3088 { "unevictable", BIT(LRU_UNEVICTABLE) },
3089 };
3090 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003091 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003092 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003093 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003094
Greg Thelen25485de2013-11-12 15:07:40 -08003095 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3096 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3097 seq_printf(m, "%s=%lu", stat->name, nr);
3098 for_each_node_state(nid, N_MEMORY) {
3099 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3100 stat->lru_mask);
3101 seq_printf(m, " N%d=%lu", nid, nr);
3102 }
3103 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003104 }
Ying Han406eb0c2011-05-26 16:25:37 -07003105
Ying Han071aee12013-11-12 15:07:41 -08003106 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3107 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003108
Ying Han071aee12013-11-12 15:07:41 -08003109 nr = 0;
3110 for_each_mem_cgroup_tree(iter, memcg)
3111 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3112 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3113 for_each_node_state(nid, N_MEMORY) {
3114 nr = 0;
3115 for_each_mem_cgroup_tree(iter, memcg)
3116 nr += mem_cgroup_node_nr_lru_pages(
3117 iter, nid, stat->lru_mask);
3118 seq_printf(m, " N%d=%lu", nid, nr);
3119 }
3120 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003121 }
Ying Han406eb0c2011-05-26 16:25:37 -07003122
Ying Han406eb0c2011-05-26 16:25:37 -07003123 return 0;
3124}
3125#endif /* CONFIG_NUMA */
3126
Tejun Heo2da8ca82013-12-05 12:28:04 -05003127static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003128{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003129 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003130 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003131 struct mem_cgroup *mi;
3132 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003133
Greg Thelen0ca44b12015-02-11 15:25:58 -08003134 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3135 MEM_CGROUP_STAT_NSTATS);
3136 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3137 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003138 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3139
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003140 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003141 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003142 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003143 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003144 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003145 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003146
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003147 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3148 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3149 mem_cgroup_read_events(memcg, i));
3150
3151 for (i = 0; i < NR_LRU_LISTS; i++)
3152 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3153 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3154
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003155 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003156 memory = memsw = PAGE_COUNTER_MAX;
3157 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3158 memory = min(memory, mi->memory.limit);
3159 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003160 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003161 seq_printf(m, "hierarchical_memory_limit %llu\n",
3162 (u64)memory * PAGE_SIZE);
3163 if (do_swap_account)
3164 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3165 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003166
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003167 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003168 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003169
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003170 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003171 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003172 for_each_mem_cgroup_tree(mi, memcg)
3173 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003174 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003175 }
3176
3177 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3178 unsigned long long val = 0;
3179
3180 for_each_mem_cgroup_tree(mi, memcg)
3181 val += mem_cgroup_read_events(mi, i);
3182 seq_printf(m, "total_%s %llu\n",
3183 mem_cgroup_events_names[i], val);
3184 }
3185
3186 for (i = 0; i < NR_LRU_LISTS; i++) {
3187 unsigned long long val = 0;
3188
3189 for_each_mem_cgroup_tree(mi, memcg)
3190 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3191 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003192 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003193
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003194#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003195 {
3196 int nid, zid;
3197 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003198 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003199 unsigned long recent_rotated[2] = {0, 0};
3200 unsigned long recent_scanned[2] = {0, 0};
3201
3202 for_each_online_node(nid)
3203 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003204 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003205 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003206
Hugh Dickins89abfab2012-05-29 15:06:53 -07003207 recent_rotated[0] += rstat->recent_rotated[0];
3208 recent_rotated[1] += rstat->recent_rotated[1];
3209 recent_scanned[0] += rstat->recent_scanned[0];
3210 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003211 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003212 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3213 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3214 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3215 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003216 }
3217#endif
3218
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003219 return 0;
3220}
3221
Tejun Heo182446d2013-08-08 20:11:24 -04003222static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3223 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003224{
Tejun Heo182446d2013-08-08 20:11:24 -04003225 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003226
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003227 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003228}
3229
Tejun Heo182446d2013-08-08 20:11:24 -04003230static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3231 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003232{
Tejun Heo182446d2013-08-08 20:11:24 -04003233 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003234
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003235 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003236 return -EINVAL;
3237
Linus Torvalds14208b02014-06-09 15:03:33 -07003238 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003239 memcg->swappiness = val;
3240 else
3241 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003242
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003243 return 0;
3244}
3245
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003246static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3247{
3248 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003249 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003250 int i;
3251
3252 rcu_read_lock();
3253 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003254 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003255 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003256 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003257
3258 if (!t)
3259 goto unlock;
3260
Johannes Weinerce00a962014-09-05 08:43:57 -04003261 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003262
3263 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003264 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003265 * If it's not true, a threshold was crossed after last
3266 * call of __mem_cgroup_threshold().
3267 */
Phil Carmody5407a562010-05-26 14:42:42 -07003268 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003269
3270 /*
3271 * Iterate backward over array of thresholds starting from
3272 * current_threshold and check if a threshold is crossed.
3273 * If none of thresholds below usage is crossed, we read
3274 * only one element of the array here.
3275 */
3276 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3277 eventfd_signal(t->entries[i].eventfd, 1);
3278
3279 /* i = current_threshold + 1 */
3280 i++;
3281
3282 /*
3283 * Iterate forward over array of thresholds starting from
3284 * current_threshold+1 and check if a threshold is crossed.
3285 * If none of thresholds above usage is crossed, we read
3286 * only one element of the array here.
3287 */
3288 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3289 eventfd_signal(t->entries[i].eventfd, 1);
3290
3291 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003292 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003293unlock:
3294 rcu_read_unlock();
3295}
3296
3297static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3298{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003299 while (memcg) {
3300 __mem_cgroup_threshold(memcg, false);
3301 if (do_swap_account)
3302 __mem_cgroup_threshold(memcg, true);
3303
3304 memcg = parent_mem_cgroup(memcg);
3305 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003306}
3307
3308static int compare_thresholds(const void *a, const void *b)
3309{
3310 const struct mem_cgroup_threshold *_a = a;
3311 const struct mem_cgroup_threshold *_b = b;
3312
Greg Thelen2bff24a2013-09-11 14:23:08 -07003313 if (_a->threshold > _b->threshold)
3314 return 1;
3315
3316 if (_a->threshold < _b->threshold)
3317 return -1;
3318
3319 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003320}
3321
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003322static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003323{
3324 struct mem_cgroup_eventfd_list *ev;
3325
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003326 spin_lock(&memcg_oom_lock);
3327
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003328 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003329 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003330
3331 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003332 return 0;
3333}
3334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003335static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003336{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003337 struct mem_cgroup *iter;
3338
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003339 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003340 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003341}
3342
Tejun Heo59b6f872013-11-22 18:20:43 -05003343static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003344 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003345{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003346 struct mem_cgroup_thresholds *thresholds;
3347 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003348 unsigned long threshold;
3349 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003350 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003351
Johannes Weiner650c5e52015-02-11 15:26:03 -08003352 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003353 if (ret)
3354 return ret;
3355
3356 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003357
Johannes Weiner05b84302014-08-06 16:05:59 -07003358 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003359 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003360 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003361 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003362 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003363 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003364 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003365 BUG();
3366
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003367 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003368 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003369 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3370
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003371 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003372
3373 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003374 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003375 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003376 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003377 ret = -ENOMEM;
3378 goto unlock;
3379 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381
3382 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003383 if (thresholds->primary) {
3384 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003385 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 }
3387
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003388 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 new->entries[size - 1].eventfd = eventfd;
3390 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391
3392 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003393 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 compare_thresholds, NULL);
3395
3396 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003397 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003398 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003399 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003401 * new->current_threshold will not be used until
3402 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403 * it here.
3404 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003405 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003406 } else
3407 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408 }
3409
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 /* Free old spare buffer and save old primary buffer as spare */
3411 kfree(thresholds->spare);
3412 thresholds->spare = thresholds->primary;
3413
3414 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003416 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 synchronize_rcu();
3418
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003419unlock:
3420 mutex_unlock(&memcg->thresholds_lock);
3421
3422 return ret;
3423}
3424
Tejun Heo59b6f872013-11-22 18:20:43 -05003425static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003426 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427{
Tejun Heo59b6f872013-11-22 18:20:43 -05003428 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003429}
3430
Tejun Heo59b6f872013-11-22 18:20:43 -05003431static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003432 struct eventfd_ctx *eventfd, const char *args)
3433{
Tejun Heo59b6f872013-11-22 18:20:43 -05003434 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003435}
3436
Tejun Heo59b6f872013-11-22 18:20:43 -05003437static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003438 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003439{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003440 struct mem_cgroup_thresholds *thresholds;
3441 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003442 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003443 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444
3445 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003446
3447 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003448 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003449 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003450 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003451 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003452 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003453 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003454 BUG();
3455
Anton Vorontsov371528c2012-02-24 05:14:46 +04003456 if (!thresholds->primary)
3457 goto unlock;
3458
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003459 /* Check if a threshold crossed before removing */
3460 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3461
3462 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003463 size = 0;
3464 for (i = 0; i < thresholds->primary->size; i++) {
3465 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466 size++;
3467 }
3468
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003469 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003470
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471 /* Set thresholds array to NULL if we don't have thresholds */
3472 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003473 kfree(new);
3474 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003475 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003476 }
3477
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003479
3480 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003481 new->current_threshold = -1;
3482 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3483 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003484 continue;
3485
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003486 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003487 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003489 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003490 * until rcu_assign_pointer(), so it's safe to increment
3491 * it here.
3492 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003493 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003494 }
3495 j++;
3496 }
3497
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003498swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003499 /* Swap primary and spare array */
3500 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003501 /* If all events are unregistered, free the spare array */
3502 if (!new) {
3503 kfree(thresholds->spare);
3504 thresholds->spare = NULL;
3505 }
3506
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003507 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003508
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003509 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003510 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04003511unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003512 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003513}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003514
Tejun Heo59b6f872013-11-22 18:20:43 -05003515static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003516 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003517{
Tejun Heo59b6f872013-11-22 18:20:43 -05003518 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003519}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003520
Tejun Heo59b6f872013-11-22 18:20:43 -05003521static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003522 struct eventfd_ctx *eventfd)
3523{
Tejun Heo59b6f872013-11-22 18:20:43 -05003524 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003525}
3526
Tejun Heo59b6f872013-11-22 18:20:43 -05003527static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003528 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003529{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003531
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003532 event = kmalloc(sizeof(*event), GFP_KERNEL);
3533 if (!event)
3534 return -ENOMEM;
3535
Michal Hocko1af8efe2011-07-26 16:08:24 -07003536 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003537
3538 event->eventfd = eventfd;
3539 list_add(&event->list, &memcg->oom_notify);
3540
3541 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003542 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003543 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003544 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545
3546 return 0;
3547}
3548
Tejun Heo59b6f872013-11-22 18:20:43 -05003549static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003550 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003551{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003552 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553
Michal Hocko1af8efe2011-07-26 16:08:24 -07003554 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003556 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003557 if (ev->eventfd == eventfd) {
3558 list_del(&ev->list);
3559 kfree(ev);
3560 }
3561 }
3562
Michal Hocko1af8efe2011-07-26 16:08:24 -07003563 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003564}
3565
Tejun Heo2da8ca82013-12-05 12:28:04 -05003566static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003567{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003568 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003569
Tejun Heo791badb2013-12-05 12:28:02 -05003570 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003571 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003572 return 0;
3573}
3574
Tejun Heo182446d2013-08-08 20:11:24 -04003575static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003576 struct cftype *cft, u64 val)
3577{
Tejun Heo182446d2013-08-08 20:11:24 -04003578 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003579
3580 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003581 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003582 return -EINVAL;
3583
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003584 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003585 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003586 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003587
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003588 return 0;
3589}
3590
Andrew Mortonc255a452012-07-31 16:43:02 -07003591#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003592static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003593{
Glauber Costa55007d82012-12-18 14:22:38 -08003594 int ret;
3595
Glauber Costa55007d82012-12-18 14:22:38 -08003596 ret = memcg_propagate_kmem(memcg);
3597 if (ret)
3598 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003599
Glauber Costa1d62e432012-04-09 19:36:33 -03003600 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003601}
Glauber Costae5671df2011-12-11 21:47:01 +00003602
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003603static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3604{
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003605 struct cgroup_subsys_state *css;
3606 struct mem_cgroup *parent, *child;
3607 int kmemcg_id;
3608
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003609 if (!memcg->kmem_acct_active)
3610 return;
3611
3612 /*
3613 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3614 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3615 * guarantees no cache will be created for this cgroup after we are
3616 * done (see memcg_create_kmem_cache()).
3617 */
3618 memcg->kmem_acct_active = false;
3619
3620 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003621
3622 kmemcg_id = memcg->kmemcg_id;
3623 BUG_ON(kmemcg_id < 0);
3624
3625 parent = parent_mem_cgroup(memcg);
3626 if (!parent)
3627 parent = root_mem_cgroup;
3628
3629 /*
3630 * Change kmemcg_id of this cgroup and all its descendants to the
3631 * parent's id, and then move all entries from this cgroup's list_lrus
3632 * to ones of the parent. After we have finished, all list_lrus
3633 * corresponding to this cgroup are guaranteed to remain empty. The
3634 * ordering is imposed by list_lru_node->lock taken by
3635 * memcg_drain_all_list_lrus().
3636 */
3637 css_for_each_descendant_pre(css, &memcg->css) {
3638 child = mem_cgroup_from_css(css);
3639 BUG_ON(child->kmemcg_id != kmemcg_id);
3640 child->kmemcg_id = parent->kmemcg_id;
3641 if (!memcg->use_hierarchy)
3642 break;
3643 }
3644 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3645
3646 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003647}
3648
Li Zefan10d5ebf2013-07-08 16:00:33 -07003649static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003650{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003651 if (memcg->kmem_acct_activated) {
3652 memcg_destroy_kmem_caches(memcg);
3653 static_key_slow_dec(&memcg_kmem_enabled_key);
3654 WARN_ON(page_counter_read(&memcg->kmem));
3655 }
Glauber Costa1d62e432012-04-09 19:36:33 -03003656 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003657}
Glauber Costae5671df2011-12-11 21:47:01 +00003658#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003659static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003660{
3661 return 0;
3662}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003663
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003664static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3665{
3666}
3667
Li Zefan10d5ebf2013-07-08 16:00:33 -07003668static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3669{
3670}
Glauber Costae5671df2011-12-11 21:47:01 +00003671#endif
3672
Tejun Heo52ebea72015-05-22 17:13:37 -04003673#ifdef CONFIG_CGROUP_WRITEBACK
3674
3675struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3676{
3677 return &memcg->cgwb_list;
3678}
3679
Tejun Heo841710a2015-05-22 18:23:33 -04003680static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3681{
3682 return wb_domain_init(&memcg->cgwb_domain, gfp);
3683}
3684
3685static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3686{
3687 wb_domain_exit(&memcg->cgwb_domain);
3688}
3689
Tejun Heo2529bb32015-05-22 18:23:34 -04003690static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3691{
3692 wb_domain_size_changed(&memcg->cgwb_domain);
3693}
3694
Tejun Heo841710a2015-05-22 18:23:33 -04003695struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3696{
3697 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3698
3699 if (!memcg->css.parent)
3700 return NULL;
3701
3702 return &memcg->cgwb_domain;
3703}
3704
Tejun Heoc2aa7232015-05-22 18:23:35 -04003705/**
3706 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3707 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003708 * @pfilepages: out parameter for number of file pages
3709 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003710 * @pdirty: out parameter for number of dirty pages
3711 * @pwriteback: out parameter for number of pages under writeback
3712 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003713 * Determine the numbers of file, headroom, dirty, and writeback pages in
3714 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3715 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003716 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003717 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3718 * headroom is calculated as the lowest headroom of itself and the
3719 * ancestors. Note that this doesn't consider the actual amount of
3720 * available memory in the system. The caller should further cap
3721 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003722 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003723void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3724 unsigned long *pheadroom, unsigned long *pdirty,
3725 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003726{
3727 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3728 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003729
3730 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3731
3732 /* this should eventually include NR_UNSTABLE_NFS */
3733 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003734 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3735 (1 << LRU_ACTIVE_FILE));
3736 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003737
Tejun Heoc2aa7232015-05-22 18:23:35 -04003738 while ((parent = parent_mem_cgroup(memcg))) {
3739 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3740 unsigned long used = page_counter_read(&memcg->memory);
3741
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003742 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003743 memcg = parent;
3744 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003745}
3746
Tejun Heo841710a2015-05-22 18:23:33 -04003747#else /* CONFIG_CGROUP_WRITEBACK */
3748
3749static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3750{
3751 return 0;
3752}
3753
3754static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3755{
3756}
3757
Tejun Heo2529bb32015-05-22 18:23:34 -04003758static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3759{
3760}
3761
Tejun Heo52ebea72015-05-22 17:13:37 -04003762#endif /* CONFIG_CGROUP_WRITEBACK */
3763
Tejun Heo79bd9812013-11-22 18:20:42 -05003764/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003765 * DO NOT USE IN NEW FILES.
3766 *
3767 * "cgroup.event_control" implementation.
3768 *
3769 * This is way over-engineered. It tries to support fully configurable
3770 * events for each user. Such level of flexibility is completely
3771 * unnecessary especially in the light of the planned unified hierarchy.
3772 *
3773 * Please deprecate this and replace with something simpler if at all
3774 * possible.
3775 */
3776
3777/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003778 * Unregister event and free resources.
3779 *
3780 * Gets called from workqueue.
3781 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003782static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003783{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003784 struct mem_cgroup_event *event =
3785 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003786 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003787
3788 remove_wait_queue(event->wqh, &event->wait);
3789
Tejun Heo59b6f872013-11-22 18:20:43 -05003790 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003791
3792 /* Notify userspace the event is going away. */
3793 eventfd_signal(event->eventfd, 1);
3794
3795 eventfd_ctx_put(event->eventfd);
3796 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003797 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003798}
3799
3800/*
3801 * Gets called on POLLHUP on eventfd when user closes it.
3802 *
3803 * Called with wqh->lock held and interrupts disabled.
3804 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003805static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3806 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003807{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003808 struct mem_cgroup_event *event =
3809 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003810 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003811 unsigned long flags = (unsigned long)key;
3812
3813 if (flags & POLLHUP) {
3814 /*
3815 * If the event has been detached at cgroup removal, we
3816 * can simply return knowing the other side will cleanup
3817 * for us.
3818 *
3819 * We can't race against event freeing since the other
3820 * side will require wqh->lock via remove_wait_queue(),
3821 * which we hold.
3822 */
Tejun Heofba94802013-11-22 18:20:43 -05003823 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003824 if (!list_empty(&event->list)) {
3825 list_del_init(&event->list);
3826 /*
3827 * We are in atomic context, but cgroup_event_remove()
3828 * may sleep, so we have to call it in workqueue.
3829 */
3830 schedule_work(&event->remove);
3831 }
Tejun Heofba94802013-11-22 18:20:43 -05003832 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003833 }
3834
3835 return 0;
3836}
3837
Tejun Heo3bc942f2013-11-22 18:20:44 -05003838static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003839 wait_queue_head_t *wqh, poll_table *pt)
3840{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003841 struct mem_cgroup_event *event =
3842 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003843
3844 event->wqh = wqh;
3845 add_wait_queue(wqh, &event->wait);
3846}
3847
3848/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003849 * DO NOT USE IN NEW FILES.
3850 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003851 * Parse input and register new cgroup event handler.
3852 *
3853 * Input must be in format '<event_fd> <control_fd> <args>'.
3854 * Interpretation of args is defined by control file implementation.
3855 */
Tejun Heo451af502014-05-13 12:16:21 -04003856static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3857 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003858{
Tejun Heo451af502014-05-13 12:16:21 -04003859 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003860 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003861 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003862 struct cgroup_subsys_state *cfile_css;
3863 unsigned int efd, cfd;
3864 struct fd efile;
3865 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003866 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003867 char *endp;
3868 int ret;
3869
Tejun Heo451af502014-05-13 12:16:21 -04003870 buf = strstrip(buf);
3871
3872 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003873 if (*endp != ' ')
3874 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003875 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003876
Tejun Heo451af502014-05-13 12:16:21 -04003877 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003878 if ((*endp != ' ') && (*endp != '\0'))
3879 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003880 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003881
3882 event = kzalloc(sizeof(*event), GFP_KERNEL);
3883 if (!event)
3884 return -ENOMEM;
3885
Tejun Heo59b6f872013-11-22 18:20:43 -05003886 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003887 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003888 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3889 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3890 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003891
3892 efile = fdget(efd);
3893 if (!efile.file) {
3894 ret = -EBADF;
3895 goto out_kfree;
3896 }
3897
3898 event->eventfd = eventfd_ctx_fileget(efile.file);
3899 if (IS_ERR(event->eventfd)) {
3900 ret = PTR_ERR(event->eventfd);
3901 goto out_put_efile;
3902 }
3903
3904 cfile = fdget(cfd);
3905 if (!cfile.file) {
3906 ret = -EBADF;
3907 goto out_put_eventfd;
3908 }
3909
3910 /* the process need read permission on control file */
3911 /* AV: shouldn't we check that it's been opened for read instead? */
3912 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3913 if (ret < 0)
3914 goto out_put_cfile;
3915
Tejun Heo79bd9812013-11-22 18:20:42 -05003916 /*
Tejun Heofba94802013-11-22 18:20:43 -05003917 * Determine the event callbacks and set them in @event. This used
3918 * to be done via struct cftype but cgroup core no longer knows
3919 * about these events. The following is crude but the whole thing
3920 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003921 *
3922 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003923 */
Al Virob5830432014-10-31 01:22:04 -04003924 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003925
3926 if (!strcmp(name, "memory.usage_in_bytes")) {
3927 event->register_event = mem_cgroup_usage_register_event;
3928 event->unregister_event = mem_cgroup_usage_unregister_event;
3929 } else if (!strcmp(name, "memory.oom_control")) {
3930 event->register_event = mem_cgroup_oom_register_event;
3931 event->unregister_event = mem_cgroup_oom_unregister_event;
3932 } else if (!strcmp(name, "memory.pressure_level")) {
3933 event->register_event = vmpressure_register_event;
3934 event->unregister_event = vmpressure_unregister_event;
3935 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003936 event->register_event = memsw_cgroup_usage_register_event;
3937 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003938 } else {
3939 ret = -EINVAL;
3940 goto out_put_cfile;
3941 }
3942
3943 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003944 * Verify @cfile should belong to @css. Also, remaining events are
3945 * automatically removed on cgroup destruction but the removal is
3946 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003947 */
Al Virob5830432014-10-31 01:22:04 -04003948 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003949 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003950 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003951 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003952 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003953 if (cfile_css != css) {
3954 css_put(cfile_css);
3955 goto out_put_cfile;
3956 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003957
Tejun Heo451af502014-05-13 12:16:21 -04003958 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003959 if (ret)
3960 goto out_put_css;
3961
3962 efile.file->f_op->poll(efile.file, &event->pt);
3963
Tejun Heofba94802013-11-22 18:20:43 -05003964 spin_lock(&memcg->event_list_lock);
3965 list_add(&event->list, &memcg->event_list);
3966 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003967
3968 fdput(cfile);
3969 fdput(efile);
3970
Tejun Heo451af502014-05-13 12:16:21 -04003971 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003972
3973out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003974 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003975out_put_cfile:
3976 fdput(cfile);
3977out_put_eventfd:
3978 eventfd_ctx_put(event->eventfd);
3979out_put_efile:
3980 fdput(efile);
3981out_kfree:
3982 kfree(event);
3983
3984 return ret;
3985}
3986
Johannes Weiner241994e2015-02-11 15:26:06 -08003987static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003988 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003989 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003990 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003991 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003992 },
3993 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003994 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003995 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003996 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003997 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003998 },
3999 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004000 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004001 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004002 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004003 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004004 },
4005 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004006 .name = "soft_limit_in_bytes",
4007 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004008 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004009 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004010 },
4011 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004012 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004013 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004014 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004015 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004016 },
Balbir Singh8697d332008-02-07 00:13:59 -08004017 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004018 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004019 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004020 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004021 {
4022 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004023 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004024 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004025 {
4026 .name = "use_hierarchy",
4027 .write_u64 = mem_cgroup_hierarchy_write,
4028 .read_u64 = mem_cgroup_hierarchy_read,
4029 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004030 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004031 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004032 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04004033 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004034 },
4035 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004036 .name = "swappiness",
4037 .read_u64 = mem_cgroup_swappiness_read,
4038 .write_u64 = mem_cgroup_swappiness_write,
4039 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004040 {
4041 .name = "move_charge_at_immigrate",
4042 .read_u64 = mem_cgroup_move_charge_read,
4043 .write_u64 = mem_cgroup_move_charge_write,
4044 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004045 {
4046 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004047 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004048 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004049 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4050 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004051 {
4052 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004053 },
Ying Han406eb0c2011-05-26 16:25:37 -07004054#ifdef CONFIG_NUMA
4055 {
4056 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004057 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004058 },
4059#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004060#ifdef CONFIG_MEMCG_KMEM
4061 {
4062 .name = "kmem.limit_in_bytes",
4063 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004064 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004065 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004066 },
4067 {
4068 .name = "kmem.usage_in_bytes",
4069 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004070 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004071 },
4072 {
4073 .name = "kmem.failcnt",
4074 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004075 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004076 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004077 },
4078 {
4079 .name = "kmem.max_usage_in_bytes",
4080 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004081 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004082 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004083 },
Glauber Costa749c5412012-12-18 14:23:01 -08004084#ifdef CONFIG_SLABINFO
4085 {
4086 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004087 .seq_start = slab_start,
4088 .seq_next = slab_next,
4089 .seq_stop = slab_stop,
4090 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004091 },
4092#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004093#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004094 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004095};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004096
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004097static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004098{
4099 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004100 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004101 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004102 /*
4103 * This routine is called against possible nodes.
4104 * But it's BUG to call kmalloc() against offline node.
4105 *
4106 * TODO: this routine can waste much memory for nodes which will
4107 * never be onlined. It's better to use memory hotplug callback
4108 * function.
4109 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004110 if (!node_state(node, N_NORMAL_MEMORY))
4111 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004112 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004113 if (!pn)
4114 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004115
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004116 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4117 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004118 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004119 mz->usage_in_excess = 0;
4120 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004121 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004122 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004123 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004124 return 0;
4125}
4126
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004127static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004128{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004129 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004130}
4131
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004132static struct mem_cgroup *mem_cgroup_alloc(void)
4133{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004134 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004135 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004136
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004137 size = sizeof(struct mem_cgroup);
4138 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004139
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004140 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004141 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004142 return NULL;
4143
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004144 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4145 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004146 goto out_free;
Tejun Heo841710a2015-05-22 18:23:33 -04004147
4148 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4149 goto out_free_stat;
4150
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004151 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004152
Tejun Heo841710a2015-05-22 18:23:33 -04004153out_free_stat:
4154 free_percpu(memcg->stat);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004155out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004156 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004157 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004158}
4159
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004160/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004161 * At destroying mem_cgroup, references from swap_cgroup can remain.
4162 * (scanning all at force_empty is too costly...)
4163 *
4164 * Instead of clearing all references at force_empty, we remember
4165 * the number of reference from swap_cgroup and free mem_cgroup when
4166 * it goes down to 0.
4167 *
4168 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004169 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004170
4171static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004172{
Glauber Costac8b2a362012-12-18 14:22:13 -08004173 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004174
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004175 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004176
4177 for_each_node(node)
4178 free_mem_cgroup_per_zone_info(memcg, node);
4179
4180 free_percpu(memcg->stat);
Tejun Heo841710a2015-05-22 18:23:33 -04004181 memcg_wb_domain_exit(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004182 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004183}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004184
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004185/*
4186 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4187 */
Glauber Costae1aab162011-12-11 21:47:03 +00004188struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004189{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004190 if (!memcg->memory.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004191 return NULL;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004192 return mem_cgroup_from_counter(memcg->memory.parent, memory);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004193}
Glauber Costae1aab162011-12-11 21:47:03 +00004194EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004195
Li Zefan0eb253e2009-01-15 13:51:25 -08004196static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004197mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004198{
Glauber Costad142e3e2013-02-22 16:34:52 -08004199 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004200 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004201 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004202
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004203 memcg = mem_cgroup_alloc();
4204 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004205 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004206
Bob Liu3ed28fa2012-01-12 17:19:04 -08004207 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004208 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004209 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004210
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004211 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004212 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004213 root_mem_cgroup = memcg;
Tejun Heo56161632015-05-22 17:13:20 -04004214 mem_cgroup_root_css = &memcg->css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004215 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994e2015-02-11 15:26:06 -08004216 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004217 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004218 page_counter_init(&memcg->memsw, NULL);
4219 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004220 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004221
Glauber Costad142e3e2013-02-22 16:34:52 -08004222 memcg->last_scanned_node = MAX_NUMNODES;
4223 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004224 memcg->move_charge_at_immigrate = 0;
4225 mutex_init(&memcg->thresholds_lock);
4226 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004227 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004228 INIT_LIST_HEAD(&memcg->event_list);
4229 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004230#ifdef CONFIG_MEMCG_KMEM
4231 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004232#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004233#ifdef CONFIG_CGROUP_WRITEBACK
4234 INIT_LIST_HEAD(&memcg->cgwb_list);
4235#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004236 return &memcg->css;
4237
4238free_out:
4239 __mem_cgroup_free(memcg);
4240 return ERR_PTR(error);
4241}
4242
4243static int
Tejun Heoeb954192013-08-08 20:11:23 -04004244mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004245{
Tejun Heoeb954192013-08-08 20:11:23 -04004246 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004247 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004248 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004249
Tejun Heo15a4c832014-05-04 15:09:14 -04004250 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004251 return -ENOSPC;
4252
Tejun Heo63876982013-08-08 20:11:23 -04004253 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004254 return 0;
4255
Glauber Costa09998212013-02-22 16:34:55 -08004256 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004257
4258 memcg->use_hierarchy = parent->use_hierarchy;
4259 memcg->oom_kill_disable = parent->oom_kill_disable;
4260 memcg->swappiness = mem_cgroup_swappiness(parent);
4261
4262 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004263 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994e2015-02-11 15:26:06 -08004264 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004265 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004266 page_counter_init(&memcg->memsw, &parent->memsw);
4267 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004268
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004269 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004270 * No need to take a reference to the parent because cgroup
4271 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004272 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004273 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004274 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994e2015-02-11 15:26:06 -08004275 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004276 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004277 page_counter_init(&memcg->memsw, NULL);
4278 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004279 /*
4280 * Deeper hierachy with use_hierarchy == false doesn't make
4281 * much sense so let cgroup subsystem know about this
4282 * unfortunate state in our controller.
4283 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004284 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004285 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004286 }
Glauber Costa09998212013-02-22 16:34:55 -08004287 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004288
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004289 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4290 if (ret)
4291 return ret;
4292
4293 /*
4294 * Make sure the memcg is initialized: mem_cgroup_iter()
4295 * orders reading memcg->initialized against its callers
4296 * reading the memcg members.
4297 */
4298 smp_store_release(&memcg->initialized, 1);
4299
4300 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004301}
4302
Tejun Heoeb954192013-08-08 20:11:23 -04004303static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004304{
Tejun Heoeb954192013-08-08 20:11:23 -04004305 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004306 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004307
4308 /*
4309 * Unregister events and notify userspace.
4310 * Notify userspace about cgroup removing only after rmdir of cgroup
4311 * directory to avoid race between userspace and kernelspace.
4312 */
Tejun Heofba94802013-11-22 18:20:43 -05004313 spin_lock(&memcg->event_list_lock);
4314 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004315 list_del_init(&event->list);
4316 schedule_work(&event->remove);
4317 }
Tejun Heofba94802013-11-22 18:20:43 -05004318 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004319
Michal Hocko33cb8762013-07-31 13:53:51 -07004320 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004321
4322 memcg_deactivate_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004323
4324 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004325}
4326
Tejun Heoeb954192013-08-08 20:11:23 -04004327static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004328{
Tejun Heoeb954192013-08-08 20:11:23 -04004329 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004330
Li Zefan10d5ebf2013-07-08 16:00:33 -07004331 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07004332 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004333}
4334
Tejun Heo1ced9532014-07-08 18:02:57 -04004335/**
4336 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4337 * @css: the target css
4338 *
4339 * Reset the states of the mem_cgroup associated with @css. This is
4340 * invoked when the userland requests disabling on the default hierarchy
4341 * but the memcg is pinned through dependency. The memcg should stop
4342 * applying policies and should revert to the vanilla state as it may be
4343 * made visible again.
4344 *
4345 * The current implementation only resets the essential configurations.
4346 * This needs to be expanded to cover all the visible parts.
4347 */
4348static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4349{
4350 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4351
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004352 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4353 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4354 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004355 memcg->low = 0;
4356 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004357 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004358 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004359}
4360
Daisuke Nishimura02491442010-03-10 15:22:17 -08004361#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004362/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004363static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004364{
Johannes Weiner05b84302014-08-06 16:05:59 -07004365 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004366
Mel Gormand0164ad2015-11-06 16:28:21 -08004367 /* Try a single bulk charge without reclaim first, kswapd may wake */
4368 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004369 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004370 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004371 return ret;
4372 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004373
4374 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004375 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004376 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004377 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004378 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004379 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004380 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004381 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004382 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004383}
4384
4385/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004386 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004387 * @vma: the vma the pte to be checked belongs
4388 * @addr: the address corresponding to the pte to be checked
4389 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004390 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004391 *
4392 * Returns
4393 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4394 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4395 * move charge. if @target is not NULL, the page is stored in target->page
4396 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004397 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4398 * target for charge migration. if @target is not NULL, the entry is stored
4399 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004400 *
4401 * Called with pte lock held.
4402 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004403union mc_target {
4404 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004405 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004406};
4407
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004408enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004409 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004410 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004411 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004412};
4413
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004414static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4415 unsigned long addr, pte_t ptent)
4416{
4417 struct page *page = vm_normal_page(vma, addr, ptent);
4418
4419 if (!page || !page_mapped(page))
4420 return NULL;
4421 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004422 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004423 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004424 } else {
4425 if (!(mc.flags & MOVE_FILE))
4426 return NULL;
4427 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004428 if (!get_page_unless_zero(page))
4429 return NULL;
4430
4431 return page;
4432}
4433
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004434#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004435static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4436 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4437{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004438 struct page *page = NULL;
4439 swp_entry_t ent = pte_to_swp_entry(ptent);
4440
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004441 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004442 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004443 /*
4444 * Because lookup_swap_cache() updates some statistics counter,
4445 * we call find_get_page() with swapper_space directly.
4446 */
Shaohua Li33806f02013-02-22 16:34:37 -08004447 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004448 if (do_swap_account)
4449 entry->val = ent.val;
4450
4451 return page;
4452}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004453#else
4454static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4455 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4456{
4457 return NULL;
4458}
4459#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004460
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004461static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4462 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4463{
4464 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004465 struct address_space *mapping;
4466 pgoff_t pgoff;
4467
4468 if (!vma->vm_file) /* anonymous vma */
4469 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004470 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004471 return NULL;
4472
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004473 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004474 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004475
4476 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004477#ifdef CONFIG_SWAP
4478 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004479 if (shmem_mapping(mapping)) {
4480 page = find_get_entry(mapping, pgoff);
4481 if (radix_tree_exceptional_entry(page)) {
4482 swp_entry_t swp = radix_to_swp_entry(page);
4483 if (do_swap_account)
4484 *entry = swp;
4485 page = find_get_page(swap_address_space(swp), swp.val);
4486 }
4487 } else
4488 page = find_get_page(mapping, pgoff);
4489#else
4490 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004491#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004492 return page;
4493}
4494
Chen Gangb1b0dea2015-04-14 15:47:35 -07004495/**
4496 * mem_cgroup_move_account - move account of the page
4497 * @page: the page
4498 * @nr_pages: number of regular pages (>1 for huge pages)
4499 * @from: mem_cgroup which the page is moved from.
4500 * @to: mem_cgroup which the page is moved to. @from != @to.
4501 *
4502 * The caller must confirm following.
4503 * - page is not on LRU (isolate_page() is useful.)
4504 * - compound_lock is held when nr_pages > 1
4505 *
4506 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4507 * from old cgroup.
4508 */
4509static int mem_cgroup_move_account(struct page *page,
4510 unsigned int nr_pages,
4511 struct mem_cgroup *from,
4512 struct mem_cgroup *to)
4513{
4514 unsigned long flags;
4515 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004516 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004517
4518 VM_BUG_ON(from == to);
4519 VM_BUG_ON_PAGE(PageLRU(page), page);
4520 /*
4521 * The page is isolated from LRU. So, collapse function
4522 * will not handle this page. But page splitting can happen.
4523 * Do this check under compound_page_lock(). The caller should
4524 * hold it.
4525 */
4526 ret = -EBUSY;
4527 if (nr_pages > 1 && !PageTransHuge(page))
4528 goto out;
4529
4530 /*
Hugh Dickins45637ba2015-11-05 18:49:40 -08004531 * Prevent mem_cgroup_replace_page() from looking at
4532 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004533 */
4534 if (!trylock_page(page))
4535 goto out;
4536
4537 ret = -EINVAL;
4538 if (page->mem_cgroup != from)
4539 goto out_unlock;
4540
Greg Thelenc4843a72015-05-22 17:13:16 -04004541 anon = PageAnon(page);
4542
Chen Gangb1b0dea2015-04-14 15:47:35 -07004543 spin_lock_irqsave(&from->move_lock, flags);
4544
Greg Thelenc4843a72015-05-22 17:13:16 -04004545 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004546 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4547 nr_pages);
4548 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4549 nr_pages);
4550 }
4551
Greg Thelenc4843a72015-05-22 17:13:16 -04004552 /*
4553 * move_lock grabbed above and caller set from->moving_account, so
4554 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4555 * So mapping should be stable for dirty pages.
4556 */
4557 if (!anon && PageDirty(page)) {
4558 struct address_space *mapping = page_mapping(page);
4559
4560 if (mapping_cap_account_dirty(mapping)) {
4561 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4562 nr_pages);
4563 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4564 nr_pages);
4565 }
4566 }
4567
Chen Gangb1b0dea2015-04-14 15:47:35 -07004568 if (PageWriteback(page)) {
4569 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4570 nr_pages);
4571 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4572 nr_pages);
4573 }
4574
4575 /*
4576 * It is safe to change page->mem_cgroup here because the page
4577 * is referenced, charged, and isolated - we can't race with
4578 * uncharging, charging, migration, or LRU putback.
4579 */
4580
4581 /* caller should have done css_get */
4582 page->mem_cgroup = to;
4583 spin_unlock_irqrestore(&from->move_lock, flags);
4584
4585 ret = 0;
4586
4587 local_irq_disable();
4588 mem_cgroup_charge_statistics(to, page, nr_pages);
4589 memcg_check_events(to, page);
4590 mem_cgroup_charge_statistics(from, page, -nr_pages);
4591 memcg_check_events(from, page);
4592 local_irq_enable();
4593out_unlock:
4594 unlock_page(page);
4595out:
4596 return ret;
4597}
4598
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004599static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004600 unsigned long addr, pte_t ptent, union mc_target *target)
4601{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004602 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004603 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004604 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004605
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004606 if (pte_present(ptent))
4607 page = mc_handle_present_pte(vma, addr, ptent);
4608 else if (is_swap_pte(ptent))
4609 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004610 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004611 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004612
4613 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004614 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004615 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004616 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004617 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004618 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004619 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004620 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004621 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004622 ret = MC_TARGET_PAGE;
4623 if (target)
4624 target->page = page;
4625 }
4626 if (!ret || !target)
4627 put_page(page);
4628 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004629 /* There is a swap entry and a page doesn't exist or isn't charged */
4630 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004631 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004632 ret = MC_TARGET_SWAP;
4633 if (target)
4634 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004635 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004636 return ret;
4637}
4638
Naoya Horiguchi12724852012-03-21 16:34:28 -07004639#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4640/*
4641 * We don't consider swapping or file mapped pages because THP does not
4642 * support them for now.
4643 * Caller should make sure that pmd_trans_huge(pmd) is true.
4644 */
4645static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4646 unsigned long addr, pmd_t pmd, union mc_target *target)
4647{
4648 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004649 enum mc_target_type ret = MC_TARGET_NONE;
4650
4651 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004652 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004653 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004654 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004655 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004656 ret = MC_TARGET_PAGE;
4657 if (target) {
4658 get_page(page);
4659 target->page = page;
4660 }
4661 }
4662 return ret;
4663}
4664#else
4665static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4666 unsigned long addr, pmd_t pmd, union mc_target *target)
4667{
4668 return MC_TARGET_NONE;
4669}
4670#endif
4671
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004672static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4673 unsigned long addr, unsigned long end,
4674 struct mm_walk *walk)
4675{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004676 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004677 pte_t *pte;
4678 spinlock_t *ptl;
4679
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004680 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004681 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4682 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004683 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004684 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004685 }
Dave Hansen03319322011-03-22 16:32:56 -07004686
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004687 if (pmd_trans_unstable(pmd))
4688 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004689 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4690 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004691 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004692 mc.precharge++; /* increment precharge temporarily */
4693 pte_unmap_unlock(pte - 1, ptl);
4694 cond_resched();
4695
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004696 return 0;
4697}
4698
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004699static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4700{
4701 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004702
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004703 struct mm_walk mem_cgroup_count_precharge_walk = {
4704 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4705 .mm = mm,
4706 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004707 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004708 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004709 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004710
4711 precharge = mc.precharge;
4712 mc.precharge = 0;
4713
4714 return precharge;
4715}
4716
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004717static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4718{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004719 unsigned long precharge = mem_cgroup_count_precharge(mm);
4720
4721 VM_BUG_ON(mc.moving_task);
4722 mc.moving_task = current;
4723 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004724}
4725
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004726/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4727static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004728{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004729 struct mem_cgroup *from = mc.from;
4730 struct mem_cgroup *to = mc.to;
4731
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004732 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004733 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004734 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004735 mc.precharge = 0;
4736 }
4737 /*
4738 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4739 * we must uncharge here.
4740 */
4741 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004742 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004743 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004744 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004745 /* we must fixup refcnts and charges */
4746 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004747 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004748 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004749 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004750
Johannes Weiner05b84302014-08-06 16:05:59 -07004751 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004752 * we charged both to->memory and to->memsw, so we
4753 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004754 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004755 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004756 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004757
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004758 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004759
Li Zefan40503772013-07-08 16:00:34 -07004760 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004761 mc.moved_swap = 0;
4762 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004763 memcg_oom_recover(from);
4764 memcg_oom_recover(to);
4765 wake_up_all(&mc.waitq);
4766}
4767
4768static void mem_cgroup_clear_mc(void)
4769{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004770 /*
4771 * we must clear moving_task before waking up waiters at the end of
4772 * task migration.
4773 */
4774 mc.moving_task = NULL;
4775 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004776 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004777 mc.from = NULL;
4778 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004779 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004780}
4781
Tejun Heoeb954192013-08-08 20:11:23 -04004782static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004783 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004784{
Tejun Heoeb954192013-08-08 20:11:23 -04004785 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo9f2115f2015-09-08 15:01:10 -07004786 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004787 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004788 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004789 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004790 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004791
Glauber Costaee5e8472013-02-22 16:34:50 -08004792 /*
4793 * We are now commited to this value whatever it is. Changes in this
4794 * tunable will only affect upcoming migrations, not the current one.
4795 * So we need to save it, and keep it going.
4796 */
Jason Low4db0c3c2015-04-15 16:14:08 -07004797 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
Tejun Heo9f2115f2015-09-08 15:01:10 -07004798 if (!move_flags)
4799 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004800
Tejun Heo4530edd2015-09-11 15:00:19 -04004801 /*
4802 * Multi-process migrations only happen on the default hierarchy
4803 * where charge immigration is not used. Perform charge
4804 * immigration if @tset contains a leader and whine if there are
4805 * multiple.
4806 */
4807 p = NULL;
4808 cgroup_taskset_for_each_leader(leader, tset) {
4809 WARN_ON_ONCE(p);
4810 p = leader;
4811 }
4812 if (!p)
4813 return 0;
4814
Tejun Heo9f2115f2015-09-08 15:01:10 -07004815 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004816
Tejun Heo9f2115f2015-09-08 15:01:10 -07004817 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004818
Tejun Heo9f2115f2015-09-08 15:01:10 -07004819 mm = get_task_mm(p);
4820 if (!mm)
4821 return 0;
4822 /* We move charges only when we move a owner of the mm */
4823 if (mm->owner == p) {
4824 VM_BUG_ON(mc.from);
4825 VM_BUG_ON(mc.to);
4826 VM_BUG_ON(mc.precharge);
4827 VM_BUG_ON(mc.moved_charge);
4828 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004829
Tejun Heo9f2115f2015-09-08 15:01:10 -07004830 spin_lock(&mc.lock);
4831 mc.from = from;
4832 mc.to = memcg;
4833 mc.flags = move_flags;
4834 spin_unlock(&mc.lock);
4835 /* We set mc.moving_task later */
4836
4837 ret = mem_cgroup_precharge_mc(mm);
4838 if (ret)
4839 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004840 }
Tejun Heo9f2115f2015-09-08 15:01:10 -07004841 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004842 return ret;
4843}
4844
Tejun Heoeb954192013-08-08 20:11:23 -04004845static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004846 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004847{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004848 if (mc.to)
4849 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004850}
4851
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004852static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4853 unsigned long addr, unsigned long end,
4854 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004855{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004856 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004857 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004858 pte_t *pte;
4859 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004860 enum mc_target_type target_type;
4861 union mc_target target;
4862 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004863
Naoya Horiguchi12724852012-03-21 16:34:28 -07004864 /*
4865 * We don't take compound_lock() here but no race with splitting thp
4866 * happens because:
4867 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
4868 * under splitting, which means there's no concurrent thp split,
4869 * - if another thread runs into split_huge_page() just after we
4870 * entered this if-block, the thread must wait for page table lock
4871 * to be unlocked in __split_huge_page_splitting(), where the main
4872 * part of thp split is not executed yet.
4873 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004874 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004875 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004876 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004877 return 0;
4878 }
4879 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4880 if (target_type == MC_TARGET_PAGE) {
4881 page = target.page;
4882 if (!isolate_lru_page(page)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004883 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
Johannes Weiner1306a852014-12-10 15:44:52 -08004884 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004885 mc.precharge -= HPAGE_PMD_NR;
4886 mc.moved_charge += HPAGE_PMD_NR;
4887 }
4888 putback_lru_page(page);
4889 }
4890 put_page(page);
4891 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004892 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004893 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004894 }
4895
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004896 if (pmd_trans_unstable(pmd))
4897 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004898retry:
4899 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4900 for (; addr != end; addr += PAGE_SIZE) {
4901 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004902 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004903
4904 if (!mc.precharge)
4905 break;
4906
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004907 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004908 case MC_TARGET_PAGE:
4909 page = target.page;
4910 if (isolate_lru_page(page))
4911 goto put;
Johannes Weiner1306a852014-12-10 15:44:52 -08004912 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004913 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004914 /* we uncharge from mc.from later. */
4915 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004916 }
4917 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004918put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004919 put_page(page);
4920 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004921 case MC_TARGET_SWAP:
4922 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004923 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004924 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004925 /* we fixup refcnts and charges later. */
4926 mc.moved_swap++;
4927 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004928 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004929 default:
4930 break;
4931 }
4932 }
4933 pte_unmap_unlock(pte - 1, ptl);
4934 cond_resched();
4935
4936 if (addr != end) {
4937 /*
4938 * We have consumed all precharges we got in can_attach().
4939 * We try charge one by one, but don't do any additional
4940 * charges to mc.to if we have failed in charge once in attach()
4941 * phase.
4942 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004943 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004944 if (!ret)
4945 goto retry;
4946 }
4947
4948 return ret;
4949}
4950
4951static void mem_cgroup_move_charge(struct mm_struct *mm)
4952{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004953 struct mm_walk mem_cgroup_move_charge_walk = {
4954 .pmd_entry = mem_cgroup_move_charge_pte_range,
4955 .mm = mm,
4956 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004957
4958 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004959 /*
4960 * Signal mem_cgroup_begin_page_stat() to take the memcg's
4961 * move_lock while we're moving its pages to another memcg.
4962 * Then wait for already started RCU-only updates to finish.
4963 */
4964 atomic_inc(&mc.from->moving_account);
4965 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004966retry:
4967 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4968 /*
4969 * Someone who are holding the mmap_sem might be waiting in
4970 * waitq. So we cancel all extra charges, wake up all waiters,
4971 * and retry. Because we cancel precharges, we might not be able
4972 * to move enough charges, but moving charge is a best-effort
4973 * feature anyway, so it wouldn't be a big problem.
4974 */
4975 __mem_cgroup_clear_mc();
4976 cond_resched();
4977 goto retry;
4978 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004979 /*
4980 * When we have consumed all precharges and failed in doing
4981 * additional charge, the page walk just aborts.
4982 */
4983 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004984 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004985 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004986}
4987
Tejun Heoeb954192013-08-08 20:11:23 -04004988static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004989 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08004990{
Tejun Heo2f7ee562011-12-12 18:12:21 -08004991 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004992 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004993
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004994 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004995 if (mc.to)
4996 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004997 mmput(mm);
4998 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004999 if (mc.to)
5000 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005001}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005002#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04005003static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005004 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005005{
5006 return 0;
5007}
Tejun Heoeb954192013-08-08 20:11:23 -04005008static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005009 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005010{
5011}
Tejun Heoeb954192013-08-08 20:11:23 -04005012static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005013 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005014{
5015}
5016#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005017
Tejun Heof00baae2013-04-15 13:41:15 -07005018/*
5019 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005020 * to verify whether we're attached to the default hierarchy on each mount
5021 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005022 */
Tejun Heoeb954192013-08-08 20:11:23 -04005023static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005024{
5025 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005026 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005027 * guarantees that @root doesn't have any children, so turning it
5028 * on for the root memcg is enough.
5029 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005030 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005031 root_mem_cgroup->use_hierarchy = true;
5032 else
5033 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005034}
5035
Johannes Weiner241994e2015-02-11 15:26:06 -08005036static u64 memory_current_read(struct cgroup_subsys_state *css,
5037 struct cftype *cft)
5038{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005039 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5040
5041 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005042}
5043
5044static int memory_low_show(struct seq_file *m, void *v)
5045{
5046 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005047 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005048
5049 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005050 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005051 else
5052 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5053
5054 return 0;
5055}
5056
5057static ssize_t memory_low_write(struct kernfs_open_file *of,
5058 char *buf, size_t nbytes, loff_t off)
5059{
5060 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5061 unsigned long low;
5062 int err;
5063
5064 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005065 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005066 if (err)
5067 return err;
5068
5069 memcg->low = low;
5070
5071 return nbytes;
5072}
5073
5074static int memory_high_show(struct seq_file *m, void *v)
5075{
5076 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005077 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005078
5079 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005080 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005081 else
5082 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5083
5084 return 0;
5085}
5086
5087static ssize_t memory_high_write(struct kernfs_open_file *of,
5088 char *buf, size_t nbytes, loff_t off)
5089{
5090 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5091 unsigned long high;
5092 int err;
5093
5094 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005095 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005096 if (err)
5097 return err;
5098
5099 memcg->high = high;
5100
Tejun Heo2529bb32015-05-22 18:23:34 -04005101 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005102 return nbytes;
5103}
5104
5105static int memory_max_show(struct seq_file *m, void *v)
5106{
5107 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005108 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005109
5110 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005111 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005112 else
5113 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5114
5115 return 0;
5116}
5117
5118static ssize_t memory_max_write(struct kernfs_open_file *of,
5119 char *buf, size_t nbytes, loff_t off)
5120{
5121 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5122 unsigned long max;
5123 int err;
5124
5125 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005126 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005127 if (err)
5128 return err;
5129
5130 err = mem_cgroup_resize_limit(memcg, max);
5131 if (err)
5132 return err;
5133
Tejun Heo2529bb32015-05-22 18:23:34 -04005134 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005135 return nbytes;
5136}
5137
5138static int memory_events_show(struct seq_file *m, void *v)
5139{
5140 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5141
5142 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5143 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5144 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5145 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5146
5147 return 0;
5148}
5149
5150static struct cftype memory_files[] = {
5151 {
5152 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005153 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005154 .read_u64 = memory_current_read,
5155 },
5156 {
5157 .name = "low",
5158 .flags = CFTYPE_NOT_ON_ROOT,
5159 .seq_show = memory_low_show,
5160 .write = memory_low_write,
5161 },
5162 {
5163 .name = "high",
5164 .flags = CFTYPE_NOT_ON_ROOT,
5165 .seq_show = memory_high_show,
5166 .write = memory_high_write,
5167 },
5168 {
5169 .name = "max",
5170 .flags = CFTYPE_NOT_ON_ROOT,
5171 .seq_show = memory_max_show,
5172 .write = memory_max_write,
5173 },
5174 {
5175 .name = "events",
5176 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005177 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005178 .seq_show = memory_events_show,
5179 },
5180 { } /* terminate */
5181};
5182
Tejun Heo073219e2014-02-08 10:36:58 -05005183struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005184 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005185 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005186 .css_offline = mem_cgroup_css_offline,
5187 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005188 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005189 .can_attach = mem_cgroup_can_attach,
5190 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005191 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005192 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005193 .dfl_cftypes = memory_files,
5194 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005195 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005196};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005197
Johannes Weiner241994e2015-02-11 15:26:06 -08005198/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005199 * mem_cgroup_low - check if memory consumption is below the normal range
5200 * @root: the highest ancestor to consider
5201 * @memcg: the memory cgroup to check
5202 *
5203 * Returns %true if memory consumption of @memcg, and that of all
5204 * configurable ancestors up to @root, is below the normal range.
5205 */
5206bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5207{
5208 if (mem_cgroup_disabled())
5209 return false;
5210
5211 /*
5212 * The toplevel group doesn't have a configurable range, so
5213 * it's never low when looked at directly, and it is not
5214 * considered an ancestor when assessing the hierarchy.
5215 */
5216
5217 if (memcg == root_mem_cgroup)
5218 return false;
5219
Michal Hocko4e54ded2015-02-27 15:51:46 -08005220 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005221 return false;
5222
5223 while (memcg != root) {
5224 memcg = parent_mem_cgroup(memcg);
5225
5226 if (memcg == root_mem_cgroup)
5227 break;
5228
Michal Hocko4e54ded2015-02-27 15:51:46 -08005229 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005230 return false;
5231 }
5232 return true;
5233}
5234
Johannes Weiner00501b52014-08-08 14:19:20 -07005235/**
5236 * mem_cgroup_try_charge - try charging a page
5237 * @page: page to charge
5238 * @mm: mm context of the victim
5239 * @gfp_mask: reclaim mode
5240 * @memcgp: charged memcg return
5241 *
5242 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5243 * pages according to @gfp_mask if necessary.
5244 *
5245 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5246 * Otherwise, an error code is returned.
5247 *
5248 * After page->mapping has been set up, the caller must finalize the
5249 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5250 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5251 */
5252int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5253 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5254{
5255 struct mem_cgroup *memcg = NULL;
5256 unsigned int nr_pages = 1;
5257 int ret = 0;
5258
5259 if (mem_cgroup_disabled())
5260 goto out;
5261
5262 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005263 /*
5264 * Every swap fault against a single page tries to charge the
5265 * page, bail as early as possible. shmem_unuse() encounters
5266 * already charged pages, too. The USED bit is protected by
5267 * the page lock, which serializes swap cache removal, which
5268 * in turn serializes uncharging.
5269 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005270 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005271 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005272 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005273
5274 if (do_swap_account) {
5275 swp_entry_t ent = { .val = page_private(page), };
5276 unsigned short id = lookup_swap_cgroup_id(ent);
5277
5278 rcu_read_lock();
5279 memcg = mem_cgroup_from_id(id);
5280 if (memcg && !css_tryget_online(&memcg->css))
5281 memcg = NULL;
5282 rcu_read_unlock();
5283 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005284 }
5285
5286 if (PageTransHuge(page)) {
5287 nr_pages <<= compound_order(page);
5288 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5289 }
5290
Johannes Weiner00501b52014-08-08 14:19:20 -07005291 if (!memcg)
5292 memcg = get_mem_cgroup_from_mm(mm);
5293
5294 ret = try_charge(memcg, gfp_mask, nr_pages);
5295
5296 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005297out:
5298 *memcgp = memcg;
5299 return ret;
5300}
5301
5302/**
5303 * mem_cgroup_commit_charge - commit a page charge
5304 * @page: page to charge
5305 * @memcg: memcg to charge the page to
5306 * @lrucare: page might be on LRU already
5307 *
5308 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5309 * after page->mapping has been set up. This must happen atomically
5310 * as part of the page instantiation, i.e. under the page table lock
5311 * for anonymous pages, under the page lock for page and swap cache.
5312 *
5313 * In addition, the page must not be on the LRU during the commit, to
5314 * prevent racing with task migration. If it might be, use @lrucare.
5315 *
5316 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5317 */
5318void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5319 bool lrucare)
5320{
5321 unsigned int nr_pages = 1;
5322
5323 VM_BUG_ON_PAGE(!page->mapping, page);
5324 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5325
5326 if (mem_cgroup_disabled())
5327 return;
5328 /*
5329 * Swap faults will attempt to charge the same page multiple
5330 * times. But reuse_swap_page() might have removed the page
5331 * from swapcache already, so we can't check PageSwapCache().
5332 */
5333 if (!memcg)
5334 return;
5335
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005336 commit_charge(page, memcg, lrucare);
5337
Johannes Weiner00501b52014-08-08 14:19:20 -07005338 if (PageTransHuge(page)) {
5339 nr_pages <<= compound_order(page);
5340 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5341 }
5342
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005343 local_irq_disable();
5344 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5345 memcg_check_events(memcg, page);
5346 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005347
5348 if (do_swap_account && PageSwapCache(page)) {
5349 swp_entry_t entry = { .val = page_private(page) };
5350 /*
5351 * The swap entry might not get freed for a long time,
5352 * let's not wait for it. The page already received a
5353 * memory+swap charge, drop the swap entry duplicate.
5354 */
5355 mem_cgroup_uncharge_swap(entry);
5356 }
5357}
5358
5359/**
5360 * mem_cgroup_cancel_charge - cancel a page charge
5361 * @page: page to charge
5362 * @memcg: memcg to charge the page to
5363 *
5364 * Cancel a charge transaction started by mem_cgroup_try_charge().
5365 */
5366void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5367{
5368 unsigned int nr_pages = 1;
5369
5370 if (mem_cgroup_disabled())
5371 return;
5372 /*
5373 * Swap faults will attempt to charge the same page multiple
5374 * times. But reuse_swap_page() might have removed the page
5375 * from swapcache already, so we can't check PageSwapCache().
5376 */
5377 if (!memcg)
5378 return;
5379
5380 if (PageTransHuge(page)) {
5381 nr_pages <<= compound_order(page);
5382 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5383 }
5384
5385 cancel_charge(memcg, nr_pages);
5386}
5387
Johannes Weiner747db952014-08-08 14:19:24 -07005388static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005389 unsigned long nr_anon, unsigned long nr_file,
5390 unsigned long nr_huge, struct page *dummy_page)
5391{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005392 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005393 unsigned long flags;
5394
Johannes Weinerce00a962014-09-05 08:43:57 -04005395 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005396 page_counter_uncharge(&memcg->memory, nr_pages);
5397 if (do_swap_account)
5398 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005399 memcg_oom_recover(memcg);
5400 }
Johannes Weiner747db952014-08-08 14:19:24 -07005401
5402 local_irq_save(flags);
5403 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5404 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5405 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5406 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005407 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005408 memcg_check_events(memcg, dummy_page);
5409 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005410
5411 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005412 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005413}
5414
5415static void uncharge_list(struct list_head *page_list)
5416{
5417 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005418 unsigned long nr_anon = 0;
5419 unsigned long nr_file = 0;
5420 unsigned long nr_huge = 0;
5421 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005422 struct list_head *next;
5423 struct page *page;
5424
5425 next = page_list->next;
5426 do {
5427 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005428
5429 page = list_entry(next, struct page, lru);
5430 next = page->lru.next;
5431
5432 VM_BUG_ON_PAGE(PageLRU(page), page);
5433 VM_BUG_ON_PAGE(page_count(page), page);
5434
Johannes Weiner1306a852014-12-10 15:44:52 -08005435 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005436 continue;
5437
5438 /*
5439 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005440 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005441 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005442 */
5443
Johannes Weiner1306a852014-12-10 15:44:52 -08005444 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005445 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005446 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5447 nr_huge, page);
5448 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005449 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005450 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005451 }
5452
5453 if (PageTransHuge(page)) {
5454 nr_pages <<= compound_order(page);
5455 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5456 nr_huge += nr_pages;
5457 }
5458
5459 if (PageAnon(page))
5460 nr_anon += nr_pages;
5461 else
5462 nr_file += nr_pages;
5463
Johannes Weiner1306a852014-12-10 15:44:52 -08005464 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005465
5466 pgpgout++;
5467 } while (next != page_list);
5468
5469 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005470 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5471 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005472}
5473
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005474/**
5475 * mem_cgroup_uncharge - uncharge a page
5476 * @page: page to uncharge
5477 *
5478 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5479 * mem_cgroup_commit_charge().
5480 */
5481void mem_cgroup_uncharge(struct page *page)
5482{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005483 if (mem_cgroup_disabled())
5484 return;
5485
Johannes Weiner747db952014-08-08 14:19:24 -07005486 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005487 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005488 return;
5489
Johannes Weiner747db952014-08-08 14:19:24 -07005490 INIT_LIST_HEAD(&page->lru);
5491 uncharge_list(&page->lru);
5492}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005493
Johannes Weiner747db952014-08-08 14:19:24 -07005494/**
5495 * mem_cgroup_uncharge_list - uncharge a list of page
5496 * @page_list: list of pages to uncharge
5497 *
5498 * Uncharge a list of pages previously charged with
5499 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5500 */
5501void mem_cgroup_uncharge_list(struct list_head *page_list)
5502{
5503 if (mem_cgroup_disabled())
5504 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005505
Johannes Weiner747db952014-08-08 14:19:24 -07005506 if (!list_empty(page_list))
5507 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005508}
5509
5510/**
Hugh Dickins45637ba2015-11-05 18:49:40 -08005511 * mem_cgroup_replace_page - migrate a charge to another page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005512 * @oldpage: currently charged page
5513 * @newpage: page to transfer the charge to
Michal Hockof5e03a42015-02-05 12:25:14 -08005514 * @lrucare: either or both pages might be on the LRU already
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005515 *
5516 * Migrate the charge from @oldpage to @newpage.
5517 *
5518 * Both pages must be locked, @newpage->mapping must be set up.
5519 */
Hugh Dickins45637ba2015-11-05 18:49:40 -08005520void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005521{
Johannes Weiner29833312014-12-10 15:44:02 -08005522 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005523 int isolated;
5524
5525 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5526 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005527 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005528 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5529 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005530
5531 if (mem_cgroup_disabled())
5532 return;
5533
5534 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005535 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005536 return;
5537
Hugh Dickins45637ba2015-11-05 18:49:40 -08005538 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005539 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005540 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005541 return;
5542
Hugh Dickins45637ba2015-11-05 18:49:40 -08005543 lock_page_lru(oldpage, &isolated);
Johannes Weiner1306a852014-12-10 15:44:52 -08005544 oldpage->mem_cgroup = NULL;
Hugh Dickins45637ba2015-11-05 18:49:40 -08005545 unlock_page_lru(oldpage, isolated);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005546
Hugh Dickins45637ba2015-11-05 18:49:40 -08005547 commit_charge(newpage, memcg, true);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005548}
5549
Michal Hocko2d110852013-02-22 16:34:43 -08005550/*
Michal Hocko10813122013-02-22 16:35:41 -08005551 * subsys_initcall() for memory controller.
5552 *
5553 * Some parts like hotcpu_notifier() have to be initialized from this context
5554 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5555 * everything that doesn't depend on a specific mem_cgroup structure should
5556 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005557 */
5558static int __init mem_cgroup_init(void)
5559{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005560 int cpu, node;
5561
Michal Hocko2d110852013-02-22 16:34:43 -08005562 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005563
5564 for_each_possible_cpu(cpu)
5565 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5566 drain_local_stock);
5567
5568 for_each_node(node) {
5569 struct mem_cgroup_tree_per_node *rtpn;
5570 int zone;
5571
5572 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5573 node_online(node) ? node : NUMA_NO_NODE);
5574
5575 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5576 struct mem_cgroup_tree_per_zone *rtpz;
5577
5578 rtpz = &rtpn->rb_tree_per_zone[zone];
5579 rtpz->rb_root = RB_ROOT;
5580 spin_lock_init(&rtpz->lock);
5581 }
5582 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5583 }
5584
Michal Hocko2d110852013-02-22 16:34:43 -08005585 return 0;
5586}
5587subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005588
5589#ifdef CONFIG_MEMCG_SWAP
5590/**
5591 * mem_cgroup_swapout - transfer a memsw charge to swap
5592 * @page: page whose memsw charge to transfer
5593 * @entry: swap entry to move the charge to
5594 *
5595 * Transfer the memsw charge of @page to @entry.
5596 */
5597void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5598{
5599 struct mem_cgroup *memcg;
5600 unsigned short oldid;
5601
5602 VM_BUG_ON_PAGE(PageLRU(page), page);
5603 VM_BUG_ON_PAGE(page_count(page), page);
5604
5605 if (!do_swap_account)
5606 return;
5607
5608 memcg = page->mem_cgroup;
5609
5610 /* Readahead page, never charged */
5611 if (!memcg)
5612 return;
5613
5614 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5615 VM_BUG_ON_PAGE(oldid, page);
5616 mem_cgroup_swap_statistics(memcg, true);
5617
5618 page->mem_cgroup = NULL;
5619
5620 if (!mem_cgroup_is_root(memcg))
5621 page_counter_uncharge(&memcg->memory, 1);
5622
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005623 /*
5624 * Interrupts should be disabled here because the caller holds the
5625 * mapping->tree_lock lock which is taken with interrupts-off. It is
5626 * important here to have the interrupts disabled because it is the
5627 * only synchronisation we have for udpating the per-CPU variables.
5628 */
5629 VM_BUG_ON(!irqs_disabled());
Johannes Weiner21afa382015-02-11 15:26:36 -08005630 mem_cgroup_charge_statistics(memcg, page, -1);
5631 memcg_check_events(memcg, page);
5632}
5633
5634/**
5635 * mem_cgroup_uncharge_swap - uncharge a swap entry
5636 * @entry: swap entry to uncharge
5637 *
5638 * Drop the memsw charge associated with @entry.
5639 */
5640void mem_cgroup_uncharge_swap(swp_entry_t entry)
5641{
5642 struct mem_cgroup *memcg;
5643 unsigned short id;
5644
5645 if (!do_swap_account)
5646 return;
5647
5648 id = swap_cgroup_record(entry, 0);
5649 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005650 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005651 if (memcg) {
5652 if (!mem_cgroup_is_root(memcg))
5653 page_counter_uncharge(&memcg->memsw, 1);
5654 mem_cgroup_swap_statistics(memcg, false);
5655 css_put(&memcg->css);
5656 }
5657 rcu_read_unlock();
5658}
5659
5660/* for remember boot option*/
5661#ifdef CONFIG_MEMCG_SWAP_ENABLED
5662static int really_do_swap_account __initdata = 1;
5663#else
5664static int really_do_swap_account __initdata;
5665#endif
5666
5667static int __init enable_swap_account(char *s)
5668{
5669 if (!strcmp(s, "1"))
5670 really_do_swap_account = 1;
5671 else if (!strcmp(s, "0"))
5672 really_do_swap_account = 0;
5673 return 1;
5674}
5675__setup("swapaccount=", enable_swap_account);
5676
5677static struct cftype memsw_cgroup_files[] = {
5678 {
5679 .name = "memsw.usage_in_bytes",
5680 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5681 .read_u64 = mem_cgroup_read_u64,
5682 },
5683 {
5684 .name = "memsw.max_usage_in_bytes",
5685 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5686 .write = mem_cgroup_reset,
5687 .read_u64 = mem_cgroup_read_u64,
5688 },
5689 {
5690 .name = "memsw.limit_in_bytes",
5691 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5692 .write = mem_cgroup_write,
5693 .read_u64 = mem_cgroup_read_u64,
5694 },
5695 {
5696 .name = "memsw.failcnt",
5697 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5698 .write = mem_cgroup_reset,
5699 .read_u64 = mem_cgroup_read_u64,
5700 },
5701 { }, /* terminate */
5702};
5703
5704static int __init mem_cgroup_swap_init(void)
5705{
5706 if (!mem_cgroup_disabled() && really_do_swap_account) {
5707 do_swap_account = 1;
5708 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5709 memsw_cgroup_files));
5710 }
5711 return 0;
5712}
5713subsys_initcall(mem_cgroup_swap_init);
5714
5715#endif /* CONFIG_MEMCG_SWAP */